01 2003 2162 1 txt EN 28-02-2005 10:05 Pagina 4 · 01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 8....

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 4

Impact of education and training

Third report on vocational training research in Europe: background report

Pascaline Descy,

Manfred Tessaring (eds)

Cedefop Reference series; 54

Luxembourg: Office for Official Publications of the European Communities, 2004

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 1

A great deal of additional information on the European Union is available on the Internet.

It can be accessed through the Europa server (http://europa.eu.int).

Cataloguing data can be found at the end of this publication.

Luxembourg: Office for Official Publications of the European Communities, 2004

ISBN 92-896-0277-5

© European Centre for the Development of Vocational Training, 2004

All rights reserved

Printed in Italy

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 2

The European Centre for the Development of

Vocational Training (Cedefop) is the European Union’s

reference centre for vocational education and training.

We provide information on and analyses

of vocational education and training systems,

policies, research and pratice.

Cedefop was established in 1975

by Council Regulation (EEC) No 337/75.

Europe 123

GR-57001 Thessaloniki (Pylea)

Postal Adress:

PO Box 22427

GR-55102 Thessaloniki

Tel. (30) 23 10 49 01 11

Fax (30) 23 10 49 00 20

E-mail: [email protected]

Homepage: www.cedefop.eu.int

Interactive website: www.trainingvillage.gr

Edited by:

Cedefop

Pascaline Descy, Manfred Tessaring, Project managers

Published under the responsability of:

Johan van Rens, Director

Stavros Stavrou, Deputy Director

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 3

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 4

Preface 7

The impact of human capital on economic growth: a review 9

Rob A. Wilson, Geoff Briscoe

Empirical analysis of human capital development and economic growth in European 71

regions

Hiro Izushi, Robert Huggins

Non-material benefits of education, training and skills at a macro level 119

Andy Green, John Preston, Lars-Erik Malmberg

Macroeconometric evaluation of active labour-market policy – a case study for Germany 179

Reinhard Hujer, Marco Caliendo, Christopher Zeiss

Active policies and measures: impact on integration and reintegration in the labour 215

market and social life

Kenneth Walsh, David J. Parsons

The impact of human capital and human capital investments on company performance 261

Evidence from literature and European survey results

Bo Hansson, Ulf Johanson, Karl-Heinz Leitner

The benefits of education, training and skills from an individual life-course perspective 321

with a particular focus on life-course and biographical research

Maren Heise, Wolfgang Meyer

List of authors 382

Content

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 5

Other volumes of the background report

The background report to the third Research report is composed of two other volumes published sepa-rately the content of which is detailed below.

The foundations of evaluation and impact

research

Descy P., Tessaring M. (eds)

Preface

Philosophies and types of evaluation researchElliot Stern

Developing standards to evaluate vocationaleducation and training programmesWolfgang Beywl, Sandra Speer

Methods and limitations of evaluation and impactresearchReinhard Hujer, Marco Caliendo, Dubravko Radic

From project to policy evaluation in vocationaleducation and training – possible concepts andtools.Evidence from countries in transition.Evelyn Viertel, Søren P. Nielsen, David L. Parkes,Søren Poulsen

Look, listen and learn: an international evaluationof adult learningBeatriz Pont, Patrick Werquin

Measurement and evaluation of competenceGerald A. Straka

An overarching conceptual framework forassessing key competences. Lessons from aninterdisciplinary and policy-oriented approachDominique Simone Rychen

Evaluation of systems and programmes

Descy P., Tessaring M. (eds)

Preface

Evaluating the impact of reforms of vocationaleducation and training: examples of practiceMike Coles

Evaluating systems’ reform in vocational educa-tion and training. Learning from Danish and DutchcasesLoek Nieuwenhuis, Hanne Shapiro

Evaluation of EU and international programmesand initiatives promoting mobility – selected casestudiesWolfgang Hellwig, Uwe Lauterbach, Hermann-Günter Hesse, Sabine Fabriz

Consultancy for free? Evaluation practice in theEuropean Union and central and eastern EuropeFindings from selected EU programmesBernd Baumgartl, Olga Strietska-Ilina, GerhardSchaumberger

Quasi-market reforms in employment and trainingservices: first experiences and evaluation resultsLudo Struyven, Geert Steurs

Evaluation activities in the European CommissionJosep Molsosa

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 6

Preface

The series of reports on vocational education andtraining (VET) research have been published byCedefop since 1998 (1). The reports provide acomprehensive review of current research ininitial and continuing VET in Europe, researchresults and their implications for policy, practiceand future research. Attention is also paid to thetheoretical and methodological foundations anddue reference given to relations with institutional,economic, social, demographic, and other fieldsof social action.

Each research report consists of a backgroundreport of several volumes with contributions fromrenowned researchers (this publication), and asynthesis report developed by Cedefop experts.

Third research report: evaluation and impact

of education and training

This third report informs on current research onevaluation and the impact of VET on individuals,enterprises and society and economy in general,including:(a) the assessment of education and training

systems;(b) the implementation and outcomes of

programmes and reforms with a VET compo-nent;

(c) the costs and benefits of skills, qualificationsand competences which accrue to differentactors at different levels;

(d) impact research, quantifying the contribution ofeducation, training and skills on, for example,earnings, economic growth, employment andsocial inclusion.

The report thus serves both to inform andimprove policy and practice, and further developresearch methodologies. It contributes toachieving the overall European goals expressedby the European Council at its Lisbon andfollow-up summits.

To demonstrate the contribution of educationand training towards achieving a knowledge-based

society and to specify the diverse benefits at alllevels – such as done by ‘summative evaluation’ orimpact research – is as important as indicating waysto improve the design and implementation ofeducation and training programmes or measuresby ‘formative evaluation’.

The background report

The background report gathers contributionsfrom renowned experts and researchers. Theyallow the reader to approach evaluation andimpact research from various angles: individual,enterprise and macro-system level by alsoconsidering essential basics on the foundation,approaches, standards – and limits – of evalua-tion and impact research.

Contributions have been regrouped into threebroad themes, published in separate volumes:(a) impact of education and training;(b) the foundations of evaluation and impact

research;(c) evaluation of systems and programmes.

Impact of education and training

The present volume addresses research work toassess the material and non-material impact ofeducation, training and skills at macroeconomicand regional levels, and the impact on enterprisesand individuals. The picture is completed by casestudies on the impact of active labour-market poli-cies (ALMP) with training components, in botheffects for the economy and social integration.

At the macro level, R. Wilson and G. Briscoediscuss the contribution of education, trainingand skills to economic growth and productivity.H. Izushi and R. Huggins complement this bymeasuring the impact of education and training atregional level. A. Green, J. Preston and L.Malmberg review research on the non-materialbenefits of education and training such as on citi-zenship, health, crime, trust and political partici-pation.

(1) Tessaring, 1998; Cedefop, 1998; Descy and Tessaring, 2001a and b.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 7

From a company perspective, B. Hansson, U. Johanson and K.-H. Leitner discuss the impactof training on company performance, in terms ofproduction growth, productivity and competitive-ness, particularly in the knowledge-intensivesectors of the economy.

Finally, M. Heise and W. Meyer review researchwork that investigates the influence of educationand training in the life course of individuals, ontheir biographies and working careers.

The volume is completed by studies on activepolicies, including active labour-market policies(ALMP) which include a training component.K. Walsh and D. J. Parsons discuss their impacton integration into the labour market, whileR. Hujer, M. Caliendo and C. Zeiss present a casestudy on macroeconomic evaluation of ALMP inGermany.

Stavros Manfred Pascaline

Stavrou Tessaring Descy

References

Cedefop. Vocational education and training – theEuropean research field: background report(2 volumes). Luxembourg: Office for OfficialPublications of the European Communities,1998 (Cedefop Reference series).

Descy, P.; Tessaring, M. Training and learning forcompetence: second report on vocationaltraining research in Europe: synthesis report.Luxembourg: Office for Official Publications ofthe European Communities, 2001a (CedefopReference series, 6).

Descy, P.; Tessaring, M. (eds) Training in Europe:second report on vocational training research inEurope 2000: background report (3 volumes).Luxembourg: Office for Official Publications ofthe European Communities, 2001b (CedefopReference series, 3008).

Tessaring, M. Training for a changing society: areport on current vocational education andtraining research in Europe (second edition).Luxembourg: Office for Official Publications ofthe European Communities, 1999 (CedefopReference series).

Impact of education and training8

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 8

The impact of human capitalon economic growth: a review

Rob A. Wilson, Geoff Briscoe

AbstractThis review provides an in-depth appraisal of a wide body of international research that examines thelinks between education and training in a country and its macroeconomic growth. An initial analysis ofbroad statistics for all EU Member States suggests a loose correlation between investment in humanresources and growth in gross national product (GNP), but clear causal relationships are difficult toestablish. Increased investment in education is shown to lead to higher productivity and earnings for theindividual and similarly, such investment results in significant social rates of return. The returns on invest-ment in vocational training are more difficult to demonstrate. This study reviews a large number of growthmodels that attempt to specify and quantify the GNP and human resource relationship. Wide differencesare found in the model specifications, the quality of the data inputs and the results obtained. Other linksbetween investment in human capital and economic performance are reviewed using diverse literaturesources on human resource management, corporate market value, company size and industry structure.The indirect impact of education on non-economic benefits is also examined in the context of the tech-nological, spatial and environmental gains to society. It is concluded that, overall, the impact of invest-ment in education and training on national economic growth is positive and significant. Some policyconclusions are drawn and directions for future research in this area are suggested.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 9

1. Executive summary: key findings 13

1.1. Background 13

1.2. Rates of return on investment in human capital 13

1.3. Skills and organisational performance 14

1.4. Macroeconomic performance and education 14

1.5. Other links between investment in human capital and performance 14

1.6. Spill-over, external effects and non-economic benefits 15

2. Introduction 16

2.1. Aims of the study 16

2.2. Content of the study 16

2.3. Approach 16

2.4. Limitations of the study 17

2.5. Structure of the present report 18

3. Statistics of growth, education and training 19

3.1. Economic growth 19

3.2. Employment and unemployment 20

3.3. Educational participation 22

3.4. Educational qualifications of the workforce 24

3.5. Vocational training 26

4. Rates of return on education and training 29

4.1. General literature strands 29

4.2. Rates of return on investment in human capital 29

4.3. Private versus social rates of return 31

5. Human resources and performance 33

5.1. Links between human resources and performance at company level 33

5.2. Management skills and organisational performance 33

5.3. Empirical studies at company level 34

5.4. Empirical studies at the industry and sector level 35

6. The role of education and training in economic growth 37

6.1. General literature strands 37

6.2. Growth accounting literature 38

6.3. Neoclassical growth models 39

6.4. Endogenous growth models 40

6.5. Empirical growth regressions 41

6.6. A recent example of growth modelling 44

6.7. The influence of data quality on results 46

6.8. Human capital externalities 47

6.9. Social capital and economic growth 47

6.10.Possible endogeneity and simultaneity bias 48

Table of contents

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 10

7. Market value and related literature 50

7.1. Literature strands 50

7.2. Some empirical results on market value models 50

7.3. Knowledge, intellectual capital and intangible asset literature 51

8. Further links between education and training and company performance 53


8.2. Company size and the problems of small firms 53

8.3. Small and medium size enterprises and the qualifications of managers 54

8.4. Structure-conduct-performance models 54

9. Spill-over effects and externalities 56


9.2. Technological spill-overs 56

9.3. Spatial and city dynamics 56

9.4. Environmental externalities 57

10. Conclusions 59

10.1.Summary of key points 59

10.2.Some policy implications 60

10.3.Possible areas of future research 61

List of abbreviations 63

Annex 64

References 65


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 11

Tables

Table 1: Estimated growth rates of GDP per capita 19

Table 2: Growth in total employment – average annual percentage changes 20

Table 3: Unemployment rates – average percentages 21

Table 4: Total expenditure on education as percentage of GNP 22

Table 5: Percentage of population aged 18-22 entering tertiary education 23

Table 6: Educational attainment of the adult population 24

Table 7: Average schooling and no schooling rates 1970-2000 26

Table 8: Comparisons of vocational training based on European labour force survey 28

Table 9: Barro growth regressions 45

Figures

Figure 1: Average growth and employment rates in EU 21

Figure 2: Highest educational attainment of the adult population 25

Figure 3: Average years of schooling 25

Figure 4: Schooling attainment 1970-2000: average years of schooling 27

Figure 5: Schooling attainment 1970-2000: percentage of population 15+ with no schooling 27

List of tables and figures

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 12

1.1. Background

Various conceptual approaches have been usedto explore the links between education andtraining and economic performance. The aim ofthe present review is to provide a critical assess-ment of this work, focusing primarily on themacroeconomic level. In order to reach a macrooverview, it is necessary to consider many otherbranches of literature than those just concernedwith macroeconomic growth models.

The study, therefore, presents an extensiveinternational review into a number of strands ofliterature that focus on the link between invest-ment in education, training and skills andeconomic performance at the macro level. Thisincludes some micro level research, where trainingand education conducted by firms and industriesis deemed to be an important contributor to overalleconomic growth. While a majority of the publica-tions examined are in English, much Europeanresearch is also incorporated in the form of empir-ical studies using European data and Europeanauthors publishing in UK and US publications.

All of the European Union (EU) 15 MemberStates have experienced significant growth overthe past 30-year period. However, there is varia-tion between countries, especially in the shortterm. Educational expenditure as a percentage ofgross national product (GNP) has been main-tained at roughly constant levels across all theMember States over the long term. This hasensured increased investment in human resourcesthroughout the EU. Other data show a stronggrowth trend in the percentage of the eligiblepopulation entering tertiary education over thethree decades. Consistent statistics on trends inEU vocational training prove more difficult toobtain, but there is evidence of significant voca-tional training inputs in a number of EU countries.A simple reading of these statistics indicates thatincreases in economic growth across the EU areassociated with increases in both education andtraining. However, more detailed comparisonsalso illustrate the difficulties in establishing acausal link between educational and training

inputs on the one hand and economic outputs (inform of growth in gross domestic product – GDP)on the other. Similar patterns are observed inother developed economies.

1.2. Rates of return oninvestment in human capital

A key body of literature relates to the rates ofreturn for the individual (and for society ingeneral) of investment in human capital, in theform of education and training. Much of thisresearch draws on the seminal work by Becker(1964), Mincer (1974) and many others. This bodyof work is founded on a microeconomicapproach. Nevertheless the results have impor-tant macroeconomic implications. They highlightthe strong links observed between education,productivity and output levels. Although somehave questioned the direction of causality andargued that much education simply acts as ascreening device to help employers to identifymore able individuals, the general consensusseems to be that education does result in higherindividual productivity and earnings. On balance,the results suggest a strong and positive causallink between investment in education and trainingand earnings. This applies both at the level of theindividual and also to the broader social returnson such investments, the evidence suggestingsubstantial social as well as private benefits. Theimplication is that what is good for the individualis also good for society at large at a macro level.

Evidence on social rates of return concentrateson the benefits to the economy arising fromincreased education by calculating all the costsof schooling and education compared to the rela-tive pre-tax earnings of individuals in receipt ofsuch education. Many qualifying assumptionsneed to be made in arriving at a quantifiable esti-mate, but several studies suggest a social rate ofreturn in the range 6 to 12 %. This compares wellwith rates of return on other capital. Comparablestudies to establish the social rate of return for

1. Executive summary: key findings

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 13


vocational training are hard to find, although thereis some evidence that such training has a signifi-cant positive impact on productivity.

1.3. Skills and organisationalperformance

Another key focus of recent research has been theuse and deployment of human resources and skillsat an organisational level and how this is linked tothe economic performance of organisations andeconomies more broadly. While much of thisresearch lies outside main economic literature,research on the management of human resourcessuggests a range of results concerning variousaspects of human capital and firm performance.This is not really surprising, as there is a widely heldbelief that people are a key source of an organisa-tion’s competitive advantage. It is the quality of thehuman resources that determines organisationalperformance. There are a number of differentemphases in this research, including the role playedby human resource management (HRM) inpromoting such resources, and the role of manage-ment itself – particularly top managers. One impor-tant theme is the role of leadership, with a focus onthe role of managers, especially top executives, ininfluencing company performance. Their skills,education and training are important in what areoften very complex processes.

1.4. Macroeconomic performanceand education

Literature on economic growth models exploresdirectly the quantitative relationship betweeninvestments in education and training and thelevel and growth of per capita GDP. There are alarge number of studies, beginning with the clas-sical growth models first developed in the 1950s,through to the so-called endogenous growthmodels that are still widely applied in manycurrent empirical studies. Both data sets andeconometric modelling techniques have devel-oped extensively over recent years and manydifferent model specifications have beenproposed and empirically tested. Typically, thesemodels use data drawn from a cross-section of

countries, sometimes only for developed coun-tries but often for a wider set. Data difficultiescommonly mean that consistent series of educa-tional variables are hard to obtain over sufficientlylong periods to facilitate econometric time seriesanalyses. For developed countries over morerecent years, cross-sectional and time series dataare combined into panel sets to enable a morecomprehensive testing of the relationshipbetween human resources and economic growth.

This body of research can be divided up into anumber of subsections. The so-called ‘growthaccounting’ literature emphasises the importanceof measuring changes in the quality of labour, asindicated by improved qualifications and higherskills, when trying to account for economicgrowth over the long term. The impact of theaccumulation of knowledge though the under-taking of research and development (R&D) hasalso been a key feature of this body of research.

Other subsections focus on technical researchinto production and related functions, which areconcerned with the relationships between factorinputs (both tangible and intangible, such asknowledge) and economic outputs. The so-called‘new growth theories’ are also very relevant here.These highlight the determinants of economicgrowth in the broadest sense, concentrating onhuman capital inputs. There are a number ofdistinguishing features of the new growth models,but, essentially, they extend the existing modelsby endogenising technological change (hence,they are also sometimes referred to as endoge-nous growth theories). This contrasts with tradi-tional neoclassical models, where economicgrowth is driven by the increase in factor inputs(i.e. population growth) and by the exogenousrate of (labour augmenting) technological change.The newer models often allow for increasingreturns to scale, where growth is unbounded andin which growth rates can continue to increaseindefinitely over time.

1.5. Other links betweeninvestment in human capitaland performance

There are a number of other areas of researchthat have a bearing on economic growth issues.Further important evidence comes from market

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 14


value literature which is concerned with howintangible assets, such as technical knowledge,influence potential performance and net worth.

While macro growth models have developedthe production function approach to explainmacroeconomic growth, market valuation hasdeveloped the same approach to explain thegrowth of individual companies. This also hasimportant macroeconomic consequences.Research studies in market value are concernedwith how intangible assets, such as technicalknowledge, influence potential performance andnet worth. These are intimately tied up withinvestment in education and training as well asskills, which may often be necessary, if not suffi-cient, to improve such intangible assets.Research and development, patents and intellec-tual capital are identified as key determinants ofcorporate market value. These same variableswere also found to be important contributors tomacroeconomic growth in the new growth theoryliterature.

Other research has emphasised the impor-tance of knowledge, intellectual capital and otherintangible assets, in influencing the performanceof companies at a micro level and economies at amacro level. The role of firm size, includinginsights from the structure-conduct-performanceliterature (which highlights the importance of sizeand monopoly power in exploiting new knowl-edge) and also the role of qualifications andquality of management in the performance ofsmaller firms, need to be considered. Work in thisarea has been carried out for some time andmany of the studies emphasise the role of thesmall firm in economic growth. Such small firmsare an extremely heterogeneous group, that havemarkedly different training needs from those oftheir larger counterparts. There is found to be farless emphasis on educational and training qualifi-cations in small firm environments. Empiricalstudies suggest that conventional theories ofcompany behaviour are far too simplistic forexplaining the goals and aspirations of thesesmaller organisations.

1.6. Spill-over, external effectsand non-economic benefits

The final strands of literature considered here detailsome of the indirect effects of investment in humanresources that are not normally captured inmeasured national economic growth. Suchspill-overs and externalities can be technological,spatial or environmental and economic andnon-economic. Each contributes significant socialgains. Many of these effects are related to otherareas of the literature covered in this review. Thespatial externalities, that embrace city dynamics,demonstrate how geographical clustering of busi-nesses employing highly qualified workersproduces high productivity and strong localeconomic growth. Other significant spill-oversrelate to health and life expectancy, which typicallyincreases as a consequence of higher levels ofeducation.

Such spill-over and related external effects arepotentially very important. In making a traininginvestment, an individual firm takes into account theimpact of the training on its performance, given thecurrent total stock of human capital in the economy.The higher the human capital in the economy, thegreater is the firm’s own performance. However, itdoes not take into account the effect of its owninvestment on the total human capital stock. As faras the individual firm is concerned, the impact ofthat investment is minuscule and need not beconsidered. However, from the perspective of theeconomy as a whole, the totality of training invest-ments by firms can further increase economicoutput and economy-wide performance. Theseexternal effects can add considerably to themacroeconomic consequences of any initial invest-ment in human capital.

External benefits are often not economic.They also include improvements to the environ-ment, better health and reduced crime rates.Serious attempts are now being made to quan-tify these in economic terms. These are majorareas of study in their own right and are onlytouched on here.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 15

2.1. Aims of the study

The main objective of this study is to provide acomprehensive and critical overview of the impactof education and training on economic perfor-mance, and, by implication, employment opportu-nities, at the macro level. In this context the term‘macro’ is defined quite widely to incorporate firmsand industries, as well as the general economy. Itis argued that there is a wide range of differentstudies that have a bearing on the link betweeninvestment in human capital and economicgrowth. These studies need to be synthesised inorder to obtain a comprehensive view of the manyways in which education, training and the accumu-lation and appropriate deployment of skills influ-ence economic performance.

The review considers the implications forpolicy and practice, drawing general lessons forfurther improvement in the economies of membercountries. While attention is given to the role ofvocational education and training, it is muchmore difficult to find relevant evidence in this areathan for more general education.

2.2. Content of the study

The expansion of formal education and training indeveloped economies in recent years has hadsubstantial and easily observed implications forthe skill levels and skill structures of the popula-tions and employed workforces of these coun-tries. However, the contribution of education andtraining to economic growth and other measuresof performance is less clear.

Simple statistics can be produced to demon-strate a well-established correlation betweeninvestment in human capital and economic indi-cators such as GDP. However, the direction ofcausality remains the subject of heated debate.The implications for future education and trainingpolicies are the subject of continuing researchand are addressed by this study.

A number of key areas can be identified, each

of which has an important bearing on the issuesconsidered here. The main ones include:(a) rates of return on investment in human capital

and the link between private rates of returnfrom education and training and social returnsfor society at large;

(b) links between the use and deployment ofhuman resources and economic performance,including literature on management skills andorganisational performance and general litera-ture on HRM, policy and performance;

(c) literature on the causes of economic growth,including the growth accounting studies,production and knowledge functions relatingGDP to various determinants and the newgrowth theories which endogenise thesources of economic growth;

(d) literature on market valuation of companiesand the role of R&D (a proxy for educationand training in higher level skills) and intellec-tual capital;

(e) the role of firm size and structure-conduct-perfor-mance theories of growth, with an implied role foreducation and training;

(f) the spill-over effects and externalities that resultfrom higher investments in human capital.

A key objective is to draw together thesedisparate strands and to identify the commonmessages about the links between education,training and skills and various indicators ofperformance.

2.3. Approach

The approach adopted is to conduct a series ofextensive international literature reviews ofresearch in each of these areas, coveringeconomic, econometric, business organisationand other related disciplines. These each focuson different aspects of the relationship betweeninvestment in education, training and skills andeconomic performance. It is argued that it is alsonecessary to consider some micro level research,which has a bearing on outcomes at a broadermacro level. In particular, research into firms and

2. Introduction

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 16


industries is considered very important in under-standing the contribution of education andtraining to overall economic growth. The paperbuilds upon and extends the analysis presentedin the two previous Cedefop reports edited byTessaring (1998) and Descy and Tessaring (2001).

The aim was to include as much material aspossible from non-English speaking countries, aswell as to provide a greater focus on VET asopposed to more general education. However,this proved much easier said than done and thepresent draft is heavily reliant on Englishlanguage material (particularly American research)and the effects of more general education. Thisreflects the balance of studies currently available.In the country comparative analyses that arereported in the section on economic growthmodelling, much of the analysis focuses on inter-national comparisons, often based on OECDstatistics. Although many of the authors emanatefrom non-English speaking countries, theypublish in English.

2.4. Limitations of the study

The research explores the many diverse studieswhich touch upon the relationship betweeneducation, training and skills and performance inall its aspects. A variety of variables are used asindicators of investment in skills and investmentin human capital. Measures of performance at amacroeconomic level are equally varied. Thediversity of skills-performance literature iscaused, in part, by the range of both performanceand input measures used. These include differ-ences in the level of aggregation and whether thefocus of the research is concerned with private orsocial returns. The latter are potentially very wide,encompassing all kinds of spill-over and externaleffects, including impacts on the general environ-ment. Many of these effects are especially diffi-cult to quantify. The associated study by Greenet al. (2003) into the non-material benefits ofeducation, training and skills focuses moredirectly on these issues.

In investigating the relationships betweeneconomic performance and education,economists have focused on a range of measuressuch as the rate of economic growth, the level andrate of growth in income and wealth, the level and

change in unemployment, export performance,and the like. At an individual level more emphasisis usually placed on the link with earnings. Empir-ical investigations have been undertaken at bothmicro and macro levels, including individual,establishment, firm, industry and economy-widelevel studies. The present study covers all ofthese.

Education and training are key contributors tothe development of skills and knowledge. Typi-cally, but not always, the latter are reflected in theaward of some form of qualification or credential.It is these measured qualifications that arenormally used to capture the human resourceinput into the production and growth process.Where such measures are not available, years ofcompleted schooling are often used instead.

At the organisational level, more complex indica-tors are often deployed. Higher levels of compe-tences should result in superior performance by theindividual and, other things being equal, by theorganisation in which the individual works. Whensuch organisations are aggregated together, thehigher competences should be reflected in highernational growth. Campbell (2001) has recentlyargued, for example, that, for the UK, long-termstudies show a considerable contribution byeducation to economic growth at a local level.

Most studies of economic performance at thelevel of individual establishments have been ofprivate sector companies. While there are exam-ples of economic studies of public organisationsand institutions, these are relatively rare and it isusually necessary to go to other disciplines, such asHRM, to find more information. There are also other,more holistic, approaches to performance thatfocus mainly on geographical areas. For example,city dynamics forms a very large and growing areaof research undertaken by economic geographers.This is given brief coverage in the present review,but the associated study by Izushi and Huggins(2003) analyses the impact of investment in humancapital on regional growth across the EU.

It is important to emphasise that the presentreview does not attempt to assess the precisevalues of the parameters reported, although it isconcerned with the direction and strength of therelationships. The focus is more on the nature ofthe key indicators and the variables used instudies and the general nature of the relation-ships established.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 17


2.5. Structure of the present report

The remainder of this paper is divided into sevenmain sections, each dealing with a particular areaof literature.

Chapter 3 presents a brief overview of statis-tics on economic growth, employment, unem-ployment and various measures of investment ineducation and training. While this highlights thestrong correlation between some of these indica-tors, it also emphasises the difficulties in identi-fying and establishing causal relationshipsbetween the two.

Chapter 4 examines evidence of such links,focussing on the conventional rate of return oneducation and training. Although much of thisliterature is now based on the use of micro level,individual data sets on earnings, its conclusionshave very important macroeconomic conse-quences. This includes the aggregate effects ofinvestments by individuals as well as the widersocial returns.

Chapter 5 concentrates on the links betweeninvestment in skills and performance at the levelof the organisation or firm. This has focussed onthe use and deployment of human resources inthe broadest sense. While much of this researchhas a micro focus, in terms of the units of obser-vation examined, it has crucial macro implicationsfor the performance of economies at muchbroader sectoral and national levels.

Chapter 6 focuses more directly on macroeco-

nomic literature. It highlights the increasing rangeof evidence of the links between investment ineducation and training and economic growth.This covers a range of different approaches fromthe classical growth models of the 1950s and1960s, through growth accounting methods tothe new endogenous growth theories, whichplace education and training and other forms ofinvestment in human capital at the centre ofthings.

There are a number of other areas of researchwhich have focussed on the link between invest-ment in human capital and economic perfor-mance. These are covered in Chapter 7, whichhighlights, in particular, some of the insights frommarket value literature. It also covers otherstudies concerned with the value of intangibleassets, much of which is tied up in some form ofhuman capital.

Chapter 8 covers a further series of researchareas, all of which have some bearing on thistopic, including structure conduct performanceliterature. Chapter 9 provides a round-up of workthat has emphasised the externalities andspill-over effects of investment in human capital.These encompass a large range of both economicand non-economic benefits. In total, these consti-tute very significant macro effects of many indi-vidual investments in education and training byindividuals, companies and other organisations.

Chapter 10 gives an overall summary of thefindings, draws some implications for policy andmakes suggestions for possible future work.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 18

Many international agencies collate data on GDP,employment and unemployment. Others providea wide array of statistics on educational spendingand qualifications within different countries.Generally there is good availability of this type ofdata for the Member States of the EU, coveringthe period after 1970. Some problems arise withconsistency and comparability of the data setsthrough time. Quantitative measures of vocationaltraining in Member States are much more difficultto obtain and, most commonly, such data areonly available for a limited number of countries,and then only for more recent years.

3.1. Economic growth

Table 1 presents estimated annual GDP per capitagrowth rates (1) for the 15 Member States for various5-year periods between 1970 and the year 2000.These data are based on the main economic indi-cators series published by OECD; the growth ratesare based on GDP per capita estimates, allexpressed in 1995 constant prices and using USDexchange rates for the same year. The averageannual growth rate for the 30-year period is given inthe right-hand column of Table 1. This shows howall countries have experienced significant

(1) GDP per capita is the most widely adopted measure of economic growth, as it standardises for the size of country. Alternativegrowth measures, such as the growth of GDP without any adjustment for population size, may well show a different picture.

EU Member 1970-75 1975-80 1980-85 1985-90 1990-95 1995-2000 1970-2000 States

Belgium 3.46 2.91 0.08 2.99 0.90 2.81 2.18

Denmark 2.19 2.50 0.47 1.41 1.63 2.85 1.84

Germany* 2.25 3.34 0.68 2.91 -3.28 1.91 1.28

Greece 5.07 4.41 0.07 1.72 0.59 3.42 2.53

Spain 5.52 1.96 1.08 4.49 0.91 3.80 2.95

France 4.00 3.27 0.35 3.20 0.38 2.68 2.30

Ireland 4.93 4.48 0.27 4.48 3.56 9.90 4.56

Italy 2.41 3.85 0.81 3.03 1.08 0.11 1.87

Luxembourg 3.23 2.61 1.18 4.69 1.59 6.33 3.26

Netherlands 3.22 2.61 -0.54 3.11 0.93 3.68 2.16

Austria 3.94 3.41 1.05 3.00 1.50 2.57 2.58

Portugal 4.73 5.27 0.75 5.11 0.98 3.82 3.43

Finland 4.09 3.20 1.87 3.39 -1.51 2.07 2.17

Sweden 2.57 1.33 1.01 2.29 -0.54 2.96 1.60

United Kingdom 2.17 1.64 1.30 3.28 0.85 2.85 2.01

EU-15 (average) 3.28 2.93 0.73 3.18 0.48 3.51 2.34

* Discontinuity in German series between 1990 and 1995 is attributable to Unification.Source: OECD main economic indicators

3. Statistics of growth, education and training

Table 1: Estimated growth rates of GDP per capita

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 19


economic growth in recent decades, with an EUannual average growth rate of about 2.3 %. Theaverage for Germany turns out to be artificially lowas a result of much slower growth (decline) in theperiod after unification. Ireland, a relatively smallcountry, has the highest growth rate of all the15 Member States. There is significant variationaround the EU long-term average. Sweden showsone of the lowest growth rates over this period.

Over time, economic growth rates differ acrossthe respective five-year periods shown in Table 1.For most of these EU countries growth was rela-tively strong in the 1970s, but the period from 1980to 1985 produced a marked slowdown for all coun-tries, with average growth rates below 1 %. Thesecond half of the 1980s resulted in much strongereconomic growth, but the early 1990s producedrelative stagnation for most Member States, withsome countries experiencing GDP per capitadeclines. There was a return to significant growthin the most recent five-year period, although whileIreland averaged annual growth rates approaching10 %, the comparable figure for Italy was onlyabout 1 %.

3.2. Employment and unemployment

Associated with the growth in GDP are changes innational employment levels. Table 2, based onstatistics provided by Eurostat, shows longer-termtrends in average annual rates of growth in totalnational employment. Most of the Member Stateshave experienced positive growth in employmentover each of the past four decades, despite the vari-ation in economic growth rates. Across all theMember States employment has grown over thelonger term, although individual rates of growthshow some variability. Table 3, using consistentEurostat definitions, presents estimates of unem-ployment rates also averaged over decades. Theseshow a significant upward trend, especially in theperiod after 1980. Lower rates of economic growthhave brought about higher unemployment levelsacross the EU, as job creation has failed to keeppace with the numbers in the labour market. Thedifference in unemployment rates between MemberStates is very marked, especially in the 1990s.

Country 1961-70 1971-80 1981-90 1991-2000

Belgium 0.5 0.2 0.1 0.5

Denmark 1.1 0.3 0.3 0.4

Germany 0.2 0.2 0.5 0.3

Greece -0.8 0.7 1.0 0.4

Spain 0.6 -0.6 0.8 1.2

France 0.6 0.5 0.3 0.5

Ireland 0.0 0.9 -0.2 3.7

Italy -0.5 0.7 0.6 0.2

Luxembourg 0.6 1.2 1.7 3.4

Netherlands 1.9 0.7 1.1 1.9

Austria -0.4 0.7 0.1 0.3

Portugal 0.2 0.1 0.2 -0.1

Finland 0.4 0.3 0.5 -0.8

Sweden 0.7 0.8 0.7 -0.7

United Kingdom 0.3 0.3 0.5 0.2

EU-15 (average) 0.3 0.3 0.5 0.4

Source: Eurostat - European economy: 2001 review

Table 2: Growth in total employment – average annual percentage changes

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 20


Country 1961-70 1971-80 1981-90 1991-2000

Belgium 1.9 4.6 9.7 8.7

Denmark 1.0 3.7 8.4 7.1

Germany 0.7 2.2 6.0 8.1

Greece 5.0 2.2 6.4 9.5

Spain 2.5 5.4 18.5 19.6

France 1.8 4.1 9.2 11.3

Ireland 5.4 7.7 14.7 11.1

Italy 4.8 6.1 8.7 10.7

Luxembourg 0.0 0.6 2.5 2.6

Netherlands 0.9 4.4 8.5 5.4

Austria 2.5 1.8 3.3 3.9

Portugal 2.5 5.1 7.3 5.6

Finland 2.3 4.0 4.7 12.5

Sweden 1.7 2.1 2.6 7.7

United Kingdom 1.7 3.8 9.8 8.1

EU-15 (average) 2.2 4.0 9.0 9.9

Source: Eurostat - European economy: 2001 review

Figure 1: Average growth and employment rates in EU

Source: Eurostat statistics

-2

0

2

4

6

8

10

12

1970-1975 1975-1980 1980-1985 1985-1990 1990-1995 1995-2000

GDP p.c. growthAverage unemployment Employment growth

Table 3: Unemployment rates – average percentages

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 21

Figure 1 provides a simple comparison of GDPper capita growth rates, total employment averageannual growth rates and unemployment averagepercentages for the 15 Member States across sixtime intervals over the period 1970-2000. The graphshows how GDP and employment growth moveclosely together. Unemployment has risen steadilyover the longer term.

3.3. Educational participation

International education statistics tend to be lessreadily comparable than economic growth dataand employment data, but organisations such asUnesco and OECD have long published figuresmeasuring educational variables. More recently,results from various European labour surveyshave supplemented these sources of information.Unfortunately, in this area, different surveys oftenseem to produce statistics that can be signifi-cantly at odds with one another; see for examplethe discussion in Barro (2000). Table 4 repro-duces data measuring total expenditure oneducation as a percentage of GNP; this is mainly

derived from various Unesco Statistical Year-books. Since the statistics in Table 4 are onlyconcerned with the 15 Member States, thesedata are considered to be relatively comparable.

Table 4 shows how, overall, EU countries havemaintained educational expenditure as a more orless constant percentage of GNP. Given that theseeconomies have been growing steadily over theperiod since 1970, educational investment hasalso been increasing in line with economic growth.Some Member States significantly increased theireducational expenditures (as percentage of GNP)over the 30-year period; Greece, Spain, Irelandand Portugal are the best examples. The averagepercentage levels were consistently higher inDenmark, Finland and Sweden than in the otherMember States. The data in Table 4 suggest somevariability in education budgets over the five-yearobservation periods. This may be as a result ofpolitical policy changes or it could simply be areflection of changing measurement practices inthe different countries. International time series ofeducation variables can sometimes be ratherinaccurate and most studies, such as thosedescribed in Chapter 6 below, use broad average

Country 1970 1975 1980 1985 1990 1995 1998*

Belgium 6.0 6.2 6.0 6.0 5.0 5.0 4.7

Denmark 6.7 7.6 6.7 7.0 7.1 7.7 7.4

Germany 4.8 5.0

Greece 1.7 1.7 1.8 2.4 2.5 2.9 3.0

Spain 2.0 1.8 2.6 3.3 4.4 4.9 4.8

France 4.8 5.2 5.0 5.8 5.4 6.1 6.0

Ireland 4.8 5.8 6.3 6.4 5.6 6.0 5.8

Italy 3.7 4.1 4.4 5.0 5.2 4.7 4.3

Luxembourg 3.6 4.7 5.7 3.8 3.6 4.1 4.3

Netherlands 7.2 8.1 7.6 6.4 6.0 5.2 5.0

Austria 4.5 5.6 5.5 5.8 5.4 5.6 5.5

Portugal 1.5 3.5 3.8 4.0 4.2 5.3 5.7

Finland 5.9 6.4 5.3 5.4 5.7 7.5 7.9

Sweden 7.6 7.0 9.0 7.7 7.7 8.1 8.5

United Kingdom 5.3 6.6 5.6 4.9 4.9 5.3 5.3

EU-15 (average %) 4.7 5.3 5.4 5.3 5.2 5.5 5.5

NB: * Estimated percentage based on OECD Education at a glance, 2001Source: Unesco statistical yearbooks


Table 4: Total expenditure on education as percentage of GNP

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 22

observations in cross-sectional comparisons,rather than time series analyses.

Comparison of the data in Tables 1 and 4 illus-trates immediately some of the difficulties inestablishing links between indicators of educa-tion and economic growth. There are no obviouspatterns. Sweden, which has one of the highestpercentages of expenditure on education, hasone of the lowest growth rates. Ireland, thefastest growing country, cannot attribute this to ahigher than average proportion of spending oneducation or growth in the percentage. The linksbetween these variables are often complex anddifficult to identify.

Educational expenditure provides only a verybroad measure of investment in humanresources. Other statistics are available on thepercentage of those of eligible age in the popula-tion enrolled in various categories of education.For most Member States, primary and secondaryeducation participation is virtually 100 % andtrends over time have very little significance.However, the percentage of the eligible popula-tion in tertiary education shows marked variabilitybetween EU countries and changes over time are

very apparent. Table 5 summarises this data, asderived from various Unesco statistical year-books. All the Member States show very signifi-cant increases in the percentages enteringtertiary education over the 25-year periodcovered by the data. By 1995, the EU averagewas approaching 50 %, showing a strong trendgrowth, especially in the period since 1985.Again, there is no obvious link between theseindicators and performance in terms of GDPgrowth.

Other indicators of educational investment arealso available, such as statistics on the averagenumber of years of education completed bymembers of the working population in differentcountries. Some detailed studies are availablecomparing qualification rates in particular areasof study, such as mathematics, sciences andforeign languages; see, for example, some of thedata reproduced in Prais (1995). Further statisticsthat try to provide some measure of the quality ofeducational input detail the qualifications of thosewho carry out the teaching in schools and relatededucational establishments.

Virtually all published education statistics show

Country 1970 1975 1980 1985 1990 1995

Belgium 17 22 26 32 40 56

Denmark 19 30 28 29 36 48

Germany 34 46

Greece 13 18 17 24 36 42

Spain 9 20 23 29 37 48

France 19 25 25 30 40 51

Ireland 12 17 18 22 29 40

Italy 17 26 27 25 32 42

Luxembourg

Netherlands 20 25 29 32 40 48

Austria 12 18 22 26 35 48

Portugal 7 10 11 12 23 39

Finland 13 28 32 34 49 55

Sweden 22 30 31 31 32 47

United Kingdom 14 19 19 22 30 50

EU-15 (average %) 15 22 24 25 35 47

NB: * Basic group is ages 18-22, but this varies slightly for different countriesSource: Unesco statistical yearbooks


Table 5: Percentage of population aged 18-22* entering tertiary education

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 23

increasing levels of investment over time acrossall the Member States. These increasing trendsare correlated with economic growth statistics.However, the direction of causality remains to beestablished. Is increased educational investmenta prime factor determining growth in GDP percapita or is this observed increase in education aresult of continuing economic growth? The litera-ture on economic growth reviewed in Chapter 6of this paper suggests that there is a growingweight of evidence that it is the former, although

there is also important feedback in the otherdirection.

3.4. Educational qualifications of the workforce

The OECD has adopted the international stan-dard classification of education (ISCED) systemto facilitate comparisons of educational attain-

Highest completed level of education

Upper secondary Tertiary (%)Average years

or above (%)of schooling

Australia 53 24 11.9

Austria 69 8 11.9

Belgium 53 25 11.7

Canada 75 47 13.2

Czech Republic 83 11 12.4

Denmark 62 20 12.4

Finland 65 21 11.6

France 68 19 11.2

Germany 84 23 13.4

Greece 43 17 10.9

Ireland 47 20 10.8

Italy 35 8 10.0

Netherlands 61 22 12.7

New Zealand 59 25 11.4

Norway 81 29 12.4

Portugal 20 11 10.0

Spain 28 16 11.2

Sweden 75 28 12.1

Switzerland 82 21 12.6

United Kingdom 76 21 12.1

United States 86 33 13.5

OECD average 62 21 11.9

NB: The estimates of average years of schooling relate to total cumulative time spent in formal education over all ISCED levelsfrom the beginning of primary level (ISCED 1) to tertiary level. These estimates are obtained by using data on educational attain-ment of each age group from the labour force survey and applying an estimated average cumulative duration for each level ofeducation. Where there are programmes of different duration at the same ISCED level, a weighted average is taken based onweights corresponding to the number of persons in each broad educational programme.Source: Education at a glance – OECD indicators (OECD, 2001a), indicator A2.1, p. 38 (using data on educational attainment ofindividuals from labour force survey sources, or in the case of Denmark, the register of educational attainment of the popula-tion).


Table 6: Educational attainment of the adult population.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 24

(2) OECD has adopted ISCED, but Unesco remains responsible. Unesco revised ISCED in 1997 with the aim of reflecting evolu-tion in education-training systems and increasing comparability of data. The new ISCED distinguishes between programmeorientation (general education, pre- and vocational education and training) and programme destination (further studies, labourmarket). This means that there exist consistent and comparable data on education and training (but the potential of new ISCEDis still unexplored).

ment between countries (2). ISCED was originallydesigned to serve as an instrument suitable forassembling, compiling and presenting statisticsof education both within individual countries andinternationally. The ISCED typology is described

in further detail in Annex 1. This classificationsystem forms the basis for much of the compara-tive evidence emanating from the OECD oneducation systems (e.g. OECD, 2001a; Steedmanand McIntosh, 2001). International comparisons

Figure 2: Highest educational attainment of the adult population

Source: OECD (1998)

Tertiary

0

25

50

75

100

AS A B CA CZ DK FIN F D EL IRL I NL NO NZ P E S CH UK US

Upper secondary or above

Figure 3: Average years of schooling

Source: OECD (1998)

0

5

10

15

AS A B CA CZ DK FIN F D EL IRL I NL NO NZ P E S CH UK US

Average years of schooling


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 25

of education are typically based on years ofschooling, and/or the highest educational levelachieved.

A comparison between OECD countries ofeducational attainment is provided in Table 6 andFigures 2 and 3. Two measures are provided – theproportions of the working age population whosehighest attainment level is upper-secondary ortertiary (primary level attainment is now almostuniversal in the OECD) and the average numberof years of schooling. As can be seen, there areonly relatively small differences between coun-tries in terms of average years of schoolingacross the OECD.

Recent work by Barro and Lee (1993, 1996,2001) has also focused on comparisons of educa-tional attainment between countries. However,given the wider set of countries they consider –including many developing and transition coun-tries – and the paucity of detailed data (such ashousehold surveys), especially for their historicalseries, they simply focus on average years ofschooling and a broad measure of education levelattained. Table 7 and Figures 4 and 5 providesome details for a range of country groups.

In general, there have been increasing averagelevels of participation in education over the lastthree decades, in all regions, but growth hasbeen particularly strong in the Middle East andNorth Africa. Mean school years have increasedin developed countries too, with the net resultthat there has been a substantial increase in theworld stock of educated people, although littleconvergence between regions. South Asia and

Sub-Saharan Africa still have substantial propor-tions of their populations with very little or noeducation.

A simple comparison of these data with thoseon economic growth rates does not reveal anyobvious and immediate links. Teasing out thecausal links between the two is a complexproblem, as the discussion in subsequentsections makes clear.

3.5. Vocational training

Consistent statistics detailing trends in vocationaltraining across EU countries are often more diffi-cult to obtain, although more recent OECD statis-tics and European labour force surveys (ELFS)provide much useful comparable information inthis area. Table 8 shows some comparative datafor selected Member States for one particularsurvey period. While consistent definitions wereagreed for the ELFS, exactly what sorts of trainingare included and which are excluded by thesurvey questions is not at all obvious. McIntosh(1999) has detailed a whole range of difficulties intrying to use these data to draw quantitativecomparisons about levels of vocational trainingacross the EU. Questionnaires used to collect thisinformation are often revised and modifiedbetween successive surveys, so that discontinu-ities in the data sets over time are commonplace.

Table 8 appears to show some marked discrep-ancies across countries between the percentagesof individuals receiving training in the month prior

Mean school years % with no education (age 15+)

1970 1980 1990 2000 1970 1980 1990 2000

Middle East/North Africa 2.07 3.29 4.38 5.44 69.8 55.5 42.8 32.0

South Asia 2.05 2.97 3.85 4.57 69.3 66.9 55.2 45.2

Sub-Saharan Africa 2.07 2.39 3.14 3.52 63.8 56.8 45.9 42.8

East Asia and Pacific 3.80 5.10 5.84 6.71 35.4 22.6 26.4 19.8

Latin American and Caribbean 3.82 4.43 5.32 6.06 31.2 23.8 17.2 14.6

Advanced countries 7.56 8.86 9.19 9.76 5.1 4.8 4.5 3.7

Transitional 8.47 8.90 9.97 9.68 3.1 2.8 1.7 2.2

World 5.16 5.92 6.43 6.66 31.4 29.5 26.4 24.2

Source: Barro and Lee (2001)


Table 7: Average schooling and no schooling rates 1970-2000

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 26

to the survey; while Sweden has just over 10 %,France and Portugal have less than 1 %.However, much of this variation is explainable interms of how the training question is interpreted indifferent countries, rather than real observable

differences. Much of Table 8 is concerned withhow the incidence of training is distributed acrossthe characteristics of individuals, their job tenuresand their places of work. In most Member States,training is concentrated on those in the younger

Figure 4: Schooling attainment 1970-2000: average years of schooling


0

2

4

6

8

10

Advanced East Asia Latin America Middle East/North Africa

Sub-SaharanAfrica

South Asia Transitional

years - 1970 years - 1980 years - 1990 years - 2000

Figure 5: Schooling attainment 1970-2000: percentage of population 15+ with no schooling


no education - 1970 no education - 1980 no education - 1990 no education - 2000

0

20

40

60

80

Advanced East Asia LatinAmerica

Middle East/North Africa

Sub-SaharanAfrica

South Asia Transitional


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 27

age groups, although Sweden is a significantexception. While in the Sweden and the UK voca-tional training is aimed more towards thoseemployees with higher prior education levels(ISCEDs), in Germany the opposite is the case,with the least qualified employees being thosemost likely to receive training. Workers who havebeen in the job for the shortest period of timereceive more training in France and the UK thanlonger-term employees, whereas in Sweden thepattern is reversed with training increasing withlength of job tenure. Table 8 also shows how inGermany, France and the UK training incidence ishighest in smaller enterprises that employ fewerworkers, while in the Netherlands training ishighest in the largest enterprises.

It is clear that these training statistics arecomplex and difficult to interpret, given the signif-icant differences that underlie the national voca-

tional training systems. There is good reason tosuppose that vocational training, like its educa-tional counterpart, has been on an increasingtrend in all Member States over recent decades.However, measurement problems mean that voca-tional training data are unlikely to be suitable forinclusion in models of the kind described later inthis study. In an earlier Cedefop report, Luttringer(1998) has detailed some of the difficulties indefining precisely investment in continuing voca-tional training and also in relating it to productivitygains, both at the micro and macro levels.

A proxy measure of skill levels in populations isprovided by the OECD’s international adultliteracy survey (IALS). However this survey onlycovers a limited number of EU countries over theperiod 1994 to 1998. It is very likely thatincreases in adult literacy are associated witheconomic growth.

Category Germany France Netherlands Portugal Sweden UK

All employees 4.9 0.5 5.3 0.1 10.5 7.3

Females 4.8 0.3 4.4 0.1 10.8 6.7

Males 5.0 0.6 5.8 0.1 10.2 7.8

Age 15-20 65.1 26.5 8.0 0.2 3.4 25.6

Age 21-30 3.5 0.5 6.7 0.1 9.0 6.6

Age 31-40 0.8 0.1 5.4 0.1 12.1 6.2

ISCED high 1.0 0.2 3.7 0.3 14.7 10.5

ISCED medium 1.9 0.2 5.8 0.2 10.1 6.0

ISCED low 24.6 1.1 5.1 0.1 6.0 6.5

Tenure +6 years 0.5 0.1 4.6 0.2 11.7 5.6

Tenure 1-5 years 9.8 0.1 8.5 0.1 10.0 6.7

Tenure <1 year 6.2 2.5 3.4 0.1 7.7 8.1

< 11 employees 5.5 1.4 3.5 0.1 - 9.3

11-19 employees 5.5 1.3 4.4 0.2 - 6.6

20-49 employees 4.4 0 3.9 0 - 6.9

50 > employees 3.4 0 5.9 0.4 - 7.0


Table 8: Comparisons of vocational training based on European labour force survey

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 28

4.1. General literature strands

There is extensive and detailed literature on ratesof return on education and training, based onhuman capital theory. Much of this literaturerelates to the rates of return for the individual andsociety in general of investment in human capital,in the form of education and training. Most of thismicroeconomic research draws on work byBecker (1964), Mincer (1974) and many others.Measuring the benefits of education and trainingis complex and can be considered at variouslevels and from differing perspectives:(a) at an individual level, examining the impact of

education or training on the chances of beingemployed or unemployed or, most commonly,earnings;

(b) from an organisational perspective, examiningthe impact of education and training onorganisational performance. The extent towhich employers engage in training their staffand provide employees with transferableskills in the labour market is critical (Stevens,1999);

(c) at a macro level, examining the role of educa-tion and training, typically using qualificationsas a measure of activity, and their impact onproductivity, output growth and employment.

The discussion in this section briefly focuseson certain aspects of the first of these, while thesecond and third form the subject matter ofChapter 5 and 6 respectively.

4.2. Rates of return oninvestment in human capital

A number of reviews on the rates of return ongeneral education and training have been under-taken; most recently both Harmon and Walker(2001) and Blundell et al. (2001). A review of themost significant literature on rates of return oncontinuous vocational training in companies hasbeen carried out by Barrett (2001), in an earlierCedefop report. Sianesi and Van Reenen (2000)

have carried out a review of the macroeconomicreturns on education, but this largely deals withresearch considered in subsequent sections ofthe present document.

General literature on the subject demonstratesmost directly the links between education andtraining and performance at an individual level.However, it is important to appreciate how someof the microeconomic theory and analysis carriesimplications for the macroeconomic level which,(subject to certain concerns about the extent towhich qualifications act as a signalling devicerather than actually enhancing productivity), canbe regarded as the sum of the individual parts.While much of the evidence relates to the bene-fits to the individual, it is clear that this has directimplications at a more macro level.

The main emphasis of most of this research ison the returns on higher education (e.g. a firstdegree), although there is fairly extensive litera-ture on the returns on training, particularly in theUS. The prime focus of most of this literature isalso upon the benefit to individuals, althoughsome explicit efforts have also been made toassess wider benefits to society as a whole thateffectively represents a macro effect, which is theprinciple concern of this study.

The key concept here is the notion of the rateof return on the investment in human capital thateducation and training represents. Rates of returnare typically estimated in two ways:(a) early studies used a discounted cash flow,

accounting framework;(b) more recently researchers have adopted an

earnings function approach.The effects on employment, productivity and

earnings can be expressed in terms of the rate ofreturn on education and training.

Typically, rates of return are computed forvarious qualifications or for the length of timespent in education or on training courses. Therate of return expresses the value of an additionalyear of education (or the value of a particularqualification) in terms of the associated increasein earnings or income. The issues addressed inthe wide variety of studies that have estimated

4. Rates of return on education and training

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 29

rates of return on education and training includevariations in rates of return according to type ofqualification, gender, age and ability; thescreening impact of additional education to filterindividuals into well-paid jobs; the effects ofover-education and the direct impact of educa-tion and training (including government trainingschemes) on the probability of employment.

The discounted cash flow accounting approachcompares income or earnings streams of individ-uals with and without the education or trainingprogramme (i.e. they compare graduate incomestreams with those whose highest qualification is Alevel (3)). The calculation generates an internal rateof return on the investment that can be compared tothe going rate of interest. It is possible to distinguishbetween two substrands: ex post rates of return(which generally compare actual incomes of indi-viduals of different ages, with and without theeducation or training) and ex ante rates of return(which generally compare an individual’s percep-tions of future income streams over future yearswith and without the education or training).

The earnings function approach, involves esti-mating an earnings function based on data on theearnings of large samples of individual workers,together with information about their personaland other characteristics, including their invest-ments in education and training. Estimates ofrates of return on years of schooling or on indi-vidual qualifications can be obtained from suchan analysis using a method pioneered by Mincer(1974). Although it has been refined in variousrespects, this basic approach has become theaccepted standard, allowing for problems ofendogeneity and spurious correlation to bedirectly addressed. Although the discounted cashflow approach still has some adherents and is stilluseful in situations where the largecross-Sectional data sets on individual earningsare not available, the earnings function model isnow by far the most popular.

In effect, both approaches attribute the differ-ence in income between individuals or groups ofindividuals to the education or trainingprogramme. This is only true if the individualsbeing compared (i.e. those with and without theeducation or training) are identical in every impor-tant respect. Finally, this approach makes no

allowance for selection biases (i.e. that the indi-viduals most likely to benefit select themselves/are selected onto the programme). Someresearchers have tried to address the issue ofselection as well as testing the biases resultingfrom ability. Their results suggest that the majorpart of the benefits that accrue to individuals as aresult of investment in education or trainingreflect enhanced productivity. Hujer et al. (2003),in an associated study, discuss some of themethods and limitations of evaluation and impactresearch.

Despite this, there have been continuingconcerns about over-education, qualificationinflation and graduate unemployment, see forexample, Büchel (2001). In the UK, two papers byBrynin, (2002a and b) focus on these concerns.They examine first, the impact of graduate densi-ties on wages, and second, the impact of overqualification on returns on first and second jobs.Graduate density is defined as the proportion ofgraduates in employment defined by both occu-pation and industry. Together with other variables,including average education, these measures arethen used as potential explanations of wageoutcomes.

The results suggest some evidence of grad-uate overcrowding but also gender-based jobsegregation that appears to operate in favour ofwomen. For men, there are diminishing returnsfrom expansion in graduate numbers overall.Brynin also examines the extent and impact ofover qualification for a job. Normally this isassessed by asking about the level of educationrequired for the job and comparing this to actualqualifications. This often indicates that a largeproportion of people have some excess educa-tion, although typically they earn more comparedto other people doing the same kind of work.Normally such individuals earn less than thosedoing the kind of job for which they consider theyare appropriately educated. The results indicatethat for younger cohorts, having excess educa-tion brings diminishing returns in their first job,but this does not apply to the older cohorts.These points were originally raised in theCedefop second research report, Part 4.

Brynin reports that, as average qualificationlevels have increased over time (as measured by


(3) The General Certificate of Education (GCE) Advanced (A) level is normally taken at the end of secondary school in the UK.A levels are the main standard for entrance to higher education.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 30

average O-level/GCSE (4) results), the status ofsecond as well as first jobs (as measured by the‘Hope-Goldthorpe scale’ of occupational status)has been declining. This does not necessarilyimply the need to put the brakes on the expan-sion of education but does demonstrate that noteveryone will benefit uniformly from furtherexpansion of education.

4.3. Private versus social rates of return

An important feature of this micro literature,carrying implications for macro studies, is the keydistinction between private and social rates ofreturn. In effect, the private rates relate only tothe costs and benefits experienced by the indi-vidual undertaking the education or training. Inparticular, the costs of education would onlyinvolve those elements paid for by the individual(i.e. they would not include fees if these werepaid by the government) and higher futureincomes would be assessed net of tax. The socialrate of return, on the other hand, takes intoaccount all costs and benefits. Private and socialreturns are generally always different for the typesof reasons set out above. An Austrian empiricalstudy by Fersterer and Winter-Ebmer (1999)demonstrates some of the differences in the ratesof return on various levels of education forAustria.

In addition, there is the issue of spill-overeffects and externalities from education andtraining, which may also have a distinct spatialaspect. An example relates to poaching. If, forexample, few firms train, then the trained workerscan be offered higher salaries by firms that do notundertake training. The firms which train lose inthis process and the firms that do not trainbenefit, although, increasingly, firms ask for arepayment of training costs if workers leave afirm. This undermines the incentive to train. If,however, all firms train (to the same standard),then, ignoring differences in types of skills, thereis no incentive to poach and, when individuals domove, those lost can be replaced by others of

equal skill level. There are other aspects of socialreturns, however, linked to new growth theories,which are developed in Chapter 9 below. Inparticular, the bridging concept is that educationspills-over from the educated to thenon-educated group. The original idea wasproposed many years ago by Blaug (1972).

Mingat and Tan (1996) produced estimates ofsocial returns on education, to incorporate exter-nalities and non-economic effects, for a widerange of countries over the period 1960-85. Theyused the overall economic performance of thevarious countries to capture these externalities.Their results confirm the social profitability ofinvesting in education, but they suggest that theresults are sensitive to the stage of economicdevelopment achieved by the individual country.The best returns for low-income countries camefrom investing in primary education, in secondaryeducation for middle-income countries, and forthe highest-income countries the optimal socialreturn was from higher education. Despite this,there is evidence from some countries, includingthe UK, that there are still significant benefits tobe obtained by investing in basic literacy andnumeracy among those segments of the popula-tion that have, for one reason or another, missedout on education, Dearden et al. (2002)

Harmon and Walker (2001), in their recentreview of evidence and issues relating to thereturns on an individual’s education, indicate aconsiderable degree of consensus on estimatesof the rate of return for an additional year’seducation. Basic specifications suggest a returnfor a year of schooling of between 7 and 9 %.Most studies suggest that women gain more fromadditional education than men, that those in thetop part of the income distribution gain higherreturns per year of education than those in thebottom part , and that the effect of underlyingability on earnings is small compared with theeffect of education. The impact of such educationon overall national output and productivity is notconsidered by such studies.

Much of the analytical debate regarding theneed for greater public investment in educationand training has rested on the social rate of returnrelative to the private one. Evidence on social rates


(4) The Ordinary (O) level is the main examination taken by secondary school pupils in the UK leading to the General Certificateof Secondary Education (GCSE). They are usually taken after four or five years of secondary schooling and can lead to moreadvanced education and training.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 31

of return looks at the benefits to the economy fromincreased years of education, typically by calcu-lating the costs of education/schooling comparedto pre-tax earnings. From a methodologicalperspective this is contentious due to the need tomake many assumptions, but available evidenceindicates returns are around 6-12 % (Chevalieret al., 2001) (Steel and Sausmann, 1997). A newstudy by Trostel et al. (2002), uses broadlycomparable data covering a number of countriesto show that recent rates of return on schoolingvary significantly across the 28 countries in thesample set but are still generally positive andlarge.

The present study is as much concerned withvocational training as with general education. Themost obvious and direct effect of training byemployers is to provide additional skills and toraise productivity. This is the central notion of thehuman capital model of training. Barrett (2001)reviewed a number of empirical studies acrossseveral countries, examining the links between

returns on continuing vocational training andcompany performance. The main results suggestthat strict rates of return, equivalent to thoseused in the higher education studies, have rarelybeen used in the evaluation of continuing voca-tional training. Such training is usually found tohave a positive effect on both productivity andwages, but there have been few attempts todemonstrate whether these are sufficient to guar-antee a good rate of return. Specific training hasthe greatest impact by raising the productivity ofthe employee within the firm providing thetraining. General training, which can be used byall companies, is still beneficial, but it tends tohave less impact on company performance,although it may have important benefits for theindividual and for society at large. There remainimportant differences in the perception of thevalue of training between the individual, thecompany investing in the training and society asa whole. These issues are more fully investigatedin the associated study by Hansson et al. (2003).


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 32

5.1. Links between humanresources and performanceat company level

Most of the studies on rates of return discussedin the previous section are concerned with issuesof appropriate measurement and analysis of thefactors that lead to variation in the measuredrates for the individual. Generally, there is noattempt to relate such rates of return specificallyto more general corporate performance. It can beassumed from human capital literature thatgreater education leads, ceteris paribus, to higherproductivity, which will be reflected in economicgrowth or other kinds of improved performancefor the individual company or organisation.

Outside mainstream economic literature, studiesin the management of human resources report arange of results concerning the impact of humancapital on firm and industry performance. Suchliterature bridges the gap between rate of returnanalysis and macro level studies. It suggests thatcompanies employing managers, professional andother staff with higher qualifications can expect toachieve better commercial performance. This liter-ature is briefly reviewed here.

There has long existed a widely held belief thatpeople are a key source of an organisation’scompetitive advantage (Prahalad, 1983) (Pfeffer,1994) (Wright et al., 1994). By implication, it may beargued that it is the quality of the management ofhuman resources that determines organisationalperformance (Adler, 1988) (Reich, 1991) (Youndtet al., 1996). There appear to be a number ofdifferent emphases in this literature, including therole played by managers, as well as the extent andnature of human resources themselves.

5.2. Management skills andorganisational performance

A key theme in the economics and broadermanagement literature has been the importantrole that managers (and administrators) play in

determining organisational performance.Amongst trading companies, for example, it hasgenerally been assumed that the capacity of thechief executive officer (CEO) or the managementteam forms the ultimate limit to the rate ofsustainable growth (Bosworth and Jacobs, 1989).

The way in which this line of thinking developedwas to look at the goals of the largeowner-managed firms and the manner in whichthese impacted upon company behaviour. Theempirical counterpart to this was structure-conduct-performance literature, where firm sizeand market power were key components. Thissection focuses on more recent literatureconcerned with the level and discipline of qualifi-cations of the workforce and, in particular, theCEO or management team. (Bosworth, 1999).

There is no real underpinning theoretical litera-ture, although there are clearly links with thetheories relating to individual rate of return,human capital and the value of intangible assets.Empirical modelling usually attempts to relate P,some measure of company performance to Q, avector of qualification and skill-related variables,and X, a vector of other variables influencingcompany performance.

A further issue is the use of qualifications as ameasure of the quality of workers or managers.Certainly education and training impart a varietyof knowledge and skills that may raise theproductivity of the individual in the job in whichthey are employed. However, these are input andnot output measures. There is no indication as tohow good or relevant the education or training isor, indeed, the capacity of the individual toabsorb the knowledge. In some ways this can beseen as an empirical issue. It is possible simply toestimate an equation where P is determined by Qand X, to see whether qualifications are signifi-cant or not, in the presence of other determinants.It should be noted that measurable qualificationsare not the same as actual skills or quality ofcompetence.

The role of leadership has been defined inseveral ways. The first is ‘stewardship’, which istypically represented as a dummy variable for a

5. Human resources and performance

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 33

specific CEO in a particular company (Liebersonand O’Connor, 1972) (Weiner and Mahoney,1981). A dummy variable is defined Dijt as takinga value of 1 when the ith CEO is headingcompany j at time t. An alternative way is toattempt to quantify CEO ‘strategy’ by observingkey variables over which the individual hascontrol, such as capital structure and retainedearnings (i.e. two financial strategy variables). Itcan be seen that the first variable is a crudeproxy, and the second echoes the Tobin’s Q typespecifications discussed in Chapter 7 below.

HRM literature is mixed in its views about theimpact of leadership. One strand argues that theevolution of formal and informal organisationalstructures in large corporations means that theylargely run themselves, so effectively limiting theinfluence of any single individual, even the CEO.Hall (1977) argues that leadership is important intimes of growth, development and crisis, but notin times when the organisation is roughly main-taining a status quo. Weiner and Mahoney (1981)point out that while the presence of a leaderremains essential, the particular leader that ischosen may be of little importance. Liebersonand O’Connor (1972) found that situational andorganisational factors were of greater importancethan leadership.

5.3. Empirical studies at company level

Issues of HRM at company level were discussedearlier in Cedefop’s second research report,Part 3 and they are also reviewed in the parallelstudy by Hansson et al. (2003). A key theme ofHRM literature is the link between HRM strategyemployed by the organisation and performance.Youndt et al. (1996), for example, distinguishbetween universal and contingency strategies (5).The authors report that their empirical studylargely supports the contingency approach. Moreimportantly, from the perspective of the presentstudy, the authors conclude that HRM systems,which focus on human capital enhancement, aredirectly related to multiple dimensions of organi-

sational performance, in particular to employeeproductivity, machine efficiency, and consumeralignment. However, in line with their conclusionsabout the contingency approach, Youndt et al.report that fuller analysis reveals that enhancedperformance is predominantly the result of linkinghuman-capital-enhancing HRM systems withquality manufacturing strategies. There is a linkbetween HRM strategy and the quality ofmanagement decisions.

An earlier survey of over 700 UK productioncompanies revealed the relatively low levels ofqualifications of both the boards of directors andtheir CEOs (Bosworth et al., 1992). However therewas strong confirmation of a link between theexistence of a R&D culture and the ability andwillingness to innovate, particularly when gradu-ates were employed to carry out R&D. Regres-sion analysis of the extent of the introduction ofnew technologies by companies in the sample (aperformance indicator) revealed a positive corre-lation between the probability of innovation andthe presence of graduates in the workforce. Firmsthat utilised these advanced technologies tendedto outperform non-users in a number of respects.In particular they worked closer to full capacityand had higher growth in turnover and marketshare. However, one consequence of workingcloser to full capacity and being more dynamicwas that such companies tended to experiencemore acute skill shortages. The study reportedboth direct and indirect (i.e. via innovation) linksbetween qualifications and performance. Two ofthe main conclusions drawn from the study, were‘[…] the presence of graduates […] improvesgeneral economic performance’ and ‘[…] thedifferential impact of graduates (generally) oncompany performance […] is much moreclear-cut than the differential performance thatcould be attributed to the employment of profes-sional scientists and engineers vis-à-vis gradu-ates from other discipline areas.’

Barry et al. (1997) update and extend the earlierresults of Bosworth et al. (1992) and compare themwith a study by Wood (1992), who linked aprofit-based measure of performance to qualifica-tions. Wood found that manufacturing companiesrun by CEOs possessing any kind of degree/profes-


(5) ‘The universal, or “best practices” perspective implies a direct relationship between particular approaches to human resourcesand performance, and the contingency perspective posits that an organisation’s strategic posture either augments or dimin-ishes the impact of HR practice on performance.’ (op cit. p. 837).

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 34

sional qualifications were more likely to achievesuperior levels of profitability. However, there wasno evidence that the precise type of degree/profes-sional qualification held by the CEO made a greatdeal of difference to the level of profitability.

The Barry et al. (1997) study focused moresimply on the top executives of the companies,rather than the earlier approach that had incorpo-rated both non-executive and executive directors.A key conclusion on the role of qualifications wasthat companies headed by qualified top execu-tives tended to outperform those with unqualifiedtop executives and many of the latter companiesexhibited significant under-performance in themarket-place. A further conclusion concerned thecontribution of the discipline background of thetop executive, whereby companies headed bytop executives with non-technical qualifications –particularly accounting – outperformed compa-nies with top executives who were qualified engi-neers or scientists.

A key question relates to what the measure ofcompany performance should be. Earlier concep-tual literature suggested that growth might be a keyperformance indicator, but there is little in empiricalliterature that has tested such a relationship. Thereare some models in structure-conduct-perfor-mance literature that have explored the influenceson growth, but these have not usually incorporatedthe human capital variables. There are, however,some studies that test alternative goals, such asprofits. Here, however, it is not possible to use, forexample, private rates of return theory as an under-pinning, as this is based upon the idea that individ-uals are paid the value of their marginal product (orsome equivalent dynamic concept). However, it isfairly easy to think of a rationale from the more insti-tutional/contracting literature that suggests, forexample, more qualified individuals are offeredcontracts that induce them to generate a surplus.This surplus is then shared between shareholdersand qualified workers in the form of both higherprofits and higher wages.

A study by Ballot et al. (2001) examined theimpact of human and technological capital onproductivity in a panel sample of large Frenchand Swedish firms. Specifically, measures of afirm’s human capital stock were constructed onthe basis of past and present training expendi-tures. The results confirm that such firm-spon-sored training, together with R&D, are key inputs

to the market performance of firms in both coun-tries. Another study by Blechinger and Pfeiffer(1998) used survey data for the German manufac-turing sector to explore the links betweenemployment growth, technological change andlabour force skill structures. Innovative firmsexperienced the highest growth rates and suchfirms tended to employ more highly skilledworkers. An HRM study by Papalexandris andNikandrou (2000) into Greek firms found that,where training was treated as a continuous, life-long learning process, it had considerable impacton the growth of firms.

5.4. Empirical studies at theindustry and sector level

A different approach to exploring the links betweeninvestment in human resources and productivity inorganisations is that developed over the last20 years by researchers at the national institute ofeconomic and social research (NIESR) in the UK.Much of this work is summarised in Prais (1995) andall the relevant papers have been published in full invarious editions of the National Institute EconomicReview. The research approach consists of carryingout empirical international comparisons of produc-tivity and associated schooling and vocationaltraining investments between Germany, France,the Netherlands, the UK and other selected coun-tries. Often matched samples of plants and firmsfrom specific industrial sectors have been used toexplore the productivity and human capital rela-tionship.

It should be noted that this body of researchdraws on statistics and information from severalEU countries. Moreover, this research is note-worthy in utilising vocational training data, as wellas the more commonly adopted years ofschooling measures. The concern is not so muchwith the statistical impact of an extra year oftraining or schooling on productivity, but ratherwith the results of different kinds and qualities ofhuman resource investment.

Prais (1995) summarises broad statistical esti-mates from various national labour force surveyscarried out in the period 1988 to 1991. This datashows marked differences in intermediate voca-tional training qualifications between selected EUcountries. While in the UK only 25 % of all


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 35

economically active persons achieved such qual-ifications, the comparable figure for France was40 % and for Germany it was 63 %. The maindifference was found in craft training rather thantechnician training. Conversely, some 64 % ofthe UK workforce were found to have no voca-tional qualifications of any sort, whereas thecomparable figure for France was 53 % and forGermany only 26 %. Prais concluded that the UKwas anomalous in the relatively low proportion ofits workforce that has received systematicallyorganised vocational preparation and attainedformally examined vocational qualifications.During the ensuing decade, investments in voca-tional training across the Member States havechanged these percentages but the differentialsbetween countries still persist and this has impli-cations for comparative productivity.

Various studies carried out by researchers atthe NIESR have sought to demonstrate the linkbetween a better educated and better trainedworkforce and realised higher output per worker.Investigations were made into manufacturingplants in the UK and their counterparts in otherEU countries. The studies looked into the metal-working, woodworking, clothing and food manu-facturing industries, as well as selected servicesector trades. It was found that across all theseindustries the acquisition of skills amongst theworkforce was a critical factor in raising produc-tivity, as measured by output per worker. Recent

studies of this kind have been carried out byMason et al. (1999) into banking services and byJarvis et al. (2002) into the ceramic tablewareindustries. Commonly, plants in Germany, Franceand the Netherlands, where the percentage ofvocational qualifications was much higher, returnsignificantly higher output per worker than theirUK counterparts.

Steedman (2001) has recently comparedsystems of apprenticeship training across variousEuropean countries (from a UK perspective) anddrawn implications for productivity. This studyidentified significant shortcomings in the UKapproach that have resulted in an inferior qualityof vocational training for young workers.

Another study by O’Mahony and de Boer(2002) confirms that the UK continues to lagbehind both Germany and France in terms oflabour productivity, and this gap is primarilyexplained by differential rates of investment inboth human and physical capital. This predomi-nantly statistical study compared labour produc-tivity not only across the aggregate economy butalso over some 10 broad industrial sectors. Itapplied education and training statistics, dividedinto higher, intermediate and lower level qualifica-tions, to quantify comparative workforce skills inthe different countries. It identified a significantassociation between labour productivity andmeasured workforce skills across the differentindustrial sectors of the comparator countries.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 36

6.1. General literature strands

Over the last three decades a large body of liter-ature has been produced examining the role ofhuman capital in determining the level and growthof GDP per capita. Much of the earlier work ismainly theoretical and deals with different growthmodel specifications and their associatedeconomic properties; for a summary, see Aghionand Howitt (1998). More recent work seeks totest empirically the different model specifications,most commonly using cross sectional data for alarge number of countries. Some researchershave attempted time series analyses for smallergroups of countries, such as those in the OECDarea, where the quality of educational data areboth more frequent and perceived to be morereliable. A few studies have combined crosssectional data with time series information toproduce panel data sets in which allowances canbe made for country-specific effects. Two recentpapers by Sianesi and Van Reenen (2000) andTemple (2000) provide useful overviews of thevarious empirical studies. The former paper looksat the links between formal education andeconomic growth across all countries, while thelatter is concerned with the impact on growth inOECD countries of both education and morewidely defined social capital. The present paperdraws on both these studies.

From a comparative international perspective,attempts to identify the extent to which educationand qualifications affect relative economic perfor-mance, including employment levels, have begunto reveal the importance of human capital.Macroeconomic models incorporate humancapital into growth specifications either throughextensions of the Solow neoclassical growthmodel (Solow, 1957) or through endogenousgrowth equations, as developed by Romer (1986),Lucas (1988) and others. A current application ofsuch a model is shown in the associated study byIzushi and Huggins (2003). Fundamentally,neoclassical models imply that a one-off increase

in the stock of human capital leads to an associ-ated one-off increase in productivity growth,whereas endogenous growth models suggestthat the same one-off increase in human capitalcan lead to a permanent increase in productivitygrowth. In the short term, both models producebroadly similar results, each dependent on theprecise specification, but, over the longer term,the newer growth models imply significantlyhigher returns on investments in human capital.

Regardless of the precise model that isadopted, there is strong evidence that highereducational inputs increase productivity and soproduce higher levels of national growth. Aftersurveying the empirical results from a wide rangeof model specifications, Sianesi and Van Reenen(2000) concluded that an overall 1 % increase inschool enrolment rates leads to an increase inGDP per capita growth of between 1 and 3 %. Anadditional year of secondary education whichincreases the stock of human capital, rather thanjust the flow into education, leads to more than a1 % increase in economic growth each year. Theresults vary dependent on the model specifica-tions and the data sets in use. Sylwester (2000)has suggested that current educational expendi-ture leads to future economic growth, so there isa significant time lag in the causal relationship.

Aside from concerns relating to the preferredmodel specification, other methodological issuesconcentrate on the most appropriate variables forinclusion and their methods of measurement. Avery important strand in the literature concernsthe definition of human capital and how farhuman capital can be practically measured. Thisissue was originally addressed in an earlierCedefop report (Westphalen, 2001), where humancapital was widely defined to include knowledge,skills, competences and similar attributes. For themost part, empirical growth models adopt amuch narrower interpretation of human capitaland usually some measure of flow or stock ofsecondary education is used to proxy humancapital. Typically, no attempt is made to incorpo-

6. The role of education and training in economic growth

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 37

rate vocational training data into growth models.Koch and Reuling (1998) pointed out some of thedifficulties of measuring training investments withany degree of consistency across different coun-tries. Recently, other authors have suggested thatit is social capital as well as human capital whichcontributes significantly to growth (Woolcock,2000), but given the measurement difficulties,social capital type variables have not, to date,been much used in growth models.

Another strand in the literature is concernedwith the selection of appropriate data formeasuring economic growth across countries andthrough time, on a consistent basis. A majority ofstudies make use of the Penn World Table seriesproduced by Summers and Heston (1991) andthis comprehensive data set has been updatedand extended by Barro and Lee (1996). Someauthors, such as Osberg and Sharpe (2000), haveargued that GDP per capita is an inadequate indi-cator of the overall economic well-being of anation and they maintain that the link betweenhuman or social capital and economic well-beingis much stronger than is often implied whensimple GDP measures are used in growth models.Such analyses lead towards the literature thatlinks investments in human capital, and educationin particular, to externalities in economic growth.Higher levels of education are typically associatedwith better environment, higher levels of publichealth and greater social cohesion, all of whichwould be expected to feed back into fastereconomic growth measured in the wider sense.This strand of literature has recently beensurveyed in OECD (1998).

Most of the literature produced on this topichas appeared in English language publications. Amajority of the studies are by UK and US authors;European authors have produced some of theOECD publications. However, the empiricalstudies use international data sets and EU andOECD countries are widely analysed within muchof the published research.

6.2. Growth accounting literature

Growth accounting literature is deemed to besufficiently well known to warrant only brief treat-ment here. Griliches (1997) has provided a usefulsurvey of literature on applied growth accounting

exercises. This has much in common with generalrates of return on education and training,although it focuses much more explicitly onmacroeconomic issues with the measured effectsof education and training in the form of higheroutput or sales rather than earnings. Growthaccounting literature has a long pedigree, whichcan be traced back to Solow (1957). In essence,this assumes a Cobb-Douglas production func-tion:

Yt = At Kta Lt

b Mtc (1)

Where Y is gross output (e.g. sales turnover), K isphysical capital, L is labour stock (employment)and M is materials and intermediate inputs. At

reflects efficiency, so that the higher At, for anygiven values of the inputs, the higher is output.The units of measurement of the inputs affect thelevel of At. This can be avoided by writing theequation in growth form (i.e. differentiating withrespect to time),

∆Yt = ∆At + a ∆Kt + b ∆Lt + c∆ Mt (2)

where the ∆ associated with each variablereflects the rate of growth. It can be seen, byrearranging equation (2), that ∆At is the rate ofchange in total factor productivity.

∆At = ∆Yt - a ∆Kt - b ∆Lt - c∆ Mt (3)

In other words, it represents the output growththat cannot be explained by changes in thegrowth of physical inputs. Hence At was chris-tened the ‘residual factor’. In the original Solowstudy this residual component accounted forsome 85 % of output growth. A follow-up studyby Jorgenson and Griliches (1967) attempted toexplain this residual factor.

Further extensions of the framework to allowexplicitly for quality change and control over inputand output prices are reported in Bosworth andGharneh (1996), using a pure accountingtautology in a dynamic context. This literature isshown to have evolved through many phases,which can be categorised as three different gener-ations. First generation models are of the formdescribed in equation (1), using highly aggregatemeasures of the inputs (i.e. all physical capital, allemployees and the total real expenditure onmaterials and intermediate inputs). Second gener-


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 38

ation models are basically the same functionalform, but utilise a much more disaggregated setof variables, many different types of capital and alarge variety of different types of hours worked.Third generation models not only disaggregate theinputs in detail, but also begin to disaggregate thesectors of the economy, separately distinguishingbetween the production sectors of the economyand the educational sector.

Jorgenson and Fraumeni (1992) use growthaccounting methodology in a third generationframework to demonstrate that investment inhuman and physical capital accounts for a veryhigh proportion of growth in both the educationand production sectors of the US economy overthe post-war period. Specifically, growth in labourinput is held to account for just over 60 % ofoverall economic growth and increases in labourquality (education and training) explain some42 % of this labour contribution. As usual in thegrowth accounting approach, the contributions ofthe various inputs to output growth are obtainedby weighting the input growth rates by theirshares in the sector’s value added. The contribu-tions of capital and labour inputs are thendecomposed into separate components ofcapital stock and quality, and of hours workedand labour quality.

Hall and Jones (1999) adopt a growthaccounting framework to explore differences inoutput per worker across different countries. Theyuse the Summers and Heston (1991) database,augmented by educational attainment statistics asproduced by Barro and Lee (1996). Their findingssuggest that differences in physical capital andeducational attainment explain only a smallamount of the differences in output per worker.International output differences are predominantlyaccounted for by differences in productivity anddifferences in growth rates derive almost entirelyfrom differences in the growth of A in equation (2).This result is at odds with the results of Jorgensonand Fraumeni (1992).

It is important not to read too much into theempirical estimates based upon the growthaccounting type of model because this approachentails imposing many assumptions rather thantesting them empirically. The methodology typi-

cally involves constructing the estimates using anaccounting tautology. The rate of change of eachinput is weighted by the share of that input withinthe total. The validity of this depends crucially onthe assumption that each factor is paid the valueof its marginal physical product. The growthaccounting approach only reveals the sources ofeconomic growth, if, amongst other things, thereare competitive factor markets. The apportion-ment of output growth to measured and residualinputs provides no real insight into the processthat determines such contributions. While humancapital has a significant role to play in explainingoutput growth in this type of model, the measure-ment of this human capital is usually a complexmix of educational, demographic and labourmarket variables.

6.3. Neoclassical growth models

Although growth accounting exercises are usefulfor exploring the links between human capital andeconomic growth, they are limited by their basicassumptions. As Griliches (1997) recognised inhis survey, the true test of the importance ofeducation to growth is to include such a variablein an estimated production function. The startingpoint for such an approach is the Solow (1957) orneoclassical model. This growth model has previ-ously been described in detail in an earlierCedefop report (Barrett, 2001), so only an outlineis provided here.

The production function can be written in theform where some measure of output is related tothe relevant inputs (as in equation (1) above) (6).Using the Cobb-Douglas form:

Yt = A egt Kta Lt

b Mtc (4)

where Y is gross output or sales, K is the physicalcapital stock, L denotes labour and M is thematerials and intermediate input variable (all vari-ables are in real or volume measures) (7). This issupposedly a technical relationship betweeninputs and outputs. The link with the growthaccounting approach is obvious, as (4) is almost


(6) Gross output is preferred. This is related to capital, labour, materials and fuels (often called the KLEM models). Net output (orvalue added) is sometimes used, which is related only to capital and labour.

(7) The A term is now a constant.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 39

identical to (1). There are two main differences.First there is an extra term, egt, which attempts topick up the effects of technological change (i.e.an exponential time trend in times series estima-tion). Second, the function is now estimateddirectly using regression techniques, rather thanimposing accounting values. The parameters a, band c are still interpreted as factor shares.

This basic model can be augmented, in approx-imation to second generation growth accountingmodels, by the inclusion of different types (orvintages) of capital, K1t

a1 … Kntan, and similarly,

different types of labour (i.e. different lengths ofschooling, qualifications, occupations, etc.).Assuming that the homogeneous nature of the vari-ables, as in equation (4) is maintained, but repre-senting their changing quality by some measure orvector of measures, H, a model is derived:

Yt = A Kta Lt

b Mtc Ht

d (5)

where H is an indicator of the changing quality ofinputs and represents some weighted average ofeducation or qualifications (8).

It is now fairly easy to develop a link to theliterature of knowledge production functions. Thebasic idea of this approach is to regress somemeasure of output, preferably gross output, ontangible and intangible inputs. Thus:

Yt = A Kta Lt

b Mtc Rt

d … Rt-ne (6)

where the R variables denote current and pastinvestments in knowledge and are virtually iden-tical to H in equation (5). One way of looking atthis relationship is that current and past knowl-edge generates current output (or sales), holdingtangible inputs constant. An alternative interpre-tation is to note that {Yt/AKt

aLtbMt

c} is a measureof total factor productivity, and that current andpast knowledge is driving total factor productivity.

Although their endogenous counterparts have,to some extent, superseded these neoclassicalmodels, they continue to be empirically tested andSianesi and Van Reenen (2000) were led toconclude that the neoclassical approach producesresults that are more consistent with the estab-lished microeconomic evidence. One of the betterknown and most influential contributions to growth

literature is the study by Mankiw et al. (1992), whouse an augmented Solow model to explain crosscountry differences in income levels from 1960 to1985. This model of equation type (5), but withoutthe materials input variable, explains over 70 % ofthe variation in income per capita across a largesample of countries. Human capital, as proxied bysecondary school enrolment ratios, accounts foralmost half the difference in per capita incomes. Fornon-oil countries, a 1 % increase in the averagepercentage of the working-age population insecondary school is estimated to lead to a 0.66 %increase in long-term income per capita.

Koman and Marin (1999) applied an augmentedSolow model, in a time series framework, toexplain recent economic growth trends inGermany and Austria. The results obtained didnot support this type of specification and theincorporation of a variable measuring the accu-mulation of broadly-defined human capital led toinsignificant estimates. Temple (1998) has shownthat the estimated parameters and convergencerates in this type of model are highly sensitive tomeasurement errors, and the obtained results arenot always statistically robust.

Neoclassical models remain restricted by theirunderlying assumptions of perfect competition andconstant returns to scale. A critical property of thebasic production function, equation (4), is that thereare diminishing returns on the accumulation ofcapital. In the absence of any technological change,diminishing returns will eventually choke off anyeconomic growth. The inclusion of human capitalaccumulation in the model, equation (5), increasesthe impact of physical investment on the steadystate level of output and it can then account for avery slow rate of convergence in income levelsacross countries. However the problem of exoge-nous technological change remains and Romer(1994), amongst others, has stressed the need tomake technological advances explainable withinthe model framework.

6.4. Endogenous growth models

In contrast to neoclassical models, endogenousgrowth models explicitly incorporate technology


(8) Some models use stock or levels of education and training, while others use flow or accumulation of the same variable. Thepreferred specification is usually empirically determined.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 40

(the A term in equations (4) through (6)) andattempt to recognise that technological changedepends on economic decisions in the same wayas capital accumulation. In particular, technolog-ical change is most commonly related to the stockof human capital, which is explicitly modelled interms of educational investments in theseendogenous specifications (9). The inclusion oftechnological change and knowledge dissemina-tion into the neoclassical framework is rendereddifficult because of the underlying competitiveassumptions, which do not allow for the possi-bility of increasing returns to scale. In the endoge-nous models, economic growth can continueindefinitely because the returns on investment in abroad class of both physical and human capitalgoods do not necessarily diminish through time.Spill-overs of knowledge across producers andexternal benefits from improvements in humancapital are part of this process because theyoffset tendencies to diminishing returns. Growthframeworks have also incorporated R&Dconcepts, as well as imperfect competition(Romer, 1986) (Barro and Sala-I-Martin, 1995).

A large number of endogenous growth specifi-cations have been put forward. A typical specification for analysing growth across severalcountries follows Barro (1997):

∆y = f ( y, y* ) (7a)

y* = f ( Z ) (7b)

where ∆y is the growth rate of per capita output,y is the current level of per capita output and y* isthe long-term or steady state level of per capitaoutput. For a given value of y, the growth raterises with y*, which is determined by a wide set ofeconomic, policy and environmental variables.These variables differ between studies, but typically Z in equation (7b) contains variablesmeasuring population (fertility and lifeexpectancy), labour supply, government expendi-ture and investment, terms of trade, inflation and,most significant for present purposes, educationalvariables. Measurement issues associated withthe educational variables are discussed below.

Barro (2000) maintains that in this model any

increase in the steady-state level y* will raise theper capita growth rate, y, over a transitioninterval. So if, for example, government improvesthe business climate by increasing its expenditureor if they decide to increase their investment ineducation by increasing enrolment rates insecondary education, this will increase the targetlevel y* and raise ∆y. As actual per capita outputincreases, diminishing returns will eventuallyrestore the growth rate to a level determined bythe long-term rate of technological progress. Inthe very long term, the impact of improved policyis on the level of per capita output rather than justits growth rate. However, transitions to the longterm tend to be lengthy and the growth effectsfrom shifts in government policy tend to persistfor a significantly long period.

6.5. Empirical growth regressions

Following the early work of Barro and his collabo-rators (Barro, 1991) (Barro and Sala-I-Martin,1995) (Barro and Lee, 1996), a large number ofgrowth regressions containing human capitalvariables in the set of regressors have appeared.Temple (2000) has distinguished two main groupsof model specification, those that link outputgrowth to some initial level or stock of educa-tional attainment, such as secondary schoolenrolment rates, and those that relate growth tothe flow of educational attainment rather than itslevel. The first group assume that the stock ofhuman capital is the engine of economic growthwhereas the second group attribute such growthto the accumulation of education and training in agiven period. This distinction is more fully devel-oped in the associated paper by Izushi andHuggins (2003), where empirical tests are carriedout on both model forms.

Most empirical models combine data from alarge group of countries, both developed anddeveloping, often using dummy variables todistinguish geographical or economic groupingsof countries. The aim of these studies is to iden-tify statistically significant and robust relation-ships between the various factors (the Z variablesin equation (7a) above) and economic growth


(9) Applied examples of such models are provided in Sections 6.5 and 6.6 below. The methods of incorporating technologicalchange and human resource investments into the models are discussed in these parts of the paper.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 41

across the full sample of countries. The estimatesseek to determine causal relationships and soestablish the sources of economic growth.

Barro’s original 1991 study used data for98 countries from 1960 to 1985 and related thereal growth rate of GDP per capita to initial humancapital, as proxied by school enrolment rates for1960, and a large set of other potential deter-mining variables. It found that output growth wassignificantly positively determined by both primaryand secondary school enrolment, in the presenceof other determinants. A one percentage pointincrease in primary school enrolment was associ-ated with a 2.5 % increase in GDP growth and asimilar increase in secondary school enrolmentproduced 3 % growth. When attempts were madeto control for measurement error, the resultsbecame weaker in magnitude but still significant.When 1950 school enrolment rates were added tothe model they become insignificant, as became avariable measuring the student-teacher ratio. Thisstudy also revealed how human capital variableswere significantly correlated with lower levels ofnet fertility and larger levels of physical capitalinvestment.

There followed several papers in the Barrotradition; see Sianesi and Van Reenen (2000) for amore complete overview. Different data sets andmeasured variables were applied in an attempt toreplicate and establish the robustness of theBarro results. Englander and Gurney (1994)re-estimated some of the early growth regres-sions using only OECD data, which are regardedas a more homogeneous and superior qualitydata set. Secondary school enrolment rates wereconfirmed as one of only three significant deter-minants of labour productivity growth, alongsidegrowth in the labour force and increases in thecapital to labour ratio. However, this study foundthat these significant regressors provided muchless explanatory power when estimated for thenarrower group of OECD countries than for theoriginal wide sample. Englander and Gurneyacknowledged difficulties with the educationproxy variables, measurement reliability issuesand problems in ensuring strict internationalcomparability: all these issues cast doubt on therobustness of the results.

Gemmell (1996) also worked on OECD datasets and his original contribution involved theconstruction of some alternative measures of

human capital based on both stocks and annualaverage growth rates at primary, secondary andtertiary education levels. He found that a 1 %increase in initial tertiary human stock was asso-ciated with a 1.1 % increase in per capita GDPgrowth, while a 1 % increase in subsequentgrowth in tertiary education (flow) was associatedwith almost 6 % output growth. While the directgrowth effects come through tertiary education,secondary education was found to have an indi-rect impact through its positive significant associ-ation with physical investment. When theseOECD results were compared to a wider sampleof countries, it was found that primary humancapital had the most impact in the poorest groupof the less developed countries and secondaryhuman capital was the most significant variablefor the intermediate group of less developedcountries.

Benhabib and Spiegel (1994) use a new set ofcountry comparative data on human capital stockinitially to test the augmented neoclassical model,given as equation (5) above and later to modifythe specification to allow human capital to enterdirectly into productivity in an endogenousgrowth specification. The neoclassical modelyields insignificant and generally negative coeffi-cients on the human capital stock variables, aresult which holds when other regressors areadded into the model and alternative proxies forhuman capital are applied. In contrast, theendogenous specification produces humancapital levels that are positive, but not alwayssignificant, determinants of per capita incomegrowth. These results suggest a distinct role forhuman capital in enabling foreign technology indeveloping countries and the creation of newdomestic technologies in more highly developedcountries, rather than entering the model on itsown as a conventional factor of production.

Barro and Lee (1993) add comprehensive neweducational data sets to the Summers and Hestondata and test endogenous specifications of equa-tion type (7) using a much wider set of potentialregressors. In particular, the average years ofsecondary schooling of the adult population at thebeginning of the data period are introduced as akey explanatory variable. Barro and Lee’s resultssuggest that an extra year of male secondaryschooling is associated with a 1.4 % increase inGDP growth per worker. In comparison, an addi-


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 42

tional year of female schooling seemingly has anegative impact on a country’s growth rate.Female education is significant in reducingnational fertility and hence population growth.Barro (1997), using modified data in panel formatand applying more sophisticated estimating tech-niques, produces a similar set of findings to theearlier paper. An extra year of male upper-levelschooling is associated with a 1.2 % increase inper capita GDP growth rate. Male primaryschooling is found to have no significant impacton growth and, again, female schooling at variouslevels have negative but insignificant coefficientsin the various equations that are reported.

Krueger and Lindahl (1998) criticise the find-ings of both Benhabib and Spiegel (1994) andBarro (1997). In particular, they focus on theresults which identify only the initial level or stockof educational attainment as the determinant ofGDP growth and the failure to find growth in suchattainment (the flow of educational investment) asa key determinant. Krueger and Lindahl show thespurious results of the earlier two papers to beattributable to the extremely high and unaccount-able measurement error in first-differencedcountry comparative education data. In a simpli-fied version of equation (7), with only levels ofschooling and changes in schooling variables onthe right hand side (10), the increase in educationas the key determining variable of GDP growth isshown. The initial level of education is not posi-tively related to future economic growth for theaverage country when assumptions on linearityand homogeneity of parameters for education arerelaxed. Unfortunately, the estimates of theimpact of changes in educational attainment onincome growth turn out to be implausibly largewhen set against findings from microeconomicstudies into private returns and the authors iden-tify endogeneity bias (richer countries tend toexpand their educational infrastructure) as thelikely cause of the problem.

A study by Judson (1998) attempts not only tosubstantiate the role of increasing investment ineducation in promoting growth, but also toexamine the importance of the allocation of thisinvestment in the growth context. In addition to

the familiar Summers and Heston data, and theBarro and Lee human capital stock statistics,Judson uses Unesco data on educational enrol-ments and spending to estimate the efficiency ofexisting educational allocations within countries.Overall, a 1 % increase in human capital growthis found to be associated with an 11 % increasein GDP growth rate (11). Judson applies a countrycomparative growth decomposition regression toshow that the correlation of human capital accu-mulation is not significant in countries with poorallocations but it is strongly significant and posi-tive in countries with better allocations (predomi-nantly richer countries). This finding that thecontribution to growth of human capital dependson the efficiency with which it is accumulated hasimportant policy implications in terms of theexact allocation of educational and trainingresources.

Graff (1995) examined the role of humancapital in explaining economic growth in some114 countries from 1965 to 1985. Generally, theresults showed the accumulation of humancapital, physical capital and technologicalprogress all to be significant determinants of thegrowth process. In a related study, Graff (1996)investigated the importance of higher levels ofeducation for a subset of poorer countries. Thisvariable was found to be important for growth, solong as the investment in higher education didnot lead to imbalances elsewhere in educationprovision, especially to the detriment of elemen-tary levels of education.

While most empirical growth studies usecountry comparative data, either averaged acrossa sample of years or taken over several years inpanel data format, a few studies attempt timeseries analysis within an individual country. Themajor problem with the time series approach isobtaining adequately long series on consistentbases; this is a particular problem for educationand training variables. One example of the timeseries approach is provided by Jenkins (1995) forthe UK. This paper uses annual data from 1971 to1992 and it proxies the stock of human capital bythree series measuring workforce qualifications.These series are used as key determinants of


(10) Some of these variables are lagged and both time and geographic-regional dummies are also included.(11) This implausibly large elasticity impact is a common finding in studies of this kind and it has led Sianesi and Van Reenen (2000)

to conclude that the effect is seriously overstated due to a number of methodological problems.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 43

aggregate output, alongside physical capital,total workforce, capacity utilisation and a timetrend. The overall result confirms the finding thatinvestment in human capital increases produc-tivity. Highly-qualified workers are found tocontribute almost twice as much to productiveefficiency as those with no qualifications at all.The relatively small number of observations (12)mean that the unrestricted estimates are impre-cisely determined and such results cannot beregarded as robust.

Another example of time series modelling isAsteriou and Agiomirgianakis (2001), who usecointegrated regressions to explore the long-termrelationship between formal education and GDPin the Greek economy. This study finds a signifi-cant relationship between primary, secondary andhigher education enrolments and GDP per capita.The main direction of causality runs through theeducation variables to economic growth, but inthe case of higher education, there exists reversecausality. This problem is more fully discussed inSection 6.10 below.

6.6. A recent example of growth modelling

One of the more recent papers in the Barro tradi-tion reports empirical results for the countrycomparative data set updated to 1995 (Barro,2000). The paper shows the results of using paneldata for more than 80 countries over 3 decades;the sample is disaggregated into three 10-yeartime periods and the countries are groupedaccording to various development criteria. Thegrowth rate of real per capita GDP is regressedagainst a large group of potential determinants,as outlined in equation (7) above. Several of thesedetermining variables are only available, at best,at five-year intervals and some less frequently, soeither average values are used for each decadeor, in the case of the key education variables,their level at the beginning of each period isapplied. The main education variable is onefavoured by Barro from earlier empirical exer-cises. It measures the average years of schoolattainment at the upper (secondary and tertiary)levels for males aged 25 and over. Subsequent

analysis introduces several alternative educa-tional measures into the model: primary schoolattainment, attainment by females and results oninternationally comparable examinations. Itshould be noted that due to lack of data avail-ability at the country comparative level, nomeasures of growth in education (flows) areincorporated, only stock levels at the beginningof each period.

The model is estimated by three-stage leastsquares, using instrumental variables. Somebasic results are reproduced as Table 9. In theoverall sample, the education variable turns outsignificantly positive, in the presence of manyother policy variables. The estimated coefficientimplies that an additional year of schooling raisesthe growth rate on impact by 0.44 % per year.Barro maintains that a possible interpretation ofthis effect is that a work force educated atsecondary and tertiary levels facilitates theabsorption of technologies from more advancedcountries. In this model, the effect from an addi-tional year of schooling impacts on the growthrate of GDP and only feeds back onto the level ofGDP slowly over time (higher levels of GDP havea negative impact on the growth rate). Usingfurther assumptions on national convergencerates and the average costs of providing an addi-tional year of schooling, Barro suggests that thecoefficient value of 0.0044 implies a real socialrate of return on schooling of the order of 7 % peryear; a figure within the range of typical microe-conomic estimates of returns on education.

Column 2 in Table 9 shows how results change,using a slightly more restricted version of themodel, when the sample is based only on OECDcountries. The result for this group of countries ismuch less satisfactory in terms of the signifi-cance of the regressors, very few of which aresignificant. In particular, the key education vari-able takes a value of zero. When other wealthier,non-OECD, countries are added to the OECD setto form the rich country sample (column 4), thereis some improvement in the results and the addi-tional year of schooling impact is now significant,but its effect on economic growth is only abouthalf as strong as was the case for the full sample.Results for the poor country sample are better interms of significant regressors, although the


(12) This is the main restraint on time series analysis in growth studies.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 44

overall coefficients of determination are relativelymuch lower for this group (below 50 %). Theimpact of the additional year of schooling ongrowth turns out to be almost four times greaterfor the poor countries compared to the richgroup.

Barro (2000) considers additional dimensionsto the years of schooling, beyond those reportedin Table 9. Female attainment in secondary andhigher levels of education become insignificantwhen added to the basic model for the overallsample. The same insignificant results apply tovariables measuring average years of primaryschooling for males and females separately,although it is clear that primary schooling is crit-ical as a prerequisite for secondary education.Problems of colinearity are present betweensome education variables and other potentialgrowth determinants; most obviously betweenfemale education and fertility rates.

Barro also attempts within the model frame-work to introduce measures of quality of educa-tion as represented by international examinationscores, where available. Such indicators of thequality of schooling capital have been claimed byHanushek and Kimko (2000) to be more impor-

tant for subsequent economic growth than yearsof educational attainment. Predominantly, dataare only available for richer countries and only forrecent years. Results suggest that science scores(as a measure of quality of education) are signifi-cantly positive for growth, mathematics scoresare also positive, but of less significant impactthan science, while reading scores apparentlyhave no significant impact. Given the dubiousquality of some of these data, it would be wrongto read very much into the results.

It is clear that modelling exercises in the Barrotradition do not produce results that can beregarded as robust. Modifications in specifica-tions and changes in data sets can lead to largedifferences in estimated coefficient values, asexemplified in Table 9. It might have beenexpected that the superior quality data availablefor the more homogeneous group of countrieswithin the OECD would have yielded betterresults than those for the full sample, but this wasnot the case. Such findings have prompted someresearchers to carry out tests of robustness onthe model results (Temple, 1998). Others haveintroduced more complex model specifications toaccommodate full sets of panel data (Islam,


Independent variable Overall OECD Rich-country Poor-country sample sample sample sample

Log GDP 0.107* -0.034* -0.0343* -0.0190*

Log (GDP)squared -0.0084*

Years of upper school for males over 25 0.0044* 0.000 0.0023* 0.0084*

Govt consumption/GDP -0.157* 0.015 -0.014 -0.167*

Rule of law index 0.0138* 0.0115 0.0116* 0.0196*

Exports + imports/GDP (Openness ratio) 0.0133* 0.0148* 0.0112* 0.0361*

(Openness ratio)* Log GDP -0.0142*

Inflation rate -0.0137* -0.0228 -0.0051 0.0033

Log fertility rate -0.0275* -0.0209* -0.0174* -0.0212*

Investment/GDP 0.033 0.045* 0.029 0.053

Growth rate terms of trade 0.110 -0.010 -0.008 0.0134*

Observations & R squared

1965-75 81, 0.62 23, 0.85 32, 0.77 49, 0.48

1975-85 84, 0.50 23, 0.65 32, 0.62 52, 0.39

1985-95 81, 0.47 23, 0.50 31, 0.52 50, 0.44

NB: * indicates coefficient is statistically significant at 5 % levelSource: Adapted from Barro (2000, Table 1, p. 19)

Table 9: Barro growth regressions

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 45

1995), while a further group have explored thedifferences that data quality makes to the results(De La Fuente and Doménech, 2000). Some ofthe results obtained from growth modelling implysome improbably large impacts of initial humanstock (education) on growth and these results aresometimes at odds with findings from microeco-nomic studies (Topel, 1999). One of the issueswhich has not been resolved is the extent towhich some of the results based on the broadergroups of countries reflect greater variability inthe data (which helps to establish the parametersof the relationships being tested) or whether it isa consequence of including extreme outlierobservations, which bias the outcomes andcause spurious results based on poor qualitydata.

6.7. The influence of data quality on results

The issue of data quality is a common theme inmany of the studies reported above; both Sianesiand Reenen (2000) and Temple (2000) raise prob-lems with data quality and suggest that empiricalrelationships between human capital and growthmay be compromised by measurement errors.De La Fuente and Doménech (2000) suggest thatsome of the insignificant relationships areattributable to some of the educational data thatare commonly used in growth regressions.Results that rely on time series variation in thedata are especially prone to error. When dataadjustments are applied in a preferred growthspecification, a much more significant relation-ship between educational attainment andeconomic growth is identified.

De La Fuente and Doménech review the variousdata sets on educational attainment that have beenconstructed on the basis of Unesco enrolmentstatistics; the best known of these is the Barro andLee (1996) series. Enrolment data are transformedinto attainment stock figures through a perpetualinventory method and interpolations betweencensus observations. The various authors applydifferent assumptions and estimating proceduresto produce series that exhibit inconsistencies, oneto another, across countries. Beyond this problem,there remain accuracy and consistency concernsabout the base Unesco enrolment data. Some

schooling levels reported for some countries do notappear to be very plausible, while others displayextremely large changes in attainment levels overshort time periods or extremely unlikely trends; seethe discussion in Behrman and Rosenzweig (1994).

While De La Fuente and Doménech (2000)conclude that the Barro and Lee series are prob-ably the most accurate source of data on humancapital stocks, these data still contain a largeamount of noise that can be traced largely toinconsistencies in the underlying primary statis-tics. De La Fuente and Doménech revise theeducational attainment data for 21 OECD coun-tries during the period 1960-90, using detailedstatistics from national sources within eachcountry. In general, the resulting series are held tobe much smoother through time and, as a result,much more plausible. The new data are thencompared to the original in various growth speci-fications (variants of equation (5)), using averageyears of schooling of the adult population as themeasure of human capital stock. The size andsignificance of the coefficient on this newlyconstructed human capital variable shows anappreciable improvement over the original Barroand Lee variable. When the estimates arerepeated with data in first differences, the gain iseven more noticeable and only the revised dataproduce a significant coefficient. Further testsusing other growth specifications confirm thisresult and the equations with the revised datameet the appropriate tests for robustness.

Bassanini and Scarpetta (2001) apply theDe La Fuente and Doménech human capital seriesin panel regressions over the period 1971-98. Thisstudy follows Islam (1995) in making full use of boththe time series and cross-sectional dimensions ofthe data. Country comparative studies that fail touse the time dimension produce results which arean average for heterogeneous countries and there-fore hard to interpret in the context of a singlecountry policy (Lee et al., 1997). The estimationmethod used by Bassanini and Scarpetta employspooled mean group techniques that allowshort-term coefficients, speed of adjustment anderror variances to differ across countries, butimposes homogeneity on long-term coefficients.Various specifications are estimated, but the empir-ical results favour an endogenous growth modelwith broad constant returns on human and physicalcapital. The results indicate a significant and posi-


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 46

tive impact of human capital accumulation on percapita output growth. One additional year of educa-tion produces 6 % growth in output in the long term,a result held to be consistent with microeconomicevidence. The findings survived a number of checksfor robustness using modified specifications andcountry comparative samples. The empiricalresults were sensitive to the use of more reliabledata and the choice of estimating approach.

6.8. Human capital externalities

It has long been recognised that the benefits ofinvesting in human capital are not restricted to thedirect recipient but spill over to others in society.A better-educated and trained workforce may wellimprove the productivity of their less educatedwork colleagues. Acemoglu (1996) argues that anincrease in the average level of human capital willoften cause firms to anticipate growth and makegreater investments in physical capital andresearch and development. Given an imperfectmatching process, firms that invest more will notnecessarily be associated with the workers whohave invested more in education. However, otherworkers will benefit, as firms using more physicalcapital than before employ them. Increases in theaverage level of human capital clearly createexternal benefits, but growth models may not becapable of capturing all these external effects.

McMahon (2000) has argued that the directeffect of education on economic growth is sepa-rable from the indirect effect or externalities. Ofthese externalities, probably about 75 % arenon-market outcomes which feed back intoeconomic growth, but are not readily measurablein the same way as GNP. The main non-marketexternalities are: health, including longevity, infantmortality and fertility; environmental impact,including various forms of pollution and defor-estation; crime, including rule of law, crimesagainst the person as well as property crime;better income distribution and the issue ofpoverty; and democratisation, including humanrights and political stability. These issues aredealt with more fully in Chapter 9 of this paperand in the fuller treatment in the associated paperby Green et al. (2003). It will be recalled howBarro regressions, such as those in Table 9,attempt to capture some of these impacts on

growth, but measurement limitations preventmodels capturing most of these influences. Thereis also likely to be much interaction between thedifferent non-market social outcomes.

It is clear that human capital externalities aremuch more than just a spill-over effect fromeducation in the economy. They are a wholeseries of net outcomes, most of which are onlypartially realised after initial impact and many takefull effect over a very long time span. These exter-nalities, along with investment in human capital,are important in offsetting diminishing returns onphysical capital and so determining positive futureper capita growth rates. Investment in educationand training is important, not only for its directcontribution to growth but also for the indirectexternalities that it creates, as these eventuallyfeed back into the growth process.

6.9. Social capital and economic growth

Most of the literature reviewed in this section isconcerned with the link between human capitaland measured per capita output or productivitygrowth. As such, human capital is narrowlydefined to be some measure of education,usually average years of schooling at primary,secondary or tertiary levels. Temple (2000) recog-nises that the failure to incorporate measures ofvocational training is a serious omission from thedefinition of human capital. Broadberry andWagner (1996) have shown that vocationaltraining is closely connected with corporateproduction strategies and also national outputgrowth. However, the extreme variability in voca-tional training approaches across countriesmakes it difficult to measure and quantify to facil-itate inclusion in country comparative growthmodelling. McIntosh (1999) demonstrates someof the problems involved in making comparisonsof vocational training between six EU countries. Awide spectrum of training exists in most countriesand often a large amount of training is informal,on-the-job training that may not be recorded andso cannot be systematically measured. It isunlikely in the near future that country compara-tive growth models, which constitute the core ofthis growth literature, will be able to incorporatemeasured training variables.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 47

Despite measurement difficulties, a number ofresearch papers have recently appeared linkingeconomic growth to social capital (e.g. Temple,2000; Osberg and Sharpe, 2000; Woolcock, 2000and the literature reviews in these papers). Socialcapital is a difficult concept to define precisely,but, most simply, it refers to the social norms andnetworks that facilitate collective action. Itembraces the nature of social ties within commu-nities, the relationship between civil society andthe state and the quality of governing institutions.The development of social capital is likely toprove highly significant for improvements ineconomic well-being; a wider concept thansimply growth in per capita GDP.

While literature on social capital lies mostlyoutside mainstream economics, there has, in recentyears, been an increasing body of material on socialcapital published by organisations such as OECD.The majority of this literature is still to be foundpredominantly in social science and political publi-cations. Two studies have attempted to constructquantitative indicators of social capital and relatethem to other economic variables, includingproductivity. La Porta et al. (1997), Knack andKeefer (1997) and, most recently, Green et al. (2003)used an index of trust, derived from a World valuessurvey, carried out across some 28 marketeconomies. La Porta et al. found the trust index tobe weakly associated with country comparativegrowth from 1970 to 1993, but the explanatorypower was low and measurement concerns serveto limit the usefulness of these findings. Knack andKeefer report a strong correlation between trust andaverage years of schooling. Education is argued asstrengthening trust and civic norms, thus producinganother external effect from investment in humancapital. Literature on social capital and growthremains at a relatively early stage and until bettermethods of measuring the social variables can bedevised, social capital, together with vocationaltraining variables, are likely to have a only marginalrole in quantitative growth analyses.

6.10. Possible endogeneity andsimultaneity bias

Sianesi and Van Reenen (2000) have identifiedpossible problems of reverse causality (i.e. growthstimulates education and training) in the links

between investment in human capital and economicgrowth. As per capita income increases, so educa-tional inputs, both in terms of quantity and quality,also grow, but it is not obvious that this economicgrowth is caused by the rising educational stan-dards. Income growth is likely to induce a higherdemand for education and training and much of thisdemand will be income elastic. In countries experi-encing significant economic growth, governmentsare more able to increase public spending on educa-tion and training, and to ensure better access tosuch education for more of the population.

A review of international literature on the demandfor higher education shows that one of the keydeterminants is real income per capita (Briscoe andWilson, 1998). In an empirical study of entrants intoUK higher educational institutions, real income wasidentified as one of the more consistently significantdeterminants, drawn from a wide list of potentialexplanatory variables. Literature on education alsosuggests a similar link between demand for bothprimary and secondary education and the level of,or growth in, national income.

For the growth models discussed in thepresent paper, the issue is whether growth in theeconomy is brought about by the accumulation ofhuman capital or whether the stimulus of growthinduces the workforce to seek after higher educa-tional and training standards. Bils and Klenow(2000) have challenged the findings of Barro andothers (Section 6.6) on the grounds of reversecausality, whereby it is shown that fastereconomic growth induces greater schoolingopportunities by raising private returns.

Sianesi and Van Reenen (2000) maintain thatthe most plausible answer to the causalityproblem is that both processes are at work simul-taneously, so that there is a bi-directionalcausality between investment in human capitaland economic growth. Such considerationssuggest that growth models ideally should recog-nise the endogenous nature of human capital andcontrol for the simultaneity bias. Unfortunately,endogeneity problems apply to other right-handside variables in the growth equations, such asphysical capital accumulation. There would seemto be a shortage of plausible instruments toensure robust empirical estimating.

Aghion and Howitt (1998) have pointed outsome of the potential difficulties in relatingeconomic growth to the initial value of an


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 48

explanatory variable, such as school enrolmentsat the start of the data period (Section 6.5 above).There exists the possibility that both the growthand the enrolment variable are jointly determinedby some external factor, such as the prevailingpolitical regime, that plays no specific part in themodel. There is also a strong likelihood thatexpectations of future economic growth may spurmany workers to invest in education and training

now, in anticipation of future economic opportu-nities. Where time series data are available on asufficiently long basis, lagged values of theseendogenous variables may be used as instru-ments in the estimating equations. However, eventhen, the exogeneity of such lagged estimatorsmay be questioned for, as Temple (1999) hasshown, there may be long and delayed impactsof human capital accumulation on growth.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 49

7.1. Literature strands

Market valuation literature has much in commonwith the production function approach on whichthe growth models discussed above are based.However, much of it is concerned with individualcompanies, although their effects can haveimportant macroeconomic consequences. Thereare two main market valuation literaturesubstrands. The archetypal market value modelsexamine the role of the stocks of intangibleassets, including human capital investments, indetermining market value (13). The estimatedcoefficients effectively provide a measure of themarket’s valuation of these intangibles. A secondsubstrand examines the different impacts of theexpected and unexpected changes in intangibleassets, for example, an unanticipated increase inthe company’s R&D expenditure or a new inven-tion that is reflected in a patent grant. A furtheremerging theme concerns the distribution ofvalues of intangibles, for example, linked to therisky nature of R&D investment.

The market value of a company is taken to be ameasure of the firm’s (potential) dynamic perfor-mance. In essence, the rationale for the model isthat the market value of the firm is higher the greaterthe future profit flows of the firm and, thereby, thehigher the future dividend payment to shareholders.On the assumption that intangible assets are themain source of abnormal profits, then market valueswill reflect the perceived returns on R&D and otherintangible capital, such as goodwill built up throughadvertising. Market value is, therefore, aforward-looking indicator of company perfor-mance, in other words, it is a dynamic measure ofperformance. In the long run, the market value isequal to the discounted sum of expected profits(although it may be higher than this, given that theholding of any share is an option).

The so-called Tobin’s Q form of this relation-ship is set out in a number of places (e.g. Hall,1999), which also describes the potential bias of

the linear approximation that is usually adoptedprior to estimating. This can be written as,

log MV = log Q + a log K + a log (1 + c {R/K}) (8)

where K denotes (the replacement value of)tangible assets and R refers to intangible assets;MV/Q is analogous to Tobin’s Q (and identical toTobin’s Q where a=1 and there is only one type ofasset, a tangible asset, K); the coefficient ‘a’describes the overall scale effect and should beequal to unity under constant returns to scale.The specification may disaggregate K and/or Rinto various components.

The link to the models outlined in Section 6.3 isfairly easy to establish. Clearly, the higher thecontribution of current and past R&D to futuretotal factor productivity, other things being equal,the higher profits and growth are likely to be. Ithas been demonstrated elsewhere that themarket value approach and the production func-tion when applied to individual companies arebroadly consistent from a theoretical perspective.Under (fairly heroic) assumptions regardingperfect information and efficient operation ofcapital markets, in principle, they will yield iden-tical results (Bosworth and Gharneh, 1996).Certainly, the empirical results from the twoapproaches available for the US look broadlyconsistent. A key distinctive feature of marketvaluation literature is that it has, to date, onlybeen investigated for quoted companies, usingfinancial accounting and stock market price data.

7.2. Some empirical results on market value models

This work has recently been reviewed by Toivanenet al. (1998) and others. Studies are usually basedon large-scale panel data sets of individualcompanies. Most of the literature focuses on therole of R&D, rather than human capital per se, and

7. Market value and related literature

(13) A minor offshoot of this literature is the work that attempts to look at the explanation of share prices. For the purposes of thepresent paper, we treat these as broadly equivalent approaches.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 50

the principal result is that the R&D has a signifi-cant positive effect on market value. In general, itmakes little difference whether the R&D flow orstock is used, as the flow is relatively constant foreach company. However, in those studies whichdistinguish between R&D anticipated, e.g. byinvestors, and unanticipated R&D, it is normallythe latter that plays a significant role.

The second variable that has been extensivelyinvestigated is the number of patents, althoughthere is a small amount of work emerging now onthe quality of patents as measured by citationcounts. Market valuation literature suggests astrong positive relationship between patents andmarket value, especially when patents are used inplace of R&D.

Patents represent an outcome of the R&Dprocess. Carrying out R&D and obtaining patentsrequires the deployment of high level skills.However, such activity involves a certain level of risk,so for any particular project there is no definite linkbetween employment of skilled labour and suchoutcomes. Nevertheless, the evidence suggeststhat, on average, they are important. When the twovariables are entered together, they tend to dobroadly the same job, although patents are some-what more noisy and the R&D variable tends todominate. Patents also have the disadvantage thatthey are less relevant outside of certain key manu-facturing sectors, and also in the case of smallcompanies. New intellectual property variables haverecently been tested, including trademark data.Anecdotal evidence suggests that, on average,trademarks are at least as valuable as patents.

7.3. Knowledge, intellectualcapital and intangible assetliterature

The growing interest in knowledge and intellec-tual capital forms one of the most exciting devel-opments in recent economics and managementstudies. It builds upon a number of the earlierdiscussions described above, including the roleof R&D knowledge, intellectual property andintangible assets. However, it should be notedthat work in this area, particularly empiricaltesting of the emerging models, is in its infancyand research seeking to identify various indica-tors or proxies is somewhat speculative.

The literature points to important distinctionsbetween information, knowledge, intellectualcapital and the broader range of intangible assets(Nonaka and Takeuchi, 1995). These distinctionsare important insofar as information and knowl-edge are only inputs into the production of intel-lectual capital, where the latter can be defined asall of the knowledge that has produced or iscapable of producing value for the company. Inother words, intellectual capital is intellectualmaterial that has been formalised, captured andleveraged to produce a higher value asset. Theprincipal focus is the efficiency with whichcompanies access information and transform itinto intellectual capital and other valuable intan-gible assets, thereby determining competitiveadvantage and improvements in performance.

The growth in the importance of intellectualcapital is a result of the shift towards knowl-edge-based, rather than manufacturing-basedproduction, with the implications that this has forthe focus on intangible rather than tangibleassets. Lynn (1998) has claimed that there hasbeen a metamorphosis from a resource andmanufacturing-based economy to one in whichknowledge and services are the key drivers ofeconomic growth. Evidence of the importance ofthe broader range of intangible assets abounds.For example, a number of pharmaceuticalcompanies are sold at many times the book valueof their tangible assets. Similarly, the market valu-ation of many companies is significantly higherthan their balance sheet valuations. In addition, itis argued that organisations have restructured inorder to cope, increasing their agility, by elimi-nating hierarchies and decentralising, in somecases creating ‘spider web’ or ‘fish net’ organisa-tions (Bartlett and Ghoshal, 1993) (Hedlund,1994).

Applying a process perspective, it is possibleto distinguish at least eight categories of knowl-edge focused activities:(a) generating new knowledge;(b) accessing valuable knowledge from outside

sources;(c) using accessible knowledge in decision

making;(d) embedding knowledge in processes, prod-

ucts and services;(e) representing knowledge in documents, data-

bases and software;


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 51

(f) facilitating knowledge growth through cultureand incentives;

(g) transferring existing knowledge into otherparts of the organisation;

(h) measuring the value of knowledge assetsand/or impact of knowledge management(Ruggles, 1998).

Much of this knowledge is created throughinvestment in human capital, with high levels ofeducation and training inputs.

The importance of the management of intellec-tual capital has been pointed out by a number ofcommentators. For example, it has been arguedthat intellectual capital is the invisible skeleton ofthe corporation; that it is, in fact, an economicoperating system, widely viewed as thecompany’s single most valuable asset (Darling,1996). It is only by formal systems of human

capital reporting that intellectual capital can bemade visible; see the discussion in Cedefopsecond research report, Part 3. At the presenttime, the management of intellectual capital is stillan experimental exercise for most organisations,although a number of key companies (e.g.Hewlett-Packard and Dow Chemicals) are recog-nised to be leading the process. The Society ofManagement Accountants of Canada is at theforefront of the promotion of measurement andreporting of intellectual capital. Currently, nocompany is likely to have a comprehensive intel-lectual capital management system in place;many will actively manage parts of their intellec-tual capital, such as their patent portfolios orbrands, or parts of their human capital throughtheir HRM systems. There is a strong link here tothe discussion in Chapter 5.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 52


This section draws together a number of moredisparate areas of literature that have some bearingon the links between education, training, skills andeconomic performance. The concern here is mainlywith company performance because, whencompanies are aggregated together, the macroe-conomy emerges. The key issues are to do withcompany size, including the problems of smallfirms, and with work on structure conduct andperformance. These are considered in turn. As withsome of the other areas of research previouslydiscussed, the importance of the links betweeninvestments in human capital (in various forms) andeconomic success are highlighted.

8.2. Company size and theproblems of small firms

There is a significant amount of literature onvarious aspects of company size and aspects ofperformance, such as growth, with importantlinks to the structure-conduct-performancehypotheses. The material can be traced back atleast to Gibrat’s law of proportionate effect.However, this literature has become increasinglyless mechanistic and more behavioural as it hasevolved over time, branching off in many direc-tions, some of which have already beendiscussed in other sections of this report. In thissection, the focus is wholly on the rather mecha-nistic Gibrat’s law. Structure-conduct-perfor-mance and the small firm elements are examinedin the following subsection.

In essence, Gibrat’s law is concerned withwhether company growth is related to size. Thus,if the following relationship is estimated:

(St+1/S t ) = aS tb-1 (9)

using data on firm size, S (where t denotes theyear), it is possible to test whether large firms

grow faster than small firms (b>1) or whethersmall firms grow faster than large firms (b<1).Simple tests of the law involve no other variables,but more recent tests have included a variety ofother right-hand side variables to control for otherinfluences on growth, including human capital.

There are a number of reviews reporting esti-mates of Gibrat’s relationship (Hay and Morris,1979). The results indicate that the simple formula-tion of Gibrat’s law does not generally hold,although it is accepted that there are some indus-tries in which company growth is significantlyrelated to size (both positively and negatively). Ageneral finding is that variance in growth rate isnegatively related to size; in other words, there is agreater diversity of performance among small firms.For example, in the small firm sector in the UK,approximately half the growth in the sector comesfrom a very small percentage of firms. One early, yetstill valid observation, from Singh and Whittington(1968) was that of the serial correlation of growthrates; in other words, firms that tended to grow fastover one period would grow faster in other periods.This links to the company-fixed effects literature tobe found, for example, in the market value basedstudies of performance.

A recent paper in this tradition, but focusing ongrowth in company size in the service sector, isJohnson et al. (1999). It estimates an equation ongrowth in employment, using existing size as one ofthe explanatory variables. It finds a non-linear rela-tionship between growth and size, with growthdecreasing with size up to five employees, thenrising with size from five to 15 employees, beforefalling again. They note that the higher profitmaximising scale of around 17 employees is in themiddle of the range at which salaried managerialappointments tend to be made by small firms. Theother variables included, however, give counterin-tuitive results, with both the education level of theowner and the number of professionals employedhaving a negative impact on growth rates. There areclearly important differences between small andlarge firms in a number of respects.

8. Further links between education and training and company performance

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 53

8.3. Small and medium sizeenterprises and thequalifications of managers

Bosworth (1999) has shown how there has been amajor growth in interest in the performance ofsmall and medium sized enterprises in recentyears. Interest arises for a number of reasons,including the large proportion of total employmentaccounted for by the small firms’ sector and alsothe high failure rate among small firms, particularlythose that are newly established. Two potentiallyrelevant areas of analysis are identifiable in litera-ture on small firms’, the role of qualifications andthe quality of management within small firms. Thishas tended to be a more qualitative literature,often descriptive or case study in nature.

There are several important problems in comingto any firm conclusions about small businesses,including the general paucity of the data, and, inparticular, the coverage of the very smallest, oftenowner-managed firms (i.e. micro-firms). A secondproblem is that both the data and case studyevidence indicate the extreme heterogeneity offirms in the small business group.

While there is little dispute about the hypoth-esis that small firms would perform better ifmanagement quality were to be increased, it isclear that this may not carry the same implica-tions for the role of qualifications leading toimproved performance. It appears to be the casethat the distinction between qualifications andskills/quality is the greatest for small firms. Smallfirms have quite different management needsand, with the exception of certain professionaland related services, qualifications per se are notseen as important, at least for theowner-manager. Qualifications may be moreimportant for non-owner managers and otheremployees of small firms, in order to gain flexi-bility in the job market.

Studies suggests that a number of qualifica-tions available in the UK, such as MBAs, do notappear to be at all relevant to the small firm. Eventhose which are potentially relevant, such asNVQs and GVNQs, do not appear to be whollyuseful for many small firms. Other countries’experiences suggest that one possible way

forward is to link the training provided to thedevelopment of the business, for example, toprovide training implicitly or explicitly whenproviding help to the business to overcome prob-lems. Demonstrating the usefulness of training tothe development of the business can encouragesmall firms to take further education and trainingsubsequently. This is often best taken forward ona regional or local basis (14).

8.4. Structure-conduct-performancemodels

The bulk of structure-conduct-performance litera-ture is now somewhat dated, although therecontinue to be articles published in this area. Anoverview of this literature suggests that there areessentially two main strands of investigation:(a) static models, which focus on the impact of

company size and market power on profits (orother current performance measures);

(b) dynamic models, which focus on the impactof company size and market power on inven-tion and innovation and, thereby, by implica-tion, on company growth and performance.

The key importance of the dynamic models isthe role that they ascribe to managers (includingtheir skills and qualifications) and the importancethey attach to intangible assets (including knowl-edge and skills) in determining company perfor-mance.

Managerial theories, first seen in the 1960s and1970s, were primarily concerned with companygoals in large, professionally managed compa-nies, as opposed to the small owner-managedcompanies. In general, the former were argued tobe sales revenue, utility or growth maximisersand the latter were depicted as profit maximisers.This strand of the literature involves some form ofoptimisation and the rationale for this literaturewas that professional managers would act in theirown best interests. The main theoretical predic-tions are that, in the main, managerial modelssuggest higher levels of labour demand than theprofit maximising case. In addition, a number ofthese models suggest a skewing of demandtowards higher-level skills and occupations.


(14) The examples of Emilia Romagna in Italy and the small firm hubs in India and Pakistan, in particular, are examples of localsmall firm synergies.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 54

Behavioural theories, which developed a fewyears after their managerial counterparts,discussed firms as ‘satisfiers’, rather than opti-mising units, operating under uncertainty andimperfect information. These behavioural theoriessaid little about labour demands per se, butsuggest that all firms operate with some degreeof inefficiency (i.e. X-inefficiency), only part ofwhich it is economic to remove.

In practice, most empirical literature is basedupon fairly ad hoc tests of relationships betweenperformance, company size and market struc-ture. Bosworth (1983) has argued that there arevarious categories of model that are potentiallyrelevant. In those which have the longest pedi-gree, causality runs from company size to perfor-mance:

P = f (S, C, X) (10)

where P is a measure of performance, such asprofitability or sales growth, S refers to size, Cdenotes the degree of concentration in themarket and X is a vector of other explanatoryvariables, which often have included factors suchas advertising expenditure. This is of someinterest, as size is purported to affect perfor-mance, and size is almost certainly a proxy forother things such as skills, knowledge and R&D.

If different sizes of companies had differentgoals, this would have direct implications for theirperformance and for their labour demands,including the demand for skills.

In the case of dynamic structure-conduct-performance models, while it is true to say thatincreases in size give rise to greater R&D, it isless clear whether the increase in R&D is justproportionate, more than proportionate or lessthan proportionate with size. In addition, thefinding that formal R&D only begins above acertain size threshold is muddied by the fact thatsmall firms may still undertake R&D, but this isnot a separate function and tends to be moreinformal in nature than in larger firms. This ismade even less clear by the fact that there isfairly strong evidence that many major inventionshave been made by individuals rather than by

large corporations. In addition, as noted above,changes in R&D expenditure (an input measure ofinvention and innovation) with company size saysnothing about the relationship between R&Doutputs and size. This issue does not appear tohave been resolved. However, the empiricalevidence seems to support the hypothesis thatlarger firms are better innovators than small ones,and that R&D expenditures and associatedinvestments in human capital help in the scaleand success of innovation activities.

The stylised views of management behaviourset out in managerial and behavioural theorieshave proved to be far too simplistic. While profitmaximisation seems a reasonable first approxi-mation of the likely goal of an owner-manager, inpractice, even here there appears to be a muchmore complex picture. Bosworth and Jacobs(1989) reviewed the various literatures aboutmanagement goals and behaviour. They found awide range of goals, including the maximisationof profits. Equally, however, they included themaximisation of the owner’s utility (which wasinfluenced by time spent working, by the rewardsgained from extra work and by the status andpower conferred by self-employment and thecontrol of others). Other managers purported tominimise their risk, growth often being associatednot only with high risk, but also with personal andsocial (family) costs.

There were some characteristics that helped todetermine which goal would be set. It was oftenthe older small business, rather than the newer,that tended to adopt more conservative and lessdynamic goals. Owner-managers themselvestended to be divided between the artisan and theentrepreneurial, an almost cultural differencebetween individuals. In addition, psychologicalmake-up of the individual was also important;often the individual’s need to maintain controlconstrained the willingness to grow. Wherecontrol was important, the manager often exhib-ited a reluctance to employ more highly qualifiedindividuals, as the latter might deploy reasoningand knowledge that the managers would notunderstand, thereby leading to a loss of control.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 55


Evidence of the positive influence of education,training and skills on performance is mostapparent at the level of the individual. Commenthas a tendency to focus on private rather thansocial performance. Some, although by no meansall, of these benefits can also be expected toaccrue at a macro level. Some of the benefitsmay, of course, be dissipated in competitionbetween individuals or companies, with no overallsocial gain.

Several literature strands indicate that theremay be important additional spill-over effects andexternalities. Spill-overs include technological,spatial and environmental aspects. Some refer-ences have already been made to externalities inSection 6.8 above.

9.2. Technological spill-overs

It is neither necessary nor appropriate to dwell ontechnological spill-overs, as they are a naturalextension of both endogenous growth theories(Chapter 6) and the market valuation approach, inwhich the performance of a particular company isinfluenced by the R&D or patents of othercompanies (Chapter 7). In the case of patents, forexample, the publication of the patent specifica-tion discloses information that may be of use tocompetitors or to companies in other productareas which use (or might use) related technolo-gies. In practice, many of the market valuestudies that have investigated the role ofspill-overs have found that they appear to bestatistically more important than the firm’s ownR&D or patents. There are a number of existingreviews (Griliches, 1981 and 1992). The key vari-ables are R&D and/or patent flows and/or stocks,although, in this case, the focus is on the generalpool of intellectual property generated by othercompanies. There are clearly potentially impor-tant links with the knowledge management area,including the sources and management of infor-

mation, and the most appropriate organisationalstructure to maximise the benefits to thecompany of the intellectual capital pool.

9.3. Spatial and city dynamics

Spatial literature considers the synergies thatstem from a particular location. One particulartheme has been the dynamics of small firm clus-ters, stimulated originally by the dynamics ofsmall firms in the Emilia Romagna region of Italy(Goodman and Bamford, 1989). However, therehas been similar work in developing countries,including India and Pakistan (Nadvi, 1998). In thecase of the small firm surgical instruments clusterin Sialkot, Pakistan, for example, ‘Clustering hasallowed small businesses to become highlyspecialised, and to benefit from externaleconomies and joint action.’ The evidencesuggests that such clusters benefit from bothvertical and horizontal synergies, from coopera-tion and other forms of spill-overs, as well asfrom intense competition between rivals. Suchclusters also benefit from the presence ofexternal institutions, such as the Metal IndustriesDevelopment Centre, in Pakistan, which is widelyused by local small and medium size enterprises.External agents form the principal sources ofinformation and technical know-how.

Spatial literature resides primarily witheconomic geographers and a comprehensivesurvey of UK urban labour markets was carriedout by Turok (1999). Some 20 cities with a popu-lation size above 250 000 are identifiable from the1991 census and of these, 8 are deemed to beconurbations, while 12 are free-standing. These20 cities accounted for more than 40 % of allnational jobs and so they exhibit a high concen-tration of both skills and companies.

During the 1980s and 1990s these citiescontinued to suffer from a relative decline in jobavailability, especially for full-time males in thelocal population. In consequence, there was aslow outward migration of skills from the urbancentres and reduction in male activity rates

9. Spill-over effects and externalities

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 56

(Turok, 1999). The long-term trend out of manu-facturing and into service sector activitycontinued and its impact was most strongly felt inthe cities. Although government provided a rangeof training schemes for the unemployed, somekey skills were eroded, as the jobs were lost,especially in the recession of the early 1990s.When the economy began to recover later in thedecade, companies found problems with skillshortages and deficiencies in the cities.

For the highly qualified workforce, the city hasincreasingly become a magnet for job opportuni-ties. Warf (1995) has noted the geographic reper-cussions of the growth of telecommunications onthe development of world cities, of which Londonis one of the foremost. There has been anagglomeration of professional jobs on offer astelecommunications offices have mushroomedacross the city. Graham (1999) has noted how ‘ahost of value-added and consultancy firms arealso concentrated in the city, supportingstate-of-the-art innovation, financial telecomsservices and corporate telematics applications’. Itis clear that synergies of all kinds stem from citylocation.

As the knowledge economy gathers strength,so cities become ever more important. It hasbeen claimed that trends towards decentralisa-tion, which were observable from 1960 to 1980,are now in the process of being reversed (Aminand Graham, 1997), although in some parts of theEU there remains statistical evidence of with-drawal of firms and people from large agglomera-tions to rural areas. Large cities at the end of thetwentieth century have become the nationaleconomic motor, acting as powerhouses for thepersonal interaction which drives the teamworkand creativity that is essential for good corporateperformance. Skill training which concentrates oncity professional workers is likely to produce thehighest social benefits, even where labourturnover is significantly high for individual compa-nies. A key feature of much of the literature in thisarea is the emphasis on professionalisation ofemployment in many cities and at the same timethe polarisation of employment structure withmany low skilled and low paying jobs beingcreated (e.g. Hamnett,1996).

A recent study by OECD (2001b) providesevidence of links between skills acquisition andeconomic growth in 180 regions across

15 Member States. Correlations were madebetween primary, secondary and tertiary educationlevels and regional GDP per capita. Whilst tertiaryeducation was important, the strongest significantassociation was with secondary education. Theformer level of education is critical for R&D andrelated innovations, but it is secondary educationthat provides the intermediate skills that are criticalto industrial know-how and learning-by-doing.Correlations at the regional level are blurred by themobility of labour between regions.

Moretti (1999) used US census data to try toestimate the external returns on education bycomparing wages for otherwise similar individualswho work in cities with differential levels ofeducation. Estimating problems arose in trying toisolate the causal impact on individual wages ofthe average level of city-specific education. Themain finding was that a one-year increase inhigher education in a city raised average wagesby between 8 and 15 %, after controlling for theprivate return on education. However, this resultdoes not necessarily point to an external effect,as it may simply be caused by complementaritiesbetween well and poorly educated workers.There are significant econometric difficulties inachieving robust results in this type of study.

9.4. Environmental externalities

Investment in education produces environmentalspill-overs in many different ways. One obviousarea is through health and life-expectancy. Higherlevels of investment in human capital creategreater awareness of the potential causes ofillness and promote increasingly healthierlife-styles. A more highly educated workforcechoose safer occupations and societies withhigher relative education bring in health andsafety laws to protect the employees. Grossmanand Kaestner (1997), after controlling for percapita income, found that people with moreeducation simply live longer. One particular mani-festation of the external impact of education onhealth is in the reduction in infant mortality ratesin developing countries. Women with higherlevels of education are more aware of thedangers to their children’s health and they aremore likely to take the steps that lead to lowerinfant mortality. Wolfe and Zuvekas (1997)


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 57

demonstrate that better education also promptslower fertility rates and smaller family sizes givethe children better health and survival prospects.

McMahon (2000) has argued that higher levelsof education will lead to the creation of moresustainable environments, as greater awarenessof physical environmental damage becomesapparent. Initially, faster economic growth isassociated with global warming, deforestation

and many forms of pollution, but educationenables populations to take steps to rectify someof this damage without sacrificing the growthgains. Most developed societies, with relativelyhigher human capital investments, have environ-mental protection laws. The spill-over of educa-tion on the physical environment is an expandingarea of specialist study that lies beyond thereaches of the present paper.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 58

This section contains an overview of the key find-ings in the paper. It also attempts to drawtogether the implications for policy and to identifypossible directions for future research.

Some of the material presented in this sectionalso appears in the executive summary.

10.1. Summary of key points

This paper has presented an extensive interna-tional review into the various strands of literaturethat bear on the links between investment ineducation, training and skills and economicperformance at the level of the macroeconomy.Where training and education conducted by firmsand industries was identified as contributing tooverall economic growth, this was also reviewed.It was not possible to segregate macro frommicro studies, as the amount of overlap wasfound to be considerable.

An initial examination of recent EU statisticsmeasuring GNP, employment, unemployment andvarious educational and vocational traininginvestments suggested that economic growthacross the 15 Member States is associated withincreases in both education and training.However, a fuller appraisal of the data revealedthe difficulties in demonstrating significant causallinks between growth and investment in humanresources. The amount of literature in this areaunderscored the complexity of this relationship.

Studies on the general rate of return werefound to demonstrate strong links betweeneducation and training and economic perfor-mance, not only for the individual, but for societyas a whole. Evidence on social rates of return hasconcentrated on the benefits to the economyarising from increased education, calculating allthe costs of schooling and education comparedto the relative pre-tax earnings of individuals inreceipt of such education. Many qualifyingassumptions needed to be made in arriving at aquantifiable estimate, but several studies havesuggested a social rate of return in the range 6 to12 %. Comparable studies to establish the social

rate of return for vocational training proved hardto find, although there was some evidence thatsuch training has a significant positive impact onproductivity.

Literature on the management of humanresources at both company and industry levelexists in both micro and macro level studies. Thisresearch stresses the belief that individualworkers are a key source of an organisation’scompetitive advantage. It is the quality of thesehuman resources that determines organisationalperformance. There appeared to be a number ofdifferent emphases, including the role played byHRM in promoting such resources, and the roleof management itself, particularly top managers.One important theme was the role of leadership.Another theme stressed the importance of work-force skill qualifications in contributing to higherlevels of productivity across different industrialsectors. This literature emphasised the impor-tance of education and training in often verycomplex industrial processes.

Economic growth models directly explore thequantitative relationship between investments ineducation and training and the level and growthof per capita GDP at national level. This covers alarge number of studies, beginning with thegrowth accounting models, through to theendogenous growth models that are widelyapplied in many recent empirical studies. Bothdata sets and econometric modelling techniqueshave developed extensively over recent yearsand many different model specifications havebeen proposed and empirically tested. Typically,these models use data drawn from across-section of different countries, sometimesonly for developed countries but often for a widerset of nations. Data difficulties commonly meanthat consistent series of educational variables arehard to obtain over sufficiently long time periodsto facilitate econometric time series analyses. Fordeveloped countries over more recent years,cross-sectional and time series data have beencombined into panel sets to enable a morecomprehensive testing of the relationshipbetween human resources and economic growth.

10. Conclusions

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 59

Overall, these growth models demonstrate thathigher educational investments have had a signif-icant impact on national economic growth.Broadly, the weight of evidence suggests that a1 % increase in school enrolment rates has leadto an increase in GDP per capita growth ofbetween 1 and 3 %. An additional year ofsecondary education, which increases the stockof human capital, rather than simply the flow intoeducation, has lead to a more than 1 % increasein economic growth each year. The results varyconsiderably, depending upon details of modelspecification and the precise data sets in use.Issues dealing with the robustness of such resultsare widely discussed. Almost all these growthmodels use only a limited set of educational vari-ables to proxy human resource input and they donot incorporate any vocational training or socialcapital variables.

Market valuation literature is an areaconcerned with the factors that determine thegrowth of individual companies and so, indirectly,the overall economy. Research studies in thisarea are concerned with how intangible assets,such as technical knowledge, influence potentialperformance and net worth. These are intimatelytied to investment in education and training, aswell as skills, which may often be a necessary, ifnot sufficient, condition for improving such intan-gible assets. Research and development, patentsand intellectual capital are key determinants ofcorporate market value.

Further studies relate to the links betweeneducation, training, skills and economic perfor-mance. This is concerned with company size andthe structure, conduct and performance. Manystudies have emphasised the role of the smallfirm in economic growth. Such small firms are anextremely heterogeneous group that havemarkedly different training needs from those oftheir larger counterparts. There is often far lessemphasis on educational and training qualifica-tions in small firm environments. Empiricalresearch studies suggest that conventional theo-ries of company behaviour ware far too simplisticfor explaining the goals and aspirations of thesesmaller organisations.

The final strand of literature deals with some ofthe indirect effects of investment in humanresources that are not normally captured inmeasured national economic growth. Such

spill-overs and externalities can be technological,spatial or environmental. Each can contributesignificant social gain. Many of these effects arefound to be related to other areas covered in thisreview. The spatial externalities, that embrace citydynamics, demonstrate how geographical clus-tering of businesses employing highly qualifiedworkers can produce high productivity and stronglocal economic growth. Environmental spill-oversare often related to health and life expectancy,which typically increase in consequence of higherlevels of education.

10.2. Some policy implications

The overall conclusion from this body of work isthat the impact of investment in education andtraining on national economic growth is positiveand significant. However, its total effect is difficultto measure with any precision, as many of theinfluencing mechanisms are indirect andcomplex. Education is certainly a key determinantof economic growth for both developed anddeveloping countries: its impact is probably moremarked in developing nations. The link betweenvocational training and economic growth is moredifficult to demonstrate at the macro level, mainlybecause of measurement difficulties associatedwith training investments. It is easier to see apositive impact from vocational training atcompany and industry levels, although againthese are difficult to quantify precisely.

Despite the qualifying assumptions and compu-tational difficulties, evidence suggests a very signif-icant return for both the individual and society as awhole from investments in higher education indeveloped countries. Computed social rates ofreturn vary between countries and over differenttime periods, but most are positive and they providejustification for government continuing investmentin higher education. However, there are emergingconcerns of graduate overcrowding, as thepercentage of graduates in the labour marketcontinues to increase, and this trend needs carefulmonitoring to ensure that social returns are notdiminished through overqualification.

The computation of social rates of return onvocational training, equivalent to that for highereducation, is very difficult to achieve because of theinconsistent data usually available on such training


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 60

investments. Research supports the claim thatvocational training improves the performance of theindividual in the workplace and so raises theproductivity of the employing organisation. Studieson HRM argue that the skill of the individual workeris often a critical factor in an organisation’s compet-itive advantage. Comparative international studiesemphasise the key importance of skills training inraising productivity. Despite this, employers inmany countries have a rather negative attitudetowards investment in skills and training. Unless theinstitutional and legal infrastructure encouragesemployers to invest (as in Germany) there is atendency to rely on others to carry out the invest-ment (as is often the case in the UK). It seems clearthat governments should provide every incentive tocompanies, either through tax allowances or directgrants, as well as through reforms to institutionalstructures, etc., to encourage them to increase theirinvestment in training. Such training can providestrong spill-over effects for the whole of society.

The weight of research evidence suggests thatwhere governments manage to increase the stockof the educated workforce, through additionalyears of secondary or tertiary education, rather thansimply enhancing the flow into education in anygiven year, the impact on economic growth is moresignificant. Such a finding directs governmentpolicy towards changes to improve the quality ofeducation rather than just the numbers passingthrough the system in the short term. For mostMember States the emphasis is on access totertiary education and the quality of the graduatesproduced by universities. For less developed coun-tries the critical factor is secondary school enrol-ment rates and the need to extend the number ofyears in such education. It is difficult to measure therelative quality of education in different countries,but this would appear to be a key issue for govern-ments to address if educational quality is to besignificantly improved.

The role of R&D in the growth of firms isstrongly underlined in various studies and it isclear that successful R&D depends on skilledhuman resources. Governments need to provideincentives to firms to continue to make invest-ments in the education and training thatenhances their innovative capabilities. In partic-ular, attention needs to be paid to the role of thesmall firm in contributing to R&D and so to overalleconomic growth. Often small firms may not have

sufficient resources to invest in training theirworkforce, but they can be expected to benefitfrom training carried out by larger firms when thetrained workers subsequently move to new jobsin the small firm sector.

There is no guarantee that any investment inhuman capital will result in a positive return,whether it be investment in training, deploymentof highly skilled labour, R&D or in some otherform. There will always be some risk and uncer-tainty. In general, the evidence suggests thatsuch investments pay off but each case needs tobe considered on its merits.

10.3. Possible areas of futureresearch

This review has drawn attention to difficulties withpublished data related to the measurement ofeducation and training provision. Many of thegrowth model studies raised problems with dataquality. In particular, consistent time series ofeducational statistics were often lacking or thetime-span coverage was too short for meaningfulanalysis. Empirical results often proved sensitive tothe exact data set in use. The compilation of accu-rate educational statistics measuring both quantityand quality of provision is an important area ofresearch. Information on the quality of teaching wasfrequently unavailable to researchers and thisundoubtedly limited the rigour of the analysis.

Most of the reviewed research studies wereconcerned only with educational variables anddid not extend to vocational training. This limita-tion arises from the restricted availability of suit-able data on training provision at industry or theeconomy level. Individual companies oftenproduce such data but they are less frequentlyavailable at higher levels of aggregation. There isa strong need to continue to produce consistentstatistics on all types of training provision and toensure that these statistics are broadly compa-rable across EU countries.

This paper touched upon insights into linksbetween skills, education and training andproductivity growth at corporate level. However,given that the prime emphasis was the link tomacroeconomic growth, it was not possible hereto explore this area fully. Specifically, researchstudies in HRM and the market valuation of


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 61

companies could be examined in greater depth toreveal the mechanisms by which education andtraining investments lead to improved corporateperformance.

Relatively recent studies on social capital andits contribution to economic growth was only

touched upon in this paper but it would certainlywarrant a much fuller exploration. Several papershave now been published on social capital but itsrole as a determinant of economic growth needsto be tested empirically. Again, some importantdata difficulties will first need to be overcome.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 62

List of abbreviations

CEO Chief executive officer

ELFS European labour force surveys

EU European Union

GDP Gross domestic product

GNP Gross national product

HRM Human resources management

IALS International adult literacy survey

ISCED International standard classification of education

NIESR National institute of economic and social research

R&D Research and development

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 63

Annex

International standard classification of education (ISCED)

Educational level Details/age range Common terminology Classification

Early childhood education Introduction of pre-school Pre-primary, nursery, ISCED 0children to a school-type kindergarten, pre-schoolenvironment, from age 3

Primary education First stage of Elementary school ISCED 1basic schooling,up to age 11 or 12

Lower secondary education Second stage of Junior high school ISCED 2basic schooling,up to age 14 or 15

Upper secondary education Stage leading to final Senior high school, Lycee, ISCED 3secondary qualification, Gymnasium, Sixth form, ISCED 4typically age 18 or 19 Further education

Tertiary education Programmes significantly Higher education, ISCED 5Amore advanced than upper College education ISCED 5Bsecondary studies ISCED 6

University level education Studies leading to a first ISCED 5Adegree, at least bachelor’s ISCED 6level or equivalent

Source: OECD (2001a).

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 64

References

Acemoglu, D. Microfoundation for social increasingreturns in human capital accumulation. Quarterly Journal of Economics, 1996, Vol. 111,No 3, p. 779-804.

Adler, P. S. Managing flexible automation. CaliforniaManagement Review, 1988, Vol. 30, No 3,p. 34-56.

Aghion, P.; Howitt, P. Endogenous growth theory.Cambridge, Ma: MIT Press, 1998.

Amin, A.; Graham, S. The ordinary city. Transactionsof the Institute of British Geographers, 1997,Vol. 22, No 4, p. 411-429.

Ashworth, J. A waste of time? Private rates of returnto higher education in the 1990s. Higher Education Quarterly, 1997, Vol. 51, No 2,p. 164-188.

Asteriou, D.; Agiomirgianakis, G. M. Human capitaland economic growth: time series evidencefrom Greece. Journal of Policy Modeling, 2001,Vol. 23, No 5, p. 481-489.

Ballot, G.; Fakhfakh, F.; Taymaz, E. Firms humancapital, R&D and performance: a study onFrench and Swedish firms. Journal of LaborEconomics, 2001, Vol. 8, No 4, p. 443-462.

Barrett, A. Economic performance of education andtraining: costs and benefits. In: Descy, P.;Tessaring, M. (eds) Training in Europe. Secondreport on vocational training research in Europe2000: background report. Luxembourg: Officefor Official Publications of the EuropeanCommunities, 2001, Vol. II, Part 4, p. 383-404(Cedefop Reference series, 3008).

Barro, R. J. Determinants of economic growth: across-country empirical study. Cambridge, MA:MIT Press,1997.

Barro, R. J. Economic growth in a cross-section ofcountries. Quarterly Journal of Economics,1991, Vol. 106, No 2, p. 407-443.

Barro, R. J. Education and economic growth. Paperpresented at the international symposium onthe contribution of human and social capital tosustained economic growth and well-being,organised by the OECD and HRDC, QuebecCity, Canada, 19-21 March 2000.

Barro, R. J.; Lee, J.-W. International comparisons ofeducational attainment. Journal of MonetaryEconomics, 1993, Vol. 32, No 3, p. 363-394.

Barro, R. J.; Lee, J.-W. International data on educa-tional attainment: updates and implications.Oxford Economic Papers, 2001, Vol. 53, No 3,p. 541-563.

Barro, R. J.; Lee, J.-W. International measures ofschooling years and schooling quality. Amer-ican Economic Review, 1996, Vol. 86, No 2,p. 218-223 (Papers and proceedings).

Barro, R. J.; Sala-I-Martin, X. Economic Growth.New York: McGraw Hill, 1995.

Barry, R.; Bosworth, D.; Wilson, R. Engineers in topmanagement. Coventry: IER – Institute forEmployment Research, 1997 (Research report).

Bartlett, C. A.; Ghoshal, S. Beyond the M-form:toward a managerial theory of the firm.Strategic Management Journal, 1993, Vol. 14,p. 13-46.

Bassanini, A.; Scarpetta, S. Does human capitalmatter for growth in OECD countries? Evidencefrom pooled mean-group estimates. Paris:OECD, 2001 (ECO/WKP(2001)8).

Becker, G. Human Capital. Chicago: University ofChicago Press, 1964.

Behrman, J.; Rosenzweig, M. Caveat emptor:cross-country data on education and the laborforce. Journal of Development Economics,1994,Vol. 44, p. 141-171.

Benhabib, J.; Spiegel, M. The role of human capitalin economic development: evidence fromaggregate cross-country data. Journal ofMonetary Economics, 1994, Vol. 34, No 2,p. 143-173.

Berndt, E. R. The practice of econometrics: classicand contemporary. Reading, MA: AddisonWesley, 1991.

Bils, M.; Klenow, P. J. Does schooling causegrowth? American Economic Review, 2000,Vol. 90, No 5, p. 1160-1183.

Blaug, M. An introduction to the economics ofeducation. Harmondsworth: Penguin Books,1972.

Blechinger, D.; Pfeiffer, F. Qualification, employmentand technical progress: further empirical

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 65

evidence based on data of the Mannheim inno-vation panel. Mannheim: ZEW – Centre forEuropean Economic Research, 1998 (Discus-sion paper No 98-04).

Blundell, R.; Dearden, L.; Sianesi; B. Estimating thereturns to education: models, methods andresults. London: CEE – Centre for theEconomics of Education (LSE), 2001 (Discus-sion paper, DP 16).

Bosworth, D. L. Management skills in the UK.Research report prepared for the Departmentfor Education and Employment. Manchester:Manchester School of Management, UMIST,1999.

Bosworth, D. L. Recent developments in theeconomics of technological change literature.Economics, 1983, Vol. 19, Part 3, No 83,p. 104-107.

Bosworth, D. L.; Gharneh, N. S. Dynamic activitiesand firm performance. In: Oppenländer, K. H.;Poser, G. (eds) Business cycle surveys: forecasting and methodological aspects. Alder-shot: Avebury, 1996.

Bosworth, D. L.; Jacobs, C. Management attitudes,behaviour and abilities. In: Barber, J.;Metcalf, J. S.; Porteous, M. (eds) Barriers togrowth in small firms. London: Routledge, 1989,p. 20-38.

Bosworth, D. L.; Wilson, R.; Taylor, P. Technologicalchange: the role of scientists and engineers.Aldershot: Avebury Press, 1992.

Briscoe, G.; Wilson, R. The demand for highereducation: a study of the size and nature of theflow of entrants into higher education in the UK.Coventry: IER – Institute for EmploymentResearch, 1998 (Research Paper).

Broadberry, S. N.; Wagner, K. Human capital andproductivity in manufacturing during the twentieth century: Britain, Germany and theUnited States. In: Van Ark, B. Crafts, N. (eds)Quantitative aspects of post-war Europeaneconomic growth. Cambridge: CambridgeUniversity Press, 1996.

Brynin, M. Graduate density, gender and employ-ment. British Journal of Sociology, 2002a,Vol. 53, No 3, p. 363-381.

Brynin, M. Overqualification in employment. Work,Employment and Society, 2002b, Vol. 16, No 4,p. 637-654.

Büchel, F. Overqualification: reasons, measure-ment issues and typological affinity to unem-

ployment. In: Descy, P.; Tessaring, M. (eds)Training in Europe. Second report on vocationaltraining research in Europe 2000: backgroundreport. Luxembourg: Office for Official Publica-tions of the European Communities, 2001,Vol. II, Part 4, p. 453-560 (Cedefop Referenceseries, 3008).

Campbell, M. Skills in England 2001: the researchreport. Nottingham: DfES Publications, 2001.

Chevalier, G. et al. The returns to higher educationteaching. London: CEE – Centre for theEconomics of Education (LSE), 2001.

Darling, M. S. Building the knowledge organisation.Business Quarterly, 1996, p. 61-66.

De La Fuente, A.; Doménech, R. Human capital ingrowth regressions: how much difference doesdata quality make? Paris: OECD, 2000(ECO/WKP(2000)35).

Dearden, L. et al. The returns to academic andvocational qualifications in Britain. Bulletin ofEconomic Research, 2002, Vol. 54, No 3,p. 249-274.

Descy, P.; Tessaring, M. (eds) Training in Europe.Second report on vocational training research inEurope 2000: background Report. Luxembourg:Office for Official Publications of the EuropeanCommunities, 2001, Vol. II, Part 4, p. 383-404(Cedefop Reference series, 3008).

Elson-Rogers, S. (ed.) Approaches and obstacles tothe evaluation of investment in continuing vocational training: discussion and case studiesfrom six Member States of the European Union.Luxembourg: Office for Official Publications ofthe European Communities, 1998 (CedefopPanorama series, 78).

Englander, A. S.; Gurney, A. Medium-term determi-nants of OECD productivity. OECD EconomicStudies. Paris: OECD, 1994, No 22, p. 49-109.

Fersterer, J.; Winter-Ebmer, R. Human capital andearnings in Austria. In: Asplund, R.; Pereira, P. T.(eds) Returns to human capital in Europe: areview of literature. Helsinki: Taloustieto Oy,1999.

Gallie, D.; White, M. Employee commitment and theskills revolution. London: Policy Studies Insti-tute, 1993.

Gemmell, N. Evaluating the impacts of humancapital stocks and accumulation on economicgrowth: some new evidence. Oxford Bulletin ofEconomics and Statistics, 1996, Vol. 58, No 1,p. 9-28.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 66

Goodman, E.; Bamford, J. Small firms and industrialdistricts of Italy. London: Routledge, 1989.

Graff, M The significance of education within theprocess of economic development. KölnerZeitschrift für Soziologie und Sozialpsy-chologie, 1996, Vol. 48, No 2, p. 274-295.

Graff, M. Education and technical progress asdeterminants of economic development: a theo-retical and empirical investigation. Heidelberg:Physica-Verlag, 1995.

Graham, S. Global grids of glass. Urban Studies,1999, Vol. 36, No 5/6, p. 929-949.

Green, A.; Preston, J.; Malmberg, L. Non-materialbenefits of education, training and skills at amacro level (in this volume).

Griliches, Z. Education, human capital and growth: apersonal perspective. Journal of LaborEconomics, 1997, Vol. 15, No 1, Part 2,p. S330-S344.

Griliches, Z. Market value, R&D and patents.Economic Letters, 1981, Vol. 7, No 2,p. 183-187.

Griliches, Z. The search for R&D spillovers. Scandi-navian Journal of Economics, 1992, Vol. 94,p. S29-S47.

Grossman, M.; Kaestner, R. Effects of education onhealth. In: Behrman, J. R.; Stacey, N. (eds) Thesocial benefits of education. Ann Arbor, MI:University of Michigan Press, 1997.

Hall, B. H. Innovation and market share. Cambridge,Ma: NBER – National Bureau of EconomicResearch, 1999 (Working Paper, w6984).

Hall, B. H. The value of intangible corporate assets:an empirical study of the components of Tobin’sQ. Berkeley, CA: University of California, 1992(Economics Working Papers, 93-207).

Hall, R. E.; Jones, C. I. Why do some countriesproduce so much more output per worker thanothers? Quarterly Journal of Economics, 1999,Vol. 114, No 1, p. 83-116.

Hall, R. H. Organisations, structure and process.Englewood Cliffs, NJ: Prentice Hall, 1977(Second edition).

Hamnett, C. Social polarisation, economic restruc-turing and welfare state regimes. UrbanStudies, 1996, Vol. 33, No 8, p. 1407-1430.

Hansson, B.; Johanson, U.; Leitner, K. H. Theimpact of human capital and human capitalinvestments on company performance:evidence from literature and European surveyresults (in this volume).

Hanushek, E. A.; Kimko, D. Schooling, labour forcequality and the growth of nations. AmericanEconomic Review, 2000, Vol. 90, No 5,p. 1184-1208.

Harmon, C.; Walker, I. The returns to education: areview of the evidence, issues, and deficienciesin the literature. Sheffield: Department forEducation and Skills, 2001 (Research report).

Hay, D. A.; Morris, D. J. Industrial economics: theoryand evidence. Oxford: Oxford University Press,1979.

Hedlund, G. A model of knowledge managementand the N-form corporation. Strategic Manage-ment Review, 1994, Vol. 15. p. 73-90.

Hujer, R.; Caliendo, M.; Radi, D. Methods and limi-tations of evaluation and impact research. In:Descy, P.; Tessaring, M. (eds) Evaluation andimpact of education and training: backgroundreport. Luxembourg: Office for Official Publica-tions of the European Communities, 2003.(Cedefop Reference series).

Islam, N. Growth empirics: a panel data approach.Quarterly Journal of Economics, 1995, Vol. 110,No 4, p. 1127-1170.

Izushi, H.; Huggins, R. Empirical analysis of humancapital development and economic growth inEuropean regions (in this volume).

Jarvis, V.; O’Mahony, M.; Wessels, H. Productquality, productivity and competitiveness: astudy of the British and German ceramic table-ware industries. London: National Institute ofEconomic and Social Research, 2002 (Occa-sional paper, No 55).

Jenkins, H. Education and production in the UnitedKingdom. Oxford: Nuffield College, 1995(Economics discussion paper, No 101).

Johnson, P.; Conway, C.; Kattuman, P. Small business growth in the short run. Small Busi-ness Economics, 1999, Vol. 12, No 2p. 103-112.

Jorgenson, D. W.; Fraumeni, B. M. Investment ineducation and US economic growth. Scandina-vian Journal of Economics, 1992, Vol. 94,p. S51-S70.

Jorgenson, D. W.; Griliches, Z. The explanation ofproductivity change. Review of EconomicStudies, 1967, Vol. 34, No 3, p. 249-283.

Judson, R. Economic growth and investment ineducation: how allocation matters. Journal ofEconomic Growth, 1998, Vol. 3, No 4,p. 337-360.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 67

Knack, S.; Keefer, P. Does social capital have aneconomic payoff? A cross-country investiga-tion. Quarterly Journal of Economics, 1997,Vol. 112, No 4, p. 1251-1288.

Koch, R.; Reuling, J. Institutional framework condi-tions and regulations of initial vocationaltraining using Germany, France and GreatBritain as examples. In: Tessaring, M. (ed.)Vocational education and training – the Euro-pean research field: background report.Luxembourg: Office for Official Publications ofthe European Communities, 1998, Vol. 1,p. 1-32. (Cedefop reference series, 3002).

Koman, R.; Marin, D. Human capital and macroe-conomic growth: Austria and Germany1960-1997. Munich: Münchener Wirtschafts-wissenschaftliche Beiträge, 1999 (Discussionpaper, 99-10).

Krueger, A. B.; Lindahl, M. Education and growth:why and for whom? Princeton: PrincetonUniversity, 1998 (Working paper, No 429).

La Porta, R. et al. Trust in large organisations. Amer-ican Economic Review, 1997, Vol. 87, No 2,p. 333-338.

Lee, K.; Pesaran, M. H.; Smith, R. Growth andconvergence in a multi-country empiricalstochastic Solow model. Journal of AppliedEconometrics, 1997, Vol. 12, No 4, p. 357-392.

Lieberson, S.; O’Connor, J. F. Leadership andorganisational performance: a study of largecorporations. American Sociological Review,1972, Vol. 37, No 2, p. 117-130.

Lubatkin, M. H. et al. Stockholder reactions to CEOchanges in large corporations. Academy ofManagement Journal, 1989, Vol. 2, No 1,p. 47-68.

Lucas, R. E. On the mechanics of economic devel-opment. Journal of Monetary Economics, 1988,Vol. 22, No 1, p. 3-42.

Luttringer, J-M. Conceptual introduction. In:Elson-Rogers, S. (ed.) Approaches and obsta-cles to the evaluation of investment in contin-uing vocational training: discussion and casestudies from six Member States of the EuropeanUnion. Luxembourg: Office for Official Publica-tions of the European Communities, 1998,p. 3-7 (Cedefop Panorama series, 5078).

Lynn, B. E. The management of intellectual capital:the issues and the practice. Hamilton (Ontario):Society of Management Accountants ofCanada, 1998.

Mankiw, N. G.; Romer, D.; Weil, D. N. A contributionto the empirics of economic growth QuarterlyJournal of Economics, 1992, Vol. 107, No 2,p. 407-437.

Mason, G.; Keltner, B.; Wagner, K. Productivity,technology and skills in banking: commerciallending in Britain, the United States andGermany. London: National Institute ofEconomic and Social Research, 1999 (Discus-sion paper, No 159).

McGivency, V. Adult participation in learning: canwe change the pattern? In: Cotfield, F. (ed.) Anational strategy for lifelong learning.Newcastle: University of Newcastle, 1997.

McIntosh, S. A cross-country comparison of thedeterminants of vocational training. London:Centre for Economic Performance (LSE), 1999(Discussion paper, No 432).

McMahon, W. W. The impact of human capital onnon-market outcomes and feedbacks oneconomic development. Paper presented at theinternational symposium on the contribution ofhuman and social capital to sustainedeconomic growth and well-being, organised bythe OECD and HRDC, Quebec City, Canada,19-21 March 2000.

Mincer, J. Schooling, experience and earnings. NewYork: Columbia University Press, 1974.

Mingat, A.; Tan, J.-P. The full social returns toeducation: estimates based on countries’economic growth performance. Human CapitalDevelopment Papers. Washington: WorldBank, 1996.

Moretti, E. Estimating the external returns to educa-tion: evidence from repeated cross-sectionaland longitudinal data. Berkeley: University ofCalifornia, 1999 (Research paper).

Nadvi, K. International competitiveness and smallfirm clusters – evidence from Pakistan. SmallEnterprise Development, 1998, Vol. 9, No 1,p. 12-24.

Nonaka, I; Takeuchi, H. The knowledge creatingcompany: how Japanese companies create thedynamics of innovation. New York: OxfordUniversity Press, 1995.

O’Mahony, M.; de Boer, W. Britain’s relative produc-tivity performance: has anything changed?National Institute Economic Review, 2002,Vol. 179, No 1, p. 38-43.

OECD. Cities and regions in the new learningeconomy. Paris: OECD, 2001b.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 68

OECD. Education at a glance: OECD indicators2001. Paris: OECD, 2001a.

OECD. Human capital investment: an internationalcomparison. Paris: OECD/CERI, 1998.

OECD. Human capital investment: an internationalcomparison. Paris: OECD/CERI, 1999.

OECD. The well-being of nations: the role of humanand social capital. Geneva: OECD, 2001c.

Osberg, L.; Sharpe, A. Comparisons of trends inGDP and economic well-being: the impact ofsocial capital. Paper presented at the interna-tional symposium on the contribution of humanand social capital to sustained economic growthand well-being, organised by the OECD andHRDC, Quebec City, Canada, 19-21 March2000.

Papalexandris, N.; Nikandrou, I. Benchmarkingemployee skills: results from best practice firmsin Greece. Journal of European IndustrialTraining, 2000, Vol. 24, No 7, p. 391-402.

Penrose, E. The theory of the growth of the firm.London: Blackwell publishing, 1959.

Pfeffer, J. Competitive advantage through people.Boston: Harvard Business School Press, 1994.

Prahalad, C. K. Developing strategic capability: anagenda for top management. Human ResourceManagement, 1983, Vol. 22, p. 237-254.

Prais, S. J. Productivity, education and training: aninternational perspective. Cambridge Univer-sity Press, 1995.

Reich, R. B. The work of nations: preparingourselves for 21st-century capitalism. NewYork: Knopf, 1991.

Romer, P. M. Endogenous technological change.Journal of Political Economy, 1990, Vol. 98,No 5, p. S71-S102.

Romer, P. M. Increasing returns and long-rungrowth. Journal of Political Economy, 1986,Vol. 94, No 5, p. 1002-1037.

Romer, P. M. The origins of endogenous growth.Journal of Economic Perspectives, 1994, Vol. 8,No 1, p. 3-22.

Ruggles, R. The state of the notion: knowledgemanagement in practice. California Manage-ment Review, 1998, Vol. 40, No 3, p. 80-89.

Satchell, S. E.; Damant, D. C. Testing for shorttermism in the UK stock market: a comment.Economic Journal, 1995, Vol. 105,p. 1218-1223.

Sawyer, M. C. The economics of industries andfirms. London: Croom Helm, 1981.

Sianesi, B.; Van Reenen, J. The returns to educa-tion: a review of the macro-economic literature.London: CEE – Centre for the Economics ofEducation (LSE), 2000 (Discussion paper, DP 6).

Singh, A.; Whittington, G. Growth, profitability andvaluation. Cambridge: Cambridge UniversityPress, 1968.

Solow, R. Technical change and the aggregateproduction function. Review of Economics andStatistics, 1957, Vol. 39, p. 312-320.

Steedman, H. Five years of the modern apprentice-ship initiative: an assessment against conti-nental European models. National InstituteEconomic Review, 2001, Vol. 187, No 1,p. 75-87.

Steedman, H.; McIntosh, S. Measuring low skills inEurope: how useful is the ISCED framework?Oxford Economic Papers, 2001, Vol. 53, No 3,p. 564-581.

Steel, J.; Sausmann, C. The contribution of gradu-ates to the economy. London: Stationery Office,1997 (NCIHE report, No 7).

Stevens, M. Human capital theory and UK voca-tional training policy. Oxford Review ofEconomic Policy, 1999, Vol. 15, No 1, p. 16-32.

Storey, D. Preface. In: Barber, J. et al. (eds) Barriersto growth in small firms. London: Routledge,1989, p. viii-ix.

Summers, R.; Heston, A. The Penn worldtable-mark 5: an expanded set of internationalcomparisons, 1950-1988. Quarterly Journal ofEconomics, 1991, Vol. 106, No 2, p. 327-369.

Sylwester, K. Income inequality, education expen-ditures and growth. Journal of DevelopmentEconomics, 2000, Vol. 63, No 2, p. 379-398.

Temple, J. Growth effects of education and socialcapital in the OECD countries. Paris: OECD,2000 (ECO/WKP(2000)36).

Temple, J. Robustness tests of the augmentedSolow model. Journal of Applied Econometrics,1998, Vol. 13, No 4, p. 361-375.

Temple, J. The new growth evidence. Journal ofEconomic Literature, 1999, Vol. 37, No 1,p. 112-156.

Tessaring, M. Training for a changing society: areport on current vocational education andtraining research in Europe. Luxembourg:Office for Official Publications of the EuropeanCommunities, 1998 (Cedefop Referenceseries).


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 69

Timmerman, A. Can agents learn to form rationalexpectations? Some results on convergenceand stability of learning in the UK stock market.Economic Journal, 1994, Vol. 104, p. 777-797.

Toivanen, O.; Stoneman, P.; Bosworth, D. Innova-tion and the market value of UK firms, 1989-95.Coventry: Warwick Business School, 1998(Working paper).

Topel, R. Labour markets and economic growth. In:Ashenfelter, O.; Card, D. (eds) Handbook oflabor economics. Amsterdam: North Holland,1999, Vol. 3c (Handbooks in Economics series).

Trostel, P.; Walker, I.; Woolley, P. Estimates of theeconomic return to schooling for 28 countries.Labour Economics, 2002, Vol. 9, No 1, p. 1-16.

Turok, I. Urban labour markets: the causes andconsequences of change. Urban Studies, 1999,Vol. 36, No 5/6, p. 893-915.

Warf, B. Telecommunications and the changinggeographies of knowledge transmission in thelate 20th century. Urban Studies, 1995, Vol. 32,No 2, p. 361-378.

Weiner, N.; Mahoney, T. A. A model of corporateperformance as a function of environmental,organisational and leadership influences.Academy of Management Journal, 1981,Vol. 24, No 3, p. 453-470.

Westphalen, S. Reporting on human capital: objec-tives and trends. In: Descy, P.; Tessaring, M.(eds) Training in Europe. Second report onvocational training research in Europe 2000:background Report. Luxembourg: Office forOfficial Publications of the European Commu-

nities, 2001, Vol. II, Part 4, p. 249-278 (CedefopReference series, 3008).

Wilson, R. A. Changing pay relativities for the highlyqualified. Review of the Economy and Employ-ment 1985/86, 1986, Vol. 2, p. 30-48.

Wolfe, B.; Zuvekas, S. Non-market outcomes ofschooling. International Journal of EducationalResearch, 1997, Vol. 27, No 6, p. 491-502.

Wolter, S. C.; Weber, B. Skilling the unskilled: aquestion of incentives. In: Lange, T. (ed.) Skillingthe unskilled: achievements and underachieve-ments in education and training. Aberdeen:Robert Gordon University, 1998.

Wood, W. J. Who is running British manufacturing?Industrial Management and Data Systems,1992, Vol. 92, No 6.

Woolcock, M. The place of social capital in under-standing social and economic outcomes. Paperpresented at the international symposium onthe contribution of human and social capital tosustained economic growth and well-being,organised by the OECD and HRDC, QuebecCity, Canada, 19-21 March 2000.

Wright, P. M.; McMahan, G. C.; McWilliams, A.Human resources and sustained competitiveadvantage: a resource-based perspective.International Journal of Human ResourcesManagement, 1994, Vol. 5, p. 301-326.

Youndt, M. A. et al. Human resource management,manufacturing strategy and firm performance.Academy of Management Journal, 1996,Vol. 39, No 4, p. 836-865.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 70

AbstractRecent discussion of the knowledge-based economy draws increasingly attention to the role that thecreation and management of knowledge plays in economic development. Development of human capital,the principal mechanism for knowledge creation and management, becomes a central issue forpolicy-makers and practitioners at the regional, as well as national, level. Facing competition both withinand across nations, regional policy-makers view human capital development as a key to strengtheningthe positions of their economies in the global market. Against this background, the aim of this study is togo some way towards answering the question of whether, and how, investment in education and voca-tional training at regional level provides these territorial units with comparative advantages.The study reviews literature in economics and economic geography on economic growth (Chapter 2). Ingrowth model literature, human capital has gained increased recognition as a key production factor alongwith physical capital and labour. Although leaving technical progress as an exogenous factor, neoclas-sical Solow-Swan models have improved their estimates through the inclusion of human capital. Incontrast, endogenous growth models place investment in research at centre stage in accounting fortechnical progress. As a result, they often focus upon research workers, who embody high-order humancapital, as a key variable in their framework. An issue of discussion is how human capital facilitateseconomic growth: is it the level of its stock or its accumulation that influences the rate of growth? Inaddition, these economic models are criticised in economic geography literature for their failure toconsider spatial aspects of economic development, and particularly for their lack of attention to tacitknowledge and urban environments that facilitate the exchange of such knowledge.Our empirical analysis of European regions (Chapter 3) shows that investment by individuals in humancapital formation has distinct patterns. Those regions with a higher level of investment in tertiary educa-tion tend to have a larger concentration of information and communication technology (ICT) sectors(including provision of ICT services and manufacture of ICT devices and equipment) and research func-tions. Not surprisingly, regions with major metropolitan areas where higher education institutions arelocated show a high enrolment rate for tertiary education, suggesting a possible link to the demand fromhigh-order corporate functions located there. Furthermore, the rate of human capital development (at thelevel of vocational type of upper secondary education) appears to have significant association with thelevel of entrepreneurship in emerging industries such as ICT-related services and ICT manufacturing,whereas such association is not found with traditional manufacturing industries.In general, a high level of investment by individuals in tertiary education is found in those regions thataccommodate high-tech industries and high-order corporate functions such as research and develop-ment (R&D). These functions are supported through the urban infrastructure and public science base,facilitating exchange of tacit knowledge. They also enjoy a low unemployment rate.

Empirical analysis of human capitaldevelopment and economic growth

in European regionsHiro Izushi, Robert Huggins

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 71


However, the existing stock of human and physical capital in those regions with a high level of urbaninfrastructure does not lead to a high rate of economic growth. Our empirical analysis demonstrates thatthe rate of economic growth is determined by the accumulation of human and physical capital, not bylevel of their existing stocks. We found no significant effects of scale that would favour those regions witha larger stock of human capital.The primary policy implication of our study is that, in order to facilitate economic growth, education andtraining need to supply human capital at a faster pace than simply replenishing it as it disappears fromthe labour market. Given the significant impact of high-order human capital (such as business R&D staffin our case study) as well as the increasingly fast pace of technological change that makes human capitalobsolete, a concerted effort needs to be made to facilitate its continuous development.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 72

Table of contents

1. Introduction 75

1.1. Human capital theory 75

1.2. Regions, states and skills: structure of the report 75

2. Models of economic growth and human capital 77

2.1. Early neoclassical perspectives 77

2.2. Limitations of the Solow-Swan model 78

2.3. Frankel-Romer model: AK approach to endogenous growth 79

2.4. The second Romer model 80

2.5. Schumpeterian growth model 81

2.6. Accumulation or stock? Source of economic growth due to human capital 83

2.7. Regional perspectives 85

2.8. Summary of literature review 88

3. Empirical analysis of regions in Europe 90

3.1. Relationship between investment in education/training and demand for human capital 90

3.1.1. Human capital development and employment patterns 90

3.1.2. Human capital development and public science base 92

3.1.3. Urban/rural settings and human capital development 92

3.1.4. Relationship between human capital development and entrepreneurship 92

3.1.5. Human capital development and unemployment rate 95

3.1.6. Human capital development and economic growth 101

3.2. Formal analysis of the relationship between human capital and economic growth at regional

and national level 101

3.2.1. Estimates at the augmented Solow-Swan model: accumulation of human capital 101

3.2.2. Estimates at the Nelson-Phelps framework: level of human capital stock 104

3.2.3. Human capital and growth: cross-country estimates 106

4. Summary and conclusions 110

4.1. Significance of economic growth models 110

4.2. The regional dimension 110

4.3. Policy implications 111


Annex 1: Technical difficulty of incorporating increasing returns to a neoclassical 114

model

Annex 2: Growth rate of consumption in Romer (1986) model 115

References 116

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 73


Tables

Table 1: Estimation of the augmented Solow-Swan model with the assumption of a steady state 102

Table 2: Estimation of the augmented Solow-Swan model 103

Table 3: Cross-regional growth accounting results: Human capital in log levels 105

Table 4: Cross-regional growth accounting results: Human capital in levels and in product of levels

and differences from the leading-edge region in labour productivity 107

Table 5: Cross-country growth accounting results, 1992-2000 108

Figures

Figure 1: A schematic representation of economic activities in the multisector model

(Aghion and Howitt, 1998, p. 86) 82

Figure 2: Interaction between the local education and training system and the locality’s stock

of highly-skilled workers (adopted from Bradley and Taylor, 1996, p. 3) 87

Figure 3: Schematic model of knowledge and production 88

Figure 4: Students in tertiary education as a percentage of working age population

and employment in ICT sectors 91

Figure 5: Students in tertiary education as a percentage of working age population and R&D staff 91

Figure 6: Students in tertiary education as a percentage of population aged 20 to 24 years

and public R&D expenditures 93

Figure 7: Students in tertiary education as a percentage of working age population and population density 93

Figure 8: Students in upper secondary education as a percentage of working age population

and population density 94

Figure 9: Change in students in tertiary education as a percentage of population aged 16 to 19 years

and population density 94

Figure 10: Students in general type of upper secondary education as a percentage

of population aged 16 to 19 years and average firm size in ICT-related services 96

Figure 11: Students in vocational type of upper secondary education as a percentage

of population aged 16 to 19 years and average firm size in ICT-related services 96

Figure 12: Students in general type of upper secondary education as a percentage

of population aged 16 to 19 years and average firm size in ICT manufacturing 97

Figure 13: Students in vocational type of upper secondary education as a percentage

of population aged 16 to 19 years and average firm size in ICT manufacturing 97

Figure 14: Students in tertiary education as a percentage of working age population

and unemployment rate 98


and unemployment rate 98

Figure 16: Unemployment rate and business R&D staff 99

Figure 17: Per worker GDP growth rate and students in tertiary education as a percentage

of working age population 99

Figure 18: Per worker GDP growth rate and students in general type

of upper secondary education as a percentage of working age population 100

Figure 19: Per worker GDP growth rate and students in vocational type

of upper secondary education as a percentage of working age population 100

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 74

1.1. Human capital theory

As the global economy shifts towards moreknowledge-based sectors (e.g. the manufactureof ICT devices, pharmaceuticals, telecommunica-tions and other ICT-based services, R&D), skillsand human capital development becomes acentral issue for policy-makers and practitionersengaged in economic development both at thenational and regional level (OECD, 1996). Yet, theimpact education and vocational training activi-ties exert upon changing national and regionaleconomies remains less than thoroughlyexplained and analysed. Since the introduction ofhuman capital theory in the 1960s, a number ofstudies have attempted to address this andrelated issues.

Human capital theory views schooling andtraining as an investment in skills and compe-tences (Schultz, 1960 and 1961) (Becker, 1964). Itis argued that, based on rational expectations ofreturns on investment, individuals make decisionson the education and training they receive as away of augmenting their productivity. A similarstrand of studies focuses on the interactionbetween the educational/skill levels of the work-force and measurements of technological activity(Nelson and Phelps, 1966). According to thistheory, a more educated/skilled workforce makesit easier for a firm to adopt and implement newtechnologies, thus reinforcing returns on educa-tion and training. Empirical studies provideevidence supporting the aggregate effects ofeducation and training. For example, Griliches(1970) estimated that one third of the Solow(1957) residual (i.e. the portion of the outputgrowth in the US economy that could not beattributed to the growth in labour hours or capitalstock) could be accounted for by the increase inthe labour force’s educational attainments. In thesame vein, Denison (1979) reported the effectupon per capita income in the US, while others –including Baumol et al. (1989), Barro (1991) andMankiw et al. (1992) – have confirmed these posi-tive relationships through a cross-section ofcountries covering all levels of development.

Bartel and Lichtenberg (1987) and Wolff (1996,2001) found that education/skill levels are posi-tively related with technological change in thesectors concerned. Also Crouch et al. (1999)provide a degree of evidence that highlyeducated workers are more likely to be employedin sectors exposed to international competition ina recent period, suggesting close associationbetween education/skill levels of workers andtechnological activity undertaken. Looking atthese impacts upon society as a whole,Abramovitz (1986) argues that education andvocational training is part of a set of key ingredi-ents sustaining society’s growth, which he terms‘social capability’.

1.2. Regions, states and skills:structure of the report

When these impacts of education and vocationaltraining are disaggregated, and their distributionsamong different segments of the society areconsidered, their effects are mixed and hence effec-tive modes for policy intervention are notadequately developed. Unequal effects of skills andhuman capital development are very noticeableamong regions, as well as among nations. A rangeof literature provides empirical evidence concerninginter-regional inequality in labour productivity. Forexample, Dunford (1997) demonstrates persistentdisparities among the UK regions in productivityand economic participation. A recent study we haveundertaken (Huggins and Izushi, 2002) confirms asignificant variation in productivity among advancedregional economies around the globe. This is asource of great concern, since many regions areincreasing their independence as political andeconomic units, and are competing against oneanother. At the same time, powerful multinationalcorporations are becoming evermore footloose,increasingly escaping from the controls of nationstates (Ohmae, 1995).

Hence an important question is whether theinfrastructure and provision of education and

1. Introduction

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 75

vocational training at the regional level providesthese territorial units with comparative advan-tages. In order to answer this question satisfacto-rily, it is necessary to understand the manner inwhich skills and human capital developmentexerts its influence on regional and nationaleconomies, as well as the links between the twogeographic layers. The structure of the report isas follows.

Chapter 2 provides an overview of literature oneconomic growth and human capital. As humancapital development and technological progressare inextricably linked, much attention is paid tothe impact of technological progress oneconomic growth. Chapter 2 starts with a reviewof key models of economic growth. This coversan early neoclassical model employed by Solowand Swan, subsequent attempts to account forthe contribution of technological progress toproductivity growth, an endogenous growthmodel developed by Romer, and a Schumpete-rian approach that places innovation at the centreof economic growth. It then looks into an issueraised by Benhabib and Spiegel (1994): howhuman capital affects economic growth. This is

followed by a review of studies that applynational models to regional economies. Thechapter also reviews other models that falloutside the neoclassical framework.

Chapter 3 presents the results of our empiricalanalysis of regions in Europe. The first part looksinto relationships between the supply anddemand of human capital, and between humancapital supply and other economic indicators atregional level. The results show that supply anddemand of human capital development havedistinct patterns relating to educational level aswell as types of regions. The second part ofChapter 3 focuses on the question of whichaffects the economic growth of regions, the rateof accumulation of human capital or the level ofstock of human capital. The empirical analysisdemonstrates that the rate of accumulation ofhuman capital has a clear impact upon economicgrowth at the regional level, while we find nosignificant impact from the level of human capitalstock. This finding provides evidence that refutesBenhabib and Spiegel’s thesis (1994).

Finally, Chapter 4 summarises key findings ofthe study and provides policy recommendations.


01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 76

2.1. Early neoclassicalperspectives

Although economists have studied economicgrowth for generations, there is still disagreementabout how it is accounted for in a formal model.While some researchers take a Keynesian route andstress the role of demand factors, other researchersfollow the neoclassical route, or more recently aSchumpeterian approach, emphasising the role offactor supplies in growth. In general, supply-sidemodels are designed to uncover production factorsfor economic growth and are hence consideredmore appropriate for the purpose of investigatingthe impact of human capital formation.

Neoclassical growth theory seeks to under-stand the determinant of long-term economicgrowth rate through accumulation of factor inputssuch as physical capital and labour. Studiesreveal a significant contribution from technicalprogress, which is defined as an exogenousfactor. Solow (1957) and Swan (1956) are amongthose who first demonstrated this.

At the heart of the neoclassical model lies anaggregate production function exhibitingconstant returns to scale in labour and repro-ducible capital. This can be written in generalform as follows:

Y = F (K, L)

where Y is output (or income), K is the stock ofcapital, and L is the labour force. The functionexpresses the output Y under a given state ofknowledge, with a given range of available tech-

niques, and a given array of different capital,intermediate goods and consumption goods.With constant returns to scale, output per worker(i.e. labour productivity) y ≡ Y / L will depend onthe capital stock per worker (i.e. capital intensity)k ≡ K / L. Under the assumption of constantreturns to scale, the relationship each unit oflabour has with capital in production does notchange with the quantity of capital or labour inthe economy.

A crucial property of the aggregate productionfunction is that there are diminishing returns onthe accumulation of capital. In other words, eachadditional unit of capital used by a workerproduces a decreasing amount of output (2). Aform called the Cobb-Douglas function usuallyexpresses the relationship:

.

Alternatively the per worker production functioncan be written as:

.

In other words, labour productivity can increaseonly if there is capital deepening (i.e. if capitalintensity increases) (3).

The crucial tenet of the neoclassical model is that,under decreasing returns on capital, output perworker does not increase indefinitely. Assuming:(a) people save a constant fraction s of their

gross income y (4);(b) the constant fraction δ of the capital stock

disappears each year as a result of deprecia-tion;

αkkfy == )(

10,1 <<= − ααα KLY

2. Models of economic growth and human capital (1)

(1) This literature review section owes much to Aghion and Howitt (1998), Armstrong and Taylor (2000), and Richardson (1979) aswell as Harris (2001) and Romer (1986 and 1990).

(2) This is expressed in mathematical terms, F’(K) > 0 and F’’(K) < 0 for all K.

(3) The idea is also expressed in mathematical terms as follows: .

(4) While the assumption of a fixed saving rate is not a bad approximation to long-term data, many argue that people save at arate that varies over their life. The permanent-income and lifecycle hypothesis presumes that people save with a view tosmoothing their consumption over their lifetimes, taking into account their preferences for consumption at different dates andthe rate of return that they can anticipate if they sacrifice current consumption in order to save for the future (Aghion andHowitt, 1998, p. 17-18). A model based on this assumption is the Cass-Koopmans-Ramsey model.

kLKLYydtdk

dtdL

dtdK

dtdL

dtdY

dtdy

αα =−

=−=

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 77

(c) the rate of population growth is n, and popu-lation growth will cause the capital stock perworker k to fall at the annual rate nk;

then the net rate of increase in k can be writtenby the following equation as:

.

While the decline in the capital stock per workerdue to depreciation and population growth isproportional to the capital stock, the growth ofper worker capital through saving is constrainedby decreasing returns on capital in production.When the marginal product of capital per workerfalls to a sufficiently low level, gross investmentwill be just sufficient to maintain the existingstock of capital. Hence, the capital stock perworker will, in the long term, converge asymptot-ically to k* that is defined by:

.

In this steady-state equilibrium, output and thecapital stock will both continue to grow, but onlyat the rate of population growth.

The model’s implication does not account forempirical evidence of long-term growth. Usingthis framework, Solow (1957) demonstrated thatan attempt to account for decades of USeconomic growth produced an astonishingresidual of approximately 85 %. Solow attributedmost of the residual to technological change (5).Accordingly, we can modify the neoclassicalmodel by supposing that there is a productivity(or technology) parameter A in the aggregatefunction that reflects the current state of techno-logical knowledge.

Y = F (A, K, L).

Assuming that productivity increases smoothlyover time at a constant growth rate g (6),

.

From this, it follows that growth in income isdetermined by productivity growth g and thegrowth of capital per worker (7). Hence, even ifthe capital stock and the labour force grow at thesame rate, output per worker will increaseprovided that the rate of technical progress ishigher than zero.

2.2. Limitations of theSolow-Swan model

An obvious limitation of the Solow-Swan model isits failure in accounting for the causes of techno-logical progress. Although the model shows thattechnological progress contributes to economicgrowth, it does not spell out why technologicalprogress takes place. The rate of technologicalprogress is set at g without any theoretical rela-tionships with other variables within the model(i.e. the rate is set exogenously). The justificationnormally given is that technological change origi-nates from knowledge produced by the publicscience base (e.g. universities, public researchinstitutes) outside the domain of the economicsystem the model expresses (Solow, 1957) (Shell,1966 and 1967).

However, there is every reason to believe thattechnological progress itself depends oneconomic decisions, to much the same degree ascapital accumulation. Entrepreneurs look forways to make a profit and one way of doing thisis to produce new ideas. Since there is a profitincentive to produce new knowledge and to inno-vate, knowledge creation and innovation need tobe incorporated into a model of economic growthin such a way that, while they spur economicgrowth, they are in turn further advanced byeconomic growth. In other words, technologicalprogress needs to be endogenised.

Another issue of the Solow-Swan model is itsassumption of constant returns to scale. There issome evidence that suggests increasing returns in

αα −= 10 LKeAY tg

0)( * =+− knsk δα

( ) knskknksfdtdk )()( δδ α +−=+−=


(5) It is worth noting that neither capital stock K nor labour force L included human capital in Solow’s calculation. The modelassumes that labour is homogeneous.

(6) Namely, .

(7) In mathematical terms, this is expressed as, .kgLKgLYydtdk

dtdL

dtdK

dtdL

dtdY

dtdy

αα +=−

+=−=

tgeAA 0=

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 78

long-term economic growth. For example,Kendrick (1976) attempted to explain US economicgrowth by adding intangible investments, such ashuman capital (e.g. R&D and education andtraining), to the capital stock that normally consistsof tangible components (i.e. physical capital andlabour). Such intangible investments can becounted as capital stock because they must have alifetime of more than one year, that is, they improvethe quality of the tangible factor over two or moreannual accounting periods. However, he found that,between 1929 and 1969, an annual growth rate inreal total capital (2.4 %) represented only 70 % ofthe 3.4 % average annual growth of real product inthe private domestic business economy (Kendrick,1976, p. 131). Romer (1986, p. 1013) suggests that,given the repeated failure of this kind of growthaccounting exercise, there is no basis in the data forexcluding the possibility that aggregate productionfunctions are best described as exhibitingincreasing returns.

The idea that increasing returns are central to theexplanation of long-term growth is at least as old asAdam Smith’s story of the pin factory (Smith, 1776).Alfred Marshall (1890) introduced the concept ofincreasing returns that are external to a firm butinternal to an industry. Allyn Young furthered theidea with his competitive equilibrium interpretation,though no formal dynamic model embodying thatinsight was developed. Kenneth Arrow (1962)assumed that the productivity of a given firm is anincreasing function of cumulative aggregate invest-ment for the industry. Avoiding issues of specialisa-tion and divisions of labour, Arrow argued thatincreasing returns arise because new knowledge isdiscovered as investment and production takeplace (Romer, 1986, p. 1005).

The failure of neoclassical models to introducetechnological progress in such a way to account forits causes (i.e. endogenise technological progress)is, in large part, due to technical difficulty dealingwith increasing returns in a dynamic general equi-librium framework. Attempts to understand

increasing returns have sought their source in tech-nological progress. However, the approach entailstechnical difficulty if it is to maintain the Walrasianframework of marginal product (8).

Arrow (1962) avoided the problem by assumingthat the growth of productivity A is an unintendedconsequence of the experience of producing newcapital goods, a phenomenon he called ‘learningby doing’. He assumed that an increase in Knecessarily leads to an equiproportionate increasein knowledge through ‘learning by doing’. In hismodel, K and L are paid their marginal productsas those firms that produce capital goods are notcompensated for their learning by doing(i.e. contribution to a growth in A). Yet, the growthof A became endogenous in the sense that itwould increase saving propensity, which would inturn affect output up to an equilibrium point. TheArrow model is fully operational in the case of afixed capital/labour ratio. This implies that themodel does not have enough increasing returns tosustain output growth in the long term withoutgrowth in labour, as in the Solow-Swan model(Aghion and Howitt, 1998, p. 23).

2.3. Frankel-Romer model: AKapproach to endogenousgrowth

More recent attempts to endogenise technolog-ical progress were spurred by Paul Romer’s twoseminal papers (1986 and 1990). Of these, thefirst 1986 paper has its theoretical origin inFrankel’s (1962) AK model (9). Frankel assumedthat each firm j in the economy has a productionfunction expressed as:

where Kj and Lj are the firm’s own employment ofcapital and labour. He then extended this produc-

αα −= 1jjj LKAY

Empirical analysis of human capital development and economic growth in European regions 79

(8) Aghion and Howitt (1998) put this as follows:... if A is to be endogenized, then the decisions that make A grow must be rewarded, just as K and L must be rewarded. Butbecause F exhibits constant returns in K and L when A is held constant, it must exhibit increasing returns in three ‘factors’ K,L, and A. Euler’s theorem tells us that with increasing returns not all factors can be paid their marginal products. Thus some-thing other than the usual Walrasian theory of competitive equilibrium, in which all factors are paid their marginal products,must be found to underlie the neoclassical model (p. 23).See Annex 1 for technical explanation of this.

(9) Aghion and Howitt, 1998, p. 25.Apparently Romer himself did not realise the theoretical lineage since he did not cite Frankel’s work in his 1986 paper.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 79

tion function to the whole economy, assumingthat all firms face the same technology and thesame factor prices, and will hire factors in thesame proportions, which obtains:

. (1)

To endogenise the productivity parameter A,Frankel assumed that it is a function of the overallcapital/labour ratio:

because in many respects the stock of knowl-edge depends on the amount of capital perworker in the economy. This is based on the ideathat technological knowledge is itself a kind ofdisembodied capital good (10).

Another assumption made in Frankel’s modelis that although A is endogenous to the economy(i.e. related to changes in K and L), it was takenas given by each firm, because the firm wouldonly internalise a negligible amount of the effectthat its own investment decisions have on theaggregate stock of capital.

When α + β = 1, equation (1) becomes Y = AK.This form of model is referred to as the AK model.Diminishing returns on the accumulation ofcapital play a crucial role in limiting growth inneoclassical models like the Solow-Swan model.However, in the Frankel model, output grows inproportion to capital because of the effect ofknowledge creation activities that counteractdiminishing returns.

In his 1986 paper, Romer in effect extendedthe Frankel model by introducing a lifetime utility

function , where c(t) is the

time path of consumption per person, u(.) is aninstantaneous utility function exhibiting positivebut diminishing marginal utility, and ρ is a positiverate of time preference. Romer assumed aproduction function with externalities of the samesort as considered by Frankel, and examined thecase in which labour supply per firm was equal to

unity (i.e. L=1) and the rate of depreciation δ waszero. If it is supposed that the productivityparameter A reflects the total stock of accumu-lated capital NK where N is the number of firms,

.In a steady-state growth, consumption (11) and

output grow at the same rate g, which isexpressed as:

if α + β. This indicates that the larger the numberof firms N, the more externalities there will be inproducing new technological knowledge andtherefore the faster the representative firm andthe economy will grow (12).

As shown above, the AK approach introducesa specific relationship between technologicalprogress and capital accumulation by assumingthat knowledge is a sort of capital good andproductivity increases with capital per labour.However, accumulation of knowledge is stillexternal in the relationship since the approachdoes not explicitly express how knowledgecreation is remunerated.

2.4. The second Romer model

Romer takes a different approach to accounting fortechnological progress in his article published in1990. While he saw knowledge as part of the aggre-gate capital K and related technological progress toan increase in capital/labour ratio in his 1986 article,Romer focused this time on the production ofknowledge by research workers. This modelassumes that technological knowledge islabour-augmented, enhancing their productivity.The production function is expressed as:

(2)

so that AL denotes a knowledge-adjusted work-force. Further, the model assumes that research

( ) αα −= 1ALKY

εραα −= − ANg

1

β)(NKAA =

( )dttcueW t )(0∫ ∞ −= ρ

( )βLKAA /=

αα −= 1LKAY


(10) Aghion and Howitt (1998) explains this:It [technological knowledge] can be used in combination with other factors of production to produce final output, it can bestored over time because it does not get completely used up wherever it is put into a production process, and it can be accu-mulated through R&D and other knowledge-creation activities, a process that involves the sacrifice of current resources inexchange for future benefits (p. 25-26).

(11) The growth rate of consumption under the model is given in Annex 2.(12) In his 1986 article, Romer in fact assumed α + β > 1, that is, increasing social returns on capital.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 80

workers create technological knowledge. In asimple form, this is expressed as:

(3)

where HA is human capital of research workers,and δ is a parameter. It is plain to see that themore researchers, the more new ideas arecreated, and the larger the existing stock ofknowledge A, the more new ideas are produced(i.e. effect of externalities).

Equation (3) shows that the rate of technicalprogress will be determined by the stock of humancapital of research workers. In other words, aneconomy with a larger total stock of human capitalwill grow faster (Romer, 1990, p. S99).

It is worth emphasising that unlike his previousmodel, the second Romer model explicitly recog-nises the role human capital plays in economicgrowth. Also the model differs from humancapital models such as the one developed byBecker et al. (1990) that treats all forms of intan-gible knowledge as being analogous to humancapital skills that are rival and excludable. Thesecond Romer model includes two distinct waysin which knowledge enters production. One is thecontribution of new ideas (or designs in Romer’sterm) to producing new goods. Research workersemployed by firms undertake the production ofnew designs. New designs are nonrival butexcludable as their property rights are protectedby patents. At the same time, new designs alsoincrease the total stock of knowledge shared bythe community of research workers and therebyincrease the productivity of human capital in theresearch sector as a whole. Knowledge spilloversimply externalities: knowledge is thus nonexclud-able in this realm (Romer, 1990, p. S84).

2.5. Schumpeterian growth model

Introducing the rival property of knowledgeprotected by property rights, the second Romermodel adopts a Schumpeterian view of innova-tion and explicitly assumes market power. Theidea was furthered in the 1990s by those modelsthat assumed imperfect competition and elabo-rated more on the process of innovation. Amongthose early attempts was that of Segerstromet al. (1990), who modelled sustained growth asarising from a succession of product improve-ments in a fixed number of sectors. However,Segerstrom et al. did not integrate the uncertainnature of innovation in their model. The introduc-tion of uncertainty had to wait for the modelproposed by Aghion and Howitt (1992). Aghionand Howitt assumed the creation of innovationsthrough research as a stochastic process inwhich the innovation quantity is expressed asflow probability. As a specific form of thestochastic process, a Poisson process isnormally adopted (13).

Aghion and Howitt (1998) extended the modelto include more than one economic sector and toconsider technology spillovers across sectors(Figure 1). In the model, there is one final goodthat is produced from a continuum of interme-diate goods. Each intermediate good can beused to produce the final good independently ofthe other intermediate goods, with no comple-mentarities between them. Each intermediatesector is monopolised by the holder of a patentto the latest generation of that intermediate good.Also each intermediate sector has its ownresearch sector in which firms compete todiscover the next generation of that particulargood. Innovations in research sectors all draw onthe same pool of shared technological knowledgethat exist beyond sectoral boundaries. The stateof this knowledge is represented by leading-edge

AHdtdA

Aδ=


(13) Suppose that events of a particular kind occur at random during a particular time. Poisson process has the probability distri-bution that meets the following four conditions:(a) The probability that each event occurs in a very short time interval must be proportional to the length of this time interval.(b) The probability that two or more events of the relevant kind occur in a very short time interval must be so small that it can

be regarded as zero.(c) The probability that a particular number of these events occurs in a particular time interval must not depend on when this

time interval begins.(d) The probability that a particular number of these events occurs in a particular time interval must not depend on the number

of these events that occurred prior to the beginning of this time interval.(Mansfield, 1980).

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 81

technology. Each innovation at date t in anysector adds an increment to the level of theleading-edge technology at date t – 1 andpermits the innovator to start producing in hissector using the new level of the leading edgetechnology. The previous incumbent in sector i,whose technology is no longer leading-edge, willbe displaced. Hence the leading-edge technologygrows gradually, at a rate that depends on theaggregate flow of innovations in the economy asa whole.

Aghion and Howitt (1998) also incorporated intheir model horizontal imitation as a source torestrict effects of increasing returns to scale.While the neoclassical theory of Solow and Swanassumes constant returns to scale, R&D modelsof growth no longer have constant returns in allthe factors that are growing: capital, knowledgeand labour. Growth models proposed by Romer(1990), Grossman and Helpman (1991), andAghion and Howitt (1992), for example, predictthat the steady-state growth rate depends on thelevel of resources devoted to R&D – if the level ofR&D resources is doubled, then per capitagrowth in output should also double. Jones

(1995) criticises this, showing the dramaticincrease of scientists and engineers in the USduring the last 40 years contradicting a constantmean of the growth rate of the economy over thesame period. To counter this, Aghion and Howittargue that a source that limits such scale effectsis imitations and a resultant growth of interme-diate goods in the economy without adding tooverall productivity.

The steady-state growth rate of per-workerincome g, which equals the growth rate ofleading-edge technology, is expressed as:

where σ is the size of an average increment ofknowledge that is added to the level ofleading-edge technology at each innovation, λ isthe productivity of R&D, n is the amount of inputin research which is adjusted by the level of theleading-edge technology (14), and φ(·) is a func-tion of the probability with which innovations takeplace (15). In other words, the steady-state growthrate g depends positively upon the productivity ofR&D (λ). Also, the flow probability of innovations

)(ng φλσ=


(14) It is presumed that as technology advances, the resource cost of further advances increases proportionally.(15) The function of innovation probability φ(·) has the property that represents a decreasing marginal product of research input n.

The function’s property is due to research congestion within a product.

Figure 1: A schematic representation of economic activities in the multisector model (Aghion and

Howitt, 1998, p. 86)

Manufacturing labour[sector-specific]

Research[sector-specific]

Public good[used in allsectors]

Innovations[sector-specific]

Technology spillover[from innovations inall sectors]

Intermediategoods

Final output

Knowledge

Labour

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 82

depends positively on technology-adjusted inputin research (n). Given the same level ofleading-edge technology, the growth ratedepends positively upon research input. Further,as an effect of horizontal imitation, Aghion andHowitt argue that the steady-state growth rate ofper-worker income also depends positively onpopulation growth (16).

2.6. Accumulation or stock?Source of economic growthdue to human capital

The above review shows that models ofeconomic growth vary in the ways they predictproduction factors can cause an economy togrow. According to the Solow-Swan model, thegrowth of per capita income arises from accumu-lation of capital until the economy reaches asteady state. In the steady state, per capitaincome growth relies solely on technologicalprogress that the model does not attempt toexplain. In contrast, endogenous growth modelsset R&D at the centre of their framework. Theypredict that per capita income growth is deter-mined by the amount of resources devoted toR&D. The neoclassical Solow-Swan model seesthe change in the amount of capital (i.e. capitalaccumulation) as the source of economic growth(until the economy reaches a steady state),whereas endogenous growth models assumethat the level of the stock of a particular capital(that is devoted to R&D) decides economicgrowth.

This disagreement about the source of economicgrowth is also found in discussions on humancapital. Broadly, there are two basic frameworkswith which to model and analyse the relationshipbetween human capital formation and economicgrowth (Benhabib and Spiegel, 1994) (Aghion andHowitt, 1998). The first approach has its origin inBecker’s (1964) theory of human capital and hasattracted attention with the 1988 article by Lucas. Itis based on the idea that growth is primarily drivenby the accumulation of human capital. According to

this approach, differences in growth rates of percapita income across economies are in large partaccounted for by differences in the rates at whichthe economies accumulate human capital. Thesecond approach dates back to the seminal paperof Nelson and Phelps (1966) and has recently beenrevived in Schumpeterian growth literature. Itcontends that the stock of human capital deter-mines the economy’s capacity to innovate or catchup with more advanced economies, which in turndrives economic growth. Hence, the level of humancapital stock is, though indirectly, a determinant ofper capita economic growth in this view.

In the economy assumed by Lucas (1988),individuals choose at each date how to allocatetheir time between current production and skillsacquisition (or schooling), taking into accountincreases in productivity and wages in futureperiods that arise from current investment of timein education or training. If h denotes the currenthuman capital stock of the representative person,and u denotes the fraction of the person’s timecurrently allocated to production, the Lucasmodel can be summarised by:

y = kβ (uh)1–β

where k denotes the per capita stock of physicalcapital, and:

, δ > 0.

While the second equation expresses that thegrowth rate of human capital is determined by timespent in education or training, the first equationdescribes the way human capital affects currentproduction. As the first equation’s similarity to theSolow-Swan model suggests, per capita incomegrowth comes from accumulation of human capital(as well as accumulation of physical capital). Inother words, the growth rate of per capita incomedepends positively on the growth rate of humancapital (as well as the growth rate of physicalcapital). Under the assumption of constant returnsto the stock of human capital, the steady-stategrowth is expressed by:

)1( ∗−= ug δ

)1( uhdtdh −= δ


(16) Aghion and Howitt (1998, p. 108) note: This new effect [of imitation] shows that what used to be thought of as an embarrassingscale effect in Schumpeterian growth theory can be seen … as a novel prediction that distinguishes it from the neoclassicaltheory of Solow and Swan. Instead of saying that growth goes up with the level of population, it goes up with the growth rateof population.

01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 83

where u* is the optimal allocation of individuals’time between production and education/training.

In contrast, Nelson and Phelps (1966)suggested that this standard view of humancapital as an additional input would represent agross misspecification of the production process.They argued that education and training facilitatethe adoption and implementation of new tech-nologies, which are continuously invented at anexogenous rate. In their view, the growth ofproductivity parameter A is expressed by:

where Tt denotes the level of theoretical knowl-edge at date t. It is evident in the specificationthat the growth rate of A depends on the gapbetween its level and the level of T, and the levelof human capital H through the function c(H)

where (17).

Extending the model, Benhabib and Spiegel(1994) substituted technology ‘catch-up’ acrossdifferent economies for the closing of a gapbetween A and T in the Nelson and Phelps frame-work. According to Benhabib and Spiegel, thegrowth rate of productivity parameter A for aneconomy i is written as:

(4)

where the endogenous growth rate g(Hi) and thecatch-up coefficient c(Hi) are non-decreasing func-tions of Hi. In other words, the level of human capitalnot only enhances the ability of an economy todevelop its own technological innovations (as in R&D-based growth models), but also its capacity to adaptand implement technologies developed elsewhere.

There is disagreement in empirical evidence as towhich influences economic growth – accumulationof human capital or level of human capital stock.

In a cross-country study of per capita GDPgrowth during two periods (from 1965 to 1975with 87 countries and from 1975 to 1985 with97 countries), Barro and Sala-i-Martín (1995)obtained the following findings:(a) educational attainment (measured by average

years of schooling) is significantly correlatedwith subsequent growth (with a correlationcoefficient at around 0.05), although if theaggregate measure of educational attainmentis decomposed by level of education, theimpact of primary education remains largelyinsignificant;

(b) public spending on education also has asignificantly positive effect on growth: a 1.5 %increase of the ratio of public educationspending to GDP during the period 1965-75would have raised the average growth rateduring the same period by .3 % per year (18).

Mankiw et al. (1992) also tested the impact ofhuman capital formation using the Solow-Swanmodel. In their test, they assumed a steadystate (19) and used a proxy for the rate of humanti

titjjii

ti

dtdA

AAA

HcHgAti −

⋅+=max)()(

0>dHdc

( )t

ttdtdA

AATHcA

−⋅=


(17) The growth rate of A settles down to the growth rate of T in the long-term.(18) Barro and Sala-i-Martín assume that a function for a country’s per capita growth rate in period t, Dy, as

Dyt = F(yt–1, ht–1; ...),where yt–1 is initial per capita GDP and ht–1 is initial human capital per person (based on measures of educational attainmentand health). The omitted variables, denoted by ..., comprise an array of control and environmental influences.

(19) Mankiw et al. use an augmented Solow-Swan model that is expressed as:

where H is the stock of human capital. When the fraction of income invested in physical capital is sk, and the fraction of incomeinvested in human capital is sh, the evolution of the economy is determined by:

where y = Y/AL, k = K/AL, h = H/AL, , n is growth rate of L, g is growth rate of A, and δ is depreciation rate. In the steady state,the following equation holds:

.( ) ( ) hktt ssgngtALY log1log1log1loglog 0 βαβ

βααδβα

βα−−+−−+++−−

+−+=

htht

ttkt

kgnysdtdh

kgnysdtdk

)(

)(

δ

δ

++−=

++−=

( ) βαβα −−= 1ttttt LAHKY

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 84

capital accumulation that measures approxi-mately the percentage of the working age popu-lation that is in secondary school. In the test thatexamined GDP per working age person in 1985for 98 non-oil countries, Mankiw et al. found thatthe coefficient on human capital accumulation issignificant, that is, human capital accumulation,along with physical capital accumulation,accounts for the growth of per capita GDP.

In contrast, Benhabib and Spiegel (1994) esti-mated the stock of human capital and tested theaugmented Solow-Swan model without theassumption of a steady state (20). From the dataof 78 countries during the period of 1965-85,Benhabib and Spiegel found that the log differ-ence in human capital in their specificationalways enters insignificantly, and almost alwayswith a negative coefficient. In other words, humancapital accumulation is found to lead to a nega-tive growth of the economy although this impactis statistically not significant.

Benhabib and Spiegel then undertook testsusing different models that included the stock ofhuman capital instead of the accumulation ofhuman capital. In the model that includes anaverage of human capital stock over the periodunder study, human capital stock enters insignifi-cantly with a negative sign (21). However, wheninitial income levels are introduced in the model,human capital stock enters significantly with thepredicted positive sign. Benhabib and Spiegelsuggest that catch-up remains a significantelement in growth, and countries with highereducation tend to close the technology gap fasterthan others. In the second model that incorpo-rates both endogenous growth and catch-upterms as in equation (4) (22), the catch-up termenters positively and significantly for the entiresample of 78 countries. However, the coefficientestimate on country-specific technological

progress is negative and insignificant. Benhabiband Spiegel tested the same model for subgroupsof their sample, assuming that the relatively strongimpact of the catch-up term may change with therelative position of the country. They found:(a) for the poorest third of their sample, the

catch-up term is positive and significant,whereas the endogenous growth term isnegative and insignificant;

(b) for the middle group, both terms are insignifi-cant;

(c) for the richest third of the sample, theendogenous growth term enters positivelyand significantly with a 6 % level of confi-dence while the catch-up term enters insignif-icantly with a coefficient estimate that is posi-tive but close to zero.

From these results, they argue that humancapital stocks in levels, rather than their growthrates, play a role in determining the growth of percapita income.

2.7. Regional perspectives

These models of economic growth are usuallydeveloped with respect to national-leveleconomic growth and treat nations as spacelessunits. The lack of attention to space attractedstrong criticism from those who study regionaleconomies. The determinants of growth overspace carry certain implications that are not easyto reconcile with the central principles of growthmodels, and particularly neoclassical models.Richardson (1979, p. 142) summarises suchneoclassical principles as:(a) reliance on the price mechanism as the

spatial allocator of resources;(b) emphasis on marginal adjustments, whereas

spatial functions are discontinuous and loca-


(20) They assumed a production function, , and used the following equation to examine the impact of humancapital accumulation:

.

(21) The model’s specification is as follows:

.

(22) The model’s structural specification is as follows:

.( ) ( ) ( ) ( )00max0 loglogloglogloglog LLKKYYmHHmgcYY TTiiiT −+−++−+=− βα

( ) ( )

+−+−+−=− ∑T tTTTT HTLLKKAAYY

00000 log1loglogloglogloglogloglog γβα

( ) ( ) ( )00000 loglogloglogloglogloglogloglog HHLLKKAAYY TTTTT −+−+−+−=− γβα

γβαttttt HLKAY =

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 85

tion changes usually mean inertia (i.e. nochange) or a long-distance jump;

(c) the assumption that growth can be construc-tively analysed with an aggregate productionfunction and a homogenous capital stock;

(d) the predilection for equilibrium solutions;(e) a greater facility with deterministic rather than

probabilistic solutions.A central weakness of neoclassical models lies

in the assumption that all factors of productionare completely mobile between regions within acountry. The weakness is particularly acute whenneoclassical models are employed to account forlong-term regional disparities in economic devel-opment. The assumption of mobile factors withina country predicts that any differences in thecapital/labour ratio, and thus labour productivity,between regions disappear in the long term ascapital and labour move to the regions that yieldthe highest returns (23).

As for short- or mid-term disparities, theassumption of nonexcludable knowledge posesanother problem. This is because technologicalknowledge is assumed to be perfectly mobilebetween regions and always available to allregions simultaneously. For instance, the partic-ular assumption limits the application of theRomer models to the world economy as a wholebecause technological progress diffuses acrossgeographical space so that even smalleconomies can benefit from it without having torely on knowledge created within their own fron-tiers (Armstrong and Taylor, 2000, p. 79) (24).However, innovations do not diffuse instanta-neously or at an even rate over the economy as awhole. They diffuse irregularly though predictably,reaching some areas very early in the adoptionstage but not being adopted in other areas untilvery late. In some cases (e.g. when a thresholdmarket is required), adoption at a particular loca-tion may never occur (Richardson, 1979,p. 125-126). The pioneering study in this area,Hägerstand (1966), focusing on agricultural inno-

vations, demonstrated the importance of thecommunications network as a determinant of thediffusion path (25). He also showed that the diffu-sion process could be understood by a model ofstochastic process (26).

Another aspect of nonexcludable knowledge isthat some types of knowledge are embodied inindividuals (i.e. tacit) and difficult to transferthrough other means than interpersonal, oftenface-to-face communications. This needs tobring another class into the second Romermodel: tacit knowledge embodied in individualsshould be distinguished from patent-protectedknowledge and shared, codified knowledge. Tacitknowledge also signifies the importance ofhuman capital that represents a carrier of suchknowledge. Some even argue about a region’sinstitutional environment as a key determinant ofits capacity to create technological progress(Rauch, 1993). According to this view, thecreation of technological progress is determinedby a collective learning process within whichmany individuals interact and exchange ideasand information (some of which are tacit). Thereare economies of scale to be gained from thegeographical concentration of highly educatedpeople as this results in a more rapid transfer ofknowledge through their proximity. In addition tothis, some regions are said to possess an institu-tional environment or culture that better facilitatessuch a collective learning process (Saxenian,1994). In addition to universities and researchinstitutes, a vertically-disintegrated industrialstructure, a high mobility of skilled workers, andan abundance of venture capital are often foundin such an environment.

Regarding mobility of human capital betweenregions, Bradley and Taylor (1996) argue thatthere is a sequential interaction between the localeducation and training system and the locality’sstock of highly-skilled workers. This is shown inFigure 2.

The rate of enrolment in education is influ-


(23) This weakness is mitigated when discussing disparities between nations because international mobility of labour is restrictedby immigration regulations.

(24) Based on the assumption, Romer argues that a country’s economic growth is correlated with the degree of its integration intoworldwide markets.

(25) The classic study of the diffusion of hybrid corn by Griliches (1957) looked at inter-state differentials but did not explicitly inves-tigate the spatial spread of the innovation (Richardson, 1979, p.125).

(26) Innovations diffuse over space and time. A common way of representing general spatial diffusion radiating out from the inno-vation source is expressed by a distance-decay function: p(r) = ae-br

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 86

enced by the socioeconomic background ofpupils, employment and career prospects in thelocal economy, and the quality of local schooling.Enrolment, in turn, determines the locality’s work-force skills, labour productivity, and economicperformance. Its economic performance thendetermines the volume and occupational mix ofinward migrant workers into the locality.Economic growth also provides employers withmore worker training, facilitating further skillsincrease. A shift in the occupational mix towardsskilled workers will have beneficial effects on thelocality’s human capital formation. While skilledworkers are often keen to invest in education fortheir children, an improved economy alsoprovides better employment opportunities andinduces other pupils to seek for education andtraining. Hence, Bradley and Taylor argue that theeducation and training system interacts with thelocal economy in such a way that spatial dispari-ties in economic well-being are exacerbatedthrough the cumulative causation mechanism.

A summary of our review of regional studies isgiven in Figure 3. While regional studies literature

shares a basic understanding of production witheconomics literature, it pays more attention to theway space affects economic production.

It particularly elaborates more on the effects ofspatial agglomeration of economic activity andattempts to identify causes of spatial agglomera-tion as well as its effects. Distinction betweencodified knowledge and tacit knowledge isemphasised in the attempt. It is argued that tacitknowledge is embodied in skilled workers andless mobile than codified knowledge. Hence, theassumption of frictionless diffusion of technolog-ical knowledge is under attack. Also drawing onmore institutional studies than mainstreameconomics, such as the theories of transactioncosts (Williamson, 1975 and 1985) and socialembeddedness (Granovetter, 1985), the literatureoften examines social aspects of relationsbetween economic agents. These efforts lead tothe conceptualisation of other types of produc-tion factors, such as social capital and networkcapital. Another area of focus is economic dispar-ities between regions that arise from spatial divi-sions of labour. It is often argued that high-order


Figure 2: Interaction between the local education and training system and the locality’s stock of

highly-skilled workers (adopted from Bradley and Taylor, 1996, p. 3)

Investmentin training

Post-schoolemploymentopportunities

Human capital formationthrough education

Haspirations, goals andmotivation of students

Workforce skills

Labour productivity

Inward migration ofhigh-skill workers

Growth of human capital stock

Growth of local economy

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 87

functions that require significant human capitalare concentrated in core regions (due to agglom-eration effects), creating regional disparities.Regional studies and economic studies of growthhence intersect at the issue of convergence.

2.8. Summary of literature review

The literature review shows that human capitalattracts more attention as economic growth

models attempt to account for technologicalprogress in greater detail. In the classicSolow-Swan model, technological progress wasidentified as a residual that is not explained bycapital and labour. Though capital in the modelcan theoretically include both human and phys-ical capital, human capital was, in practice, notconsidered in many empirical studies employingthe model. As a step towards accounting betterfor the role of technological progress in economicgrowth, the Fankel’s AK model related it to an


Figure 3: Schematic model of knowledge and production

Education and trainingLabour mobilityLearning and imitation

DIFFUSION OF

KNOWLEDGE

PRODUCTION OF

KNOWLEDGE

ResearchLearming by doing

PRODUCTION

Knowledge

General knowledge

Accumulated stockas public goods

Private knowledge

Appropriated bybusinesses

Social Environment

Disposition tosocial interactionHuman Capital

Embodyingknowledge througheducation & training

Network Capital

Knowlefge ofknow-who and

state of economicgovernance

LabourPhysical

Capital

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 88

increase in capital per worker, seeing knowledgeas a sort of disembodied capital. Romer (1986)refined the model by incorporating maximisationof lifetime utility with an intertemporal utility func-tion. Yet, both the original Frankel AK model andthe first Romer model did not give any explicitrole to human capital.

In contrast, Romer’s second endogenousgrowth model (1990) recognises human capital asa primary source of technological progress and,therefore, economic growth. Romer viewsresearch workers as the source of new ideas andhence profits. In the model, Romer also distin-guishes patent-protected technology from thestock of knowledge that is shared by the commu-nity of research workers. Other endogenousgrowth models, including the Aghion-Howittmodel, also set R&D at the centre of their frame-works. Such R&D-based growth models produceimplications that are distinct from the neoclas-sical Solow-Swan model. An example of this isprediction of scale effects.

There remains disagreement on how humancapital affects economic growth. While theapproach initiated by Lucas (1988) views accu-mulation of human capital as the source ofeconomic growth, the approach of Nelson and

Phelps (1966) and Benhabib and Spiegel (1994)assumes that stock of human capital determinesthe ability of an economy to develop and assimi-late technologies and thus produce economicgrowth. This difference in their positions mirrorsdifferent treatments of technological progress andR&D in the Solow-Swan model and R&D-basedendogenous growth models.

These economic growth models are, however,criticised for their spaceless analysis. Their weak-ness is said to be acute when accounting fordisparities in economic development betweenregions in which all production factors areassumed to be mobile. The assumption of fric-tionless diffusion of technological knowledge isunder particular attack. It is argued that tacitknowledge is embodied in skilled workers andless mobile than codified knowledge. Further-more, a mechanism of cumulative causation issaid to work in the location of highly-skilledworkers through education and training. Suchdiscussions in spatial studies imply increasingdisparities between regions. This implication is instark contrast to neoclassical models ofeconomic growth predicting convergence due todecreasing returns on capital.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 89

This chapter presents results of an empiricalanalysis of regions in EU Member States. The firstpart looks into relationships among aspects suchas investment in education/training, demand forhuman capital, and economic indicators ofregions. The second part provides empiricalevidence relating to the debate between theLucasian approach and the Benhabib-Spiegelapproach.

Unless otherwise noted, we use data providedby Eurostat that covers regions and countries inthe EU (Belgium, Denmark, Germany, Greece,Spain, France, Ireland, Italy, Luxembourg, theNetherlands, Austria, Portugal, Finland, Sweden,and the UK). We use the EU’s definition ofregional units, NUTS (Nomenclature of territorialunits for statistics) level 1 (27). Because of thedefinition, some nations are included as regions(i.e. Denmark, Ireland, Luxembourg). Regions inSweden, as well as regions in some parts ofPortugal and Finland, are based on NUTS level 2,a lower level of units.

Data availability restricts the majority of theanalysis in the second part to Denmark, Germany(excluding regions in ex-German DemocraticRepublic), France, Ireland, and Italy whereas thefirst part covers a greater proportion of theMember States. Also a majority of data, includingnumbers of students enrolled in education byeducational level, are available at the regionallevel only for short periods in the mid 1990s. Thisconstrained our analysis in the majority of the firstchapter.

3.1. Relationship betweeninvestment ineducation/training anddemand for human capital

3.1.1. Human capital development and

employment patterns

First, we looked at relationships between invest-ment by individuals in education/training andemployment size of sectors/functions that requirehigh-order human capital. We use the numbers ofstudents as a percentage of the working agepopulation (15 to 64 years old) as a proxy forinvestment by individuals in human capital devel-opment (28). We chose ICT-related sectors,including manufacturing of ICT devices and ICTservices, as requiring high order humancapital (29). The two variables are correlated andresults are as follows, quantified by enrolments:(a) for both general and vocational types of

upper secondary education, no significantcorrelation is found between students as apercentage of the working population andemployment in the high-tech sectors;

(b) for tertiary education, correlation betweenstudents as a percentage of the workingpopulation and employment in the high-techsectors is positive (0.41) and significant(0.005) – see Figure 4.

Similarly there is strong association, as shownin Figure 5, between students in tertiary educationas a percentage of working age population andvolume of R&D staff (including business, govern-ment, and higher education institutions). Correla-tion between the two is positive (0.54) and signifi-cant (0.000). We found close association (correla-tion: 0.31; significance: 0.01) between students intertiary education as a percentage of the working

3. Empirical analysis of regions in Europe

(27) The NUTS was set up at the beginning of the 1970s as a single, coherent system for dividing up the EU to produce regionalstatistics. Of its three levels of regions, level 1 is the largest unit that can be compared in size to some smaller member coun-tries of the EU.

(28) Mankiw et al. (1992) used this proxy in their study.(29) NACE 30 (office machinery and computers), 32 (telecommunications equipment), 64 (post and telecommunications), 72

(computer-related services) and 73 (R&D services) are included. In Eurostat, the data for NACE 64, 72 and 73 (and NACE 30and 32 as well) is collated and made available as a single group, not allowing NACE 73 to be separated from the rest.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 90


Figure 4: Students in tertiary education as a percentage of working age population and

employment in ICT sectors

Students in tertiary education as a percentage of working age population,

average of 1995-97

Employment

in ICT sectors

as a percentage

of total

employment,

average of

1996-99

0

2

4

6

8

10

0 2 4 6 8

Figure 5: Students in tertiary education as a percentage of working age population and R&D staff


average of 1995-97

R&D staff

(all sectors)

as a percentage

of total

employment,

average of

1995-97

0

1

2

3

4

0 2 4 6 8

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 91

population and number of business R&D staff aswell. In contrast, no significant association wasfound between number of R&D staff and studentsin both general and vocational types of uppersecondary education as a percentage of theworking population (not in Figure 5).

The results show the tendency that investmentby individuals in development of high-orderhuman capital (i.e. tertiary education) is strong inthose regions where there is a strong demandfrom activities requiring it (e.g. high-tech indus-tries, R&D departments).

3.1.2. Human capital development and public

science base

We saw close association between human capitaldevelopment and R&D capacity of private-sectorfirms above. The R&D capacity of regions is alsoinfluenced by public R&D effort from govern-ments and higher education institutions. Anexamination of the relationship is shown inFigure 6.

As expected, students in tertiary education asa percentage of the population aged 20 to24 years and public R&D expenditures have closeassociation, with a positive correlation (0.43) atthe 1 % significance level (0.000). In contrast,there is no significant correlation betweenstudents in upper secondary education as apercentage of the population aged 15 to 19 yearsand public R&D expenditure (30).

3.1.3. Urban/rural settings and human capital

development

One of the arguments found in regional studies isthat urban areas in which economic activityconcentrates tend to facilitate diffusion of knowl-edge (particularly tacit knowledge) through theease of face-to-face interactions. With agglomer-ation effects, such urban areas hence often hosthigh-order corporate functions (e.g. headquartersand R&D departments) and firms operating inhigh-tech industries as well as universities andresearch institutes (31).

Given this association between investment in

human capital development and high-tech indus-tries/R&D functions, close association is expectedbetween investment in human capital develop-ment and density of economic activity. We usepopulation density as a proxy for economicactivity density. Figure 7 shows the relationship.

It is clear that individuals in regions with largemetropolitan areas, such as Berlin, Bremen,Hamburg, Île de France, and Comunidad deMadrid, heavily invest in tertiary education. Corre-lation between the two variables is positive (0.27)and significant (0.02). Metropolitan areas hostinghigh-order corporate functions have a dispropor-tionate supply of universities which provide suit-able human capital.

In contrast, regions with large metropolitanareas have a relatively small number of studentsin upper secondary education as a percentage ofthe working population. This is shown inFigure 8 (32).

Whereas individuals in regions with largemetropolitan areas heavily invest in higher educa-tion, there is some evidence that gaps betweenregions are closing. Figure 9 shows the relation-ship between population density and change instudents in tertiary education as a percentage ofthe population aged 16 to 19 years from 1995 to1997. Correlation between the two variables isnegative (-0.23) and significant at the 10 % level(0.07). As the period is short, the change in enrol-ment rate is susceptible to short-term shocks.Accordingly, the finding is only suggestive. Thefigure indicates that an increase in enrolment ratein regions with large metropolitan areas is morelikely to be slower while the rate is growing fasterin some non-metropolitan regions (i.e. low popu-lation-density regions) (33).

3.1.4. Relationship between human capital

development and entrepreneurship

The creation of new businesses is an importantsource of economic dynamism. High-technologyindustries in particular evolve through tech-nology-based start-ups and spin-offs byentrepreneurs who are often supported by venture


(30) The same relationships are found between the numbers of students as a percentage of the working population by level ofenrolment and public R&D expenditure.

(31) Lucas (1988) discusses this, referring to the work of Jane Jacobs.(32) Correlation between the variables is negative (-0.20) at the 10 % significance level (0.09).(33) As might be expected, no significant correlation was found between population density and change in students in upper

secondary education as a percentage of the population aged 16 to 19 years.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 92


Figure 6: Students in tertiary education as a percentage of population aged 20 to 24 years and

public R&D expenditures

Students in tertiary education as a percentage of population aged 20-24 years,

average of 1995-97

R&D

expenditures by

government

and HEs as a

percentage of

GDP, average

of 1995-97

0.0

0.5

1.0

1.5

2.0

0 20 40 60 80


population density


average of 1995-97

Population

density,

average of

1995-97,

persons

per square

kilometer

0

1000

2000

3000

4000

0 2 4 6

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 93


Figure 8: Students in upper secondary education as a percentage of working age population and

population density

Students in upper secondary education as a percentage of working age population,

average of 1995-97

Population

density, average

of 1995-97,

persons per

square kilometer

0

1000

2000

3000

4000

4 6 8 10 12 14 16

Figure 9: Change in students in tertiary education as a percentage of population aged 16 to

19 years and population density

Change in students in tertiary education as a percentage of population aged 16-19 years

from 1995 to 1997

Population

density, average

of 1995-97,

persons per

square

kilometer

0

1000

2000

3000

4000

-2 0 2 4 6 8 10 12 14

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 94

capital. We examined the relationship betweenhuman capital development and entrepreneurship,using the average size of firms in selected high-techsectors as proxy for the rate of new business forma-tion.

We focused upon ICT-related industries astheir importance in the economy dramaticallyincreased in the 1990s along with the develop-ment of the Internet and related technologies.The industries are divided into ICT-relatedservices and ICT manufacturing (34).

For ICT-related services, we found no signifi-cant association between average firm size andstudents in tertiary education as a percentage ofthe population aged 20 to 24 years. However,general and vocational types of upper secondaryeducation show contrasting relationships withaverage firm size in the sectors. Students ingeneral types of upper secondary education as apercentage of the population aged 16 to 19 yearsshow a tendency to increase with the averagefirm size in ICT-related services (Figure 10). Theircorrelation (0.62) is significant at the 1 % level(0.001). In contrast, negative association is foundbetween students in vocational types of uppersecondary education as a percentage of thepopulation aged 16 to 19 years and the averagefirm size of ICT-related services (Figure 11). Theircorrelation (-0.82) is significant at the 1 % level(0.000) again.

These figures suggest that those regions with ahigh rate of new firm formation in ICT-relatedservices have a high enrolment rate in vocationaltypes of upper secondary education and a lowenrolment rate in general types of uppersecondary education.

We obtained the same findings for ICT manu-facturing, although their significance is weaker.

The correlation between the enrolment rate ofgeneral types of upper secondary education andthe average firm size in ICT manufacturing is 0.34and significant at the 10 % level (0.09). For voca-tional types of upper secondary education, corre-lation is –0.36 and significant at the 10 % level(0.07).

We also examined the same relationship withrespect to traditional manufacturing indus-tries (35). No significant association was found

between average firm size and enrolment rate ofany educational level examined.

In summary, human capital developmentappears to have significant association with therate of entrepreneurship in emerging industriessuch as ICT-related services and ICT manufac-turing, whereas such association is not found intraditional manufacturing industries. The associa-tion with entrepreneurship in ICT sectors is,however, found at the level of vocational types ofupper secondary education. The implications ofthis, along with the lack of significant associationwith higher education, are subject to interpreta-tion and not clear. Does the finding suggest thatgraduates from vocational types of uppersecondary education start new firms withoutobtaining higher education degrees? It is oftenassumed (in the US) that qualified engineers withhigher education degrees start new ICT-relatedbusinesses with support from venture capital. Theabove finding seems to contradict this assump-tion but clarification will require further research.


unemployment rate

Before moving on to the analysis of economicgrowth, we examine the relationship betweenhuman capital development and unemploymentrate. Figure 14 shows the relationship betweenstudents in tertiary education as a percentage ofthe working age population and unemploymentrate. The correlation is positive (0.19) but its levelof significance is slightly over 10 % (0.11). Signif-icant correlation is found between students inupper secondary education as a percentage ofthe working age population and unemploymentrate (Figure 15). The correlation is 0.35 and signif-icant at the 1 % level (0.003). In other words, indi-viduals in regions that have high unemploymentrate are more likely to invest in education andtraining. This fits with the general observation thatindividuals are more likely to invest in educationand training when facing unfavourable employ-ment situations at economic downturns.

Those regions enjoying low unemployment ratetend to have more R&D functions (Figure 16). Thecorrelation between unemployment rate andbusiness R&D staff as a percentage of total


(34) ICT-related services include NACE 64 (post and telecommunications), 72 (computer-related services) and 73 (R&D services).ICT manufacturing includes NACE 30 (office machinery and computers) and 32 (telecommunications equipment).

(35) The industries examined include general, electrical, and transport engineering.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 95


Figure 10: Students in general type of upper secondary education as a percentage of population

aged 16 to 19 years and average firm size in ICT-related services

Students in general type of upper secondary education as a percentage of population aged 16-19 years,

average of 1995-97

Average

firm size in

ICT-related

services,

1997

0

10

20

30

40

0 10 20 30 40 50 60 70

Figure 11: Students in vocational type of upper secondary education as a percentage of

population aged 16 to 19 years and average firm size in ICT-related services

Average

firm size in

ICT-related

services,

1997

Students in vocational type of upper secondary education as a percentage of population aged 16-19 years,

average of 1995-97

0

10

20

30

40

0 10 20 30 40 50 60 70

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 96


Figure 12: Students in general type of upper secondary education as a percentage of population

aged 16 to 19 years and average firm size in ICT manufacturing

Students in general type of upper secondary education as a percentage of population aged 16-19 years,

average of 1995-97

Average firm

size in ICT

manufacturing

1997

0

50

100

150

200

250

0 10 20 30 40 50 60 70

Figure 13: Students in vocational type of upper secondary education as a percentage of

population aged 16 to 19 years and average firm size in ICT manufacturing

Students in vocational type of upper secondary education as a percentage of population aged 16-19 years old,

average of 1995-97

Average firm

size in ICT

manufacturing

1997

0

50

100

150

200

250

0 10 20 30 40 50 60 70

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 97



and unemployment rate

Unemployment rate,

average of 1991-2000

Students in

upper secondary

education

as a percentage

of working age

population,

average of

1995-97

4

6

8

10

12

14

0 5 10 15 20 25 30


unemployment rate

Unemployment rate,


Students in

tertiary education

as a percentage

of working age

population,

average of

1995-97

0

2

4

6

8

0 5 10 15 20 25 30

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 98


Figure 16: Unemployment rate and business R&D staff

Unemployment rate,


Business R&D

staff as a

percentage

of total

employment,

average

1995-97

0.0

0.5

1.0

1.5

2.0

2.5

0 5 10 15 20 25 30

Figure 17: Per worker GDP growth rate and students in tertiary education as a percentage of

working age population


average of 1995-97

Per worker

GDP growth

rate (%),

average

of 1991-99

-5

0

5

10

15

20

0 2 4 6 8

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 99


Figure 18: Per worker GDP growth rate and students in general type of upper secondary

education as a percentage of working age population

Students in general type of upper secondary education as a percentage of working age population,

average of 1995-97

Per worker

GDP growth

rate (%),

average of

1991-99

-5

0

5

10

15

20

0 1 2 3 4 5 6 7 8

Figure 19: Per worker GDP growth rate and students in vocational type of upper secondary

education as a percentage of working age population

Students in vocational type of upper secondary education as a percentage of working age population,

average of 1995-97

Per worker

GDP growth

rate (%),

average

of 1991-99

-5

0

5

10

15

20

0 1 2 3 4 5 6 7 8

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 100

employment is negative (-0.27) and significant atthe 1 % level (0.01). This, along with the abovefindings, suggests that agglomeration ofhigh-order corporate functions such as R&Ddepartments has a closer association with lowunemployment rate than investment by individ-uals in higher education.


economic growth

An initial look at the relationship between humancapital development and economic growth(Figures 17, 18 and 19) shows relationshipsbetween per worker GDP (i.e. labour productivity)growth rate from 1991 to 1999 and students bylevel of education/training as a percentage of theworking age population.

The top five regions in terms of labour produc-tivity growth rate are those in the ex-GermanDemocratic Republic (GDR) (36). Their extraordinarygrowth is due to the opening of their economy tothe West at the beginning of the 1990s, which gaverise to an influx of capital and technology. As canbe seen from the figures, the numbers of studentsas a percentage of working age population in rela-tion to labour productivity growth does not followthe pattern of the rest of our sample. Accordingly,we considered the ex-GDR regions as exceptionalcases and removed them from the ensuing analysisof economic growth.

Results of the analysis of our sample excludingex-GDR regions are as follows:(a) negative association (-0.26) is found between

per worker GDP growth rate and students intertiary education as a percentage of theworking age population. The association issignificant at the 5 % level (0.05);

(b) there is negative association (-0.35) betweenlabour productivity growth and students invocational type of upper secondary education

as a percentage of the working age popula-tion. It is significant at the 1 % level (0.009);

(c) the correlation between labour productivitygrowth and students in general types ofupper secondary education as a percentageof the working age population is positive(0.21). However, the correlation is not signifi-cant at the 10 % level (0.13).

The results contradict the generally assumedassociation between human capital developmentand economic growth. This clearly indicates thata better understanding of the relationshiprequires a more formal analysis that takes intoaccount the impact of physical capital.

3.2. Formal analysis of therelationship between humancapital and economic growthat regional and national level

3.2.1. Estimates at the augmented

Solow-Swan model: accumulation of

human capital

Formal analysis based on the Solow-Swan frame-work requires data on stocks of physical andhuman capital (e.g. average schooling years of thelabour force). With respect to regions in Europe,data on average schooling years of labour force isnot available (37). Mankiw et al. (1992) overcomesthis requirement by assuming that countries intheir sample are at a steady state in theSolow-Swan framework (38). We used the samemethod and obtained the results in Table 1.

In both models, the fraction of income investedin physical capital (I/GDP), and (n + g + δ), has awrong sign. Furthermore, the fraction of incomeinvested in human capital (SCHOOL) has a wrong


(36) Namely, Thüringen, Sachsen-Anhalt, Sachsen, Mecklenburg, and Brandenburg in order of labour productivity growth rate.0

(37) We attempted to estimate average schooling years of the labour force at the regional level, following the method used by Kyri-acou (1991). In his cross-country study of the growth effects of human capital, Kyriacou examined relationships betweenschooling years and enrolment rates by educational level and devised the following equation,

,

where H75 represents average years of schooling in the labour force, PRIM60 represents the 1960 enrolment rate of primaryeducation, SEC70 represents the 1970 enrolment rate of secondary education, and HIGH70 represents the 1970 enrolment rateof higher education. We evaluated a similar relationship using data for countries in Europe. However, we found no significantrelationship between average schooling years and enrolment rates, so, we were unable to estimate average schooling yearsfor regions using their enrolment data.

(38) See footnote 23.

70706075 0918.86645.24390.40520.0 HIGHSECPRIMH +++=

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 101

sign in model 2, too. This suggests that theassumption of a steady state is not applicable toour sample.

Accordingly, we decided to use the businessR&D staff numbers as an indicator of humancapital stock (39). This is consistent with the study’spurpose as R&D staff embodies high-order humancapital that is deemed critical to technologicalprogress in endogenous growth literature.

We applied the standard augmentedSolow-Swan model Y = AKα Lβ Hγ to the data ofphysical capital stock, total employment, and

business R&D staff in 1990 and 1997 withdifferent depreciation rates (40).

The period from 1990 to 1997 is adopted forthe following reasons:(a) the EU economy entered a new era of integra-

tion after the full liberalisation of capital move-ments in eight Member States in 1990 as well aspolitical landmarks such as the fall of the BerlinWall in 1989 and unification of Germany in 1990.In tandem with the revolutionary effect of ICTsincluding the Internet, economic integrationspurred movement of money and people within


(39) Because total R&D staff data availability is restricted to a smaller number of regions, we could not undertake the same anal-ysis for total R&D staff (including staff in government institutes and higher education institutions) and compare the results withthose for private R&D staff only.

(40) Stocks of physical capital at the regional level are estimated using the Penn World Tables 5.6. A description of the tables isfound in Summers and Heston (1991). First, based on a standard three-factor neoclassical aggregate production function withconstant returns, Y = Kα Lβ Hγ, we estimated at the national level a model that accounts for GDP of EU nations. We obtaineda model whose coefficients are all significant at the 1 % level. Then using the model, we estimated the stock of physical capitalfor each of the regions in our sample. Finally we adopted the perpetual inventory method to produce physical capital stocksfor 1986 and onwards. (See Barro and Sala-i-Martín, 1995 about the perpetual inventory method.) As a depreciation rate δ, weused 0.04. See footnote 37 above. To see effects of a different depreciation rate, we also tested 0.07, following Benhabib andSpiegel (1994).

Model 1 Model 2

Constant 4.49 (a) 4.59 (a)

(0.36) (0.30)

log (I/GDP) -0.84 (b) -0.91 (a)

(0.32) (0.25)

log (n + g + δ) 0.36 0.0004(0.27) (0.25)

log (SCHOOL) 0.05 -0.67 (a)

(0.16) (0.20)

Observations 32 32

R square 0.28 0.48

F statistic 3.58 8.53

NB: - Dependent variable is log of GDP per worker in 1997.- Standard errors are in parentheses.- I/GDP represents gross fixed capital formation as a percentage of GDP (average of 1987-97).- Figures of GDP and gross fixed capital formation are at constant prices in 1995.- (n + g + δ) represents the sum of growth rate of population, growth rate of technical progress, and rate of depreciation.- (g + δ) is assumed to be 0.06 (1).- SCHOOL is students in tertiary education as a percentage of working age population (average of 1995-97) in model 1

and students in upper secondary education as a percentage of working age population (average of 1995-97) in model 2.- (a) and (b) represent 1 % and 5 % significance levels respectively.- Regressions were run using ordinary least squares method.

(1) For a broad sample of countries, Romer (1989) finds that δ is about 0.03 or 0.04. Mankiw et al. (1992) note that growth inincome per capita averaged 1.7 % in the US and 2.2 % in their sample of intermediate countries, and thus suggest that gis about 0.02.

Table 1: Estimation of the augmented Solow-Swan model with the assumption of a steady state

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 102

EU. This, in turn, facilitated the rapid exchangeof information and knowledge, makingleading-edge technology, which firms within aregion aim for in their innovation efforts, moregenerally available. Prevalence of leading-edgeknowledge or technology is often assumed ingrowth model literature;

(b) we used 1985 as the base year to which theperpetual inventory method was applied toestimate physical capital stock in eachregion. This is the earliest year we could useto estimate each region’s capital stock from

the national data. The 1985 capital stock dataat the regional level is subject to estimatingerror. Because of the cumulative calculationof the perpetual inventory method, the laterthe start year of the Solow-Swan analysis, themore reliable the regional capital stock data.At the same time, a reasonably extensiveperiod is necessary for the Solow-Swan anal-ysis to measure long-term growth.

By taking log of the standard augmentedSolow-Swan model, the following equation isobtained:


(41) The equation implies that the first term of the right-hand side of the equation, log AT – log A0, is constant and common for alleconomies under analysis. In other words, technology or knowledge is spread and available to all economies.

.)log(log)log(log)log(logloglogloglog00

000HHLL

KKAAYYTT

TTT

−+−+−+−=−

γβα

Estimated coefficients are shown in Table 2 (41).The augmented Solow-Swan model that

includes physical capital, labour, and businessR&D staff accounts for the economic growth

between 1990 and 1997 with R square of 0.71and 0.72 for models 1 and 2 respectively.Although the estimated coefficient for log differ-ence in physical capital is significant only at the

Model 1 Model 2 Model 3 Model 4

Constant -0.04 -0.03 0.72 (a) 0.73 (a)

(0.05) (0.04) (0.16) (0.16)

dK 0.40 (c) 0.42 (c) 0.36 (b) 0.38 (b)

(0.22) (0.21) (0.17) (0.16)

dL 0.53 (a) 0.49 (a) 0.67 (a) 0.63 (a)

(0.16) (0.17) (0.13) (0.04)

dH 0.16 (a) 0.15 (a) 0.08 (b) 0.08 (b)

(0.04) (0.04) (0.03) (0.03)

Y0 -0.16 (a) -0.16 (a)

(0.03) (0.03)

Observations 32 32 32 32

R square 0.71 0.72 0.84 0.84

F statistic 22.8 23.6 36.6 38.3

NB: - Dependent variable is the log difference in GDP.- Standard errors are in parentheses.- dX refers to the log difference in variable X.- The period for comparison is 1990-97.- K, L, H, and Y0 represent stock of physical capital, number of workers, and size of business R&D staff,

and per worker GDP in 1990 respectively.- Figures of physical capital are at 1995 constant prices.- (a), (b), and (c) represent 1 %, 5 %, and 10 % significance levels respectively.- Models 1 and 3 use a depreciation rate of 0.04.- Models 2 and 4 use a depreciation rate of 0.07.- Regressions were run using ordinary least squares method.

Table 2: Estimation of the augmented Solow-Swan model

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 103

10 % level (0.08 in model 1 and 0.05 in model 2),log differences in labour and business R&D staffenter highly significantly. The significance of logdifference in labour is 0.003 in model 1 and 0.008in model 2, and the significance of log differencein private R&D staff is 0.001 in both models 1 and2. The results suggest that the change in busi-ness R&D staff during the period varies in a waydistinct from those in physical capital and labourand accounts for a significant part of the variationof the change in GDP. The positive sign for thecoefficient of the log difference in business R&Dstaff means that an increase in business R&Dstaff led to an increase in the region’s per workerGDP (i.e. labour productivity).

We also entered initial income per worker (GDPin 1990 worker), Y0, in the equation (model 3 andmodel 4). In the neoclassical Solow-Swan frame-work that supposes decreasing returns oncapital, the level of per capita GDP will convergetoward its steady state asymptotically. The speedof convergence increases with the distance to thesteady state. In other words, when the determi-nants of the steady state are controlled for, thelower initial values of per capita GDP, the highertransitional growth rates. This is called ‘condi-tional convergence’. As expected from previousempirical studies that support conditional conver-gence, initial per worker GDP in 1990 enters witha negative sign at the 1 % significance level (0.00in both model 3 and model 4). More importantly,the coefficient for log difference in business R&Dstaff remains positive and significant at the 5 %level (0.03 in both model 3 and model 4) althoughits value becomes less than in models 1 and 2.

The results of Table 2 suggest that high-orderhuman capital represented by business R&D staffcontributes to economic growth in the same way asother production factors (i.e. physical capital,

labour). That is, it is the accumulation of the humancapital that affects economic growth. This isconsistent with the neoclassical Becker-Lucasframework predicting that the accumulation ofhuman capital determines the marginal productivityof education and maintains it at a positive level.

3.2.2. Estimates at the Nelson-Phelps

framework: level of human capital stock

Nelson and Phelps, and Benhabib and Spiegelprovide a different framework, according to whichit is the level of human capital stock that affectsan economy’s capacity to develop and implementnew technologies. The level of human capitalstock is positively related to the rate of technicalprogress, or the growth rate of productivityparameter A, in their view.

R&D staff certainly reflects such a capacity in aneconomy. In their cross-country study, Benhabiband Spiegel used average years of schooling inthe labour force as an indicator of human capital,focusing on investment in education as a whole.Unlike schooling years, R&D staff are a smallsegment of the labour force that embodieshigh-order human capital. Though small in size,this sector is considered a good representation ofan economy’s capacity to develop new technolo-gies (i.e. knowledge) as argued in an endogenousgrowth literature. Also, R&D staff are frequentlyinvolved in transferring and implementing newtechnologies developed in other sectors or under-standing new technologies invented by other firmsin their own sectors (e.g. reverse engineering).Hence the framework of Nelson and Phelps maybe applicable to R&D staff.

We tested two equations based on this frame-work. The first equation takes an average of humancapital levels during the period under examination.


.)log1()log(log)log(logloglogloglog

00

000

∑+−+−+−=−T

tT

TTT

HTLL

KKAAYY

γβα

Table 3 shows results of the estimates of coef-ficients.

In model 1, all terms, including log of anaverage of business R&D staff in level from 1990to 1997, enter the equation significantly at the1 % level. However, whereas the coefficients forlog differences in physical capital and labour arepositive (as expected), the coefficient for the

average business R&D staff is estimated as nega-tive. In other words, regions with a higher level ofbusiness R&D staff experienced a slower growthfrom 1990 and 1997.

In model 2, we included initial per worker GDP(labour productivity) in 1990 in the equation tosee the effects of conditional convergence. Logdifferences in physical capital and labour enter

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 104

significantly (0.02 and 0.00 respectively), taking aproper positive sign. The term for initial labourproductivity, Y0, also enters significantly at theconventional 5 % level (0.03). Its negative signshows that the lower initial per worker GDP, thefaster an economy grows. In other words, condi-tional convergence took place among theregions. As for the average business R&D staff inlevel, its coefficient takes a negative sign.Furthermore, it fails to enter significantly (0.32).

The failure of business R&D staff to enter the equation significantly in model 2 is due to close asso-ciation between business R&D staff in level and per

worker GDP. As the level of business R&D staff hasclose association with the level of labour produc-tivity (42), the average R&D staff from 1990 to 1997,AH, in model 1 acts as a proxy for the level of labourproductivity and enters the equation with a negativesign, suggesting conditional convergence among theregions. However, when initial labour productivity, Y0,is included in model 2, it accounts for the conver-gence better than level of R&D staff does. Hence levelof R&D staff loses its significance in model 2.

The second equation we tested includes twoterms of human capital that represent its differenteffects.


(42) We examined contributions of physical capital, labour, and business R&D staff in levels to the level of income of regions in1990 and 1997, using a standard three-factor neoclassical aggregate production function with constant returns, Y = Kα Lβ Hγ.The level of business R&D staff enters the equation significantly with a positive sign, suggesting that regions with a higher levelof business R&D staff have a higher level of income per capita (labour productivity).

Model 1 Model 2

Constant -0.16 (a) 0.58 (c)

(0.04) (0.30)

dK 0.92 (a) 0.58 (b)

(0.20) (0.24)

dL 0.50 (a) 0.62 (a)

(0.14) (0.14)

AH -0.06 (a) -0.02

(0.01) (0. 02)

Y0 -0.14 (b)

(0.06)

Observations 32 32

R square 0.78 0.82


NB: - Dependent variable is the log difference in GDP.- Standard errors are in parentheses.- dX refers to the log difference in variable X.- The period for comparison is 1990-97.- K, L, and Y0 represent stock of physical capital, number of workers, and per worker GDP in 1990 respectively.- AH is an average of business R&D staff during 1990-97.- Figures of physical capital are at 1995 constant prices.- Depreciation rate is 0.04.- (a), (b), and (c) represent 1 %, 5 %, and 10 % significance levels respectively.- Regressions were run using ordinary least squares method.

Table 3: Cross-regional growth accounting results: human capital in log levels

[ ]

( ).)log(log)log(log

)()log(log)log(log

/)(loglog

00

max

00

max0

LLKKYYmHHmgc

LLKKYYYmHgHcYY

TT

iii

TT

iiiiT

−+−++−+=

−+−+−++=−

βα

βα

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 105

The term, gH, represents endogenous develop-ment based on the level of human capital, andthe term, mHi [(Ymax – Yi)/Yi], represents catch-upof region i with the region leading in terms of percapita GDP, Ymax (i.e. the region with the highestlabour productivity in the data set).

Table 4 shows the results.In model 1, both log differences in physical

capital and labour enter the equation at the 1 %significance level (0.001 and 0.009 respectively)with a proper positive sign. The coefficient esti-mate for (g – m) on H is negative and significantat the 10 % level (0.08). The negative signsuggests that regions with a higher level of busi-ness R&D staff in 1990 experienced a slowergrowth from 1990 to 1997. In contrast, thecatch-up term of business R&D staff, H0 (Ymax/Yi),fails to enter the equation significantly (signifi-cance: 0.79) though it takes a positive sign.

In model 2, we added the initial ratio of thelabour productivity of the leading-edge region tothat of region i to see the effects of conditionalconvergence (43). As in model 2 in Table 3, theinitial position of a region in terms of labourproductivity, Ymax/Yi, enters at the 1 % significancelevel (0.002). Its sign is positive, suggesting condi-tional convergence: the larger the initial gap withthe leading region, the faster an economy grows.On the other hand, neither Y0 nor Y0(Ymax/Yi) enterssignificantly (0.62 and 0.34 respectively). In otherwords, conditional convergence that is due to aregion’s initial position in per worker GDP exertsmore significant effects upon its change in GDP.Although the size of business R&D staff, H0, playsthe role of a proxy for labour productivity level, itseffects as an economy’s capacity to develop andimplement new technology in the Nelson-Phelpsframework are found insignificant.

These findings are in stark contrast to those ofBenhabib and Spiegel. In their cross-countrystudy, Benhabib and Spiegel found that thecatch-up term of human capital enters signifi-cantly. The catch-up term maintains its propersign (i.e. positive) and significance (though signif-icance level drops from 1 % to 5 %) even if the

variable for initial position of income Ymax/Yi isincluded. In contrast, our results show thatneither of the human capital terms is found signif-icant, whereas initial position as well as log differ-ences in physical capital and labour enter at anequal or higher level of significance than in thecase of Benhabib and Spiegel (44).

Even if we focus on wealthier economies in thesample of Benhabib and Spiegel, their results aredifferent. In their study, Benhabib and Spiegeldivided their sample into 3 groups (26 countrieseach) according to wealth. For the wealthiest thirdof the sample, they found that the endogenousdevelopment term of human capital H enters signif-icantly while the catch-up term is found insignifi-cant. However, the sign for H is found positive intheir study suggesting that the capacity to developand implement new technology, which H repre-sents, contributes to economic growth positively. Incontrast (as shown in model 1 in Table 4), the sign ofthe term in our results is negative, suggesting thatthe greater the volume of human capital in level(i.e. business R&D staff in our case), the slower aneconomy grows.

Most important, accumulation of human capital(i.e. business R&D staff in our case) accounts foreconomic growth at a significant level along withaccumulation of physical capital. In the study ofBenhabib and Spiegel, accumulation of humancapital fails to enter significantly with respect toall three measures of human capital theytested (45) as well as alternative subsamples ofthe data. In addition, the sign of human capitalaccumulation was found negative in most of thecases they examined.

3.2.3. Human capital and growth:

cross-country estimates

An obvious, possible source for the differencebetween the findings is the different types ofhuman capital examined. While Benhabib andSpiegel adopt average schooling years of thelabour force, we use the number of private sectorworkers engaged in R&D which endogenousgrowth literature focuses upon as a source of


(43) This follows the equation adopted by Benhabib and Spiegel. Adding initial labour productivity in 1990 (as in model 2 in Table 3as well as many studies of conditional convergence) produces similar results. In that case, the coefficient for the variable takesa negative sign.

(44) In their estimates, log difference in labour fails to enter the model significantly whereas the coefficient for log difference inphysical capital was found significant at the 1 % level.

(45) Namely, average schooling years used by Kyriacou (1991), the Barro and Lee (1993) estimate of human capital, and literacy.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 106

technical progress. To see if the different type ofhuman capital affects the results, we undertookan analysis based on average schooling years.Since data on schooling for the labour force isnot available at regional level, the analysis is onlyat national level. The figures for persons in the

labour force (25 to 59 years old) who completededucation, given by educational level, are avail-able for 11 countries in the EU for the period from1992 to 2000. From the data, we calculated anestimate of average schooling years (46) andtested the following three models:


(46) We gave 9 years, 12 years and 16 years for those who completed lower secondary education, upper secondary education, andtertiary education respectively.

Model 1 Model 2

Constant -0.11 (b) -0.15 (a)

(0.04) (0.04)

dK 0.88 (a) 0.59 (a)

(0.23) (0.21)

dL 0.41 (a) 0.65 (a)

(0.14) (0.14)

H0 -0.05 (c) 0.01

(0.03) (0.03)

H0 (Ymax/Yi ) 0.006 (0.02)

-0.02 (0.02)

Ymax/Yi 0.04 (a)

(0.01)

Observations 32 32

R square 0.78 0.85


NB: - Dependent variable is the log difference in GDP.- Standard errors are in parentheses.- dX refers to the log difference in variable X.- The period for comparison is 1990-97.- K and L represent stock of physical capital and number of workers respectively.- H0 is the size of business R&D staff in 1990.- Ymax/Yi represents the ratio of the labour productivity of the leading-edge region to that of region i in 1990.- Figures of physical capital are at 1995 constant prices.- Depreciation rate is 0.04.- (a), (b), and (c) represent 1 %, 5 %, and 10 % significance levels respectively.- Regressions were run using ordinary least squares method.

Table 4: Cross-regional growth accounting results: Human capital in levels and in product of

levels and differences from the leading-edge region in labour productivity

,)log(log)log(log)log(logloglogloglog00

000HHLL

KKAAYYTT

TTT

−+−+−+−=−

γβα

,)log1()log(log)log(logloglogloglog

00

000

∑+−+−+−=−T

tT

TTT

HTLL

KKAAYY

γβα

and

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 107

Table 5 shows the results.The estimates show lower significance, in large

part due to the small sample size. However, thepatterns of significance are consistent with theresults for business R&D staff except for insignif-icant estimates for labour. In model 1, only logdifference in physical capital enters significantly

(0.014) although all variables take a proper, posi-tive sign. The significance of log difference inschooling years is 0.14, being above the conven-tional maximum cut-off point of 10 %. However,in model 2 that includes initial level of income, logdifference in schooling years and as log differ-ence in physical capital, enter significantly at the


( ).)log(log)log(log

)(loglog00

max0LLKK

YYmHHmgcYYTT

iiiT

−+−++−+=−

βα

Model 1 Model 2 Model 3 Model 4

Constant -0.06 -0.15 -0.05 0.01

(0.04) (0.17) (0.09) (0.16)

dK 0.61 (b) 0.75 (b) 0.55 (b) 0.49

(0.19) (0.21) (0.22) (0.28)

dL 0.002 -0.14 0.02 0.06

(0.30) (0.31) (0.36) (0.41)

dH 0.33 0.56 (a)

(0.20) (0.26)

AH 0.0009

(0.006)

H0 0.005

(0.02)

H0 (Ymax/Yi ) -0.008

(0.03)

Y0 -0.05

(0.04)

Observations 11 11 11 11

R square 0.65 0.73 0.52 0.52

F statistic 4.3 4.1 2.5 1.6

NB: - The following 11 countries were included: Belgium, Denmark, Germany, Greece, Spain, France, Ireland, Italy, the Nether-lands, Portugal, and the UK.

- Dependent variable is the log difference in GDP.- Standard errors are in parentheses.- dX refers to the log difference in variable X.- The period for comparison is 1992-2000.- K, L and H represent stock of physical capital, number of workers, and average schooling years of the labour force

respectively.- AH is an average of H during 1992-2000.- H0 is the value in 1992.- Ymax/Yi represents the ratio of the labour productivity of the leading-edge region to that of region i in 1992.- Y0 is per worker GDP in 1992.- Figures of physical capital are at 1995 constant prices.- Depreciation rate is 0.04.- (a), (b), and (c) represent 1 %, 5 %, and 10 % significance levels respectively.- Regressions were run using ordinary least squares method.

Table 5: Cross-country growth accounting results, 1992-2000

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 108

levels of 10 % and 5 % (0.07 and 0.011) respec-tively. In other words, an improvement of educa-tional level in the labour force led to highereconomic growth during the period. In contrast,model 3 and model 4, which take schooling yearsin levels, fail to be significant (47). Estimates of thecoefficients for human capital terms in the twomodels are very small relative to standard errors,showing little significance. Though the results arenot conclusive due to the small sample size, theysuggest that, even if schooling years are adopted

as an indicator of human capital (as in the studyof Benhabib and Spiegel), their accumulation, nottheir stock level, accounts for economic growth.

It is most likely that catch-up effects of humancapital Benhabib and Spiegel observed are found inthe case of extremely wide gaps between devel-oped countries and developing countries, includingones in Africa and Latin America. When we focusupon developed economies such as regions in theEU, the effects of human capital upon economicgrowth derives from its accumulation.


(47) The significances of the models gained from analysis of variance (F statistic) are 0.15 and 0.28 respectively.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 109

This report has reviewed literature and empiricallyexamined the way in which human capital devel-opment is associated with economic growth inregions in Europe in the 1990s. While powerfulmultinational corporations are becoming ever-more ‘footloose’ and escaping from the control ofnation states, regions are ‘competing’ againstone another, within and across nations, atattracting investments and supporting local busi-nesses to increase competitiveness in the globalmarket. In this, many regions, and particularlyones in peripheries, are facing the reality ofpersistent regional disparities in productivity andcorporate functions located. By examining therelationship between human capital formationand economic performance in European regions,this study aims to go some way towardsanswering the question whether, and how, humancapital development provides these territorialunits with comparative advantages over others.

In growth model literature, human capital hasgained increased recognition as one of keyproduction factors following physical capital andlabour. The original Solow-Swan model consistedof physical capital and labour alone, failing toaccount for a significant part of income growth asresidual (called the ‘Solow residual’). This residualwas considered to be an exogenous factor thatderives from technical progress taking placewhere there is a public science base. In themeantime, a number of studies, most notably theseminal work of Becker, drew attention to invest-ment in skills and knowledge as another factor ofproduction. As a result, the Solow-Swan modelcame to be augmented with the inclusion of anadditional term of human capital.

4.1. Significance of economicgrowth models

Another development in growth model literatureis endogenous growth models. Whereas theneoclassical Solow-Swan model treats technicalprogress as an exogenous factor, endogenous

growth models take into consideration theprocess of innovation and technology diffusion,making a departure from the assumption ofperfect competition. The mechanism of investingin research takes centre stage in the model. As aresult, R&D workers, who embody high-orderhuman capital, are often included in the models.

An issue identified in growth model literature isthe way human capital development affectseconomic growth. According to the frameworkdeveloped by Lucas on the basis of Becker’shuman capital theory, growth is primarily drivenby the accumulation of human capital. Incontrast, Nelson and Phelps contend that thestock of human capital determines the economy’scapacity to innovate or catch up with moreadvanced economies, which in turn driveseconomic growth. Hence, the level of humancapital stock is, though indirectly, a determinantof per capita economic growth in this view. Whilethe first framework is in agreement with neoclas-sical growth models, the second framework hasrevived in the Schumpeterian growth literature.

4.2. The regional dimension

These economic models are criticised ineconomic geography literature for their failure toconsider spatial aspects of economic develop-ment. Unlike the neoclassical assumption thatfactors of production are completely mobile,knowledge and innovations do not diffuse instan-taneously or at an even rate over the economy asa whole. Tacit knowledge in particular, whosetransfer often relies on interpersonal, oftenface-to-face communications, is argued to beconcentrated in major metropolitan areas, signi-fying the importance of human capital that repre-sents a carrier of such knowledge. Furthermore,there is a view that the position of metropolitanareas is strengthened by the cumulative causationmechanism in which the education and trainingsystem interacts with the local economy to furtherspatial disparities in economic well-being.

Our empirical analysis of European regions

4. Summary and conclusions

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 110

shows that investment by individuals in humancapital development has distinct patterns. Thoseregions with a higher level of investment intertiary education tend to have a larger concen-tration of ICT sectors (including provision of ICTservices and manufacture of ICT devices andequipment) and research functions. On the otherhand, there is no significant association betweensuch high-order functions and investment inupper secondary education, both general andvocational types.

In relation to the density of economic activity,those regions that include major metropolitanareas show a high enrolment rate for tertiaryeducation. While this is, to some degree, due to aconcentration of higher education institutions inthose areas, the association also suggests apossibility of link to high-order corporate func-tions that tend to concentrate in high-densitymetropolitan areas. However, some low-densityregions have made progress in take-up of tertiaryeducation at a faster pace, closing a gap in theformation of high-order human capital.

Generation of new firms is an important sourceof economic dynamism. In our empirical analysis,the rate of human capital development appearsto have significant association with the rate ofentrepreneurship in emerging industries such asICT-related services and ICT manufacturing,whereas such association is not found with tradi-tional manufacturing industries. The associationwith entrepreneurship in ICT sectors is, however,found at the level of vocational type of uppersecondary education, not at the level of tertiaryeducation. The implications of this are subject tointerpretation and require further research.

Individuals in regions that suffer from highunemployment rate tend to invest more in educa-tion and training. This fits with the general obser-vation that individuals are more likely to invest ineducation and training when facing withunfavourable employment situations at economicdownturns. Another finding in relation to this isthat individuals in regions with lower unemploy-ment rate tend to have a higher level of R&Dfunctions.

In short, a high level of investment by individ-uals in tertiary education is found in those regionsthat accommodate high-tech industries andhigh-order corporate functions like R&D. Regionssupport such high-order functions through the

urban infrastructure, facilitating exchange of tacitknowledge, as well as through a public sciencebase. They also enjoy a low unemployment rate.

However, the existing stock of human capitaldoes not lead to a high rate of economic growth.We did not find any significant effects of scalethat would favour those regions with a largerstock of human capital. Instead, our empiricalanalysis demonstrates that the rate of economicgrowth is associated with the accumulation ofhuman capital. Furthermore, those regions with alower per worker GDP at the beginning of theperiod of our analysis tend to show a fastergrowth rate.

4.3. Policy implications

The primary policy implication of our study is theneed to support continuous human capital devel-opment. The growth of computers and digitaltechnology has led a number of observers to arenewed enthusiasm for the Schumpeterianvision of capitalist creative destruction (Harris,2001). With the notion of the ‘knowledge-basedeconomy’ they view that creation of knowledgeand its conversion to commercial use plays agreater role than ever in economic development.Knowledge is distinct from natural resources in itsnon-rival nature: its use by one firm or person inno way limits its use by another. Once it iscreated, knowledge is not depleted: it adds to theexisting stock of knowledge.

This non-excludable nature of knowledge inturn allows research workers to share the stockof knowledge and act on it to create new ideas.New ideas will be then embodied in new orimproved products and production processes,bringing about economic growth. A predictiondrawn from this view of the knowledge-basedeconomy is its ever-continuing growth. As aneconomy shifts its primary activities to knowl-edge creation, with its conversion to commercialvalue, the economy is more likely to benefit fromthe existing stock of knowledge that onlycontinues to grow. The view also predicts theadvantage of an economy with a large stock ofresearch workers: the more workers are engagedin research, the more new ideas are likely to becreated. Accordingly, a rosy picture of continued


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 111

growth in advanced economies emerged with thelong-term boom in the US economy in the 1990s.

In contrast to the prediction of theknowledge-based economy, our findings showthat there is not any significant associationbetween the existing stock of research workersand economic growth. Instead, economic growthis found to be associated with accumulation ofresearch workers. This suggests that a key toeconomic growth is continuous development ofhigh-order human capital. While there is littledoubt that advanced economies in WesternEurope are becoming more knowledge-based,they cannot rest on the existing research base togrow further. Given the increasingly fast pace oftechnological change that makes human capitalobsolete, a concerted effort needs to be made tofacilitate continuous development of high-orderhuman capital.

The development of high-order human capitaldoes not mean education and training to increaseresearch workers alone but refers to developmentof ‘knowledge workers’ in a broad spectrum ofeconomic activities. As Kline and Rosenberg (1986)demonstrate, knowledge creation has shifted froma traditional linear process of innovation to a morecomplex chain-linked model based on interactionsbetween knowledge workers. In this new mode,creation of new ideas takes place throughout theentire value chain spanning an organisation’sdifferent functions (e.g. R&D, production,marketing, sales) and its external partners (e.g.suppliers, customers, universities, research insti-tutes, government organisations). Development ofsuch an organisation-wide innovation capacity inan economy will require education and trainingpolicy that aims to upgrade continuously a broadrange of human capital to a higher level.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 112

GDR German Democratic Republic

ICT Information and communication technology

NUTS Nomenclature of territorial units for statistics



01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 113

Suppose the prices of output F and factors K, L and A be p, wK, wL, and wA respectively. If each factoris paid its marginal product,

Output F can be then written as:

(5)

According to Euler’s Theorem, if f(x1, ..., xn) has continuous first partial derivatives and is positivelyhomogeneous of degree k, then:

... ...

This suggests that equation (5) holds if F(K, L, A) is positively homogeneous of degree 1, that is:

It is easy to see that this would mean constant returns in K, L and A. Hence under increasing returns, allfactors cannot be paid their marginal products.

)( )( ALKFttAtLtKF ,,,, =

)nx,) (n xfkx ,, 1=(i

n

ii xfxx ,1

1K∑

= ∂∂

( ) AFAL

FLKFKAwLwKwpF ALK ∂

∂∂∂

∂∂ ++=++= 1

AFpwA ∂

∂=LFpwL ∂

∂=KFpwK ∂

∂=

Annex 1Technical difficulty of incorporating increasing returns toa neoclassical model

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 114

The rate of saving is determined by the owner of the representative (i.e. average) one-worker firm whotries to maximise his lifetime utility W. Thus the problem is expressed as:

subject to (i.e. investment equals net product minus consumption) and .

Assuming a constant intertemporal elasticity of substitution (i.e. ), the above dynamicoptimisation problem yields the Euler equation:

(a )(because of L = 1, F = A

_Kα in the case of the representative one-worker firm).

ρα −−1KAερ∂∂

ε =

−= 11KF

cdtdc

εε

−−= −

11)( 1ccu

0≥dtdKcKAdt

dK −= α

( ) dtecu tt

ρ−∞∫ 0max

Annex 2Growth rate of consumption in Romer (1986) model

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 115

Abramovitz, M. Catching up, forging ahead, andfalling behind. Journal of Economic History.Cambridge: Cambridge University Press, 1986,Vol. 46, p. 385-406.

Aghion, P.; Howitt, P. A model of growth throughcreative destruction. Econometrica. Oxford:Blackwell Publishing, 1992, Vol. 60, p. 323-351.

Aghion, P.; Howitt, P. Endogenous growth theory.Cambridge, MA: MIT Press, 1998.

Armstrong, H.; Taylor, J. Regional economic andpolicy, 3rd ed. Oxford: Blackwell Publishing,2000.

Arrow, K. J. The economic implications of learningby doing. Review of Economic Studies. Oxford:Blackwell Publishing, 1962, Vol. 29, issue 3,p. 155-173.

Barro, R. J. Economic growth in a cross section ofcountries. Quarterly Journal of Economics.Cambridge: MIT Press Journals, 1991, Vol. 106,p. 407-443.

Barro, R. J.; Lee, J.-W. International comparisons ofeducational attainment. Journal of MonetaryEconomics. Amsterdam: Elsevier Science B.V.,1993, Vol. 32, issue 3, p. 363-394.

Barro, R. J.; Sala-i-Martín, X. Economic Growth.New York: McGraw-Hill, 1995.

Bartel, A. P.; Lichtenberg, F. R. The comparativeadvantage of educated workers in imple-menting new technology. The Review ofEconomics and Statistics. Cambridge, MA: MITPress Journals, 1987, Vol. 69, issue 1, p. 1-11.

Baumol, W. J.; Blackman, S. A. B.; Wolff, E. N.Productivity and American leadership: the longview. Cambridge, MA: MIT Press, 1989.

Becker, G. S. Human capital: a theoretical andempirical analysis, with special reference toeducation. New York: National Bureau ofEconomic Research, 1964.

Becker, G. S.; Murphy, K. M.; Tamura, R. Humancapital, fertility, and economic growth. Journalof Political Economy. Chicago: University ofChicago Press, 1990, Vol. 9, issue 5,p. S12-S37.

Benhabib, J.; Spiegel, M. M. The role of humancapital in economic development: evidencefrom aggregate cross-country data. Journal of

Monetary Economics. Amsterdam: ElsevierScience B.V., 1994, Vol. 34, issue 2, p. 143-173.

Bradley, S.; Taylor, J. Human capital formation andlocal economic performance. Regional Studies.Philadelphia: Carfax publishing, 1996, Vol. 30,No 1, p. 1-14.

Crouch, C.; Finegold, D.; Sako, M. Are skills theanswer? The political economy of skill creationin advanced industrial countries. Oxford:Oxford University Press, 1999.

Denison, E. F. Accounting for slower economicgrowth: the United States in the 1970s. Wash-ington, DC: Brookings Institution, 1979.

Dunford, M. Divergence, instability and exclusion:regional dynamics in Great Britain. In: Lee, R.;Wills, J. (eds) Geographies of Economies.London: Hodder Arnold, 1997, p. 259-277.

Frankel, A. The production function in allocationand growth: a synthesis. American EconomicReview. Nashville: American Economic Associ-ation, 1962, Vol. 52, p. 995-1022.

Granovetter, M. Economic action and social struc-ture: the problem of embeddedness. AmericanJournal of Sociology. Chicago: University ofChicago Press, 1985, Vol. 91, issue 3,p. 481-510.

Griliches, Z. Hybrid corn: an exploration in theeconomics of technological change. Econo-metrica. Oxford: Blackwell Publishing, 1957,Vol. 25, p. 501-522.

Griliches, Z. Notes on the role of education inproduction functions and growth accounting.In: Hansen, W. L. (ed.) Education, income, andhuman capital. New York: National Bureau ofEconomic Research, 1970 (Studies in incomeand wealth, Vol. 35).

Grossman, G. M.; Helpman, E. Innovation andgrowth in the global economy. Cambridge, MA:MIT Press, 1991.

Hägerstand, T. Aspects of the spatial structure ofsocial communication and the diffusion of infor-mation. Papers and Proceedings. RegionalScience Association, 1966, Vol. 16, p. 27-42.

Harris, R. G. The knowledge-based economy: intel-lectual origins and new economic perspectives.International Journal of Management Reviews.

References

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 116

Oxford: Blackwell Publishing, 2001, Vol. 3,issue 1, p. 21-40.

Huggins, R.; Izushi, H. World knowledge competi-tiveness index 2002: benchmarking the globe’shigh performing regions. Cardiff: RobertHuggins Business & Economic Policy Press,2002.

Jones, C. I. R&D-based models of economicgrowth. Journal of Political Economy. Chicago:University of Chicago Press, 1995, Vol. 103,issue 4, p. 759-784.

Kendrick, J. W. The formation and stocks of totalcapital. New York: National Bureau of EconomicResearch, 1976 (General series, No 100).

Kline, S. J.; Rosenberg, N. An overview of innova-tion. In: Landau, R.; Rosenberg, N. (eds) Thepositive sum strategy: harnessing technologyfor economic growth. Washington, DC:National Academies Press, 1986, p. 275-305.

Kyriacou, G. Level and growth effects of humancapital: a cross-country study of the conver-gence hypothesis. New York, NY: C.V. StarrCenter, 1991 (Working paper, No 91-26).

Lucas, R. E. Jr. On the mechanics of economicdevelopment. Journal of Monetary Economics.Amsterdam: Elsevier Science B.V., 1988,Vol. 22, issue 1, p. 3-42.

Mankiw, N. G.; Romer, D.; Weil, D. N. A contributionto the empirics of economic growth. QuarterlyJournal of Economics. Cambridge: MIT PressJournals, 1992, Vol. 107, issue 2, p. 407-437.

Mansfield, E. Statistics for business andeconomics: methods and applications, 3rd ed.New York: W. W. Norton, 1980.

Marshall, A. Principles of economics. London:Macmillan, 1890.

Nelson, R. R.; Phelps, E. Investment in humans,technological diffusion and economic growth.American Economic Review. Nashville: Amer-ican Economic Association, 1966, Vol. 61, No 2,p. 69-75.

OECD – Organisation for Economic Cooperationand Development. The knowledge-basedeconomy. Paris: OECD, 1996.

Ohmae, K. The end of the nation-state. New York:Free Press, 1995.

Rauch, J. E. Productivity gains from geographicconcentration of human capital: evidence fromthe cities. Journal of Urban Economics. SanDiego, CA: Academic press inc., 1993, Vol. 34,issue 3, p. 380-400.

Richardson, H. W. Regional Economics. Urbana:University of Illinois Press, 1979.

Romer, P. M. Increasing returns and long-rungrowth. Journal of Political Economy. Chicago:University of Chicago Press, 1986, Vol. 94,issue 5, p. 1002-1037.

Romer, P. M. Capital accumulation in the theory oflong run growth. In: Barro, R. J. (ed.) Modernbusiness cycle theory. Cambridge, MA: HarvardUniversity Press, 1989, p. 51-127.

Romer, P. M. Endogenous technological change.Journal of Political Economy. Chicago: Univer-sity of Chicago Press, 1990, Vol. 98, issue 5,p. S71-S102.

Saxenian, A. Regional advantage: culture andcompetition in Silicon Valley and Route 128.Cambridge, MA: Harvard University Press,1994.

Schultz, T. W. Capital formation by education.Journal of Political Economy. Chicago: Univer-sity of Chicago Press, 1960, Vol. 68, p. 571-583.

Schultz, T. W. Investment in human capital. Amer-ican Economic Review. Nashville: AmericanEconomic Association, 1961, Vol. 51, No 1,p. 1-17.

Segerstrom, P. S.; Anant, T.; Dinopoulos, E. ASchumpeterian model of the product life cycle.American Economic Review. Nashville: Amer-ican Economic Association, 1990, Vol. 80,p. 1077-1092.

Shell, K. Toward a theory of inventive activity andcapital accumulation. American EconomicReview. Nashville: American Economic Associ-ation, 1966, Vol. 56, No 2, p. 62-68.

Shell, K. A model of inventive activity and capitalaccumulation. In: Shell, K. (ed.) Essays on thetheory of optimal economic growth.Cambridge, MA: MIT Press, 1967, p. 67-85.

Smith, A. An inquiry into the nature and causes ofthe wealth of nations. Harmondsworth:Penguin, 1776 (reprinted 1970).

Solow, R. M. Technical change and the aggregateproduction function. Review of Economics andStatistics. Cambridge, MA: MIT Press Journals,1957, Vol. 39, p. 312-320.

Summers, R.; Heston, A. The penn world table(mark 5): an expanded set of internationalcomparisons, 1950-1988. Quarterly Journal ofEconomics. Cambridge: MIT Press Journals,1991, Vol. 106, issue 2, p. 327-368.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 117

Swan, T. W. Economic growth and capital accumu-lation. Economic Record. Oxford: BlackwellPublishing, 1956, Vol. 32, p. 334-361.

Williamson, O. E. Markets and hierarchies: analysisand antitrust implications. New York: FreePress, 1975.

Williamson, O. E. The economic institutions of capi-talism. New York: Free Press, 1985.

Wolff, E. N. Technology and the demand for skills.STI – Science, Technology and Industry Review.OECD, 1996, No 18.

Wolff, E. N. The role of education in the postwarproductivity convergence among OECD coun-tries. Industrial and Corporate Change. Oxford:Oxford University Press, 2001, Vol. 10, issue 3,p. 735-759.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 118

Non-material benefits of education,training and skills at a macro level

Andy Green, John Preston, Lars-Erik Malmberg

AbstractThe macrosocial, as opposed to the microsocial or economic, benefits of education are often neglectedby researchers and policy-makers. An emphasis on individual actors and communities as the foundationsof macrostructures has limited potential when understanding societal properties. Macrosocial conceptssuch as social cohesion and societal values have received much attention in sociological theory as itrelates to education and this paper is part of what we identify as a wider rediscovery of such conceptsin educational research and policy.The macrosocial is conceptually different from the microsocial in terms not only of level, but also in termsof its emphasis on the relational properties of social functioning. These conceptual differences are high-lighted through discussion of various macrosocial indicators – crime, social cohesion, societal values andcitizen participation.Beginning with an analysis of crime, we discuss why developmental and economic perspectives on crim-inal behaviour offer a limited understanding of crime, particularly in a comparative context. However, thestructural antecedents of crime (particularly inequality and labour market position) are shown to bepowerful explanatory tools.Next we move to consider social cohesion, trust and values. Classical sociological accounts clearlyconstrue cohesion as a macrosocial issue but current policy debates rarely refer to classical conceptionsand freely conflate societal aspects of cohesion with micro and meso concepts of social capital andcommunity. Through a discussion of these themes we ascertain that national and historical differences inthese properties cannot be explained as aggregated microphenomena. Moreover, the educationalsystem, including vocational education and training (VET), has a pivotal (yet sometimes contradictory)role to play in the construction of social cohesion.It is not as clear that active citizenship and the antecedents of civic and political participation are macrophenomena in the same way as social cohesion. There are clear mechanisms involving resources andstatus which link VET to all aspects of participation. However, there are also cultural differences in thenature of participation and the relationships between participation and general trust. There is, therefore,a need for micro relationships between education and citizenship outcomes to be understood in termsof local and national context as part of a more general macro-micro synthesis.From this review of literature we discuss the ways in which we may evaluate the relationship betweenVET and macrosocial outcomes. Although an evaluation in a summative and final sense would beill-conceived in terms of societal outcomes, a hybrid of formative evaluation and macro causal forms ofcomparison is potentially fruitful. We employ a macro causal and evaluative form of comparison throughthe use of both macrosocial and microsocial data. These enable us to test hypotheses related to distri-butional considerations and the primacy of cultural over individual factors. Although the results are

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 119

exploratory, there is some evidence that distribution considerations are important factors in producingsocial cohesion.We conclude by restating the case for comparative approaches to these issues. Both distribution consid-erations and the values transmitted through education are important in realising macrosocial benefits, inparticular social cohesion. In terms of implementation, although there are cultural limits on the extent that‘policy borrowing’ is appropriate, there are clear lessons for policy-makers; improving the distributionalequality of educational outcomes is as important as raising average skill levels.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 120

Table of contents

1. The macrosocial benefits of education 123

1.1. What are macrosocial benefits? 123

2. Crime in comparative context 126

2.1. Introduction 126

2.2. Individual theories of crime – the rediscovery of the offender 126

2.3. Cultural, ecological and grand theories of crime 128

2.4. Indirect effects of ET on crime: unemployment, community effects and inequality 131

2.5. Conclusion 132

3. Social cohesion: values, trust and tolerance 134

3.1. Defining social cohesion 134

3.2. Trust 136

3.3. Tolerance 137

3.4. Social cohesion and VET 141

4. Active citizenship, civic and political participation 144

4.1. Conclusion 147

5. Evaluating the macrosocial benefits of education, vocational education and training 149

5.1. Is it possible to evaluate the macrosocial? 149

6. The macrosocial benefits of education: a preliminary investigation 152

6.1. Education and social cohesion: a macrosocial approach 152

6.2. Correlation between education and social cohesion measures 154

6.3. Educational inequality and social cohesion 154

6.4. Education and income inequality 156

6.5. Income inequality and macrosocial outcomes 158

7. Macrosocial benefits through microsocial data? 159

8. Conclusions 162


Annex 1: Macrosocial data 166

Annex 2: Details of SEM modelling 168

Bibliography and references 171

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 121


Tables

Table 1: Structural and cultural dimensions of social capital 146

Table 2: Pearson correlation coefficients and levels of significance for social cohesion aggregates 153

Table 3: Pearson correlation coefficients and levels of significance for mean level of upper secondary

attainment and social cohesion aggregates 154

Table 4: Mean proficiency scores and test-score ratio for countries in sample 155

Table 5: Pearson correlation coefficients and levels of significance for distribution of educational

attainments and social cohesion aggregates 156

Table 6: GINI coefficients (mid 1990s) 157

Table 7: Pearson correlation coefficients and levels of significance for distribution of income and social

cohesion aggregates 158

Table 8: Pearson correlation coefficients and levels of significance for distribution of income and social

cohesion aggregates with controls for GNP/capita 158

Table 9: Raw weighted means and standard deviations (in brackets) within each country 161

Table 10: Macrosocial aggregates for 15 countries 166

Table 11: Descriptions of, and goodness of fit indices for models 1 to 5 169

Table 12: Descriptions of, and goodness of fit indices for models 6 to 11 169

Figures

Figure 1: A coherent syndrome? Civic participation and general trust at the national level 154

Figure 2: Educational equality and general trust 156

Figure 3: Educational inequality and income inequality 157

Figure 4: Effects of education on social cohesion in four countries (non standardised and standardised

regression coefficients) with controls for socioeconomic status and age 161

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 122

In this report we examine what we call the macro-social benefits of education. The question of delim-iting between micro- and macrosocial benefits isdifficult. For many economists and social scientists,steeped in the traditions of methodological individu-alism, there is no macrosocial, and any benefitsderived at that level are simply the aggregation ofmicrosocial benefits. This view in extremis can befound in the ‘Austrian school’ economists’ belief thatmacro aggregates are not meaningful, being simplydescriptive compositions of heterogeneous micro-processes. This view is the antithesis of that of theo-rists such as Lockwood (1992) who reject the idea ofreducing macrosocial phenomena to the behaviourof individual actors (Mortensen, 1999). Here we takethe view that societal effects are, at least in part, irreducible to individual level phenomena sinceunderstanding at the macro societal level may requireanalysis of the effects of some structures and charac-teristics which are integral to the collectivity or societyitself, and which have meaning only at that level.

In terms of what might be called the microsocialbenefits of learning, there is much evidence thateducation impacts upon individual health, propensityto commit crime and general quality of life(McMahon, 2000). Much of this evidence is groundedin the theoretical perspective of human capital, in thatindividuals are theorised to make decisions on thebasis of present and future costs and benefits in orderto maximise their net economic welfare. This wouldinclude any non-monetary benefits of education, whatBecker (1993) refers to as psychic income, in additionto monetary benefits. Blaug specifies the contempo-rary definition of human capital theory to includenon-monetary benefits: ‘People spend on themselvesin diverse ways, not only for the sake of present enjoy-ment, but also for the sake of future pecuniary andnon-pecuniary returns.’ (Blaug, 1992, p. 207).

1.1. What are macrosocialbenefits?

Many of the social benefits of learning can belocated in a microeconomic or microsocial frame-

work and are embedded in human capitalconceptions of education. However, when weprogress to wider units of analysis such as thefamily, community and nation, the limitations ofsuch an individually-based framework becomeapparent. This is not only in terms of whether thenature of the benefit can be reduced to individualdecision-making, but if it can be expressed as anidentifiable individual benefit. In some ways,macrosocial benefits may be thought of asecological in that they provide a framework inwhich other microsocial benefits may beenhanced or diminished.

There are three primary features of macrosocialbenefits of education to consider. First, theypossess the characteristic of non-identifiability.Many micro-social benefits, such as improvedhealth or crime reduction, can be attached to anindividual, or a community. Macrosocial benefitsdo not have the property of identifiability in thatthey cannot necessarily be attached to a person,or community at lower than the national level(although their impact may be felt at this level).For example, social cohesion may be measur-able, or at least proxied, but it is not possible toquantify the social cohesiveness of an individual.In economic terms, macrosocial benefits are apure public good as the benefits derived arenon-rival and non-excludable.

Second, they are often relational or positionalin nature. Benefits such as improved literacy canbe expressed in terms of one individual. However,educational equity in terms of the distribution ofeducational test scores cannot. They may beexpressed only in terms of relations between indi-viduals and, in this sense, may be thought to bemacrosocial in nature.

Third, they are system level or social integra-tion benefits (Mortensen, 1999). Macrosocialbenefits cannot necessarily be aggregated fromother social indicators, although their propertiesmay well be related to microsocial indicators. Forinstance, individuals’ answers to the general trustquestion in the world values survey (WVS) (1) do

1. The macrosocial benefits of education

(1) ‘Do you trust people in general?’

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 123

not necessarily represent societal trust, althoughthey can be used as a rough proxy for it. Obvi-ously, there will be measurement error as indi-vidual perceptions of trust are not the same asthe actual level of trust. However, even given thiserror, the level of social trust comparativelyspeaking is more than the aggregation ofexpressed individual trust as it comprises histori-cally and culturally derived norms of trust which,although they determine individual responses andare captured in their measures, will deviate fromthose in other countries.

We can use human capital, social capital andsocial cohesion to illustrate arguments of identifi-ability, position and level. Human capital is clearlyidentifiable – the benefits can be attached to anindividual, although it may make some sense todiscuss the skills base of a nation. It is partly rela-tional in that there is a relationship between anindividual’s relative human capital stock and hisor her access to resources such as politicalpower (Nie et al., 1996) or employment asevidenced by the burgeoning literature on overe-ducation. Human capital can be aggregated to anational skills base, although whether this skillsbase has macrosocial properties separate fromthe individual stock of capital is questionable. Sohuman capital accumulation is not necessarily aclear example of a macrosocial benefit as bene-fits are primarily realised at the individual level.However, recent developments in the economicsof education such as endogenous growth theoryand on the social benefits of education such asdemocratisation (McMahon, 2000) indicate thatthere is potential for an engagement withmacrosocial issues by human capital theorists.For example, endogenous growth theorists relatehuman capital accumulation to economy-wideinnovation and lower crime (Sianesi andVan Reenen, 2000).

It is not as clear that social capital, defined byPutnam as ‘social networks and the norms ofreciprocity and trustworthiness that arise fromthem’ (Putnam, 2000, p. 19) is a microlevel benefitof education. The ability of social capital to spanmicro-, meso- and macrolevels of analysis ismuch noted by its advocates (Lin, 2001) (Baron

et al., 2000). Social capital cannot necessarily beattached to one individual (although Glaeseret al., 2000, attempt to distinguish between indi-vidual and community social capital) and from thewritings of Putnam (1993, 2000) it is clearlyecological in nature. There are relational or posi-tional elements of social capital in terms of itsdistribution or individuals’ access to social capital,although it is rarely expressed in these terms.However, the role of social capital in system func-tioning or integration is not clear. In certain cases,the formation of civic associations and trust maywork against forces for social cohesion (Skocpoland Fiorna, 1999) (Green and Preston, 2001).

Although it is not as clearly defined as eitherhuman or social capital, social cohesion can beconsidered to be a macrosocial benefit of educa-tion. It is clearly non-attributable to individuals orcommunities (although individuals or communi-ties may have characteristics conducive to it). Ithas relational or positional properties in that thepositioning of various groups within society isinstrumental in producing social cohesion (2).

In this report, we begin by examining themacrosocial benefits of VET, and education andtraining (ET) more generally, by examining threerelated sets of literature. First we review literaturerelated to education, crime and the formation ofdelinquent attitudes. Crime can be considered tobe an important element in the potential for socialcohesion. In particular, certain types of crime,such as hate crime, may be considered particu-larly relevant to the struggle for integration. Next,we examine literature related to education, socialcohesion and the formation of values such associal trust and tolerance. This is diverse. And inparticular, work relating to the influence of educa-tion on social cohesion is often of a qualitativenature. It is therefore often difficult to opera-tionalise and test hypotheses related to learningand social cohesion within a positivist framework.We then move to consider the impact of educa-tion on the formation of an active citizenry interms of civic and political life.

In the last part we examine how far quantitativeand qualitative techniques can be used to inves-tigate the impact of education on the macrosocial


(2) Social cohesion is also synonymous with social integration and system maintenance, at least in functionalist theories of society(Parsons, 1951). Parsonian or neo-Parsonian views of societal functioning are regarded critically by Marxist theorists such asLockwood (1992) who stress the role of structural contradictions within capitalism as the basis for social conflict or socialdisintegration.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 124

benefits of learning. We critically overview litera-ture related to the comparative method, econo-metric techniques (such as the social rate ofreturn) and evaluation methodologies. We thenpresent two related pieces of quantitative work toinvestigate how far different methodologies

(aggregate analysis and microsocial analysiswithin a comparative context) can be used toascertain the impact of education on macrosocialbenefits. We also test some of the hypothesesraised in this report related to the role of educa-tion in comparative context.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 125

2.1. Introduction

We begin our report by examining the literatureon the impact of ET on crime. As with allmacrosocial indicators, comparisons of crimestatistics are troublesome in comparativecontext. Differences in legislation, recording andeven cultural differences in the perception ofcriminal activity mean that cross-nationalcomparisons should be made with caution.

There is little evidence that the collection ofcross-national crime statistics has improved overtime (Jousten, 1998, p. 282). One frequently citedexample of a measurement problem is that thereis less variation between countries in victimisa-tion rates (self-reported crime) than in police dataowing to differences in how the police define,handle and count offences (Killias and Aebi,2000, p. 45). Particular problems occur inresearching criminal subpopulations such as drugusers where there are ‘rare and hidden subpopu-lations …. hard drug users are difficult to reachand not very willing to co-operate’ (Ødegård,1998, p. 358).

Measurement problems are of a cultural as wellas a legalistic nature. Even given differencesbetween actual crime, reported crime, policerecords and judicial interventions there is a socialdimension to perceptions of crime (Jousten,1998, p. 283) which can be seen as a driver ofnational policies, including those concerningeducation or resocialisation of offenders. Garland(2000) identifies what he calls a culture of highcrime societies in the UK and the US and alsopotentially in other Northern European countriessuch as Germany. This is not necessarily a recentphenomenon. Historically public perceptions ofcrime have partly driven crime policy (Waltonet al., 1999).

Competing definitions and interpretations ofcrime are rife in literature which, given its multi-disciplinary nature, makes a generalised assess-ment of the effects of ET on crime difficult. Inter-nationally there have been attempts to collectcrime statistics with cross-cultural differences inmind. For example, the European sourcebook on

national crime, the International crime victimisa-tion survey (ICVS) and to a lesser extent Interpolstatistics provide a basis for internationalcomparisons. Moreover, qualitative work, or thatwhich examines trends (rather than absolutelevels of crime), enables us to understand in partthe effects of ET on crime at an aggregate level.

Given these qualifications concerning measure-ment and interpretation, we now turn to theresearch evidence connecting ET to crime. We haveidentified two broad areas of research: individualexplanations of the ET-crime relationship; andcultural, ecological and grand theories of crime.

2.2. Individual theories of crime –the rediscovery of theoffender

There has recently been a return to individualexplanations of crime phenomena or a redis-covery of the offender (Kaiser, 1997). Theseapproaches eschew grand theories of crime andtheories which emphasise cultural rather thansituational factors. Most influential in the currentcriminological literature is Gottfredson andHirschi’s (1990) ‘general theory of crime’ which,although using a single explanatory mechanismcalled ‘control theory’ to explain individuals’propensity to commit crime, involves whatGottfredson and Hirschi (1990) call a culture-freeperspective. They argue that: ‘cultural variabilityis not important in the causation of crime, [and]that we should look for constancy rather thanvariability in the definition of and causation ofcrime, and that a single theory of crime canencompass the reality of cross-cultural differ-ences in crime rates’ (Gottfredson and Hirschi,1990) (quoted in Vazsony et al., 2001).

Individuals who are inadequately socialised inearly life and fail to bond with their parentsthrough lack of adequate family structures arebelieved to lack self-control as adults. This lackof self-control is thought to result in antisocialbehaviour and crime across all categories. As

2. Crime in comparative context

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 126

self-control is formed in early childhood, there islittle role for formal education in influencing this,except where educational interventions may beapplied later after identification of supposedlyinadequate family structures. Other develop-mental approaches to criminology, such as crim-inal careers research, are similarly scepticalregarding the role of later education in addressingbio-social and early childhood antecedents ofcrime.

Although there is a substantial number ofnational studies which support the tenets ofcontrol theory (Lanier and Henry, 1998,pp. 164-165), there has been little cross-culturalvalidation or testing of the concept. Most of whatthere has been has involved studies of youth anddelinquency or crime-analogous behaviours suchas deviance rather than more serious adultcrimes. However, in a study of adolescents inHungary, the Netherlands, Switzerland, and theUS (Vazsony et al., 2001) the reliability ofself-control and a scale for deviance (the norma-tive deviance scale, NDS) is shown to be withinacceptable limits for all countries. Moreover,self-control is the most powerful predictor ofdeviance, with differences between countriesaccounting for only 0.6 % of the variance. Thisimplies that most variation is within countriesrather than between countries (although theoutcomes are crime-analogous behaviours ratherthan crime as legally recorded).

While control theory and other developmentalperspectives emphasise the role of individualdifference as a cause of crime, routine activitiestheory (Felson, 1999) stresses the supply ofopportunities for crime. Routine activities theoryis an economic theory of crime which argues thatcriminal decisions are part of the category ofmore general economic behaviour based onrational perceptions of the costs and benefits ofcriminal activity (Becker, 1968), although strictassumptions regarding the degree of individualrationality are not necessarily supported incontemporary economic criminology (Lanier andHenry, 1998, pp. 76-77). A lowering of the costsof crime, in terms of a greater availability of crim-inal opportunities or reduced punishments,produces a greater amount of criminal behaviour.The recent surge in mobile phone theft in the UKcould be seen to be explained by such a theory

given the increased use of these phones andhence opportunities of theft.

A routine activities approach is employed byKillias and Aebi (2000) in an analysis of Europeancrime trends from 1990 to 1996. Compared withthe US experience, there has not been a signifi-cant decline in European crime trends in recentyears although aggregate levels for most crimesare still higher in the US. Despite having demo-graphic trends similar to those in the US, mostEuropean Union (EU) countries saw rises incrimes of all types, with a particularly sharp rise inproperty crimes and drug offences. According toKillias and Aebi (2000) there is no need to utilisegrand theory, involving concepts of anomie ordemography (which, in any case, are reasonablyconvergent across the EU, and not substantiallydifferent from those in the US) to explain thesetrends. Rather, the opening up of markets acrossEurope and wealth inequalities between East andWestern Europe have offered new markets forstolen goods and new supply lines for drugs.Killias and Aebi (2000) do not suggest thateducation could play a major role in the contextof this recent increase in crime. However, theymention that a lack of educational and labourmarket opportunities for migrants in the EU andEastern European youth in general may havefacilitated this process (Killias and Aebi, 2000,p. 52). This signals a possible role for targeted ET.

The lack of emphasis which both develop-mental (namely control theory) and economic(namely routine activities) theories of crime placeupon formal education is borne out in nationalempirical literature. There is little evidence thateducational level has an influence on an indi-vidual’s propensity to engage in crime indepen-dently of other factors (Witte, 1997), althoughlevel of education has a powerful effect on otherfactors, such as income, which are related. Othereducational outcome measures show a directcorrelation. Time spent in education (Witte, 1997,p. 227) is associated with crime (but note thattime spent in education is strongly correlated withsocial class), as is early childhood education(Rutter, 1994). Educational failure (dropping out)is also related to crime (Lochner and Moretti,2001 reviewed in Feinstein, 2002).

Furthermore, there is significant evidence thatdelinquency is linked with poor performance atschool or early school leaving (Friday, 1980,


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 127

pp. 117-120). However, there is little evidencethat delinquency necessarily leads to crimeexcept for those individuals already at risk ofcriminal behaviour through poor early socialisa-tion (Farrington and Loeber, 1999).

This relative neglect of ET or educational levelis not true of all individual-orientated theories ofcrime. In the life-course approach, such as thatadopted in the Tübingen studies (Kerner et al.,1995) ET is one possible road out of criminality(Kaiser, 1997, p. 373). Laub and Sampson (1993)perceive training and work as possible turningpoints in the criminal life course through socialcontrol and the creation of informal social bonds.The life-course approach has also been used toassess the impact of education on victimisation(the probability of being a victim, rather than aperpetrator of crime). Using reconstructed longi-tudinal data from the Netherlands, Wittebroodand Nieuwbeerta (1999) find that those who areof higher education and status are more likely tobe victims of violent victimisation, particularlyrobbery and property crime. As they are lesslikely to commit crime, though, this has someeffect on reducing their chances of being a victimof violent crime.

Individualistic criminological theories have littleto say (either theoretically or empirically) about therole of ET as opposed to early and targeted inter-ventions (Witte, 1997, pp. 220-233). They havealso been challenged regarding their methodologyand socially authoritarian potential. Haines (1999)criticises criminal careers and life-course researchas being based on small, unselective sampleswhich quickly become less relevant for policyconcerns as respondents age. For example, thethree major UK birth cohort studies started withsamples born respectively in 1946, 1958 and1970. This means that even in the latest cohortmost individuals would have progressed throughformal education in the years 1975-86 with VETfollowing this period. The implication is that anumber of important UK policy changes in ET,such as reforms in the qualifications system andthe introduction of internal markets in ET provi-sion, could not be tracked in this cohort. For theolder cohorts (those born in 1946 and 1958) thepolicy context in which they undertook theireducation is even further removed from theconcerns of contemporary policy-makers.

In addition Haines (1999) follows Garland’s

(2000) contention regarding the manner in whichcultural perceptions of crime are reified as socialfact. There is little in individualistic theories ofcrime, other than in the life-course approach,about rehabilitation in adulthood. Policy-makershave capitalised on the more authoritarianaspects of these theories with an emphasis onearly identification (youth crime policies, identifi-cation of potential juvenile offenders by teachersin primary schools in the UK) and incarceration(zero tolerance in the UK, three strikes policy inthe US).

Therefore, the role of ET in mitigating criminalactivity is seldom identified in individual theoriesof crime. As we have shown, there is littleresearch evidence linking ET to individual propen-sity to commit crime, independent of its indirecteffects (although there is more evidence on delin-quency and crime-analogous behaviour). More-over, the emphasis in these theories on the earlyformation of criminality seems to rule out the effi-cacy of later interventions. However, life-coursetheories offer some space where we may at leasttheorise a role for ET and there is some evidencethat increased training may (given the availabilityof labour market positions) offer a route out ofcriminality (Laub and Sampson, 1993). Here,though, evidence is tentative.

2.3. Cultural, ecological andgrand theories of crime

Unlike individualistic theories of crime, in cultural,ecological and grand theories the role of educa-tion systems and national educational strategiesis more apparent, and necessary. Althoughstudies of this nature tend to be based on a smallnumber of cases – if nations or regions are theunit of analysis – they enable us to move beyondcontext-free examination of simple causalitybetween education and crime. As Ødegård (1998,p. 365) states with regard to drug use: ‘For twodifferent nations one and the same characteristiccan generate opposite effects in the same waythat some people can become alcoholicsbecause of their milieu, while others can becometotal abstainers because of the same milieu’.

We will begin this section by examining theevidence relating national cultures and ETsystems to crime. We will then move on to


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 128

discuss the position of ET in relation to theformation of certain criminal subcultures, namelythose which perpetrate hate crimes and footballhooliganism. Finally, we will examine the indirecteffects of education on crime through meso andmacro level mediators such as community,employment levels and income equality.

The importance of culture and institutions inunderstanding relationships between crime andother variables, such as education, is explored byJunger-Tas (2001) in her report of the preliminaryresults from the International self-report delin-quency study (ISRD) which involves surveys ofyouth aged 14-21 in 13 countries including 11 EUMember States. Although there is similarity in therelationship between self-control and delinquencyin different countries, there are also importantnational differences. For example, in England andGermany father absence was associated withhigher delinquency, whereas in Nordic countriesthis was not the case. This is possibly due todifferent welfare arrangements between countrieswhereby single parent families receive moresupport in Nordic states (Junger-Tas, 2000,p. 323). Similarly, whereas there was a relationbetween large peer groups and delinquency insome countries, this was not the case in southernEurope where, arguably, these are often the norm.This reveals, again, the importance of differentsocial and cultural contexts. With regard toeducation, Junger-Tas (2001) reports from acontinuing survey of Dutch youth that shows thatschool achievement and parental supervision areimportant factors in reducing juvenile delinquency.However, for ethnic minorities youth parentalsupport is greater and they are more likely to bevictims of physical abuse than Dutch youths.

In another example, Eisner and Wikstrom (1999)compare violent crime in Stockholm and Basel.They find that the temporal and spatial patterns ofcrime are similar, occurring more frequently in theevenings and early mornings, particularly in areaswith high social disorganisation resulting frompoverty, unemployment and transient populations.However, there are differences in the levels andtypes of crime, with violent crime particularly highon weekend nights in Stockholm. Eisner andWikstrom (1999) suggest that this is due to the

cultural norms amongst young men in Stockholm.However, they find that education has a perverseeffect on the rate of violent crime in Basel, with thepercentage of university students in each districtbeing positively correlated with the rate of violentcrime (an effect which may be due to the nature ofundergraduate fraternities there?). This effectpersists even when controlling for indicators oflocal deprivation, although other educationalcontrols (such as the number of educational estab-lishments in a district) are not used.

The relationship of European ET systems tojuvenile crime is further discussed by Estrada(1999). Estrada distinguishes between twomodels of post-war juvenile crime – one in whichthe trend has been a general and continuingincrease (in England, Finland, Germany (3)) andone in which it levelled off in the 1970s (in Austria,Denmark, the Netherlands, Norway, Scotland,Sweden, Switzerland). Given these differentialtrends, Estrada does not consider that eitherroutine activities or control theory are suitableexplanations. The availability of crime opportuni-ties did not necessarily differ from country tocountry. Indeed, most countries adopted a puni-tive approach to juvenile crime prevention whichwould not explain the divergent trends (Estrada,1999, p. 37). In addition, those factors whichwould suggest a deterioration in family func-tioning, such as increases in the divorce rate,occurred mainly from 1965 to 1975 which is notconsistent with the increases in crime from 1950to 1965 nor with the levelling off in juvenile crimewhich occurred in most countries thereafter.

Estrada argues against an individually-basedexplanation of crime and in favour of one basedmainly on the social control functions of ET.According to him, the segregation of youngpeople from adult society through ‘the increasinglength of educational careers, a later and laterentrance into the labour market and the growth ofa youth focused popular culture’ have beenresponsible (Estrada, 1999, p. 38). This processhas occurred at various times and with variablerates of consistency across Europe and this, heargues, explains the variation in crime trends. Forexample, the crime trend in Sweden can, heargues, be explained by the fact that


(3) To construct a German time series, Estrada (1999) uses crime statistics for that part of the country known as West Germanyup to 1989.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 129

post-compulsory education and late entry into thelabour market increased dramatically from 1950 to1960, peaking in the late 1960s and stabilisingthereafter. To support this point Estrada citesself-report evidence from both Denmark andSweden which suggests that youths are actuallymore disciplined now than in the 1970s (Estrada,1999, p. 38). The fact that juvenile crime trendshave not levelled off in some countries can beexplained by the continuing extension of educa-tion and youth separation from the labour market,according to Estrada (1999, p. 38). However, thisis somewhat belied by the rise in youth crime inGermany where the ET system is generallybelieved to provide a highly structured entry intoadult life (Brown et al., 2000). In addition, it wouldbe interesting to know whether changes in thesegregation of youth correspond with changes inother social indicators such as the distribution ofskills or income inequality.

In his long-range sociohistorical study of Euro-pean homicide rates, Eisner examines the deter-mining role of education systems in the light ofNorbert Elias’ theory (1978) on the effects ofmodernisation in engendering self-control (Eisner,2001). From a variety of historical records, Eisnerconstructs a time series data set for Belgium,England, Germany, Italy, the Netherlands,Switzerland, and the Scandinavian countries fromaround 1200 to the present day. Althoughcautious of measurement issues, he is able toidentify a downward secular trend in homiciderates where the ‘phases of accelerated decline[…] often seem to coincide with periods of rapidexpansion and stabilisation of state structures’(Eisner, 2001, p. 630). For example, the decline inSwedish rates coincides with the establishmentof centralised bureaucratic structures and thedecline in Italian rates with national unification inthe 1870s. Eisner uses Oestreich’s (1968, 1982)concept of social disciplining resulting from aperiod of state intrusion into everyday life in orderto explain these phenomena (Eisner, 2001,p. 631). In particular, he argues that ‘the expan-sion of literacy and schooling and early capitalistexpansion of work constitute independentsources of the disciplining process in the earlymodern age […] Their effects on the structures ofthe self were both rigidly to enforce self-controland to provide the social and cultural resourcesfor a more orderly conduct of life’ (Eisner, 2001,

p. 631). Education did not act independently ofcontext but ‘these effects may have been partic-ularly penetrating among those groups and areaswhere intensified moral control by the church,expanded schooling, pervasive state structures,and work discipline intertwined into mutually rein-forcing power structures’ (Eisner, 2001, p. 631).At the same time he sees the rises in homiciderates which occurred at the end of the16th century and in the period around 1800 asreflecting social and cultural transformation inEuropean societies. Similarly, the rise in homiciderates since the 1960s may reflect the transitionfrom modern to post-modern society (Eisner,2001, p. 633).

These studies show how general theories ofcrime are limited by their blindness to culturaldifferences. Different cultural antecedents ofdelinquency operate not only for national popula-tions (Estrada, 1999) (Eisner, 2001) (Junger-Tas,2001), but also for subpopulations (Junger-Tas,2001) (Eisner and Wikstrom, 1999). We also seethe influence of distinct national cultural contextsin examining criminal subcultures.

The role of educational arrangements and insti-tutions within national cultural contexts isapparent in comparative work on criminal subcul-tures. For example, there has been recent policyand research interest in what are called hatecrimes in the EU. These are crimes involvingviolence against a specifically targeted racial ornational group which have specific implicationsfor social cohesion. According to Levin andRabrenovic (2001, pp. 584-585) various commen-tators argue that hate crimes are ‘more harmful tothe social fabric of society than comparablecrimes without a bias motive’. This is due tovictim interchangeability (victims are chosenbecause of their membership of a specific group,not because of prior actions or opportunity),secondary victimisation (that attacks against thevictims family and community escalate due tohate crimes) and escalation (that hate crimes mayescalate into large scale social conflict).

There are obvious cultural differences in termsof legal definitions, types and targets of hatecrime. In a study of aggressive youth cultures andhate crime in Germany, Watts (2001) contrastshate crime in Germany and the US. He arguesthat although the historical roots of hate crime aredifferent in both countries, the structural


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 130

antecedents of right-wing violence amongst theskinhead subcultures are similar across cultures.These antecedents are status anxiety, decline ofworking-class culture and unemployment due tomarginalised positions in education, labour andhousing markets (Watts, 2001, p. 612).

Similarly, Dunning (2000), in an analysis of foot-ball hooliganism as a world phenomenon, identi-fies cultural differences between the nature ofhooliganism but identifies similarities in structuralantecedents. In Belgium, for example,Van Limbergen et al. (1987) cite unemploymentand a short and frustrating school career asresponsible, whereas in Holland typical Dutchhooligans ‘tend to resent and resist formal educa-tion’ (Van der Brug, 1986, cited in Dunning, 2000,p. 160). Dunning states that hooliganism is situ-ated around fault lines in each country ‘inEngland, that means social class and regionalinequalities […] in Italy city particularism […] inGermany the relations between East and West’(Dunning, 2000, p. 161).

In all cases economic (and educational)inequalities are realised and expressed indifferent cultural forms of hooliganism. Althoughthe role of education in hate crimes and hooli-ganism is not direct, educational inequality(particularly in terms of lack of access to labourmarkets) can be seen as one part of various inter-related inequalities.

However, explanations of subcultures of crimewhich are orientated around working class orunderclass social and economic disadvantageare problematic. Crime statistics tend to overrep-resent working class crime. Moran (2000) withreference to UK and US data demonstrates howthe criminal justice systems over-record thenumber of working class men as perpetrators ofhomophobic violence (see also Lees, 1996 for afeminist critique of the construction of rape anddomestic violence as a predominantly workingclass crime). Additionally, working class crimesare often perceived to be a greater social problemthan so called ‘white collar crimes’. Robson(2000) examines how football hooliganism hasbeen identify as a problem among white workingclasses (at least in the UK) whereas statisticsdemonstrate it features a range of social classes(as conceded by Dunning, 2000). Such studiesmay cause us to treat direct causal relationshipsbetween economic and social disadvantage and

crime with due caution, although there is someevidence of these effects, as discussed in thenext section.

2.4. Indirect effects of ET oncrime: unemployment,community effects andinequality

What we have seen is that the influence ofeducation on crime is not direct and that culturalfactors and institutional arrangements are impor-tant. However, there is some evidence that thereare indirect effects of education on crime. Thelabour market advantages associated with ahigher level of education, namely a lower proba-bility of unemployment and a higher salary, maybe mechanisms through which ET has an impact.We have already explained how unemployment ispart of the social exclusion involved in hatecrimes and hooliganism. Indeed, at a microsociallevel of analysis, longitudinal studies reveal anassociation between early school leaving, unem-ployment and crime (Farrington et al., 1986)although crime may paradoxically rise in times ofhigh employment. Dunning (2000) shows thatfootball hooliganism in the UK was much higherin the 1960s when there was virtually full employ-ment when compared to the economic depres-sion of the 1930s when hooliganism was virtuallynon-existent.

Such paradoxes may be resolved by referenceto macroanalysis. At an aggregate level, the influ-ence of unemployment on crime operates in twoopposing directions (Beki et al., 1999). Firstly,unemployment reduces general economic activityand reduces the value of goods to be stolen. Thissuggests an opportunity effect which would leadto a negative relationship between unemploymentand crime. As low levels of economic activity arecredibly associated with lower levels of conspic-uous consumption and business activity, theopportunity and temptation for crime falls.Second, unemployed individuals may have moreincentive to steal as they have lower absolute andrelative incomes. Through a time series analysisof the Netherlands from 1950 to 1993, Beki et al.,(1999, p. 410) show that unemployment has anegative effect on the aggregates for most theftcrimes including burglary (the opportunity effect).


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 131

The only positive relationship between unemploy-ment and crime is in terms of fraud where theremay be an unsurprising motivation effect.

Possible area, or ecological, effects of educa-tion are also indirect. Social disorganisation theo-ries examine the influence on crime of variablessuch as poverty, ethnic heterogeneity and resi-dential mobility. In practice, proxy measures forthese variables such as indices of deprivation,ethnic mix and unemployment are used (Meschand Fishman, 1999). The use of these proxies hasled to some debate as to whether social disor-ganisation is a single characteristic of areas, orrather a cluster of unrelated variables. Mesch andFishman (1999) dispute that a latent variable forsocial disorganisation fully accounts for the directeffects of urbanisation and family disruption oncrime whereas Glaeser et al. (1996) contend thatthe quality and quantity of social interactions aremore important than social disorganisation. Interms of ET, social disorganisation theory may becriticised as neglecting the indirect effects ofeducation on values, as opposed to their labourmarket functions. For example, in socially disor-ganised areas the weak socialisation capacity ofET institutions are rarely examined explicitly(Bursik, 1998). Indeed, there is some evidencethat peer group effects are predictors of juveniledelinquency (Gaviria and Raphael, 2001),although again whether this leads to subsequentcriminal activity is questionable.

Rather than examine the intrinsic characteris-tics of communities in order to identify crimino-genic elements, a strand of research involvesanalysing the relative standing of countries andcommunities in terms of income inequality. Usingcross-national data, Braithwaite, J. and Braith-waite, V. (1980) show a statistically significantcorrelation between greater inequality of earningsand higher homicide rates across countries.Messner (1982) found that the extent of incomeinequality accounted for 35 % of the differencesin homicide rates among the 39 countries forwhich he had data.

Research based on US state level data alsosuggests a link between inequality and crime.Kelly (2000) shows that even controlling for otherfactors such as poverty, race and family compo-sition there are strong associations betweeneconomic segregation and crime. Lee (2000)shows how the spatial isolation of poor individ-

uals from the wealthy is a more powerfulpredictor of crime than the intrinsic properties ofindividuals and communities. He uses this findingto criticise researchers who attempt to identifyessentialist explanations of crime among AfricanAmericans, rather than examine relative inequali-ties. Kelly (2000) also indicates the potential forreplication of inequality and crime studies usingEuropean data. Examining inequalities and therelative position of individuals and groups mayenable us to investigate how far social disorgani-sation theories of crime have validity.

Econometric literature shows that income andeducation inequality are strongly associated(Nickell and Layard, 1998). Countries with widerdispersion of skills and qualifications, as we willshow later, also tend to have greater inequality ofincome. If income inequality is related to higherlevels of certain types of crime at regional andnational levels, it may be that societal levels of crimeare indirectly affected by education inequality.

2.5. Conclusion

In conclusion, there is relatively little quantitativeevidence at the individual level to support a directrelationship between levels of ET, or evenpost-compulsory education, and crime whenother factors, such as social class, are controlledfor. However, there are clearly indirect effects.

In comparative context, the effects of educa-tion systems (and particularly the relationshipbetween education and labour markets) are moreapparent and probably derive as much from thecontent and distribution of educational outcomes,as from the average levels in any given country.The effects of education on crime are highlymediated by their national context. Indeed, to acertain degree how we perceive crime isconstructed differently between nations(Garland, 2000). The macrosocial effects ofeducation on crime are perhaps best perceived interms of long periods of time (Eisner, 2001) andthere has been relatively little research conductedinto long duration relationships. However, we maymake a number of tentative generalisations.

First, the marginalisation of individuals fromlabour markets and the norms of society is one ofthe features of criminal subcultures across Euro-pean societies, although the form of this marginali-


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 132

sation differs between countries. Second, althoughthere are distinct cultural realisations of crime, thereare identifiable common structural antecedents.Although the relationship between unemployment,

social disorganisation and crime is unclear, there isemerging evidence that income inequality, and byimplication education inequality, is an antecedentof some types of crime.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 133

3.1. Defining social cohesion

Social cohesion is a concept with a long andcomplex history. All societies have beenconcerned with problems of social order and theirphilosophers have written extensively about themfrom Aristotle to Hobbes. During the 19th centuryan explicitly sociological approach to the problemwas developed which examined the forces, insti-tutions and values which hold – or fail to hold –society together. It might be said that social orderand social cohesion represented the definingproblems of the new discipline of sociologydeveloped by Comte, Saint-Simon, Durkheim,Spencer, Weber and Tönnies in 19th centuryEurope. The founding fathers of the new scienceof society (or secular religion as detractors werelikely to call it) concerned themselves with socialcohesion because they were aware that they livedin an era of rapid transition when traditionalbonds and ties were being eroded and where thecentrifugal forces of industrialisation and democ-racy could rip apart previous social connections.As Marx, contemplating the whirlwind of capi-talism, famously wrote: ‘All that’s solid melts intoair.’. We are currently living in a similarly transfor-mative age and ask similar questions.

The answers provided by the 19th century socialthinkers to the problem of social cohesion werevaried, as they are today. All noted that industriali-sation and the division of labour were transformingsocial and spatial relations from societies based onface-to-face community (what Durkheim called‘mechanical solidaritiy’) to some new form of orderwith more diverse and distributed social connec-tions. To Durkheim this meant the erosion of thecollective conscience and close-binding values oftraditional society and their replacement by newforms of organic solidarity based on the functionalmutual interdependencies created by the division oflabour. To Tönnies it meant the shift from societybased on community (Gemeinschaft) to societybased on contract (Gesellschaft). Such changeswere seen to be inevitable, but they did not guar-antee that social cohesion and order would prevail.For Spencer unfettered market relationships were

enough to hold society together, but for the conti-nental thinkers no such benevolent hidden handexisted. For Comte and Tönnies it was ultimatelyonly the state that could hold society together. ForDurkheim, who criticised Comte’s insistence onmoral consensus and both Comte’s and Tönniesreliance on the state, there had to be other forces,beyond market and state, which maintained cohe-sion, although he recognised that the state had animportant role to play in promoting core values ofmorality and meritocracy. In times of rapid transition,and particularly when technological change outransociety’s moral capacities for adaptation, patholog-ical social disorders arose which required new reme-dies. Primary among Durkheim’s candidates for thiswere the new intermediary associations of civilsociety that stood between the state and the market– most notably professional associations (Lukes,1973). Education also had a key role, and Durkheimbecame a key advocate of the Third Republic’s char-acteristic educational policy of promoting socialsolidarity through schooling. ‘Society,’ he wrote,‘can only exist if there exists among its members asufficient degree of homogeneity.Education perpet-uates and reinforces this homogeneity by fixing inthe child, from the beginning, the essential similari-ties that collective life demands.’ (Durkheim, 1977).

Durkheim wrote as a liberal socialist republican inlate 19th century France (Lukes, 1973), but his theo-ries left a complex legacy informing both left andright notions of social order and social cohesion. Inthe American liberal tradition, a particular strand ofDurkheim’s thought was appropriated by theschool of structural functionist sociology devel-oped by Parsons and Merton (Parsons, 1951). Thisstressed the idea of the market division of labourand functional interdependence in complexmodern societies as a source of self-reproducingorder but failed to address processes of change.Continental social democratic traditions, on theother hand, have placed more stress on the role ofstate and organised intermediary associations asthe basis of cohesion in modern societies. Indeed,it is hard to separate the idea of the modern welfarestate and social partnership from continentalconceptions of social cohesion.

3. Social cohesion: values, trust and tolerance

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 134

Both traditions have stressed in their differentways the importance of education to social cohe-sion. In Parsonian theory, schools have the vitalrole of ensuring efficient allocation of skills in thelabour market as well as being a major socialisa-tion agency for children into the key normativevalues of society, not least by promoting loyaltyto a meritocratic belief system which is taken tobe the main ideological cement of society. Socialdemocracy, and particularly the Nordic variety,has, on the other hand, placed more stress on therole of education in fostering social solidaritythough common experience and learning(Boucher, 1982).

It is fair to say that education has been given amajor role in traditional sociological accounts ofsocial cohesion. However, it should be stressedthat most theories accord equal importance tofull employment, welfare, crime, industrial rela-tions, community relations, national identity, andcitizenship. There is literature for instance on: thewelfare State (Mortensen, 2000), governance(Ritzen et al., 2000b), equity (Ritzen et al., 2000b)(Heynemann, 2001) and opportunity structures(Mann, 1999), value formation (Parsons, 1951),gender relations (Siim, 1999), crime and corrup-tion (Ritzen et al., 2000b) and industrial conflict(Mouzelis, 1999).

Classical sociological accounts construe socialcohesion as a macro societal issue. However,current policy debates rarely refer to classicalconceptions and theories and freely conflatesocietal aspects of cohesion with micro andmeso conceptions of social capital and commu-nity. The Canadian Policy Research Institute(1997, p. 2), define social cohesion as ‘theongoing process of developing a community ofshared values, shared challenges and equalopportunity within Canada, based on a sense oftrust, hope and reciprocity amongst all Cana-dians’, whereas Ritzen et al. (2000b, p. 6) write of‘a state of affairs in which a group of people(delineated by a geographical reason, like acountry) demonstrate an aptitude for collabora-tion that produces a climate for change’.

The use of terms such as trust, reciprocity andcollaboration provides a parallel with writing onsocial capital and, for some theorists, socialcohesion is little more than a special case ofsocial capital, whereby linking rather thanbridging social capital ties groups into the

nation-state. Implicit in the definition is the impor-tance of a sense of consensus, shared valuesand shared challenges in the formation of socialcohesion. Indeed, in recent writings on global civilsociety (Anheier, 2001), the social capitalmetaphor has been applied to the (trans) nationaldimension (Dasgupta and Serageldin, 2000).

The applicability of social capital theory or ageneral sense of community to this level ofaggregation may be questioned as themes ofstructural inequalities or how shared values suchas trust and reciprocity come to be arrived at insociety, are not tackled. For example, even aneoliberal perspective on social cohesion Ritzenand Woolcock (2000a, p. 6) finds it necessary toincorporate a macro political component.

Questions of macro relationships, equity andhegemony are central to an analysis of the impactof education on social cohesion. Social cohesionrequires more than neighbourhood or regionalstability (what has come to be called communitycohesion) or the inflation of communitarian idealsto macrosocial objectives.

Social cohesion, in terms of both values andmacrosocial outcomes, has been a long standingnational objective of both general education andET. According to Heynemann and Todoric-Bebic(2000, p. 161) the social cohesion function ofeducation is at the heart of each countries educa-tion system: ‘at the end of the 20th century,public schools are asked to perform more or lessthe same task as they were at the beginning ofthe 17th century – or trying, anyway’. Heynemannand Todoric-Bebic (2000) state that the meaningof social cohesion as an objective of nationaleducation systems is not uniform. In some coun-tries, such as the newly independent ex-Sovietstates, reducing public corruption and fosteringcivil society may be seen as primary social cohe-sion objectives. In others, such as Europe andthe US, ethnic and supernational identity may beconcerns (Hepburn, 1992).

Although social cohesion objectives varybetween countries, the importance of fostering(potentially) socially cohesive values such associal trust and tolerance is a clearly statedobjective in many national curricula and it is tothese areas which we now turn our attention.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 135

3.2. Trust

Trust is a notion which is frequently associatedwith both social capital and social cohesion.Social capital theory treats trust as a keyconstituent of social capital, seeing it as theproduct of iterated face-to-face interactionsbetween individuals engaged in common pursuitswithin associations and networks. In Putnam’saccount, trust is the outcome of association(Putnam, 2000), rather than the cause, althoughlittle evidence is cited to justify this interpretation(Green and Preston, 2001). Putnam also assumesfor the most part that trust within bounded asso-ciative groups spills out into a wider trustthroughout society as a whole. However, thismore diffuse form of societal trust may be betterthought of as a qualitatively distinct characteristic.A thinner trust which Newton (1999, pp. 17-20)refers to as abstract trust may be a more appro-priate analytical concept than general trust ininvestigating macrosocial benefits such as socialcohesion. Abstract trust is not necessarily basedon repeated face-to-face interactions, but oftenon the limited and sporadic contacts which takeplace continuously within modern industrial soci-eties. It also reflects trust in imagined, or empathiccommunities, such as trust with other Europeans,and therefore connects with notions of identity.

The basis of this abstract trust, according toNewton, may lie in education. He writes:

‘education which teaches the young to under-stand and operate the abstract principles of suchthings as trust, fairness, equality and universalism[…] Education also provides the disparate citi-zens of modern society with a common set ofcultural references without which daily under-standing would be impossible […] Education, it issaid, is what is left after people have forgottenwhat they have been taught. A willingness to trustand reciprocate may be among the things whichstick when all else has been forgotten.’ (Newton,1999, p. 18).

This socialisation and value formation functionof education is somewhat contrary to theresource-based function of education supposedby Putnam (2000), although it is central to bothDurkheimian and Parsonian conceptions (Morrowand Torres, 1995). Rather than educationenhancing personal resources, which in turn arethe antecedents of organisational membership

and eventually generalised trust, education in thisconception acts directly on higher-order trust inabstract systems. We discuss the socialisationfunctions of ET later in this section.

It may be additionally argued that in manycountries, and in certain areas of industrialisedcountries, ‘thick’ trust, defined in terms of inten-sive, daily contacts with community or familymembers, is of greater significance than thintrust. This concept of thick trust is proximate towhat Durkheim (in Giddens, 1972) refers to asmechanical solidarity (as opposed to organic soli-darity, which is analogous to thin trust) and whatTönnies (1957) refers to as Gemeinschaft.Although thick trust can probably not be used asan explanatory variable for macrosocial outcomesin advanced, industrial societies, pockets of thicktrust will exist among isolated, tightly-knit urbanand regional communities and ethnic enclaveswithin countries.

In addition to the various kinds of interpersonalor general trust discussed above, there are alsofurther measures of trust in institutions (institu-tional trust) and in democratic processes (demo-cratic trust).

Trust is generally considered an importantproperty for social capital, social cohesion andthe health of civil society generally (Almond andVerba, 1963). Consequently it has been measuredover a number of years across many countries,usually by asking respondents if they generallytrust other people and specified institutions.There has been some debate about whether thefirst question is construed by respondents interms of trusting in close friends and family ormore widely, but it seems likely that most peopleunderstand it in the second sense as intendedsince trust levels are extremely low in some coun-tries (10 % in Brazil on WVS figures). Aggregateresponses change over time within countries, butslowly, and there is a remarkable consistency inthe country levels in different surveys (Inglehart,1990); so it would appear that measurement isreasonably robust. In WVS and other surveysthere is also a striking difference in levelsbetween countries – from 70 % plus trusting inNordic countries to single figure levels in somecountries. This suggests that the question ismeasuring a fairly central and durable feature ofcultural life in different countries.

There is some debate about how closely these


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 136

various concepts of trust relate to each other.Social capital theorists claim that generalised trustis the basis for other forms of trust, particularlyabstract trust in systems, national and suprana-tional entities, but there is little comparativeevidence that types of trust are correlated withinor across countries (Prakash and Selle, 2001)(Norris, 2000). Even where correlations are foundto exist, for example between general and demo-cratic trust (Green and Preston, 2001), it is notnecessarily the case that organisational member-ships are the cause of higher levels of trust in bothcases as Putnam contends. There is an equallyvalid argument for claiming that the causationruns from trust to associational membership,rather than from membership to trust. Trust isseen as an important factor contributing to otherdesirable socioeconomic outcomes such aseconomic growth and strength of democracy(Norris, 2000). However, despite the strength ofthese relationships in contemporary advancedsocieties (Norris, 2000) over the longer historicalterm more complex relationships emerge. It maybe, for instance, that distrust and conflict betweenvarious interest groups have been instrumental inthe production of both European democracy andeconomic dynamism at various periods (Skocpoland Fiorna, 1999, p. 14). There is clearly an histor-ical and institutional element to trust which cannotnecessarily be identified within one country orthrough statistical analysis.

Evidence on the trends in levels of trust inmodern societies produces a mixed picture.Putnam provides substantial evidence thatgeneral trust is eroding for various differentgroups in the US (Putnam, 2000). Hall’s (1999)analysis of data for the UK between 1959 and1990 shows that levels of trust have declined forall groups defined by age, gender, class andeducation, although the decline within workingclass groups has been greater than amongmiddle class groups. Other studies, however,have found more mixed effects in different Euro-pean countries (de Hart and Dekker, 1999 for theNetherlands; Van Deth, 1999, for Finland, Norwayand Spain).

We have little evidence on the relationshipsbetween ET and trust at the macro level, and itmay well be that this depends very much on whattype of trust is being considered. ET may beparticularly effective, for instance, in promoting

thin trust in terms of abstract notions of generaltrust, fairness and universalism, but to substan-tiate this would clearly require more research. Aswe show later in this report, there is no significantcorrelation across countries between levels ofeducation and levels of trust, although there is astrong correlation between distribution of educa-tion outcomes and trust. Given that levels of trustvary hugely between countries, and rather morethan levels of education, we may assume thatother factors are involved in generating trust aswell as education.

3.3. Tolerance

Another commonly assumed component of socialcohesion is tolerance and, like trust, this is ahighly contested concept. Tolerance may beunderstood as acceptance of intragroup lifestyledifferences (permissiveness), or it may be under-stood as openness towards other cultures (as inethnic tolerance). These propensities may notnecessarily coincide. Equally there may be liber-tarian conceptions of tolerance as acceptance ofall values, no matter how abhorrent, which arequite different from liberal notions which acceptvalue differences but only where they do nottransgress certain core values. Libertarian atti-tudes may involve a general permissivenesstowards deviant majority group behaviour but maynot necessarily include attitudes conducive toethnic or racial tolerance. Research evidencesuggests that, at the individual level, in certainsocial contexts, education is associated both withmore permissive attitudes and with greater accep-tance of other cultures (Putnam, 2000) (Inglehart,1990). However, effects at the societal/nationallevel may be much more complicated.

Halman (1994) reports on the results of Euro-barometer surveys across Member States whichseek to gauge attitudes towards foreigners. In the1988 survey, 37 % of those surveyed thoughtthat there were too many people of a foreignnationality living in their country while 33 %thought there were too many of another race, and29 % too many of another religion. There weresubstantial differences in responses across coun-tries, but with responses relating to foreignnationality and other races covarying. The mostlikely to believe there were too many people of a


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 137

foreign nationality in their country were respon-dents from (in descending order) Belgium, theUK, France, West Germany and Denmark. Leastlikely (in ascending order) were those fromIreland, Spain, Portugal and the Netherlands.Most likely to believe that there were too manyfrom other races were respondents from (indescending order) West Germany, the UK, Franceand Belgium and least likely from (in ascendingorder) Ireland, Portugal, Spain and the Nether-lands. The Danish respondents were most likelyto be concerned about the numbers from otherreligions and cultures, but least likely to beconcerned about the numbers from differentsocial classes. In the 1988 data there is a closecorrespondence between the proportion in eachcountry believing that there are too manyforeigners and the proportion saying that theirlives were disturbed by their presence, although itwas the other races which were perceived asmost disturbing rather than the foreign nationals.

The 1993 survey shows considerable changesin levels of intolerance in a number of countries,with declines in West Germany but an overallincrease in most countries. Most marked werethe increases in Denmark, where the proportionsfinding the presence of foreigners disturbing rosein respect of other nationals (from 10 to 21 %),other races (from 13 to 20 %) and other religions(from 15 to 19 %). By 1993 the Danish respon-dents were far more likely than those in othercountries to be disturbed by those of another reli-gion (39 % as against 19 % in the next highestcountry – Belgium) and most likely overall to bebothered by people of different nationalities,races or religions. However, the European valuessurvey (EVS) – which asks respondents whetherthey dislike having different categories of peopleas neighbours – shows Danes to be the mosttolerant as regards such groups as drinkers, drugaddicts and political extremists, suggesting that itis quite possible to combine intragroup permis-siveness with closure towards foreign cultures.The EVS data for 1981 and 1990 show increasesin levels of tolerance in Ireland, the Netherlands,the UK and West Germany, and decreases inBelgium, Denmark, France and Italy.

Although these data suggest interestingregional variations in attitudes, with southernEurope coming off apparently better in relation totolerance than northern Europe, they may not tell

us much about how far different national groupsare, inherently or culturally prone to intolerance.Levels of discomfort with foreigners appear to bequite situational as they correspond closely toactual levels of immigration and to perceptions ofdifficulties arising from the presence of immi-grants. They also change rapidly from one periodto another, presumably in response to circum-stantial events – like unification in Germany,which initially seems to have had a positive effect– or to political climate shifts. They may, there-fore, tell us very little about whether one nationalpopulation will respond more intolerantly thananother to the presence of a given proportion offoreigners under similar circumstances. It shouldalso be noted that although the proportionsfeeling discomforted by foreigners have risenacross EU countries, the vast majority still saythat they are not disturbed by the presence. In asmuch as intolerance appears to have risen, andduring a period of rising levels of education, wemay conclude from this analysis that it is wise tobe cautious about assuming any direct effect ofaverage education levels on aggregate levels oftolerance. If there are such effects they may beoverwhelmed by other more powerful contextualeffects.

In a wide-ranging review comprising researchfrom several European countries, Hagendoornand Nekuee (1999) collected evidenceconcerning ET and racial tolerance based largelyon microdata in individual countries. According toHagendoorn (1999) there are two main causalmechanisms by which education may lead toincreased racial tolerance.

First, education leads to increased cognitiveskills involving enhanced abilities to categorise,understand causal relationships and perceivestates of the world. Hence individuals will beincreasingly able to understand that potentiallyracist statements, for example blaming migrants forunemployment, are based on faulty reasoning. Thesecond mechanism is through the formation ofracially tolerant values as part of socialisationthrough schooling. There is much researchevidence that years and levels of schooling have animpact on subscribing to racist views. Althoughthere is little evidence to suggest that particularinterventions or types of curriculum lead to a reduc-tion in racism (Hagendoorn, 1999, p. 5) there issome evidence that courses which stress individ-


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 138

uals’ critical capacities seem to have a greatereffect than other courses (Hagendoorn, 1999, p. 6).

Given the quantity of research evidence,Hagendoorn (1999) admits that there are anumber of paradoxes in education-racism litera-ture. For example, despite rising education levelsin the US and many education interventionsaimed at increasing racial tolerance, there isevidence that US youths are as racist as theywere after the Second World War. Hagendoorn(1999) explains that rising levels of education maysimply moderate the expression of racism.Although educated individuals may not wish tostate racist views in public (or in a survey) theymay be racist in their private lives and in infor-mally supporting discriminatory practices. Petti-grew and Meertens (1995) refer to this as a differ-ence between blatant and subtle forms of racismamongst those of different social classes andeducational levels. From a longitudinal survey ofevidence from the Netherlands, Verberk andScheepers (1999) show that those with interme-diate education are not likely to be blatantly racistbut are more likely to be subtly racist than thosewith lower levels of education.

The effects of education are also dependent onother factors, not least socioeconomic context. Forexample, those with low relative levels of educationmay face actual (or at least perceived) competitionfor unskilled jobs (Cox, 1970, pp. 392-422)(Roediger, 1991) (Hagendoorn, 1999, p. 3) or inhousing and community politics (Rex andTomlinson, 1979). In an analysis of pooled Europeandata (EVS), Jasinska-Kania (1999, p. 90) shows thatthe impact of education on racial tolerance is greaterin those countries with greater levels of immigration,whereas in countries with small proportions of immi-grants in the population the impact of education onracism is much smaller. This may be because morecircumstantially-driven racism provides moreopportunity for educational attenuation, whilehard-core (under-any-circumstances) racism isimpervious to educational mitigation.

Given the impact of actual or perceivedcompetition for economic resources on racism,we suggest that aside from values and cognitiveresources there is a third possible pathwaythrough which ET may effect racism – whatVerberk and Scheepers (1999) refer to as realisticconflict theory. As they argue: ‘A central assump-tion of realistic conflict theory is that socioeco-

nomic competition for scarce resources betweengroups such as ethnic groups leads to the forma-tion of negative attitudes of the other groups. Thecompetition may be concrete such as housing orlabour, or abstract such as culture, power andstatus’ (Verberk and Scheepers, 1999, p. 179).Cognitive resources may be implicated in thismechanism in that the source of resource conflict(unequal distribution of resources by the state orbusinesses) may be wrongly attributed tomigrants rather than state or business intereststhemselves.

Although little attention has been paid to real-istic conflict theory in mainstream research litera-ture, it has been a current of Marxist andneo-Marxist thought for some decades, at leastsince the black American Marxist Oliver CromwellCox developed these ideas in the 1940s (Cox,1970). Realistic conflict theory provides us with atheoretical framework for examining racism interms of structural inequalities (in which educa-tion is implicated) rather than in terms of indi-vidual moral and cognitive deficits. However, thetheory clearly has limitations. Historically, racismhas often developed among the most affluent andpowerful groups with limited reasons to feelcompetition over scarce resources with immi-grant or ethnic minority groups. They may,however, have been in exploitative relationshipswith members of these groups, in their positionsof slave owners, colonial administrators, orlow-paying employers, which it may have beenexpedient to rationalise through racist ideologies.Moreover, it is important to note that workingclass groups in potential material conflict withimmigrants and minorities will not always develophostile views. Within education in contemporaryEngland, for instance, working class studentsfrequently resist, rather than accept, racistdoctrines (Gillborn, 1995). This leads us to becritical of left realist perspectives which citeracism within the working class aside from amore dialectical view of class and race relations.

Given the three possible mechanisms by whichET may influence racial tolerance – value forma-tion, cognitive and realistic-conflict – we nowexamine the empirical literature on education andtolerance.

There are a number of national differences inthe influence of ET on values conducive to racialtolerance. In some countries (such as Italy), the


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 139

influence of education on tolerant values hasbeen found to be small and indirect (Peri, 1999),whereas in others (France, Germany) large effectshave been identified (Haegel, 1999) (Winkler,1999). Moreover, the relationship between racialintolerance and other personality characteristics(namely authoritarianism) differs from country tocountry.

Peri (1999) finds that in Italy the direct impactof education is small. The influence of educationon tolerance is indirect, operating through chan-nels of conformism, traditional values and profes-sional employment. However, Peri’s study is bothcorrelational and cross-sectional, so strict claimsof causality cannot be made. We do not know, forexample, if education or the family was impli-cated in developing conformism and traditionalvalues. In a French study, Haegel (1999) exam-ines the influence of education on both authori-tarian values and racial tolerance. She finds apositive association of education with tolerance,although the effects are weaker for those withvocational qualifications. Interestingly, there aredifferent levels of ideological consistencydepending on the individual’s level of education.For those individuals with low levels of education,there is little relationship between authoritari-anism and racial tolerance, although those insuch a position are likely to feel insecure aboutthe future. Although individuals with higher levelsof academic education are more likely to betolerant, those who are racially intolerant arelikely to hold authoritarian attitudes.

As Haegel (1999) shows, individuals withdifferent levels of education may exhibit differentclusters of values. She relates this to the Frencheducation system and sees the coercive flip-sideof the French model of assimilation as being therejection of certain other ethnic differences(Haegel, 1999, p. 34). Similarly, for Germany,Winkler (1999) shows that there are differentpathways for racism between individuals withdifferent levels of education, although he arrivesat different conclusions from Haegel concerningthe relevance of authoritarianism. Through struc-tural equation modelling, Winkler (1999, p. 126)demonstrates that there are different pathwaysexplaining racism for highly and less highlyeducated people. He suggests that socioculturalinsecurity, comprising right-wing views, nationalpride and authoritarianism, is a powerful predictor

of racism for those with lower levels of education,whereas for those with higher levels of educationauthoritarianism is not significantly related toracism. Winkler’s (1999) study additionallyprovides some support for realistic conflict theoryas sociocultural insecurity is a particular predictorof racism of those with lower levels of education.

There is a difference in the role which authori-tarianism plays in the formation of racism amongthose with low levels of education in the twocountries. In France, Haegel (1999) shows thatauthoritarianism does not correlate with racistvalues, whereas in Germany Winkler (1999)shows that there is a strong correlation. The rele-vance of this factor may result from the historicaldevelopment of racism in the two countries.

In terms of the cognitive mechanism by whicheducation influences racism, De Witte (1999)distinguishes between various forms of racism.He refers to these as general racism (negativeattitudes towards migrants), biological racism (abelief in the hereditary superiority of one’s ownrace) and cultural and economic racism (a beliefthat the cultural habits of migrants differ and thatthey expose nationals to resource competition). Itis this last form that De Witte refers to aseveryday racism as it is the least ideologicallyformed and most prevalent. Although everdayracism has shown little change in Belgium overtime, it is at a higher level than in other Europeancountries. De Witte (1999) contends that cogni-tive capacity is a strong mechanism in the reduc-tion of everyday racism. He argues that researchin Belgium (Gavaert, 1993) and the Netherlands(Raaijmakers, 1993) has shown that thosefollowing vocational courses are more likely toexhibit everyday racism and that this may be dueto the greater attention paid to cognitive skills inBelgian academic education. It might also beargued that this is a class effect, since thosefollowing vocational courses are likely to comefrom less affluent social groups and thus morelikely to perceive competition over scarceresources. However, De Witte (1999, p. 68) doesnot necessarily reject the socialisation function ofeducation and believes that there is a differencein the emphasis placed on values in the academictrack (De Witte, 1999, p. 69).

In the debate concerning the influence ofeducation on tolerance, there may be little tochoose between whether education influences


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 140

values or cognitive skills. Values are obviouslyimportant, as are national characteristics and thenature of the education system in each country.So too is the role of the curriculum in buildingindividual resources. It may be helpful, then, tosee values and cognitive resources as joint partsof a process of formation of racial tolerance.Sniderman and Gould (1999) see the process ofracial tolerance as the interaction of valuesacquired through socialisation, values invoked atthe moment of choice, and cognitive sophistica-tion. Education has an influence not only onlong-term value formation but also on the exer-cise of values at the moment of choosing whetherto express a racist opinion or action. In addition,reasoning is involved in both the long-termformation and short-term exercising of values.

To processes of cognitive and value change,we would add that the formation of values takesplace within a historical and cultural context. AsHalman (1994) shows, rising absolute educationlevels have not led to an increase in racial toler-ance. Education does not remove the individualfrom society – individual values are embedded ina social context. For example, the role of author-itarianism in the formation of racist values maydiffer from one culture to another. We must alsoremember that under certain historical conditionsthere may be a perverse relationship betweeneducation and tolerance, at least in terms ofsupport for intolerant regimes. As Abramson andInglehart remind us: ‘the assertion that educationhas some inherent tendency to instill democraticvalues does not stand up in historical perspec-tive. In Germany during the Weimar era, forexample, the National Socialists won studentelections in eight universities, at a time when theNazis won only 18 % of the vote in national elec-tions… Today, higher education does tend tosupport democratic values, but this relationshipreflects specific historical conditions and is notan automatic consequence of education.’(Abramson and Inglehart, 1994, p. 800).

3.4. Social cohesion and VET

We have shown that social cohesion is historicallyderived and culturally specific, involving equity,values and macrosocial actors such as the welfarestate. The specific role of education in bringing

about social cohesion in a society depends notonly on the level of qualifications, but potentiallyon the distribution of skills and opportunities aswell as the transmission of values. Moreover, therole of education systems is both historically andculturally situated (Abramson and Inglehart, 1994).Given the multiple and embedded functions ofeducation there is a limit to which we can makegeneralisations concerning its role across the EU,let alone more specifically in relation to VET.However, there are a number of qualitativecomparative studies of vocational education andapprenticeships which have interesting things tosay about value formation.

As Aldrich (1999) explains, the nature ofapprenticeship has historically involved the socialand legal integration of youth into society.However, this occurs in quite nationally specificways. For example, in the UK, apprenticeship isnow largely considered to be part of vocationaltraining whereas in France and Germany theroute to a Beruf or profession involves a morestructured process of social and legal transition,at least in theory. Young (2000) refers to thesedistinctions in terms of differences in assessmentregimes. In the UK, an outcomes-based systemof assessment does not involve the same type ofintegration into adult life as the institutionalapproach favoured by Germany.

The German dual system of apprenticeship isoften held up as an ideal model of the relation-ship between training and economic and socialintegration. As Green and Sakamoto (2001,pp. 69-74) explain, embedding the dual systemwithin a neocorporatist system involving work-place codetermination, sectoral agreements andother aspects of social partner regulation of workand training, has delivered not only high skills butan upgrading of skills and jobs throughout theeconomy. In terms of citizenship in the widersocial sense, the system also enables widercommunity acceptance of youth, aiding transitioninto adult life (Evans, 1998).

This emphasis in the dual system on integra-tion into both economic and social spheres isphilosophically underpinned by the work of thelate 19th century Bavarian writer GeorgKerschensteiner who was concerned with orien-tating education systems around both civicresponsibility and work. In Kerschensteiner’stheory, work schools are required to develop both


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 141

manual and intellectual skills. Students wouldlearn within work groups which would developthe basic rules for civic cooperation andcommunal life (Röhrs, 1993). Although Kerschen-steiner placed an emphasis on the duties (ratherthan rights) of citizens, the practice of thoseduties within the work school would be throughindependent, responsible work.

The contemporary German dual system stillmanifests this concern with civic responsibility; inits broad curriculum, encompassing preparationfor both work and citizenship, and in the way inwhich it seeks to provide a structured transitionfor young people into the highly regulatedGerman labour market (Brown et al., 2001). Theschool (Berufsschule) component of the appren-tice has a particularly broad mission includinggeneral education and occupational theory.According to the general 1991 framework agree-ment for vocational schools set out by the LandMinister and the BIBB, Berufschulen haveamongst their objectives:(a) ‘to impart professional competence,

specialised competence in conjunction withhuman and social capabilities;

(b) to develop occupational flexibility in order tocope with the changing demands of theworking world and of society, as well as havingregard to the growing together of Europe;

(c) to encourage preparedness for continuingand further professional training;

(d) to provide the ability and willingness to actresponsibly in terms of the individual shapingof one’s own life and in the public sphere’(quoted in Brown et al., 2001).

The final emphasis on the public sphere isindicative. However, despite these foundations,the dual system may also have certain character-istics which can have negative consequences forsocial cohesion.

In a comparative study, Evans and Heinz (1994)consider youth transitions through a comparativestudy of vocational preparation in England andGermany. They use the term active-individualisa-tion to describe an ideal type transition involving aprocess of self-determination and planning. Thisis opposed to passive-individualisation involvingweak specification of goals. On these groundsone would expect the German system of voca-tional preparation to be superior in its social inte-gration function as routes are clearly specified,

with clear links between employment and citizen-ship. In the UK youth entered the labour markettwo years before their German counterparts andprogression routes were poorly defined. This ledto some individuals in the UK reporting a lack ofintegration and a sense of powerlessness inattempting to gain employment or citizenship.However, although in Germany progression routeswere much clearer, for those individuals whoexperienced difficulties or dropped out, the lack offlexibility meant that it was difficult to achieve rein-tegration. The Evans and Heinz study (1994) indi-cates the difficulty of equating an historically wellfunctioning system of vocational preparation withone which delivers social integration.

Even in those systems of vocational prepara-tion which appear to deliver a smooth transitioninto adult roles and citizenship we need to beaware of the latent functions that deliver incomeand social equality. For example, the Germanapprenticeship system may perpetuate labourmarket inequalities, with girls and immigrant chil-dren typically finding places – and hence laterjobs – in only the lower status (Bynner, 1994)(Brown, 2001). The role of education (particularlyvocational preparation) in maintaining inequalitiesin terms of economic and cultural reproduction israrely referred to in policy discourses on socialcohesion, although it is central to current educa-tional theory (Morrow and Torres, 1995).

In a review of the European literature on therole of social class in the reproduction of educa-tional inequalities, Hatcher (1998) cites evidencethat in only two EU countries (the Netherlandsand Sweden) did social class inequalities ineducation decline between the First and SecondWorld Wars with only limited improvement forsome countries since the Second World War.Moreover, even in a country often cited as anexemplar of egalitarian educational and welfarepolicy – Sweden – there has been very littlemovement in the pattern of class inequality. Thepattern of educational inequality remainedreasonably constant in Sweden from 1970 to1990. However, although inequality of socialopportunity in Sweden is roughly the same asthat of other EU countries, welfare provisionmeans that there is lower inequality in terms ofstandards of living (Erikson and Jonsson, 1996aand 1996b) (cited in Hatcher, 1998). Thisevidence points to the intractability of class


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 142

inequalities and the difficulties of educational orVET reform more generally in addressing theseinequalities. This represents the counterpoint toeducational reform which seeks to address socialcohesion as an issue purely of increasing educa-tional access.

However, even if we should not expect VET toaddress class inequalities, there are opportunitiesfor such education to lead to active or critical citi-zenship. As shown by the ETGACE project(Education and training for active citizenship inEurope) work related ET may open up criticalspaces for discussion and dissemination of ideas.

The project’s case studies show the close linksbetween lifelong learning and various interlockingareas of civic life. This includes VET and theworkplace. For example, in Belgium a case isprovided of workers cooperative (De Wrikker)where the relationship between VET and citizen-ship occurs in terms of making choicesconcerning alternative conceptions and practiceof work (ETGACE, 1992). This particular notion ofactive citizenship as solidaristic and sociallytransformative differs substantially from thatoffered by many contemporary theorists, as weshall see in the next chapter.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 143

The position of an active citizenry as a macro-,rather than microsocial, benefit of education isunclear in empirical literature. Although there aremany studies which indicate that education isassociated with varied civic and politicalbehaviour (Emler and Fraser, 1999, provide auseful summary of this literature) it is not clearhow education is expected to impact on suchbehaviour. Most studies assume that educationhas a role in increasing the resources of individ-uals and that this leads to an increase in variousforms of participation. However, some studies areconcerned with the positional aspects of educa-tion (Nie et al., 1996).

Nie et al. (1996) examine the positional natureof political participation in the US. Through ordi-nary least square (OLS) regressions over timethey find that it is the relative, rather than abso-lute, level of education that is important in deter-mining access to network central positions andpolitical influence. As the general level of educa-tion increases, the value of each qualificationlevel in gaining network centrality and politicalinfluence declines. They use preliminary evidencefrom the EVS to indicate that the results of theirstudy may be generalised beyond countries otherthan the US. Although Nie et al. (1996) take intoaccount context in terms of educational level,there are few other studies which examineeducation for citizenship comparatively. In partic-ular, emphasis on microsocial or institutional casestudies has meant that there are relatively fewstudies concerned with the influence of nationaleducation systems or contexts on citizenshipoutcomes. Those studies which do exist aremainly historical, concerning the development ofeducation and the nation-state (Green, 1990).

One contemporary study that does take intoaccount the impact of national systems of educa-tion on citizenship is the IEA (International associ-ation for the evaluation of educational achieve-ment) citizenship study of 28 countries(Torney-Purta et al., 2001). This cross-nationalstudy of 90 000 14 year-olds attempted to ascer-tain processes and outcomes of citizenship

formation through qualitative and quantitativedata from students, teachers and schools.

Students demonstrated a basic knowledge ofdemocratic processes, although their under-standing was often at a superficial level, and apositive relationship between civic knowledge at14 and future preference for voting was identified.Many students also rejected conventional polit-ical routes in favour of non-violent political actionand collecting money for charities or environ-mental causes. Schools with democraticprocesses and an open climate were found to beparticularly effective in inculcating civic knowl-edge and activity in all countries.

The distribution of civic knowledge withincountries was not as unequal as the distributionof other educational outcomes such as mathe-matics or literacy. Although this could result fromthe nature of civic knowledge (there is naturallyless variation than in other types of knowledge) itcould also signify that civic knowledge is notnecessarily a process in which schooling plays amajor part (less than common accessible media).However, despite the narrow distribution of civicknowledge between countries, significant differ-ences were found between both knowledge andactivity. Although, in general, transition countriesand older democracies scored more highly oncivic knowledge, there are some interestingcontradictions within the results.

In terms of civic knowledge, the Czechrepublic, Finland, Greece, Italy, Norway, Polandand Slovakia were significantly above the interna-tional mean. Many of these countries have incommon a high level of reading literacy(Torney-Purta et al., 2001, p. 78), and this mayindicate a relationship between civic knowledgeand general cognitive skill. Belgium and Portugalwere significantly below the international civicknowledge mean. However, there is not neces-sarily a relationship between civic knowledge andcivic engagement, at least at the national level.For example, Portugal was significantly above themean in terms of belief that conventional forms ofcivic engagement are important (despite beingbelow the country mean for civic knowledge)

4. Active citizenship, civic and political participation

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 144

whereas the Czech Republic was below thecountry mean in terms of belief that non-conven-tional forms of civic engagement are important(despite being above the country mean for civicknowledge). This may reflect the respective histo-ries of the two countries. For example, in theCzech Republic transition to a market economyand democracy may mean that there is less needto support unconventional forms of politicalengagement in order to effect change. Interest-ingly, many of the Nordic countries (Denmark,Finland and Sweden, but not Norway) alsoscored below the mean in terms of support for allforms of political participation, which may indi-cate preferences for more consensual or institu-tionalised forms of political action or a perceptionthat there are not so many injustices to contest.

The data also show that there is not neces-sarily a relationship between country levels ofliteracy, civic knowledge and support for rightsfor women and ethnic minorities. Slovakia, forexample, scored significantly above the interna-tional mean in terms of civic knowledge, but interms of support for rights for women and ethnicminorities was significantly below the interna-tional mean alongside other transition economies(Bulgaria, Estonia, Hungary, Latvia, Lithuania).This result is not necessarily surprising given theconservative forms of nationalism which currentlypredominate in these countries (Brubaker, 1996).

There are various messages from the IEA civicparticipation study, but one which resonates withthis report is the importance of national context inexamining microsocial relationships. The indi-vidual level relationship between civic knowledgeand civic activity, for example, is well establishedin various studies (Emler and Fraser, 1999). Thisrelationship does not necessarily hold at anational level, however. For example, those coun-tries with high levels of civic and reading literacydo not necessarily have high levels of support forpolitical activity. This seems to imply that a

resource, or cognitively based model of politicalparticipation, is inadequate in explaining varia-tions in the level of political activity (rather thanthe relationship between learning and politicalactivity) internationally. It seems that regionalpatterns of civic knowledge, attitudes andbehaviour are important.

The importance of comparative researchbecomes clear when examining the effect ofeducation on civic participation in other studies.In particular, national differences in the causes ofcivic participation, in forms of civic participationand in the relationships between civic participa-tion and other values, such as trust, becomeclear. There are different causes and conse-quences of civic participation – education beingonly one possible route. The Putnamesque (4)model of social capital may be seen to be partic-ularly narrow when applied to countries outsideof the US (Prakash and Selle, 2001).

The sub-elements of what has come to becalled social capital, both structural (socialnetworks and civic participation) and cultural(localised and generalised trust), are not neces-sarily correlated at national level (Prakash andSelle, 2001). Norris (2000) provides evidence forthe lack of correlation between levels of associa-tional memberships and general trust (5) across47 countries using evidence from the WVS (6).Moreover, there is a strong tendency for thedistribution of trust and associational member-ships in countries to follow patterns which mightreflect underlying cultural values of the countriesconcerned, rather than a random distribution ofsocial capital. In terms of the distribution of asso-ciational memberships and trust Norris cate-gorises countries as belonging to one of fourtypologies (Table 1).

As can be seen in Table 1, not all countrieshave either rich or poor social capital. Many fallinto the mixed category with no positive correla-tion between trust and associational membership.


(4) For Putnam (2000) civic associations are the root of social capital formation. Individuals participate in associational life whichleads to the development of localised trust (in other associational members) and then to more generalised trust (in people asa whole). Hence general trust (and more universal social benefits such as democratisation) develop as a result of civic life.Although this somewhat simplifies Putnam’s (2000) argument, the emphasis on civic association as the key to general trust anda stream of social benefits is, we believe, the core of ‘Putnamesque’ social capital.

(5) As measured by the percentage of individuals answering positively to the question ‘Generally speaking would you say thatpeople can be trusted or that you can’t be too careful in dealing with people’.

(6) As measured by the percentage of individuals in each country who were a member of at least one voluntary associationincluding church and religious organisations, sports or other recreational associations, labour unions, professional associa-tions, charitable organisations and any other voluntary associations.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 145

We may have reason to suspect that the distribu-tion of countries to each quadrant is non-random– countries within each quadrant follow cleargeographical groupings. A Putnamesque explana-tion for this distribution would involve theorisingabout distributions of social capital in terms ofsimilarities in social and generational trends ineach country, for example, explaining low levels oftrust in Latin America on the basis of increasedtelevision viewing. However, this would notexplain the coexistence of low levels of socialtrust in Latin American countries with relativelyhigh levels of associational membership – asrevealed by Table 1. The two should comprisepart of a coherent syndrome and there is noreason to expect imbalance between the two topersist, particularly across a range of Latin Amer-ican countries. In the absence of the ability ofsocial capital to explain this phenomenon, wemay pursue alternative explanations such asincome inequalities, national culture or differentialimpacts of supranational phenomena, such asglobalisation. We may also speculate on thehistorical role of national education systems infostering these types of relationship – a point towhich we will return in our conclusion.

In Chile, for example, explanation of thrivingcivil society with low trust could be based uponthe recent history of a corrupt and dictatorialregime. As in former Eastern Bloc countries inEurope (such as the former German DemocraticRepublic) associational memberships may have

been a necessity for many citizens in securing abasic standard of living.

In common with other authors (Knack andKeefer, 1997) (La Porta et al., 1997), Norris (2000)also finds little correlation between the associa-tional membership component of social capitaland macrosocial outcomes. It appears that anyapparent relationships are driven by general trust.For example, although there are strong andsignificant correlations between social trust andvarious macrosocial benefits such as educationalenrolments, life expectancy, the human develop-ment index, per capita gross domestic product(GDP), economic growth, democratisation, polit-ical involvement, ownership of televisions, news-paper readership, and use of the Internet, onlyper capita GDP and internet usage are correlatedwith membership of voluntary associations(although interestingly, tolerance is also signifi-cantly correlated with organisational membershiprather than trust).

Aside from the difficulty in identifying coherentpatterns of relationship between associationalmembership, trust and macrosocial outcomes aswould be expected by Putnam, there are also issuesconcerning the meaning of these terms in compar-ative context, or even when they are investigatedwithin a national context, such as the US.

A point explored in part by Putnam (2000) isthat different types of membership and associa-tion may result in different macrosocial outcomes.Like any form of capital, social capital can be


Social capital – mid 1990s

Structural dimension(Associational activism)

Weak Strong

Mixed Rich social capital

(e.g. East Asian countries (e.g. Nordic countries Strong including China, Japan, Taiwan) including Finland, Norway,

Sweden, alsoCultural dimension West Germany)

(Social trust) Poor social capital Mixed

(e.g. Spain and acceding (e.g. Latin American Weak EU countries plus Bulgaria countries including Brazil,

and Turkey) Chile and Mexico, also perhaps the UK and the US)

Source: adapted from Norris (2000, p. 23)

Table 1: Structural and cultural dimensions of social capital

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 146

used for malign purposes – whether to excludeothers, practice intolerance or for criminal orterrorist purposes – what Putnam refers to as itsdark side (Putnam, 2000, pp. 350-363). However,Putnam also states that social capital withoutsocial mixing is better than no social capital at allas a second-best solution. Hence, separateschools, churches and associations are seen asbeneficial in building a fraternal society (Putnam,2000, p. 362), although these institutional divideshave been widely accepted as contributing tolack of community cohesion (Home Office, 2001)and even institutional racism (Gillborn andYoudell, 2000) in the UK. Such a binary divisionbetween useful (i.e. most) social capital and itsdark side is not necessarily helpful in under-standing the relationship between associationalmembership and social outcomes. Possibleconflicts between social capital and what wemight refer to as social cohesion are clear.

Prakash and Selle (2001) suggest that a subtlercharacterisation of associational groupings is morehelpful in attempting to ascertain the types of bene-fits produced. This may mean distinguishingbetween those groups with a political as opposedto a social purpose, between those who are delib-erative and interest based, between those with ahierarchical and egalitarian structure and betweenthose associations where membership is voluntary,ascribed or a hybrid of the two. Fukuyama (2000),on the other hand, makes a primary distinctionbetween forms of association on the basis theirdegrees of moral authority and ideological reach. Ina recent study of social cohesion and fragmentationin modern societies (Fukuyama, 2000) he considersthe changing nature of associational life and theapparent paradox, in America, of the coexistence ofrelatively high levels of association and growinglevels of distrust and social fragmentation. Theanswer, he says, ‘has to do with moral miniaturisa-tion: while people continue to participate in grouplife, the groups themselves are less authoritativeand produce a smaller radius of trust. As a whole,then, there are fewer common values shared bysocieties and more competition amongst groups.’(Fukuyama, 2000, p. 49). Whether or not one sharesFukuyama’s socially conservative analysis of thecauses of societal fragmentation, he has certainlypointed to a dilemma at the heart of social capitaltheory and one which underlines the importance ofsocietal explanations of social cohesion.

However, even if we distinguish between types ofassociation, we cannot necessarily ever under-stand their purposes or outcomes outside thenational sociopolitical and cultural context in whichthey are embedded. As Prakash and Selle (2001)explain, associations and their aims evolve in rela-tion to the development of political institutions,societal change and mobility. This means thatorganisations which may seem to represent asuboptimal form of social capital in one socialcontext – such as chequebook memberships in theUS – may lead to social benefits in others, such asthe UK (Maloney, 1999). In particular, informalnetworks and associations have been shown toresult in specific macrosocial outcomes. Forexample, Gundelach and Torpe (1996) find thatinformal, network-type associations have a greaterimpact on Danish attitudes than classical, formalpolitical associations of the de Toquevillian model.Additionally, Parry, Moyser and Day (1992) showthat ad hoc, rather than formally constituted, polit-ical groups are important in producing further polit-ical involvement and altering attitudes in the UK.

4.1. Conclusion

It seems that there are problems in applyingmodels which at the individual level imply a rela-tionship between education, training and skillsand citizenship benefits in a comparative context.First, as shown by exhaustive comparativestudies (Torney-Purta et al., 2001) there are notnecessarily relationships at the national levelbetween skills, knowledge and civic outcomes.There is considerable variation in civic outcomesbetween countries with those nations scoringmost highly on civic knowledge not necessarilybeing favoured with high levels of civic engage-ment. Indeed, there appear to be clear regionalgroupings, with a different group of countries withhigh levels of civic knowledge (the Nordic coun-tries and transition countries) as compared tothose with high levels of civic activity (a range ofEU countries, excluding the Nordic countries).There is also considerable variation in the rela-tionships between civic activity and other values,such as trust. Again, the story seems to indicatethat simple causality between education andmacrosocial outcomes is somewhat misplaced.

Our conclusions to this chapter are therefore


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 147

cautious. In signalling the role of specific histor-ical conditions on the relationship between ET,trust, tolerance and social cohesion more gener-ally, we would call for a reappraisal not only of theimportance of historical contexts but also of theconnection between structural relationships andsocial cohesion outcomes. Education may haveimportant effects on many of the outcomes underconsideration under certain conditions. However,many of the effects are indirect and conditional

on other – often more powerful – contextualdeterminants. To study these effects thereforerequires attention to time and place and theexhaustive analysis of a range of factors and vari-ables operating at the macro level. Quantitativecomparative analysis at the macro level will havesome value where the variables are sufficientlycarefully specified. However, much of the work ofexplaining complex interactions will require morein depth comparative qualitative analysis.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 148

5.1. Is it possible to evaluate themacrosocial?

In examining the impact of ET on macrosocialbenefits there are obvious limits to how far anysingle technique can hope to capture the range ofbenefits, their qualitative dimension and thehistorical and cultural context in which they areembedded. In this report, we have made refer-ence to various techniques for ascertaining themacrosocial benefits of learning, involving qualita-tive and quantitative techniques. It seems that amixed-methods approach to examining thesebenefits is required. For example, Eisner (2001)incorporates both statistical analysis and histor-ical studies in examining trends in Europeanhomicide rates over time. Here statistical data isused descriptively in the analysis of trends.Eisner’s approach enables us to explain the trendsinvolved with reference to factors other than vari-ation between individuals, namely differencesbetween education systems and other associatedentities involved in the formation of nation-states.Although such an approach is probably optimal, inthis report we use mainly statistical techniques inthe evaluation of macrosocial benefits. However infuture, the Centre for the wider benefits of learning(WBL) will develop the interrelationships betweenquantitative and qualitative techniques (Schulleret al., 2001).

When examining the macrosocial benefits ofeducation, training and skills there are limits tohow far the concept of evaluation can be used, atleast in its summative sense. It is perhaps helpfulto distinguish between three types of activity(Plewis and Preston, 2001): measuring, modellingand evaluation.

Measuring macrosocial outcomes is an activityengaged in by many international bodies such asthe EU, OECD and the World Bank. Accuratemeasurement is central to any evaluation butmeasuring alone tells us little concerning the rela-tionship between educational activities or systemsand these benefits. For example, we cannot

assume post hoc that an increase in life-expectancyin a country arises from increasing levels of educa-tion. To make such statements requires somepreliminary form of modelling. At its most basic, thismay involve descriptive comparison of aggregatesthrough scatter plots or correlations. Moreadvanced techniques such as multilevel modelling(MLM) or structural equation modelling (SEM) arealso employed. The extent to which these representanalysis of actual macrosocial aggregates ratherthan contextualisation of microsocial relationshipsis a matter for debate.

One method of modelling would be to examinemacrosocial and macro educational aggregates.Using time series data on educational levels,educational distributions, macrosocial aggregatesand appropriate controls one would model rela-tionships over time. This would allow speculationon causality – whether changes in education vari-ables cause changes in social cohesion variablesindependently of other influences. Time seriesdata on macrosocial and educational indicatorsover time is difficult to obtain, although McMahon(2000) constructs time series for a number ofcountries. In the EU, for example, measures ofskill distribution (derived from the Internationaladult literacy survey – IALS) or values (derivedfrom the WVS) are only available over short timeperiods, where causality would be difficult todetermine.

The ecological fallacy is often referred to as areason why such macrosocial analysis is inappro-priate. The ecological fallacy holds when thewrong units of analysis are considered in makingan interpretation of data. For example, that boththe mean level of education and the mean level oftolerance are high in a country does not indicatethat there is a relationship between educationand tolerance for all individuals in that country.Alternatively, that respondents in a country aregenerally more trusting does not mean that thereis a national culture of trust as a property of thatcountry – this is the fallacy of aggregation.

Although it is important to be aware of the

5. Evaluating the macrosocial benefits of education,vocational education and training

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 149

ecological fallacy, it is equally important to under-stand that the fallacy operates both ways. It is theimportance of choosing an appropriate unit ofanalysis, not the automatic acceptance ofmethodological individualism which is implied bythe fallacy. The aggregate level of trust elicitedthrough individual trust levels may be meaningfulin itself, and useful in analytical work. There arealso some indicators which it is impossible toexpress at an individual level such as skill distri-butions, income distributions, ethnic conflict,industrial disputes or government corruption.

Another technique would be to use microsocialdata to model the relationships between educa-tion and social outcomes in various countries andthen to compare effect sizes. There are variousapproaches which could be utilised here such asregression analysis or multiple comparison ofgroups using SEM. The interpretation of suchfindings is a matter of some difficulty, though.That education has a bigger effect on socialoutcomes in one country than another does notnecessarily mean that the education system ofthat country is better, or that conditions in thatcountry are more adequate in facilitating theeffects of education. Differences in the absolutelevels of education and the social outcome maymean that education has a greater effect in thosecountries where general educational levels orlevels of the social outcome are lower. It wouldalso be possible to use MLM to examine interac-tions between countries, regions (where possible)and individual effects. Data considerations arevery important when attempting to use MLM, forexample a sufficient number of sampling units arerequired. This means that MLM is not necessarilyan efficient technique when making comparisonsamong a small number of countries.

If robust estimates of effect sizes are obtainedit may also be possible to monetarise thesebenefits in order to calculate a social rate ofreturn. The social rate of return on education isan expression of the relative benefits and costs ofan educational input. More precisely, it is the rateof discount at which the current and futurestream of educational benefits for an individualand society are equal to the current and futurestream of educational costs. Although manycalculations of social rates of return are actuallyfiscal, concentrating only on the costs and bene-fits of education in terms of government expendi-

ture and tax revenue, there has been significantprogress by economists towards monetarisingother social benefits such as intergenerationaltransfers, health and crime (McMahon, 2000).

How far social rates of return can be used asan indicator of the macroeconomic, or macro,rather than micro, social benefits of education isa matter for debate. Even if such an indicatorwere to be calculated, the difficulty of makinginternational comparisons using the social rate ofreturn is well documented (Bennell, 1998). As thesocial rate of return shows the marginal rate ofreturn to education within a country, it makes littlesense to aggregate or compare these marginalrates. If country A has a social rate of return oneducation of 8 % and country B a social rate ofreturn of 6 %, it does not make sense to say thatcountry A is more (or less) efficient at producingsocial outcomes of education than country Bunless human capital assets are fully mobilebetween the two countries. Moreover, as thesocial rate of return is a marginal indicator, wecannot be sure how far additional investment ineducation will depress this rate. Clearly economicrates of return on learning are not necessarilymacrosocial in that consequences for the indi-vidual taxpayer do not necessarily have animpact on social structures and the organisationof social life in general.

As social rates of return cannot specifically beconsidered a macrosocial property of educationwe have not discussed them at length in thisreport. Clearly, in bringing together both thesocial costs and social benefits of educationalinvestment (and in reconciling the social andeconomic) they are potentially a powerful microe-conomic tool. This potential has only beenpartially realised at present as those studieswhich do exist tend to focus on monetarisingone, or at most two, social benefits of education(McMahon, 2000).

It is difficult to see how the calculation of a fullsocial rate of return, using all social benefits, couldbe achieved. Many of the macrosocial effects ofeducation such as social cohesion and changes inattitudes and values, which we have identified inthis report could not be easily incorporated within asocial rate of return (although this does not rule outother forms of economic modelling). Moreover, asthe macrosocial effects arising from ET take effectover long periods of time, the social rate of return


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 150

would need to be calculated with reference to inter-generational considerations.

Modelling, whether in terms of regression,SEM, MLM or calculation of the social rate ofreturn does not represent the same sort ofactivity as evaluation. Evaluation implies asystematic analysis of the effects of a particularprogramme or activity which should usually bebuilt into the programme design. Modelling is notnormally built into the programme design anddoes not usually involve an analysis of the effectsof a particular programme. For example,modelling may examine the effects of educationallevel on social outcomes, whereas evaluationmay involve the effect of a specific socialprogramme on targeted social outcomes.

Comparatively, summative evaluation seems tobe a poor model for research of this type.National education systems are of a differentorder to educational qualifications and certainlydo not represent targeted programmes. This isnot to say that specific programmes or policiescannot be evaluated in a summative sense.However, the evolution of education systems overtime and their contested nature make us scep-tical about claims that they are designed in orderto meet discrete social objectives (although theremay be general aims underlying educationsystems). Moreover, the embedded nature ofeducation systems within national cultures andinstitutional structures means that it is difficult toseparate out the effects of education. It may evenbe counterintuitive to do so given that the func-tions of education are so tightly embedded withinother national systems such as the welfare state.

However, the identification of differences andsimilarities between countries and their systems isa staple of comparative research. For exampleRagin (1981) explains how by systematicallystating similarities and differences between coun-tries one may arrive at a series of logical state-ments regarding country properties. This enablesus to test hypothesis, or at least answer research

questions, comparatively. The most powerful,macrocausal forms of comparison (Skocpol andSomers, 1980) involve logical analysis of multipleinstances where a particular phenomenon occursand the conditions they have in common, and thecomparison of these with a range of instanceswhere the phenomenon does not occur. Wherecertain condition(s) are common to the first setand are absent in the second set, and if the casesare otherwise similar, it can be concluded thatthese conditions represent causes of thephenomena in question. The procedure is, as withquantitative methods, open to the charge thatthere are possible causes which remained unob-served, but the logical comparative method hasthe distinct advantage that it treats each case asa totality, seeking to explain causal processes inthe real context. This procedure is not totallycontrary to evaluation. Moreover, the criteria ofsummative (final) evaluation are probably too strictfor comparative work. Evaluation can be formativeand examine development, rather than targets.

Hence it is probably best to consider researchquestions involving the macrosocial benefits ofeducation as a hybrid somewhere betweenmodelling and evaluation. Comparative researchof this type cannot be evaluative in its summativesense as the targets to be met (the macrosocialbenefits) cannot be modelled in such a way as toremove confounding influences. Countries arehistorically produced, open systems, and it is notpossible to subject education systems ornation-states to control or experiment (at leastsocial-democratic ones). It is possible, though, togain understanding of developmental processesand to identify similarities and differencesbetween countries. In this way, a hybrid repre-senting modelling and formative judgements mayrepresent the closest comparative research ofthis type can get to evaluation. With this in mind,we therefore proceed to a preliminary investiga-tion of the effects of education, training and skillson macrosocial benefits.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 151

From our review of literature, it would seem thatthere are three hypotheses worthy of investigation:(a) macrosocial indicators do not form a

coherent syndrome at national level;(b) distribution of education is as important as

educational levels in influencing macrosocialbenefits;

(c) education/skill has a differential impact onvarious macrosocial indicators according tocountry context.

We will use two quantitative methods of investiga-tion in exploring these hypotheses. First, we willprovide an analysis involving country level aggre-gates from a combined data set. Second, an analysiswill be conducted involving comparison of nationalgroups using individualised respondent data.

6.1. Education and socialcohesion: a macrosocialapproach

We begin by examining the ways in whichmacrosocial indicators relate to each other and tovarious educational aggregates. We have shownin our discussion how it is not only the propertiesof national education systems that are importantin determining macrosocial outcomes, but alsothe distribution of educational qualifications andother related properties such as income distribu-tion. As our discussion has shown, there is noreason to expect macrosocial indicators tocovary across countries (Knack and Keefer, 1997)(Norris, 2000) (Newton and Norris, 2000).

In this analysis we select countries which are notonly advanced democracies, but from which data isreadily available from various international surveys.Specifically, we will be using data from the WVS, theIALS, Interpol crime statistics and the ICVS. Our

country set includes mainly EU countries (Belgium,Denmark, Finland, Germany, Ireland, the Nether-lands, Norway, Portugal, Sweden and the UK)although a small number of non-EU countries werealso included (Australia, Poland and the US). Weuse various macrosocial indicators: general trust(GENTR) and trust in democracy (DEMTR); civiccooperation in terms of attitudes to cheating ontaxes and public transport (TAXCH and TRANCH);a civic participation measure (GROUP); a toleranceindicator (TOLER); and measures of violent crimeand a perception of risk of assault in the localcommunity (CRIME and RISK). Note that thesecrime and community safety variables are coded sothat a reduction in crime or risk would be thought tobe socially beneficial.

As Table 2 shows (7) there is no significant rela-tionship (8) between general trust (GENTR), asso-ciational memberships (GROUP) and oppositionto cheating on public transport fares (TRANCH) atan aggregate level. Figure 1 shows in the form ofa scatter plot the lack of clear relationshipbetween group memberships (GROUP) andgeneral trust (GENTR) at national level – elementswhich are often taken to be centrally coherentsyndromes of social capital.

However, as Table 2 also shows, there aresignificant correlations between general trust(GENTR) and trust in government (DEMTR) (9)(r=.563, p=.029). There are also strongly signifi-cant relationships between general trust and afeeling of local safety (RISK) (r=-.724, p=.005)(p<0.001) and between norms of civic cooperationsuch as never cheating on taxes (TAXCH) andnever cheating on public transport (TRANCH)(r=.592, p=.020) (p<0.001). These do not signifi-cantly correlate in our analysis with trust andmembership, but in Knack and Keefer’s (1997)analysis of the same data, which uses an aggre-

(7) Estimates were computed using the SPSS computer package.(8) The test for significance is p<0.05 for a two-tailed test.(9) This conflicts with Knack and Keefer’s (1997) findings from WVS that aggregate national levels of civic cooperation covary with

social trust scores.

6. The macrosocial benefits of education: a preliminaryinvestigation

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 152

gated factor based on answers to a larger numberof questions about honesty and civic cooperation,there is a correlation between trust and civiccooperation values. Civic cooperation mighttherefore be included as a covariant of trust,although we have not done so in this analysis.

Table 2 also reveals a significant negative correla-tion between tolerance (TOLER) and never cheatingon public transport (TRANCH) which might indicatethat there is a perverse relationship between these

liberal attitudes at a national level (r=-.526, p=.044).Those countries with a higher proportion of thepopulation tolerating people in general are alsosignificantly more likely to have a high proportion ofthe population who are prepared to cheat with theirpublic transport fares. Interestingly, we also find apositive and significant relationship between civicparticipation (GROUP) and only one measure ofcivic cooperation, namely a belief that it is never rightto cheat on taxes (TAXCH) (r=.592, p=.020).


GENTR GROUP DEMTR TAXCH TRANCH CRIME TOLER RISK

Pearson

GENTRcorrelation 1Sig. (2-tailed) .N 15

Pearson .003

GROUPcorrelation 1Sig. (2-tailed) .990 .N 15 15

Pearson .563 (a) -.226

DEMTRcorrelation 1Sig. (2-tailed) .029 .417 .N 15 15 15

Pearson .077 .592 (a) -.312

TAXCHcorrelation 1Sig. (2-tailed) .786 .020 .257 .N 15 15 15 15

Pearson .195 .071 .223 .554 (a)

TRANCHcorrelation 1Sig. (2-tailed) .486 .801 .425 .032 .N 15 15 15 15 15

Pearson -.146 .407 -.177 .430 -.087

CRIMEcorrelation 1Sig. (2-tailed) .603 .132 .528 .110 .757 .N 15 15 15 15 15 15

Pearson .095 .351 -.262 .121 -.526 (a) .250

TOLERcorrelation 1Sig. (2-tailed) .737 .200 .345 .667 .044 .370 .N 15 15 15 15 15 15 15

Pearson -.724 (b) .013 -.372 .026 .075 .012 -.266

RISKcorrelation 1Sig. (2-tailed) .005 .965 .210 .932 .808 .970 .380 .N 13 13 13 13 13 13 13 13

(a) Correlation is significant at the 0.05 level (2-tailed)(b) Correlation is significant at the 0.01 level (2-tailed)

Table 2: Pearson correlation coefficients and levels of significance for social cohesion

aggregates

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 153

6.2. Correlation betweeneducation and socialcohesion measures

In order to test correlation across countriesbetween education and our measures of socialcohesion we need some valid national measures ofeducation. Here we use the data on literacy in theIALS. This survey has been criticised by some(Blum et al., 2001) for cultural bias, but it at leasthas the merit of attempting to provide directmeasures of skills, rather than proxies such asschooling years or qualifications. One may assumethat the skills that it is measuring are related toboth the quantity and quality of education received.

As Table 3 shows, there are no significant corre-lations (p<0.05) across countries between aggre-gates for education levels (PROUS – the meanlevel of upper-secondary attainment in literacy)and measures for social cohesion although there

are significant correlations for TOLER, TRANCHand RISK at the 10 % significance level (p<0.10).These results should be no surprise given what wehave said already about the likelihood that nationalcultural and institutional factors greatly outweighgross education effects on social cohesion. Wetherefore look next at the impact of educationalinequality on social cohesion, on the basis thatcomparative historical and theoretical literaturesuggests that social cohesion is highly sensitive todistributional effects.

6.3. Educational inequality andsocial cohesion

We have used results from all cycles of the IALS toascertain the distribution of educational outcomes,in terms of literacy skills, across a number of coun-tries also included in the WVS. Using a similar


Figure 1: A coherent syndrome? Civic participation and general trust at the national level

PLP

B

UK

FINIRL

DK

NLCA S

NO

US

AU

CH

D

Memberships

.00

.20

.40

.60

.80

1.00

1.20

1.40

1.60

1.80

.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00

General trust


Pearson .354 (a) -.120 .244 -.376 -.487 -.055 .491 -.505

PROUScorrelationSig. (2-tailed) .196 (b) .670 .381 .167 .066 .845 .063 .078N 15 15 15 15 15 15 15 13

(a) Correlation is significant at the 0.05 level (2-tailed)(b) Correlation is significant at the 0.01 level (2-tailed)

Table 3: Pearson correlation coefficients and levels of significance for mean level of upper

secondary attainment and social cohesion aggregates

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 154

methodology to Nickell and Layard (1998, p. 67)we calculated a test score ratio based on thedifferences between the average literacy levels ofthose who attended minimal compulsory educa-tion for that country and those who continued theireducation after the upper secondary level.Following the method used by the OECD (2000)when assessing the social consequences ofinequalities in literacy, we used the measure ofprose literacy rather than the measure of quantita-tive literacy employed by Nickell and Layard (1998,p. 67). There may be questions about the suit-ability of these measures, or a combined measureas a proxy for skill distribution in the labourmarket, and this is an issue for further debate.

Table 4 shows the mean prose scores for thosewhose educational level is less than uppersecondary (PROLEUS), for those who haveattained upper secondary education (PROUS) andfor those who have attained some tertiary educa-tion (PROTERT). The test score ratio (P3) is theratio of the score of those attaining tertiary educa-tion (PROTERT) to those attaining lower thanupper secondary education (PROLEUS). Hence itis the ratio between the level of attainment ofthose who have experienced post-compulsoryeducation and those who have only attained thelowest level of secondary education.

The results show that measures of inequality inskills outcomes are rather higher in English-speakingcountries such as Canada, the UK and the US than insome continental and Nordic countries such asGermany and Sweden. The relative positions of coun-tries here confirm some of the findings on skillsspreads by Brown et al. (2001), based on analysis ofIEA data for test scores at 14 years old, and Green andSakamoto (2001) based on adult distributions of qual-ifications. They have also been broadly confirmed inthe recent OECD PISA study (OECD, 2001).

If we correlate national measures of skills distri-bution against national measures of social cohesion(Table 5) we find that there is a significant (p<0.05)correlation (r=-.592, p=.020) between educationalinequality (P3) and one commonly used macrosocialmeasure, the general level of trust (GENTR). Hence,the higher the level of educational inequality, thelower the level of general trust. This is demonstratedin Figure 2 which shows a scatter plot of generaltrust, as measured in the WVS, against skills distri-bution, as measured by the skills ratio (P3). There isclearly a negative relationship between the two vari-ables with those countries with low inequality ofskills (such as Denmark, Norway and Sweden) alsohaving high levels of trust and those countries withhigh inequality of skills (such as Portugal, the UKand the US) having low levels of trust.


CountryCountry PROLEUS PROUS PROTERT P3

Code

AU Australia 250.60 280.00 310.40 1.24

B Belgium 242.50 281.00 312.30 1.29

CA Canada 233.40 283.80 314.80 1.35

CH Switzerland 228.10 274.10 298.30 1.31

D Germany 265.60 283.80 310.10 1.17

DK Denmark 252.80 278.10 298.50 1.18

FIN Finland 261.60 295.90 316.90 1.21

IRL Ireland 238.80 288.20 308.30 1.29

NL Netherlands 257.50 297.00 312.10 1.21

NO Norway 254.50 284.40 315.10 1.24

P Portugal 206.60 291.50 304.80 1.48

PL Poland 210.50 252.70 277.30 1.32

S Sweden 275.40 302.30 329.10 1.19

UK United Kingdom 247.90 281.90 309.50 1.25

US United States 207.10 270.70 308.40 1.49

Table 4: Mean proficiency scores and test-score ratio for countries in sample

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 155

6.4. Education and incomeinequality

In our literature review we have referred to theinfluence of the distribution of income, or at leastof opportunity, on macrosocial outcomes. We cantest the effects of educational distributions onincome distributions using a method adaptedfrom Nickell and Layard (1998).

In measuring the degree to which skill differen-tials correspond to income inequality, Nickell andLayard (1998) employ earnings ratios, i.e. theratio of incomes between individuals of differingeducational levels. In this paper, we employmeasures of income inequality (GINI) coefficientswhich are a more general measure of earningsinequality for the whole population. GINI coeffi-cients employed are provided in Table 6.

Figure 3 shows the relationship between

educational inequality, measured by the testscore ratio P3, and income inequality for the 15countries in our sample. As can be seen, there isan association between distributions of literacyskills and income inequality. Economies with ahigh degree of skill disparity also have highdegrees of income inequality and vice-versa. Asshown in Figure 3, there is a clear relationshipbetween the test score ratio (P3) and the GINIcoefficient. This relationship is statistically signifi-cant (p<0.01) with a positive and large correlationcoefficient (r=.650, p=.009).

We cannot know from this correlation which wayany causality may run. It is quite probable thatincome equality impacts on educational equalitythrough equalising access to education. It is alsolikely that social cohesion and solidaristic culturesand political ideologies promote both incomeequality and educational equality through equal-ising aspirations and supporting certain types of


Figure 2: Educational equality and general trust

P

US

UK BPL

CHAU

IRL

D

FIN

CA

NLSDK

NO

15

25

35

45

55

65

1.0 1.1 1.2 1.3 1.4 1.5 1.6

Skills distribution

General trust


Pearson-.592 (a) .333 -.283 .265 .171 .398 -.060 .404

P3 correlationSig. (2-tailed) .020 .225 .307 .340 .543 .142 .831 .171N 15 15 15 15 15 15 15 13

(a) Correlation is significant at the 0.05 level (2-tailed)

Table 5: Pearson correlation coefficients and levels of significance for distribution of educational

attainments and social cohesion aggregates

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 156


Country code Country GINI

AU Australia 35.20

B Belgium 25.00

CA Canada 31.50

CH Switzerland 33.10

D Germany 30.00

DK Denmark 24.70

FIN Finland 25.60

IRL Ireland 35.90

NL Netherlands 32.60

NO Norway 25.80

P Portugal 35.60

PL Poland 32.90

S Sweden 25.00

UK United Kingdom 36.10

US United States 40.80

Source: World Bank (2001)

Table 6: GINI coefficients (mid 1990s)

Figure 3: Educational inequality and income inequality

FIN

AUB

DKS D

NL

PLCH

USP

CA

IRL

UKNO

1.00

1.10

1.20

1.30

1.40

1.50

1.60

20.0 25.0 30.0 35.0 40.0 45.0

Test score ratio

Income inequality

policy interventions. Minimum wages and otherforms of labour market regulation that make wageagreements binding and inclusive for entire sectorsmay well enhance income equality (Blau and Kahn,1996) (Nickell and Layard, 1998). Measures toequalise resources for, and admissions to, schoolsmay make educational outcomes more equal, as

may shared aspirations about the value ofschooling; this has been argued in the case of the atleast until recently highly egalitarian Japaneseeducation system (Green, 1999). These relation-ships are still to be investigated but our analysishere of correlations at least suggests that there is anissue to be explored.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 157

6.5. Income inequality andmacrosocial outcomes

The next stage in the model requires that we testwhether there is an association between incomeinequality and macrosocial outcomes.

Table 7 provides the results of the analysis ofcorrelations between income inequality and oursocial cohesion aggregates. For the 15 countriesin our sample we failed to find a significant rela-tionship between income inequality and associa-tion membership. However, a significant positiverelationship between income inequality andviolent crime (CRIME) (r=.640, p=.010) and theperceived risk (RISK) of assault in the community(r=.636, p=.020) was identified in addition to asignificant negative relationship between incomeinequality and general trust (GENTR) (r=-.547,p=.035).

The effect of income inequality on thesemacrosocial outcomes persists even when wecontrol for the general level of economic activity. Inthis model (Table 8), the gross national product(GNP) per capita is used as a control, hence thecorrelation coefficients presented are partial corre-lation coefficients. The measure of GNP per capitaused was taken from the purchasing power parityindex employed by the World Bank (2001). After

introducing controls, the partial correlation coeffi-cients between income inequality and general trust(GENTR) (r=-.562, p=.037) remain significant. Asbefore, we find that inequality decreases generaltrust but increases violent crime (CRIME) (r=.660,p=.010) and increases perceptions of risk of crime(RISK) (r=.628, p=.029). We also find that control-ling for GNP per capita means that the associationbetween income inequality and civic participationbecomes significant (GROUP) (r=.595, p=.025).Hence, even in our reduced sample, it is possibleto locate a positive relationship between incomeinequality and civic participation.

Our analysis of macrosocial indicators hasshown, as the literature discussed abovesuggested, that there are no necessary correlationsbetween macrosocial outcomes at national level.Although we have not examined trends over time inthis paper, it is difficult to accept that increasinglevels of education over time have had a uniformlypositive effect on macrosocial indicators. The meanlevel of skill in literacy in each country does notcorrelate with any measure of macrosocial welfare,although the distribution of this skill does correlatewith general trust. The distribution of this skill is alsocorrelated with income inequality, which in turnseems to be negatively correlated with manymacrosocial benefits including crime.


GENTR GROUP DEMTR TAXCH TRANCH CRIME RISK TOLER

Pearson -.547 (a) .414 -.305 .403 -.009 .640 (a) .636 (a) .240

GINI correlationSig. (2-tailed) .035 .125 .269 .136 .975 .010 .020 .389N 15 15 15 15 15 15 13 15


Table 7: Pearson correlation coefficients and levels of significance for distribution of income and

social cohesion aggregates


Pearson -.562 (a) .595 (a) -.032 .430 -.004 .660 (a) .270 .628 (a)

GINI correlationSig. (2-tailed) .037 .025 .293 .125 .989 .010 .350 .029N 15 15 15 15 15 15 15 15


Table 8: Pearson correlation coefficients and levels of significance for distribution of income and

social cohesion aggregates with controls for GNP/capita

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 158

As a second approach to assessing the effects ofeducation on macrosocial outcomes, we adoptan approach using microsocial data withincomparative context. We tackle similar questionsas in the macrosocial analysis. First, we examinewhether there are differences in the compositionof various social indicators between countries.That is if the measurement indicators areinvariant. Second, we examine the effects ofeducational level on these social indicators inorder to ascertain whether there are systematicdifferences between the four countries. As we arenow dealing with microsocial indicators we arelimited both in terms of the indicators we use andthe data set(s) employed. As we are dealing withindividual level data we cannot use indicatorssuch as the distribution of skills or income. Inaddition, we cannot merge data sets such as theWVS and IALS which utilise different samples ofrespondents.

The 1995 sweep from the WVS probablypresents the optimum choice both in terms ofsocial variables and a standard measure ofeducational level. We included data from fourcountries: Sweden representing a Nordic country(N=1 009), Western Germany representing aNorthern continental country (N=1 017), Spainrepresenting a Mediterranean country (N=1 211),and Poland representing a former Eastern Euro-pean block (and future EU acceding) country(N=1 153). In total there were 4 390 respondents.The WVS recommends weighting of the samplesin order to arrive at equal sample sizes. Usingdifferent weights within each country the samplesizes were boosted to 1 503 for Sweden, 1 495for Western Germany, 1 503 for Spain and 1 499for Poland, in total 6 000 subjects.

We analyse four social outcome variables.Political action is a measure of to what extentpeople have carried out or would potentially carryout actions directed against the establishment inorder to improve the current state of affairs.Respondents indicated on three point scales (1-3)whether they have done, might do or would nevercarry out a certain action. We based the scale onthree items indicating ordinary action; sign a peti-

tion, joining a boycott and attending a lawfuldemonstration. Initial analysis showed thatincluding two more extreme activities (as indi-cated by the skewnesses of the items), joiningunofficial strikes and occupying buildings orfactories, fitted SEM models less well thanmodels in which these two items were notincluded. Hence, the scale was assumed toprobe ordinary instead of extreme action. Thescale was reverse coded with higher values indi-cating more action. The mean internal consis-tency, as measured by Cronbach’s alpha, was .71across the four countries.

Trust in institutions was measured by responsesto the question ‘could you tell me how much confi-dence you have in (institution)?’ Answers weregiven on four point scales: 1 = a great deal, 2 =quite a lot, 3 = not very much, 4 = none at all. Weselected three institutions, the legal system, parlia-ment and civil service, and reverse coded thescale, higher values indicating more trust. Themean internal consistency across the countries asmeasured by Cronbach’s alpha was .67.

Support for democracy was measured withthree items. The respondents rated on four pointscales to what extent they agreed with each state-ment (1 = agree strongly; 2 = agree; 3 = disagree;4 = disagree strongly). The three statements were‘in democracy, the economy system runs badly’,‘democracies are indecisive and have too muchsquabbling’ and ‘democracies aren’t good atmaintaining order’. After initial analysis it wasdecided to drop a fourth item ‘democracy mayhave problems but it’s better than any other formof government’ which, in some countries, did notload on the same factor as the three firstmentioned items. Given that the statements werenegative, negative responses to these (i.e., highervalues) were taken to be indicative of support fordemocracy in general. The mean internal consis-tency was a Cronbach’s alpha of 0.77.

Race tolerance was measured with threeitems. The respondents were asked to sort outwho they would not like to have as a neighbourfrom a list of distinctive groups. We usedresponses to other races, immigrants and

7. Macrosocial benefits through microsocial data?

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 159

Muslims. Responses for each group were coded0 if they mentioned the group and 1 if they didnot. The scale was thus formed as an indexranging from 0 to 3, higher values indicating moretolerance (i.e., they had nominated fewer of thementioned groups as potential neighbours). Thethree items were all skewed. The mean internalconsistency was a Cronbach’s alpha of 0.67.

Our main predictor of these variables is educa-tional level, while we controlled for the effects ofsocioeconomic status and age. Education level wasrecorded according to the structural options avail-able within each country and standardised withineach country in order to make these comparable. Wealso standardised socioeconomic status (SES),which was measured in different ways across coun-tries and ages. Initial analyses showed that age, SESand education level were interrelated in twoexpected ways. First, older participants within eachcountry had a lower level of education and second,educational level and SES were positively correlated.

A technique known as structural equationmodelling (SEM) was employed to model theeffect of education on each of the socialoutcomes. SEM allows us to analyse the impactof education on various latent variables,measuring the social outcomes simultaneously,and to compare effects to identify systematicdifferences between countries. The full details ofthe modelling are provided in Annex 2 where weexplain how the models were specified andconfirm that the models were an adequate fit tothe data across all four countries. Here we willconcentrate on the substantive results.

Figure 4 shows the effect of education on variousmicrosocial indicators of social cohesion: politicalaction (POL), institutional trust (INST), trust in democ-racy (DEM) and race tolerance (RACE). The diagramsshow the effect of education having controlled forsocioeconomic status and age. Each arrow showsthe effect size in terms of both standardised and nonstandardised regression coefficients. For example, inSweden, the effect of education on political action is0.09, or 0.24 when standardised. As explained inAnnex 2, we used significance tests to show wherethe effect sizes differed significantly between coun-tries. Results show that effect sizes for both Sweden

and Poland are significantly different from those inthe other countries, namely, the effect of educationon both trust in institutions and trust in democracywas stronger in Sweden. In addition, the effect ofeducation on political action was lower in Sweden.Somewhat surprisingly, the effect of education onracial tolerance was higher in Poland than the othercountries.

There are two possible explanations for theseeffects. First, that the effects simply representdifferences in country means (10). That is, in coun-tries with a low mean score for racial tolerance(for example), education will have a strongereffect as there is more room for potentialincreases in the tolerance score. This argumentmay apply in the case of Poland where, as theIEA study shows, levels of racial tolerance arelow. This is also confirmed in the data used in thisstudy. Table 9 shows the raw weighted means foreach country, and Poland has the lowest meanlevel of racial tolerance (2.33). This may also bethe case for institutional trust in this study, wheredescriptive statistics show that the Swedish levelof institutional trust is below that of the other fourcountries in the study with a mean value of 2.51.

Second, it could be tentatively argued thatdifferences in education effects represent differ-ences in the socialisation effects of education (asopposed to other influences) in these countries.In Sweden, for example, it could be argued thateducation has a central role in imparting trustwhereas socialisation into forms of political actionmay occur through involvement in the communityor workplace (given Sweden’s high rates ofunionisation and the importance of unions innational policy determination). In Poland, theeducation system may play an important functionin terms of increasing racial tolerance, whereasfor the other three countries the family or commu-nity may have a more important role.

These conclusions are tentative in that furtheranalysis and more qualitative studies arerequired. However, we have shown here thepossibilities of using microsocial analysis to illu-minate, although not explain, macrosocial differ-ences in education systems in terms of theirsocialisation functions.


(10) As the variables used in this SEM are latent variables we can not readily convert effect sizes into marginal effects (e.g. theeffect of a level of education on a unit change in tolerance, for example) although we may compare effect sizes between coun-tries. Mean values are given for the sample population as a whole and are indicative of the general level of the variables in eachcountry.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 160


POL INST DEM RACE

Sweden 2.30 (.52) 2.51 (.55) 2.82 (.71) 2.79 ( .58)

Western Germany 2.18 (.51) 2.60 (.50) 3.01 (.54) 2.84 ( .47)

Spain 1.65 (.60) 2.68 (.62) 2.75 (.60) 2.73 ( .69)

Poland 1.57 (.55) 2.68 (.64) 2.14 (.65) 2.33 (1.04)

NB: POL = political action, INST = trust in institutions, DEM = support for democracy. RACE = race tolerance

Table 9: Raw weighted means and standard deviations (in brackets) within each country

Figure 4: Effects of education on social cohesion in four countries (non standardised and

standardised regression coefficients) with controls for socioeconomic status and age

West Germany

Spain

Sweden

Poland

EDUC

SES

AGE

POL

INST

DEM

RACE

EDUC

SES

AGE

POL

INST

DEM

RACE

EDUC

SES

AGE

POL

INST

DEM

RACE

POL

INST

DEM

RACE

EDUC

SES

AGE

.09 / .24

.07 / .16

.22 / .36

.01 / .10

(a)

(a)

(a)

.16 / .41

.00 / .00

.13 / .19

.01 / .15

.16 / .33

.00 / .00

.13 / .25

. 01 / .08

.16 / .34

.00 / .00

.13 / .23

.06 / .22

(a)

Note: EDUC = Educational level standardised within each countrySES = Socioeconomic status standardised within each countryAGE = Age standardised within each countryPOL = Political actionINST = Trust in institutionsDEM = Support for democracyRACE = Race tolerance

(a) regression path that is significantly different from path in the other three countries as tested in a nested model comparison(Dc2; p <.05). For sake of readability regression paths from SES and AGE, as well as latent correlations were omitted.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 161

In this paper we have explored the effects ofeducation, training and skills on a wide range ofmacrosocial outcomes including crime, socialcohesion, citizenship, civic and political participa-tion. Through a review of literature and statisticalmodelling we have mapped out both the problemaspects of methodological individualism and thecomparative approach. In particular, the possibilityfor summative evaluation (rather than measuringor modelling) of education systems appears to belimited, although the component parts of nationaleducation systems are open to evaluation. Wesuggest instead that evaluation of macrosocialbenefits requires a hybrid approach to evaluationinvolving both modelling and formative evaluation.The procedures advanced by Ragin (1981) appearto be helpful in this respect as they enable us,through the method of similarity and difference, toconstruct workable research questionsconcerning the outcomes of education.

In terms of the status of various models of therelationship between education and socialoutcomes, we are troubled that many of thesemodels are specified at the individual level. Whilemany microsocial impacts of education withincountries are best specified as relations involvingindividual resources, knowledge and skills, whencomparing countries the country context matters.Moreover, country context is more than justanother variable and is not ecological in a simplesense. There are expressions of national contextsuch as distributions of skills and income, educa-tion systems and culture which cannot be speci-fied except in comparison to other countries.

With regard to macrosocial benefits of educa-tion, training and skills there are a few generalisa-tions which can be made across all EU countries.

First, that for some macrosocial benefits (orcosts) there are common structural antecedents.Various forms of criminal activity can be regarded aslocal manifestation of structural phenomena. Forexample, with regard to work on football hooli-ganism (Dunning, 2000), juvenile delinquency andhate crime (Watts, 2001) and possibly tolerance,similar structural antecedents (unemployment andalienation) related to education are implicated. The

relevance of income distribution (and by implicationthe distribution of skills) and the spatial character-istics of high crime areas are also potentially similarstructural antecedents of crime (Kelly, 2000) (Lee,2000). As shown by our work using macrosocialaggregates, there are clearly relationships betweeneducational inequality, income inequality andoutcomes such as general trust, crime and feelingsof community safety.

Second, the clustering of social benefits andeducational level which one may see at the microlevel does not necessarily hold at the macro.Educational level and social benefits such asgeneral (and other forms of) trust, associationmemberships and crime are not necessarilyrelated at the national level as evidenced by ourliterature review and our modelling of macrosocialoutcomes. Macrosocial outcomes are not relatedat the country level and do not form a coherentsyndrome (Putnam, 2000). The implications foreducation systems are that generalisationsconcerning the role of education in the rebuildingof civil society (social capital investment) or infostering widespread political involvementthrough civics education are not applicableacross nation-states. While such policies are notnecessarily misguided, in that some individualsmay benefit, the effect on national levels of socialoutcomes may be small or non-existent. As ourliterature review has shown, there is muchevidence that increases in the general level ofeducation have not had any effect on nationallevels of tolerance, crime or social cohesion.Evidence from Wilkinson (1996, p. 20) supportsthis general conclusion with respect to health: ‘Ineffect, the extent of the variation around asociety’s norms is fixed so that the proportion ofpeople with bad diets, who are heavy drinkers,who have high blood pressure etc. is a reflectionof where the society’s norms are […] it was easierto change the societal norms than to leave themunchanged while trying to reduce the proportionof the population over some level of risk.’

However, that societal norms and inequalitiesare hard to change does not mean that there isno role for education or training. As Eisner’s work

8. Conclusions

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 162

(2001) explains, the effects of various institutionsimplicated in state formation (such as education)may only be seen to have effects over a longperiod of time. Moreover, we may speculate as tothe role of education systems in various regionalgroupings on the formation of values and theconstruction of inequalities.

To illustrate this point, it is clear that the Nordiccountries form a group of high trust, mainly lowcrime countries where general levels of civic partic-ipation are also moderate. In the Danish case, thisis combined with high levels of lifestyle permissive-ness but rather low levels of tolerance towardsforeigners (at least on Halman’s evidence, 1994).The high levels of trust may be associated withvarious non-education macro factors such as thestrongly solidaristic welfare states and historicallyrelatively high levels of ethnic/cultural homogeneity(Knack and Keefer, 1997), although we have notexamined these factors here. They may also relateto relatively high levels of income equality. Lowerlevels of ethnic tolerance in Denmark may be asso-ciated, paradoxically, with that same emphasis oncultural homogeneity that may be conducivetowards high trust in this case (although we do notsuggest that these relationships would hold in allnational contexts). We may hypothesise that rela-tive equality of educational outcomes promotestrust and lower crime through its impact on incomeequality. The strong effect in Sweden of educationon trust in institutions and in democracy may beattributable to the strong solidarity principlesenshrined in curricula and in the universal nature ofprimary and secondary school systems whichremain comprehensive and non-selective up to theend of upper secondary schooling (gymnasieskola).However, the lower levels of civic association(compared to other – mostly European – countriesin our sample) may result from the historicallyprominent role of the state in Nordic social democ-racy in promoting equity and inclusion and to thesuccess of this in promoting social equity. This mayseem to obviate the need to take political actionoutside mainstream channels. The evidence fromour micro level analysis that education has a weakeffect on civic association in Sweden may reflectthe fact that civic participation in Sweden is highlyinstitutionalised, not least with the prominent role oftrades unions within the social partnership system.

In contrast, the UK has high levels of crime andscores low relative to the other countries in our

sample on both trust and association, while alsobeing low on tolerance according to Halman’sevidence (1994). Although (due to dataconstraints) the UK was not included in ourmicrosocial analysis, an historical overview shedslight on why this might be the case. Non-educa-tion macro factors associated with higher crimeand lower trust may include the high rates ofincome inequality (among the highest in the EU)and higher levels of intolerance may relate in partto historically high levels of immigration over thepast 40 years (Halman, 1994). Education may playa part in generating lower levels of trust andhigher crime through its impact on incomeinequality. A high level of market influence, relativeto the rest of Europe, with high levels of inequalityin outcomes between schools and regions, andconsequently wide distributions of educationaloutcomes, may be significant in generatingincome inequality and lower trust. The latter arisesboth through its effect on income inequality andmore directly through the competitive values itpromotes, which are not counteracted by anyNordic style emphasis on social solidarity in theschool curriculum. Since differences betweenschools reflect ethnic differences, given thetendency for increasing ethnic concentration/segregation in schools in a quasi-market system,this may also play a part in both decreasing trustand increasing intolerance (the latter because ofthe reduction in ethnic mixing). The low tomoderate UK level of association relative to therest of the countries in our reduced sample isharder to explain given Britain’s history of valuingcivil society and intermediate associations(Gramsci, 1971), although the reduction in tradeunion membership and activity following therestrictive laws brought in by the Thatcher govern-ment after 1979 may have had some effect.

These are obvious generalisations and somewhatstereotypical depictions of these countries’ educationsystems. We do contend, though, that the macroso-cial benefits of ET are rooted both in the distribution ofeducational outcomes and in the values transmittedthrough education systems. They are also contingenton the relationship between education and the labourmarket and other parts of the welfare state. Althoughthere are cultural limits to the extent to which policyborrowing is appropriate with regard to educationsystems, there are clear lessons for policy-makers. Inparticular, raising educational skills and training levels


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 163

is neither a necessary nor sufficient condition forpromoting macrosocial benefits. However, improvingthe distribution of educational outcomes may be one

way in which education and training can make somecontribution to more general economic and socialredistribution.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 164

AGE Age standardised

AIC Akaike information criterion

CFI Comparative fit index

CRIME Violent crime

DEM Trust in democracy/support for democracy

DEMTR Trust in democracy/trust in government

EDUC Educational level standardised

ET Education and training

ETGACE Education and training for active citizenship in Europe

EVS European values survey

GDP Gross domestic product

GENTR General trust

GINI Measures of income inequality

GNP Gross national product

GROUP Civic participation measure/associational memberships/group membership

IALS International adult literacy survey

ICVS International crime victimisation survey

IEA International association for the evaluation of educational achievement

INST Institutional trust

ISRD International self-report delinquency study

MLM Multi-level modelling

P3 Test score ratio/measure of educational inequality

POL Political action

PROLEUS Educational level lower than upper secondary

PROTERT Tertiary education attained

PROUS Upper secondary education attained/the mean level of upper-secondary attainment in literacy

RACE Race tolerance

RISK Risk of assault/perceived risk/feeling of local safety

RMSEA Root mean square error of approximation

SEM Structural equation modelling

SES Socioeconomic status

TAXCH Cheating on taxes

TOLER Tolerance indicator

TRANCH Cheating on public transport

VET Vocational education and training

WVS World values survey


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 165

As no one data set could satisfy the internationalcomparisons required, a combined data set wasconstructed using data from the WVS, IALS,World Bank, Interpol statistics and the ICVS. Alldata used was from the years 1990 to 2000.Fifteen countries were included in the core dataset being Australia, Belgium, Canada, Denmark,Finland, Germany, Ireland, the Netherlands,Norway, Poland, Portugal, Sweden, Switzerland,the UK and the US. Social cohesion variablelevels for each country are given in Table 10.

Social cohesion measures were obtained fromthe most recent country sweep available of theWVS. In most cases, data used was from the1995 to the 1997 sweep, although when data forthese years was not available, data from the 1990sweep was substituted.

General trust (GENTR) was measured by thepercentage of individuals sampled in eachcountry who agreed that most people could be

trusted when asked: ‘Generally speaking, wouldyou say that most people can be trusted or thatyou can’t be too careful in dealing with people?’(WVS, question V27).

Associational memberships (GROUP) wasmeasured by the mean number of associationalmemberships for sampled individuals in eachcountry, not including memberships of sportingassociations (WVS, questions V28 and V30-V36).

Trust in government (DEMTR) was measuredby the percentage of individuals sampled in eachcountry who agreed or strongly agreed that theyhad confidence in their parliament (WVS, ques-tion V144).

Civic cooperation measures cheating on publictransport fares (TRANSC) and cheating on taxes(TAXCH). They were measured by the percentageof individuals in each country who stated thatsuch actions were never justifiable (WVS, ques-tions V193 and V194).

Annex 1Macrosocial data

CountryCountry GENTR GROUP DEMTR TAXCH TRANCH CRIME TOLER RISK

Code

AU Australia 39.90 1.06 30.60 62.10 62.80 37.38 95.42 2.25

B Belgium 30.60 .28 42.80 33.90 57.70 29.84 82.28 1.89

CA Canada 50.70 .47 37.90 59.20 61.90 109.21 94.30 1.78

CH Switzerland 37.80 .68 43.90 53.70 59.30 34.40 89.99 1.87

D Germany 41.80 .54 29.40 40.10 38.60 86.92 95.60 N/A

DK Denmark 57.70 .18 42.00 57.30 74.50 47.69 88.95 1.67

FIN Finland 46.90 .32 32.40 57.40 62.60 45.42 85.44 1.77

IRL Ireland 46.80 .23 50.30 48.80 57.50 96.88 93.78 1.99

NL Netherlands 55.80 .36 51.60 42.90 55.80 121.46 88.64 1.83

NO Norway 64.80 .61 69.50 47.50 70.20 31.26 81.75 N/A

P Portugal 20.70 .19 33.50 39.90 53.40 62.57 90.53 2.18

PL Poland 34.50 .03 34.50 55.20 68.10 71.04 75.89 2.29

S Sweden 56.60 .52 44.60 49.30 47.00 85.38 95.34 1.68

UK United Kingdom 29.10 .20 46.10 53.90 59.40 144.83 88.30 2.10

US United States 35.00 1.63 30.30 73.60 66.50 209.85 90.31 1.95

Table 10: Macrosocial aggregates for 15 countries

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 166


The measure of educational inequality (P3) wasobtained from IALS secondary data by dividingthe mean prose score of those individuals whohad completed tertiary education by the meanprose score of those who had completed uppersecondary education only. To compute thesescores, we utilised the most recent sweep ofIALS data (OECD and Statistics Canada, 2000).

Measures of income inequality (GINI) and GNPper capita (GNPCAPIT) were taken from the mostrecently available World Bank Statistics (WorldBank, 2001, pp. 282-283).

The measure of crime (CRIME) was obtainedfrom Interpol statistics for 1996 (InternationalCriminal Police Organisation, 1996). The measure

of crime used being the sum of homicides,robberies and violent thefts per 10 000 inhabitants.

The measure of tolerance (TOLER) was obtainedfrom question V57 of the WVS and measures thepercentage of respondents in each country whowould not mind having an immigrant as a neighbour.

The measure of perceived risk of crime (RISK)was obtained from the mid-1990s, or most recentpossible, sweep of the ICVS and measures themean score for each country from respondents’feelings of safety when walking alone after dark inthe area (very safe=1, very unsafe=4). Figureswere not available for Germany and Norway andso these countries were not included in analysisinvolving this variable.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 167

The raw data was first screened for missing values,which were replaced using regression imputationtechniques as available in the SPSS MVA module.Regression imputation techniques provide lessbiased parameter estimates than do list wise or pairwise deletion or mean value imputation of data.After the data imputation, correlation matrices, andmean and standard deviation moments were esti-mated in weighted samples, the data was importedinto the AMOS software.

We then conducted the following analyses. First,by using SEM techniques we set up a Confirmatoryfactor analyses (CFA) in order to test the structuralvalidity of a model including four latent constructsand three manifest indicators for each construct.This means that we wanted to investigate whetherthe items actually measured the underlying propertythat we hypothesised, by testing this empirically. Thelogic of SEM is to test whether one’s theory (herethree items loading on each of the four constructs)as defined in a measurement model fits the data well.If there is no significant difference between data andmodel then one’s theory receives support. We eval-uated model fit by the Comparative fit index (CFI),values over .95 indicating good model fit (Bentler,1990), and the Root mean square error of approxi-mation (RMSEA) values lower than .05 indicatinggood model fit (Brown and Cudeck, 1993). Eventhough the χ2 and χ2/df are susceptible to samplesize and model complexity these are reported. Addi-tionally we report the Akaike information criterion(AIC) (Brown and Cudeck, 1989) measure, which isnot affected by the particularities of the defaultAMOS null models when estimating means andintercepts (Arbuckle and Wothke,1999). Separatemodels fitted the data well within each country. Wethen proceeded to testing a four-country modelaccording to the following specifications.

We wanted to investigate whether our latentconstructs measured the same thing across the fourcountries, which is the cornerstone of cross-culturalresearch, or when comparing measurements acrosssubgroups (Little, 1997). If the researcher candemonstrate measurement invariance across the

subgroups, the comparison of means or regressioncoefficients across the subgroups becomes mean-ingful. The strictest test of measurement invarianceis to constrain both factor loadings (regression pathfrom latent construct to manifest item) and inter-cepts (mean level of the manifest indicator). A lessstrict test would be to constrain the factor loadings(but not the intercepts), which would still be indica-tive of structural validity across contexts. If theconstrained model holds, we can thereafter proceedto include educational level (with controls forsocioeconomic status and age) as predictor of thefour latent constructs.

We proceeded by first testing a freely esti-mated model (model 1, Table 11) including fourcorrelated latent constructs, each with threemanifest items. All factor loadings were allowedto vary across the four groups. The model fitteddata well (RMSEA = .016; CFI = .983). Next, weconstrained the factor loadings across the fourgroups in model 2, which fitted the data less wellthan model 1 (the freely estimated one), but stillhad a good model fit (RMSEA = .021; CFI = .960).

In the next series of models we included thelatent means to be estimated. In model 3 weconstrained the intercepts while allowing themeans to vary, in model 4, both the interceptsand factor loadings while the means varied, andin model 5 intercepts, factor loadings, and latentmeans were constrained. The equal interceptmodel held (RMSEA=.028; CFI=.941) but themodel with equal intercepts and factor loadingsdid not (RMSEA=.038; CFI=.886). Model 5 inwhich the latent means were constrained indi-cated that there were cultural differences withregard to the four outcomes, however, given thepoor fit of model 4 we did not pursue testingthese differences. In sum the first set of models(model 1-5) demonstrated that it is possible toconstrain the factor loadings, which is indicativeof structural validity across the four countries.However, the misfit of models 4 and 5 indicatedthat it is not possible to estimate latent meandifferences due to incompatible intercept struc-

Annex 2Details of SEM modelling

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 168

ture. Nevertheless, we deemed equal factorstructures as a plausible way forward for testingthe effect of educational level on the fouroutcomes, with controls for socioeconomic back-ground and age, across the four countries.

In order to investigate the main research ques-tion of this study, that is how educational levelaffects social cohesion across four countries, weincluded regression paths from educational quali-fications, socioeconomic status and age uponeach of the four latent constructs, thus creating amodel with three latent predictor variables andfour latent outcomes.

Next, model 6 included three latent predictorsand four latent outcomes with freely estimatedfactor loadings for the outcomes (the latent predic-tors only included one manifest indicator each,which were set to 1.00). In model 7 the factor load-ings of the four latent outcomes were constrained

across the countries while the regression paths wereallowed to vary. As we can see in Table 11, model 7fitted less well than model 6, but still at an accept-able level, whereby we in the following modelsconstrained the factor loadings. We tested whetherthe effect of education was the same in all four coun-tries by comparing the model in which the regres-sion paths in the four countries were constrained tobe equal as compared against a model in whichthese were freely estimated (model 7), in nestedmodel comparisons, as a form of post hoc test. Wethen observed the difference between the twonested models according to the difference of χ2

between the two models (∆χ2). In order to improve fitbetween the two nested models, some parameterswere relaxed until the ∆χ2 became non-significant.This means that there are significant differences inthe regression paths between the countries. Theseregression coefficients are presented in Table 12.


Model Verbal description χ2 df χ2/df RMSEA AIC CFI

1 free factor loadings 472.98 192 2.46 0.016 712.98 0.983

2 constrained factor loadings 872.21 216 4.04 0.023 1064.21 0.960

3 equal intercepts 1196.64 216 5.54 0.028 1484.64 0.941(a)

4 equal intercepts 2137.04 240 8.90 0.038 2377.04 0.886(a)

and factor loadings

5 equal intercepts, factor 5129.48 252 20.36 0.057 5345.48 0.706(a)

loadings and latent means

(a) In models in which means and intercepts were estimated, the CFI fit index was calculated using a self-defined null-model

Table 11: Descriptions of, and goodness of fit indices for models 1 to 5

Model Verbal description χ2 df χ2/df RMSEA AIC CFI

6 free factor loadings 1579.04 324 4.87 0.025 1891.04 0.939and free regression paths

7 constrained factor loading 1992.85 348 5.73 0.028 2256.85 0.920and free regression paths

8 constrained regression 1998.77 350 5.71 0.028 2258.77 0.920EDUC -> POL (not Sweden)

9 constrained regression 1996.62 350 5.71 0.028 2256.62 0.920EDUC -> INST (not Sweden)

10 constrained regression EDUC -> DEM (not Sweden) 1997.50 350 5.71 0.028 2257.50 0.920

11 constrained regression EDUC -> RACE (not Poland) 1997.21 350 5.71 0.028 2257.21 0.920

Table 12: Descriptions of, and goodness of fit indices for models 6 to 11

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 169

In model 8 (Table 12) we first constrained theeffect of education on political action across thefour countries. When we set the regression pathin the Swedish group free we found that theeffect was lower (the non standardised coefficientwas βunstand=.09; p<.001) and standardised coeffi-cient was βstand=.24; p<.001), than in the otherthree countries (βunstand=.16; p<.001; βstand=.41)which were not different from each other.

In model 9 we tested, in the same manner,whether there were differences in effects of educa-tion on trust in institutions. The nested modelcomparison showed that the effect of educationwas stronger in Sweden (.07/.16) than in the otherthree countries (.00/.00), which were not differentfrom each other. The regression effects were notsignificant in Poland, Spain or Western Germany.

In model 9, the regression path from educa-tional level to trust in democracy in Sweden(.22/.36), had a stronger effect than in the otherthree countries (.13/.19-.25), which were notdifferent from each other.

Finally in model 10, the effect of educational

level on race tolerance was stronger in Poland(.06/.22), than in the other three countries, inwhich the effects were not different from eachother (.01/.08-.15).

In our analysis we compared a number ofmodels in which different parameters wereconstrained (Table 11 and 12). Importantly, themodels in which intercepts and factor loadingswere constrained across countries did not reachan appropriate model fit. This means that thereare cultural differences in which people perceive,interpret and give weight to the different itemsthat constitute the four social cohesion measures.We were not able to compare the properties ofthe construct in the strictest sense. Previousstudies have acknowledged the difficulty ofattaining measurement invariant constructs ofsocial capital (Lillbacka, 2002). However, in a lessconstrained model the factor loadings were foundto be equal across countries, demonstrating thatthe structural validity of the constructs arecompatible (i.e., the items measure that certainlatent construct, and not another construct).


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 170

Abramson, P.; Inglehart, R. Education, security andpostmaterialism: a comment on Dutch andTaylor’s ‘Postmaterialism and the economiccondition’. American Journal of PoliticalScience. Bloomington, IN: MPSA – MidwestPolitical Science Association, 1994, Vol. 38,No 3, p. 797-814.

Aldrich, R. The apprentice in history. In: Ainley, P.;Rainbird, H. (eds) Apprenticeship: towards anew paradigm of learning. London: KoganPage, 1999.

Almond, G.; Verba, V. The civic culture. Princeton:Princeton University Press, 1963.

Anheier, H. Measuring global civil society. In:Anheier, H.; Glasius, M.; Kaldor, M. (eds) Globalcivil society. Oxford: Oxford University Press,2001.

Arbuckle, J. L.; Wothke, W. Amos 4.0 users’ guide.Chicago: Smallwaters, 1999.

Baron, S.; Field, S. J., Schuller, T. Social capital: critical perspectives. Oxford: Oxford UniversityPress, 2000.

Becker, G. S. Crime and punishment: an economicapproach. The Journal of Political Economy.Chicago: University of Chicago Press, 1968,Vol. 76, p. 169-217.

Becker, G. S. Human capital: a theoretical andempirical analysis, with special reference toeducation (third ed.). Princeton NJ: PrincetonUniversity Press, 1993.

Beki, C.; Zeelenberg, K.; van Montfort, K. An anal-ysis of the crime rate in the Netherlands,1950-93. British Journal of Criminology. Oxford:Oxford University Press, 1999, Vol. 39, No 3,p. 401-415.

Bennell, P. Rates of return to education in Asia: areview of the evidence. Education Economics,August 1998, Vol. 6, No 2, p. 107-120.

Bentler, P. M. Comparative fit indexes in structuralmodels. Psychological Bulletin. Washington,DC: American Psychological Association,1990, Vol. 88, p. 588-606.

Blau, F.; Kahn, L. M. International differences inmale wage inequality: institutions versusmarket forces. The Journal of PoliticalEconomy. Chicago: University of Chicago

Press, 1996, Vol. 104, No 4, p. 791-837 (alsoNBER working paper, No w4678).

Blaug, M. The methodology of economics, or howeconomists explain. Cambridge: CambridgeUniversity Press, 1992.

Blum, A.; Goldstein, H.; Guérin-Pace, F. Interna-tional adult literacy survey (IALS): an analysis ofinternational comparisons of adult literacy.Assessment in Education. Philadelphia: Carfaxpublishing, 2001, Vol. 8, No 2, p. 225-246.

Boucher, L. Tradition and change in Swedish education. London: Elsevier Science, 1982.

Braithwaite, J.; Braithwaite, V. The effect on incomeinequality and social democracy on homicide.British Journal of Criminology. Oxford: OxfordUniversity Press, 1980, Vol. 20, No 1, p. 45-53.

Brown, M. W.; Cudeck, R. Single sample cross-validation indices for covariance structures. Multivariate Behavioral Research. LawrenceMahwah, NJ: Erlbaum Associates, 1989,Vol. 24, p. 445-455.

Brown, M. W.; Cudeck, R. Alternative ways ofassessing model fit. In: Bollen, K. A.; Long, J. S.(eds) Testing structural equation models.London: Sage, 1993. (Sage Focus Editionsseries, 154).

Brown, P.; Green, A.; Lauder, H. High skills: globali-sation, competitiveness and skill formation.Oxford: Oxford University Press, 2001.

Brubaker, R. Nationalism reframed: nationhood andthe national question in the new Europe.Cambridge: Cambridge University Press, 1996.

Bursik, R. J. Social disorganization theories ofcrime and delinquency: problems andprospects. Criminology. Columbus, Ohio:American Society of Criminology, 1988, Vol. 26,No 4, p. 519-552.

Bynner, J. M. Changing needs, new structures?Lessons for vocational preparation. CEDARInternational Conference, University ofWarwick, 15 April 1994.

Canadian Policy Research Institute. Social cohe-sion research workplan. Canadian PolicyResearch Institute working paper, 1997.

Clarke, L. The changing structure and significanceof apprenticeship with special reference to

Bibliography and references

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 171

construction. In: Ainley, P.; Rainbird, H. (eds)Apprenticeship: towards a new paradigm oflearning. London: Kogan Page, 1999.

Cox, O. C. Class, caste and race. New York: MonthlyReview Press, 1970.

Dasgupta, P; Serageldin, I. Social capital: a multi-faceted perspective. Washington DC: WorldBank Office of the Publisher, 2000.

de Hart, J.; Dekker, P. Contracting out the participa-tion function: social capital and chequebookparticipation. In: Van Deth J. et al. (eds) Socialcapital and European democracy. London:Routledge, 1999.

De Witte, H. Everyday racism in Belgium: anoverview of the research and an interpretationof its link with education. In: Hagelndoorn, L.;Nekuee, S. (eds) Education and racism: across-national inventory of positive effects ofeducation on ethnic tolerance. Aldershot:Ashgate, 1999 (Research in migration andethnic relations series).

Dean, D. Education for moral improvement, domes-ticity and social cohesion: expectations andfears of the labour government, 1945-51. OxfordReview of Education. Philadelphia: Carfaxpublishing, 1991, Vol. 17, No 3, p. 269-286.

Dunning, E. Towards a sociological understandingof football hooliganism as a worldphenomenon. European Journal on CriminalPolicy and Research. Dordrecht: Kluwer, June2000, Vol. 8, No 2, p. 141-162.

Durkheim, É. The evolution of educational thought.London: Routledge & Kegan Paul, 1977.

Eisner, M. Modernization, self-control and lethalviolence: the long-term dynamics of Europeanhomicide rates in theoretical perspective.British Journal of Criminology. Oxford: OxfordUniversity Press, 2001, Vol. 41, No 4,p. 618-638.

Eisner, M.; Wikstrom, P. Violent crime in the urbancommunity: a comparison of Stockholm andBasel. European Journal on Criminal Policy andResearch. Dordrecht: Kluwer, 1999, Vol. 7, No 4,p. 427-442.

Elias, N. The civilizing process. Oxford: OxfordUniversity Press, 1978, Vol. I, II.

Elliot, D. Longitudinal research in criminology:promise and practice. In: Westekamp, G.;Kerner, H. (eds) Cross national longitudinalresearch on human development and criminalbehaviour. Dordrecht: Kluwer, 1994.

Emler, N.; Fraser, E. Politics: the education effect.Oxford Review of Education. Philadelphia:Carfax publishing, 1999, Vol. 25, No 1 and 2,p. 251-273.

Erikson, E.; Jonsson, J. The Swedish context:educational reform and long-term change ineducational inequality. In: Erikson, R.;Jonsson, J. (eds) Can education be equalised?The Swedish test case in comparative perspec-tive. Boulder, CO: Westview Press, 1996a.

Erikson, R.; Jonsson, J. Introduction: explainingclass inequality in education: the Swedish testcase. In: Erikson, R.; Jonsson, J. (eds) Caneducation be equalised? The Swedish test casein comparative perspective. Boulder, CO: West-view Press, 1996b.

Estrada, F. Juvenile crime trends in post-warEurope. European Journal on Criminal Policyand Research. Dordrecht: Kluwer, 1999, Vol. 7,No 1, p. 23-42.

ETGACE. Becoming active citizens: Europeansreflecting on their practice. ETGACE – Educa-tion and training for governance and active citi-zenship in Europe, 2002 (working paper).

Evans, K. Shaping futures: learning for competenceand citizenship. Aldershot: Ashgate, 1998.

Evans, K.; Heinz, W. Becoming adults in Englandand Germany. Rochester: Anglo-German Foundation, 1994.

Farrington, D. P. et al. Unemployment, schoolleaving and crime. British Journal of Crimi-nology. Oxford: Oxford University Press, 1986,Vol. 26, No 4, p. 335-356.

Farrington, D. P.; Loeber, R. Transatlantic replica-bility of risk factors in the development of delin-quency. In: Cohen, P.; Slomkowski, C.;Robins, L. N. (eds) Historical and geographicalinfluences on psychopathology. New Jersey:Lawrence Erlbaum, 1999.

Feinstein, L. Quantitative estimates of the socialbenefits of learning, 1: Crime. London: TheCentre for research on the wider benefits oflearning, 2002 (research report No 5).

Felson, M. Crime rate trends and cycles: changes inmodern society. Paper presented at the confer-ence ‘Why is crime diminishing in theUnited States?’ Bologna: Instituto Carlo,Cattaneo, May 1999.

Freeman, R. Why do so many young American mencommit crimes and what might we do about it?


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 172

Journal of Economic Perspectives, 1996,Vol. 10, No 1, p. 25-42.

Friday, P. International review of youth, crime anddelinquency. In: Newman, G. (ed.) Crime anddeviance: a comparative perspective. London:Sage, 1980.

Fukuyama, F. The great disruption: human natureand the reconstruction of social order. London:Profile Books, 1999.

Garland, D. The culture of high crime societies.Some preconditions of recent ‘law and order’policies. British Journal of Criminology. Oxford:Oxford University Press, 2000, Vol. 40, No 3,p. 347-375.

Gavaert, A. Ethnocentrisme bij scholieren. Eenonderzoek naar de geslachtverschillen inzakede houding ten opzichte van migranten.Unpublished report of Psychology Department,Leuven, 1993 (cited). In: Hagendoorn, L.;Nekuee, S. (eds) Education and racism: across-national inventory of positive effects oneducation and racial tolerance. Aldershot:Ashgate, 1999.

Gaviria, A., Raphael, S. School-based peer effectsand juvenile behavior. Review of Economicsand Statistics. Cambridge: MIT Press, 2001,Vol. 83, p. 257-268.

Giddens, A. (ed.) Emile Durkheim: selected writings.London: Cambridge University Press, 1972.

Gillborn, D. Racism and antiracism in real schools:theory, policy, practice. Buckingham: OpenUniversity Press, 1995.

Gillborn, D.; Youdell, D. Rationing education: policy,practice, reform and equity. Buckingham: OpenUniversity Press, 2000.

Glaeser, E. L.; Sacerdote, B.; Scheinkman, J. Crimeand social interactions. Quarterly Journal ofEconomics. Cambridge: MIT Press, 1996,Vol. 111, No 2, p. 507-548.

Glaeser, E.; Laibson, D.; Sacerdote, B. Theeconomic approach to social capital.Cambridge: National Bureau of EconomicResearch, 2000 (NBER working paper 7728).

Gottfredson, M. R.; Hirschi, T. A general theory ofcrime. Stanford CA: Stanford University Press,1990.

Gramsci, A. Selections from the prison notebooks.New York: International Publishers, 1971(Translated and edited by Hoare, Q.;Smith, G. N.).

Green, A. Education and state formation: the rise ofeducation systems in England, France and theUSA. Basingstoke: Macmillan, 1990.

Green, A.; Preston, J. Education and social cohe-sion: re-centering the debate. Peabody Journalof Education, 2001, Vol. 76, No 3-4, p. 247-284.

Green, A.; Sakamoto, A. Models of high skills innational competition strategies. In: Brown, P.;Green, A.; Lauder, H. (eds) High skills: globali-sation, competitiveness and skill formation.Oxford: Oxford University Press, 2001.

Gundelach, P.; Torpe, L. Voluntary associations.new types of involvement and democracy.Paper prepared for the ECPR workshop on‘Social involvement, voluntary associationsand democratic politics’, Oslo, 29 March to3 April 1996.

Haegel, F. The effect of education on the expressionof negative views towards immigrants inFrance: the influence of the republican modelput to the test. In: Hagendoorn, L.; Nekuee, S.(eds) Education and racism: a cross-nationalinventory of positive effects on education andracial tolerance. Aldershot: Ashgate, 1999.

Hagendoorn, L. Introduction: a model of the effectsof education on prejudice and racism. In:Hagendoorn, L.; Nekuee, S. (eds) Educationand racism: a cross-national inventory of posi-tive effects on education and racial tolerance.Aldershot: Ashgate, 1999.

Hagendoorn, L.; Nekuee, S. (eds) Education andracism: a cross-national inventory of positiveeffects on education and racial tolerance. Alder-shot: Ashgate, 1999.

Haines, K. Crime is a social problem. EuropeanJournal on Criminal Policy and Research.Dordrecht: Kluwer, 1999, Vol. 7, No2, p.263-275.

Hall, P. Social capital in Britain. British Journal ofPolitics, 1999, Vol. 29, p. 417-461.

Halman, L. Variations in tolerance levels in Europe:evidence from the eurobarometers and Euro-pean values study. European Journal on Crim-inal Policy and Research. Dordrecht: Kluwer,1994, 2-3, p. 15-38.

Hatcher, R. Class differentiation in education:rational choices? British Journal of Sociology ofEducation. Philadelphia: Carfax publishing,1998, Vol. 19, No 1, p. 5-24.

Hepburn, M. Multiculturalism and social cohesionin a democratic society: is the United Statesexperience a model or example? Prospects.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 173

Geneva: IBE – International Bureau of Educa-tion 1992, Vol. 22, No 1, p. 80-90.

Heynemann, S. Measuring the impact of educationon social cohesion. Paper presented to theOxford International Conference on Educationand Development, September 2001.

Heynemann, S.; Todoric-Bebic, S. A renewed sensefor the purpose of schooling: the challenges ofeducation and social cohesion in Asia, Africa,Latin America, Europe and Central Asia.Prospects. Geneva: IBE – International Bureauof Education, 2000, Vol. 30, No 2, p. 145-166.

Home Office. Community cohesion: a report of theIndependent Review Team chaired byTed Cantle. London: Home Office, 2001.

Idriss, C. Challenge and change in the Germanvocational system since 1990. Oxford Reviewof Education. Philadelphia: Carfax publishing,2002, Vol. 28, No 4, p. 473-490.

Inglehart, R. Culture shift in advanced industrialsociety. Princeton: Princeton University Press,1990.

Interpol – International Criminal Police Organisa-tion. International crime statistics. Paris:Interpol, 1996.

Jasinska-Kania, A. The impact of education onracism in Poland compared with other Euro-pean countries. In: Hagendoorn, L.; Nekuee, S.(eds) Education and racism: a cross-nationalinventory of positive effects on education andracial tolerance. Aldershot: Ashgate, 1999.

Johnston,M.; Jowell,R. ‘Social capital and the socialfabric’ in national centre for social research.British social attitudes, 16th edition: who sharesnew labour’s values? Aldershot: Ashgate, 2000.

Jousten,M. Proceedings of the sixth European collo-quium on crime and criminal policy, Helsinki10-12 December 1998. European Journal onCriminal Policy and Research. Dordrecht:Kluwer, 1998, Vol. 7, No 2, p. 277-288.

Junger-Tas, J. Similarity and difference. EuropeanJournal on Criminal Policy and Research.Dordrecht: Kluwer, 2001, Vol. 9, No 3,p. 321-325.

Kaiser, G. The rediscovery of the offender: is there arevitalization of individualistic theory? Euro-pean Journal of Crime, Criminal Law and Crim-inal Justice, 1997, p. 364-376.

Kelly, M. Inequality and crime. The Review ofEconomics and Statistics. Cambridge: MITPress, 2000, Vol. 82, No 4, p. 530-539.

Kerner, H.-J.; Weitekamp, E.; Stelly, W. From childdelinquency to adult criminality: first results ofthe follow up of the Tübingen criminal behaviourStudy. Eurocriminology. Tübingen, Institut fürKriminologie, 1995, No 8-9, p. 127-162.

Killias, M.; Aebi, M. F. Crime trends in Europe from1990 to 1996: how Europe illustrates the limitsof the American experience. European Journalon Criminal Policy and Research, Dordrecht:Kluwer, 2000, Vol. 8, No 1, p. 43-63.

Knack, S.; Keefer, P. Does social capital have aneconomic payoff? A cross-country investiga-tion. Quarterly Journal of Economics.Cambridge: MIT Press Journals, 1997, Vol. 112,No 4, p. 1251-1288.

La Porta, R. et al. Trust in large organisations. Amer-ican Economics Association Papers andProceedings, May 1997, p. 333-338.

Lanier, M. M.; Henry, S. Essential criminology.Boulder, CO: Westview Press, 1998.

Laub, J. H.; Sampson, R. J. Turning points in the lifecourse: why change matters to the study ofcrime. Criminology. Columbus, Ohio: AmericanSociety of Criminology, 1993, Vol. 31, No 3,p. 301-325.

Lee, M. R. Concentrated poverty, race and homi-cide. The Sociological Quarterly. Berkeley: TheUniversity of California Press, 2000, Vol. 41,No 2, p. 189-206.

Lees, S. Carnal knowledge: rape on trial. London:Penguin, 1996.

Levin, J.; Rabrenovic, G. Hate crimes and ethnicconflict: an introduction. American BehavioralScientist. London: Sage Publications, 2001,Vol. 45, No 4, p. 574-587.

Lillbacka, R. On the nature of social capital: sometentative inquiries aiming at an individual-levelmeasurement strategy. Pro Facultate. Vasa:Faculty of Social and Caring Sciences at ÅboAkademi University, 2002, No 6 (In press).

Lin, N. Social capital theory: a theory of social struc-ture and social action. Cambridge: CambridgeUniversity Press, 2001.

Little, T. D. Means and covariance structures(MACS) analyses of cross-cultural data: prac-tical and theoretical issues. Multivariate Behav-ioral Research, 1997, Vol. 32, p. 53-76.

Lochner, L.; Moretti, E. The effect of education oncrime: evidence from prison inmates, arrestsand self-reports. National Bureau of EconomicResearch, 2001, working paper 8605.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 174

Lockwood, D. Solidarity and Schisms, ‘the problemof disorder’ in Durkheimian and Marxist Soci-ology. Oxford: Clarendon Press, 1992.

Lukes, S. Emile Durkheim: his life and work.Harmondsworth: Penguin, 1973.

Maloney, W. Contracting out the participation func-tion: social capital and chequebook participa-tion. In: Van Deth J. et al. (eds) Social capital andEuropean democracy. London: Routledge,1999.

Mann, K. Critical reflections on the ‘underclass’ andpoverty. In: Gough, I.; Olofsson, G. (eds) Capitalism and social cohesion: essays onexclusion and integration. London: MacmillanPress, 1999.

McMahon, W. Conceptual framework for the anal-ysis of the social benefits of lifelong learning.Education Economics. London: Routledge,1998, Vol. 6, No 3, p. 309-346.

McMahon, W. Education and development:measuring the social benefits. Oxford: OxfordUniversity Press, 2000.

Mesch, G.; Fishman, G. Entering the system, Araband Israelis differences in record closure.Journal of Research in Crime and DelinquencyLondon: Sage, 1999, Vol. 36, No 2, p. 175-193.

Messner, S. F. Societal development, social equalityand homicide. Social Forces. Chapel Hill, NC:University of North Carolina Press, 1982,Vol. 61.

Moran, L. J. Homophobic violence: the hiddeninjuries of class. In: Munt, S. (ed.) Culturalstudies and the working class: subject tochange. London: Cassell, 2000, p. 206-218.

Morrow, R. A.; Torres, C. A. Social theory andeducation: a critique of theories of social andcultural reproduction. Albany: State Universityof New York Press, 1995.

Mortensen, N. Mapping system integration andsocial integration. In: Gough, I.; Olofsson, G.(eds) Capitalism and social cohesion: essays onexclusion and integration. Basingstoke:Macmillan, 1999.

Mouzelis, N. Differentiation and marginalization inlate modernity. In: Gough, I.; Olofsson, G. (eds)Capitalism and social cohesion: essays onexclusion and integration. Basingstoke:Macmillan, 1999.

Newton, K. Social capital and democracy inmodern Europe. In: Van Deth J. et al. (eds)Social capital and European democracy.London: Routledge, 1999.

Newton, K.; Norris, P. Confidence in public institu-tions: faith, culture or performance? In: Susan J.et al. (eds) Disaffected democracies: what’stroubling the trilateral countries? Sussex:Princeton University Press, 2000.

Nickell, S.; Layard, R. Institutions and economicperformance. London: LSE – London School ofEconomics, 1998 (Discussion Paper).

Nie, N., Junn, J., Stehlik-Barry, K. Education anddemocratic citizenship in America. Chicago:University of Chicago Press, 1996.

Norris, P. Making democracies work: social capitaland civic engagement in 47 societies. Paperpresented at the Euresco conference on socialcapital: interdisciplinary perspectives, Univer-sity of Exeter, 15-20 September 2000.

Ødegård, E. Comparative research in the drug field.European Journal on Criminal Policy andResearch, Dordrecht: Kluwer, 1998, Vol. 6, No 3,p. 357-367.

OECD. Knowledge and skills for life. Paris: OECD,2001.

OECD and Statistics Canada. Literacy in the infor-mation age: final report of the International adultliteracy survey. Paris and Ottawa: OECD andStatistics Canada, 2000.

Oestreich, G. Neostoicism and the early modernstate. Cambridge: Cambridge University Press,1982.

Oestreich, G. Strukturprobleme des europäischenAbsolutismus. Vierteljahrschrift für Social- und Wirtschaftgeschichte, 1968, Vol. 55,p. 329-347.

Parry, G.; Moyser, G.; Day, N. Political participationand democracy in Britain. Cambridge:Cambridge University Press, 1992.

Parsons, T. The social system. New York: The FreePress, 1951.

Peri, P. Education and prejudice against immi-grants. In: Hagendoorn, L.; Nekuee, S. (eds)Education and racism: a cross-national inven-tory of positive effects on education and racialtolerance. Aldershot: Ashgate, 1999.

Pettigrew, T.; Meertens, R. Subtle and blatant prej-udice in Western Europe. European Journal ofSocial Psychology. New York: John Wiley andSons, 1995, Vol. 25, p. 57-75.

Plewis, I.; Preston, J. Evaluating the benefits of life-long learning: a synthesis. London: Institute ofEducation, 2001.

Prakash, S.; Selle, P. Investigating social capital.Comparative perspectives on participation, civil


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 175

society and local governance. Nordic Associa-tion for South Asian Studies conference, Voss,Norway, 20-22 September 2001.

Putnam, R. Bowling alone. New York: Simon andSchuster, 2000.

Putnam, R. Making democracy work: civic tradi-tions in modern Italy. Princeton: PrincetonUniversity Press, 1993.

Raaijmakers, Q. Opvattingen over politek enmaatschappij. In: Meeus, W.; Hart, H. (eds)Jongeren in Nederland. Een nationaal surveynaar ontwikkeling in de adolescentie en inter-generationele overdracht. Amersfoort:Academische Uitgeverij, 1993.

Ragin, C. Comparative sociology and the compar-ative method. International Journal of Compar-ative Sociology. Willowdale, Ontario: De SitterPublications, 1981, Vol. 22, No 1-2, p. 102-120.

Raphael, S.; Winter-Ebmer, R. Identifying the effectof unemployment on crime. Journal of Law andEconomics, 2001, Vol. 44, p. 259-283.

Rex, J.; Tomlinson, S. Colonial immigrants in aBritish city: a class analysis. London: Routledgeand Kegan Paul, 1979.

Ritzen, J.; Woolcock, M. Social cohesion, publicpolicy and economic growth: implications forcountries in transition. Address prepared for theABCDE-Europe (Annual Bank Conference onDevelopment Economics), Paris, 26-28 June2000a.

Ritzen, J.; Easterly, W.; Woolcock, M. On ‘good’politicians and ‘bad policies’: social cohesion,institutions and growth. Keynote address:56th Congress IIPF, Seville, Spain.28-31 August. 2000b.

Robson, G. Millwall football club: masculinity, raceand belonging. In: Munt, S. (ed.) Cultural studiesand the working class: subject to change.London: Cassell; 2000.

Roediger, D. R. The wages of whiteness: race andthe making of the American working class.London: Verso, 1991 (first edition).

Röhrs, H. Georg Kerschensteiner (1852-1932).Prospects: the Quarterly Review of Compara-tive Education. 1993, Vol. 23, Nos 3 and 4,p. 807-822.

Rutter, M. Beyond longitudinal data: causes,consequences, change and continuity. Journalof Consulting and Clinical Psychology, 1994,Vol. 62, p. 928-940.

Schuller, T. et al. Modelling and measuring the widerbenefits of education: an initial synthesis.London: Institute of Education, 2001.

Selle, P. The transformation of the voluntary sectorin Norway: a decline in social capital? In:Van Deth J. et al. (eds) Social capital and Euro-pean democracy. London: Routledge, 1999.

Sianesi, B; Van Reenen, J. The returns to education:a review of the macro-economic literature.London: Centre for the Economics of Educa-tion, 2000, paper 6.

Siim, B. Gender, citizenship and empowerment. In:Gough, I.; Olofsson, G. (eds) Capitalism andsocial cohesion: essays on exclusion and inte-gration. Basingstoke: Macmillan, 1999.

Skocpol, T.; Fiorna, M. Making sense of the civicengagement debate. In: Skocopol, T.;Fiorna, M. (eds) Civic engagement in Americandemocracy. Washington DC: Brookings Institu-tion Press, 1999.

Skocpol, T.; Somers, M. The uses of comparativehistory in macrosocial inquiry. ComparativeStudies in Sociology and History. Cambridge:Cambridge University Press, 1980, Vol. 22,No 2, p. 1743-197.

Sniderman, P.; Gould, E. R. Dynamics of politicalvalues: education and issues of tolerance. In:Hagendoorn, L.; Nekuee, S. (eds) Educationand racism: a cross-national inventory of posi-tive effects on education and racial tolerance.Aldershot: Ashgate, 1999.

Stolle, D.; Rochon, T. The myth of American excep-tionalism: a three-nation comparison of associ-ational membership and social capital. In:Van Deth J. et al. (eds) Social capital and Euro-pean democracy. London: Routledge, 1999.

Tönnies, F. Community and society [Gemeinschaftund Gesellschaft] (translated and edited byCharles P. Loomis). Michigan: University ofMichigan Press, 1957.

Torney-Purta, J. et al. Citizenship and education intwenty-eight countries. Civic knowledge andengagement at age fourteen. Amsterdam: Inter-national Association for the Evaluation ofEducational Achievement, 2001.

Van der Brug, H. Voetbalvandalisme. Haarlem: DeVriesborch, 1986.

Van Deth J. et al. (eds) Social capital and Europeandemocracy. London: Routledge, 1999.

Van Limbergen, K.; Colaers, C.; Walgrave, L.Research on the societal and psycho-sociolog-


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 176

ical background of football hooliganism.Leuven: Catholic University of Leuven, 1987.

Vazsony, A. et al. An empirical test of a general theoryof crime: a four-nation comparative study ofself-control and the prediction of deviance.Journal of Research in Crime and Delinquency.London: Sage, 2001, Vol. 38, No 2, p. 91-131.

Verberk, G.; Scheepers, P. Education, attitudestowards ethnic minorities and opposition to affir-mative action. In: Hagendoorn, L.; Nekuee, S.(eds) Education and racism: a cross-nationalinventory of positive effects on education andracial tolerance. Aldershot: Ashgate, 1999.

Walton, J. et al. Crime, migration and social changein north-west England and the Basque country.British Journal of Criminology. Oxford: OxfordUniversity Press, 1999, Vol. 39, No 1, p. 90-112.

Watts, M. Aggressive youth cultures and hatecrime: skinheads and xenophobic youth inGermany. American Behavioral Scientist.London: Sage Publications, 2001, Vol. 45, No 4,p. 600-615.

Wilkinson, R. Unhealthy societies: the afflictions ofinequality. London: Routledge, 1996.

Winkler, J. Explaining individual racial prejudice incontemporary Germany. In: Hagendoorn, L.;Nekuee, S. (eds) Education and racism: across-national inventory of positive effects oneducation and racial tolerance. Aldershot:Ashgate, 1999.

Witte, A. D. Crime. In: Behrman, J. R.; Stacey, N.(eds) The social benefits of education. AnnArbor: Michigan University Press, 1997.

Wittebrood, K.; Nieuwbeerta, P. Wages of sin? Thelink between offending, lifestyle and violentvictimisation. European Journal on CriminalPolicy and Research. Dordrecht: Kluwer, 1999,Vol. 7, No 1, p. 63-80.

World Bank. World development report: attackingpoverty. Oxford: Oxford University Press, 2001.

World Bank, Serageldin, I. Social capital: A multi-faceted perspective. Washington, DC: WorldBank, 1999, p. 1-424.

Young, M. Contrasting approaches to the role ofqualifications in the promotion of lifelonglearning. In: Evans, K.; Hodkinson, P.; Unwin, L.(eds) Working to learn: transforming learning inthe workplace. London: Kogan Page, 2002.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 177

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 178

Macroeconometric evaluation ofactive labour-market policy –

a case study for GermanyReinhard Hujer, Marco Caliendo, Christopher Zeiss

AbstractMost evaluation studies of active labour-market policies (ALMPs) focus on the microeconometric evalua-tion approach using individual data. However, as the microeconometric approach usually ignores impactson non-participants, it should be seen only as a first step to a complete evaluation to be followed byanalysis at macroeconomic level.Our study evaluates the impact of active labour-market programmes in Germany on the whole economy.The evaluation focuses on vocational training and additionally on subsidised employment programmes inorder to compare them to vocational training. To do this we outline the institutional structure of ALMPsin Germany and conduct a theoretical discussion that enables us see how ALMP might influence thewhole economy. This framework serves as a starting point for the empirical analysis, which uses regionaldata from 175 German labour office districts and a dynamic panel data estimator to evaluate the impactsof ALMP in Germany from 1999 to 2001.The empirical analysis relies on the Beveridge curve relationship, explaining total unemployment as afunction of vacancies and ALMP activity. With this empirical model we are able to estimate the overalleffect of ALMP on the economy. We use a dynamic specification that enables us to control for the highlypersistent pattern of German unemployment. Since the usual estimation methods are not appropriate indynamic panel data models, we use a system generalised method of moments (GMM) estimator that wasfound to perform well in the case of highly persistent data.Within the system GMM framework we account for the inherent simultaneity problem resulting from thefact that ALMP would normally be determined by a policy reaction function, i.e. the decision on howmuch is spent on ALMP is driven by the economic situation. This would result in a causal link from thedependent variable (unemployment or employment) to the ALMP measures which would result inmisleading estimates.Our results indicate different pictures in East and West Germany. Whereas in West Germany we find apositive result for vocational training, the results in East Germany do not look favourable. The results forEast Germany should be treated with caution since adequate estimators cannot be implemented due toa too small number of regional units. Comparing vocational training programmes in West Germany withjob creation schemes (Arbeitsbeschaffungsmaßnahmen, JCS), we additionally find that vocationaltraining is the more efficient measure in reducing unemployment.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 179

1. Executive summary 182

2. Introduction 184

3. ALMPs in Germany 186

4. Theoretical issues on the evaluation of ALMP 190

5. Previous empirical findings for Germany 194

6. Estimation methods 195

7. Empirical analysis 199

8. Implications for policy and future research 206

9. Conclusions 207


Annex 209

References 213

Table of contents

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 180

Tables

Table 1: Spending on labour-market policies in Germany, 1998-2000 187

Table 2: Unemployment rate in Germany, 1994-2000 188

Table 3: Estimate results for West Germany (OLS and within estimator) 199

Table 4: Estimate results for West Germany (GMM estimator) 200

Table 5: Estimate results for East Germany 203

Table 6: Descriptive statistics for Germany 209

Table 7: Descriptive statistics for West Germany 209

Table 8: Descriptive statistics for East Germany 210

Table 9: Cumulative lag coefficients for West Germany 210

Table 10: Cumulative lag coefficients for East Germany 211

Figures

Figure 1: The revised Beveridge curve 192

Figure 2: Cumulative lag coefficients for JCS in West Germany (GMM estimates) 201

Figure 3: Cumulative lag coefficients for FbW in West Germany (GMM estimates) 202

Figure 4: Cumulative lag coefficients for JCS in East Germany 204

Figure 5: Cumulative lag coefficients for SAS in East Germany (within estimates) 204

Figure 6: Cumulative lag coefficients for FbW in East Germany (within estimates) 205

Figure 7: Participants in ALMP and unemployed in East Germany 212

Figure 8: Participants in ALMP and unemployed in West Germany 212


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 181

We were interested in estimating the net effectsof ALMP in Germany. As microeconometric eval-uations usually ignore impacts on non-partici-pants, we used a macroeconometric approach toanalyse the effects of JCS, structural adjustmentschemes (Strukturanpassungsmaßnahmen, SAS),only for East Germany, and continuing vocationaltraining (Förderung der beruflichen Weiterbildung,FbW) on employment in Germany from 1999 to2001.

Our analysis starts with an overview of theinstitutional structure of ALMP in Germany.Labour-market policies in Germany are organisedby the Federal Employment Service (Bunde-sanstalt für Arbeit). Up to 1998, the legal basis forpolicy has been the work promotion act (Arbeits-förderungsgesetz, AFG), established in 1969.From that point, the new social welfare codeSGB III (Sozialgesetzbuch) took over this role.Changes have been made both in objectives,such as a more intensive focus on problemgroups in the labour market, and also in the insti-tutional organisation of labour-market policy,leading to decentralisation and more flexibility inthe regional allocation of resources to differentmeasures.

Since our data ranges from 1999 to 2001, wefocus only on SGB III which was created in aneconomic situation with high unemployment ratesand narrow budget constraints. The most impor-tant goal of ALMP within the framework of SGB IIIis the (re)integration of problem groups in theregular labour market while using resources in anefficient way. With respect to the organisationalset up of ALMP, SGB III allows local labour officesto allocate their budgets relatively freely todifferent measures. This leaves the decision on themix of instruments to the relevant regional branchof the Federal Employment Service (Brinkmann,1999). Therefore, any evaluation of the efficiencyof labour-market policy must give more considera-tion than before to regional flexibility.

In terms of spending, we find that FbW is themost important ALMP measure for both East andWest German regions, followed by traditionalJCS. In West Germany there is a big difference

between spending on FbW and JCS, whereas inEast Germany the situation is much morebalanced. Finally, we also consider structuraladjustment schemes which are similar to tradi-tional JCS but with less severe eligibility criteria.Since structural adjustment schemes are of minorimportance in West Germany they are only evalu-ated for East Germany.

As a starting point for our empirical analysis wepresent a theoretical framework which aims toidentify the various channels through whichALMP might influence the whole economy.Leaving aside the traditional way of ‘cheating thePhillips curve’, i.e. improving the unemploy-ment-inflation trade-off and thereby reducing thenon-accelerating inflation rate of unemployment,a model is needed that explains the relevantlabour-market variables (e.g. regular employmentor unemployment) and is also capable of incorpo-rating ALMP.

The most important requirement of a macroe-conomic model used to analyse ALMP is that it isable to explain a positive equilibrium unemploy-ment rate. Theoretical considerations of ALMPare, therefore, mostly based on the Layard andNickell framework or the search model frame-work.

Considering the traditional aim of ALMP, andespecially vocational training programmes, toovercome structural imbalances in the labourmarket, the search model framework seems mostadvantageous for analysing impacts. The basicmechanism within the search model framework isthat ALMP helps to increase the searching effi-ciency of participants and thus lowers unemploy-ment for a given set of vacancies, i.e. it improvesthe matching process. This might also affectlabour demand since vacancies become moreprofitable if they are filled more quickly. Otherimportant effects are reduced welfare, produc-tivity effects and effects on labour force participa-tion. Since data limitations do not allow estimatesof structural relationships we cannot analysethese effects separately.

Flow data is required in order to estimate theeffects of ALMP on the matching process. Unfor-

1. Executive summary

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 182

tunately such data is not available so we are util-ising the Beveridge curve relationship for ourempirical analysis. The Beveridge curve can beused as an indirect way of estimating matchingefficiency since it is the steady state relationshipbetween unemployment and vacancies resultingfrom the matching process.

For our empirical analysis we use a revisedBeveridge curve that relates job seekers, i.e. theunemployed and programme participants, tovacancies. Using the revised Beveridge curve isimportant since it avoids estimating the book-keeping effect and thus focuses on the net effectof ALMP.

Immense differences are found in terms ofprevious empirical findings from macroecono-metric evaluations for Germany. As Calmfors andSkedinger (1995) note, the exact specificationand the methods of estimation seem to be crucialto the results of a macroeconometric evaluation.This may result from the fact that ALMP only havemarginal impacts on the whole economy sincethey are mostly designed to affect a particulargroup in the labour market.

The availability of quarterly data allowed us to

use a dynamic panel data model, i.e. to takedynamics and persistency in the labour marketinto account. In order to control for the problemsarising in the dynamic panel data context weapplied a system GMM estimator where we alsoaccounted for the inherent simultaneity problemof ALMP. In order to solve the simultaneityproblem we use lagged levels of the ALMPmeasures as instruments. Although this choice ofinstruments may not be the best, it is a validoption. Other appropriate sets of instruments arenot available since they are either endogenous orsuffer from too little variation.

The fact that there are major differencesbetween the East and the West German labourmarkets made a separate estimate necessary forboth areas. But since the number of regionsavailable for East Germany is too small, thesystem GMM estimator cannot be implemented.Therefore our analysis for East Germany reliesonly on the within estimator (see box inChapter 6) and should be treated with cautionsince the within estimator does not account forendogenity of the lagged dependent and theALMP measures.

Macroeconometric evaluation of active labour-market policy

– a case study for Germany 183

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 183

In view of the immense spending on ALMPs inGermany (about DEM 43 billion in 2000) and theirdebatable success, evaluation literature has beengrowing in recent years (1). Most studies focus onthe microeconometric approach using individualdata. The importance of this approach is straight-forward and the framework for such an analysis iswell developed (Heckman et al., 1999). However,as the microeconometric approach usuallyignores impacts on non-participants it shouldonly be seen as a first step to a complete evalua-tion, to be followed by a macroeconometric anal-ysis. Instead of looking at the effect on individualperformance, we would like to know if ALMPrepresent a net gain to the whole economy. Thisis likely to be the case only if the total number ofjobs is positively affected by ALMP.

Two major strands of literature on how to esti-mate macroeconomic effects can be distinguished.First are the structural, general equilibrium models(e.g. Davidson and Woodbury, 1993; or Heckmanet al., 1998). The advantage of these studies is thatthey make explicit assumptions about the mecha-nism generating the general equilibrium effects andalso provide a framework that allows estimates ofmany evaluation parameters (Smith, 2000). Oneobvious disadvantage, however, lies in the strongidentifying assumptions they require (2). Second,authors like Forslund and Krueger (1994) prefer touse variation in programme scale across regionalunits (jurisdictions) combined with data at theregional level to estimate the effects. This can, inprinciple, also be done at national level withcross-country data, even though such an analysismight suffer from the heterogeneous policy

measures between the countries (e.g. Jackmanet al., 1990).

As we want to evaluate the effects of ALMP inGermany for recent years, the second strandseems more appropriate and will be described inChapter 4. In 1998 the legal basis for thelabour-market policy in Germany changed to thenew social welfare code SGB III. Changes havebeen made not only in the objectives, such as amore intensive focus on problem groups of thelabour market, but also in the institutional organi-sation of labour-market policy, leading to decen-tralisation and more flexibility in the regional allo-cation of resources to different measures. Thisdecentralisation allows an adjustment to the situ-ation in local labour markets while requiring anyevaluation to give more consideration to regionalflexibility than before. The importance of suitabledata, allowing regional heterogeneity to be takeninto account, has to be stressed. This is espe-cially problematic in Germany due to permanentadjustments in the regional delimitations of thelabour office districts (Arbeitsamtsbezirke). Incontrast, in other evaluation studies this is notproblematic because no such changes occurredin the time under consideration.

The aim of the study is to add a new perspec-tive to the evaluation of ALMP in Germany. This isdone by using regional data to obtain macroeco-nomic or net effects of these measures. Thereforewe focus especially on training programmes andcompare these with the alternative ALMPprogrammes JCS and SAS.

The remainder of this paper is organised asfollows. In the next chapter we give an overview ofALMP in Germany before we present some theoret-

2. Introduction

(1) See Hagen and Steiner (2000) or Hujer and Caliendo (2001) for extensive overviews regarding micro- and macroeconometricevaluations of ALMP in Germany.

(2) Those studies of the general equilibrium effects of ALMP are based on a general equilibrium model of the whole economy.Behavioural equations that are obtained from standard micro foundations describe the general equilibrium of the economy. Theeffects are obtained by calibrating the model, i.e. by imposing explicit functional forms for the basic equations and numericalvalues for the exogenous parameters. Solving the model allows the effects of ALMP to be estimated.Problems arise for the following reasons: first, the assumptions the model is built on must be correct; second, the imposedfunctional forms must be correct, e.g. should a utility function impose risk aversion or should it impose risk neutrality; third, thetrue values for the exogenous parameters must be used. Hypothetically these parameters should be estimated with econo-metric methods, but this often fails due to data limitations. Therefore, often arbitrary choices on those parameters are madeand thus the estimated effects of ALMP are questionable.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 184



ical issues on the macroeconometric evaluation ofALMP, which also includes a short comparisonbetween micro- and macroeconometric evaluationapproaches. Highlighting the advantages andshortcomings of each approach makes clear theirnecessity as additional ingredients to a completeevaluation. Furthermore we will present macroeco-

nomic theory on ALMP that is needed to build up theeconometric model. Following that, we reviewprevious empirical findings of macroeconomicevaluations in Germany before we offer our empir-ical analysis, where we will briefly present themethods of estimating and the results from sepa-rate estimates for East and West Germany.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 185

Labour-market policies in Germany are organisedby the Federal Employment Service (Bunde-sanstalt für Arbeit). Up to 1998, the legal basis forlabour-market policy in Germany was the workpromotion act, established in 1969. From 1998the new social welfare code SGB III took over thisrole. Changes have been made both in the objec-tives, such as a more intensive focus on problemgroups in the labour market, and also in the insti-tutional organisation of labour-market policy,leading to decentralisation and more flexibility inthe regional allocation of resources to differentmeasures.

Since the data we analyse ranges from 1999 to2001, we focus on SGB III and discuss AFG verybriefly. A good overview of AFG’s historical evolu-tion can be found in Staat (1997) (3). Theimprovement of the labour force structure, i.e. theadjustment of the labour supply to the changinglabour demand, was the primary goal in the earlyyears. It was intended to deal with the continualgrowth of the economy that changed thelabour-market conditions permanently, with acontinuous adjustment to the labour force struc-ture to fulfil the new requirements. A shortage ofjobs with specific (high-level) skills and an excessof jobs requiring low skills was to be avoided.

These goals had to be quickly revised at theend of 1973 when sharply rising unemployment,connected to the first oil price shock, refocusedattention. This becomes clear when we look atALMP participation. In the early 1970s, less than15 % of all participants had been unemployedbefore participation, whereas in the 1980s thiswas the case for almost 80 %.

After some innovations and amendments, AFGhad been replaced by SGB III in 1998. A goodoverview of the most relevant reforms can befound in Fitzenberger and Speckesser (2000) whopresent an extensive discussion of the newSGB III, especially regarding the responsibility ofemployees for their own labour-market success.

Fertig and Schmidt (2000) explain and classify thedifferent measures of employment promotion andexplicitly distinguish between non-discretionaryand discretionary measures. Brinkmann (1999)discusses aspects of decentralisation and region-alisation as well as the now mandatory outputevaluations.

Whereas the AFG was implemented under fullemployment conditions, SGB III came into beingin a harsher economic climate with labour-marketpolicy affected by narrower budget constraints.Some of the AFG’s objectives, such as securing ahigh employment rate and the avoidance oflow-quality employment, were dropped. The mostimportant goal (Paragraph 7.3, SGB III) is the(re)integration of problem groups into the regularlabour market while using resources in an effi-cient way (Grundsatz der Wirtschaftlichkeit undSparsamkeit). As the government sees itself onlyin a promoting role, SGB III places particularemphasis on the fact that employees have to acton their own authority regarding theirlabour-market success. This comes together witha tightening of the ‘reasonableness’ clause(Zumutbarkeitsklausel), which makes it harder forthe unemployed to turn down job offers.

Besides the change of the objectives, there havealso been organisational changes increasing theflexibility of ALMP on a regional and local level.Local labour offices are now allowed to allocatetheir budgets relatively freely to different measures.According to paragraph 71b, SGB IV, several cate-gories of ALMP must be financed by one singlebudget item (Eingliederungstitel), which is thenassigned to the regional labour office. The newfeature of SGB III is now that the labour offices arefree to set their priorities in relation to differentprogrammes. This leaves the decision on the mix ofinstruments to the regional branch of the FederalEmployment Service (Brinkmann, 1999), allowingadjustment to local labour-market conditions.Furthermore, 10 % of the budget can be used for

3. ALMPs in Germany

(3) The main goals can be found in paragraphs 1,2 AFG and have been: (a) securing high employment; (b) avoidance oflow-quality employment; (c) improvement of the structure of the labour force; (d) promotion of mobility; (e) social goals and (f)promotion of target groups.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 186

‘free promotion’ (Freie Förderung, paragraph 10,SGB III), allowing more individualised support. Eachlabour office has considerable flexibility to act withlocal focus, e.g. by implementing measures whichare custom-made for the local labour market.

Another promising feature is the so-called ‘inte-gration plan’ (Eingliederungsplan) to avoidlong-term unemployment. Under this new plan, thelocal labour exchange and the unemployed must,within six months of unemployment beginning,

establish which active measures or which action onthe part of the unemployed will best avoid a drift into(long-term) unemployment. Other interesting newmeasures, such as the special programme tocombat youth unemployment (JUMP), measuresimplemented in pathfinder regions aiming topromote employment of low-qualified individuals orthe long-term unemployed (CAST), and the reformlaw regarding the ALMP instruments (JOB-AQTIV),cannot be discussed here. For a comprehensive



1998 1999 2000

in billion % of in billion % of in billion % of DEM total DEM total DEM total

Germany

Total spending 133.18 135.29 125.96

Passive labour-market policies 85.32 64.1 81.19 60.0 73.93 58.7

Active labour-market policies 39.40 29.6 45.30 33.5 43.04 34.2

Continuing vocational training (FbW) 12.51 9.4 13.20 9.8 13.31 10.6

Job creation schemes (JCS) 7.43 5.6 7.81 5.8 7.20 5.7

Structural adjustment schemes (SAS) 1.47 1.1 1.84 1.1 1.40 1.1

Sam-East for private firms (SAS-East) 3.13 2.4 3.57 2.5 1.27 1.0

Free support (FS) 0.55 0.4 1.09 0.8 1.13 0.9

West Germany

Total spending 82.17 83.25 78.14







Free support (FS) 0.22 0.3 0.50 0.6 0.54 0.7

East Germany

Total spending 51.01 52.04 47.83







Free support (FS) 0.33 0.6 0.59 1.1 0.58 1.2

Source: Bundesanstalt für Arbeit (2000)

Table 1: Spending on labour-market policies in Germany, 1998-2000

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 187

overview see Fitzenberger and Hujer (2002).Another new point in the SGB III is mandatoryoutput evaluation. Labour offices are now requiredto draw up integration balances (Eingliederungsbi-lanzen, paragraph 11, SGB III), including, mostimportantly, the employment status of each partic-ipant some time after completion of a measure.Unfortunately the integration balance cannot beseen as a proper evaluation approach, since it doesnot compare the status of participants with thestatus of comparable non-participants. Thereforethe question of how the participants would havedone without the programme cannot be answeredwith an integration balance.

Any evaluation of the efficiency oflabour-market policy must give more considera-tion than before to regional flexibility and takeinto account the various support strategies devel-oped by the labour offices (Brinkmann, 1999).

Table 1 shows the spending on labour-marketpolicies in Germany from 1998 to 2000. Whereas in1998 only 29.6 % of total spending was dedicatedto active measures, the proportion rose to 34.2 % in2000. In West Germany, the proportion of ALMPsrose from 23.5 % to 30.6 %, though it remained rela-tively stable in East Germany at around 40 %. Oneobvious reason for the limited success in switchingresources into active measures is the constantlyhigh unemployment rate in East Germany (Table 2).As unemployment benefits are entitlementprogrammes and most active measures are discre-tionary in nature, the former increase automaticallywith a rising unemployment rate, whereas the latterare more easily discarded.

The most important measures in 2000 were thepromotion of further FbW with DEM 13.31 billionand subsidised employment, consisting of tradi-tional JCS with DEM 7.2 billion and SAS withDEM 2.67 billion.

In principle, public vocational training underAFG comprised three types of training measures,namely further training (Fortbildung), retraining(Umschulung) and training to familiarise with anew occupation (Einarbeitung) (4). The first twotypes have been summarised in one item (para-graphs 77-96, 153-159, 517, SGB III). The latter isnow part of employment subsidies and will not bediscussed here (5). The Federal EmploymentService pays the costs of the training measuresand a subsistence allowance (Unterhaltsgeld) tothe participants, which amount to 60 % (67 %with one or more children) of the previous netincome (equal to unemployment benefit). Themain goals are to reintegrate the unemployed byimproving their skills and to avert the danger ofunemployment for employees at risk.

The German subsidised employmentprogrammes consist of JCS and SAS. JCS is themore important programme in East and WestGermany, with expenditure in 2000 ofDEM 7.2 billion. JCS (paragraphs 260-271, SGB III)are normally only available to non-profit organisa-tions. They should support activities which are ofvalue to society and additional in nature, that iswithout the subsidy they could not be executed.Priority is given to projects which improve the chanceof permanent jobs, that support structural improve-ment in social or environmental services or that aimat the integration of extremely hard-to-place individ-uals. SAS (paragraphs 272-279, SGB III) is particu-larly prominent in East Germany, with spendingamounting to DEM 2.4 billion in 2000. Their goal is,analogous to JCS, integration into regular employ-ment, but less severe eligibility criteria apply toparticipants, so both the unemployed and individ-uals threatened by unemployment may participate.The SAS consist of a wage subsidy equal to theaverage amount of unemployment allowance or


1994 1995 1996 1997 1998 1999 2000

Germany 10.6 10.4 11.5 12.7 12.3 11.7 10.7

West Germany 9.2 9.3 10.1 11 10.5 9.9 8.7

East Germany 16 14.9 16.7 19.5 19.5 19 18.8

Source: Bundesanstalt für Arbeit (2000)

Table 2: Unemployment rate in Germany, 1994-2000

(4) See Hujer and Wellner (2000) for an overview of vocational training under the AFG.(5) See Hujer et al. (2001) for an overview.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 188

assistance (including contributions to the socialsecurity system) paid in the Federal territory.

Looking at the distribution of spending ondifferent measures, considerable differencesbetween East and West Germany become clear.The main measure used in the West is FbW,where expenditures amount to DEM 7.94 billion,corresponding to a share of 10.16 % of totalspending. The next important measures are JCSwith a share of 2.56 % and SAS with 0.36 %. InEast Germany, the situation is much morebalanced. Again, FbW is the most importantprogramme (DEM 5.37 billion, 11.24 %) but JCS(DEM 5.2 billion, 10.87 %) follow closely.

The discrepancy between both parts alsobecomes clear in Figures 7 and 8 in the annex.They show the number of participants in the threemost important programmes for East and WestGermany from April 1997 to April 2000. In WestGermany, an average of 200 000 individuals

participated in FbW in every quarter, 50 000 inJCS and around 8 000 in SAS. The ratio of partic-ipants to unemployed is 1:10.

As expected, the situation in East Germany isdifferent. First, participation in the differentmeasures is more balanced. The most importantmeasures are JCS with 143 000 participants onaverage, followed by SAS with 137 000 and FbWwith 134 000. The ratio of participants to unem-ployed is 1:3.

After looking at spending on the differentmeasures and the participating individuals, it isalso interesting to look at the average duration ofthe measures. This should give us important hintson the lag structure for our latter analysis. Theaverage duration for the measures under consid-eration in 1999 is between 8 and 10 months. JCShave the shortest duration with 8.3 months,followed by FbW (8.4 months) and SAS with9.8 months (Bundesanstalt für Arbeit, 2000).



01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 189

The ideal evaluation process consists of threesteps. First, the impact of the programme on theparticipating individual should be estimated.Second, it should be examined whether theimpacts are large enough to yield net socialgains. Finally, it should be determined if this is thebest outcome that could have been achieved forthe money spent (Fay, 1996).

The main question of microeconometric evalua-tion is whether the particular outcome variable foran individual is affected by participation in an ALMPprogramme. That being so, the direct gain can becompared with the associated costs and thesuccess of the programme can be judged.However, microeconometric approaches estimate,in nearly all cases, the effect of treatment on thetreated. One important concept in this context is thestable unit treatment value assumption (SUTVA)(Rubin, 1980). One implication of SUTVA is that theeffect of the intervention on each individual is notaffected by the participation decision of any otherindividual, i.e. the treatment effect for each personis independent of the treatment of other individuals.This assumption guarantees that average treat-ment effects can be estimated independently of thesize and composition of the treatment population.Among other things SUTVA excludes cross-effectsor general equilibrium effects. Even though itsvalidity facilitates a manageable formal setup, inpractical applications it is frequently questionablewhether it holds. If one looks at the immenseamounts spent on ALMP in Germany and the largescale of the programmes, spill-over effects onnon-participants are very likely. Therefore themicroeconometric approach is partially analyticaland should only be seen as one step in a completeevaluation or, as Heckman (1999) puts it, microdataare no panacea and must be used in conjunctionwith aggregate time-series data to estimate the fullgeneral-equilibrium consequences of policies.

Important negative effects that a microecono-metric evaluation ignores are deadweight lossesand substitution effects (6). If the outcome of theprogramme is not different from what would have

happened in its absence, we talk about a dead-weight loss. A common example is the hiringfrom the target group that would have occurredalso without the programme. If a worker is takenon by a firm in a subsidised job instead of anunsubsidised worker who would have been hiredotherwise, we talk about a substitution effect. Inthis context it is also possible that a worker isdismissed and replaced by a subsidised worker.The net short-term employment effect in this caseis zero. Such effects are likely in the case ofsubsidies for private-sector work. There is alwaysa risk that employers hold back ordinary jobcreation in order to be able to take advantage ofthe subsidies. Another problem might be thatALMP crowd out regular employment. This canbe seen as a generalisation of the so-calleddisplacement effect. This effect typically refers todisplacement in the product market, e.g. wherefirms with subsidised workers increase output butdisplace (reduce) output among firms that do nothave subsidised workers. Calmfors (1994) alsostresses the importance of tax effects in thesense that programmes have to be financed bytaxes which distort the choices of both partici-pants and non-participants.

To see how problems can arise if one neglectssuch effects let us discuss an example. As theoutcome variable of interest in Germany is usuallyemployment for participants, we consider a wagesubsidy programme that aims to increase theemployment probability of the long-term unem-ployed by giving employers a subsidy if they hire anindividual out of the target group. The usual microe-conometric approach analyses the effects on thelabour supply side of the market by looking at theindividual’s performance. Where the individual got ajob because of the subsidy, the programme was asuccess. The shortcoming is, however, the possibleoccurrence of substitution or displacement effects,e.g. if an unsubsidised worker is fired to hire asubsidised worker. Hujer et al. (2001) suggest esti-mating the success of wage subsidies on the labourdemand side of the market by looking at the employ-

4. Theoretical issues on the evaluation of ALMP

(6) For details see Hujer et al. (2002).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 190

ment situation within the firm. In this way, substitu-tion effects within the firm already ‘net out’ and givea clearer picture of the net effects. Displacementeffects between firms, however, cannot be detectedwith this approach. Clearly, these effects have to betaken into account if one intends to make state-ments about the net effect of ALMP.

As the outcome variable in a macroecono-metric evaluation is an aggregated variable,substitution and displacement effects plus thedeadweight losses can be taken into account.

Estimating the effects on aggregate variables isnot straightforward and, compared to the amountof micro-analysis, existing literature is relativelysmall. This might have several causes, such asthe inherent simultaneity problem or the avail-ability of suitable data. The issue of simultaneityarises because spending on ALMP should influ-ence the labour-market situation but might alsobe determined by it. The major obstacle,however, is the absence of an obvious theoreticalframework within which to couch the analysis.Leaving aside the traditional way of ‘cheating thePhillips curve’, i.e. improving the unemploy-ment-inflation trade-off and thereby reducing thenon-accelerating inflation rate of unemployment(Baily and Tobin, 1977), a model is needed thatexplains the relevant labour-market variables (e.g.regular employment or unemployment) and isalso capable of incorporating ALMP.

The most important requirement of a macroe-conomic model used to analyse ALMP is that ithas to be able to explain a positive equilibriumunemployment rate. Theoretical considerationson ALMP are, therefore, mostly based on theLayard and Nickell framework or the searchmodel framework (e.g. Pissarides, 2000). Bothmodels differ in their primary reason for equilib-rium unemployment. In the framework presentedby Layard and Nickell (1986), unemployment isgenerated through a wage setting process thatpushes wage rate over the equilibrium rate gener-ated by labour demand and supply. One possibleexplanation for these wage distortions is thepower of unions in the wage bargaining processor efficiency wages. Search models on the otherhand assign the cause of unemployment to atime and cost consuming matching process.Theoretical considerations on the impacts ofALMP in these two frameworks are given forexample by Calmfors and Lang (1995) and by

Holmlund and Lindén (1993). Additionally, bothframeworks can be combined to approximateusual labour-market situations more realistically(Calmfors and Lang, 1995).

Considering the traditional aims of ALMP, andespecially vocational training programmes, toovercome structural imbalances in the labourmarket, the search model framework seems mostadvantageous for analysing the impacts of ALMP.The core element of these models is a matchingfunction that determines the number of newrecruitments with respect to the stock of theeffective job seekers and the stock of vacancies.Effective job seekers are mostly defined as theweighted sum of the programme participants andthe unemployed, where the weights are given bythe search efficiency of the respective group(Holmlund and Lindén, 1993) (Calmfors and Lang,1995). Search efficiency can be thought of as aparameter that summarises various characteris-tics that determine the efficiency of job seekers infinding a successful match. This includes searchintensity, i.e. how much effort a job seeker putsinto actively searching for a job, or the level ofeducation or experience in the occupationalcareer of the job seeker. If ALMP is able toimprove the search efficiency of the group ofparticipants and if this group is large enough, theprogramme would be able to improve thematching process, i.e. to increase the number ofnew recruitments for a given stock of vacanciesand job seekers.

Calmfors (1994) gives three explanations ofhow ALMP can help to improve search efficiency.First, ALMP can increase the search intensity ofprogramme participants by encouraging them tosearch more actively. Second, ALMP canupgrade the skills of participants and adjust themto labour demand. Third, participation can serveas a substitute for regular work experience whichcan reduce the employer’s uncertainty about theemployability of the job applicant.

However, there are also negative effects onsearch efficiency if the programme implies fulltime engagement. In the case of full time employ-ment, ALMP programmes should be expected toreduce search efficiency because the effectivetime available for an active job-search would bereduced. But this locking-in effect reduces thesearch efficiency only when participants are in theprogramme. After the programme expires, the



01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 191

search efficiency of participants should be higherrelative to other unemployed.

The usual strategy for evaluating the effect ofALMP on the matching process would be to esti-mate the matching function itself, i.e. to regressnew recruitments against the number of unem-ployed, programme participants and vacancies.Unfortunately, no reliable data on new recruit-ments are available to allow for unbiased esti-mates for the parameters of the matching func-tion. Furthermore, an estimate of the matchingfunction would not deliver a complete picture ofthe effects of ALMP, as it only considers theeffects on inflows into employment; the effects ofALMP on existing jobs are omitted. If, forexample, ALMP helps to increase the level ofhiring, we do not know if such hiring only replacesexisting jobs, leaving employment constant.

The Beveridge curve (i.e. the unemployment-

vacancy relationship) can be used to circumvent thisproblem. The Beveridge curve is a steady statecondition where the outflow from employmentequals new intake. Since the matching functionensures the simultaneous coexistence of unemploy-ment and vacancies, the Beveridge curve resultsdirectly from the existence of the matching function.Therefore, as Petrongolo and Pissarides (1999) note,the Beveridge curve can be used as an indirect wayestimating matching efficiency.

For our empirical analysis we will follow Calm-fors (1994) and use the revised Beveridge curvewhich relates job seekers, i.e. the unemployedand programme participants to vacancies. Theutilisation of the revised Beveridge curve isimportant since it avoids estimating the book-keeping effect. In particular, a rise in programmeparticipation is usually associated with a propor-tional decrease in the unemployed (7). Therefore,


(7) Note that this depends if unemployment is an eligibility criteria for being placed into a programme.

Figure 1: The revised Beveridge curve

Vacancy rate

Unemployed and programme participants as proportion of the labour force

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 192

an application of the conventional unemploymentvacancy relationship would lead to an overesti-mation of the effects of ALMP. The revisedBeveridge curve (i.e. the relationship between thevacancies and the sum of openly unemployedand programme participants) therefore allows anestimate of the net effect of ALMP.

In Figure 1 we have plotted the revisedBeveridge curve, i.e. the relationship between theunemployed plus the programme participantsand the vacancies. If the programme participantscome from the unemployed, a programme expan-sion would not lead to a shift of the revisedBeveridge curve in Figure 1.

But if ALMP improves search efficiency, andtherefore matching efficiency, there would be ashift of the revised Beveridge curve to the originbecause the number of vacancies and jobseekers is reduced simultaneously.

But there may also be consequences forexisting jobs, in that ALMP can tend to increasewages or to change productivity (8). The mostimportant effects of ALMP with respect to wagesare reduced welfare losses where ALMP partici-pation is associated with an expected higherincome compared to being unemployed, so thewelfare losses of being unemployed are reduced.The expected higher income might result from ahigher compensation level compared to unem-ployment benefits and/or from better job oppor-

tunities resulting from programme participation.In a wage bargaining process between workersand firms the expected higher income leads to anincrease in wage claims of workers, and so tohigher wage rate.

The productivity effect of ALMP programmesmay be assigned to the fact that participationserves as a substitute for work experience. In thecase of training programmes that are aimed atimproving skills, an impact on the productivity ispart of the objectives.

An important point in this context is thatproductivity effects depend heavily on the type ofparticipant. If the skill level of participants is rela-tively low, a training programme has greatereffects on productivity than if the skill level ofparticipants is relatively high.

Since the effects on productivity and wage ratealso affect the Beveridge curve we cannot observethe pure effect resulting from changes in thematching process. Implementing the revisedBeveridge curve as an econometric model relatesmore to a reduced form model which can be usedto estimate the overall effect. In theory, the impor-tant effects of ALMP besides the effect on thematching process are reduced welfare losses ofunemployment, effects on productivity and labourforce participation. Since data limitations do notallow for the estimation of structural relationshipswe cannot analyse these effects separately.



(8) See Calmfors (1994) for an extensive theoretical discussion of the effects of ALMP.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 193

We will now give a brief overview of empirical find-ings from macroeconometric evaluations forGermany on a regional level.

Büttner and Prey (1998) use yearly data (from1986 to 1993) from 74 planning regions of WestGermany to evaluate the effects of trainingprogrammes and public sector job creation onlabour-market efficiency. They use a disequilibriumapproach and their results suggest that trainingprogrammes have no effect and job creationprogrammes have a significant positive effect onmatching efficiency.

Prey (1999) extends this work by additionallycontrolling for regional age structure and recipientsof welfare assistance and estimating separately formen and women. She finds that vocational trainingincreases (decreases) the mismatch for women(men), whereas JCS decrease the mismatch formen. Thus she finds an opposite effect for men andwomen.

Pannenberg and Schwarze (1998) use the datafrom 35 local labour office districts to evaluatetraining programmes in East Germany. They usemonthly data from 1992 to 1994 and find that theprogrammes have negative effects on regionalwages.

Steiner et al. (1998) examine the effects of voca-tional training on labour-market mismatch usingdata from 35 local labour office districts in EastGermany. They observe only very small effects onmatching efficiency which disappear in thelong-run.

Schmid et al. (2000) use yearly data from142 local labour office districts to estimate theeffects of further training, retraining, public sectorjob creation and wage subsidies on long-termunemployment from 1994 to 1997. They find thatjob creation programmes reduce only ‘short’

long-term unemployment (6-24 months), whereasvocational training reduces long-term unemploy-ment (>24 months).

Hagen and Steiner (2000) evaluate vocationaltraining, JCS and SAS in East and West Germanyusing data from local labour office districts. Thetime period under consideration differs and rangesfrom 1990 to 1999. The estimated net effects are notvery promising as all measures increase unemploy-ment in West Germany. Only SAS reduces theunemployment rate slightly in East Germany,whereas JCS and vocational training increase it.

Blien et al. (2002) analyse the effects of ALMP onthe development of regional employment in EasternGermany and find positive impacts. They usedetailed data from employment statistics covering1993 to 1999. Their regional units are 112 districts(Landkreise/ kreisfreie Städte). Their method is aneconometric equivalent to conventional shift-shareanalysis (based on constrained regression), whichis extended to include many determining variables.

These studies demonstrate immense differencesin the results. The inherent simultaneity problem ofALMP may serve as an explanation, since it is notevident how to tackle this problem. Furthermore allempirical models relate to different theoreticalbackgrounds, i.e. they use different specificationsfor the empirical analysis. As Calmfors andSkedinger (1995) note, the exact specification andthe methods of estimation seem to be crucial for theresults of a macroeconometric evaluation. This mayresult from the fact that ALMP have marginalimpacts on the whole economy, since they aremostly designed to affect a particular group in thelabour market. Therefore, it is not only an appro-priate econometric model that is necessary to iden-tify the effects of ALMP on the whole economy butalso appropriate methods of estimating.

5. Previous empirical findings for Germany

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 194

The empirical analysis undertaken by the authorsexploits a pooled time-series cross-section dataset for the German labour office districts. This datawas taken from administrative processes in theFederal Employment Service. The cross section inthis data set comes from the administrative areas ofthe local offices of the Federal Employment Servicein Nuremberg. Putting East and West Germanytogether we have 175 regional units. As discussed

in Chapter 3, the immense differences between theEast and the West German labour markets make itnecessary to analyse both areas separately. Thenumber of cross sections for West Germany areN=141 and N=34 for East Germany (9).

The time span ranges from the first quarter1999 to the fourth quarter 2001, leaving us T=12observations for each labour office district. Thebasic equation we want to estimate is given by

6. Estimation methods

Ψ ,ititit uXLb ++ )(Ø 3itit LbvLbasLc ++= )()()( 210

where sit denotes the total rate of job seekers(JSR) at time t for region i, that is defined as thesum of unemployed and participants in ALMPprogrammes relative to the labour force. νit is thevacancy rate (VR), i.e. the number of vacanciesrelative to the labour force. Ψit is a vectorcontaining the measures for the activity of ALMPprogrammes JCS, SAS and FbW. ALMP activityis measured with so-called accommodation ratiosthat are defined as the sum of programme partic-ipants relative to the stock of job seekers. Χit is avector that summarises other explanatory vari-ables, such as national unemployment (NUR) andnational vacancy rate (NVR) as well as seasonaldummies to control for the cyclic pattern in quar-terly data.

c(L) = 1 – c1L – c2L2 – … – cpL

p is a polynomialin the lag operator associated with the job seekerrate, with p as maximum lag, and bj(L) = bj0 + bj1L+ bj2L

2 + … + cjqLq is a polynomial in the lag oper-

ator associated to the vacancy rate, the ALMPmeasures and the other explanatory variables,with q as maximum lag. Therefore q need not bethe same for all variables. The imposed dynamicspecification enables us to control for the highlypersistent pattern of the unemployment rate.Since the average duration of German ALMPprogrammes is about 8 to 10 months, the inclu-sion of several lags for the ALMP measuresseems to be crucial for identification.

The job seeker rate is included as explanatorywith four lags. The vacancy rate and ALMPmeasures are included up to the fourth lag andthe national variables are included with the firstand the second lag. Therefore, we cover exactlyone year using quarterly data.

While attempting to make clear statementsabout the effects of ALMP measures on the jobseeker rate, the interpretation of the parameterswith the inclusion of four lags may be a bitcumbersome.

A more straightforward interpretation of resultsis the use of so-called lag coefficients. Lag coef-ficients describe the impact of an ALMP expan-sion in t on the job seeker rate after t+q quarters,e.g. the second lag coefficients gives the changeof the job seeker rate resulting from an ALMPexpansion two quarters ago (10).

The overall effect of an ALMP expansion, i.e.the total change in job seeker rate after q quar-ters is then simply given by the sum of the lagcoefficients up to the q’th quarter. We use thecumulative lag coefficients to make a generalstatement about the effects of an ALMP expan-sion through time.

For the residual uit we assume a one way errorcomponent structure, i.e. uit

= µi+ εit, with µi as an

unobserved regional specific effect and εit as aresidual varying over the regions and time.

The empirical model relates to a revised

(9) Due to data limitations, Berlin is excluded from the analysis.(10) See Greene (2000) for the calculation of the lag coefficients.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 195

Beveridge curve, explaining total unemployment(i.e. the job seeker rate) as a function of vacanciesand the ALMP measures. There are two majorconcerns in estimating the effects of ALMP. First,our model is a dynamic panel data model whereusual estimation methods such as ordinary leastsquares (OLS) or the within transformation arebiased (e.g. Baltagi, 2001). The bias of the OLS esti-mator results from the non-zero correlation of theexplanatory variables with regional specific effects,i.e. we would face a violation of the classical OLSassumptions (Greene, 2000). As discussed inBaltagi (2001) the OLS estimator tends therefore tobe upward biased with respect to the autoregres-sive parameters. In contrast to the OLS estimator,the within estimator accounts for the correlationwith the regional specific effect by transforming thedata so that the regional specific effect is removed.

This transformation is appropriate for static models(i.e. no lagged dependent variable is included in theeconometric model) but leads to biased estimatesin the case of dynamic panel data models.

A second issue is that the ALMP measures inΨit should be expected to be determined by apolicy reaction function, i.e. the job seeker rateand the ALMP measures are determined simulta-neously.

In order to handle both problems we apply thesystem GMM estimator suggested by Blundelland Bond (1998). In our macro application wemust expect that those variables that vary overthe regions (i.e. JSR, VR, JCS, SAS and FbW) areall correlated with the regional specific effect µi. Ifwe consider the first differences residual

∆uit = c(L)∆sit – a0 – b1(L)∆vit – b2(L)∆Ψit – b3(L)∆Χit


The econometric model used for estimating the effects of ALMP is a dynamic panel data model.Dynamic refers to a specification where the lagged values of the dependent variable (the job seekerrate) are used as explanatory variables. A panel data model uses data where cross sectional data(observations for different regions) and time series data (a sequence of observations in time) arecombined.

OLS estimator

The OLS (ordinary least squares) estimator is a simple estimation method that ignores the regionalspecific differences in the data. In particular, if these differences are not fully represented by theexplanatory variables but are correlated with them, the OLS estimator leads to biased estimates. Inthe case of a dynamic model this problem is particularly severe since a regional specific effect that isconstant in time affects the actual and lagged job seeker rate. Therefore the explanatory variable is apriori correlated with the regional specific effect.

Within estimator

The within estimator utilises a data transformation that ensures that the regional specific effect isremoved. Application of the OLS estimator on the transformed data leads to the within estimator.Unfortunately, in the case of a dynamic model, the within transformation produces a correlationbetween the lagged dependent variable and the residual term. This correlation leads to biased esti-mates when the within estimator is used for dynamic models.

System GMM estimator

Usual panel data estimation methods are not valid for dynamic models. In order to obtain reliable esti-mates the system GMM estimator is a useful alternative. The basic idea is to combine the modelequation in first differences with the model equation in levels. Since there is a considerable loss ofinformation due to the first differencing the additional equations in levels improve the performance ofthe System GMM estimator. In contrast to the OLS and the within estimator the system GMM esti-mator results in consistent estimates for dynamic panel data models.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 196

where ∆ is the first differenced operator, i.e. ∆Χit=

Χit– Χit–1, we see that the regional specific effect is

removed from the first differenced residual.Therefore the system GMM estimator usesmoment conditions that are based on the equa-tions in first differences. Following Arellano andBond (1991) we can set up the following momentconditions for the lagged dependent variable:

∆ ... .

Note that these moment conditions require thatthere is no serial correlation in εit. We assume thatthe ALMP measures are endogenously deter-mined, and thus we can set up the momentconditions analogously to the lagged dependentvariable.

Ψ ∆ ... .

Due to the utilisation of this moment condition weuse lagged levels of the ALMP measures asinstruments for the current and the first laggedALMP measure. Thinking about ALMP determinedby a policy reaction function, the question ariseswhether there may exist better instruments totackle the simultaneity problem. Calmfors andSkedinger (1995) for example proposed the utili-sation of political measures such as the share ofseats in the parliament assigned to the left parties.Unfortunately, for Germany such a measure wouldsuffer from too small variation. Since electionsusually take place every four years, the timedimension of the data set does not even coverone legislature period. Furthermore, the regionalunits are the administrative offices of the labouroffice whereas political data is on country orLänder level. Therefore political data would notvary significantly in the time dimension as well asin the cross section. In addition, the labour officeis a tripartite body and not governed by nationalgovernment, but also by social partners.

Other possible instruments could be regionalvariables resulting from the administrativeprocesses of local labour offices, such as youth

unemployment or long-term unemployment.Unfortunately, a strong interdependency is likelybetween these variables and the job seeker rate,i.e. these variables are also endogenous and thuscannot be valid instruments.

Therefore we use only lagged levels of theALMP measures as instruments, bearing in mindthat although these instruments may not be thebest instruments, they are valid.

We assume that the vacancy rate is alsoendogenously determined. Standard searchtheory predicts that the number of vacanciessupplied by firms depends on the labour market.In particular the number of vacancies supplied byfirms depends on the unemployment rate since ahigher unemployment rate implies a shorter timein which vacancies remain open. This is equiva-lent to a cost reduction that makes a vacancymore profitable. Bearing these theoretical consid-erations in mind we assume that the vacancy rateis endogenously determined and thus thefollowing moment conditions are valid:

∆ ... .

For the variables in Χit that contain no regionalinformation we simply use them to instrumentthemselves, i.e. we imply that the followingorthogonality conditions must hold:

∆ ∆ .

Furthermore the system GMM estimator usesmoment conditions that are based on the equa-tions in levels, i.e. the moment conditions refer tothe level residual given by:

uit = c(L)sit – a0 – b1(L)vit – b2(L)Ψit – b3(L)Χit,The following moment conditions relate to theassumption that the regional variables sit, Ψit andνit in first differences are not correlated with theregional specific effect µit. In this case we can usethe variables in first differences as instruments,i.e. the following moment conditions are valid:

0) =ituÄÄ itX(E

12 and , −≤≤ tgT,6for ); =tuit( −vE git

12 and , −≤≤ tgT,6for );Ä =tuitØ − gitØ(E

12 and , −≤≤ tgT,6for ); =tu it( −sE git



∆Ψ ... ...

∆ ... ...

∆v ... ... 4,,1 and , =gT,6for ); =− tu itgitÄ(E4,,1 and , =gT,6for ); =− tus itgitÄ(E4,,1 and , =gT,6for );Ø =− tuitgitÄ(E

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 197

Finally, we again use the variables in Χit simply asinstruments for themselves, i.e. we impose themoment conditions:

This moment condition states simply that wepredict the Χit with themselves since they are notcorrelated with the residual uit.

For the calculation of the system GMM esti-mator, the indicated set of moment conditions isused (11). Furthermore, Monte Carlos simulationshave shown that the heteroscedasticity consis-tent two-step estimates for the system GMM esti-mator are downward biased (12). In order to makereliable inference we use a finite sample correc-tion for the asymptotic variance of the two stepestimates as proposed by Windmeijer (2000). In

order to test the sensitivity of the results from thesystem GMM estimator, we additionally presentthe results from the OLS and the within estimator.

Unfortunately, the system GMM estimatorcannot be calculated for East Germany sincethere are not enough cross sections at hand. Aswe have only 34 cross sections but 12 timeobservations at hand, the number of momentconditions becomes too large relative to thenumber of cross sections and thus the estimatesare not reliable. In order to make some state-ments about East Germany, we calculate OLSand the within estimator. But the results fromboth estimators should be treated with cautionbecause they do not control for the problems inthe dynamic panel data context as well as for thesimultaneity problem of ALMP.

0)( =itituXE


(11) For details see Blundell and Bond (1998) or Blundell et al. (2000).(12) For system GMM estimator there exist a one- and a two-step estimator. Whereas the one-step estimator is obtained from a

restricted variance-covariance matrix, the two-step estimates are obtained from an unrestricted one. The unrestricted variance-covariance matrix is estimated from the one-step estimates (for details see Blundell et al. (2000).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 198

Estimating the effects of ALMP on the job seekerrate is done separately for East and West Germany.For both regions the estimates are done with thesame model, with the exception that for West

Germany the SAS are not included. This is becausethe SAS are of minor importance in West Germany.Tables 7 and 8 in the annex show that there are, onaverage, only 62 people in SAS programmes in

7. Empirical analysis

OLS estimator Within estimator

Variable Param. t-value Param. t-value

Const. 0.2778 6.773 - -

Job seeker rate JSRt-1 0.8480 16.587 0.4508 5.509

JSRt-2 -0.1208 -1.603 -0.2583 -4.034

JSRt-3 0.1222 4.051 -0.0121 -0.476

JSRt-4 0.1157 0.974 -0.2020 -0.667

Vacancy rate VRt 0.0661 0.330 0.0007 0.006

VRt-1 -0.1083 -1.502 -0.1962 -3.356

VRt-2 0.0579 1.750 -0.0645 -1.430

VRt-3 0.0226 0.894 -0.0359 -0.618

VRt-4 -0.0987 -0.596 -0.1971 -1.709

Participants in job creation schemes JCSt -0.3983 -6.192 -0.3753 -5.251

JCSt-1 0.3966 4.132 0.2655 3.701

JCSt-2 0.1783 2.048 0.2293 1.774

JCSt-3 -0.0742 -1.581 0.0010 0.017

JCSt-4 -0.0783 -1.460 -0.1318 -1.280

Participants in vocational training FbWt -0.1119 -4.862 -0.1449 -4.351

FbWt-1 0.1487 5.522 0.1144 6.322

FbWt-2 -0.0541 -3.226 -0.0537 2.680

FbWt-3 0.0327 2.136 0.0106 0.745

FbWt-4 -0.0472 -2.305 -0.0710 -2.567

National unemployed rate NURt-1 -1.9988 -3.708 0.0309 0.034

NURt-2 -0.1583 -0.547 1.1980 1.874

National vacancies rate NVRt-1 -3.0450 -2.527 2.3923 0.709

NVRt-2 -2.9899 -2.494 -0.7527 0.923

Seasonal dummy 1 SD1 -0.0094 -7.332 0.0063 0.897

Seasonal dummy 2 SD2 0.0058 0.676 -0.0022 -0.295


Wald test of joint significance [26] 156544.178 [26] 7716.9457

First-order serial correlation [141] 3.95 [141] -4.136

Second-order serial correlation [141] 1.97 [141] -0.133

Table 3: Estimate results for West Germany (OLS and within estimator)

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 199

West Germany, whereas East Germany has 3 105.The other ALMP measures, namely JCS and FbWare included for both regions.

The results for West Germany resulting fromthe OLS and the within estimator are reported inTable 3. Considering the coefficients for the jobseeker rate, we find a strong persistent pattern

for the OLS estimator and a weak persistentpattern for the within estimator. Therefore theestimates from the within estimator predict thatafter a unit shock (i.e. a one-shot increase) thejob seeker rate adjusts much faster to the primarylevel (i.e. the level before the shock) thancompared with the OLS estimates. This coincides


System GMM estimator

Variable Param. t-value

Const. 0.3267 6.123

Job seeker rate JSRt-1 0.8591 21.909

JSRt-2 -0.0928 -1.237

JSRt-3 0.0895 2.759

JSRt-4 0.1044 0.998

Vacancy rate VRt 0.1784 0.526

VRt-1 -0.1086 -1.090

VRt-2 0.0438 0.873

VRt-3 0.0303 0.760

VRt-4 -0.1825 -0.699

Participants in job creation schemes JCSt -0.7612 -5.568

JCSt-1 0.7873 5.198

JCSt-2 0.0004 0.004

JCSt-3 0.0833 1.316

JCSt-4 -0.0941 -1.686

Participants in vocational training FbWt -0.2068 -4.712

FbWt-1 0.2217 6.617

FbWt-2 -0.0707 -3.308

FbWt-3 0.0245 1.176

FbWt-4 -0.0293 -1.334

National unemployed rate NURt-1 -2.0112 -3.143

NURt-2 -0.4238 -1.051

National vacancies rate NVRt-1 -1.7853 -1.417

NVRt-2 -5.7245 -3.786

Seasonal dummy 1 SD1 -0.0105 -6.362



Wald test of joint significance [26] 34822.626

Sargan test [241] 124.871

First-order serial correlation [141] -2.189

NB: Parameters are the two-step estimates of the system GMM estimatorCorrected standard errors as suggested by Windmeijer (2000)

Table 4: Estimate results for West Germany (GMM estimator)

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 200

with the insight that, with respect to the autore-gressive parameters, the OLS estimator tends tobe upward biased and the within estimator down-ward biased.

Turning to the results from the system GMMestimator in Table 4, we find that the cumulativecoefficients for the lagged dependent variableimply a persistent pattern that is slightly belowthe pattern of the OLS estimator. This means thatthe estimates from the system GMM estimatorsuggest a faster adjustment after a unit shockcompared to the OLS estimates. Therefore theresults from the system GMM estimator lookmost favourable.

There is no clear negative relationship betweenthe vacancy rate and the job seeker rate for allestimators. In particular, the coefficients on thevacancies resulting from the system GMM esti-mator are entirely insignificant and thus the exis-tence of any relationship between the total unem-ployment rate and the vacancy rate as predictedby the Beveridge curve is not observable. It isimportant to note in this context that the vacancyrate resulting from the administrative data doesnot contain all vacancies in the economy. This isbecause not all open vacancies are reported tothe labour office and therefore the vacancy rate

describes only a fraction of the labour market.Hence one should be cautious when making anystatements about the existence of the Beveridgecurve when relying on administrative data for thevacancy rate.

Turning to the coefficients for the ALMPmeasures, we find that the signs of the coeffi-cients for JCS differ only for the third lag, but thecoefficient is not significant. Considering themagnitude, the system GMM estimator impliesmuch stronger effects from the contemporaneousand the first lagged job creation schememeasure. For the coefficients on FbW, there areno differences in the signs and the magnitude israther similar between the different estimators.

We choose to calculate the lag coefficients upto t+6 quarters and to summarise the effectbeyond this timeframe with the long run effect (13).Therefore we can analyse the effects up to 1.5years after a programme expansion has occurred.The cumulative lag coefficients resulting from thesystem GMM estimates are plotted in Figure 2 forthe JCS and in Figure 3 for FbW.

For the JCS we observe an immediate signifi-cantly negative impact on the job seeker rate (hereand in the following a negative implies an intendedpositive effect, i.e. reduction of unemployment).



(13) See Greene (2000) for the calculation of the long run effect.

Figure 2: Cumulative lag coefficients for JCS in West Germany (GMM estimates)

-1.20

-0.80

-0.40

0.00

0.40

0.80

1.20

t t+1 t+2 t+3 t+4 t+5 t+6

Quarters

Jo

b s

ee

ke

r ra

te

Upper confidence band (95 %)

Cumulated lag coefficients

Lower confidence band (95%)

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 201

The effect remains negative for two following quar-ters and becomes insignificant in the long run.Therefore we can conclude that the JCS in WestGermany are able to improve the situation in thelabour market only in the short run, whereas thisnegative effect vanishes very quickly.

For the West German FbW programmes weobserve a more favourable picture. The cumulativelag coefficients indicate that the immediate signifi-cant negative impact on the job seeker rate remainsnegative through time. Furthermore the negativeeffect becomes larger, which could be seen asevidence for locking-in effects. As the long runeffect for FbW is also negative significant, we canconclude that FbW programmes improve thelabour-market situation in all respects, indicatingthat in West Germany FbW is the most effectiveprogramme. This is especially true with respect tomedium-term and long-term effects.

The results from the OLS and the within estimatorfor East Germany are reported in Table 5. Unfortu-nately, the results from the OLS estimator depict anunstable path for the job seeker rate. This should beseen as an evidence that the OLS estimates for theautoregressive parameters are upward biased.Furthermore, due to the unstable pattern of the OLSestimates the lag coefficients resulting from theseestimates are not reliable. Therefore we will baseour analysis only on the within estimator, although

this estimator is also known to be biased in thedynamic panel data context.

As with West Germany, we do not find a clearpicture in relation to vacancies and so the exis-tence of the Beveridge curve may be question-able. We have plotted the cumulative lag coeffi-cients for the ALMP measures JCS, SAS andFbW resulting from the within estimator inFigures 4, 5 and 6.

For the JCS we find a near zero and insignifi-cant effect, indicating that they have hardly anyimpact on the East German labour market. In thecase of the SAS we find throughout a negativecumulative lag coefficient, although it is signifi-cant only for t+4 up to t+6. This provides someevidence that the SAS are able to reduce the jobseeker rate in the medium- and long-term. Finally,we find for FbW an insignificant positive effect,i.e. there is no evidence of an effect on the jobseeker rate.

Summarising our results, we find very differentsituations in East and West Germany, though theestimates for East Germany should be treatedwith caution. Comparing the performance of FbWprogrammes with other ALMP measures, we findthat in West Germany FbW is the most efficienttype of programme. In East Germany this doesnot hold since the effects are, at best, insignifi-cant.


Figure 3: Cumulative lag coefficients for FbW in West Germany (GMM estimates)

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

t t+1 t+2 t+3 t+4 t+5 t+6

Quarters

Jo

b s

ee

ke

r ra

te



Lower confidence band (95 %)

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 202



OLS estimator Within estimator

Variable Param. t-value Param. t-value

Const. 0.2680 3.933 - -

Job seeker rate JSRt-1 0.6729 10.042 0.4644 8.133

JSRt-2 -0.2104 -2.231 -0.2176 -3.221

JSRt-3 0.1139 1.352 0.0318 0.492

JSRt-4 0.4274 5.631 0.4768 6.168

Vacancy rate VRt -0.4437 -2.491 -0.2064 -0.857

VRt-1 0.3060 1.261 0.1673 0.824

VRt-2 -0.0719 -0.248 -0.0517 -0.187

VRt-3 0.2385 0.990 0.2198 1.244

VRt-4 -0.1483 -0.786 -0.0132 -0.073

Participants in job creation schemes JCSt -0.0118 -0.362 0.0009 0.028

JCSt-1 0.0127 0.317 0.0099 0.305

JCSt-2 0.0020 0.063 -0.0022 -0.076

JCSt-3 0.0196 0.750 0.0155 0.687

JCSt-4 -0.0165 -0.988 -0.0125 -0.835

Participants in structural SASt -0.1011 -2.801 -0.0886 -2.142adjustment schemes SASt-1 0.0927 1.869 0.0647 1.522

SASt-2 -0.0038 -0.084 0.0027 0.065

SASt-3 -0.0220 -0.604 -0.0058 -0.157

SASt-4 -0.0165 -0.494 -0.0839 -2.377

Participants in continuing FbWt -0.0573 -1.234 -0.0766 -1.439

vocational training FbWt-1 0.0452 0.871 0.0926 1.760

FbWt-2 -0.0776 -1.694 -0.0131 -0.321

FbWt-3 -0.0415 -0.996 -0.0359 -0.967

FbWt-4 0.0323 0.951 0.0908 2.105

National unemployed rate NURt-1 -3.2415 -3.785 -1.8423 -2.373

NURt-2 0.4996 0.671 0.9055 1.224

National vacancies rate NVRt-1 0.7601 0.423 3.4963 2.154

NVRt-2 -2.1817 -1.090 -1.0512 -0.655

Seasonal dummy 1 SD1 0.0088 2.553 0.0150 4.245

Seasonal dummy 2 SD2 0.0204 1.746 0.0139 1.488


Wald test of joint significance [26] 691131.047 [26] 78622.047

First-order serial correlation [34] 3.982 [34] -3.769

Second-order serial correlation [34] 2.543 [34] 0.455

Table 5: Estimate results for East Germany

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 203


Figure 4: Cumulative lag coefficients for JCS in East Germany

-0.40

-0.30

-0.20

-0.10

0.00

0.10

0.20

0.30

0.40

t t+1 t+2 t+3 t+4 t+5 t+6

Quarters

Jo

b s

ee

ke

r ra

te


Cumulated lag coefficient


Figure 5: Cumulative lag coefficients for SAS in East Germany (within estimates)

-0.40

-0.30

-0.20

-0.10

0.00

0.10

0.20

0.30

0.40

t t+1 t+2 t+3 t+4 t+5 t+6

Quarters

Jo

b s

ee

ke

r ra

te




01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 204



Figure 6: Cumulative lag coefficients for FbW in East Germany (within estimates)

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

t t+1 t+2 t+3 t+4 t+5 t+6

Quarters

Jo

b s

ee

ke

r ra

te




01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 205

What policy implications can be drawn from ourresults? Empirical evidence suggests that ifpolicy in West Germany has the objective ofreducing unemployment, spending on ALMPshould be shifted from JCS to vocational training.This conclusion can be drawn simply from thefinding that vocational training is the more effi-cient ALMP measure in West Germany. For EastGermany we cannot make similar statementssince we did not find a favourable picture for theeffects of ALMP, and the results are questionablefrom an econometric point of view.

An explanation for the non-significant result inEast Germany may be the fact that thelabour-market conditions in East Germany arevery different from West Germany. Unemploy-ment in East Germany is more severe and wherethis results from a shortage in labour demand,vocational training cannot affect thelabour-market situation. This is because voca-tional training affects the structure of laboursupply but cannot solve demand side problems.

This may also be an issue for West Germanysince major difficulties on the labour demand sideseem to be responsible for recent unemployment.Therefore, if policy intends to shift resources tovocational training it should be consideredwhether such a shift is practical in the currentsituation.

Additionally, the objective of ALMP is not solelyto increase reintegration into regular employment.Job creation programmes, in particular, haveobjectives that are not directly associated witheconomic success. If ALMP is designed to fulfil aset of objectives of which only a proportion are ofan economic nature, more than economic conse-

quences should be considered. As our study canonly assess the economic consequences ofALMP, the policy implications drawn from ourresults can only be associated with economicobjectives of the programmes.

If we consider only the economic aspects, wecan state that the shift of recourses to FbWprogrammes should have favourable effects onemployment in West Germany. If there is thequestion of what type of programme should beextended, FbW programmes seem to be themost promising option. Our results for EastGermany do not encourage policy statementssince they are unreliable.

It is recommended for future research thatmore work should be carried out on estimationtechniques for dynamic panel data models. Thereshould also be some consideration of how thesimultaneity problem of ALMP can be solved.This is directly related to the estimation tech-niques. In the context of instrumental variables orGMM estimators it is a question of finding appro-priate instruments, i.e. instruments that are notonly not correlated with the residual of the regres-sion equation but also have a good predictivepower with respect to ALMP measures. To do thisit might be useful to consider the decisionprocess of how money is allocated to differentALMP measures.

Finally, it would be informative to compare amacroeconometric evaluation with a microecono-metric evaluation. If the data is generated fromthe same sources it enables the researcher tomake statements on how the same programmeaffects individuals and how it affects the wholeeconomy.

8. Implications for policy and future research

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 206

Our results indicate for West Germany that FbWhas a negative significant effect, that is FbW isable to reduce the job seeker rate. JCS also hasa negative effect but only in the short run. There-fore FbW seems to be the more efficient ALMPmeasure in West Germany for tackling unemploy-ment. In East Germany we do not find anyevidence of a significant effect for JCS and FbW,whereas for SAS there is weak evidence for anegative effect in the medium-term.

The results from our empirical analysis indicatethat there are major differences with respect tothe effectiveness of ALMP between East andWest Germany. That ALMP seems to work inWest but not in East Germany may be caused bythe fact that the labour market in East Germany ismuch worse.

Another interesting result is the clear differencesin the effects of FbW and JCS in West Germany.

Since FbW programmes have a persistently nega-tive effect on unemployment, they seem to bemore effective in improving the labour market.

Finally, we can state that a macroeconometricevaluation should be seen as an important ingre-dient of a complete evaluation, since deadweightlosses, substitution and displacement effects area major problem of ALMP. Important issues withrespect to empirical analysis are the inherentsimultaneity problem and an appropriate under-lying theoretical framework for the econometricspecification. Furthermore, the application ofdynamic models seems to be essential becauseof the high persistency of labour-market data.Basically, the ideal macroeconomic evaluationrequires three elements: a well developedmacroeconomic theory, which is applicable in aneconometric framework and which does not faildue to data limitations.

9. Conclusions

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 207

AFG Work promotion act [Arbeitsförderungsgesetz]

ALMP Active labour-market policy

CAST measures which are implemented in pathfinder regions and aim to promote the employment of low-qualified individuals or long-term unemployed

FbW Continuing vocational training [Förderung der beruflichen Weiterbildung]

FS Free support

GMM Generalised method of moments

JCS Job creation schemes [Arbeitsbeschaffungsmaßnahmen]

JOB-AQTIV The reform law regarding the ALMP instruments

JSR Job seeker rate

JUMP Special programme to combat youth unemployment

NUR National unemployment

NVR National vacancy rate

OLS Ordinary least squares

SAS Structural adjustment schemes [Strukturanpassungsmaßnahmen]

SAS-East Sam-East for private firms

SGB Social welfare code [Sozialgesetzbuch]

SUTVA Stable unit treatment value assumption

VR Vacancy rate

VT Vocational training


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 208

Annex

Variable Mean Std Min Max

No of labour office districts: 175

No of observations: 2 100

Time range: 1999:1 – 2001:4

Quarterly average measures

Participants in JCS 1 041 1 682 3 12 547

Participants in SAS 653 1 468 0 9 091

Participants in FbW 1 890 1 407 255 8 500

Unemployed 20 855 14 108 3 331 88 317

Dependent labour force 200 124 112 471 54 113 991 637

Regional information

Area (in hectare) 203 502 134 599 16 406 813 574

Additional national data

National unemployed 3 943 250 206 213 3 688 358 4 402 852

National vacancies 492 768 49 220 413 141 562 266

National labour force 41 668 281 41 116 42 068

Table 6: Descriptive statistics for Germany



No of observations: 1 692

Time range: 1999:1 – 2001:4


Participants in JCS 356 339 3 2 181

Participants in SAS 62 107 0 1 241

Participants in FbW 1 227 930 255 8 500

Unemployed 17 092 11 282 3 331 88 317



Area (in hectare) 176 204 95 609 16 406 478 275

Table 7: Descriptive statistics for West Germany

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 209




No of observations: 408

Time range: 1999:1 – 2001:4


Participants in JCS 3 882 2 019 1 048 12 547

Participants in SAS 3 105 1 894 514 9 091

Participants in FbW 3 813 1 422 1 139 8 197

Unemployed 36 462 13 946 10 671 74 296



Area (in hectare) 316 708 198 754 56 906 813 574

Table 8: Descriptive statistics for East Germany

Job creation schemes

Quarter Lag coefficients Standard error Cumulative lag Standard errorcoefficients

t -0.7612 0.137 -0.7612 0.137

t+1 0.1333 0.056 -0.6279 0.145

t+2 0.1856 0.055 -0.4423 0.186

t+3 0.1623 0.065 -0.2801 0.200

t+4 -0.0395 0.090 -0.3196 0.208

t+5 -0.0185 0.065 -0.3380 0.240

t+6 0.0217 0.031 -0.3163 0.266

Long run effect - - 0.3930 1.142

FbW


t -0.2068 0.044 -0.2068 0.044

t+1 0.0440 0.021 -0.1629 0.057

t+2 -0.0137 0.022 -0.1766 0.072

t+3 -0.0098 0.023 -0.1864 0.088

t+4 -0.0542 0.015 -0.2406 0.096

t+5 -0.0423 0.018 -0.2829 0.109

t+6 -0.0336 0.020 -0.3164 0.127

Long run effect - - -1.5228 0.385

Table 9: Cumulative lag coefficients for West Germany

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 210



Job creation schemes


t 0.0009 0.034 0.0009 0.034

t+1 0.0103 0.020 0.0113 0.026

t+2 0.0024 0.018 0.0136 0.038

t+3 0.0144 0.018 0.0281 0.041

t+4 -0.0055 0.018 0.0226 0.052

t+5 -0.0007 0.010 0.0218 0.056

t+6 0.0025 0.009 0.0243 0.063

Long run effect – – 0.0477 0.130

Structural adjustment schemes


t -0.0886 0.0413 -0.0886 0.0413

t+1 0.0236 0.0389 -0.0650 0.0564

t+2 0.0329 0.0359 -0.0321 0.0552

t+3 0.0015 0.0413 -0.0306 0.0607

t+4 -0.1318 0.0409 -0.1624 0.0692

t+5 -0.0493 0.0247 -0.2117 0.0773

t+6 0.0215 0.0238 -0.1901 0.0817

Long run effect - - -0.4534 0.251

FbW


t -0.0766 0.0532 -0.0766 0.0532

t+1 0.0571 0.0478 -0.0195 0.0714

t+2 0.0300 0.0354 0.0105 0.0812

t+3 -0.0368 0.0333 -0.0263 0.0808

t+4 0.0325 0.0355 0.0063 0.1015

t+5 0.0513 0.0377 0.0575 0.1295

t+6 0.0299 0.0205 0.0874 0.1439

Long run effect - - 0.2366 0.275

Table 10: Cumulative lag coefficients for East Germany

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 211


Figure 7: Participants in ALMP and unemployed in East Germany

Source: Bundesanstalt für Arbeit, 2001.

ABM SAM FBW AL

Un

em

plo

ye

d

Pa

rtic

pa

nts

in

AL

MP

200000

1050000

1400000

180000

160000

140000

120000

100000

80000

60000

40000

20000

0

1150000

1250000

1350000

1300000

1200000

1100000

01/99 02/99 03/99 04/99 01/00 02/00 03/00 04/00 01/01 02/01 03/01 04/01

Figure 8: Participants in ALMP and unemployed in West Germany

Un

em

plo

ye

d

Pa

rtic

pa

nts

in

AL

MP

250000

0

3000000

200000

150000

100000

50000

0

1000000

2000000

2500000

1500000

500000

01/99 02/99 03/99 04/99 01/00 02/00 03/00 04/00 01/01 02/01 03/01 04/01

Source: Bundesanstalt für Arbeit, 2001.

JCS SAM FBW Unemployed

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 212

Arellano, M.; Bond, S. R. Dynamic panel data esti-mation using DPD98 for Gauss, 1998. Availablefrom Internet: http://www.ifs.org.uk/staff/steve_b.shtml [cited 13.3.2003].

Baily, M.; Tobin, J. Macroeconomic effects of selec-tive public employment and wage subsidies.Brooking Papers on Economic Activity. Wash-ington DC: The Brooking Institution Press,1977, Vol. 2.

Baltagi, B. H. Econometric analysis of panel data.Chichester: John Wiley and Sons, 2001.

Blien, U. et al. Die Entwicklung der ostdeutschenRegionen. Beiträge zur Arbeitsmarkt- undBerufsforschung. Nürnberg: IAB – Institut fürArbeitsmarkt- und Berufsforschung der Bunde-sanstalt für Arbeit, 2002 (forthcoming).

Blundell, R. W.; Bond, S. R. Initial conditions andmoment restrictions in dynamic panel datamodels. Journal of Econometrics. Amsterdam:Elsevier Science B. V., 1998, Vol. 87, Issue 1,p. 115-143.

Blundell, R. W.; Bond, S. R; Windmeijer, F. Estima-tion in dynamic panel data models: improvingon the performance of the standard GMM esti-mators. London: IFS – The Institute for FiscalStudies; 2000 (Working paper, W00/12).

Brinkmann, C. Controlling and evaluation ofemployment promotion and the employmentservices in Germany. Nuremberg: IAB – Institutefor employment research, 1999 (Labour marketresearch topics series, No 36).

Büttner, T.; Prey, H. Does active labour-marketpolicy affect structural unemployment? Anempirical investigation for West Germanyregions, 1986-1993. Journal of Applied SocialScience Studies. Berlin: DIW – German Institutefor Economic Research, 1998, Vol. 188,p. 389-413.

Bundesanstalt für Arbeit. Daten zu denEingliederungsbilanzen 1999. Nürnberg:Bundesanstalt für Arbeit, 2000.

Bundesanstalt für Arbeit. Arbeitsmarkt 2000. Nürn-berg: Bundesanstalt für Arbeit, 2001.

Calmfors, L. Active labour market policy and unem-ployment: a framework for the analysis of

crucial design features. OECD EconomicStudies. Paris: OECD, 1994, Issue 22, p. 7-47.

Calmfors, L.; Lang, H. Macroeconomic effects ofactive labor market programmes in aunion-wage-setting model. Economic Journal.Oxford: Blackwell publishing, 1995, Vol. 105,Issue 430, p. 601-619.

Calmfors, L.; Skedinger, P. Does activelabour-market policy increase employment?Theoretical considerations and some empiricalevidence from Sweden. Oxford Review ofEconomic Policy. Oxford: Oxford UniversityPress, 1995, Vol. 11, Issue 1, p. 91-109.

Davidson, C.; Woodbury, S. A. The displacementeffect of re-employment bonus programs.Journal of Labor Economics. Chicago: Universityof Chicago Press, 1993, Vol.11, No4, p.575-605.

Fay, R. Enhancing the effectiveness of active labormarket policies: evidence from programmeevaluations in OECD countries. Paris: OECD,1996 (Labour market and social policy occa-sional papers series, No 18).

Fertig, M.; Schmidt, C. Discretionary measures ofactive labour market policy – The Germanemployment promotion reform in perspective.Heidelberg: Department of Economics, Univer-sity of Heidelberg, 2000 (Working Paper).

Fitzenberger, B.; Hujer, R. Stand und Perspektivender Evaluation der Aktiven Arbeitsmarktpolitikin Deutschland. Perspektiven der Wirtschaft-spolitik. Oxford: Blackwell Publishing, 2002,Vol. 3, No 2, p. 139-158.

Fitzenberger, B.; Speckesser, S. Zur wissen-schaftlichen Evaluation der AktivenArbeitsmarktpolitik in Deutschland: EinÜberblick. Mitteilungen aus der Arbeitsmarktund Berufsforschung, Nuremberg: IAB – Insti-tute for employment research, 2000, Vol. 33,Issue 3, p. 532-549.

Forslund, A.; Krueger, A. An evaluation of theSwedish active labor market policy: new andreceived wisdom. New York: NBER – NationalBureau of Economic Research, 1994 (Workingpaper No 4802).

Greene, W. H. Econometric Analysis. Upper SaddleRiver, N. J.: Prentice-Hall, 2000 (4th ed.)

References

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 213

Hagen, T.; Steiner, V. Von der Finanzierung derArbeitslosigkeit zur Förderung von Arbeit –Analysen und Empfehlungen zur Arbeitsmarkt-politik in Deutschland. Baden-Baden: NomosVerlagsgesellschaft, 2000.

Heckman, J. J.; Lochner, L.; Taber, C. General-equi-librium treatment effects: a study of tuitionpolicy. American Economic Review. Nashville:American Economic Association, 1998, Vol. 88,p. 381-386.

Heckman, J. J. Accounting for heterogeneity, diver-sity and general equilibrium in evaluating socialprograms. New York: National Bureau ofEconomic Research, 1999 (Working paper,No 7230).

Heckman, J. J.; Lalonde R. J.; Smith, J. A. Theeconomics and econometrics of active labormarket programs. In: Ashenfelter, O.; Card, D.(eds) Handbook of Labor Economics,Amsterdam: Elsevier, 1999, Vol.3A, p.1865-2097.

Holmlund, B.; Lindén, J. Job matching, temporarypublic employment and equilibrium unemploy-ment. Journal of Public Economics.Amsterdam: Elsevier Science B.V., 1993,Vol. 51, Issue 3, p. 329-343.

Hujer, R.; Caliendo, M. Evaluation of active labourmarket policy – methodological concepts andempirical estimates. In: Becker, I.; Ott, N.;Rolf, G. Soziale Sicherung in einer dynamischenGesellschaft, Frankfurt: Campus-Verlag, 2001,p. 583-617.

Hujer, R.; Caliendo, M.; Radic, D. Estimating theeffects of wage subsidies on the labour demandin West Germany using the IAB establishmentpanel. Ifo-Studien. Nashville: AmericanEconomic Association, 2001, Vol. 47, No 2,p. 163-199.

Hujer, R.; Caliendo, M.; Radic, D. Methods and limi-tations of evaluation and impact research. In:Descy, P.; Tessaring, M. (eds) Third researchreport: background report. Luxembourg: Officefor Official Publications of the EuropeanCommunities, 2003 (Cedefop Reference series).

Hujer, R.; Wellner, M. The effects of public sectorsponsored training on individual employmentperformance in East Germany. Bonn: IZA, 2000(Discussion paper, No 141)

Jackman, R.; Pissarides, C.; Savouri, S. Labourmarket policies and unemployment in theOECD. Economic Policy, 1990, Issue 11,p. 450-490.

Layard, R.; Nickell, S. Unemployment in Britain.Economica. Oxford: Blackwell Publishing,1986, Vol. 53, p. 121-170.

Pannenberg, M.; Schwarze, J. Labor market slackand the wage curve. Economic Letters.Amsterdam: Elsevier Science B.V., 1998,Vol. 58, Issue 3, p. 351-354.

Petrongolo, B.; Pissarides, C. Looking into theblack box: a survey of the matching function.Journal of Economic Literature. Nashville:American Economic Association, 1999, Vol. 39,p. 390-431.

Pissarides, C. Equilibrium unemployment theory.Cambridge: MIT Press, 2000 (2nd edition).

Prey, H. Wirkungen staatlicher Qualifizierungsmaß-nahmen. Eine empirische Untersuchung für dieBundesrepublik Deutschland. Bern: Verlag PaulHaupt, 1999.

Rubin, D. Comment on Basu, D. – Randomizationanalysis of experimental data: the Fisherrandomization test. Journal of the AmericanStatistical Association. Alexandria, VA: ASA,1980, Vol. 75, p. 591-593.

Schmid, G.; Speckesser, S.; Hilbert, C. Does activelabour market policy matter? An aggregateimpact analysis for Germany. In: de Koning, J.;Mosley, H. Labour market policy and unem-ployment. Evaluation of active measures inFrance, Germany, the Netherlands, Spain andSweden. Cheltenham: Edward Elgar, 2000,p. 77-114.

Smith, J. A Critical survey of empirical methods forevaluating active labor market policies.SJES – Swiss Journal of Economics and Statis-tics. Zurich: SSES – Swiss Society ofEconomics and Statistics, 2000, Vol. 136, No 3,p. 1-22.

Staat, M. Empirische Evaluation von Fortbildungund Umschulung - Schriftenreihe des ZEW 21.Baden-Baden: Nomos Verlagsgesellschaft,1997.

Steiner, V. et al. Strukturanalyse der Arbeitsmark-tentwicklung in den neuen Bundesländern.Baden-Baden: Nomos Verlagsgesellschaft,1998.

Windmeijer, F. A finite sample correction for the vari-ance of linear two-step estimators. London: TheInstitute for Fiscal Studies, 2000 (WorkingPaper 00/19).


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 214

Active policies and measures:impact on integration and reintegration

in the labour market and social lifeKenneth Walsh, David J. Parsons

AbstractThis paper forms a contribution to the third research report on vocational education and training (VET)research in Europe. The objective is to provide a comprehensive and critical overview of the success ofactive labour market and social policies (with a VET component), and to outline the implications for policyand practice, as well as indicating ways for further improvement. The focus of this report, as with othercontributions to the third research report, has been secondary analysis. The literature review has involvedthe examination of references over a long period of time, but cannot be considered exhaustive in itscoverage of all relevant material. More specifically, we have sought material post-1999 for detailed anal-ysis, but refer to previous composite studies. Relevant material has been identified through nationalexperts in EU Member States and also through various Internet sites of variable quality and usefulness.The report covers the methodological issues in the evaluation of active labour-market policies (ALMPs),including the main approaches and the problems that are likely to arise that can influence the validity ofthe results. The results of selected evaluations from various countries are reviewed, looking for common-alties in approach and the efficacy of the measures in terms of positive VET outcomes. Tentative conclu-sions on further research, policy stance and evaluation practice are made, aimed at all those involvedwith the design, implementation, monitoring and evaluation of ALMPs with a VET content.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 215

1. Executive summary 219

2. Introduction 221

2.1. Background to the study 221

2.2. The study 221

2.3. Approach and focus 221

2.4. Report structure 222

3. Methods and definitions 224


3.2. ALMP evaluation in context 224

3.3. Evaluation methodology 224

3.4. Coverage of the active policies 227

4. Evaluation findings 231


4.2. Composite studies 231

4.2.1. Training/retraining for the long-term unemployed 231

4.2.2. Retraining programmes for those laid off en masse 232

4.2.3. Training programmes for youth 232

4.3. Experience in the transitional countries 236

4.4. Selected national studies 240

4.4.1. Sweden 241

4.4.2. Denmark 243

4.4.3. Belgium 244

4.4.4. Germany 245

4.4.5. Ireland 245

4.4.6. United Kingdom 246

4.5. EU Evaluations 246

4.5.1. European Social Fund (ESF) 246

4.5.2. European employment strategy (EES) 248

4.5.2.1. Spain 248

4.5.2.2. The Netherlands 248

4.5.2.3. Ireland 249

4.5.2.4. United Kingdom 249

4.5.2.5. Austria 250

4.5.2.6. Finland 250

4.5.2.7. Other Member States 251

4.6. Lessons to be learnt 251

Table of contents

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 216

5. Summary and conclusions 252


5.2. Methodologies of evaluation 252

5.3. Programme structures 253

5.4. Forward perspectives 254

5.4.1. Developments in methods of evaluation 254

5.4.2. Further research 255


Bibliography and references 257

Active policies and measures: impact on integration and reintegration in the labour market and social life 217

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 217

Tables

Table 1: Expenditure on ALMPs 1997-2000 (Percentage of GDP) 229

Table 2: ALMP evaluations: country characteristics (1996) 236

Table 3: National evaluation of training programmes in brief 241

Table 4: Sweden: summary details of ALMP evaluations involving training in the 1990s 243

Table 5: ESF evaluations: summary of main points 247

Figures

Figure 1: Expenditure on training for adults and those at risk (percentage of total spend 228

on active measures)

Figure 2: Spending on ALMPs as proportion of total spending on labour market policy 235


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 218

The subject of the report is the evaluation andimpact of VET. It aims to describe and discussthe types, methodologies and results of evalua-tion research at different levels of analysis andthe specific problems VET research is facing.Particular attention is paid to the implications ofthe research for policy, evaluation practice andfurther research.

In preparing our report, we have concentratedon a review of published and grey literature andaccessing Internet-based material. In particular,we have sought material pre-1999 thatsummarises a range of studies from a range ofcountries, and within the limitations of secondaryanalysis we have brought together evaluationmaterial from a diverse range of sources andattempted to extract the most robust and relevantfor the focus of this study. The limitations of theanalysis are largely the constraints of the content,access and transparency of the source material.

The main element we are concerned with islabour-market training for unemployed adults andthose at risk, since this is likely to capture most ofthe relevant training/retraining measures.

Methodologically, the evaluations mostly tendto fall into the quasi-experimental sort, with anemphasis on the econometric elaboration ofprogramme outcomes. However, there is someevidence of the innovative use of administrativeand available survey-based data (such as fromlabour force surveys – LFSs) to underpin theseanalyses. The quality and validity of these evalua-tions are significantly dependent on the quality ofthe data used.

The problems in comparing programmesacross countries are caused by exogenouscircumstances affecting the relative effectivenessof programmes. This is particularly so fortraining/retraining measures, where a commit-ment can involve a substantial lead-time for aforecast future skill need. Effectiveness will,therefore, depend on factors such as the accu-racy and timeliness of labour-market information,the ability of employers to assess and communi-cate their future needs, the ability of trainingproviders to meet the articulated skill needs and

appropriate learning context, and changes inbusiness and political circumstances.

The growing emphasis on activation measuresand the coordination of active and passivemeasures further complicates effective evalua-tion. It leads to a lack of clarity about the struc-ture and objectives of individual programmes andcauses difficulties in gathering reliable informa-tion on participation.

There seem to be growing tensions betweenaccountability-led evaluations, those justifyingthe value secured from public funds, and thecontinuous improvement model. Evaluationframeworks and/or research designs seem tohave a narrowness of focus built into them, andthis does not encourage the sort of breadth ofscope and analysis needed to provide the qualityof feedback on VET impacts with ALMP neededto policy-makers, national or transnational.

Some of the best studies are those that take awider methodological perspective, certainly usingexperimental approaches where feasible, butcomplementing this with the use of administrativedata and more qualitative information on, forexample, processes and the perceptions ofprogramme participants. It suggests that evalua-tion hitherto has been more academically drivenrather than policy driven, which is not so much acriticism of the former as a lack of proper atten-tion in the latter.

The lack of an evaluation culture in some coun-tries (including some Member States) means thatprogramme evaluation is not conducted effec-tively, if at all. In addition, there are other caseswhere ALMP evaluation has been carried out onbehalf of countries under the auspices of interna-tional agencies, which is not the best basis foreffective policy follow-through.

It was disappointing to note that there waslittle evidence of what might be called a ‘holistic’approach to programme evaluation. These widerevaluations are often too easily dismissed asimpossible to measure, and so the challengeremains largely untouched. This seems a particu-larly important limitation in understanding VETimpacts.

1. Executive summary

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 219

As the mix between active and passivemeasures becomes blurred, the focus of evalua-tion is also likely to suffer. Furthermore, some ofthe existing studies are often only focused onpilot or demonstration projects, to the extent thatthey are never given the opportunity for extensive(wider and longer-term) application.

Many evaluation studies point to the need forprogrammes to have labour-market relevance.Good forward-looking labour-market informationcan provide strong indications of what the charac-teristics of the demand for labour are likely to be andthis can help inform programme provision (and indi-vidual choice). Also, the involvement of employersin the design and delivery of training programmesprovides a more direct way of ensuring that the skillsacquired by programme participants are likely tohave real currency in the labour market.

The evaluations generally agree that the bestprogrammes are those that are small in scale andhighly targeted, and this ties in with some keydevelopments in the approach to ALMPs, mostnotably the trend for offering customised support

to the unemployed. Typically, this will offer amenu of active measure options, one of which isnormally training of various sorts.

Those training programmes with job guaranteesseemed to show more positive outcomes andsuggested greater participant engagement (moti-vation and commitment) in the learning processesinvolved. This raises issues of how such interven-tions need to allow for the socialisation of skillsdevelopment and learning in programmes.

The review shows that, while evaluations aretaking place, it is neither a consistent nor uniformactivity. We conclude that policy andpolicy-makers, looking to improve the quality,efficacy and cost-effectiveness of VET interven-tions in ALMPs, are not being very well served bycurrent evaluation approaches.

This raises serious questions about just howmuch we know concerning the effectiveness ofthese mechanisms, and calls for reform of currentpractices involving not just evaluations, butcrucially those specialising in evaluation frame-works and selecting/commissioning evaluations.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 220

2.1. Background to the study

This paper forms a contribution to the thirdresearch report on VET research in Europe. Thesubject of the report is the evaluation and impactof VET and it aims to describe and discuss thetypes, methodologies and results of evaluationresearch at different levels of analysis, as well asthe specific problems VET research is facing.Particular attention is paid to the implications ofthe research for policy, evaluation practice andfurther research.

2.2. The study

The overall objective of this paper is to provide acomprehensive and critical overview of thesuccess of active labour-market and social poli-cies (with a VET component), and to outline theimplications for policy and practice as well asindicating ways for further improvement.

Our initial thoughts on the parameters of theresearch were set out in the Abstract of plannedresearch presented at the first meeting of thethird report (1). Subsequent discussions at theconference and comments by Cedefop on theseparameters were refined and set out in an interimreport (2) along with discussion of the plannedapproach.

The feedback from Cedefop at these variousstages highlighted a number of key issues toconsider in this final report and these have beentaken on board insofar as this has been possible.In particular, we have:(a) limited our review of evaluation methodolo-

gies and descriptions, given the focus ofother papers in the third research report;

(b) opened the scope of our paper to all activemeasures with a VET element, even where itmay not be the main part of the programme;

(c) concentrated on post-1999 material in the

interests of using the limited resources for thework as efficiently as possible, while recog-nising that there have been some usefuloverviews of relevant evaluations pre-1999;

(d) ensured that our search for relevant materialis not restricted to Member States;

(e) examined evaluations conducted in thecontext of EU programmes such as thosefunded through the European Social Fund(ESF) and in the review process for the Euro-pean Employment Strategies (EES).

Cedefop was also keen to encourage liaisonwith other relevant papers of the third researchreport. However, building this complementarityhas proved difficult; in particular, the Internetportal created to facilitate this communicationhas not been extensively used by any of thecontributors.

2.3. Approach and focus

The focus of this report, as with other contributionsto the third research report, has been secondaryanalysis. In preparing our report, we have thereforeconcentrated on the following tasks:(a) a review of published and grey literature,(b) access to Internet-based material.

The literature review has involved the examina-tion of references over a long period of time, butcannot be considered comprehensive or exhaus-tive in its coverage of all relevant material. Morespecifically, we have sought material pre-1999that summarises a range of studies from a rangeof countries, so that the limited project resourcescan be concentrated on accessing newer mate-rial. This was seen as the best way we could addvalue to current understanding on the evaluationof ALMPs.

Relevant material has been identified in anumber of ways. First, national experts werecontacted in Member States (e.g. through theemployment observatory Sysdem network) for

2. Introduction

(1) Held at Cedefop, Thessaloniki, from 28.2.2002 to 1.3.2002.(2) Presented to Cedefop in May 2002

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 221

advice on details of suitable studies. This proveda fruitful source of information, limited only by therange and quality of evaluations available in eachMember State, which sometimes differed consid-erably.

Member States such as Ireland, the Nether-lands, Sweden (especially) and the UK all have atrack record of robust and relevant work. In manyother Member States, information either lackstransparency or is extremely limited or non-exis-tent (as far as we could tell). Inevitably, our anal-ysis of this material was restricted by languagelimitations – basically, anything in English orFrench could be tackled (3) and fortunately somereports in other languages were either translatedinto English or had an appropriate summary.

Second, information was identified throughvarious Internet sites of variable quality andusefulness. Those sites of particular relevance tothis study included transnational organisations,and in particular:(a) European Commission,(b) International Labour Office (ILO) – particularly

the Labordoc database,(c) Organisation for Economic Cooperation and

Development (OECD),(d) the World Bank.

The material available for downloading throughthese (and other) agencies has expanded rapidlyand will continue to do so, thereby facilitatingwider search parameters through appropriatelinks (e.g. with national agencies, governmentdepartments, etc.) and cross-referencing.

Within the EU, much of the relevant evaluationeffort in Member States has been a response tothe demands of the ESF and its constituentprogrammes (such as Employment and ADAPT).This has generated many national studies andEU-wide appraisals (e.g. European Commission,2001) that provide a rich source of information.Unfortunately for both the mid-term and finalevaluations, the national reports were not alltranslated into English or French and nor werethey uploaded to the Internet, so access to themis restricted. Nevertheless, the EU-wide reportsprovide a good summary of the findings and havebeen an important contribution to this review.

More accessible are the national reports fromthe review of the EES which are available in

English and/or French on the EU website. Thesereports are, of necessity, wide-ranging, coveringas they do all four pillars of the EES. Neverthe-less, extensive coverage is given to pillar 1,employability, and it is here that some usefulmaterial has been collected for this study.However, the quality and empirical scope of thenational reports varies a great deal. Some providesystematic and thorough evaluations, whileothers are essentially little more than uncriticaldiscussions of activity, without any real attempt atanalysis of key issues such as impact andsuccess factors.

In summary, within the limitations of secondaryanalysis, we have brought together evaluationmaterial from a diverse range of sources andattempted to extract the most robust and relevantfor the focus of this study. The limitations of theanalysis are largely the constraints of the content,access and transparency of the source materialand, of course, the resources available to do thework. Other omissions and errors of interpretationare those of the authors alone.

2.4. Report structure

This introductory chapter of the report is followedby an examination of the methodological issuesin the evaluation of ALMPs, as seen through theliterature. This covers the main approaches andthe problems that are likely to arise that can influ-ence the validity of the results. There is alsodiscussion of the general approach to ALMPs inan international context, including the relativeimportance of those focused on VET.

Then the report examines the results ofselected evaluations from various countries,looking for commonalties in approach, the effi-cacy of the measures (as shown through the eval-uations) and summarising what we interpret asworking well and what does not, in producingpositive VET outcomes.

The final chapter of the report draws togetherthe findings from the previous analysis andprovides a series of tentative conclusions onfurther research, policy stance and evaluationpractice aimed at all those involved with thedesign, implementation and monitoring and eval-


(3) These parameters were agreed with Cedefop at the outset of the study.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 222

uation of ALMPs with a VET content. The relativeefficacy of the different approaches to ALMPs isexamined further through an exploration of whatcould be construed as the best options forALMPs with a VET focus. While not attempting to

establish a model for such programmes, whichwe believe would be premature given theevidence constraints, we try to isolate thoseingredients that are likely to be more successfulin terms of VET outcomes.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 223

3.1. Introduction

It is important to consider the basic methodologicalissues that underpin evaluations of ALMPs. Thisprovides an important context for the analysis andassessment of VET outcomes later in this paper. Indoing so, we are aware that other papers in the thirdresearch report (see in particular Hujer et al., 2003a)are concentrating more on the methodologicalissues to do with evaluation generally, and so herewe have tried to be very focused. However, it isreasonable to set out the main approaches used forthe active measures in general, and those concen-trating on training/retraining in particular, in order toprovide a backdrop for understanding what hasactually been done.

3.2. ALMP evaluation in context

In general, there seems to be disappointment withboth the number and quality of ALMP evaluations.Martin and Grubb (2001) make the point, in theiroverview of OECD countries, that much of the eval-uation work that exists emanates from Canada andthe US, where there has been a ‘long-standingtradition of evaluating labour-market programmes’(p. 10), going on further to suggest that there is a‘mandatory requirement on the public authorities toevaluate their programmes.’

These authors suggest that this good practicehas rubbed off on some countries (Australia,Belgium, Germany, Ireland, the Nordic countries,Switzerland and the UK are mentioned) who arenow more inclined to do thorough evaluations,whereas in others there is less evidence of suchpractice. Here, the suggestion is that monitoringof individual VET outcomes post-programmealone is not enough to constitute proper evalua-tion. Nonetheless, this focus on individual orparticipant gains remains quite common practice.

In a similar vein, the OECD (2001) thought thenumber of rigorous evaluations of ALMPs wasinadequate (p. 29) and this reflects our review ofthe evidence. In some countries, this seems to

stem from a weak evaluation culture, let aloneevaluation of available ALMPs. However, theremay be mitigating circumstances. For example,Kluve et al. (1999) point out that in the case ofPoland, rigorous evaluations have been curtailedthrough a lack of appropriate data (p. 5).

This situation regarding robust official data forsource or benchmark evidence is not unique toPoland. To a greater or lesser extent it applies to alltransitional countries. It is likely that problems withdata extent, utility of classifications and accuracyalso inhibit thorough evaluations of programmes insome Member States. However, the lack of criticalattention to this in the source material makes it diffi-cult to decide, in such cases, whether it is due to alack of evaluation culture impeding gathering betterdata or vice versa. Cause and effect here seem veryconfused.

The rigours of the funding requirement andimplementation of ESF and the EES have helpedraise the profile of proper evaluation in MemberStates. However, there is still a great deal ofdiversity evident in the approach and content ofnational reports, which makes the job of pullingtogether an EU-wide perspective particularlydaunting. This is well illustrated by one of theprincipal conclusions of the review of EES poli-cies in Greece (OMAS, 2001), which stated that a‘thorough assessment of the employment poli-cies’ (p. 97) should be carried out, as well as anevaluation of the EES in the Greek labour market(presumably one leading to the other).

By contrast, the Dutch EES review is much morethorough, albeit relying on pre-existing evaluationstudies (with which the Netherlands is reasonablywell endowed). The Belgian EES review is similarlydependent on previous research, while those forIreland and the UK are perhaps less rigorous interms of using previous evaluations and are morereliant on programme monitoring information.

3.3. Evaluation methodology

There are various general sources of information onthe approach to ALMP evaluation. The most recent

3. Methods and definitions

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 224

from the World Bank (2001) is particularly useful insetting out in simple terms the types of approach,from setting up and monitoring appropriate perfor-mance indicators, to carrying out cost-benefit andcost-effectiveness analysis. It usefully highlightsthe distinction between monitoring and evaluation,often muddled in such source material, whilestressing their complementarity when assessingthe viability of any activity. The tools described inthis study are generic and have been proposed asapplicable to a much wider range of situations thanactive labour-market policies. However, this showsthat there is no particular mystique to the evaluationof such measures.

For a more thorough discussion of the options,the collection of papers edited by Schmid et al.(1996) is a rich source of case study, as well asmethodological material, which includes someuseful work programmes relevant to our research.Another methodological report we have found tobe of particular value is that of Pierre (1999) inwhich the discussion on the strengths and weak-nesses of the various approaches to evaluation isparticularly informative. This study has the virtueof being (unusually) focused on the evaluation ofALMPs. Its conclusions are perhaps controversialin that the author finds that evaluation ‘has notbeen systematically carried out’ (p. 30), especiallyin Europe (the US fares a little better). Further-more, the author laments the lack of attentiongiven to evaluation of ALMPs by policy-makers,despite the cost and political prominence of suchinterventions.

A distinction needs to be made between thetwo principal types of policy evaluation asfollows:(a) impact of policies on the individual,(b) effect on aggregate employment and unem-

ployment.The first of these essentially seeks to measure

the individuals’ labour-market outcomes oncompletion of their programme participation andis most frequently focused on employment andearnings. ALMPs, where the objective is toimprove the employability of the participant, arewell suited to this form of evaluation, and thoseinvolving training/retraining particularly so.

The second type is more concerned withoverall employment effects, with attention givento any negative outcomes that can diminish theefficacy of the programme. While some of these

potentially negative effects can be present intraining/retraining ALMPs, they are more likely tobe present in programmes centred on subsidisedjobs.

In the evaluation of relevant activelabour-market policies, it is possible to categorisetwo main approaches as follows:(a) scientific:

(i) experimental;(ii) quasi-experimental.

(b) non-scientific.This is not presented as a typology, but more a

simple reference framework with which tocompare different emphases in the differentALMP contexts.

Scientific evaluation essentially uses the tech-nique of comparing the outcomes of those takingpart in a programme (the treatment group) andthose not taking part (the control group). Underexperimental techniques (sometimes calledcontrolled social experiments), the two groupsare established prior to the programme interven-tion and should be based on a randomisedsample distribution. However, bias can be intro-duced in the selection of programme participants(e.g. the programme administrators may selectthe more able and thus exclude those who theyfeel may not benefit from the programme) that willultimately diminish the results. Similarly, theactions of the control group cannot be guaran-teed, in that, for example, some may choose tofollow a different training/retraining course thatwould affect (positively) their performance in thelabour market.

These and other restricting methodologicalfactors (see Hujer et al., 2003a for a discussion ofthese matters), coupled with the potential difficul-ties of setting up such experiments (e.g. they canfall foul of equal opportunity legislation) and theassociated costs, means that there are onlylimited examples of this technique being appliedand most of them originate from North America.This may be due to a greater willingness ofpolicy-makers in the US and Canada to counte-nance such experiments, whereas in Europethere may be a greater reluctance to excludecertain eligible groups or individuals from aprogramme (e.g. based on an unwillingness todeny the unemployed access to opportunities).However, it is difficult to corroborate the reasonspostulated with hard evidence.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 225

It is not surprising, therefore, that the morecommon impact evaluation approach is of thequasi-experimental kind. Here, the treatment andcontrol groups are derived after the programmeintervention, with statistical techniques used tocorrect for any differences in the characteristicsof the two groups. However, such differences canbe minimised through the use of the matchedpairs approach, whereby the influential character-istics of the treatment and control groups (suchas gender, age, unemployment duration, location,etc.) can be selected to form the basis for bothsamples. Matched pairs can provide a robustbasis for contrasting impact, but at a cost thatsome policy-makers or fund holders would findprohibitive. However, it should be noted that thereare other techniques sometimes used (such asbefore- and after-studies, cross-sections andrepeated cross-sections, etc.) and for a moredetailed discussion see the work by Hujer et al.(2003a) for this third research report.

Both experimental and quasi-experimentalapproaches have their limitations. They essen-tially measure the impact of participation in theprogramme, and this is most frequently articu-lated in terms of the following two quantifiableindicators of impact on the individual:(a) employment,(b) earnings.

Clearly, there is a fundamental need to know ifa programme increases the likelihood of partici-pants finding work. The earnings outcomeprovides a crude proxy for job quality in that it canshow if the programme has helped participantsimprove their capacity to earn. This can be animportant consideration for a training/retrainingprogramme, given the investment required by theindividual in their training, and can add a powerfulincentive effect when present (and proven).

One of the big disadvantages with the scien-tific approach to evaluation is that it cannotmeasure the negative effects that inevitablyaccrue from active policies. Most significantly,any displacement or substitution (4) effectscannot be picked up through these methods and

can only be gauged through other techniques(such as employer surveys). Similarly, othereffects from the implementation of a programme,such as spill-overs, crowding out and similarunintended attributes, are not usually measurablethrough the generalised approach to evaluationand can only be picked up through a combinationof methods applied either simultaneously orpossibly consecutively.

Non-scientific evaluation techniques are muchmore useful in these cases. Typically, theseinclude using official, non-official and/orprogramme statistics to examine programmeparticipation and outcomes. Approaches are alsolikely to include some combination of surveysand interviews with employers, employees,programme participants, training providers, etc.Clearly some of these techniques (e.g. anemployer survey) could be carried out underrigorous controls in terms of sampling, question-naire design, etc., and can render robust results.However, in terms of the evaluation of the impactof active measures in most cases they are seenas complementary to the scientific techniques.Their use in addressing some of the negativeeffects from programmes has already beenpointed out, and it is perfectly feasible to includequestions in a quasi-experimental survey ofprogramme participants (and the control group)that will provide the raw material for bothapproaches (Walsh et al., 2000). This can, forexample, provide strong indicators of the deadweight effect in programmes (5).

However, this is too simplistic a notion ofpolicy effectiveness. Introducing a particularprogramme could have effects reaching beyondthe individual participant, or even aggregateemployment and unemployment. Here, we aretalking about the effects on society as a whole,such as changes in attitudes, culture and maybethe way labour markets or businesses operate.These could be construed as indirect effects butare still important.

Societal effects arising from an ALMP could,for example, encompass such indicators as crime


(4) These negative effects occur when the programme participants take the place of existing workers. Substitution takes place inthe firm where the programme participant finds a job and displacement may occur when jobs are lost in another firm becauseof the unfair cost advantage gained by the firm taking on the programme participant. See Hujer et al. (2003b) for a case studyof evaluation trying to cope with such bias.

(5) The deadweight effect essentially measures the proportion of programme participants that would have followed similar activi-ties (such as retraining) in the absence of the programme.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 226

rates, incidence of poverty and health problems.But no matter how important these might be in aholistic appraisal of programmes, they remainvirtually untouched by evaluation literature andare often dismissed as all but impossible tomeasure (e.g. the Dutch EES report). The paperby Green et al., (2003) for the third researchreport also covers this area.

For this study, we considered relevant evalua-tions that are based on any of the techniquesoutlined above, as well as combinations of these.Within this, we have concentrated on the effectson the individual, which fits with the remit of theresearch. The contribution of Hujer et al. (2003b)has more to say on the macro effects of activemeasures in the case of Germany.

Another important issue to consider in anyALMP evaluation is whether it is feasible toisolate the different elements of the programmes(and, for our use, those concerned with trainingand retraining).

Another potential problem for the evaluatorraised by Martin and Grubb (2001) is the problemof ever-changing programmes. The conditionsunder which the various programmes operate areunder constant review by national governments,who respond not so much to the results of moni-toring and evaluation, but to the exigencies ofbudgetary constraints and political demands.Whatever the cause of these changes, theirimpact on the operation of the programmesthemselves is undeniable and only serves tomake rigorous evaluation even more difficult.

Martin and Grubb go on to highlight otherlimiting factors. For example, evaluations tend tocover only short-term outcomes from programmeparticipation, maybe up to a maximum of one ortwo years’ post-participation in the programme.Also, many evaluations target pilot or demonstra-tion programmes which are, by nature, smallscale and often experimental in scope (the impli-cation being that they do not, therefore, test thereal world circumstances that the programmeswould have to operate in).

Another complicating factor is the develop-ment of so-called coordinated active and passiveprogrammes. These effectively bring together thepayment of unemployment (and subsequentsocial) benefits with conditions on job search,training and other programme participation.Some of the major developments are listed in

OECD (2001) and include reforms to the Cana-dian employment insurance system, the USwelfare system and the Danish and Irish unem-ployment benefit systems. In the UK, variants onthe New Deal programme have taken thisapproach for a few years now, and in France, thePARE (Plan d’aide au retour à l’emploi) seeks totake a similar approach.

Essentially, these initiatives aim at linking thepayment of unemployment benefits (and subse-quent social benefits) to evidence of active jobsearch or, where job search is either not feasible(i.e. the unemployed person is not yet ready tolook for a job on the regular labour market) orineffective, to participation in an active measure.This often involves a customised approach forthe unemployed, with particular attention paid toindividual abilities and motivation. In the UK, forexample, this has been taken a step further withthe merging of the traditionally separate activitiesof the payment of benefits and employmentservices. Under Jobcentre Plus (the title appliedto the initiative), the hitherto separate serviceshave been physically brought together under thesame roof, with staff dealing with all aspects ofthe unemployed client’s needs.

There are other examples of such coordinatedapproaches and it is proving to be a strong trendin the structure of ALMPs. The downside for eval-uators is that the divisions between the differentparts of the programme become obscured. It isoften not possible to isolate a particular effectbecause of its interdependence on other factors.For example, when looking at the New Dealprogramme in the UK, Walsh et al. (1999) foundthat participants were dipping in and out ofvarious options under the initiative, such that theseparate effects became virtually impossible toisolate.

3.4. Coverage of the activepolicies

The OECD employment outlook (1993) helpfullyset out three main elements that condition activelabour-market policies, as follows:(a) mobilising labour supply with job creation

schemes, job subsidies and similarapproaches;


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 227

(b) developing employment-related skills withmeasures such as retraining;

(c) promoting efficient labour markets withemployment services, job matching andcounselling.

It is the second of these main elements that weare concerned about in our research. However,the degree of importance attached to this policy(as proxied by the proportion of total spending onactive measures devoted to it) varies consider-ably between countries. Table 1 plots the infor-mation from the OECD (2001), showing variousmeasures of ALMP expenditure. This draws onthe well-established OECD Labour-marketprogrammes (LMP) database (6) which chartsmainly public expenditure on active measures.

The list of measures covered extends to thefollowing:(a) public employment services and administra-

tion,(b) labour-market training,(c) youth measures,(d) subsidised employment,(e) measures for the disabled,

(f) unemployment compensation,(g) early retirement for labour-market reasons.

The categories we are interested in are thesecond and third in the list and need some furtherexplanation as to their coverage, as follows:(a) labour-market training: this includes both

course costs and subsistence allowances,but specifically excludes special trainingprogrammes for youth and the disabled;

(b) youth measures: these include support forapprenticeships and related forms of generalyouth training.

Both categories exclude any unemployment orsocial benefits paid to the participants, whichwould be included in the expenditure on passivemeasures.

According to the table, total expenditure onactive measures ranged from a high of 1.54 % ofGDP in Denmark, to a low of 0.15 % in the US.Both countries have similar levels of unemploy-ment and so there is no real correlation betweenthese two indicators. The differences are theconsequence of policy emphasis.

The main element we are concerned with is


(6) The LMP database contains data from 1985.

Figure 1: Expenditure on training for adults and those at risk (percentage of total spend on active

measures)

Source: Martin & Grubb (2000)

Percentage of totalspend on active measures

4.3

11.910.7

15.9

10.8

30.6

26.7

21.721.4

32.7

42.9

18.4

27.6

0

5

10

15

20

25

30

35

40

45

50

AU A B DK F D I JP NL NZ E S UK US

1999-2000

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 228

labour-market training for unemployed adults andthose at risk, since this is likely to capture most ofthe relevant training/retraining measures. Expen-diture in this category varies from between0.66 % of GDP in Denmark to 0.03 % in Japan.However, a much clearer picture of the relativeimportance attached to training/retraining comesfrom the comparisons in Figure 1. This plotslabour-market training for unemployed adults as aproportion of the total spend on active measuresand shows that most emphasis was again inDenmark (42.9 %), followed some way behindwith New Zealand (32.7 %) and Austria (30.6 %).Least emphasis was in Australia (4.3 %), Japan(10.7 %) and the UK (10.8 %).

According to the OECD (2001), there has beenlittle overall change in the pattern of expenditureon ALMPs, with minor changes showing a smalldecline in the proportion of expenditure on youth

and disabled programmes and a small increase inemployment subsidies. However, the general trendin many countries towards activation measures ofthe integrated sort makes the measurement ofexpenditure on sub-elements hazardous.

Typically, the transitional countries tend tospend less on training/retraining, favouring otherpolicies such as public works job creation andsubsidised jobs. However, these proportions willtend to shift as policy emphases change. Forexample, some countries spent a large proportionof their ALMP budget on developing theiremployment services in the 1990s, only to switchemphasis to other policies once the required levelof service had been established.

Elements of training/retraining may also be inte-grated within other programmes such as job subsi-dies (where the employer might be required to offersome structured training to the participant) and


Total TotalLabour- Passive

Countryactive active

market Youth expenditurespend spend

training 20001997 2000

Training for Measures for unemployed Training for

unemployed andGeneral

adults employed Totaldisadvantaged

youth Totaland those adults

youthtraining

at risk

Australia* 0.63 0.46 0.02 0.02 0.04 0.01 0.07 0.08 1.05

Austria 0.45 0.49 0.15 0.02 0.17 0.02 0.02 0.04 1.07

Belgium** 1.46 1.35 0.16 0.09 0.25 0 0 0 2.34

Denmark 1.66 1.54 0.66 0.18 0.84 0.1 0 0.1 2.96

France** 1.34 1.36 0.25 0.03 0.28 0.21 0.19 0.41 1.76

Germany 1.23 1.23 0.34 0 0.34 0.07 0.01 0.08 1.89

Italy** na na 0.08 0.04 0.12 0.01 0.23 0.25 0.64

Japan* 0.34 0.28 0.03 0 0.03 0 0 0 0.54

Netherlands 1.47 1.57 0.25 0.05 0.3 0 0.04 0.04 2.08

New Zealand* 0.77 0.55 0.18 0 0.18 0.07 0.07 0.14 1.62

Spain 0.49 0.98 0.21 0.09 0.29 0.06 0 0.06 1.34

Sweden 2.03 1.38 0.3 0.01 0.31 0.02 0 0.02 1.34

United Kingdom*** 0.39 0.37 0.04 0.01 0.05 0.04 0.11 0.15 0.58

United States* 0.17 0.15 0.15 0.04 0 0.04 0.03 0.03 0.23

* 1996/97 and 1999/2000 figures** 1997 and 1999 figures*** 1997/98 and 1999/2000 figuresSource: OECD (2001)

Table 1: Expenditure on ALMPs 1997-2000 (percentage of GDP)

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 229

more directly in job search/counselling activities(Bysshe et al., 1998). However, to avoid the compli-cation of separating out the training/retrainingelement in such programmes, we will not be specif-ically including these hybrid approaches in ourresearch. While the application of training in employ-ment clearly falls within the remit of VET, we feel thatit would not be feasible to isolate the trainingelement from other aspects of the programme.

In further refining what we cover in this report,a study by Dar and Tzannatos (1999) provides auseful typology for training/retraining programmes.The authors suggest the three following cate-gories:(a) training for the long-term unemployed,(b) retraining for those laid off (en masse),(c) training for youth.

The rationale for this split from an evaluationperspective is that many of the relevant policiesare targeted at one of these subgroups. Further-more, the influential factors that need to be takeninto account when comparing outcomes tend tovary between them. For example, the previousemployment experience of the differentsubgroups is potentially important, with thelong-term unemployed by definition having beenaway from the labour market for some time (7),those facing redundancy clearly in a job, and, foryouth, any experience of a job possibly still someway off. These three strands should enable us toencompass the wider constituency affected byALMPs. However, as with all matters insecondary research, this will only be so far as theevaluation literature permits it.


(7) Here, the conventional definition of long-term unemployment is periods of 12 months or longer.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 230

4.1. Introduction

In this chapter of the report, we review selectedevaluations of ALMPs with a training/retrainingelement for different countries, bringing outwhere possible the methodological approach, thefindings, and exploring any observable connec-tions between these elements.

In doing so, we have adopted the followingcriteria in the selection of material, given thefocus of our study:(a) coverage of different countries, including

southern Member States;(b) evaluations carried out in transitional coun-

tries;(c) concentrating on post-1999 material.

We have examined the options available tomake best use of the secondary research focusand limited resources available, and also theneed to balance breadth with depth of coverage.

There is also much historical information avail-able on the evaluation of active measures thatneeds to be acknowledged, though many of thestudies are generalist in nature and do not neces-sarily cover in detail either the measures them-selves (e.g. a description of targets, operation,etc.) or the evaluation findings. Nevertheless,some are worthy of closer inspection, and fortu-nately there have been some extremely usefulreviews that we can use both to take on boardthe general points and to identify those studiesfor closer scrutiny.

4.2. Composite studies

A good starting point is reviewing the principalcomposite studies of the 1990s. These provide aparticularly useful overview of policy and prac-tice.

The study by Dar and Tzannatos (1999) is acomprehensive review of the evaluations of activelabour-market policies, drawing on work as far backas 1979, though principally covering work done in the1990s (up to 1999, the publication date of the study).

It has a bias towards studies from NorthAmerica and,to a lesser extent, western Europe. This focus basi-cally reflects the reality of the situation (there is astronger track record of evaluations in NorthAmerica). The work extends beyond training/retraining measures to include other active measures,though each is segmented for analysis.

For the training/retraining measures, theauthors use the familiar three categories(discussed above) and the coverage of studies isquite extensive as follows:(a) training for the long-term unemployed –

23 studies covered:(i) 6 experimental;(ii) 13 quasi-experimental;(iii) 4 non-scientific.

(b) retraining for those laid off en masse –11 studies covered:(i) 5 quasi-experimental;(ii) 5 non-scientific;(iii) 1 using a variety of techniques.

(c) training for youth – 7 studies covered:(i) 5 experimental;(ii) 2 quasi-experimental.

It is not possible from the report to determinehow comprehensive this coverage really is ofcontemporary ALMPs in these areas. However,the review by Dar and Tzannatos is the mostextensive in its coverage of individual studies.

In terms of those focused on training/retraining,the authors generally felt that, while popular, theydisplayed many shortcomings. Looking at theirfindings using the three categories alreadyexplained, a number of common themes emerge:

4.2.1. Training/retraining for the long-term

unemployed

(a) Programmes can have a positive effect onemployment and earnings, but it is likely to besmall;

(b) programme success depends to a greatextent on the business cycle. When theeconomy is buoyant, prospects for partici-pants will be greater;

(c) generally, programmes are more effective forwomen participants;

4. Evaluation findings

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 231

(d) programmes can cost between USD 900 and12 000 (8) per participant.

4.2.2. Retraining programmes for those laid

off en masse

(a) Retraining for those made redundant has ahigh dead weight negative effect;

(b) programmes are most successful if they aresmall scale and targeted towards the mostvulnerable groups;

(c) programmes can cost between USD 3 500and 25 000 per participant (9).

4.2.3. Training programmes for youth

(a) Usually targeted at disadvantaged groupssuch as school dropouts;

(b) in virtually all cases, programme participantsdid no better than their counterparts in thecontrol groups at improving their employmentprobability or earnings;

(c) where programmes had been subject tocost-benefit analysis, social rates of returntended to be negative.

Overall, the youth programmes gave the mostdisappointing set of results in virtually all thecountries examined. However, the review foundthat, in general, training/retraining programmeswere difficult to justify except for those targetedon certain subgroups. Another interesting conclu-sion was that most of the programmes were nomore effective at improving job prospects orearnings than job search assistance measures,though this was qualified with the remark that, forthe subgroups affected, the two types ofprogramme were not necessarily substitutable.

Another composite study by Fay (1996)predates the former and is less detailed and lessspecific in its coverage of training/retrainingmeasures. However, it does have the virtue of acomprehensive bibliography from which it hasbeen possible to extract particular studies formore detailed analysis. For example, the Cana-dian employability improvement programme wascovered by a study in 1995, and employmenttraining in Sweden by a study in 1995 (Tamáset al., 1995). The review by Fay has the advan-tage of the including more European studies,

though with the disadvantage that only a few arefocused on training/retraining.

Nevertheless, those that are reviewed providesometimes conflicting signals as to the effective-ness of training-based active measures.However, in interpreting this difference, Faychooses to attribute the reasons to either theevaluation methods or the programmes them-selves (or presumably to a combination of both,though this is not really explored in the paper). Inthe former instance, he suggests three possibleproblems, summarised as follows:(a) the evaluation periods may be too short,

because those on training programmes tendnot to have the opportunities for job searchthat non-participants have, so the gapbetween completing a programme and evalu-ating the employment outcome needs to beadequate to reflect this;

(b) the amount of training undertaken by thecontrol or comparison groups needs to betaken into account, since even non-participa-tion on a particular training programme doesnot exclude some other forms of trainingbeing done;

(c) the fact that training sometimes leads tofurther training (and not employment immedi-ately) is not really considered, but could beconstrued as a very positive outcome that inthe longer term could lead to favourable joboutcomes.

Unfortunately, the analysis of the possibleproblems is not necessarily matched by thepotential solutions. So, for example, Fay cannotgive any firm indication on what might be theideal gap between programme completion andevaluation.

In terms of the problems arising from theprogrammes themselves, he sets out five possibleareas to consider, summarised as follows:(a) the perceived quality of certain training

programmes by employers may suggest thatparticipants are not really increasing theiremployability;

(b) training programmes that allow participantsto requalify for unemployment benefit oncompletion may not offer the best image to


(8) Costs shown here (and for subsequent categories) should be treated as indicative only and may relate to various years priorto 1998.

(9) Indicative figures only – see previous footnote.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 232

employers and may also encourage participa-tion on the basis of prolonging benefit entitle-ment;

(c) the type of training undertaken may notreflect the needs of the participant and thiscould lead to setbacks;

(d) evaluations may not reflect the changes in thestructure and content of training courses asthey adapt to different circumstances (e.g.labour-market needs);

(e) while training programmes can be among themost expensive of the ALMP options, expec-tations of a return for the individual (e.g. inearnings) should be modest.

These factors provide a useful input in thedesign of programme evaluation and Fay usesthem to suggest how training programmes mightbe designed in the most optimal way. He sets outfive main points to consider, not only in thedesign of programmes but also in how theyshould be evaluated, summarised as follows:(a) training programmes should not allow partici-

pants to requalify for benefits and nor shouldthey be seen as a large-scale solution tounemployment;

(b) small-scale, targeted programmes offer thebest prospect of positive outcomes, andshould reflect the needs of both employersand the job seeker;

(c) better understanding of the best ways todesign courses is needed, including the useof public or private providers and their dura-tion (with longer courses normally leading to arecognised qualification);

(d) measures aimed at youths should be consid-ered in association with general policies oneducation;

(e) evaluations should be carried out over longerperiods to see if the short-term effects onearnings and employment prevail in thelonger term.

As will be seen, some of the evaluation work invarious countries, done subsequent to Fay’sstudy, show some of these elements, though it isdifficult to identify any that could be consideredcomprehensive in their coverage of the points.

The review by Martin and Grubb (2001) onOECD countries, despite its publication year,

covers national studies carried out to 1999. It is awide-ranging review in that all types of ALMPsare considered and there is not much detail onindividual programmes, which, of course, limitsits usefulness. In addition, the authors’ objectiveswere to highlight those components of ALMPsthat seem to work and those that do not. There isa great deal of discussion on the use of theOECD statistics on active measures (see Box 1for further details of this) and on the measure-ment of programmes in terms of, for example,negative effects.

The authors go into some detail on the prob-lems and limitations of programme evaluation,citing the following factors as being particularlytroublesome:(a) by focusing on specific programmes, evalua-

tions tend to ignore other important policies(they call them active measures) such as eligi-bility for unemployment benefit, the registra-tion and matching of job seekers to vacan-cies, and job search assistance;

(b) during long spells of unemployment, in manycountries job seekers are often obliged toparticipate in some active measure, the indi-vidual motivational effect of which is generallynot captured by evaluations (10);

(c) drawing inferences on the aggregate effectsof active measures is difficult from the rangeof microeconomic evaluations that tend totake place.

The problem of frequent programme redesignand reinvention has already been alluded to andis a growing trend in many countries. While manyof these new approaches put the individualunemployed person at the centre of a customisedservice, it does not always mean that they aremaking free choices from the menu of activitiesavailable.

In terms of training programmes, Martin andGrubb point out that they tend to be among themost expensive of the active measures (for an indi-vidual participant). This is reflected in their share oftotal spend on ALMPs (Figure 2) (11). On average,spending on training measures in OECD countriesin 2000 amounted to 23 % of all expenditure onALMPs, and the authors point out that this propor-tion has changed little over the past 15 years.


(10) The authors give the example here of the Mutual obligation in Australia, which required the young unemployed to participatein one or more of 15 different activities and programmes.

(11) This may not be the case everywhere. For example, in some of the transitional countries, individual direct training costs aremodest when compared to some other programmes (Walsh et al., 2000).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 233

In their overall assessment of training andretraining programmes in selected OECD coun-tries, the authors found that there were mixedsignals on their efficacy when using the traditionalmeasures of earnings and employment effects.Countries where programmes have yielded partic-ularly low or even negative results includeCanada, Ireland, Sweden and the US. Thisconclusion is derived from comparison of therates of return for programme participants againstthe costs of achieving these effects (mainly thecost of programme provision and not the social oropportunity costs involved, which are generallynot measured). Unfortunately, the authors do notgo into great detail on where such programmesactually work, limiting their comments to thesuggestion that some public training programmesshow some useful outcomes.

However, the authors single out the US, whererecent studies quoted have found more encour-aging results. The authors mention work by Fried-lander et al. (1997), Heckman et al. (2000) and

Stanley et al. (1998), which collectively suggestthat there have been employment and earningsgains for some participants in training andretraining programmes. The authors point out thatthe studies also suggest that the labour forcesubgroup most consistently gaining from theirparticipation in these programmes is adult women.

Martin and Grubb emerge from their analysis ofthe training/retraining ALMPs with the followinggeneralised observations:(a) the subgroup gaining most from participation

in programmes were adult women;(b) for out-of-school youths, virtually none of the

training programmes rendered positiveresults (12);

(c) most gains (for whatever subgroup) were inthe form of enhanced employment potentialrather than increases in hourly earnings;

(d) where there was some observable positiverate of return for programme participants, theearnings gains were not sufficient to lift fami-lies from poverty.


The OECD maintains a database on public expenditure on labour-market programmes which providesthe basis for its regular comparisons of expenditure in member countries (cf. the annual Employmentoutlook). The information is as comprehensive as possible in that it covers all types of spending,including national and regional, local, etc. insofar as this is available (there are problems in ensuringall levels of expenditure are included).

The information is presented in five categories of spending on active measures:– public employment services and administration,– labour-market training,– youth measures,– subsidised employment,– measures for the disabled.

In addition, there are two categories for spending on passive measures:– unemployment benefits,– early retirement pensions paid for labour-market reasons.

The information on training is limited to public spending in the context of labour-market policies andtherefore excludes private sector spending on apprenticeship and other training (which can besubstantial in countries such as Austria, Denmark, Germany and Switzerland). Similarly, trainingfinanced through payroll tax is excluded.

The database is the subject of collaborative work with Eurostat aiming to extend the range of infor-mation available on public spending on ALMPs and participants, etc.

Box 1: Labour-market programmes database from the OECD

(12) The disappointing results for youth extend beyond training programmes, with Martin and Grubb suggesting that the majorityof evaluations have shown that virtually none of the ALMPs work well for this subgroup.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 234

The authors find little in the evaluations to showwhy the different sub groups should derive differentbenefits. In other words, the evaluations fail toexplore the determinants of the different outcomes,compared across subgroups of the labour market.

This leads to the conclusion that the evalua-tions are disappointing in their contribution todesigning the most cost-effective public trainingprogrammes, though they are able to derive whatthey describe as crucial features in the design ofsuch programmes, and these are as follows:(a) programmes should be targeting tightly on

participants;(b) programmes should be kept relatively small

scale;(c) programmes should lead to a recognised

qualification or certification that is recognisedin the labour market;

(d) programmes should have a significanton-the-job component so that links withemployers are established (13).

These are all sensible ingredients to look for intraining/retraining programmes, though are noguarantee of success, and have the furtheradvantage that they are broadly consistent with


Figure 2: Spending on ALMPs as proportion of total spending on labour-market policy

Percentageof total

Source: OECD (2001)

0

10

20

30

40

50

60

70

80

AU B CA D DK E EL F JP NL P S UK US

1985 2000

Tight targeting:– customisation costs more;– need for competent personal advisers;

Small scale:– causes problems with delivery;– difficult to link with labour-market need;

Local focus:– job opportunities may be elsewhere;– local training provision may not be available;

Qualifications/certification:– current system may not be able to cope;– programme duration may be too short;

Employer involvement:– can be difficult to get private sector

involved;– danger of substitution and displacement

effects.

Box 2: Key features and problems

of programme design

the findings of other studies such as that of Fay(1996) referred to earlier.

(13) Here, the authors also allude to the problems that fostering closer employer involvement in training programmes could have incausing some displacement of existing workers.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 235

4.3. Experience in the transitionalcountries

In theory, evaluations of ALMPs in transitionalcountries represent a purer basis on which towork, given that the programmes are often of thetraditional sort, unencumbered by the trend indeveloped countries towards the coordination ofactive and passive measures. There is also theadvantage (strictly from an empirical perspective)that unemployment is often high and so gives theprospect of a real test of efficacy in difficultlabour-market conditions.

Those countries relatively new to the use ofactive measures also tend to present a morestraightforward approach to their design andimplementation. The labour market, social andpolitical contexts in which they operate are oftenfar removed from those found in the MemberStates, for example, and these factors need to betaken into account when interpreting the results,but active measures in any country are adaptedto, and influenced by, such factors. It is thereforeimportant that, in looking at individual nationalstudies, these potentially influencing factors areassessed.

Fretwell et al. (1999) concentrates on the eval-uation of programmes in certain transitionalcountries and includes evaluations up to 1999. Itfocuses on four countries, the Czech Republic,Hungary, Poland and Turkey. This is useful in thatthey display different size characteristics (in termsof population and territory) and socioeconomicindicators as summarised in Table 2 below. Thereare also labour-market policy contrasts; of thefour, Turkey does not pay unemployment benefit.

The authors compare the results of evaluationsin the four countries and give useful detail on themethodologies deployed in each case, covering

such important factors as sample selection. Thelimitations of the study include reliance onquasi-experimental methods only, limited togauging the net employment and earnings effectsof participation in a range of active measures(including training).

The overall findings from the analysis arehelpful in coming to some notion of what worksand what does not work. The authors broadlyconcluded that active measures could have somesignificant positive benefits on both employmentprobability and earnings potential, but only forsome subgroups of the labour market. Sometraining programmes that were of longer durationand included group training were found to bemore effective, though the reasons for this wereonly guessed at. The case for group training, forexample, was thought likely to be influenced bythe characteristics of those entering the training,which could not be accounted for in the simula-tions.

Furthermore, the case for longer durationtraining was far from clear-cut. In Poland andHungary, there was some indication that shorterperiods of training (under six months) could havegreater impact on employment probability thansome of the longer training courses, which mustlead to the conclusion that it is more the designand content rather than just duration that deter-mines its value to the individual participant.However, given that there tends to be a relativelyhigh level of formal qualifications in these coun-tries, it may be that shorter periods of training aremore than adequate to adapt from the base qual-ifications held by programme participants. Theauthors also suggested that private sector provi-sion may be more effective than public sector.

Other indications from the review suggestedthat there was some evidence (though perhapsnot always conclusive) that certain subgroups


Indicator Czech Rep Hungary Poland Turkey

Population (millions) 10.3 10.1 38.6 62.7

Labour Force (millions) 5.1 4.5 17.6 22.2

GDP per capita (USD) 4 740 4 340 3 230 2 838

Unemployment (%) 3 11.2 13.6 6

Expenditure on ALMPs (% GDP) 0.14 0.43 0.32 na

Source: Fretwell et al. (1999) and OECD (2001).

Table 2: ALMP evaluations: country characteristics (1996)

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 236

benefited more from participation in a trainingprogramme, summarised as follows:(a) men and women can benefit, but the impact

may be higher for women;(b) young and middle-aged participants benefit

more than older participants;(c) impact may be greater for the short-term

unemployed;(d) those with primary and secondary education

only benefit more than those withpost-secondary educational attainments.

While the above findings generally concur withother cross-country reviews, the one possibleexception is educational attainments, with someother studies referred to indicating that the moreeducational attainments the participants have,the more likely they are to benefit from theprogramme.

A particularly useful single country study ofALMPs in Poland by Kluve et al. (1999) is notfeatured in the composite study by Fretwell et al.(1999). Setting the scene, the authors show thatPoland has a relatively long experience in theapplication of active measures, with training andretraining being particularly important. Seen as away of solving skill mismatches in the labourmarket, as well as a way of moving the unem-ployed into jobs, training or retrainingprogrammes were proposed as responsive to the

identified needs of employers. Training is one ofthree elements among the ALMP armoury used inPoland and it is also interesting to note that theauthors ascribe some human capital develop-ment benefits to them all. They argue that eventhe intervention works (14) and public worksprogrammes were designed to ‘enhance or main-tain the human capital of participants’ (p. 5).

However, because of difficulties with accessingsuitable data on the outcomes of programmeparticipation, the extent (and quality) of evalua-tions has been severely restricted. For this partic-ular evaluation, the authors took advantage of anenhancement to the Polish LFS to derive useabledata (Box 3).

Training and retraining has better results thanany of the other programmes. In particular, bothmale and female programme participants hadhigher employment rates post-programme thanthe control group, but only after locallabour-market conditions had been controlled for.The figures are that average employment ratesincrease by 15 percentage points for men and13 for women (but no significant effect on theunemployment rate). This leads the authors toconclude that participation in a trainingprogramme tends to prevent flows out of thelabour force rather than lowering unemploymentrates among the economically active. Neverthe-


The study by Kluve et al. (1999) made innovative use of the regular LFS by using a special supple-ment to gather retrospective information on respondents. The key question to be answered was:– For those that participated in an ALMP programme, whether they find themselves in a better posi-

tion in the labour market than would have been the case had they not participated?This was done by comparing employment and unemployment rates of those who had participated inan ALMP programme (treatment group) with the corresponding rates of those who did not participate(in effect a control group). Comparability between the groups was achieved by matching persons withthe same observable characteristics (e.g. age, gender, educational attainment, etc.) andlabour-market history before participation in a programme.In an effort to minimise any effects from the macroeconomic climate on labour-market outcomes, theauthors were careful to select those from both groups at identical points in time and economiccircumstances.The Polish LFS has been held quarterly since May 1992 with a rotating panel structure since May1993. There have been supplements on ALMPs in August 1994 and August 1996. The latter supple-ment provided a detailed monthly retrospective on the labour-market histories of participants over theperiod January 1992 to August 1996.

Box 3: Use of the LFS in Poland for ALMP evaluation

(14) Intervention works are programmes that provide for wage or job subsidies to the level of the unemployment benefit that wouldhave been received by participants.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 237

less, this leads them to conclude that thismeasure appears to improve the efficiency of thePolish labour market and so should attract moreresources.

However, the results should be treated withcaution for a number of reasons, even if the use ofthe LFS managed to provide reliable data, as theauthors claim. First, there is the problem of thedynamics of the policy arena, with programmeslikely to be changed on a regular basis, soaffecting aspects such as eligibility for participa-tion and overall number of places. Second, in tran-sitional countries in particular, the way in which theunemployed are assigned to programmes coulddistort the results of any evaluation.

For example, the selection of participantscould be biased towards (male) heads of house-holds who may be deemed (by the local labouroffices) deserving of prolonged income support,as derived directly from the structure of the bene-fits system. In Poland at the time of the study,unemployment benefits were paid for a maximumof 12 months and so those who are still withoutwork after the expiry of this period would tend torely on social assistance. However, participationin an ALMP means that, provided the programmeis completed, the participant could revert to anew 12-month period of unemploymentbenefit (15).

Another problem such evaluations have tograpple with is overcoming any distortionsbetween treatment and control groups from theirjob search activities during the review period (amore general problem, not just for transitionalcountries). Those participating in trainingprogrammes would tend to have less time tosearch for work than those not on a programme,and this could influence the employmentoutcomes of both groups. However, this wouldbe balanced, to some extent, by the trainingprogramme participants’ greater prospects in thelabour market due to enhanced skills (16).

The series of programme studies (largelysupported by the World Bank) carried out in thelate 1990s in such countries as Poland andHungary (e.g. O’Leary, 1998a and b), established

a precedent for other transitional countries tofollow under encouragement from the WorldBank. The 2000 report on the situation in Bulgaria(Walsh et al., 2000), for example, essentiallyfocused on two aspects of ALMPs, their netimpact and their cost-effectiveness. This involvedthe use of a national survey of programme partic-ipants approximately 12 months after they hadcompleted their active measure, alongside acontrol group of those who shared the samecharacteristics but who had not participated in aprogramme.

The survey was carried out through personalinterview at the home of the respondent, andefforts were made both to ensure a high responserate and good coverage of all the regions. Animportant feature of the questionnaire design wasto elicit quantitative and qualitative information,the latter covering such factors as the respon-dent’s perceptions of how their participation in aparticular programme helped them, their assess-ment of how well it was run, etc. This softer infor-mation was used extensively in the report tocomplement the econometric analysis.

Employment outcomes were tested under fourdifferent possibilities, as follows:(a) ever re-employed in a regular job or

self-employment,(b) ever re-employed in any job or self-employ-

ment,(c) re-employed in a regular job or self-employ-

ment on the survey date,(d) re-employed in any job or self-employment

on the survey date.This was to distinguish between those moving

into the regular labour market from those findingemployment in another sphere such as withinanother active measure, which was quitecommon, particularly in the regions with higherunemployment (17).

The broad findings from the study showed thattraining programmes generally have positive neteffects on employment, with the added advan-tage that they are among the least expensiveoption among the menu of active measures. Inparticular, they work very well for both men and


(15) The unemployed participating in training and retraining programmes received an allowance equivalent to 115 % of the flat rateunemployment benefit.

(16) The study also found that those targeted for the training programmes tended to have more existing human capital than thoseon other programmes.

(17) In 1999, most regions of Bulgaria had double digit unemployment rates and many had over 20 or 30 % unemployment.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 238

women aged below 30 and above 44, for virtuallyall educational levels except those with highereducation, and best of all in those regions withmedium to high levels of unemployment. Forexample, in the highest unemployment regions,the net effect was found to be around 16.8 %,which fell to just 1.9 % in low unemploymentregions.

A review of the effects of ALMPs in Slovakia byvan Ours (2000) focuses on the individual treat-ment effects of training and subsidised jobsprogrammes using regional data in selecteddistricts. In these areas, information was collectedon all the registered unemployed so that individuallabour-market histories could be compiled (basedon three possible states – employment, unem-ployment and out of the labour force), as well asany participation in the relevant active measures.The author goes on to explain that the informationwas sorted with respect to the duration of timethe unemployed spent in the various states tocapture the labour-market dynamics of theseevents over time (18).

In Slovakia, the use of training for the unem-ployed has been a relatively small part of the totalresources devoted to ALMPs (the training orretraining of those still in employment waspreferred). However, during the 1990s, it seemsthat there was a requirement that training for theunemployed should be conditional upon thepromise of a job at the end of it. This being so, itis not surprising to note that the programmedelivered a large positive employment effect inevaluation. The results in van Ours study show

that participation in training has a very high posi-tive effect on employment probability, around sixtimes the rate for others, though the findings areheavily qualified by the job guarantee provision.

Finally, we refer to a study by Schwegler-Rohmeis(2000) for the World Bank but focusing onKyrgyzstan, a country in the early stages of itslabour-market development. Here, the author wascharged with the task of evaluating all activemeasures and this he tackled with a combination ofqualitative and quantitative information. In the lattercase, this involved using administrative data, inparticular the following:(a) the proportion of ALMP of total expenditure

from the Employment Fund (i.e. the govern-ment budget allocation for all passive andactive measures),

(b) the expenditure per capita in each of theactive measures,

(c) the average duration of participation in eachmeasure,

(d) the placement rates post-participation in aprogramme.

This was a case of making the best use of theinformation available and the author categorisedthe programmes’ performance into three levels(strong, moderate or weak). Training andretraining attracted a moderate performance levelas did most programmes; only special job place-ment activities and employment promotion agen-cies resulted in a strong performance.

The findings led the author to outline recom-mendations on how the programmes might beimproved and in the case of training and


(18) Here, the author was keen to overcome some of the inherent dangers in this type of analysis where those moving into anALMP may have unobserved characteristics. To overcome this, transitional rate models are estimated that allow for the threetransitional rates to be interdependent.

The response to the survey of participants from the training programmes provides a particularly inter-esting insight into the quality of provision and complements the rather one-dimensional econometricanalysis. Some of the key points that emerged are summarised below:– 23 % claimed that participation in the programme gave them more confidence;– 43 % felt that their prospects of finding work were poor, but most attributed this to the generally

low level of labour demand;– over 69 % of those finding work did so in the private sector (which was far higher than for other

programme areas);– 73 % of those in a job felt that their new skills were being used, at least some to extent, and the

figures were higher for those involved with additional training or retraining.

Box 4: Bulgaria – participants’ perceptions of the training programmes

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 239

retraining, the principal suggestion was thatprovision needs to be labour market driven ratherthan relying on what has been provided in thepast, or what clients prefer.

The evaluations carried out in the transitionalcountries are instructive in a number of ways. First,they show that useful evaluation of active measuresis feasible in sometimes difficult labour-marketconditions. Second, the actual results suggest thattraining and retraining programmes can haveimportant positive effects on employment proba-bility, albeit with some limitation imposed by thegeneral state of the labour market (i.e. where thedemand for labour is low, then employment proba-bility will tend to be lower, though those job seekerswith relevant training will tend to fare better). Third,training/retraining frequently emerges as a rela-tively low-cost option from the array of activemeasures, and this differs from the situation in theMember States, for example, where training is oftenat the upper end of the cost spectrum.

4.4. Selected national studies

The above studies provide a valuable overview ofevaluations throughout the 1990s, particularly inEurope and North America. In this section of thereport, we focus on a selection of national evalu-ations that we consider to be instructive in boththe methods of ALMP evaluation and theoutcomes from programmes.

The studies chosen are selected from a shortlist of relevant studies we have identified. They arebiased in favour of the Northern MemberStates (19) simply because this is where we havefound the most relevant and robust work. This isnot to say that relevant studies are not available inother Member States, and we make no pretencethat this is an exhaustive examination of allsources. However, our research has pointedalready to the lack of transparency of (or accessto) some of these sources to external researchers.Consequently, we have not uncovered any partic-ularly useful seam of information (notwithstandinglanguage restrictions) in many Member States.Much of the evaluation effort across the EU isnow stimulated by the requirements of the Euro-

pean Commission, and in particular the ESF andEES, and these have a wider constituency that isdiscussed in Section 4.5.

As a prelude to discussing some of these nationalstudies, Table 3 contains a summary of some of themore prominent evaluations of trainingprogrammes in the 1990s. Again, it is not anexhaustive list but is, we feel, illustrative of what hasbeen done and, more importantly, what has beenfound through these studies. Most have used aquasi-experimental methodology, though in thecase of the study of various training programmes,reliance was placed on an analysis of follow-upsurveys of participants, while in the US the class-room training element of the Job Partnership Actwas subject to a random assignment experiment.

Looking at the outcomes from these evaluations,in Canada, Ireland and Norway the results wereencouraging in that participation in trainingprogrammes had positive employment, and in somecases earnings, effects, particularly in Canada. InIreland, it is interesting to note that the positive effectswere not uniform across all sectors and, furthermore,were to some extent dependent on whether they ledto recognised qualifications. However, the findingsfrom the two programmes in the UK (Employmenttraining and Employment action) were mixed, withthe indication (perhaps obviously) that traininginvolving job placement was more likely to lead toemployment afterwards. The findings from the USstudy were the least encouraging, with no significantimpact detected on earnings.

Here, we also make reference to an interestingstudy in Australia. Stromback et al. (1998)reviewed the effect of labour-market programmeparticipation using the Survey of employment andunemployment patterns (SEUP) databasepublished by the country’s national statisticalagency. This is particularly noteworthy in that itcontains matched records of programme partici-pants, employment support services and socialsecurity records, thus providing a ready source ofinformation on assessing the effectiveness ofprogramme participation.

In this paper, the authors considered the impactfrom three different perspectives: the effect on theprobability of participating in the labour force at theend of the period; the conditional effect on partici-


(19) Given the extensive coverage in the papers by Hujer et al. (2003b) of the situation in Germany in the third research report, wehave chosen not to replicate them here.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 240

pating in the labour force; and the effect onpost-programme earnings. All types of activemeasure were covered by the ‘Working Nation’initiative (introduced in 1994), one of which wasemployment training. Training emerged positivelywhen the probability of working was considered,but interestingly this did not translate to actual posi-tive job participation. Similarly, the authors couldfind no effect of programme participation onpost-programme hourly earnings.

4.4.1. Sweden

There is a disproportionate amount of evaluationresearch coming out of Sweden, no doubt due tothe commitment shown through the dedicatedInstitute for labour-market policy evaluation(IFAU) and the availability of a database goingback to 1991 (Box 5). Recent work by Calmforset al. (2002) is illustrative of this.

In their paper, they review previous evidence ofALMP evaluation and usefully plot the changes inthe orientation of training programmes.Labour-market training originally concentrated onvocational training programmes, but the authors


Country Programme Evaluation Outcome Referencemethod source

Canada Employability Quasi- Significant effects Human improvement experimental on employment duration Resourcesprogramme and earnings for Development

job-related training Canada, 1995

Ireland Various training Follow up surveys Positive effects of O’Connelland employment of programme training with labour et al., 1997

programmes participants market relevance on employment

and wages

Norway Labour-market Quasi- Significant positive Raaumtraining programmes experimental effects where training et al., 1995

led to qualificationsin some sectors

United Kingdom Employment training Quasi- Employment training had Payneand Employment experimental positive effect on job et al., 1996

action programmes probability but Employment action did not. Training with

job placement better

United States Job training Random No significant Bloom, 1994partnership act – assignment impact of training

classroom training experiment on earnings

Table 3: National evaluation of training programmes in brief

point out that this has changed. There is now moreemphasis on general education through theseprogrammes and it includes such aspects asSwedish language training for immigrants, ITtraining (starting with Computer activity centres in1995) and so on. This complicates any comparativeevaluation of training effects.

For recent research on the labour market, ithas been possible to make use of a longitu-dinal data set containing the event history ofall the unemployed registered with offices ofthe National labour-market board (AMS) since1991 (called Handel). This allows evaluators inparticular to examine the background andparticipation in ALMPs over a long period, aswell as facilitating the selection of controlgroups. Information is held on such character-istics as education, work experience and indi-vidual labour-market histories so the reasonsfor ending an unemployment spell are logged.

Box 5: Sweden: a unique labour-market data

source

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 241

However, they reveal that studies that attempt touse quasi-experimental methods (mainly treatmentto control group comparisons) to measureprogramme effectiveness can encounter the problemof finding a suitable control group. This is because ahigh proportion of the unemployed participate in oneor more programmes during their unemploymentspell (20) and for the long-term unemployed it iscommon for a string of programmes to be followed.As they put it, ‘the choice for an unemployed is toparticipate in a programme now or later, rather thannow or never’ (Calmfors et al., 2002, p. 26).

Looking at the long history of ALMP evaluationin Sweden (Table 4) the older studies of trainingprogrammes suggested that, in the early 1980s,there were positive effects observed on theemployment and earnings potential of partici-pants. However, by the 1990s, the evaluationswere clearly tending to show insignificant or evennegative effects. It is not wholly clear why this isthe case, except to allude to the problems ofgetting usable data. Here, a study by Sianesi(2001) attempted to tackle this problem by esti-mating the effects of joining a programme at acertain point as opposed to later, with the controlgroup selected from the unemployed who werenot participating in an ALMP at that time.

In the event, the results were not wholly satis-fying, in that the short-term employment andearnings effects of training were still insignificantor negative, though more encouragingly theywere less insignificant (but still not positive) overa few years (of the 1990s).

To some extent, the data problems can be put toone side by considering other evaluationapproaches. For example, Calmfors et al. (2002)mentions the more positive results emerging fromsurveys of employer attitudes towards the unem-ployed in Sweden. Here, there is a tendency foremployers to ascribe some value to those who haveparticipated in ALMPs, and labour-market trainingis given preference above other programmes.

However, this upbeat perspective has to beseen in the context of a generally disappointingperformance from Swedish training programmesin the 1990s. Calmfors et al. come to the conclu-sion that training programmes ‘seem not to have

enhanced the employment probabilities of partic-ipants’ (p. 2). Moreover, it is suggested that partof this inefficiency was due to the scale of theprogramme, with a lack of infrastructure to copewith the expansion of training programmes in the1990s. Part of the problem was ascribed to thelow demand for labour in an economic downturnwhich inevitably had an effect on placement aftertraining. Also, there were the usual problems ofpredicting not only the timing of the economicupturn but also where any skill gaps might be.

A different objective motivated the evaluation byJohansson and Martinson (2000), again in Sweden.They started off from the premise that they wanted totest whether increased employer contact andinvolvement within a labour-market programme hadany measurable effect on participant outcomes,using survey-based data on participants to comple-ment the administrative statistics on participation.As the test bed, they chose the Swit programme,introduced in 1997 to develop IT competences withinthe labour-market training programme (for the unem-ployed) (21), an integral component of which was theinvolvement of industry which would host elementsof the training for the individual.

One specific programme was the Traineereplacement scheme. This allowed employers toreceive subsidies for a maximum of six months tocover the cost of training for an existingemployee, provided that they also recruited areplacement worker. The programme, therefore,provided training and new jobs.

The authors wanted to compare participationin the normal labour-market training programmeand the Swit, and so used data from two mainsources to carry out their evaluation. First, thedatabase of the unemployed (Handel) wasaccessed for a sample of participants. Second, atelephone survey was carried out. This coveredaround 1000 individuals who participated in eachof the relevant programmes during two months in1999 and was conducted in June 2000 – aboutsix months after the participation (22).

Overall, the evaluation indicated that being on theSwit course as opposed to other labour-markettraining increased the chances of finding a job (sixmonths) afterwards by as much as 20 %. The


(20) The authors suggest that at their peak in the mid 1990s, ALMPs covered more than 5 % of the Swedish labour force.(21) The background to the programme included a growing awareness in industry about the lack of workers with IT skills

somewhere between specialists and users of IT.(22) Response rates were high in both the Swit and LMT courses, with 79 and 80 % respectively.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 242

authors dismissed any effect on this figure arisingfrom any differences in the way participants wereselected for the programmes, and thereforeattributed the good performance almost wholly tothe employer contact that provided practical expe-rience for the participants. However, what is notclear from the research is whether the Swit partici-pants found jobs in IT (using their new skills) or inanother unrelated job. Similarly, it is unclear if partic-ipants worked with the employers with whom theyhad established contact during their programme. Ifso, this suggests that the benefits of the programmeextended beyond the acquisition of a skill in demandin the labour market, to job search and placement.

4.4.2. Denmark

Staying with the Nordic countries, but this timelooking at Denmark, a study emanating from theMinistry of Labour (2000) provided a thorough

appraisal of the Danish employability enhancementprogrammes (EEPs) over the 1990s. The EEPsinvolve the usual range of ALMPs, includingtraining, and a key objective is to bring about acti-vation early in the unemployment period to preventthe drift into long-term unemployment, after whichit becomes more difficult for the unemployedperson to return to the regular labour market.

A number of studies were carried out duringthe 1990s, most using treatment and controlgroup comparisons of various sorts drawing onspecially commissioned surveys of participantsas well as using administrative data. However,one approach for gauging the effect of upgradingthe qualifications of the unemployed (a key partof the EEPs) was examining whether participationin a programme led to a reduction in the paymentof benefits compared with non-participation (theso-called fixed-effect model (23)).


Programme Measure Outcomeand date

LMT, 1993 Regular employment six and Positive effect only if potentialtwo and a half months after programme selection is not considered

LMT, 1994 Regular employment Significant negative effect oftwo years after programme training below 100 days; no

significant effect above 100 days

LMT, 1989-91 Yearly income Significant negative effect one yearand insignificant negative effect

three years after programme

LMT & Computer Regular employment Significant positive effect ofActivity Centres, 1996 one year after programme LMT; no significant effect

of computer activity centres

LMT, 1992-93 Yearly income; and probability to Significant negative effectsa) obtain a job or

b) proceed to regular education one to two years after programme

Swit, 1999 Regular employment Significant positive effectsix months after programme

LMT, 1998-99 Regular employment Significant, large positive effectssix months after programme

LMT = Labour-market trainingSwit = IT competence training for the unemployedSource: Calmfors et al. (2002)

Table 4: Sweden: summary details of ALMP evaluations involving training in the 1990s

(23) The fixed-effect model takes account of the problem posed by the control group by allowing programme participants to repre-sent their own control group. It adopts a counterfactual approach whereby assumptions are made about what would havehappened to the individual if they had not participated in a programme.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 243

The report also describes an evaluation carriedout in 1997 that attempted to look at the effec-tiveness of programmes post-1994 when reformsshifted the emphasis towards decentralisation,individualisation and targeted initiatives. Thisinvolved using the national labour-marketauthority’s (NLMA) administrative database onthose registered as unemployed, alongside aspecial survey of participants in EEPs (24) whowere questioned about their perceptions of thebenefits from the programme. Further surveyswere carried out for the NLMA in 1997 and 1999and the most recent results tend to corroboratewhat emerged from earlier studies, with theeffects summarised as follows:(a) overall, there were positive employment

effects from participation in EEPs;(b) while both public and private sector training

programmes have significant positive effects,that for private training programmes is better;

(c) training programmes under EEPs reducegross unemployment;

(d) by 1999, the positive effects of programmeparticipation had increased to the extent thatsome 32 % of participants eventually entereda job;

(e) the positive effects were less significant asthe age of the participant increased;

(f) the positive effects increased with higherlevels of education among the participants.

The better performance in finding employmentof those following training programmes in theprivate sector was largely due to more of theparticipants finding work with the employer wherethey did their training. Furthermore, one studyconfirmed that better outcomes came fromprivate sector training when it was initiated rela-tively early in the individual’s unemploymentperiod, whereas it was the opposite in publicsector training. The reasons for this are far fromclear, though in the former case it is likely thatthose selected for training in the private sectorare likely to have short unemployment experienceand so may be able to settle in with theiremployer much more easily.

4.4.3. Belgium

A much wider cost-benefit analysis of the BelgianTOK programme by Nicaise (2000) drawstogether information from various evaluations ofthe programme. The TOK programme has beenrunning for some years and aims to offer employ-ment and training opportunities for the underpriv-ileged, most of whom are the long-term unem-ployed. It principally involves participants beingoffered temporary employment contracts on suchtasks as construction, building renovation work,assistance to families and the elderly, all adminis-tered by local welfare centres. A review of theprogramme in 1989 discovered that the numbersof programme participants moving into regularemployment were not encouraging, and so theprogramme was revamped to include moreemphasis on counselling and training (25).

Various evaluation methods have subsequentlybeen deployed to examine the performance ofthe programme, with perhaps the most thoroughbeing that of Wouters et al. (1994). This studyinvolved interviews with a group of TOK partici-pants from 1990 (in effect, around two years aftercompletion of the programme) and includedsome who had failed to complete theirprogramme (for various reasons). A control groupwas selected from non-participants but withsimilar characteristics based on age, gender, levelof education, nationality, marital status and dura-tion of welfare support, and matching with thetreatment group was carried out.

The results of the analysis showed that TOKparticipants had performed much better thannon-participants, in both employment opportuni-ties and earnings, in the two years after theycompleted the programme. Furthermore, theprofile showed that the programme participantstended to have a weaker employment historypre-programme than the control group, whichadded to the positive outcomes from participa-tion. However, the findings were tempered by thefact that only just over two-fifths of theprogramme participants were in work 12 monthsafter completion of the programme, a situationlargely attributed to the relatively poorlabour-market conditions prevailing at the time.


(24) A representative sample of 5000 unemployed participants were selected from the administrative records with a response rateof over 84 %.

(25) The programme combines 400 hours of general training, 400 hours of practical professional knowledge and 40 hours of famil-iarisation with new technology, with an average programme duration of 14 months.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 244

The author attributes much of the success ofthe TOK programme to the fact that it offerstraining geared to labour-market needs andfulfilling work for the participants, which isreflected in aspects such as low dropout.However, it is considered to be an expensiveprogramme (partly because of its relatively longduration) in strictly financial terms, but with muchwider social benefits. These wider benefits arediscussed in the report rather than quantified,and include the following:(a) programme participants can fill vacancies

that are not liable to be filled by other jobseekers;

(b) the sorts of training given can lead to efficien-cies in the labour market by helping avoidskill shortages which, in turn, could reducewage pressures and lead to aggregateemployment benefits;

(c) the training of programme participants canlead to the creation of jobs in related fields;

(d) productivity gains for employers from thetraining can lead to lower costs and prices,etc.;

(e) overall the returns to government in terms ofwelfare benefits saved, income tax paid, etc.,can be positive.

The point being made is that a thoroughcost-benefit appraisal of the effects of the TOKprogramme would bring together all these (andpossibly more) issues, but this is never done.

In a similar way, the extended benefitsaccruing to the programme participants (aside,that is, from the real benefits of income) areseldom assessed in evaluations. With the TOKprogramme, aiming as it does at particularlydisadvantaged groups, benefits were identified asincluding increased self-confidence, greatersocial contact, follow-up training, the regularisa-tion of social security rights, etc. Furthermore,there are benefits for the municipality in terms ofa reduced workload for the welfare centres and(probably) a lower incidence of crime and healthproblems.

4.4.4. Germany

A number of other studies that we have comeacross tend to dwell less on the actualprogrammes and more on the methodologicalapproach to evaluation. We have not gone intogreat detail on the German examples because thisis the subject of a separate focused paper by Hujer

et al. (2003b) in the third research report. However,Fertig and Schmidt (2000), for example, begin theirpaper by outlining the lack of proper evaluation ofGerman ALMPs, even though the process of gath-ering data through local labour offices hasundoubtedly improved. However, the point theymake is that this information is not systematicallyused to evaluate programmes, and so they then goon to describe what should be done.

Their basic requirement for effective evaluationis to establish what they call a credible counter-factual situation (p. 17), or what would havehappened in the absence of the intervention (this,of course, is a general requirement for most eval-uation approaches where control groups orbefore-and-after estimates are used). Neverthe-less, this is easier said than achieved and theauthors admit that ‘at best the effect can only beestimated with confidence, but never measuredwith certainty’ (p. 17).

When looking at the costs of German ALMPsthat aim to improve the skills of the unemployed,they outline three strands as follows:(a) costs of treatment (expenditures incurred by

the participants such as course fees, trans-port, subsistence, etc.);

(b) time costs of participation (the opportunitycost foregone by the individual);

(c) time costs involved with programme adminis-tration and delivery (e.g. the opportunity costsof administrators).

The problem is that opportunity costs arealways difficult to quantify and so tend to beignored in any evaluation exercise.

4.4.5. Ireland

A more divergent approach to evaluation is takenby Conniffe et al. (2000) in a study delving intopast evaluations in Ireland. Here, they explore theuse of natural experiments and propensity scoresas alternative (or parallel) activity. The authorspoint out that the propensity score approach isoften used in the field of epidemiology (andothers) and it rests on the notion that matchingindividuals in groups should be compared witheach other, with the usual basis for inclusionbeing the prior probability of programme partici-pation. The approach is seen as an alternative tothe use of control groups with all the attendantproblems of making assumptions on who shouldbe in them. The natural experiment occursthrough using observational data on what are, in


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 245

effect, treatment and control groups, but withoutwhat the authors call deliberate randomisation ofindividuals to groups.

In order to test the effect of their approach, theauthors apply the technique to previously gath-ered data on training programme participants(and specifically those following generaltraining) (26) and reveal that their findings corre-spond with those of the traditional evaluation.The advantage in this, they argue, is that thepropensity score approach makes fewerdemands on the evaluator to make assumptionsabout the control group in particular and soshould lead to less spurious results.

4.4.6. United Kingdom

While the UK has a fairly consistent track recordof programme evaluation, much of it is of thequalitative sort with few quasi-experimentalstudies in evidence. Nevertheless, a key strengthof the work here is its attention to all stakeholdersinvolved in programme take-up and delivery,approaching the holistic perspective that is nowbecoming recognised as desirable.

A good example of this approach is the evalu-ation of the prototype employment zones (PEZs)by Haughton et al. (2000). These were set up onan experimental basis in 1998 in five areas of highunemployment across the UK, and the evaluationwas commissioned at the same time to runalongside the development of the PEZs. Thebasic idea behind them was to bring greater flex-ibility and local innovation to helping the adultunemployed (those over 25 years old) return tothe labour market and was the first real attemptat turning the provision of employment servicesround to a more client-focused approach.

The evaluation was essentially a longitudinalstudy tracking the development of the PEZs andinvolved specific fieldwork at various points in thetwo years of research. Interviews with key stake-holders – including clients (mainly in focusgroups) – was the core activity and the resultsprovide some interesting conclusions on the pilotactivities. However, one problem encountered(which is consistent with many complex evalua-tions of this sort) was that it was difficult to sepa-rate out the relative effects from such elementsas programme design, rules of engagement and

the prevailing labour-market conditions, forexample. Nevertheless, they were able toconclude that the new approach tended to over-come a key problem of the traditional approachthat tended to offer training for training’s sake.Under the client-centred approach in the PEZs,learning for work became the most popularoption and flourished partly because local provi-sion developed to meet the challenge, with newcourses and greater flexibility offered.

4.5. EU Evaluations

While in some of the Member States there is atradition of rigorous ALMP evaluation, in themajority this is not the case. However, theemphasis now given to the evaluation of projectsin the ESF and to reviewing progress with theEES has introduced a greater discipline toMember States, such that all provide some formof evaluation. There are, however, still substantialdifferences between the rigours applied to theevaluations across the EU. In this section of thereport, we look at some of the recent studiesdone both individually and cross-EU in ESF, andin the next section for EES.

4.5.1. European Social Fund (ESF)

The various labour-market programmes fundedunder the ESF should, in theory, provide a richseam of evaluation material on three levels:(a) project level: each project funded is required

to have an explicit monitoring and evaluationstrategy, with resources attached to makesure it happens;

(b) national level: Member States were involvedin mid-term and final evaluations of theirnational ESF programmes in the 1990s;

(c) EU-wide: the European Commission, drawingon the national reviews, provided EU-wideperspectives.

As far as the individual project evaluations areconcerned, these will inevitably be variable andwould be much too detailed for use in ameta-analysis of this nature. However, theyshould contribute to the national studies and so itis reasonable to assume that in those Member


(26) For a discussion of the more traditional approach and the data mentioned here see O’Connell and McGinnity (1997).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 246

States where this was given a strong profile, thenational studies would be stronger as a result.

The national studies provide useful materialdespite the fact that they were not generallyexecuted to a consistent set of methodologicalguidelines. The Portuguese evaluations, forexample, focused on measures rather thanprogrammes, while in Ireland the attention was oncontinuous evaluation. The range of techniquesdeployed also included quantitative and qualita-tive methods, with the widespread use of discus-sions (interviews and focus groups) with the keystakeholders. The main outcome measured wasemployment, though not exclusively so, withother issues such as the impact on beneficiariesand systems also receiving attention in somestudies. Furthermore, the reviews were muchmore wide-ranging than our focus in this reporton training and retraining measures.

Unfortunately, accessing the national studies ishampered by their availability only in thelanguages they were written in and the fact that

they have not been uploaded to the Internet. Wehave, therefore, largely relied on the compositecross-EU reports for our information, notablyEuropean Commission (1998) for the mid-termevaluation, consolidated in the EuropeanCommission (2001) report that covers measuresunder Objectives 1, 3 and 4 (27). Table 5 belowsummarises some of the main findings relating tothe impact on beneficiaries.

The ESF interventions are generally morebroad ranging than simply focusing on trainingand retraining, which should be seen as one oftheir strengths. Furthermore, they do not neces-sarily apply exclusively to the unemployed,though a high proportion of the beneficiaries willbe without work or face the threat of redundancy.Nonetheless, the findings indicate some of thesame problems that have emerged from nationaland other evaluations, namely that the targetingof policies always works better and linkingtraining to what employers and the labour-marketneed is always likely to improve the impact.


(27) Objective 1 – to develop regions which are currently underdeveloped;Objective 3 – to tackle long-term unemployment, promote equal opportunities, improve lifelong learning, encourage entrepreneur-ship and adaptability and improve the role of women in the workforce;Objective 4 – to improve the qualifications and prospects of all those in employment.

ESF programme Main points

Objective 1 • Net impact was greater when programmes concentrated on the most disadvantagedgroups;

• In some cases, combining measures increased their net impact;

• Programmes offering work experience are best when combined with some form ofskills training.

Objective 3 • overall, gross employment impacts ranged from 30 % to 80 %; clients showed highlevels of satisfaction with their programmes;

• the chances of findings work after a programme depended less on the beneficiaries’personal characteristics than the availability of jobs in the labour market and thetype of project.

Objective 4 • some evaluations suggested that employers benefited more than workers fromsupported activities;

• there was a lack of attention to identifying those skill areas of most need in thefuture and applying this knowledge to project support;

• some of the training supported through projects was considered to be too generalor, in some cases, not sufficiently transferable between employers.

Source: European Commission (2001)

Table 5: ESF evaluations: summary of main points

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 247

4.5.2. European employment strategy (EES)

The EES began in 1998 with the aim of improvingthe performance of labour-market policy inMember States. It achieves this through what aresometimes referred to as soft learning instru-ments, essentially based on peer reviews andbenchmarking policy and practice betweenMember States. Within this, there is a require-ment that Member States review theirlabour-market policies on a regular basis, the firstthorough review being carried out in 2001-02 andcovering the first three years of the EES.

Each of the Member States submitted a review tothe European Commission (28). The papers are ofvariable depth and quality and not all could beconsidered evaluations, with some simply descrip-tive reviews of what has been going on inlabour-market policy, without any critical appraisal.

These papers have a much wider coveragethan ALMPs and our concerns tend to beconcentrated in the discussion of Pillar I: Employ-ability (29). Nevertheless, they provide a usefulinsight into the approach to ALMPs, and help fillthe large gap between those Member States witha tradition of ALMP evaluation and those wherethis is still an underdeveloped activity.

4.5.2.1. SpainThe Spanish EES review is a fairly thoroughattempt at evaluating the main measures fundedby the public employment service (see Box 6 fordetails of training measures in Spain). It includesan econometric analysis of the administrativedatabase in order to derive the impact of trainingmeasures using two key indicators:(a) the chances of participants having a job at

the time of the analysis (called the hiring rate);(b) the chances of participants having had a job

at one point during the accumulated observa-tion period (called the employability rate).

The report claims that the two measures offercomplementary information on the success ofparticipation in the programme, with the better ofthe two being the employability rate, which offersa better indicator of the longer-term jobprospects of the participants. This is consideredparticularly important in the context of theSpanish labour market, where there are relatively


(28) Available from Internet: http://europa.eu.int/comm/employment_social/news/2002/may/eval_en.html.(29) Theme 1: Prevention and activation policies for the unemployed, being particularly relevant within the first Pillar.

high levels of labour turnover making categorisa-tion difficult.

The EES evaluations for Spain involved twokey approaches, the first of which relied on aninterpretation of the hiring and employability ratesfrom the administrative data alone (but includingprogramme participants and non-participants),while the second used estimation techniques tomeasure the performance of the treatment groupagainst a control group. The former analysis gavedisappointing results, albeit positive, but theeconometric analysis was more encouraging,suggesting that training significantly improved theemployability of certain subgroups.

Those most benefiting from the occupationaltraining programmes were the disadvantagedyouth, disabled persons and ex-convicts, whichequated to those with the lowest levels of educa-tional attainment. However, the report indicatesthat there was some positive effect for mostparticipants in the training programmes.

4.5.2.2. The NetherlandsThe Dutch review of the EES also provides a thor-ough overview of the performance of ALMPs inthe context of a strong tradition of evaluation.However, within the paper, there is a lengthydiscussion of evaluation methodology and howthis has been used in the Netherlands. This leadsto some critical comment on the track record,with the suggestion that too many of the previous

Recent changes in the approach to profes-sional training have been based on decentrali-sation, with the management of programmesdone through accords between the govern-ment and the key social actors. Similarly, thedecentralisation of the public employmentservice has also led to regional authoritiesbecoming actively involved in the design anddelivery of training for the unemployed. Along-side this has been a more targeted approachtowards the more disadvantaged subgroups ofthe labour market, such as the short-termunemployed youths and adults.

Box 6: Training measures in Spain

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 248

evaluations of policy have relied on a simplisticapproach, providing only measures of grosseffectiveness and not net effectiveness.

The report itself does not engage in any newevaluation of ALMPs, but instead provides anoverview of the wide range of training measuresthat have been tried (and often tested). Theseinclude programmes aimed at individuals andemployers, with some involving private fundingas well as public (such as direct payments, taxsubsidies, etc.) and this leads the report toconclude that these factors contribute to thecomplexity of the task for the evaluator.

Drawing on previous evaluations (and here theoverview by De Koning (1998) is useful) andparticularly those of the quasi-experimental sort,the report suggests that the effects of participa-tion in a training programme have changed signif-icantly throughout the 1990s. In the first fewyears of the decade (to 1993), the indications arethat there were important positive effects, espe-cially for the long-term unemployed where thechances of finding work were around 10 %higher. However, by 1997, the evaluation literaturewas indicating that there was no significant effectof programme participation.

4.5.2.3. IrelandThe review for Ireland provides a thoroughappraisal of current ALMPs implemented underthe national employment action plan (NEAP).Recent emphasis has been on giving support tothe unemployed at an early stage in their unem-ployment period, and this can include a place ona training programme. However, the emphasis isinitially on the provision of a job (if suitable andavailable), followed by guidance and counselling,and it is within this process that a place on atraining programme might be offered.

The Irish review uses administrative data gath-ered through the NEAP, as well as consultationswith key players in the social and labour-marketfields (the latter particularly to understand theeffects of the EES itself), to assess the effects ofpolicy. In the case of ALMPs, the report affirmsthe country’s commitment to an array of suchmeasures and their contribution to helping alle-viate social exclusion, labour shortages and skillsdevelopment, though at the same time questionswhether they always work. As the labour markethas become tighter, the report claims that theprobability of leaving unemployment becomes

more dependent on an individual’s capacity tocompete in the labour market, citing the mainexits from unemployment as evidence of this.Those faring best include the following:(a) younger people;(b) those with educational qualifications;(c) those who had recently participated in voca-

tional training;(d) those with previous work experience;(e) those who began with a short spell rather

than a longer spell of unemployment.These factors combined provide a blueprint for

moving people from unemployment into regularwork and are evidenced in the Irish approach toALMPs. Furthermore, training programmes haveassumed greater importance in the recent lowunemployment scenario, offering some palliativefor skill shortages.

4.5.2.4. United KingdomThe UK has also faced similar problems to theIrish labour market, with relatively low unemploy-ment and skill shortages, but with the unem-ployed needing to be equipped to compete inthis dynamic job market. The EES review forthe UK provides a useful overview of the evolu-tion of current policy, drawing on some previousevaluation work, but providing no new analysis.

The foundation for ALMPs is the New Dealprogramme, launched in 1998 and continuallydeveloped ever since, with the introduction ofdifferent strands targeting different labour-marketsubgroups. With some exceptions, New Dealkicks in for young people (aged 18-24) once theyreach 6 months’ unemployment, whereas forthose over 25, the trigger is 18 months.

One exception to the rule is the New Deal50 Plus, which, as the name suggests, is targetedat those unemployed aged 50 and over with theaim of moving people back into sustainable jobs.In this case, participation in the programme isvoluntary and eligibility is after six months ofunemployment. It includes the usual customisedsupport package, and in this case one of theoptions is an in-work training grant. Earlyresearch on the programme suggested that mostparticipants were male, but there were reason-able signs of success with 57 % of the sample inwork, the majority of who were towards theyounger end of the age group, had shorter spellsof unemployment, and with a higher representa-tion of women (Employment Service, 2001).


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 249

4.5.2.5. AustriaA distinctly quantitative approach was taken inthe EES review for Austria. This made use of theconsiderable database available to the researcherthat contains information on the workforce andemployers, as well as those unemployed partici-pating in active measures. Using some of thisinformation, the report concentrated on a micro-analysis of programme participants andnon-participants, adjusting the gross findings bythe application of econometric techniques toarrive at estimates of net impact. Individual find-ings were also used as a basis for estimating themacroeconomic effects of policies.

The impact of participation in trainingprogrammes was found to be positive in the macrosense, with the upgrading of the skills of the unem-ployed responding directly to employer needs. Thisleads to the conclusion that the programmes hadincreased aggregate employment by over 9 800each year which worked to a contribution of 0.32 %to GDP growth. Over 74 000 of the unemployed hadbenefited from a public employment service quali-fication enhancement programme and participantshad managed to increase their earnings potential bysome EUR 2 870 per annum, compared to thosewith similar characteristics who did not participate.This was also expected to have added to GDPgrowth.

4.5.2.6. FinlandThe Finnish evaluation of the impact of the EES(Finnish Ministry of Labour, 2002) is a particularlythorough appraisal of all four pillars. Persistentlyhigh unemployment has meant a significant

number of those affected have participated inactive measures which, at their peak in 1996,affected around 5 % of the labour force. The eval-uation of these active measures essentially usesadministrative data combined with sources suchas the LFS and other surveys, with various manip-ulations (including quasi-experimental techniques)carried out by independent researchers.

Different studies tended to show slightlydifferent results on the effects of trainingmeasures, but the overall employment effectswere only modest for the individuals concerned.Longer-term job impact was assessed from theextensive panel data available to the researchersusing matched control groups (to minimise selec-tion bias). However, labour-market training (aswell as subsidised jobs in the public sector andcertain other measures) had a weak net impacton the employment probability of participants,though the outcomes were slightly better forthose with lower than average employability atthe start of the programme.

The Finnish study draws some interestingconclusions. It suggests that interview-basedsurveys of programme participants show consid-erable social benefits to them, though these canbe short-lived if they do not eventually find a job.Also, there needs to be a distinction betweenlabour-market training designed for immediateemployment and that for developing long-termcompetence. In the former case, the argument isthat this type of training should be responsive tothe economic cycle, being reduced when thereare fewer jobs available. In contrast, training thataims to increase skills such as language or IT, or


Work-based learning for adults (WBLA) has been an important plank of active policy in the UK sincethe late 1990s. Its primary aim is to assist unemployed adults (within the age range 25-63) intosustainable jobs through participation in training programmes that are responsive to locallabour-market needs. Participants must have been unemployed for at least six months, though thereis a long list of special groups that can be considered before this threshold, such as people withdisabilities, those needing help with basic skills, lone parents, ex-offenders, etc.The programme is delivered by the Employment Service through its local offices. A key feature of theprogramme is its flexibility and customisation to the client’s needs. So the training can bejob-specific, working towards a national qualification, developing basic employability, work experi-ence or self-employment. Participants receive an allowance equivalent to their benefit entitlement,plus an additional GBP 10 per week and they are also eligible to claim assistance with expenses suchas childcare and travel to their training.

Box 7: Work-based learning for the unemployed in the UK

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 250

gain qualifications, should move inversely to theeconomic cycle. It is argued that introducing suchresponsiveness into this aspect of labour-marketpolicy should lead to greater programme effec-tiveness.

4.5.2.7. Other Member StatesSome of the other Member State EES reviewswere disappointing in their lack of specificcoverage of ALMPs and attention to detail. Forexample, the Greek review was principally a list ofcurrent and previous legislative instruments withno attempt at evaluation. However, one of themain conclusions of the report was that properinvestigation of the policies was required. TheItalian review is a little more detailed in terms of itsinvestigation into policies, but is rather too broadfor our use, failing to cover ALMPs in sufficientdetail. Nevertheless, it makes the point that theEES may have put undue emphasis on the role ofthe public employment service in solving thestructural problems of the Italian labour marketrather than seeing it as part of a wider network oforganisations, local and national, and policies,social and economic, that should be integrated.

4.6. Lessons to be learnt

The foregoing analysis of evaluation studies is, ofnecessity, selective. However, there is sufficientbreadth of coverage in terms of countries to allowsome useful issues to be raised.

Methodologically, the evaluations tend to fallinto the quasi-experimental sort, with an emphasison the econometric elaboration of programmeoutcomes. However, there is some evidence of theinnovative use of administrative and availablesurvey-based data (such as from LFSs) tounderpin these analyses, which is important. Thequality and validity of these evaluations will besignificantly dependent on the quality of the data.

It is interesting to note that the studies carriedout in the transitional countries, where it might bepostulated that ALMPs are less complicated in

their design and operation, in general tend toshow more positive impacts than, for example, inthe Member States. The literature does not fullyexplore the likely reasons for this, though onefactor is certain to be the generally more difficultlabour-market conditions found in the transitionalcountries (as proxied by higher levels of unem-ployment – particularly long-term unemployment).However, there are also plenty of similarities inthe evaluation findings, relating to aspects suchas which subgroups do best out of participationin training programmes.

This, to some extent, reinforces the problems incomparing programmes across countries. Exoge-nous circumstances will greatly affect the relativeeffectiveness of programmes. This is particularlyso for training/retraining measures, where acommitment can involve a substantial lead-timefor a forecast future skill need. Effectiveness will,therefore, depend on factors such as:(a) the accuracy and timeliness of labour-market

information;(b) the ability of employers to assess and

communicate their future needs;(c) the ability of training providers to meet the

skill needs articulated within an appropriatelearning context;

(d) changes in business and political circum-stances.

Endogenous variables that might influenceprogramme outcomes also need to be taken intoaccount and this depends on how much detailthe individual evaluations have on the structureand operation of the programmes. From the eval-uations accessed here, there was often a disap-pointing level of detail on the actual programmes.

This raises the issue of how the approach toALMPs is changing in many countries, with thegrowing emphasis on activation measures and thecoordination of active and passive measures. Itleads to a lack of clarity about the structure andobjectives of individual programmes and associ-ated difficulties in gathering reliable information onparticipation. It suggests the job of the evaluator inthe future will be that much more difficult.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 251

5.1. Introduction

In this final chapter of the report, we reflect on thepreceding analysis of the evaluations and attemptto extract the key messages for both the method-ology of evaluation and the design of training andretraining programmes within the context ofALMPs. This is inevitably a difficult task given thediversity of the studies examined, the contrasts inscope (and robustness) and their national back-grounds. Drawing out the common denominatorsis therefore tricky, but necessary if we are tomake a contribution to improve future evaluation.

In general, it can be said that evaluation has asomewhat mixed perception among the variousinterest groups. To those being evaluated, it canseem to be an intrusive activity that is demandingof time and resources, the end result of which isinevitably some criticism. To government agen-cies responsible for ALMPs, it is often referred toas an integral component of a programme, yetcan often have few resources devoted to it.Results can be ignored or often dismissed on thebasis of their timing or design. For many here, itmay be a process of accountability more thanpolicy improvement. To the evaluators, there isnormally some intrinsic attraction in the work,offering the opportunity to experiment with socialdata. Unfortunately, this can lead to the results ofsome evaluations being too impenetrable forthose who really have the power to change theway a programme is structured. There is also thedual problem of policy-makers wanting quickresults from evaluations that might be bestconducted over medium to long periods.

The overwhelming need is to balance thesepotentially conflicting perceptions to ensure thatpolicy evaluation is both integral and useful, withthe benefits to current and future programmeparticipants being the primary focus.

Here, there seem to be growing tensionsbetween accountability-led evaluations - justi-fying the value secured from public funds - andthe continuous improvement model. The latter ismore in accord with the emphasis in some coun-tries for evidence-led public policy developments,

but is not apparently well served by the sorts ofevaluation shown in this review. In short, evalua-tion frameworks and/or research designs seem tohave a narrowness of focus (e.g. accountabilityevidence) built into them. This does notencourage the sort of breadth of scope and anal-ysis needed to provide the quality of feedback onVET impacts with ALMP needed topolicy-makers, national or transnational.

5.2. Methodologies of evaluation

The analysis has shown quite clearly that trainingprogrammes are subjected to both monitoringand evaluation, both of which have their place.However, there is perhaps too much attention inthe literature to the differences between the two,with the basic point that only experimental, or themore likely quasi-experimental, approaches arelikely to have the necessary rigour.

This is disappointing, since there is a strongcase to make (for understanding VET effects inparticular) that any single approach to evaluationis limiting. Some of the best studies are thosethat take a wider methodological perspective,certainly using experimental approaches wherefeasible, but complementing this with the use ofadministrative data and more qualitative informa-tion on, for example, processes and the percep-tions of programme participants. It suggests thatevaluation hitherto has been more academicallydriven rather than policy driven, which is not somuch a criticism of the former as a lack of properattention in the latter.

On a number of occasions, this report hasalluded to the lack of an evaluation culture insome countries to the extent that it is notconducted effectively. This is certainly the case insome Member States, where we could find littleevidence of useful programme evaluation. Alter-natively, there are other cases where ALMP eval-uation has been carried out in countries under theauspices of international agencies such as theWorld Bank, and this is probably the case with

5. Summary and conclusions

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 252

most of the transitional countries with evaluationsto show.

In our analysis, it was disappointing to notethat there was little evidence of what might becalled a holistic approach to programme evalua-tion. Here we are not so much referring to a fullyblown cost-benefit analysis (though there may bea strong case for this approach) but more to awider evaluation that attempts to measure someof the effects other than employment probabilityor earnings. Additional benefits that might accrueto the programme participant include, forexample, measurable things such as hours ofwork (and other conditions such as holiday enti-tlements, etc.), non-pecuniary benefits, and thecomponents of earnings (such as bonuspayments, etc.). Similarly, there are the lesstangible benefits that might accrue such asincreased self-worth, better health, wider socialand behavioural gains, etc. These are more diffi-cult to measure and compare (with non-partici-pants), and are explored in greater detail in thecontribution by Green et al. (2003) in this thirdresearch report.

However, as with full cost-benefit appraisals,these wider evaluations are often too easilydismissed as impossible to measure, and so thechallenge remains largely untouched. This seemsa particularly important limitation in under-standing VET impacts. Furthermore, there is littleevidence of empirical support to say whetherthese assumptions on measurement are valid.

Of course, one of the major problemspreventing more extensive evaluation of ALMPs isthe volatile environment in which they arecurrently operating in many (though not all) coun-tries. As the mix between active and passivemeasures becomes blurred, the focus of evalua-tion is also likely to suffer. In any case, some ofthe existing studies have drawn attention to thefact that evaluations are often only focused onpilot or demonstration projects, to the extent thatthey are never given the opportunity for extensive(wider and longer-term) application. This probablyarises from the perception by some that evalua-tion has this limited role, whereas the need is forevaluation to become in-built to programmesfrom their outset so that the requisite informationis gathered in an ongoing process.

Information is a key to effective evaluation andsome of the foregoing studies have demonstrated

that there is potential in accessing useful datafrom regular sources. Supplementary questionsto a regular LFS (e.g. in the form of repeatedcross-sections) offer a particularly effective (andrelatively cheap) way of gathering relevant data,yet it is surprising how little this is done (notwith-standing the inherent methodological problems).Special surveys are also an option, but to bemost cost-effective, the best examples incorpo-rate attitudinal information alongside the factual,opportunities that are often missed by evaluatorsor those specifying evaluation frameworks.

5.3. Programme structures

It is difficult to extract the best elements from thedifferent training programmes covered by theevaluations referred to in this paper. Part of this isdue to the lack of detail given on theprogrammes, though even with sufficient detail itis not always appropriate to transfer models.Local circumstances, including the prevailinglabour-market and administrative conditions, limitthe degree of transferability and we are really leftwith identifying common denominators that hintat best practice.

However, one important and clear issue foreffective training relates to the quality of trainingprovided under the active measures. Indicationsfrom the evaluations are that this is not always ashigh as it should be, to the extent that it compro-mises the individual outcomes. Most studies thathave tackled this issue point to the need forprogrammes to have labour-market relevance.This might best be achieved in two ways.

First, good forward-looking labour-marketinformation can provide strong indications ofwhat the characteristics of the demand for labourare likely to be, and this can help inform provision(and individual choice). However, the inherentdifficulties in predicting future needs, coupledwith the lead times involved in training, mean thatit is never going to be a precise activity. At best,such information can be indicative of futuredemand and can lay the foundations for aninformed choice.

Second, and complementary to the first point,the involvement of employers (or the social partnerswhere this is feasible) in the design and delivery oftraining programmes provides a more direct way of


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 253

ensuring that the skills acquired by programmeparticipants are likely to meet some real need. Thevery few evaluation studies that have examinedprogrammes where there is employer involvementhave been clear in their conclusions that it leads toa better quality programme (whatever the measureof success). However, it is true to say that theinvolvement of all parties (or as many as possible)will raise the profile and ultimately the validity ofprogramme evaluations.

However, this positive outcome is not simplyconfined to the particular employers involved inthe programme. The indications are thatprogrammes with strong employer involvementcan lead to skills that have a certain currency inthe labour market, thus enhancing participants’longer-term employment prospects measured interms of job progression as well as employmentsustainability. This is further enhanced when theprogramme training leads to a recognisednational qualification, though this tends to be onlypossible for the longer periods of training (usuallyin excess of six months).

It is difficult to discern any clear, overridingeffects across the various studies on any partic-ular subgroup of the labour market, with perhapsthe exception of adult women, who featureconsistently in those gaining most from participa-tion in training programmes. Few of the evalua-tions go into great detail about the reasons whywomen should do better, but it must be influ-enced by the nature of labour demand that inmost countries has seen a disproportionategrowth in service sector jobs, many of themoffering fractional time contracts. Here, there maybe tensions between success in terms of joboutcome and quality of employment.

However, it is by no means obvious that thoseALMPs conducted in a more favourable labourclimate (i.e. low or moderate unemployment,improving economic cycle, etc.) will be morelikely to show stronger effects related to traininginterventions. The evidence of evaluations fromthe transitional countries clearly demonstratesthat participation in training programmesproduced significant positive employment effectsin the presence of (and maybe because of) higherunemployment.

Nevertheless, the evaluations generally agreethat the best programmes are those that aresmall in scale and highly targeted, and this ties in

with some key developments in the approach toALMPs, most notably the trend for offeringcustomised support to the unemployed. Typically,this will offer a menu of active measure options,one of which is normally training of various sorts.However, offering choice is not without its prob-lems, notably the increased costs of meeting indi-vidual needs (measured in terms, for example, ofthe advice and counselling given and the actualcost of training provision) and the greater diffi-culty in organising the provision. The key in allthis is balancing labour-market needs with indi-vidual expectations, which is again an area whereevaluations can display programme weaknesses.

There is also an issue around programmedesign reducing participants’ sense of risk. Forexample, those training programmes with jobguarantees seemed to show more positiveoutcomes, and perhaps this engenders greaterparticipant engagement (motivation and commit-ment) in the learning processes involved. If so,this also raises issues of how such interventionsneed to allow for the socialisation of skills devel-opment and learning in programmes, questioningthe effectiveness of some intensive programmesand suggesting that shorter rather than longerprogrammes, or those with follow up/after care,will produce stronger positive impacts.

5.4. Forward perspectives

In summarising how to move forward with theevaluation of training active measures, weconsider the options from two different perspec-tives as follows:(a) developments in methods of evaluation;(b) emphases in future research.

The key issues as we see them aresummarised below and presented as a stimulusto further debate outside the confines of thispaper.

5.4.1. Developments in methods of evaluation

Some of the common issues arising appear to be:(a) covering the correct programme, though not

too tightly focused to restrict consideration ofknock-on effects or wider benefits;

(b) the gap between completing the programmeand concluding the evaluation needs to besufficiently long for the labour-market effect


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 254

to have worked and for gains other thanshort-term ones to be observed;

(c) the need to capture or allow for how partici-pants decide on their participation in aprogramme (motivational factors);

(d) the evaluation has to be seen as indepen-dent, objective and credible to all relevantstakeholders, placing demands on evalua-tors/contracting bodies which go beyondrobustness;

(e) there needs to be evidence that evaluationfindings will be, and have been, followedthrough by policy-makers (and social partners– particularly employers) if the profile andperceived value of the work is to be increased;

(f) information on programmes needs to besystematically kept from the beginning, withthe needs of the evaluation in mind;

(g) those specifying evaluation frameworks needto be clear about the different design implica-tions of formative and summativeapproaches.

5.4.2. Further research

Some of the issues arising include the need for asearch on and around evaluation methods to lookat and provide:(a) more understanding on those conditions that

affect how evaluations are determined, suchas commissioning arrangements, manage-ment, design and dissemination, the compe-tences of evaluators and the elaboration ofand agreement on evaluation standards;

(b) how and to what extent evaluations are usedby policy-makers, programme managers,training providers, etc., and what the obsta-cles are to more effective use of them;

(c) greater analysis of the reasons why somelabour-market subgroups gain more fromtheir participation in training measures toinform how future programmes can be opti-mally structured;

(d) what the longer-term gains (e.g. over three tofive years) are from programme participationin terms of job and learning progression, aswell as earnings and related benefits;

(e) what scope there is for further work with thenewer approaches to evaluation, such asnatural experimentation/epidemiology and formore holistic studies.

This review has provided a valuable and timelyopportunity to assess the extent and nature ofthe evaluation of those ALMPs with atraining/retraining focus. It shows that, while eval-uations are taking place, it is neither a consistentnor uniform activity. We conclude that policy andpolicy-makers, looking to improve the quality,efficacy and cost-effectiveness of VET interven-tions in ALMPs, are not being very well served bycurrent evaluation approaches.

This raises serious questions about just howmuch we know concerning the effectiveness ofthese mechanisms, and calls for reform of currentpractices involving not just evaluations, butcrucially those specialising in evaluation frame-works and selecting/commissioning evaluations.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 255

ALMP Active labour-market policy

AMS National labour-market board

EEP Employability enhancement programme

EES European employment strategy

ESF European social fund

ILO International labour office

IT Information technologies

LFS Labour force survey

LMP Labour-market programmes

LMT Labour-market training

NEAP National employment action plan

NLMA National labour-market authority

OECD Organisation for economic cooperation and development

PEZs Prototype employment zones

SEUP Survey of employment and unemployment patterns


WBLA Work-based learning for adults


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 256

Barrett, A.; Whelan, C. T.; Sexton, J. J. ‘Employa-bility’ and its relevance for the management ofthe live register. Dublin: ESRI – Economic andSocial Research Institute, 2001 (Policy researchseries No 40).

Black, D. et al. Is the threat of training more effectivethan training itself? Experimental evidence fromthe UI system. University of Western Ontario,department of Economics, 1999 (Workingpaper No 9913).

Bloom, H. The national JTPA study: overview ofimpacts, benefits and costs of Title II-A.Canada: Abt. Associates, 1994.

Brodsky, M. M. Public-service employmentprogrammes in selected OECD countries.Monthly Labor Review. US Department ofLabor, 2000, Vol. 123, No 10, p. 31-41.

Bysshe, S. et al. The role of guidance in tacklinglong-term unemployment [Galtun project].Brussels: European Commission, 1998.

Calmfors L.; Forslund, A.; Hemström, M. Doesactive labour market policy work? Lessons fromthe Swedish experiences. Uppsala: IFAU – Institute for labour market policy evaluation,2002 (Working paper series, No 2002:4).

Carling, K.; Richardson, K. The relative efficiency oflabor market programs: Swedish experiencefrom the 1990s. Uppsala: IFAU – Institute forlabour market policy evaluation 2001 (Workingpaper series, No 2001:2).

Conniffe, D.; Gash, V.; O’Connell, P. J. Evaluatingstate programmes: natural experiments andpropensity scores. Economic and SocialReview. Dublin: Economic and Social Studies,2000, Vol. 31, No 4.

Dar, A.; Tzannatos, Z. Active labor marketprogrammes: a review of the evidence fromevaluations. Washington DC: The World Bank,1999 (Social protection discussion paperseries, No 9901).

De Koning, J. Training of the unemployed in theNetherlands: what do we know after more than50 evaluation studies? Rotterdam: NEI, 1998(Working paper, No 2).

Denny K.; Harmon, C.; O’Connell, P. J. Investing inpeople: the labour market impact of human

resource interventions funded under the1994-1999 community support framework inIreland. Dublin: ESRI – Economic and SocialResearch Institute, 2000 (Policy research seriesNo 38).

DEWRSB – Department of Employment, WorkplaceRelations and Small Business. Job network eval-uation. Stage two: progress report, evaluationand program performance. Canberra: Common-wealth of Australia, 2001 (Branch report No 2).

Employment Service. Evaluation of the New Deal 50Plus: research with individuals (Wave 1).Sheffield: NIESR – National Institute ofEconomic and Social Research, 2001(Research & Development Report, ESR91).

European Commission, DG V-Employment, Indus-trial Relations and Social Affairs. Conclusions ofthe ESF mid-term evaluations: overview.Luxembourg: Office for Official Publications ofthe European Communities, 1998.

European Commission, Directorate General forEmployment and Social Affairs. Conclusions ofthe ESF final evaluations. Luxembourg: Officefor Official Publications of the EuropeanCommunities, 2001.

Fay, R. G. Enhancing the effectiveness of activelabour market policies: evidence fromprogramme evaluations in OECD countries.Paris: OECD, 1996 (Labour market and socialpolicy occasional papers, No 18).

Fertig, M.; Schmidt, C. M. Discretionary measuresof active labor market policy: the Germanemployment promotion reform in perspective.Bonn: IZA, 2000 (Dicussion paper, No 18).

Finnish Ministry of Labour. Impacts of the EES –National evaluation of the effects of the Euro-pean employment strategy. Helsinki: Ministry ofLabour, 2002.

Forslund, A.; Krueger, A. B. An evaluation of theSwedish active labor market policy: new andreceived wisdom. In: Freeman, R. B.,Topel, R. H.; Swedenborg, B. (eds) The welfarestate in transition: reforming the Swedishmodel. Chicago: University of Chicago Press,1997 (NBER-Conference Report series, No 6).

Bibliography and references

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 257

Fretwell, D.; Benus, J.; O’Leary, C. J. Evaluating theimpact of active labor market programs: Resultsof cross-country studies in Europe and CentralAsia. Washington DC: The World Bank, 1999(Social protection discussion paper series,No 9915).

Friedlander, D.; Greenberg, D. H.; Robins, P. K. Eval-uating government training programs for theeconomically disadvantaged. Journal ofEconomic Literature. Nashville: AmericanEconomic Association, 1997, Vol. 35, No 4,p. 1809-1855.

Green, A.; Preston, J.; Malmberg, L. Immaterialbenefits of education, training and skills at amacro level (in this volume).

Grubb, W. N. Evaluating job training programs in theUnited States: evidence and explanations.Berkeley: NCRVE – National Center forResearch in Vocational Education, 1995 (Tech-nical assistance report, MDS-1047).

Grubb, W. N. Lessons from education and trainingfor youth: five precepts. Washington D.C.Conference, 23-24 February 1999 on‘Preparing youth for the 21st century: the tran-sition from education to the labour market.’Paris: OECD, 1999.

Harkman, A.; Jansson, F.; Tamás, A. Effects,defects and prospects – An evaluation of labourmarket training in Sweden. Solna: The NationalLabour Market Board, 1996 (Working PaperNo 5).

Haughton, G. et al. Prototype employment zones: aqualitative and contextual evaluation. Sheffield:DfEE – Department for Education and Employ-ment, 2000 (Research Report, No 232).

Heckman, J. J.; Lochner, L. Rethinking educationand training policy: understanding the sourcesof skill formation in a modern economy. In:Danziger, S.; Waldforgel, J. (eds) Securing thefuture: investing in children from birth to adult-hood. New York: Russel Sage Foundation,2000, p. 47-86 (Ford foundation series on assetbuilding, Vol. 1).

Human Resources Development Canada. Evalua-tion of the employability improvementprogramme. Ottawa: HRDC, 1995.

Hujer, R.; Caliendo, M.; Radic, D. Method and limi-tations of evaluation and impact research. In:Descy, P.; Tessaring, M. (eds) Evaluation andimpact of education and training: volume 1.Luxembourg: Office for Official Publications of

the European Communities, 2003. (CedefopReference series).

Hujer, R.; Caliendo, M.; Zeiss, C. Macroeconomicevaluation of active labour market policy - a casestudy (in this volume).

Johanson, R. Retraining adults in Germany.Geneva: ILO – International Labour Office, 1994(Training policy study, No 4).

Johansson, P.; Martinson, S. The effect of increasedemployer contacts within a labour markettraining program. Uppsala: IFAU – Institute forlabour market policy evaluation, 2000 (Workingpaper series, No 2000:10).

Kluve, J.; Lehmann, H.; Schmidt, C. M. Active labormarket policies in Poland: human capitalenhancement, stigmatization or benefitchurning? Bonn: IZA, 1999 (Discussion paperNo 30).

Lalive, R.; van Ours, J.; Zweimüller, J. The impact ofactive labor market programs and benefit enti-tlement rules on the duration of unemployment.Tilburg: CentER – Center for EconomicResearch, 2000 (Discussion Paper, No 41).

Larsson, L. Evaluation of Swedish youth labourmarket programmes. Uppsala: IFAU – Institutefor labour market policy evaluation, 2000(Working paper series, No 2000:1).

Martin, J. P. What works among active labourmarket policies: evidence from OECD coun-tries’ experiences. Economic Studies. Paris:OECD, 2000, Vol. 30, p. 79-113.

Martin, J. P., Grubb, D. What works and for whom:A review of OECD countries’ experiences withactive labour market policies.Uppsala: IFAU –Institute for labour market policy evaluation,2001 (Working paper series, No 2001:14).

Meager, N.; Evans, C. The evaluation of activelabour market measures for the long-termunemployed. Geneva: ILO – InternationalLabour Office,1998 (Employment and trainingpapers, No. 16).

Ministry of Labour. Effects of Danish employabilityenhancement programmes. Copenhagen:Ministry of Labour, 2000 (Working group paper).

Nicaise, I. Giving fish or teaching to fish? Acost-benefit analysis of Belgianemployment-training projects for minimumincome recipients. Public Finance and Manage-ment. Harrisburg: SPAEF.com – Southern PublicAdministration Education Foundation, 2000,Vol. 2, No 2.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 258

O’Connell, P. J.; McGinnity, F. Working schemes:active labour market policies in Ireland. Alder-shot: Ashgate, 1997.

OECD – Organisation for Economic Cooperationand Development. Employment outlook: 1993.Paris: OECD, 1993.



O’Leary, C. J. Evaluating the effectiveness of activelabor programs in Poland. Kalamazoo: UpjohnInstitute, 1998a (Technical report No. 98-012).

O’Leary, C. J. Evaluating the effectiveness of activelabor programs in Hungary. Kalamazoo: UpjohnInstitute, 1998b (Technical report No. 98-013).

OMAS. Review of the employment policies inGreece during the period 1994-2001. Athens:OMAS Ltd, 2001.

Orr, L. L. et al. Does training for the disadvantagedwork? Evidence from the National JTPA Study.Washington DC: The Urban Institute, 1996.

Park, N. et al. An assessment of the impact ofgovernment-sponsored training. CanadianJournal of Economics. Oxford: BlackwellPublishing, 1996, Vol.29, supplement1, p.93-98.

Payne, J. et al. Employment training and employ-ment action: an evaluation by the matchedcomparison method. Sheffield: Department forEducation and Employment, 1996. (ResearchSeries, No 74).

Pierre, G. A framework for active labour marketpolicy evaluation. Geneva: ILO – InternationalLabour Office, 1999 (Employment and trainingpapers, No 49).

Puhani, P. A.; Steiner, V. The effectiveness and effi-ciency of active labour market programmes inPoland. Empirica. 1997, Vol. 24, p. 209-231.

Raaum, O.; Torp, H.; Goldstein, H. Employmenteffects of labour market training in Norway.Oslo: Department of Economics, University ofOslo, 1995 (Memorandum No 8).

Riley, R.;Young, G. New Deal for young people:implications for employment and the publicfinances. Sheffield: NIESR – National Institute ofEconomic and Social Research, 2000(Research & Development Report, ESR62).

Schmid, G.; O’Reilly, J.; Schömann, K. Internationalhandbook of labour market policy and evalua-tion. London: Edward Elgar, 1996.

Schwegler-Rohmeis, W. Review of active labourmarket policy in Kyrgyzstan. Berlin: ISA Consult,2000.

Sianesi, B. An evaluation of the active labour marketprogrammes in Sweden. Uppsala: IFAU – Insti-tute for labour market policy evaluation, 2001(Working paper series 2001:5).

Stanley, M; Katz, L; Krueger, A. Developing skills:what we know about the impacts of Americanunemployment and training programmes onemployment, earnings and educationaloutcomes. Cambridge, MA: Malcolm WeinerCenter for Social Policy, 1998 (Working paperseries, No H-98-2).

Stromback, T; Dockery, A. M.; Ying, W. Labourmarket programs and labour market outcomes.Perth: CLMR – Centre for labour marketresearch, 1998 (Discussion paper series,No 98/1).

Tamás, A., Harkman, A.; Jansson, F. The effect ofvocationally oriented employment training onincome and employment. Solna: Arbetsmark-nadsstyrelsen, 1995.

van den Berg, G. J.; van der Klaauw, B. Counsellingand monitoring of unemployed workers: theoryand evidence from a controlled social experi-ment. Free University of Amsterdam, 1995(Working paper).

van Ours, J. C. Do active labor market policies helpunemployed workers to find and keep regularjobs? Bonn: IZA, 2002 (Discussion paperNo 121).

Walsh, K.; Atkinson, J.; Barry, J. The new dealgateway: a labour market assessment. Brighton:The Institute for Employment Studies, 1999.

Walsh, K. et al. Evaluation of active labour marketpolicies in Bulgaria. Washington DC: The WorldBank, 2000.

World Bank. Monitoring and evaluation: some tools,methods and approaches. Washington DC: TheWorld Bank, 2001.

Wouters, M.; Van Meensel, R.; Nicaise, I. DeTOK-projecten en hun cursisten, drie jaar later.Follow-up onderzoek van de projecten van1989. Leuven: HIVA – Hoger Instituut voor deArbeid, 1994.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 259

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 260

The impact of human capital andhuman capital investments on company performance

Evidence from literature andEuropean survey results

Bo Hansson, Ulf Johanson, Karl-Heinz Leitner

AbstractThis study consists of a literature review and an analysis of an existing database on human resourcemanagement (HRM) (the Cranet survey). It focuses on research that connects human capital with the firmand asks whether education, skills/competence and training have any impact on company performance.The main results may be summarised as follows.It appears that training provided by firms for employees is not characterised by being general or specificbut by what is needed to stay ahead of competitors. There is a growing body of literature suggesting thatfirms are financing both types of training.More recent research also suggests that investments in training generate substantial gains for firms evenif employees can use this training in other firms. The evidence that employers profit from training invest-ments comes from different countries including Britain, France, Ireland, the Netherlands, Sweden, andthe US. Most of these studies indicate that training affects performance and not the other way around.The effects of education and skills/competence on aspects such as productivity and innovations aregenerally found to be positive and significant, though the connection with profitability might be lessexpected. Firms can also extract profit from prior education as they do from general training investments.Supporting employee development through training policies and methods for analysing training needs isimportant to explain the provision of training and training outcomes. Similarly, innovative (and compre-hensive) HRM practices tend to be associated with positive company performance.Innovation and information technology (IT) both result in greater investment in training and also dependon education and skills in generating profits. Other findings suggest that training and comprehensiveHRM practices are closely related to firms’ innovative capacity.The lack of studies connecting small and medium enterprises (SMEs), labour market conditions(systems), and social partners with company training policies and performance measures such asproductivity or profitability, makes it difficult to draw conclusions on these aspects. This suggests anincentive to research such matters more thoroughly in the future.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 261

1. Introduction 264

2. Method and research problem 265

2.1. Inference and causality problems 265

3. Overview of research and findings 267

3.1. Labour economics 267

3.1.1. Recent advances on the effect of training for firms 268

3.2. HRM and HPWS literature 274

3.3. National European surveys and cross-national comparisons 279

3.4. Small and medium enterprise (SME) surveys 282

3.5. Other training and impact studies 283

4. Cranet survey results 287

4.1. Selected Cranet survey questions 287

4.2. Cranet survey results 288

4.2.1. The scope of training 288

4.2.2. Correlation between variables 290

4.2.3. What determines training? 291

4.2.4. What factors are indicative for top (10 %) performers? 293

4.2.5. Main findings 295

5. Results 297

5.1. The effects of training on company performance 297

5.1.1. The impact of training 297

5.1.2. Formal/informal and general/specific training 298

5.1.3. Timing of the effects of training 298

5.1.4. Timing of training investments 298

5.2. The effects of education, skills and competences on company performance 299

5.3. How firms profit from general human capital 300

5.4. Training and HRM practices 301

5.5. Innovation, technological change and training 302

5.6. Specifics of SMEs 303

5.7. The influence of labour markets and social partners 304

6. Summary and discussion 306

6.1. Implication of firm financed general human capital investments 306

6.2. Information on training and intangibles to capital markets 307

6.3. Strength and weakness of data and methods 309

6.4. Policy implications and future research 310

List of selected abbreviations 312

Annex – Selected Cranet questions 313

References 315

Table of contents

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 262

Tables

Table 1: Labour economics 272

Table 2: HRM (HPWS) studies 276

Table 3: National and cross national studies 280

Table 4: SMEs and other training studies 284

Table 5: Cranet survey, descriptive statistics 289

Table 6: Correlations between main explanatory variables (private sector) 290

Table 7: Training regression (private sector) 293

Table 8: Mean difference between top 10 % and lower half of firms in sector (private sector) 295

List of tables

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 263

Human capital is a major factor in generatingfuture growth and prosperity. Human capitalinvestments such as education and training are,therefore, a main concern for individuals, firmsand governments. According to Becker (1993),human capital is the key determinant inexplaining the rise and fall of nations as well as amain factor in determining individual income. Theimpact of human capital on enterprises is lessclear. This is because the attributes of humancapital and human capital investments areascribed to the individual and not the firm. Thisstudy concerns research that connects humancapital with the firm. The question is whethereducation, skills/competence and training haveany impact on company performance. The focusis on continuous vocational training that takesplace inside companies and is paid for, in part orwhole, by the employer.

A considerable volume of employer sponsoredor company training takes place each year. TheEuropean Union continuing vocational trainingsurvey 1994 suggests that more than half of allfirms with 10 or more employees provided sometraining during 1993 and that about 1.6 % oflabour costs are spent on training (EuropeanCommission, 1999). More recent figures fromSweden suggest that company training plays anincreasingly important role in creating new knowl-edge and skills in society. Working time spent oncompany training in Sweden has increasedconsiderably in recent years, from about 2.5 % in1999 to roughly 3.5 % in 2001. A Norwegianstudy estimates the time spent on formal andinformal training as high as 4-6 % of working time(Hagen et al., 2001).

Although these investments amount to consid-erable sums, until recently little has been knownabout the return for firms (1). The impact ofeducation and training on company performanceis an important issue not only because of thelarge amount invested each year in knowledgeand skills, but also because it is pertinent toknow who benefits from these investments. Thelatter question has a bearing on who should carrythe costs of training investment, to what extentwe have under-investment in training, whetherthere is a need for policies to improve the currentsituation in regard to company training, etc.

The aim of this study is to provide an overviewof research that connects education, training orskills/competence with the impact of these activ-ities on productivity, profitability or other variablesof firm performance. Besides reviewing associ-ated literature, the study also involves an analysisof an existing database (Cranet survey) withregard to education and training.

The remainder of the paper is organised asfollows. The next chapter introduces the methodused in gathering studies for this review andprovides a short introduction to some statisticalproblems encountered in this line of research.Chapter 3 gives an overview of findings indifferent research disciplines. Chapter 4 takes acloser look at a European human resourcemanagement (HRM) survey (Cranet survey) in rela-tion to employee development issues. Chapter 5gives the combined findings of this paper andpresents the major results in regard to the impactof training on company performance. Chapter 6suggests some policy and future research impli-cations based on the main findings of this study.

1. Introduction

(1) An estimate by SCB (Statistics Sweden) considers as much as 3.8 % of GDP was spent on company training in 2001 whichis roughly SEK 80 billion or in the order of EUR 9 billion. The amount spent on company training is close to what is spent oncompulsory and secondary school in Sweden 2001 (SEK 88 billion). Source: The National Agency for Education. The amountspent on formal training in the US was close to USD 59 billion in 1997 (Bartel, 2000).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 264

Literature on the impact of education, training, andskills/competence on firm performance has beenreviewed from several different sources, includingpublished material. We have scanned publishedliterature mainly through channels such asABI/Inform and other university library databases. Incollecting the most recent research papers we havealso surveyed different web based databases, thebest known being the social science researchnetwork (SSRN). This is a major source of informa-tion with over 3 million downloaded working paperssince the start. Another important source is theIDEAS (UQÀM – Université du Québec à Montréal)database with over 100000 working papers and arti-cles from 1 000 universities and research centresaround the world. It includes papers from IZA (Insti-tute for the study of labor), NBER (National bureau ofeconomic research), CESifo (Center for economicstudies and institute for economic research), andseveral other important institutions. Other sourcesof information include the Cedefop bibliography onimpact research. We have also collected researchpapers directly from different universities andresearch centres. These efforts have been made toprovide the latest working papers on the matter.

The focus of the survey lies on more recentstudies and on European based research. Whilethis is an area that clearly needs much additionalresearch, we have included all papers that wehave come across that connect education, skills,and training with measures of company perfor-mance. In addition to labour economics literature,we also include findings in areas such as HRM,high performance work systems (HPWS), innova-tion studies, accounting and finance basedstudies, SME based research, as well as otherreviews and meta-analyses. The focus in thereview has been on quantitative research from theeconomic perspective of training. For instance,we have not surveyed literature on psychologicalaspects of training nor with a more ad hocapproach to training issues. A prerequisite ofinclusion is that the paper uses a statisticalapproach and is concerned with economicaspects of human capital and human capitalinvestments. We have also made an effort to find

research published in other languages, thoughmost research papers included in this review arein English.

The review of the literature is divided intodifferent sections. Labour economics forms amain segment since human capital and humancapital investment has been a major researchquestion for a considerable time. The contributionof HRM literature in regard to education andtraining is also given a section of its own. In thissection, the literature on HPWS plays an impor-tant role because of the ability to connect HRMpractices with company performance measures.We also decided to give national, cross-national,and SME based research their own sections.Since the focus is on empirical findings, theoret-ical considerations are (briefly) dealt with inconnection with empirical findings or in relation totheir respective research area.

A brief description of previous work in eachresearch area is provided, followed by in-depthcoverage of more recent papers with an analysisof methods used, data, and results. The tablesthat summarise each section include only themore recent papers and those that we regard ascontributing substantially to the existing literature.

A short description of statistical problemstypical of this type of impact research follows toprovide an understanding of the problems dealtwith in the review.

2.1. Inference and causalityproblems

Heterogeneity problem poses statistical difficultiesin estimating the impact of training on any outcomevariable. Differences between those who receiveand those who do not receive training make it diffi-cult to maintain that training alone causes the entireeffect on a dependent variable. In laboureconomics, heterogeneity among workers is typi-cally controlled for by including proxies for differ-ences in human capital accumulation and othervariables (age, tenure, education, occupation, etc.).

2. Method and research problem

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 265

In addition to factors that we normally can controlfor in empirical work, we have a number of factorsfor which it is hard to attain estimates (unobservablefactors). The main concern in statistical analysis isunobservable factors that are correlated with theregressors of the estimated equation. There areseveral remedies for this problem. One is to findsome broad approximation for the unobservablefactor, as in firm-level data which typically includesindustry dummies to control for differences inproductivity, profitability, market valuation, etc.,between firms in different industries.

If the effect on the uncontrolled (unobserved)variable is considered fixed over time, it is normalto use changes in the variable, instead of theoriginal, level data. This procedure needs dataover time (panel data). An example of afixed-effect problem that can be solved by usingchanges in the variables (first difference) is that ofunobserved ability among individuals. Forinstance, if more capable individuals are morelikely to receive training, the return on pasttraining will be upward biased, measuring, in part,ability instead of human capital investments. Aslong as the effects of the unobserved variable (inthis case ability) are stable over time (timeinvariant) taking the first-difference (the change inthe variables) will mitigate the problem.

Another problem is that not all explanatoryvariables in an equation can be considered tohave a one-way relationship with the dependentvariable. Explanatory factors that also are deter-mined by the dependent variable (the variable wetry to explain) are called endogenous and pose aproblem in estimating the returns on training. Thefollowing example, given in Dearden et al. (2000),details the problem with mutually dependent vari-ables (endogenous variables) when trying toassess the impact of training on aggregatedindustry productivity.

‘Transitory shocks could raise productivity andinduce changes in training activity (and of courseother inputs, labour and capital). For example,faced with a downturn in demand in its industry,a firm may reallocate idle labour to training activ-ities (the pit stop theory). This would then meanthat we underestimate the productivity effects oftraining because human capital accumulation willbe high when demand and production is low. Iffirms train when production and demand is highthen the opposite applies.’ (p. 25).

The remedy here is usually to estimate theequation in a system that considers the two-wayrelationship between the dependent and explana-tory variables. Typically this procedure includes asearch for instrumental variable(s) that are corre-lated with the explanatory variable but not withthe dependent variable. Another way to mitigatethe problem is to include lagged variables of theendogenous variable, as the lagged variable canpartly alleviate the problem of simultaneity.

Few studies have been able to explore the effectof mutual dependence between training and produc-tivity or profitability. While this is a potential problemin most impact research, the results of Dearden et al.(2000) are particularly interesting, assessing theimpact of training on productivity from different esti-mation procedures. In their study, the impact fromincreasing the proportion of workers trained by 5 %would result in a 31 % increase in productivity usingonly the raw correlation between training andproductivity. ‘We account for an overwhelmingproportion of this correlation, however, by our controlvariables. The 31 % effect in model A (no controls)falls to 8.5 % in model B (some controls) and 2.6 %in model C (include controls for occupation). Dealingwith endogeneity through general method ofmoments (GMM) (model D) increases the effect to4.1 %.’ The results in Dearden et al. (2000) suggestthat a well-specified regression model with adequatecontrols works quite well even in the presence ofsimultaneity problems. Given the data set used intheir study, the main issue in a well specified regres-sion model is not whether there is an overestimationof the impact of training but to what extent the modelunderestimates the impact (due to the assumptionthat training is determined exogenously).

It is important that studies address the ques-tion of heterogeneity by including adequatecontrol variables in statistical models. Usingchanges in variables instead of level datanormally gives a better base for conclusion.Lagging the effect of the impact of training furtherstrengthens the basis for cause and effect rela-tionships. The results of Dearden et al. (2000)suggest that the problem of endogeneity intraining might be of a lesser concern, at least in awell-specified regression model. In our review ofliterature we have made an attempt to addressthese issues by examining the regression modelsused and by examining the variables included inthe estimates.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 266

3.1. Labour economics

There has been continuing debate in laboureconomics literature on the subject of whetherfirms can profit from training investments. BeforeBecker’s (1962) theory on company training, mosteconomists saw education and training as theinvestment decisions of individuals. From acompany perspective, investments in humancapital (on-the-job training) differ from invest-ments in other assets because the employee hasan option to leave the firm, engage in wagebargaining and, in other ways, influence theoutcome of the investment decision. Becker(1962) advanced a theory on investment inhuman capital, explaining levels of investmentand predicting who should pay for, and who willbenefit from, the completed training.

Becker divided on-the-job training into generaland specific. General training is useful not only tothe firm providing the training but to other firmsas well. Because of this, employers are lessinclined to invest in this type of training. In acompetitive labour market, general training wouldlead to a wage increase for the employee andwould offset the profit for the firm providing thetraining. In other words, general trainingincreases the market value of the employee,suggesting that the employee should pay for thistype of training, for example, by receiving wagesbelow his or her productivity. Specific training, onthe other hand, does not benefit other firms and,subsequently, the trainee’s market value is notaffected. Because specific training does not influ-ence wages, the employee is not willing to payfor it. The firm pays for specific, on-the-jobtraining and increased productivity is accrued bythe firm providing the training. The employer mayshare some of the increased productivity with the

employee to prevent the trainee from leaving thefirm before the specific training investment isrecouped (2).

The main idea of Becker’s theory is that theparty that is most likely to benefit from the invest-ment also pays for it. The basic reasoning thatemployers are unable to benefit from generalhuman capital such as schooling, apprenticeshipprogrammes, and company training that are alsouseful to other firms, might be mitigated by thefunction of different labour markets or the degreeof competition for the skills in the market (3).

Theoretically, specific training poses noproblem for firms as these investments are nottransferable to other firms. However, most of thetraining provided by companies is to be consid-ered general in nature. About 60-70 % of allcompany training is classified as general training(e.g. Barron et al., 1999; Loewenstein andSpletzer, 1999). The study by Loewenstein andSpletzer (1999) also indicates that the generalityof training increases with more complex jobs,which suggests that most of the trainingcompleted in human capital-intensive firms isuseful to other companies. While research indi-cates that most company training has a value toother employers, the question is who is actuallypaying for this type of training?

Because most training has a value to otheremployers, theory predicts that the individualshould pay directly (by bearing the full costs) orindirectly by accepting a wage below his or herproductivity. So even if firms pay for all explicitcosts such as trainers, course fees, allowances,the individual still has the potential to pay throughwages below productivity. Testing the theorydirectly requires not only data about wages butalso, more importantly, information about produc-tivity.

3. Overview of research and findings

(2) If one introduces turnover into the equation, this will result in joint investments in firm specific human capital. This is becausethe higher wage for employees receiving specific training leads to an excess supply of workers willing to be trained. To bringsupply more in line with demand some of the costs for specific training are shifted onto the workers.

(3) The division into general and specific training may be too rigid. Company training might better be viewed as training withdiffering degrees of generality (marketability), from benefiting only the current employer, to benefiting competitors, industriesand companies in general. Company training that has varying potential to suit other employers is of interest when consideringthe ability of employers to benefit from general (marketable) training investments.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 267

The absence of measurement of individualproductivity guides labour economics studies tousing data on wages to examine the question ofpayment for company training. Asking who paysfor company training also implies an answer tothe question of who will benefit from the training,i.e., if firms pay for general company training italso suggests that firms are able to capture thereturns from such investment. By this reasoning,a first step in establishing whether training hasany impact on performance is to establishwhether firms fund such investments.

Empirical studies that focus on wage profilesappear to confirm the general human capitalprediction (Neumark and Taubman, 1995;Reilly, 1995), as well as the specific humancapital prediction (Topel, 1991). In addition to thedivision into general and specific human capital,Neal (1995) suggests that an industry-specificfactor constitutes an important component of thehuman capital stock. Neal (1995) investigateddisplaced workers and found that wages partlyreflect compensation for industry-specific skills.

Observations made on wage profiles seem tosupport predictions of human capital theory. Theproblem with inferences from wage profiles is that anumber of other theories and explanations alsopredict an upward sloping wage curve. Forinstance, wage growth is produced in job-matchingmodels because of imperfect information about theemployee’s productivity (Jovanovic, 1979).Self-selection models use back-loaded compensa-tion to discourage ‘movers’ from applying for jobs(Salop J. and Salop S., 1976). Implicit contractingmodels explain the firm’s future wage commitment(rigidity) as a consequence of an income insuranceagreement between the employer and theemployee (e.g. Azariadis and Stiglitz, 1983;Marcus, 1984). The forced savings explanationjustifies an upward sloping wage curve by workers’preferences (Loewenstein and Sicherman, 1991).

Apart from the problem of alternative theories ofwage growth, empirical studies that used a moredirect test by utilising training data have failed tosupport the predictions of human capital theory.Indications that firms invest in general training are

sometimes indirectly revealed in studies on theimpact of training on wages. For instance Lenger-mann (1996) found that recipients of what appearsto be general company training benefited fromincreased earnings during the training period. Veum(1995), after studying more recent data from thenational longitudinal survey of youth (NLSY), cameto the conclusion that firms pay for general training.Loewenstein and Spletzer (1998) conclude thatemployers pay for general training and contend thatfirms are able to obtain some return from generaltraining investments (4). More recent studies(Lowenstein and Spletzer, 1999; Barron et al., 1999)argue convincingly that firms pay for generaltraining and that firms are also able to benefit fromthese investments. Other studies that also suggestthat employers pay for general training includeAcemoglu and Pischke (1998; 1999a) on appren-ticeship programmes and Autor (2001) on tempo-rary help firms. Literature connecting training withwage effects indicates that firms pays for generaltraining. This is an important finding because itsuggests that firms benefit from all types of traininginvestments (specific as well as general training).

Whether this is the case is the main question ofthis review of literature. A prerequisite for a directtest of Becker’s theory is performance data suchas productivity or profitability, though dataconnecting training with productivity or profitabilityis hard to come by. The absence of company datais striking and few studies have had access toperformance data until very recently. These studieswill be examined in greater detail below.

3.1.1. Recent advances on the effect of

training for firms

In a review of the literature on effects of companytraining for employers, Bartel (2000) concludesthat econometric analysis of a large sample offirms provides little guidance on the question ofthe employer’s rate of return on training. Thereason given by the author is that few data setsinclude the cost of training, that few studies havebeen able to control for heterogeneity amongfirms or addressed the question regarding theendogeneity of training.


(4) In the shared investment model of Loewenstein and Spletzer (1998), the employer shares general training investments with theemployee as a consequence of the employer’s inability to commit credibly to future wages. The employer, instead, commits toa minimum guaranteed wage and shares the investment in general training and realises the returns if the minimum wage guar-antee is binding. The wage floor is only binding for less productive workers. If the wage guarantee is not enforced, the workerthen realises all returns and incurs all the costs of the training.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 268

One reason for this lack of research results is thedifficulty in estimating the amount invested intraining. The definition of what to include in estimatesof the time spent on training is unclear (e.g.informal/formal training). Similarly, the costs to beincluded in calculating training investments (e.g.direct/indirect costs) are not standardised. The lackof a coherent definition of training and of a standard-ised way of measuring training investments hampersefforts to address the question of the actual benefitof training. In many cases training is measured as theproportion of employees being trained in a yearinstead of the actual amount invested.

The review by Bartel (2000) is concerned withthe return on investment (ROI) in training. Thepresent overview of training literature is focusedon whether there is an impact on performanceand is less concerned with estimating the rate ofreturn on these investments. However, the issueof the return on training investments is still verymuch an open question as few of the studiescited include the cost of training.

European research has made some importantadvances by incorporating measures of produc-tivity in the statistical models. The major studiesare summarised in Table 1. The results of an Irishstudy by Barrett and O’Connell (1999) suggest thatthe amount of general training has a significantpositive impact on productivity, in contrast withspecific company training. These results comefrom a first-difference approach that cancels outtime-invariant effects. Barrett and O’Connell arguethat a plausible explanation why only generaltraining is significant is that general trainingprovides a greater incentive for employees tospend more effort in the learning process.

It is worth noting that these results are realisedin a model with controls for changes in corporateinnovations and the introduction of newpersonnel policies. These two control variablesshow no significant relationship to changes inproductivity. Because few HRM studies haveused the change in HRM policies, this study alsocontributes to discussion of whether personnelpractices or human capital investments are themain factor in generating the effects on companyperformance. Other interesting findings in thestudy by Barrett et al. (1998) include:(a) that training and the change in productivity is

significant whereas training and the level ofproductivity is not;

(b) that the correlation between training andtangible investments is relatively low (Rs 0.13)which suggests that tangible investments canonly explain a small part of the impact oftraining in a first-difference model;

(c) an interaction term between tangible invest-ments and general training renders thetangible investment coefficient insignificantwhile the general training variable remainssignificant.

The combined result of the study by Barrett andO’Connell suggests that training has a major influ-ence on productivity and that other plausible andnormally uncontrolled factors have little or no influ-ence on productivity effects ascribed to training.

A study by Dearden et al. (2000) suggests thatcompany-sponsored training generates substan-tial gains for employers in terms of increasedproductivity. Different methods are used tocontrol for unobserved heterogeneity and poten-tial endogeneity of training, including GMMsystem estimation. Their estimates consistentlyshow that the impact of training on productivity isabout twice as large as the impact on wages.Their results also suggest that formal training hasa larger impact on productivity than informaltraining. These results are obtained by examiningthe direct impact of training on industrial produc-tion. They also argue that treating training asexogenous leads to an underestimation of thereturns on training for employers. This is animportant observation since few studies havecontrolled for the possibility of two-way relation-ships between training and company outcomevariables such as productivity and profitability.

The results from Groot (1999) suggest that thereis a rather weak connection between whocontributes to training investment and who benefitsfrom it. The study by Groot is based on telephoneinterviews with 479 Dutch firms. In about 43 % of allcases the workers either contributed through use oftheir leisure time and did not receive benefits or didnot contribute to the training but reaped some of thebenefits. Only 5 % of the workers contributed finan-cially but more than 75 % of the workers contributedwith leisure time. Groot concludes that the benefit ofenterprise-related training is high, both in terms ofproductivity and wage effects. Average productivitygrowth following training was found to be 16% whileaverage wage growth was 3.3 %. The difference inproductivity between trained and non-trained


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 269

workers was 8 %. These effects are based on esti-mates (0-100 % scale) of productivity growth bycompany personnel. The average length of trainingwas found to be close to six months.

In a study of programming consultants inSweden, Hansson (2001) found strong evidencethat the employer paid for all programming trainingeven though the resultant skills were highly attrac-tive to other firms. This study is unique in thesense that it had access to employee measuressuch as profitability, amount of training, wages,and each employee’s acquired human capitalstock (approximated by the individual’s compe-tence profile). The results indicate that theemployer not only paid for all direct costs associ-ated with the training (course fees, travelexpenses, etc.) but also lost a considerableamount of profit during the training. Hanssonfound no evidence that the individual contributedto the training investment by receiving a wagebelow his or her productivity. The findings alsosuggest that the employer recovered the invest-ment in programming training in the long run, asindividual programming skills (competence) weresignificantly associated with profitability. Theseresults were realised in an environment with similarworking conditions such as type of job, customerbase, etc., and with a number of control variables(including a control for differences in ability amongemployees). Hansson argues that the investmentin general human capital largely looks like anyother investment scheme that firms normallyundertake in their business operations (with aninitial investment and a payoff in the future).

The work of Gunnarsson et al. (2001) suggeststhat the increased educational level of theSwedish workforce between 1986 and 1995 is animportant factor in explaining IT-related produc-tivity growth during these years. Gunnarssonet al. examined the IT productivity paradox byincluding measures of interaction between IT andeducational level in 14 industries (manufacturingsector). The IT productivity paradox relates to thefact that massive investments in IT in the 1980sdid not have any positive effects on productivityuntil the beginning of the 1990s. The interactionbetween IT and educational level is significantand contributes significantly in explaining produc-tivity growth during this period. As the inclusion

of the human capital measures increases theexplanatory power substantially, the authorsconclude that human capital is a key in explainingthe IT productivity paradox. Other interestingfindings in Gunnarsson et al. are that a marginalskill upgrading has the same effect acrossdifferent levels of education and that IT-relatedproductivity growth occurs in several industriesoutside the IT sector.

Other European based studies include those ofKazamaki Ottersten et al. (1996) who studied theimpact of training in the Swedish machine toolindustry, using an evaluation drawn upon costfunctions and productivity estimations. Their anal-ysis is based on a formal model that was appliedto the panel data of eight Swedish machine toolfirms between 1975 and 1993. Their results implythat training expenditures result in net decrease intotal costs. The estimates of productivity effectsare also positive, but smaller in magnitude (alsoKazamaki Ottersten et al., 1999).

US studies that have had access to perfor-mance data on either employees or firms aremainly from the mid 1990s (5). Some results aregiven next.

Krueger and Rouse (1998) investigated work-place education programmes in two Americancompanies in the service and manufacturingsectors respectively. The programmes includedlearning of generic skills such as reading, writing,and mathematics as well as more occupationalskills such as blueprint maths and blueprintreading. The results indicated that participating ingeneric training classes had no significant impacton employee wage growth. Occupational trainingclasses, on the other hand, yielded a positiveimpact. Training influence on the available perfor-mance measures was generally weak. In theservice company, classes had no significantimpact on whether employees received perfor-mance awards or not. The effect on absenteeismduring the training period was positive in bothcompanies but not statistically robust. Theauthors also conducted a survey of the personnelat both companies. With two exceptions, the vari-ables showed no difference between employeesparticipating in the programme and thenon-participants. Participants were more likely toreport that they would take additional training


(5) Since we have not been able to find more recent papers we are uncertain whether this is caused by a lack of research orwhether we have missed out on important publications or working papers.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 270

classes in the future and they were also morelikely to report that their supervisor would saythat they were doing better than a year ago. Thelatter result might be interpreted as an improve-ment of self-reported job performance.

Another study focused on basic skills trainingis the investigation by Bassi and Ludwig (2000) ofdifferent school-to-work (STW) programmes inthe US. The purpose of the study is to analysewhether those programmes providing generaltraining also can be cost-effective for firms spon-soring them. Data comes from case studies ofseven STW programmes representing a diverseset of industries and regions. Data was collectedthrough interviews. The authors found that mostof the STW firms studied were willing to pay forgeneral training, though it is less clear whetherfirms would be able to recoup the full costs ofthis training, given labour market institutions andpublic policies in the US. Contrary to the predic-tions of the classic Becker model, the authorsfound that in all but one case the firm paid forsome or all of the costs of general training. Theresults show a substantial variation incost/benefit ratios across the STW programmes.The discounted cost/benefit ratios variedbetween 0.69 and 1.81. One explanation for thisvariation is that firms with relatively high ratiosmay be the ones that provide little training. Otherexplanations for the large variation in ratios arethe ability of students to pay for training and theability of firms to extract profits from trainedworkers. The findings also suggest that Americanlabour markets are imperfect enough to motivatefirms to participate in STW programmes.

The findings of Black and Lynch (1996) indicatea somewhat mixed result with regard to humancapital and productivity. This study used leveldata in estimating the impact of human capitalinvestments on (log) sales and a regressionmodel with a number of control variablesincluded in the regression. The results indicatethat human capital in the form of education had asubstantial impact on productivity. Formaltraining conducted outside the company had asignificant impact on productivity for manufac-turing firms whereas computer training had asignificant impact for non-manufacturing firms.The proportion trained did not yield any signifi-cant relationship. Their results also indicated thattraining appears to have a lagged impact on

productivity. Black and Lynch (1997) reworkedtheir regression model with access to longitudinaldata on productivity. The results in this estimationprocedure indicated no significant relationshipbetween training and productivity. The authorsattributed this insignificant impact to increasedmeasurement errors.

The results of Bartel (1995) indicate thatreceiving training increased the probability of apositive change in performance the following year.Bartel investigated 1 487 professional employeesin a manufacturing firm. Different types of trainingand the amount of training (days) showed nosignificant impact. The author attributed the ratherweak impact on employee performance to thesample and scale used in this specific case. Thesample consisted only of employees whoremained in the same job (e.g. employees whogot promoted were excluded from the sample).The performance rating was executed on a singleitem (7-point scale). Because of these twoconstraints the author argues that training effectsprobably underestimate the real impact thattraining has on employee performance.

Another study by Bartel (1994) suggests thatimplementing training programmes generatesconsiderable productivity effects measured as thechange in log sales. This finding is robust fordifferent personnel categories (professionals,clerical staff, etc.) and changes in personnel poli-cies (the results are not caused by a Hawthorneeffect). In addition, the results are robust to meanreversion of productivity between firms and showthat low productivity firms were more likely toimplement training programmes. That lowproductivity firms implement training programmesto a larger extent than other firms can producedownward biased or insignificant effects oftraining programmes in cross-sectional regres-sions. The effect of using cross-sectional dataappears to underestimate the impact of trainingon productivity growth. It is important to note thatthe training effect on productivity is achieved inexcess of changes in personnel policies, indi-cating once again that training is a major factor toconsider in HPWS literature.

The main findings in labour economics withregard to the impact of human capital oncompany performance can be summarised asfollows. Previous research on training and wageeffects has established that firms pay for all type


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 271


Study Database/ Data Sample Aim/ Methodsurvey and size subject

Dearden et al. British labour Incidence of 94 industries, Impact of First difference,(2000) force survey training, informal maximum 12 training on fixed effect,UK (LFS), COP and formal years, 818-970 productivity GMM system

consensus of training, industry years and wages equationproduction longitudinal

–industry level data

Barrett and EU survey In company 215 firms, Impact of First differenceO’Connell and a follow up training two points training on regression(1999) survey of Irish (continuous in time productivity, IRL business vocational impact of

training), general general and and specific specific trainingtraining, panel data

Groot Telephone and Duration of 479 firms with Impact of Frequency/(1999) questionnaire formal training 10 or more training on ordinary least NL survey employees wages and square (OLS)

productivity difference growth approach

Hansson Company Type of training, 132 Impact of OLS(2001) database training days, programming training and S competence, consultants competence

education (skills) onprofitability and wages

Gunnarsson Labour force Proportion with 14 industries Impact of Industry weighted et al. survey, different over 10 years human capital least square (2001) employment educational and IT regression, S register, levels on productivity interaction

investment growth education and ITsurvey, etc.

Black EQW national Company 1346 Impact of OLSand Lynch employers’ training. manufacturing training on(1996) survey Number trained, and non- productivity, US type of training, manufacturing impact of types

level data establishments of training

Bartel Columbia Implementation 180 firms in Impact of Level and(1994) Business of formal training manufacturing training first difference US School survey programmes sector over programmes regression

and compustat (manag., profes., three years on productivityclerical, produc. workers) paneldata

Krueger Company Type of training, 800 (of which Impact of OLS, Probit, and Rouse personnel basic skills 480 workers training on random and(1998) record education, attending wages fixedUS (manufacturing occupational training) and employee effect models

and service courses, performance

Bartel Company Incidence of 1 478 Impact of First difference.(1995) personnel formal training professional training on Two–stepUS records and days employees productivity multinomal

(manufacturing in formal training, and wages logit modelfirm) type of training

Table 1: Labour economics

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 272


Control Outcome Strength/ Findingsvariables measures weakness

Tenure, age, Productivity Extensive Training has a positive impact on productivityeducation, measured as robustness tests and wages, with a twice as large effect on occupation change in log and econometric productivity. Formal training has larger industry, R&D, real value added modelling/data impact on productivity than informal capital intensity, per employee on incidence training.firm size, etc. of training

(not days in training)

Change in Log sales Days in training, General training has a positive impact on personnel policy, growth panel data, productivity, specific training has no impact.corporate important restructuring/ control variablesorganisation, assets, employees, etc.

Tenure, age, Estimates Direct estimates Average productivity growth about 4-5 timestime since (100 % scale) of differences larger than wage growth. Weak connection training, of productivity in productivity/ between who contributes to training education, growth before estimates of investment and who benefits from the trainingmobility, etc. and after, trained productivity

and not-trainedTenure, age, Profitability, Direct measure The concurrent impact of training on profitgross (revenues net of profitability is negative and impact on wages positive.contribution, of wage and and human The skill/competence of the individual is ability, education, overhead costs) capital significantly related to profitabilitygender, etc. stock/level data,

single employerBusiness cycle, Total factor Lagged IT and Interaction term IT and educational levelnon-computer productivity human capital is highly significant indicating that theequipment, IT effects increase in productivity growth is largelygrowth, gender, tied to increase in higher educational leveletc.

R&D, capital, Log sales A number of Education positively related to productivity, TQM, multiple growth control variables, formal training off working hours and establishment, type of other training variables computer training

training/level significant with productivity, not (number data trained)

Capital assets, Log sales Controls for Implementing training programmes positively no. employees, growth personnel related tochange in productivity, not due unions, raw policy, relation to mean reversion orchange in personnel policy.material, age between training Low productivity firms more likely to implementof firm, industry, programme training programmeschange in and personnel policy productivity/weak

training measureEducation, Performance Homogenous The work place education programme hadtenure age, awards workers, generally a weak effect on the employee gender, absenteeism, compare performancemeasures, except for company) self-reported results at perceived performance. The effect of the type of performance two companies/ training was positive but not significant in department, weak many cases.type of work, etc. performance dataOccupational Change in Controls for Individuals receiving training significantly dummies, performance determinants associated with probability of increased education, rating of training/weak performance score. Days in training and type tenure, etc. outcome variable of training not significantly associated

(rating by managers) increased performance.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 273

of training no matter whether it is general orspecific. More recent findings also indicate thatfirms benefit from such training investments.There are even signs that general (formal) trainingproduces greater benefits than specific training.There are also some indications that the educa-tion and skills of the individual are associated withproductivity and, in some cases, with profitability.

3.2. HRM and HPWS literature

The impact of human resource management(HRM) practices on company performance hasattracted considerable attention. Many specialissues of management literature are devoted toHRM practice and company performance, forinstance the Academy of Management Journal(Vol. 39, No 4, 1996), International Journal ofHuman Resources (Vol. 8, No 3, 1997), HumanResource Management (Vol. 36, No 3, 1997),Human Resource Management Journal (Vol. 9,Issue 4, 1999). The argument put forward in thisline of research is that advanced HRM practicesproduce a higher level of productivity. The find-ings suggest that there is a connection betweenHRM practices or what is often referred to ashigh performance work systems (HPWS), andcompany performance indicators such as sales,market values, market-to-book values, prof-itability, productivity, etc. (6).

Generally, this research area has good access tocompany based performance measures. Thedisadvantage is often that the statistical methodsare based on level data, which makes it difficult toestablish causality. That most of the research isbased on level data is largely a consequence of thefact that firms seldom make any large changes totheir HRM policies. Measuring changes in HRMpractices, therefore, requires extensive measure-ment periods (longitudinal data). Much of the infer-ence about the impact on firm performance is thusconfined to cross-sectional data. However, manypapers use a research design that accounts for theheterogeneity among firms, which makes the statis-tical models more robust.

In HPWS literature, education and training ispart of a larger package of the activities of a

human resource function. The areas that are typi-cally covered in these studies are screening andemployee selection, compensation systems,employee communication, teamwork practices,etc. In many cases studies also examine howaligned or integrated these practices are with theobjectives or strategy of the company. Much ofthe current debate centres on whether bundles ofhuman resource practices are the source of valuecreation in firms or whether certain practicescontribute more than others. There is also thequestion of whether there is a HRM practice thatis generally applicable to most enterprises orwhether HRM practices are firm-specific orcountry-specific.

For instance, a Dutch study by Boselie et al.(2001) argues that the institutional setting inEurope affects the potential to create high perfor-mance work practices because of the presenceof strong labour regulations and the interaction ofsocial partners. They maintain that to applyresearch on high performance work practices weneed to adjust the theoretical framework to suitthe European situation. Boselie et al. (2001) alsoprovide an overview of the findings that HRMresearch has produced in the last decade. Theresults with regard to the effects of training oncompany performance are reproduced below.Some of these papers will be examined in greaterdetail in this section:(a) training has a positive impact on the different

dimensions of the performance of the firm:product quality, product development, marketshare and growth in sales (Kalleberg andMoody, 1994);

(b) higher investment in training results in higherprofits (Kalleberg and Moody, 1994; d’Arci-moles, 1997);

(c) higher investment in training results in a lowerdegree of staff turnover (Arthur, 1994);

(d) training has a positive impact on the relation-ship between management and the otheremployees (Kalleberg and Moody, 1994);

(e) training has a positive impact upon perceivedorganisational performance (Delaney andHuselid, 1996);

(f) management development is positivelyrelated to profit (Leget, 1997);


(6) These human resource management practices are sometimes referred to as human capital enhancing systems, or highcommitment policies (systems).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 274

(g) focus on training is positively related toperceived profit, market share and investmentin the near future (Verburg, 1998);

(h) training practices affect perceived organisa-tional performance positively (Harel andTzafrir, 1999).

(Boselie et al., 2001, p. 1112)

Besides the issue of whether there is a genericHRM practice, there is continuing debate aboutwhether employee development is the key factorin the HRM bundles (e.g. Barnard andRodgers, 2000). The results presented in theprevious section of this review suggest thattraining is a main factor in generating productivityeffects, as training yields a significant impactwhile controlling for changes in personnel prac-tices (Bartel, 1994; Barrett, 2001). Other studiesmaintain that the impact on company perfor-mance is caused by the combined effect of HRMpractices (Becker and Huselid, 1997;Huselid, 1995; Becker and Gerhart, 1996). Thiscontroversy is also reflected in the selectedstudies of our review of this research area.

HPWS might appear to be connected mainlywith the knowledge intensive sector, but humanresource policies to enhance efficiency andworker commitment can work equally well inmore mature sectors of the economy. Ichniowskiet al. (1995) investigated HRM policies in the USsteel industry. Their findings indicate that innova-tive human resource practices have a significanteffect on productivity. Ichniowski et al. (1995) wasone of the few studies to examine the produc-tivity effect of firms changing their humanresource practices. Interestingly, the impact ofthe first-difference (change) approach supportedthe results of original estimates on level data. Thebenefits, in the form of increased revenues, farout-weighted the costs associated with thesehuman resource programmes. Ichniowski et al.(1995) also argue that complementarity betweendifferent human resource practices has a signifi-cant effect on worker performance, whilechanges in individual employment practices havelittle or no effect (training by itself is not enough).

D’Arcimoles (1997) utilises the disclosure ofinformation in French company personnelreports. These reports are sanctioned by law andinclude unmatched firm-based information on themain aspects of HRM such as compensation,training, recruitment, dismissal, and general

working conditions. Apart from having access todata on variables that researchers normally havea great difficulty in obtaining through surveys, thisstudy also has access to data over time (paneldata). The panel with training and HRM measuresincludes six two-year periods. The main resultswith regard to training are that the level of traininginvestment is consistently associated with boththe level of, and changes in, current and futureproductivity and profitability. Profitability isapproximated by the return on capital employedand productivity by value added per employee.The impact from the change in training on thechange in productivity seems to appear with aconsiderable lag. The results presented by d’Arci-moles suggest that the effect of training invest-ments might take as long as two to three yearsbefore they emerge in form of increasing produc-tivity. The results between the change in trainingand change in profitability are less precise. Still,these findings indicate that there exists a causallink between training and firm performance in thesense that firms invest in the current period andharvest the benefits in future periods. One mightadd that these results are achieved while control-ling for absenteeism, hiring/dismissal, work acci-dents, and total rate of resignation (all controlvariables are considered proxies for working andsocial climate at the firm).

Laursen and Foss (2000) studied the relation-ship between HRM practices and innovationperformance based on the data of the DISKOproject, a large survey on innovation behaviour in1 900 Danish firms, cofunded by the OECD. Thesample includes nine sectors in manufacturingand service industries. Laursen and Foss alsopropose some theoretical explanation as to whyHRM practices influence innovation performance,e.g. new HRM practices often increase decentral-isation, in the sense that problem-solving rightsare delegated to the shop floor, which might facil-itate the discovery and utilisation of local knowl-edge and thus enhance innovation. Due to thecomplementarities between HRM practices, theyalso state that systems of HRM will be signifi-cantly more conducive to innovation than indi-vidual practices.

Laursen and Foss used principal componentanalysis based on nine HRM factors and identi-fied two different HRM systems that areconducive to innovation. In the first system all


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 275

nine HRM variables are relevant; in the secondonly performance-related pay and internaltraining are dominant. In the latter system onlythese two factors out of nine have individualimpact, but when all factors are combined into asingle variable this is highly significant. Thisfinding supports their thesis of the complemen-tarities of HRM practices. In addition, they alsoidentified some sector-specific patterns, such asthe fact that firms in wholesale trades tend tobelong to the second system. They conclude thatthe application of HRM practices is important tothe likelihood of a firm being an innovator.

Table 2 presents the studies by Ichniowskiet al. (1995), d’Arcimoles (1997), and Laursen andFoss (2000). These studies have made an effortto disentangle training effects from other HRMpractices and two of them have also been able toexamine the impact on company performance ofchanges in HRM and training.

Other HRM studies also measuring the effectof training separately from HRM practices includeDelaney and Huselid (1996) based on US data.The training measurement in their study shows aconsistent and significant relationship withperceived organisational performance (irrespec-tive of the statistical model). The results forperceived market performance are less clear butsuggest a significant influence of at least 10 %.

These training effects are demonstrated in thepresence of other HRM aspects such as staffing,compensation, degree of internal labour market,etc. Delaney and Huselid used cross-sectionaldata on 590 firms to estimate organisationalperformance and 373 firms to estimate stockmarket performance. Because this study is basedon cross-sectional data it is difficult to establishthe relationship between organisational perfor-mance and training measurement.

Michie and Sheehan (1999) studied the data ofthe UK’s workplace industrial relations survey(WIRS) with regard to the impact of HRM practiceson innovation. They separated three different typesof HRM practice using variables such as payment,worker involvement in teams, incentives, informa-tion sharing, etc. However, they did not explicitlyintegrate the degree of training. Innovation ismeasured by research and development (R&D)expenditure and the introduction of new micro-elec-tronic technologies. The workplace industrial rela-tions survey contains information on 2061 firms withmore than 25 employees in various sectors. Michieand Sheehan were able to use 274 data sets thatcontained information on HRM as well as innova-tion. Based on an econometric model, whichexplained the probability of innovating, they wereable to identify some significant HRM factors. Theirresults suggest that low road HRM practices – short



Ichniowski Interviews Proportion Longitudinal Impact Level andet al. (1995) and al. employees data on 36 of HRM first difference US company received steel practices (fixed effect)

specific off-the-job production on regressiondata training lines, productivity

(dummy 2 190 monthly variable) observations

d’Arcimoles French Formal 61 firms Impact Level (1997) company training level data, of HRM, regressionFR personnel expenses, 42 firms wage growth, and first

report level and panel data and training differencechange (7 years) on firm OLS data performance regressions(panel data)

Laursen Data Classification 1 900 firms HRM Probit and Foss exaggerated of two HRM Practices model(2000) from the practices and their to explain theDK DISKO impact probability

project on innovation to be an (Database) performance innovator

Table 2: HRM (HPWS) studies

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 276

term contracts, etc. – are negatively correlated withinvestment in R&D and new technology, whereashigh road practices are positively correlated withR&D investment and the introduction of new tech-nology. This study also shows that skill-shortage isa serious obstacle for innovation and for movementtowards differential and higher-priced products.Their findings also deliver evidence that the strategyto increase employment flexibility by short-termcontracts, weakening trade unions, etc., does notenhance the innovation performance of firms.

MacDuffie (1995) tested the impact of humanresource activities in automotive production. Thesample consisted of 62 car assembly plants inthe US, Asia, Europe and Australia. The studywas influenced by the organisational contingencytheory and the hypothesis that the internal fitbetween different organisational strategies andcharacteristics affects performance (7). MacDuffieseparated two production systems, massproduction and flexible production. Disruptions tothe mass production process prevent the realisa-tion of economies of scale with the use of buffersas an indicator for the prevalence of this system,

whereas under flexible production buffers areseen as costly.

Within flexible production, the link betweenminimisation of buffers and the development ofhuman capabilities is driven by the philosophy ofcontinuous improvement. MacDuffie separatedproduction organisation, work systems and HRMpolicies as three interrelated independent variablesand creates corresponding indices to capturesystemic differences in organisational logicbetween mass production and flexible production.He used different human resource variables suchas job rotation, recruitment policy, training of newemployees, etc., based on a cluster analysis, toclassify human resource practices characterisedthrough a consistent bundle. Performance hasmeasured by labour productivity and quality,expressed as defects per 100 vehicles.

Regression analysis indicated that indices ofmass production, flexible production, transition,and intermediate stage, were statistically signifi-cant predictors of productivity and quality.High-commitment human resource practices,such as contingent compensation and extensivetraining, in flexible production plants, were char-


(7) The contingency theory stresses the importance of the link between human resource practices and other factors such asorganisational strategy. In contrast, the universal approach maintains that human resource management practices have a posi-tive effect on performance in general (independent of other factors).


HRM controls, Productivity A number of Adopting a coherent system of HRMcontrols for measured robustness tests, practices produces significant production line by uptime of same production productivity effects. Adopting (maintenance, production line process, panel individual work practice in age, raw data/weak isolation has no effect on material, training data productivity (training).etc.)Wages, social Productivity Impact of Level of training is consistently climate (change in change correlated with level and change (absenteeism, value added), in training in productivity and profitability,work accident, profitability measured with change in training is associated social (return on lag, HRM with change in performance expenditures), capital controls/ weak with a two to three year lagemployment employed) firm and (less stable result)variables, etc. industry controlsSector, firm Innovative Only innovation The application of HRM practices size, performance performance is does matter for the likelihood cooperation the dependent of a firm being an innovator

variable

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 277


acterised with low inventories and repair buffers,and consistently outperformed mass productionplants. MacDuffie found empirical evidence thatbundles of interrelated and internally consistenthuman resource practices, rather than individualpractices, are better predictors of performance.Overall, evidence supports the thesis thatassembly plants using flexible productionsystems, which bundle human resource practicesinto a system that is integrated with productionstrategy, outperformed plants using more tradi-tional mass production in both productivity andquality. While other academics believe that eithermass or flexible production will perform well ifthere is a good fit between their human resourceand production strategies, MacDuffie found thatflexible production leads to better performancefor automotive plants.

A similar matter was raised by Arthur (1994) instudying the impact of human resource systemsacross steel mini-mills in the US. His main hypoth-esis was that specific combinations of humanresource policies and practices are useful inpredicting differences in performance andturnover. In the tenor of the contingency theory, hestated that congruent human resource and organi-sational policies are more significant than separatehuman resource practices. He separated twohuman resource systems, control and commit-ment, which stress the importance of cost reduc-tion and commitment maximisation respectively,and shape employee behaviour and attitudes atwork. Despite the contingency view of humanresource systems, he stated that, in general,commitment human resource systems will beassociated with higher performance, especiallybecause of the high control and monitor cost inthe control system. In addition, he was interestedin the question of the impact on turnover.

Arthur expects higher turnover in firms withcontrol systems because of the lower cost forwages and training. Empirical data were based ona questionnaire from human resource managersof 30 US steel mini-mills. Based on a cluster anal-ysis, he separated the two human resourcesystems, using labour efficiency, scrap rate(number of tons of raw steel to melt one ton offinished product) and turnover as performancemeasures. Regression analysis indicated that thepresence of commitment human resourcesystems was significantly related to fewer labour

hours per ton and lower scrape rate, whereasturnover was higher in control systems. Arthuralso found, that human resource systemsmoderate the relationship between turnover andperformance, since there is a negative relationshipbetween turnover and performance in commit-ment systems. However, the results have to betreated with caution due to the small sample size.

Baldwin and Johnson (1996) studied businessstrategies in Canadian small and medium-sizedfirms, with less than 500 employees, demon-strating varying degrees of innovation. Besidesmarketing, finance, and production strategiesthey also investigated whether innovative firmsalso followed specific human resource strategies.The sample size was 850, including all majorindustrial sectors. Innovation classification wasbased on the traditional question of R&D intensityand additional variables regarding innovationbehaviour (patenting, source of innovation, etc.).

Baldwin and Johnson were able to verify theirthesis that greater innovation accompaniesgreater emphasis on human resources, stressingtraining. They found statistical evidence that moreinnovative firms offered formal and informaltraining more often and with greater continuity,accompanied by innovative compensation pack-ages. While almost three-quarters of the group ofmore innovative firms offered some form oftraining, just over half of the group of less innova-tive ones engaged in training. The authors alsoused quantitative data to estimate the amount oftraining and found that the more innovative onesspent CAD 922 on average per employee, signifi-cantly more than the CAD 789 spent by less inno-vative firms. Moreover, the former firms usedproduction employees more often as a source ofinnovation.

In addition to linear correlation analysis, theauthors used multivariate models to establishwhether certain combinations of factors or allfactors combined contributed to a given humanresource strategy. A probit model and principalcomponent analysis indicated that the firms thatfollowed the most comprehensive humanresource strategy (stressing all factors simultane-ously) are most significant. Considering all factorsstudied (human resources, marketing, etc.) theyfound that all areas are important for innovationsuccess and that more innovative firms take abalanced approach to their business by striving

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 278


for excellence in a number of different areas.However, Baldwin and Johnson did not carry outanalysis of how far specific human resourcestrategies are related to other business strategies.Finally, they analysed the relationship betweeninnovation behaviour and company performance,based on administrative data sources. Variousperformance criteria and indices (sales, prof-itability, market share, employment, and assets)suggested that more innovative firms performedbetter. Thus, the study delivers evidence for arelationship between sophisticated humanresource strategies and company performance.

In conclusion, HRM literature stresses theimportance of comprehensive HRM practices ingenerating effects on company performance(training by itself is not enough). However, fewstudies measure training separately and fewstudies have had the opportunity to examinechanges in HRM practice and in performance.Nevertheless, there are indications that training ismore efficient or generates larger effects inconnection with other HRM practices.

3.3. National European surveysand cross-nationalcomparisons

Many surveys and studies conducted by nationalinstitutions attempt to answer questions such aswhat generates innovation and what createsgrowth in their respective countries. Similarly,many cross-national comparisons are aimed atunderstanding the reasons for differences ingrowth and innovation between countries.Measurements of training and education are typi-cally included in these types of study as part of thefirm’s innovative capacity. Besides human capital,studies typically include variables that areassumed to generate innovations and growth,such as investments in IT, R&D, technology, capitalintensity, etc., and a number of control variables.The aim in many instances is to understand theinnovative capacity of firms and not human capitalper se, though this is an advantage as the resultsfor education and training are generally morerobust (as the inclusion of other variables controlsfor the influence of these factors).

A study by a Swedish business developmentagency (NUTEK, 2000) on different learning

strategies shows that competence developmentactivities are associated with both productivityand profitability. Training is measured in a broadsense by three establishment level activities(planning, learning at the job, and proportion ofemployees trained). In this study the effects oftraining activities are observed after holding otherlearning strategies constant (e.g. R&D, innova-tions, cooperation, etc.). Other findings indicatethat the effect is more pronounced for larger firmsand that higher educated personnel is associatedwith both productivity and profitability. This studyuses level data, which render causality difficult toachieve. However, since both profitability (whichis net of investment costs) and productivity aresignificantly influenced by training activities, itstrengthens the interpretation that trainingincreases productivity and that employers areable to capture some of the returns generated.

In a study of Finnish companies, Leiponen(1996a) finds a significant association betweeneducational level and profitability. Other findingssuggest that strong complementarities existbetween different educational factors. It appearsthat a sufficient number of more highly educatedemployees is a prerequisite for the profitability ofdoctoral level researchers. The results alsosuggest that innovative firms are more dependenton educational competence in generating profit.The sample used in this study consists of paneldata for 209 firms. The author deals with theendogeneity problem (caused by the effect ofprevious economic performance on the explana-tory variables) by applying a two-stage regressionprocedure. An interesting finding is that withoutaddressing the problem of endogeneity betweenprofitability and other human capital variables,the results are largely insignificant. Leiponenconcludes that general competences acquired ineducation, notably higher and post-graduateeducation, are beneficial for the profitability of thefirm. In another study of Finnish companies,Leiponen (1996b) comes to the conclusion thatinnovative firms have a more educated workforceand that they are more profitable than non-inno-vating firms.

The German Institute for employment research(IAB) has, since 1990, conducted a large establish-ment survey with over 9 200 participating establish-ments. This is one of the most comprehensive estab-lishment panel surveys in Europe (Bellmann, 2001).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 279



NUTEK Flex-2 survey Learning 911 The impact OLS level(2000) (telephone strategies, establishments of learning regressionsS and competence strategies

questionnaire development on firm’ssurvey) activities competitiveness

(measured 0-3 scale), level data

Leiponen Survey data, Educational 209 firms, Impact of First difference(1996a) compiled by level, type of time-series education on GMM, two-stageFIN Statistics of education data profitability weighted least

Finland (15 (technical, (1985-93) and innovations square regressionmanufacturing natural industries) science)

Bellmann IAB Amount 3 400 Impact of OLS level and Büchel 1997 invested cross-sectional training regressions, two (2000) Survey in training observations on productivity stage least D squareDoucouliagos Case Cost and 7 cases Application Design of a et al. study benefits of a four-step cost-effective (2000) of training evaluation training evaluationAU programmes process of model

training investments

Blandy Questionnaire Training 41 Effects of the Regressionset al. similar to UK quantity on-the-job ‘matched plant’(2000) CEP survey in hours training on methodology AU by the LSE productivity between hotel and

and earnings kitchen furniture manufacturers

Maglen Case study Training 30 case studies Evaluation Comparativeet al. based on expenditure in four sectors of the return case studies(1999) interviews with on training AU managers and depending on

employees other factors

Table 3: National and cross national studies

However, the research that has come out of the IABsurvey so far is less focused on the effects oftraining investments for firms and more on theeffects for individuals. An exception is the study byBellmann and Büchel (2000) examining the effectsof continuous vocational training on productivity.Their result is based on 3 400 cross-sectionalobservations from the 1997 IAB survey. The authorsapply different models in estimating the impact oftraining, including an OLS and a two-stage regres-sion model as well as including controls for industry,size, and employee characteristics. The initialregression results for both parts of reunitedGermany indicate a significant relationshipbetween how much is invested in training and

productivity (log annual sales per employee).However, the authors argue that this finding islargely a consequence of a selection problem, suchas the individuals receiving training having moreability and that certain firms are more capableproviding adequate training for their employees.Bellmann and Büchel stress the importance ofstrategic HRM practices in generating productivityeffects from training investment.

Some interesting studies have also been carriedout in Australia. Blandy et al. (2000) found a posi-tive relationship between a firm’s profitability andthe quantity and quality of training offered by thefirm, the latter also being correlated with otherforms of investment. A profitability index, based on

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 280



Other learning Productivity Includes a Competence development activities strategies, R&D, (value added), large number have a significant effect IT, innovations, profitability of control on productivity and profitability. cooperation (revenues to variables and Larger effects for larger firms. with partners, cost ratio) competing Education is associated with profitability.decentralisation, learning etc. variables/level

data, weak training data

Sales, market Net profit Use first Educational competence is significantly share, capital margin, difference, associated withprofitability. intensity, innovations control variables, Complementarities exist betweenindustry such as two-stage different general skills dummies patents, weighted least acquired in higher education.

improvements, square to handle Innovative firms are etc. simultaneity more dependent on educational

problems and level in generating profitability.GMM

Industry, size, Productivity Large number Initial results indicate a strong associationemployee measured by of observations/ between training and productivity. However,characteristics log annual level data the results are likely driven by ability.

sales Authors stress HRM practices as main factor.(case study) Calculations Quantitative Return on investment (%) calculated

of return on estimation of is between 70 and 7000investment returns on

training investments

Industry Productivity The study uses Profitability is directly related toand quantitative i) quantity and quality of training,profitability data on ii) firms paying above market wage rates,

training and iii) firms’ difficulties in finding suitable employeesperformance/ iv) no clear picture regarding the impact of small sample size training onthe productivity

Firm strategy Labour Control variables In most cases training investmentsand human productivity are only used had led to positive returns, though dependent resource policy for a qualitative on human resource practices; the bundling

interpretation of human resource policies is crucial; better of the differences performers also planned strategically

firms’ statements on profitability compared withtheir main competitors, was found to be positivelyassociated with indices of volume (expenditure ontraining relative to main competitors) and thequality of company training (existence of a formaltraining strategy, written commitments to trainingin workplace agreements, etc.). In addition, themore profitable firms pay above market wage ratesand operate in labour markets where suitablelabour is hard to find and keep. The authorsconclude that the principal reason why training isprofitable for firms is that it increases the produc-tivity of their employees more than it raises theiremployees’ wages.

In another Australian study, Maglen and Hopkins

(1999) integrated variables such as work organisa-tion, job design, employment practices and othercompany-specific variables to analyse returns ontraining. They compared enterprises that producedsimilar goods or services, and were similar in size.Their findings suggest that there is no key set of rela-tionships that could be translated into a series ofbest practice procedures, but that the main factorsin optimising business performance are linksbetween strategic objectives and practice within theenterprise itself. This means that the effectiveness oftraining is contingent upon the idiosyncratic circum-stances of the firm, a theory based on Becker et al.(1997) and known as ‘idiosyncratic contingency’. Acomparison between seven firms in four sectors

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 281

found that better performers, measured by labourproductivity, included training as an integral part ofstrategic planning. In contrast, the poorestperformers lacked these characteristics.

Doucouliagos and Sgro (2000) developed atraining evaluation model, which they tested onseven Australian firms with longitudinal firm-leveldata, aiming to calculate the return on investmentin training, both financially and non-financially.Collected data indicated that the return on invest-ment in training programmes varied between 30and 7.125 % for seven different trainingprogrammes. Performance and benefits wereestimated by indicators such as the saving ofenergy after training of train drivers or increase insales growth after the training of store managers.

Carr (1992) studied productivity differentials inthe automotive sector between Britain, Germany,the US and Japan and the impact of skills differ-ences. He based the investigation on 56 matchedvehicle component manufacturers in the four coun-tries in 1982. The study included interviews withchief executive officers and other personnel downto shop floor level. In 1990 Carr carried out 45 moreinterviews to gauge the effects of past differencesand to capture the impact of increased labour flex-ibility in Britain. Despite differences in product andtechnical characteristics between the productareas studied, the data clearly showed productivitydifferentials between Britain, Germany, the US, andJapan. In nearly all areas Britain lagged behind,measured by sales per employee. The studyprovides some evidence that the highly educatedand trained German workforce (craft apprentice-ship, the high qualification of the foremen, andvocational training organised by firms) explains theproductivity advantage of Germans firms. However,Carr was not able to carry out a statistical analysisregarding the relationship between specific trainingand its volume, measured by factors such asexpenditure, and its impact on productivity.Compared with the 1980s, the data showed the UKto be catching up slowly, which he ascribes to anincrease in labour flexibility in Britain, though unableto verify this statistically. Finally, he concludes thatthe UK has to place more emphasis on high stan-dards of basic education and programmes aimed atcontinuous employee development.

Mason et al. (1992) carried out a study similarto Carr (1992), where they compared the produc-tivity differences between Britain and the Nether-

lands and the impact of vocational education. Likeother studies, their investigation deals with lowerworkforce qualifications in Britain. Mason et al.compared the skills and productivity in a matchedsample of British and Dutch manufacturing plants.The study was conducted on 36 plants in twoindustries, engineering and food-processing. Theauthors also summarised the main specifics of theDutch vocational education and training system,which relies principally on full-time vocationalcolleges. This system is nearer to the Frenchschooling-based system than to the Germanapprenticeship system. They concluded that thehigher average level of skills and knowledge in theDutch workforce contributes to higher productivitythrough better maintenance of machinery, greaterconsistency of product-quality, greater workforceflexibility, and less learning-time on new jobs.These findings are not based on a statistical anal-ysis regarding the two samples, but rathercomparing the general profiles and characteristicsof workforces in the two countries. Substantiallyhigher proportions of vocationally qualifiedpersonnel were found at virtually all levels inDutch plants in both industries.

Concluding, the indications are that the educa-tional level appears to be a factor in profitabilityof firms. There are also some indications thatdifferences in educational level (system) betweenEuropean countries might explain productivitydifferentials among the countries. Most studiesstress the importance of training as an integralpart of strategy and other HRM practices.

3.4. Small and mediumenterprise (SME) surveys

Even though some of the above mentionedstudies include smaller firms in their sample anduse firm size as a (control) factor, few studiesexist which deal explicitly with the impact oftraining investment on SME performance.

Leitner (2001) carried out a study on AustrianSMEs (between 20 and 50 employees) with theaim of investigating the impact of differentstrategic investments on company performance.He also analysed training investments, measuredwith an ordinal variable, in the context of variousinternal strategic factors, endogenous factorsand their impact on performance. He found that


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 282

the amount of training was one of the few internalfactors that had a direct impact on companyearnings, this correlation being highly relevant forsmaller firms with less than 50 employees. Ingeneral, smaller firms invested less than biggerfirms. However, no positive relationship betweenthe different kinds of strategies pursued and theamount of training could be identified.

Furthermore, Leitner discovered a positive rela-tionship between training and the corporateculture and communication within the company.Given the impact of external (endogenous)factors, he found that training was an essentialaspect of profitability for firms in dynamic envi-ronments. He concluded that, in general, traininginvestments allow firms in different competitiveand rather hostile environments (mature productlife cycles, conjuncture-dependent life cycles,dynamic environment) to perform better thanother firms in similar environments. His findingspartly support the idea that the impact of trainingon company performance depends on endoge-nous and firm specific factors.

Romijn and Albaladejo (2000) investigated therole of various internal and external sources of inno-vation capability in SMEs in the UK. Besides factorssuch as R&D investment and interaction withresearch institutions, they found that a range ofinternal factors, including the owners’ technicaleducation and their prior work experience, the tech-nical skills of the workforce (measured by univer-sity-trained engineers as % of total workforce) andtraining (measured as training expenditure peremployee and % of sales) have a significant effecton innovation capability. Furthermore, a close linkto nearby training institutions also had a positiveimpact. Innovation capability was measured by thepresence of innovations during the preceding threeyears and their technological complexity. However,the authors did not report any specific analysisregarding firm size and its impact.

There is considerable literature on the importanceof human capital for the success of businessstart-ups (e.g. Trouvé, 2001). In these studies formaleducation, skills and experience and the talent of thefounder are integrated to explain business success.However, the role that the training of the workforceor teams plays for a company’s success is scarcelyinvestigated. Bosma et al. (2002) studied the valueof human capital for start-up companies inthe Netherlands. The study separated general,

industry-specific and entrepreneurship-specifichuman capital investments by the founder andmeasured performance according to survival, profitand employment generated. The main findings arethat investments in industry-specific human capital,such as former experience, and entrepreneur-ship-specific human capital, such as experience inbusiness ownership, contribute significantly to theperformance of small firm founders. General invest-ments, such as the level of higher education, play aminor role. A methodological problem of the study isthat investments are only operationalised by theexperience of the founder, without a direct analysisof training expenditure during the firm’s life.

The Irish study by Barrett and O’Connell (1999)cited in Section 3.1.1, also analysed whether afirm’s size had an impact on the relationshipbetween general or specific training and perfor-mance but they did not find any significant differ-ences based on the size of the firm. The study byMason et al. (1992), carried out in British andDutch SMEs with up to 400 employees, indicatedthat the Dutch productivity advantage wasgreatest in product areas where small- ormedium-sized batches were in demand by themarket. In engineering plants they found no vari-ation in productivity with firm size, but in thefood-processing industry (biscuits) they foundsome differences. While the largest British biscuitplants, which were highly automated, had thesame productivity as the Dutch firms, smallerDutch plants were almost twice as productive ascorresponding British plants. Mason et al. ascribethis difference in productivity to the lack of tech-nical competence of the workforce.

While few studies have examined the effects ofhuman capital and human capital investment on theperformance of smaller firms, there is nothing indi-cating that experience, skills and training shouldhave any less impact on company performance. Onthe contrary, many of the studies emphasise theimportance of these factors for smaller firms.

3.5. Other training and impactstudies

The research done at the American society oftraining and development (ASTD) is an importantsource of information. ASTD perform annualbenchmarking studies on employee development


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 283

and training. A clear advantage of the datacollected by ASTD is the quality of the informationon training. All companies subscribing to ASTD’sbenchmarking survey gather information on typesof training, amount spent on training, etc. Allmeasures of training are clearly defined andcompanies participating in this survey providetraining information that is collected in a similarway. Studies based on ASTD data thus have farless variance caused by measurement errors intraining than most other studies. The importanceof committing firms to a common definition andstandard of measuring training cannot be over-stated. The lack of a common definition of what toregard as training will be discussed in more detailin the last chapter of the present study.

The possibility of connecting training data withcompany outcome measures is also an importantadvantage of the ASTD database and offersadvantages to investigations of company training.The study by Bassi et al. (2002) shows that firmswith higher training investments also have higherstock returns the following year, higher grossprofit margins, higher return on assets, higherprice to book ratio, as well as higher income peremployee. The results are very similar whenexamining changes in performance measure-ments. An important aspect of being able toconnect training investment with profitability andstock market performance is that we aremeasuring the impact net of the cost of theinvestment. The study by Bassi and van Buren



SME

Leitner Empirical Importance 100 Intangibles, Correlation(2001) study of training SMEs strategy and Anova-A based on a and firm Analysis

questionnaire performance

Bosma Panel survey General, 1 100 Does human Regressionset al. among 1 100 industry-specific firms capital (2002) new business and investmentNL founders entrepreneurship- enhance

between specific entrepreneurial 1994-97 investment in performance?

human capital

Romijn and Survey of Skills of 50 information Internal and Correlation Albaladejo 50 SMEs workforce and computer external sources analysis(2000) (interviews) technologies of innovation UK (ICT) and capability in

electronic firms SMEs

Other studies

Bassi ASTD survey, Training 314 Impact of OLS, first et al. Compustat investment publicly training difference(2001) per employee listed firms investment regressionsUS on firm’s

(stock market) performance

Büchel German Educational 800-1 900 Impact of Logit,(2000) Socio-economic level individuals overeducation survivalD panel (GSOEP) on productivity model

Table 4: SMEs and other training studies

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 284

(1998) indicates a similar result, that training haspositive effects on company performance.

That training investments are associated withstock market performance is more rigidly demon-strated in Bassi et al. (2001). In a well-specifiedregression model the authors demonstrate thatthe level of training expenditure (investment) peremployee is associated with next years’ stockmarket return. This training effect is demonstratedin the presence of variables capturing stockmarket risk and known stock market anomaliessuch as momentum and the book-to-marketeffect. While controlling for momentum effect byincluding a lagged dependent variable, theauthors also account for (eliminate) the potentialeffect that training might have had on stock

returns during the investment period. It is alsoimportant to note that a first difference approach(change in training investment and stock marketreturn) gives substantially the same result. Theresults also indicate that the effects of trainingemerge with some lag and that training appears tohave long-term effects on profitability.

An important aspect of training and the impacton stock returns is that this information is valuerelevant and that investors are currently unable toget hold of this type of information. In a perfectworld, a well-informed investor would anticipatethe increased earnings and returns from suchhuman capital investments at the moment theywere made. Because of the absence of informa-tion on human capital investments in corporate



Industry, size, Earnings, sales No quantitative Firms with regular trainingstrategy and employment measurement of have higher returns

growth training investments

Various controls Survival, profit, Human capital Human capital (experience of the founder) such as industry, employment is measured influences the entire set gender, labour only through of performance measuresmarket histories experience

None with Innovation Captures various Skill of workforce (share of university-trained) respect to capability indicators to has a positive impact training and (index) explain on innovation performanceperformance innovation

performance, not possible to relate training directly to performance

Industry, R&D, Stock market Good quality Training investments are associated assets, returns, income-, training data, with next year’s stock market performance.investment, price Tobins Q-, and firm performance Same result for level and change data. to book, beta sales per measures, and Changes in training investment price/earnings, employee control variables can predict future stock returnsprevious performance, etc.

A number of Job satisfaction, Large number Overeducated employees are healthier, employee tenure, of observations/ have longer tenure and are characteristics, participation in self-reported more work and career minded.age, gender, on-the-job- variableshealth status, etc. training, health

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 285

reports it appears that investors are unable togather this type of information. However, ifinvestors had knowledge of these investmentsand anticipated the impact on the share price, wewould be unable to document any effect on shareperformance the following year.

Another study that connects human capital withstock market performance is Hansson (1997) onSwedish stock market information. Because of theunavailability of training data, the author createdthree stock market portfolios reflecting depen-dence on human resources and investment intraining. One portfolio mainly consisted of knowl-edge-intensive firms, a second reflected lessdependence on human resources and consistedlargely of capital-intensive firms. The third portfoliocontained a mix of firms from both knowledge-and capital-intensive sectors. The basic idea of thestudy was that these portfolios would mimic unob-served training expenditure, with relatively higherhuman capital investment in knowledge-basedfirms compared with capital-intensive firms. Thefindings indicated that knowledge-based firmsconsistently earned higher risk-adjusted returnsthan the more capital-intensive firms. The authorascribed the higher returns in the portfolio withknowledge-based firms to greater investment inunaccounted (unobserved) human capital.

In a German study on overeducation andemployee performance, Büchel (2000) comes tothe conclusion that overeducated employees inlow-skill jobs tend to be more productive than theircorrectly positioned colleagues. Büchel used ques-tions from the German socio-economic panel(GSOEP) and found that overeducated personnelhad better health, received more on-the-jobtraining, had longer tenure, and were not moredissatisfied with their job than personnel with theright educational level. This study usedself-reported questions on productivity. The resultsindicated that the risk of employing too highlyeducated personnel might be exaggerated sincethere are no indications that overeducatedpersonnel are more negative towards their work.These findings also suggest that we may possiblybe less concerned with the potential negative

effects that education can have on company perfor-mance, i.e., education might have an upside but aless pronounced downside effect.

The effects of web-based training are studiedby Schriver and Giles (1999) within a nuclearfacility, Oak Ridge, Tennessee, US. The organisa-tion has about 14 000 employees, with a trainingbudget that reflects their high qualificationrequirements. The training for Oak Ridge ismainly organised by the centre for continuouseducation, which serves as corporate university.In 1995 the management decided that significantsavings could be realised by using the Intranet todeliver selected courses and qualification tests.Schriver and Giles evaluated this new project,introduced in 1997. Calculating the investmentsin the new system such as technology, downloadcosts, etc., and savings such as travel costs,fewer instructors, physical copies, classrooms,etc., they found that the cost-benefit ratio was1:9.5. The return on investment was calculated at845 %. However, despite this convincing data,the analysis should be taken with caution, since itdid not assess the effectiveness of training andany intangible effects, such as networking oppor-tunities. Neither did it calculate how much timethe employees spent in front of the computer.

In summary, studies based on stock marketdata suggest that human capital and investmentin it are important factors in understanding stockreturns. Some of the studies based on the ASTDdata provide convincing evidence that traininggenerates substantial gains for firms. The studyby Büchel (2000) also suggests that we might beless concerned with the potential negative impactof overeducation on firm performance.

It is important to point out that additional refer-ences to impact research can be found in Barrett(1998, 2001). While these overviews also largelycover the impact that training and skills have onfirm performance, there are some overlaps withour study. We have, however, tried to distinguishour review not only by including other researchpapers but also by looking at some of the prob-lems associated with this kind of research from adifferent perspective.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 286

The Cranet network was established in 1989 byfive founder countries (Germany, Spain, France,Sweden and United Kingdom). It is coordinatedby the Centre for European Human ResourceManagement at Cranfield School of Manage-ment. The Cranet survey is now the largest andmost representative independent survey of HRMpolicies and practices in the world. The networkitself is a collaboration between 34 universitiesand business schools, which carries out a regularinternational comparative survey of organisationalpolicies and practices in HRM across Europe.

Cranet has been running the survey since 1990using standardised questionnaires sent to privateand public organisations in different countries. Thestandardised questionnaire is translated into eachmember country’s language and adapted to thedifferent national contexts (taking into considera-tion such factors as legislation, labour markets andculture). During each round of the survey, amend-ments are made to capture new developments butthe questionnaire stays mainly unchanged in orderto observe developments over time.

The questionnaire is distributed by post, exceptin Greece where interviews are used to gather theinformation. The questionnaire is distributed toorganisations with 200 or more employees and isaddressed to the most senior human resource/personnel specialist in the organisation. The 1999survey was distributed to over 50 000 organisationsand 8 050 responses were received giving a totalresponse rate of 15 %. The willingness of compa-nies to respond was higher in Scandinavian coun-tries than in Southern European countries. Thecurrent version of the 1999-survey databaseincludes 8 487 observations from 27 countries, 17 ofwhich are European countries. The number ofobservations in the forthcoming tests variesbecause not all companies have answered all ques-tions. The survey is sent to both private and publicorganisations. The descriptive statistics in thepresent investigation include responses from bothpublic and private organisations. For the analysisthat includes performance variables, the sample isrestricted to private organisations.

It is also important to note that the provider of

the information in this survey is the firm and thefigures presented here can deviate from otherstudies. As noted by Barron et al. (1997)employer-based surveys typically report moretraining than individual-based surveys. Becausethe survey is focused on larger organisations it isexpected that the incidence and amount oftraining will be higher than in surveys conductedon a more distributed population.

4.1. Selected Cranet surveyquestions

A clear advantage of the Cranet survey is theaccess to two direct questions on companytraining. The first question concerns the amountspent on training in proportion to annual salaries.The second question is related to the proportionof employees that participated in training duringthe year. As with most training studies, we havevery little knowledge of what the respondentsregard as training. It can be anything from contin-uing vocational training, initial training, toin-company apprenticeship training. Other ques-tions related to employee development includewhether the organisation has a written policy fortraining and development, whether it analysesemployees’ training needs, and whether it moni-tors training effectiveness.

Questions related to company performance areweaker. Variables are perceptions of organisa-tional performance (fairly common in studies)typically measured at different levels. One ques-tion is related to the performance of the organisa-tion for the past three years. This variable ismeasured at five levels and could be included inthe analysis as a measurement of whether prof-itability affects the provision of training. The mostinteresting performance measurements are ques-tions on the rating of the organisation’s perfor-mance compared with other firms in the samesector. These industry-adjusted questionsconcern productivity, rate of innovation, servicequality, profitability, and stock market perfor-

4. Cranet survey results

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 287

mance. Variables are measured at three levels,according to whether the firm belongs to:(a) the top 10 % of the firms in the sector;(b) the upper half of the firms in the sector, or;(c) the lower half of the firms in the sector.

Preferred outcome variables in this survey aremeasurement of stock market performance andprofitability, as these are net of the investmentcost of training. It is important to note that we arenot working with actual performance but withperceptions of performance, which are both a‘noisier’ measurement and demonstrate less vari-ation than actual performance. Performancemeasurement in the survey has the advantage ofbeing relative to other firms in the same sector(industry). Controlling for heterogeneity betweenindustries or sectors has proved to be importantin most firm-based studies of company training.

Two variables related to internal market andunionisation are utilised in the survey. The vari-able on unionisation is measured at six levelsfrom 0 % to 100 %. A measure of internal marketcan be constructed from the question on howmanagerial vacancies are filled. The respondentsare asked how three different levels of managersare recruited: senior, middle, and juniormanagers. As with most proxies this variable isonly a rough measurement of the actualconstruct, which also includes other characteris-tics such as promotion, seniority, on-the-jobtraining, etc. The respondents can select fromfour different types of recruitment:(a) internally;(b) head hunters or recruitment consultancies;(c) advertising in newspapers;(d) word of mouth.

From this it is possible to construct a roughmeasurement of the extent the firm’s internalmarket. The number of ticked suggestions inoption (a) (maximum three) related to the totalticked suggestions (maximum twelve) can workas an approximation for the degree of internalmarket. It is arguably a coarse measurement butit indicates whether the recruitment of managersis focused on internal employees or whether it isfocused on attracting employees externally.

Other variables that can be used in explainingtraining are educational structure (% graduates),age structure (% above 45 years), proportion of

manual workers, number of employees, and theimportance of innovations. The questions used inthe present study involve mainly those from theemployee development section, organisationaldetails and some questions on unions, staffingpractices and the human resource function ingeneral. The Cranet database is a rich source ofinformation and only a small section is analysed inthis report. Please refer to Annex 1 for the exactwording of all variables included in this analysis.

4.2. Cranet survey results

4.2.1. The scope of training

Table 5 provides descriptive statistics on thetraining variables used in this study. The tableshows the mean values for the percentage ofwage bill spent on training and the proportion ofemployees trained in each country. The averageamount of wage bill spent on training in 1995survey was 3.1 %; the figure was 2.94 % in the1999 survey. The proportion trained in 1999 issomewhat lower than the figures given by Euro-stat in the second continuous vocational trainingsurvey (8). The countries that deviate most fromthe Eurostat figures are Belgium (-14 %), Norway(-11 %), and Ireland (-10 %) while other countriesshow more or less the same result. The restrictedsample in the Eurostat survey is slightly differentbecause it contains enterprises with250 employees or more compared with 200 andabove in the Cranet survey. The difference in theamount of training between the original (Euro-pean) countries in 1995 and the new countries in1999 is also marginal. More striking is the factthat the proportion of employees trained hasincreased quite dramatically in all countries, theoverall increase being 11 % since 1995 to about45 % of the employees receiving training eachyear in 1999. Again there is no marked differencebetween European countries and other countriesin the proportion of employees trained.

The number of firms answering each questionis given in parenthesis. In the 1999 survey, (whichis the foundation for this investigation)5 463 public and private organisations answeredthe question on the amount spent on training and


(8) Eurostat: The second continuous vocational training survey, 1999 (Newcronos).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 288

6 685 organisations answered the question onthe proportion trained during the year. For somecountries the number of observations is low. Thatsome countries have too few observations isevident when the sample is restricted to privatefirms in the analysis of training and performance.As can be seen in Table 5, the survey is mainly

focused on European countries with most of theanswers from countries within the EuropeanUnion.

If one compares these figures with the 1999ASTD survey it appears that European companiesspend considerably more on training than theirAmerican counterparts. In 1999 US firms spent


1995 1999 1995 1999Country Code % spent on % spent on Change Proportion Proportion Change

training training trained % trained %

United Kingdom UK 2.6 (813) 2.9 (644) + 0.3 42.1 (903) 52.9 (784) + 10.8

France F 4.8 (499) 4.2 (374) - 0.6 44.2 (471) 49.5 (355) + 5.3

Germany D 2.8 (321) 2.8 (311) UNCH 26.4 (357) 32.8 (415) + 6.4

Eastern Germany D (east) 3.2 (155) 2.5 (151) - 0.7 27.7 (189) 31.8 (218) + 4.1

Sweden S 4.3 (186) 3.7 (157) - 0.6 51.5 (266) 66.1 (239) + 14.6

Spain E 2.1 (213) 2.0 (241) + 0.1 37.6 (247) 51.1 (268) + 13.5

Denmark DK 2.8 (427) 2.8 (306) UNCH 37.2 (496) 49.6 (355) + 12.4

Netherlands NL 3.8 (238) 3.4 (202) - 0.4 34.0 (258) 42.2 (193) + 8.2

Italy I 1.9 (65) 2.2 (63) + 0.3 21.4 (86) 36.2 (71) + 14.8

Norway NO 2.7 (331) 3.3 (325) + 0.6 40.0 (317) 41.5 (344) + 1.5

Switzerland CH 2.9 (170) 2.6 (113) - 0.3 36.0 (189) 42.8 (128) + 6.8

Ireland IRL 3.7 (249) 3.2 (278) - 0.5 36.2 (308) 47.1 (381) + 10.9

Portugal P 3.0 (128) 36.9 (128)

Finland FIN 2.7 (262) 2.5 (215) - 0.2 45.2 (269) 61.1 (229) + 15.9

Greece EL 2.5 (81) 36.0 (111)

Austria A 2.2 (153) 36.4 (181)

Belgium B 2.4 (252) 3.3 (182) + 0.9 27.9 (311) 45.5 (223) + 17.6

Northern Ireland NI 3.1 (107) 54.4 (157)

Estonia EE 4.3 (304) 47.0 (388)

Bulgaria BG 2.9 (70) 17.4 (107)

Czech Republic CZ 2.3 (141) 45.3 (176)

Cyprus CY 1.4 (52) 34.1 (73)

Turkey TR 3.8 (96) 3.9 (128) + 0.1 27.9 (148) 49.4 (191) + 21.5

Tunisia TN 4.3 (45) 24.6 (54)

Israel IL 3.3 (89) 47.5 (148)

Japan JP 1.7 (423) 31.6 (582)

Australia AU 3.0 (180) 56.3 (186)

Average 3.10 (4277) 2.94 (5463) -0.08 35.7 (4815) 45.22 (6685) +11.0

NB: The table shows the average percentage of wage bills spent on training in each country (% spent on training) and theaverage proportion of employees trained during the year (Proportion trained %) in the 1995 and in the 1999 Cranet survey.Number of firms answering each question in parenthesis.

Table 5: Cranet survey, descriptive statistics

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 289

1.8 % of payrolls on training (van Buren andErskine, 2002) compared with 2.9 % for the firmsincluded in our sample. It is important to notethat both surveys are completed by companiesand that the question regarding the amountinvested in training is similar in both studies. Themeasurement errors in the Cranet survey areprobably larger because of less rigorous trainingdefinitions in the questionnaire. Whether thismeasurement problem inflates or deflates thereported figures in the Cranet survey comparedwith the ASTD survey is difficult to determine.

4.2.2. Correlation between variables

The investigation that follows is based on thesample of private companies for which there areperformance measurements. The remainder of theanalysis in this section is restricted to 5 824 privatecompanies that answered the 1999 survey. Table 6shows the correlation between the main explana-tory variables used in the present study (number ofobservations in parenthesis). We have chosen vari-ables in an effort to reflect factors used in bothlabour economics and human resource literature.To increase readability, only significant correlationsare shown in the table. A number of interestinginitial observations can be made. First, the amount

spent on training is positively correlated withpersonnel turnover. This is an observation thatgoes against common knowledge that turnoverreduces training. The proportion of employees intrade unions is negatively correlated with amountof training and staff turnover. That training is less inmore unionised companies is possibly the result ofa greater proportion of manual workers. Moreinteresting is that turnover appears to be lower inmore unionised establishments (especially since% manual workers are positively correlated withturnover). This result is in line with the argumentthat unions reduce personnel turnover and therebypromote more training (e.g. Booth et al., 1999).

The internal labour market measurement isnegatively related to the amount of training andproportion trained, contradictory to expectationswhich assume that firms invest in trainingbecause internal labour markets reduce the riskof poaching. However, the prediction thatturnover is lower in firms with more focus oninternal promotion appears to be confirmed in thepresent material (negative correlation betweenturnover and internal market). That the internallabour market is positively correlated with theproportion of graduates and negatively correlatedwith the proportion of manual workers seem


% Turnover Absenteeism Unionisation Internal % %trained L.M. graduates manual Size

% spent 0.179 (a) 0.041 (b) -0.109 (a) -0.105 (a) 0.102 (a) -0.110 (a)

on training (3419) (2998) - (3217) (3645) (2726) (2774) -

% trained- - - -0.031 (b) 0.137 (a) -0.167 (a)

(4672) (3418) (3502) -

Staff turnover0.089 (a) -0.179 (a) -0.131 (a) -0.066 (a) 0.053 (a)

(2550) (3757) (4258) (3314) (3273)

Absenteeism0.158 (a) -0.133 (a) -0.266 (a) 0.191 (a) 0.047*

(2586) (2885) (2336) (2261) (2829)

Unionisation0.059 (a) -0.232 0.245 (a)

(4908) (3451) (3635) -

Internal Labour 0.207 (a) -0.057 (a)

Market (3879) (4069) -

% graduates-0.468 (a)

(3072) -

% manual -

(a) denotes significance at 1 % level(b) denotes significance at 5 % level

Table 6: Correlations between main explanatory variables (private sector)

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 290

plausible. While our measure of internal labourmarket is at odds with the prediction related totraining, the remaining correlations appear to bein line with expectations. However, a more rigidanalysis of what determines training will beperformed in Section 4.2.3.

Other correlations that are also less consistentwith prior expectations relate to graduates andmanual workers. The size of the organisation,measured by the number of employees, is notcorrelated with the two training variables. Thisapparent inconsistency with prior findings mightbe due to restricting the sample to firms with200 or more employees. A somewhat surprisingresult is also the very low correlation between theproportion of wage bills spent on training and theproportion of employees trained. This resultsuggests that these two factors measure differentaspects of training. We will elaborate on thisargument in more detail in the next section. Inconclusion, the results of Table 6 indicate thatthere is also little or only moderate correlationbetween our explanatory variables.

4.2.3. What determines training?

A number of questions in the Cranet survey offerinteresting perspectives on the amount or inci-dence of training. As well as including customaryfactors such as education, age, company size,unionisation, and manual labour, the Cranetquestionnaire includes variables that mightprovide a better understanding of what influencesthe decision to train people. First are questions ofwhether the company has a written trainingpolicy, analyses employee training needs, andhas an internal labour market. Other variablesinclude personnel turnover and whether innova-tions are important. Perhaps the most interestingvariable is the one that indicates prior profitability,as this describes the impact of prior performance(profitability) on the decision to train. This variablemay assist in understanding the causality oftraining, i.e., whether profitable firms can affordtraining or whether training generates profitability.Finally, two different training measurements(proportion trained and proportion of wage billsspent on training) might provide some idea ofwhether these two common estimates of training

measure the same thing. To account for whatdetermines the amount of training and proportiontrained in a company the following trainingregression is estimated (9):

TRAINING = POLICY + NEEDS + INTERNAL +UNION + AGE45 + MANUAL + GRADUATES +TURNOVER + SIZE + PRIORPROFIT +INNOVATION (1)

where:

TRAINING is the amount of training or propor-tioned trained (all subscripts suppressed);

POLICY is a dummy variable taking the value ofone if the firm has a written training policy;

NEEDS is a dummy variable equal to 1 if the firmanalyses employee training needs;

INTERNAL and UNION are internal labour marketand degree of unionisation;

AGE45 is the proportion of employees 45 yearsor older;

MANUAL, GRADUATES, TURNOVER, SIZE arepercentage of manual employees, percentage ofgraduates in the workforce, percentage ofpersonnel turnover and number of employees atthe firm;

PRIORPROFIT is the measurement of the perfor-mance over the past three years;

INNOVATION is a dummy variable taking thevalue of 1 if the firm considers innovation as veryimportant to the organisation.

The results for estimating equation 1 areshown in Table 7. We have chosen to analyse theresults of Table 7 in the light of the findings inliterature that employers pay for all types oftraining no matter whether the training is specific,industry-specific (occupational) or general innature. The first variable indicates that firms witha written training policy are more likely to providetraining to their employees (proportioned trained)but having a written training policy is not associ-ated with how much training is provided. Thesecond variable indicates that firms that analysetheir employees training needs also train their


(9) Because less than 1 % of the firms answered that no employees received any training and just over 1 % of the firms answeredthat 0 % of salaries is spent on training, an OLS regression is estimated in the training regression.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 291

employees to a greater extent than firms notconducting this type of analysis.

The impact from the internal labour marketmeasurement is negative on both the amount oftraining and number of people trained, which indi-cates that firms focusing more on internal promo-tion provide less training. This is contrary to thefindings of Delaney and Huselid (1996) for the USmarket. The correlation between training and theirmeasure of internal labour market was high (0.56)indicating a that firms with a higher degree ofinternal labour market also provide more training.Whether this deviation in our result is caused bydifferent measures of internal labour market orwhether there exist differences between US andother (predominantly) European countries is diffi-cult to gauge. However, the negative impact of theinternal labour market measurement in our studymight also be interpreted along the lines thatthese types of internal structures do not provideenough incentive to train or to be trained. Internallabour markets are typically based on seniority inpromotion and pay levels are based on position(post occupied) rather than on competence andskills. Both the employee and the employer haveless incentive to invest in training. For a morethorough discussion on the subject see Hanchaneand Méhaut (2001).

The degree of unionisation at the firm has anegative impact on the proportion of the wage billspent on training whereas the impact on thenumber of employees trained is positive (but notsignificant at 5 % level). Though we have controlsfor both educational level and the proportion ofmanual workers at the firm, the negative associa-tion between unionisation and the amountinvested in training is probably caused by ourinability to control for industry differences in thetraining regression. The impact from having moreold employees in the organisation is negative onboth training measures but not significant. Thisresult might be taken as an indication thatcompany training persists throughout theemployee’s career. The proportion of manualworkers is associated with fewer workers being

trained but not with the amount of trainingprovided at the firm. The proportion of graduatesat the firm has a positive but not significantimpact on training measurements. The size of theorganisation is not associated with any of thetraining measurements, possibly explained by oursample of larger firms.

Personnel turnover appears not to be a factordetermining training since it is not significant inthe training measurements. This result is qualita-tively similar to those presented in Goux andMaurin (2000) for France and in Green et al.(2000) for Britain. Both studies indicate thattraining measured at an aggregated level hadlittle impact on mobility. Considering the impor-tance of staff turnover to the potential for compa-nies to benefit from training investments we alsoconducted a simple analysis of aggregatedcountry data and found a positive relationshipbetween average personnel turnover and averageproportion of wage bills spent on training. Thisoutcome is slightly contrary to what is expected,as turnover of personnel is normally consideredto lower firms’ willingness to invest in training.However, turnover might force firms to increasetheir spending on training newly hired employees.A division of what type of training is provided bythe firm might thus give a different result (10).

The variable that indicates past profitabilityshows an interesting division in the impact on thetwo training measurements. Prior profitability ispositively and significantly associated with theproportion of employees being trained but notassociated with the proportion of wage bills spenton training. This result indicates that the propor-tion trained in a firm is largely conditioned by pastperformance. It seems that measurements oftraining based upon proportion of employeesbeing trained contain an element of reward or, atleast, dependence on past performance. That thedecision regarding the number of employeesbeing trained is endogenous to, or mutuallydependent on, past profitability makes it impor-tant to address the problem of endogeneity instudies using this measure of training.


(10) The country comparison is based only on univariate regression and there could thus be several factors such as economicconditions, unemployment rate, etc., that might drive this outcome. However, Green et al. (2000) also come to the conclusionthat different types of training have different impacts on the individual’s decision to search for a new job. The large differencein personnel turnover between different countries might still indicate that this is an important variable to consider in cross-country comparison. There are considerable differences in personnel turnover between European countries, with the lowestturnover in the Netherlands (4.69) and Germany (5.70) and the highest turnover in the UK (15.06) and Portugal (13.24).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 292

Because the proportion of wage bills spent ontraining is not associated with past performance,it might indicate that how much firms invest intraining their employees is not dependent onwhether they can afford training or not. Takingthis reasoning a bit further suggests that theinvestment volume is not a reward for pastperformance but more likely seen as an invest-ment with future benefits for the firm. That theamount spent on training is not associated withthe dependent variable (prior profitability) alsogives us a better basis for making interpretations


Dependent % spent Proportionvariable on training trained

INTERCEPT2.795 (a) 14.637 (a)

(5.53) (3.25)

POLICY0.233 9.434 (a)

(1.22) (5.60)

NEEDS0 844 (a) 17.575 (a)

(4.05) (9.51)

INTERNAL-1.501 (a) -6.883 (b)

(-4.92) (-2.51)

UNION-0.126 (a) 0.725

(-2.74) (1.75)

AGE45-0.004 -0.072

(-0.85) (-1.88)

MANUAL-0.003 -0.135 (a)

(-0.84) (-4.53)

GRADUATES0.008 0.070(1.89) (1.79)

TURNOVER0.005 0.090(0.70) (1.40)

SIZE-0.000 -0.000(-0.27) (-0.30)

PRIORPROFIT0.099 4.465 (a)

(1.21) (6.02)

INNOVATION0.169 0.637(1.03) (0.43)

F-statistics 7.827 29.556

R2 (adjusted) 0.05 0.167

N 1359 1566

T-values are in parentheses.(a) Denotes significance at the 1 % level.(b) Denotes significance at the 5 % level.

Table 7: Training regression (private sector) in, for instance, cross-sectional estimates. Inother words, examination of the amount investedin training might suggest that it is not profitablefirms that can afford training but it is training thatgenerates profitability.

The last variable, the importance of innova-tions, shows a positive but not significant impacton training measurements. Given the generallevel of analysis used in this study, the resultsshould be interpreted with some care, particularlybecause the regressions do not include importantvariables such as controls for industries orcontrols for countries. However, an impression ofthe results presented in Table 7 is that trainingappears to be discretionary within firms. Themeasurements that show the strongest influenceson training are largely determined by the firmitself. The difference between what determinesthe proportion being trained and what determinesthe amount of money spent on training is alsoworth noting. The large difference in impactamong the explanatory variables indicates thatthese two measurements indicate quite differentdimensions of training decisions. This finding is inline with the arguments in Orrje (2000) that thedeterminants of the probability of receivingtraining and the determinants of the amount oftraining are not the same.

4.2.4. What factors are indicative for top

(10 %) performers?

This part of the analysis utilises the questions onhow the organisation is performing relative toother firms in the same sector. Table 8 shows themean difference between organisations belongingto the top 10 % of performers and organisationsconsidered performing below average in thesector. Five industry-adjusted performancemeasures are shown in Panel A (profitability,productivity, innovations, service quality andstock market performance). The minimum andmaximum number of observations used in theanalysis is given in brackets in the first column.We have chosen to examine whether topperformers show any differences with respect tovariables normally considered in studies. Themean difference between top performers andthose performing below average is shown in thetable together with t-values (in parenthesis).

Panel A gives the results for the whole sampleof private organisations and Panel B gives theresults on profitability for United Kingdom, France

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 293

and Germany. The lack of observations for otherEuropean countries rendered further divisionsunworkable. Since we are not working with actualperformance but perceptions of performance weneed to be cautious with interpretations. Oneproblem with the data is that around 30 % of thefirms responded that they belonged to the top10 % of the firms in their sector, suggesting thatwe either have a response bias or that we haveover-sampled firms performing well. To whatextent this caveat influences the resultspresented in Table 8 is difficult to gauge, but aresponse bias is likely, rendering the statisticsless significant.

The first column shows that top performingfirms spend more on training compared with firmsperforming below average. This is true for allperformance measurements except for servicequality where there is no significant difference.For instance, firms belonging to the top 10 %with regard to profitability on average spend0.6 % more of their wages on training than firmshaving a profitability below average in theirrespective industry. Considering that prior prof-itability did not have any significant impact on theamount spent on training (Table 7) it might beassumed that top performing firms are topperformers in part because of their investment intraining (11).

The difference between top performers and thebelow 50 % performers with regard to theproportion trained in a year is significant for allperformance measurements (including servicequality). That high performers in service qualitytrain significantly more of their employees eachyear most likely indicates that achieving topservice quality requires all staff to undergo regulartraining. Top performers train close to 10 % moreof their staff in a year compared with firmsperforming below average on service quality.

Having a written training policy and analysingemployee training needs appear to be indicativefor top performing firms regardless of perfor-mance measurement. These two variables areimportant determinants of training, training poli-cies and support functions to provide the righttype of training, and they are characteristic ofhigh performing firms whether measured by prof-

itability, productivity, innovations, service quality,or stock market performance.

The next column indicates that firms consideredas more innovative also employ more highlyeducated personnel (percentage of graduates inthe workforce). This result is reasonably expectedbecause of the complementarity between innova-tion and education (e.g. Leiponen, 1996b). Thatfirms with better profitability and stock marketperformance also employ more graduatescompared with firms performing below average ismore interesting. This result is in line with the find-ings presented earlier that the educational level ofemployees is positively associated with produc-tivity and profitability (Black and Lynch, 1996;NUTEK, 2000; Gunnarsson et al., 2001). That firmsare able to extract higher profitability from moreskilled or more educated workers is an argumentput forward by those who propose that wagecompression is a major reason for firms to invest ingeneral skills (Acemoglu and Pischke, 1998, 1999a;Booth and Zoega, 1999; Brunello, 2002).

The last columns indicate that staff turnover doesnot vary between high and low performers, but thathigh performing firms have significantly less absen-teeism among employees than low performingfirms. An exception occurs in the results for servicequality where there is no difference between the twogroups with regard to absenteeism.

Panel B indicates that some of the results inPanel A might be driven by country specificcircumstances since not all of the three countriesshows the same response on profitability.However, the lack of significance between prof-itability and some of the variables is possiblymore a consequence of fewer sample cases.France, with few cases shows a significant differ-ence only in the proportioned trained whereasthe UK and Germany, with more cases, also showmore significant results. It is interesting to notethat staff turnover is significant in Germany yetnot for the total sample or for France and the UK.Another observation is that absenteeism is onlysignificantly different in the UK. Because of thelow numbers it is difficult to draw any generalconclusions regarding differences between thethree countries. One lesson that can be drawnfrom the Cranet survey is that when the respon-


(11) Because it appears that the decision on how much to spend on training is not conditioned by how profitable the firm has been,it is more likely that training generates profitability and not the other way around that profitability generates training (since thereis no association between amount of wage bills spent on training and past three years profitability in Table 7).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 294

dents are requested to provide actual figures on,for instance, training, staff turnover, absenteeism,etc., the response rate drops dramatically.

4.2.5. Main findings

The investigation is based on a crude statisticalanalysis, compounded by the problem of percep-tions of organisational performance, which tendto increase measurement errors and decrease thetrue variation among firms. This analysis of theCranet data should, therefore, be seen as a firstrough attempt to explore this database. However,some interesting indications have emerged fromthe present analysis, as summarised below:(a) the proportion of the wage bill spent on training

and of employees being trained appears not to

measure the same thing because the determi-nants of these two variables are quite different(the correlation between the two trainingmeasures is also weak);

(b) the proportion of employees trained appearspartly dependent upon whether the firm canafford the training, as indicated by the annualnumber of employees being trained corre-lating with prior profitability;

(c) the proportion of the wage bill spent ontraining does not correlate with past prof-itability, which might indicate that it is not aprofitable firm that can afford training but it istraining that generates profit;

(d) analysis of employee training needs and exis-tence of a written training policy are two


% wages Proportion

Written Analyse % % staff Absenteeism

spent on trained %

training traininggraduates turnover days

training policy needs

Panel AAll countries

Profitability 0.60 (a) 9.21 (a) 0.08 (a) 0.06 (a) 2.46 (b) -1.22 (a)

N (1498-2795) (4.29) (6.81) (4.26) (3.58) (2.23)–

(-3.21)

Productivity 0.43 (b) 7.46 (a) 0.05 (b) 0.06 (a) -1.84 (a)

N (1243-2389) (2.27) (4.36) (2.12) (2.65)– –

(-3.67)

Innovations 0.73 (a) 9.80 (a) 0.12 (a) 0.09 (a) 6.62 (a) -1.26 (a)

N (1253-2375) (4.52) (6.63) (6.00) (4.90) (5.31)–

(3.05)

Service Quality 9.92 (a) 0.10 (a) 0.11 (a)

N (1549-2983)–

(3.90) (3.16) (3.48)– – –

Stock Market 0.55 (a) 6.34 (a) 0.06 (b) 0.06 (b) 6.04 (a) -1.87 (a)

N (617-1174) (2.74) (3.04) (2.26) (2.23) (3.42)–

(-2.95)

Panel B

Specific countries

UK – Profitability 0.63 (c) 10.17 (a) 0.10 (b) -1.19 (b)

N (189-381) (1.79) (2.64)–

(2.52)– –

(-2.18)

F – Profitability 15.14 (a)

N (91-184)–

(3.79)– – – – –

D – Profitability 13.26 (a) 0.25 (a) 0.15 (b) -2.60 (b)

N (127-196)–

(3.39) (3.68) (2.36)–

(-2.40)–

T-statistics (parenthesis) whether the mean is different from zero.(a) denotes significance at 1 % level(b) denotes significance at 5 % level(c) denotes significance at 10 % level

Table 8: Mean difference between top 10 % and lower half of firms in sector (private sector)

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 295

important factors associated with the numberof employees trained at the firm and, for theformer, also the amount of training providedby the firm;

(e) most education and training-related measure-ments are significantly higher or morefrequent in high performing firms comparedwith firms performing below average in theirrespective sector.

An interesting aspect of these results is that thedecision to train is largely determined by companyspecific factors. Without taking the interpretation ofthe results too far, a general picture seems to

emerge where factors that one can consider asapproximations for good working conditions or the‘good employer’ are also largely connected withperformance measurements such as productivityand profitability. It is also important to note thatfirms with better profitability and stock marketperformance also provide more training, and trainmore of their employees, compared with firmsperforming below average in their respectiveindustry sector. Another notable observation is thatfirms with better sector-adjusted profitability andstock market performance also have moreeducated employees.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 296

The amount of money spent by companies onwages (the rent for human capital) suggests thatmore efficient use of this production factor wouldhave a significant impact on most firm-basedperformance measurements. The flaw in thisreasoning is that firms do not own the labour oftheir employees but pay a rent for its hire. Thequestion is whether increased efficiency arisingfrom company training benefits the hirer of thelabour or the provider. That at least some of theincreases in efficiency are captured by firms is aprerequisite for any impact on company perfor-mance measurements such as profitability orstock market performance.

Research on the effects of education, training,and skills/competence on company performancehas made some important advances in recentyears. Increasing evidence that employers benefitfrom human capital, and investment in humancapital, has led to a substantial increase in thenumber of theoretical papers seeking to explainthese findings. In the following section weprovide information on what empirical researchhas achieved in different areas.

5.1. The effects of training oncompany performance

5.1.1. The impact of training

A somewhat crude description of the researchagenda is that the general understanding hasmoved from regarding all investments in employeetraining as unprofitable for firms, to regardingspecific training as viable, and now to consideringall types of training as potentially profitable.Considering that 40 years have passed sinceBecker (1962) wrote his seminal paper on humancapital investment, research has moved slowly onthe question of whether firms can benefit fromtraining their employees. Only very recently havewe seen papers showing that employers profitfrom investment in all kinds of training.

The majority of the papers included in this

review point toward substantial gains foremployers from continuous vocational training.The absence of studies indicating that employersdo not profit from training investments maygenerate some concern over whether we have apotential bias in reporting only significant andpositive results for company training. However,because research on training has largelyaccepted that employers pay for companytraining (no matter how general) the finding thatemployers also benefit from such investmentseems progressively more plausible. Despiteacceptance that firms pay for (general) training,there is still a need for additional research on theeffects of training to understand better thecomplexities of company-provided training.

Increasingly, studies provide evidence thattraining generates substantial gains foremployers. The most compelling evidence ispresented in several recent papers connectingtraining investment with changes in productivity,profitability and stock market performance. Themajority of these studies also establish the direc-tion of the relationship, i.e., we can, with reason-able confidence, maintain that training generatesperformance effects and not the other wayaround. The studies that provide the strongestevidence of this are (dependent variable in paren-thesis):(a) Barrett and O’Connell (1999) based on

215 Irish firms (sales growth);(b) Dearden et al. (2000) based on 94 British

industries over 12 years (value added);(c) Groot (1999) based on 479 Dutch firms

(productivity estimates);(d) Hansson (2001) based on a Swedish case

study of programmers (net profitability);(e) d’Arcimoles (1997) based on French firm-level

data (value added, return on capitalemployed);

(f) Bassi et al. (2001) based on 314 US firms(stock market return, sales per employee,etc.).

There are also studies that have only been able todemonstrate a weak connection between companytraining and company performance (Black and

5. Results

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 297

Lynch, 1997; Bartel, 1995). However, the authors ofthese studies attribute weak or insignificant resultsto measurement problems. The main impressionfrom research is that firms profit from training theiremployees regardless of whether the trainingprovided is useful to other firms. Other tentativeconclusions that can be drawn from the review arebriefly presented below.

5.1.2. Formal/informal and general/specific

training

Few studies have been able to examine theeffects of different types of training. Besides diffi-culties in acquiring more specific training infor-mation, the distinction between different types oftraining appears somewhat arbitrary since mostdefinitions are not mutually independent. Never-theless, the results concerning formal andinformal training suggest that formal trainingcourses have more impact on productivity thaninformal training (Dearden et al., 2000; Black andLynch, 1996). This finding is puzzling, because itis likely that formal training relates more togeneral training and informal training more tospecific training. This result would indicate thatgeneral training might be more profitable for firmsto invest in than specific training. The study byBarrett et al. (1998) suggests that this might bethe case. The results of Barrett et al. show thatgeneral training has a significant positive impacton productivity whereas specific trainingappeared to be insignificant. The authors explainthis result by reasoning that general trainingprovides greater incentive for employees tospend more effort in the learning process.

However, the results are not entirely consistent.Bosma et al. (2002) conclude that their findingssupport the thesis that specific investments aremore influential for company success in start-upsthan general investment. This difference in resultswith regard to general and specific training can beaffected by weak definitions and measurementproblems, or the difference between entrepreneurialperformance and employee performance.

Other studies that can be distinguished interms of type of training are the studies focusedon teaching basic skills to employees (work placeeducation programmes) or students (STWprogrammes). The impact of generic skillsprogrammes is ambiguous in that it is difficult toassess the real pay-off for employers from thistype of training investment. Though much

research is needed on this subject, there are indi-cations that basic skills training can influence firmperformance positively (Bassi and Ludwig, 2000;Krueger and Rouse, 1998).

5.1.3. Timing of the effects of training

An important issue related to the impact oftraining is when one can expect to see the effectsof the training investment. There is evidence thateffects of training emerge some time after it takesplace. The results and arguments forwarded inBlack and Lynch (1996) and Bassi et al. (2001,2002) based on US data, d’Arcimoles (1997)based on French data, and Hansson (2001)based on Swedish information, suggest that theeffects of training materialise one to two yearsafter the training period. The results presented inthese studies suggest that we should measurethe effects of training after at least one year fromthe point of the investment and possibly also overa longer time horizon.

However, typically the effects of training areregistered in cross-sectional estimates, whichimplies an instant effect of the training investment.The question is whether this impact is caused by animmediate effect from the training or whethercross-sectional estimates capture the return onpast training investments. This question is validbecause it is likely that the level of training iscontinual in firms. In other words, firms that investmore in training in one period (t-1) continue to investmore in training in the following period (t). Becauseof delayed training effects, we might measure theeffect of prior training investments incross-sectional estimates. The question of when toexpect the effects of training investment to materi-alise is by no means clear and it would be beneficialto have this matter determined in future research.

If the productivity effects from training lag, thishas implications for the conclusions we can drawon instant wage effects from training. It hasgenerally been accepted that wage increases inconnection with training arise from resultingincreased productivity. The results of the abovestudies imply that wage increases during trainingmight have some other basis. Recent access todata matching employees’ and firms’ characteris-tics will possibly shed more light on this question.

5.1.4. Timing of training investments

The amount of training firms undertake ispossibly affected by the economy. The general


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 298

understanding is that expansionary economicconditions, with firms hiring new employees, alsoare associated with an upsurge in firm-sponsoredtraining. The results of Dearden et al. (2000) andBartel (1994) imply the opposite – that firms trainwhen production is low (the pit stop theory). Onereason why favourable economic conditions donot produce more company training may be thatonly a portion of all company training is gearedtowards new employees. For instance, only 18 %of all training provided by publicly held compa-nies in Sweden was introductory training fornewly hired employees (12). Since firms train whenthey have slack time, we also typically underesti-mate the impact of training on productivity incross-sectional analyses (13).

Another important finding is that the timing oftraining appears not to depend on tangible invest-ments (Barrett and O’Connell, 1999). The finding thatinvestments in training and investments in tangibleassets are only weakly correlated suggest thattangible investments do not cause the trainingeffects observed at company level (industry level).Apart from tangible investments and training, theCranet survey results also indicate that the amountof training provided by firms is not dependent onprevious profitability. Both results indicate that thedecision on how much training to provide has little todo with whether the firm has done well in the past orwhether the firm increases its tangible investments.These findings have importance for what conclu-sions that can be drawn from statistical models,especially cross-sectional regression estimates.

5.2. The effects of education,skills and competences oncompany performance

The effects of education or skills/competence oncompany performance are generally more difficultto establish, as these factors are accumulatedmeasures of human capital stock. Compared withcompany training that normally varies from year toyear, educational levels are much more constant.Because of this we are typically restricted to leveldata (with some exceptions) in analysing the impact

that human capital stock might have on companyperformance. Educational or skills levels arenormally included as control variables in mostimpact research but less frequently used as a mainvariable (at least in firm-based research). Still, this ispossibly one of the more interesting areas inexplaining company performance as the studiesincluded in this review indicate that education andskills are important factors in understanding differ-ences in performance among firms.

There is research to connect the effect ofeducational level or skills/competence level withproductivity, with positive association in the workof Black and Lynch (1996) and NUTEK (2000).Indications that skills are an important factor inproductivity are presented in Carr (1992) andMason et al. (1992). A significant paper notrestricted to cross-sectional data is the study byGunnarsson et al. (2001) examining educationallevel and productivity growth over a ten-yearperiod. Their findings suggest that the increase inthe educational level between 1986 and 1995explains a large part of the IT-related productivitygrowth. Their results also suggest that a marginalskill upgrading has the same effect acrossdifferent levels of education. Gunnarsson et al.conclude by stating that ‘measures to promoteincreased use of IT should be followed up bymeasures promoting skill upgrading. Our resultsactually show that, in general, upgrading skills ata given level of IT (i.e. share of computers in totalcapital) has a much stronger growth-enhancingeffect than increasing IT investments at a givenhuman capital structure.’ (p. 44).

The findings of Leiponen (1996b) indicate thatinnovative firms have a more educated workforceand that innovative firms are more dependent oneducational competence in generating profit(Leiponen, 1996a). The study by Michie andSheehan (1999) also suggests that skill shortageis a severe obstacle to innovation. Similarly, thefindings of Romijn and Albaladejo (2000) in SMEspropose that the owners’ technical education andtheir prior working experiences, in addition to thetechnical skills of the workforce, have a signifi-cant effect on innovative capability.

Taken together, the results indicate that we havecomplementarities between different types of


(12) IPF – Institute of personnel and corporate development. The human capital survey 2002. Uppsala: IPF, Uppsala University(please contact the authors for results).

(13) However, as noted in the previous section this conclusion depends to some extent on when the training effects materialises.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 299

education (Leiponen, 1996a) and between educa-tion and IT investments (Gunnarsson et al., 2001)that generate significant synergies or externalities.The evidence that the level of education or skills isrelated to innovation and productivity might not betoo surprising since education and skills are gener-ally considered as associated with more complexjobs and increased flexibility. What is moresurprising is that we start to see studies relatingeducation and skills to profitability.

The results of Leiponen (1996a) show thateducational level is associated with profitability(net profit margin). Leiponen uses panel data anda two-stage procedure to handle problems ofendogeneity, providing results that are morerobust than ordinary cross-sectional estimates,especially on arguments that profitability causesfirms to hire more highly educated personnel. Thestudy by NUTEK (2000) shows that the proportionof higher educated employees is significantlyassociated with both productivity (value added)and profitability (revenues to cost ratio). Becauseeducational level is associated with both produc-tivity and profitability it gives us a more solidbasis for inferring that higher education generateshigher productivity and that firms are able tocapture some of the benefits.

The idea that skills in the form of programmingcompetence are associated with how much theindividual produces in net contribution (profit) tothe firm is presented in Hansson (2001). Thisinvestigation is based on a single firm and it isthus difficult to draw any far-reaching conclu-sions. Nevertheless, because the examination isbased on differences in employees’ net contribu-tion this study avoids the argument that onlyprofitable firms can afford to hire more skilledworkers. Similarly, the results of the Cranetsurvey suggest that the more profitable firms andfirms with better stock market performance intheir respective industry sector also have morehighly educated personnel than firms performingbelow average in their respective industry sector.

From the above, it is possible to speculate onthe degree to which firms are able to capturereturns on human capital investments thatnormally are considered to belong to the indi-vidual. Because prior education is a function of

the individual it is assumed that the individualaccrues the benefit of these investments throughhigher wages. However, indications that investorsand beneficiaries are not always the same arepresented in Groot (1999) who points out the weakconnection between those who contribute totraining investment and those who benefit from it.

That firms are able to extract higher profitabilityfrom more highly skilled or educated workers isan argument put forward by those proposing thatwage compression is a major reason for firms toinvest in general skills (e.g. Acemoglu andPischke, 1998, 1999a). The basic reasoning isthat individuals are not paid their marginalproduct and that firms are able to extract higherprofits from more skilled workers who are notpaid what they are worth for the company. Wagecompression is not only a European phenomenonbut can also be found in the US (Bewley, 1998).The findings of Bewley suggest that the internalequity (fairness and moral) in firms’ pay structurerestrains managers from paying employees thefull value of their contribution. Consequently, highperforming employees are more valuable to thefirm. Bewley takes this reasoning one step furtherby arguing that low performing employees areseldom fired and, if they are, it is for grossmisconduct rather than for under perfor-mance (14). Wage compression is one of severalrecent theories that try to reconcile the empiricalfindings that firms both invest in, and extractprofit from, general human capital. The nextsection provides some further explanation.

5.3. How firms profit fromgeneral human capital

Explanations of why firms invest in, and are able toprofit from, marketable human capital are abundant.Based on differences in bargaining power, Glick andFeuer (1984) propose that general training is superiorto straight money payment as an insurance againstpersonnel turnover and that firms should invest ingeneral training to safeguard joint investments inspecific training. In the shared investment model ofLoewenstein and Spletzer (1998), the employer


(14) However, low performers are the first to be laid off when firms are reducing their workforce and the pay for low-performingemployees is often allowed to fall behind the rate of inflation. Bewley interviewed over 270 executives and managers of USbased firms about wages and layoffs.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 300

shares general training investment with theemployee as a consequence of the employer’sinability to commit credibly to future wages. Theemployer, instead, commits to a minimum guaran-teed wage, shares the investment in general trainingand realises the returns on the training if the minimumwage guarantee is binding. Autor (2001) proposes amodel in which firms offer general training to induceself-selection and perform screening of workerability. In this model general training and ability arecomplementary and it is assumed that more ableworkers self-select to receive general training to agreater extent than low ability workers. In the modelof Acemoglu and Pischke (1999a) firm-financedgeneral training is a result of compressed wagestructure. Wage compression makes employersmore willing to invest in general training as firmsextract higher profits from more skilled workers withincreased human capital.

Another response to rent extraction fromgeneral human capital investments is thatmobility thresholds reduce the ability of the indi-vidual to capitalise on such investments. Argu-ments against turnover (mobility) include:(a) the loss of firm-specific investments for the

individual when changing work (Glick andFeuer, 1984). Or that a recruiting (raiding) firmneeds to make additional investments infirm-specific knowledge;

(b) firms use back-loaded compensation schemesthat induce costs for individuals who changeemployer (Salop J. and Salop S., 1976).Back-loaded compensations are typicallydetected or defined as increasing wages withseniority over and above productivityincreases;

(c) workers have incomplete information aboutpay elsewhere (Polachek and Robst, 1998;Bewley, 1998);

(d) individuals are constrained by liquidity oraversion to risks, forcing firms to carry theseinvestments (Bishop, 1994);

(e) firms have superior information about theprofitability (payoff) of training investments(Green and Kahn, 1983).

If it is possible for firms to redistribute invest-ment risk through capital markets, this mightcause employers of larger firms to be more willingto invest in general human capital. However, Katz

and Ziderman (1990) (15), have attracted attentionwith the argument that information asymmetrybetween the training firm and a recruiting (raiding)firm about training received reduces the potentialbenefits that a worker with general training canobtain by moving to another firm. Consequently,information asymmetries render general trainingspecific in the sense that the investment is notobservable (verifiable) to other firms.

Less attention has been given to findings thatboth employers and employees benefit from theseinvestments and both parties would be worse offif they did not take place. It is tempting to assumethat employers increase wages for individualsreceiving marketable training sufficiently to offsetthe increased probability of turnover. That boththe employer and the employee benefit from theseinvestments also implies that individualsemployed in firms that provide training receivehigher wages in the long run compared withemployees of firms that do not provide training.Higher growth in wages provides a strong incen-tive to stay with an employer who continuouslyupgrades human capital instead of possiblymoving to a new employer with an unknownhuman capital investment strategy.

Also, the employer-employee relationship iscomplex and it might be myopic to focus only onmonetary gains. Part of the rent extractionconsideration might be that these investmentsrepresent good working conditions and that theemployers are committed to their employees. Inthis sense, training, no matter how general, is ameasure of employer commitment, which is likelyto reduce the probability (threat) of changingemployers.

5.4. Training and HRM practices

The basic question of whether the combinedeffect of human resource management (HRM)practices produces good performance or whethercertain practices, such as employee develop-ment, generate effects on company performanceis not easy to answer. In general, there isevidence that training has a greater impact whenundertaken in connection with supporting HRMpractices, in particular existence of a formal


(15) Schlicht (1996) has also covered asymmetric information and its impact on a firm’s training decisions.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 301

training strategy, written commitments to training,methods for analysing training needs, linkingtraining and strategic objectives (e.g. Maglen andHopkins, 1999; Blandy et al., 2000; Baldwin andJohnson, 1996). The Cranet data also indicatesthat support functions are important to training,with the variables on training needs and writtentraining policy significantly associated withtraining provision and industry-adjusted perfor-mance measurements (profitability and stockmarket return) (16).

Whether training has an additional effect overand above high performance work practices isdifficult to determine. In high performance worksystem (practice) literature, training is generallyincorporated as a factor in the larger construct ofHPWS (with some exceptions). The bundling ofdifferent human resource practices is typicallybased upon factor analysis or analysis of theinternal consistency of the total HPWS measure-ment. Because training is generally part of alarger construct it is difficult to find studies thatmeasure the additional effect of each individualvariable. However, some studies account for theimpact of different variables. The study byLaursen and Foss (2000) highlighted training andperformance-related pay as important factors forinnovation. However, the combined effect of allnine factors proved highly significant, indicatingthe existence of complementary HRM practices.

Other studies also concluded that internallyconsistent or congruent HRM practices are betterpredictors of performance than individual prac-tices, for example MacDuffie (1995) and Arthur(1994). These findings do not mean that training byitself does not have a predictive ability for perfor-mance, but only that the whole set of practicescombined were more informative. The study byDelaney and Huselid (1996) illustrates the problemof focusing solely on the effect of total HRM prac-tices since the training measurements in their studyconstantly appeared to be related to performancemeasurements, even with a significant volume ofvariables capturing other HRM practices.

High performance work system (practice) litera-ture is normally restricted to level data, which makesit difficult to establish the direction of the relation-ship. There are some exceptions to this rule. The

studies by Barrett and O’Connell (1999), d’Arcimoles(1997), Ichniowski et al. (1995), and Bartel (1994) allhave measurements of HRM practices and trainingover time. The results do not agree on whethertraining or HRM generates the effects on companyperformance. The study by Barrett and O’Connellsuggests that training causes productivity effectswhereas introduction of new personnel policies didnot show any significant impact on productivity. Theresults of Bartel propose that training programmesgenerate considerable productivity effects, inexcess of changes in personnel policies. The mainresults of d’Arcimoles indicate that trainingproduces substantial effects on both productivityand profitability. The study by d’Arcimoles includedcontrols for working and social climate at the firmsinvestigated. A contradictory view is presented inIchniowski et al. where innovative HRM practiceshave a large effect on productivity while individualemployment practices had little or no effect. Thisresult suggests that training by itself is not enough.

The common denominator of these studies isthat we can, with reasonable confidence, main-tain that a cause and effect relationship existsbetween the variables studied (training and HRM)and performance measurements. The line ofresearch or the tradition within which the studywas performed might explain the somewhatcontradictory results. For instance, researchers inlabour economics are possibly more used tomodelling training compared with HRM variables,the converse being true for researchers in theHRM tradition. In conclusion, it seems appro-priate to reason that both training and other HRMpractices are important factors in explaining whysome firms do better than others.

5.5. Innovation, technologicalchange and training

There is a multiple relationship between innova-tion behaviour, innovation performance andtraining investment. The consequences of tech-nological change and the introduction of processand product innovation, and the relationship withtraining investment, are important issues. Theinternal organisation and human resources were


(16) With one exception, the variable ‘written training policy’ was not significantly associated with the amount of wage bills spenton training.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 302

neglected for a long time in traditional,economic-oriented innovation literature. In morerecent literature attention to the role of HRM andits impact has increased. In the knowledge-basedeconomy, training investment and HRM practicesare prerequisites of innovation and are necessaryto realise the productivity potential of new infor-mation or advanced manufacturing technologies.In order to use the potential of new technologies,complementary investments in training arecrucial. There is a growing awareness of the roleof internal or organisational factors that mediatethe relationship between innovation and companyperformance. Some authors, such as Laursenand Foss (2000), stress the importance of thecomplementarities between technology andlearning.

A similar view is presented by Baldwin andJohnson (1996). Their findings suggest that moreinnovative firms place greater emphasis on humanresources. Innovation requires a human-resourcestrategy that stresses training. They found statis-tical evidence that more innovative firms offeredformal and informal training more frequently, thatthe training was more often continuous in char-acter, and that these firms had a greater tendencyto innovative compensation packages. The studyby Michie and Sheehan (1999) also stresses theimportance of HRM in generating an innovativeenvironment.

That both training and skills are importantdeterminants of innovative capability is offered inRomijn and Albaladejo (2000). The owners’ tech-nical education and their prior working experi-ence, the technical skills of the workforce, and theamount of training provided proved to be impor-tant aspects of innovation capability. Their find-ings also suggest that interaction with researchinstitutions and a close link to nearby traininginstitutions enhance innovation capacity. As notedearlier, the composition of the workforce is impor-tant for innovating firms (Leiponen, 1996a). Thefindings of Leiponen indicate that innovative firmsare more dependent on educational competencein generating profit.

Training associated with the introduction ofnew technologies or new work practices is likelyto have high productivity effects. Blandy et al.(2000) found evidence for this relationship withinthe case studies they carried out in addition tothe questionnaire survey. That IT generates a

substantial amount of training is clear. In the Insti-tute of personnel and corporate development’s(IPF, Uppsala University) human capital survey2002 (see footnote 12) about 41 % of allcompany-provided training was considered to berelated to information technology.

In addition, innovation is frequently used as aperformance measurement for companies. Inno-vation itself is related to various financial returnsbut there is no definite association between inno-vation performance and the financial performanceof firms. However, there is broad empiricalevidence that innovation is associated with thegrowth of firms and that, in specific industries,more innovative firms yield higher financialreturns. The results of the Cranet survey alsoindicate a connection between innovation and anumber of personnel related variables. The top(10 %) performing firms in their sector had moretraining, trained more of their employees, hadgreater supporting HRM policies for employeedevelopment, and employed more graduatesthan low performing firms.

5.6. Specifics of SMEs

Small and medium sized enterprises (SMEs) areusually defined as firms with less than250 employees. They represent more than 95 %of all firms and employ around 65 % of allemployees in the EU. Given their importance tothe economy, there are surprisingly few empiricalstudies dealing explicitly with the impact oftraining on SME performance. Despite greatheterogeneity across Europe and the diversity ofindustries, most surveys in different Europeancountries show that SMEs make fewer invest-ments in vocational training and do not useformalised forms of training. While SMEsgenerate many jobs and attract young people,most do not provide them with skills and improvetheir long-term employability. Apprenticeship isan important form of SME initial vocationaltraining of the workforce. In general, most SMEshave difficulty in appropriating codified forms oftraining (Trouvé, 2001), which, in turn, leads toadditional methodological problems in measuringinformal training within SMEs. The organisationalstructure of SMEs is an important reason for thelack of vocational training by them.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 303

The survey on continuing vocational training inenterprises, carried out by the EU in 12 MemberStates, shows that larger firms offer continuingvocational training more often (EuropeanCommission, 1999). Small firms, especially thosewith 10-49 employees, seldom offer continuingvocational training; only half the firms of this sizeprovide it, compared with more than 90 % offirms with more than 250 employees. Corre-spondingly, expenditure on vocational training inSMEs is lower than in larger firms, though this ishighly dependent on country, size and sector.Similarly, SMEs rarely have a clear humanresource strategy, training plan, or advancedHRM practices.

The importance of human capital investment isalso of interest in relation to other forms of intan-gible and tangible investments in explainingcompetitiveness and performance. Flexibility,entrepreneurship, close relations with partnersand customers, motivation of the workforce, andthe realisation of niche strategies are importantSME strengths in comparison with larger firms(for more details see Descy and Tessaring, 2001).In addition, there is empirical evidence thatsmaller firms use their R&D investments moreefficiently. Nevertheless, there are no theoreticalor empirical facts stating that this holds fortraining investments. Yet, it is probable that addi-tional training investment could yield higherreturns in SMEs than in larger firms given the lowlevel of training of the workforce.

It is difficult to deal with return on traininginvestments in more detail, leaving us with thequestion why SMEs do not invest more intraining. We conclude by stating that there isnothing in the current review suggesting thatskills or training in SMEs have any less impact oncompany performance. The study by Leitner(2001) indicates that training is one of the fewvariables associated with company earnings. Thefindings of Bosma et al. (2002) suggest thatcertain skills, such as experience in businessownership and industry experience, contribute tothe success of start-up companies. The study byRomijn and Albaladejo (2000) indicates that bothskills and training are associated with the innova-tion capability of SMEs.

5.7. The influence of labourmarkets and social partners

Research that connects education, training, orskills/competence with different labour marketconditions and the impact that they may have ontraining strategies and company performance isnot very common. We have not seen any paperexamining the effect of different labour marketconditions on, for instance, the ability of firms toprofit from training investments. However, somegeneral remarks can be made on existing trainingliterature and the influence of labour markets. Ithas been argued that differences between the USlabour market and labour market in Germany another European countries in regard to mobilityand wage structure would indicate that traininginvestment is more likely in Europe (e.g.Acemoglu and Pischke, 1998; 1999a). Thesearguments have been supported by differentempirical reasoning based mainly upon observa-tions that firms invest in general training, such asin the German apprenticeship system (17). Never-theless, there are no clear empirical results indi-cating that less efficient labour markets maketraining investments more profitable for firms,though mobility has the ability to reduce thispotential. Similarly, it is likely that more equality inwages (compressed wage structure) has a posi-tive influence on the ability to extract profits fromtraining investments (and prior education). Still,we have not been able to locate research veri-fying differences in return on company trainingbetween different labour market systems.

There are some indications that differences inproductivity between countries can be explainedby differences in national education and trainingsystems. For instance, Mason et al. (1992) foundsubstantially higher proportions of vocationallyqualified personnel on all job levels in the Nether-lands compared with Britain. They also argue thatthe higher average level of skills and knowledgein the Dutch workforce contributes to higherproductivity through better maintenance ofmachinery, greater consistency of product quality,greater workforce flexibility, and less learningtime on new jobs. Similarly, Carr (1992) found that


(17) Other features of the German apprenticeship system are that firms screen potential employees and that the apprentice systemprovides a better matching of employee and employer. According to Euwals and Winkelmann (2001) apprentices staying withintheir firm after graduation have higher wages and longer first job duration than apprentices leaving the training firm.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 304

Britain had substantially lower productivity(measured by sales per employee) than, forinstance, Germany. Carr maintains that the highlyeducated and trained German workforce, thehigh qualification of the foremen, and extensivevocational training by firms, explains the produc-tivity advantage of German firms. Carr alsoascribed improved productivity growth in Britainduring the 1980s to increased labour flexibilityduring this period. However, it is important tonote that neither study is based on any (signifi-cant) statistical test but more on reasoning basedupon gathered information.

A somewhat contradictory argument on theeffect of increased labour flexibility is put forwardin Michie and Sheehan (1999). Their findingssuggest that strategies to increase employmentflexibility by short-term contracts, weakeningtrade unions, etc., do not enhance innovationperformance. This study leads us to the effectsthat social partners may have on trainingoutcomes. Again we have not come across anystudy that connects the influence of social part-ners on the decision to train employees and whateffect this may have on company performance.The role of the social partners is not explicitlytreated in any of the analysed studies.

Nevertheless, there are some general observa-tions that can be made. The role of social partnersis strongly connected with the different nationalfunding systems for education and training. Asargued by Mason et al. (1992) and Carr (1992),education and vocational training may, in turn, havean effect on productivity. Other observations thatcan be made are, for instance, that higher wagesand lower mobility in unionised establishmentstypically promote training (Booth et al., 1999).

The Cranet survey results also cast some lighton the question of whether different labourmarket systems or the social partners have anyeffect on the decision to train. The trainingregression is based on about 1 400 companyresponses mainly from European countries. The

results indicated that unions might have a nega-tive impact on the amount of training providedbut a positive (not significant) impact on thenumber of employees being trained in a year. Asmentioned earlier, the negative associationbetween unionisation and the amount invested intraining is probably caused by our inability tocontrol for industry differences in the trainingregression. The indication that unions mightaffect the training decision by providing moreemployees with company training is more inter-esting. However, this is not statistically verified inthe analysis of the Cranet survey. The correlationanalysis also indicated a lower personnelturnover in more unionised establishments, whichis in line with the findings of Booth et al. (1999).

Our measurement of the degree of internallabour market appears to confirm that these typesof structures do not promote training and learning.In spite of the indication of lower personnelturnover in firms with more internal promotion,training is less provided in these types of estab-lishments. The result of the training regressionalso revealed that personnel turnover itself doesnot determine training. This finding might be inter-preted as an indication that turnover does notreduce the incentive to train employees or thatpersonnel turnover induces training by forcingfirms to train newly hired employees. We also seelarge differences in turnover among different Euro-pean countries but this seems to affect theamount of training provided to a minor degree.

It is important to note that these findings onlyconcern the provision of training and not thepotential effect that different labour marketssystems may have on training outcomes. This isclearly an area that needs much additional work.We conclude this section by merely stating thatthere are indications that labour market condi-tions and the role of, for instance, unions mayhave an effect on company training outcomes bytheir effect on mobility, wages, and the incentivesto train and be trained.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 305

A growing number of papers focus on the effects ofhuman capital and human capital investments oncompany performance. Previously, this subject waslargely disconnected from company based impactresearch as human capital (investments) are notowned or controlled by firms. However, morestudies are needed to understand how education,training and skills/competence affect firms, in aneffort to comprehend fully what generates profitsand growth. The main findings of this review of liter-ature on the impact of education, training, andskills/competence may be summarised as follows.(a) The type of training firms provide to their

employees is not so much a question ofwhether the training is general or specific butpossibly more a question of what is needed tostay ahead of competitors. A growing body ofliterature suggests that firms are financing alltypes of training (general as well as specific).

(b) More recent research findings also suggestthat investments in training generate substan-tial gains for firms irrespective of whether thetraining is useful to other firms. The evidencethat employers profit from training investmentcomes from different countries includingIreland, Britain, the Netherlands, Sweden,France, as well as the US. In most of thesestudies we can, with reasonable confidence,maintain that training generates performanceeffects and not the other way around.

(c) The effects of education and skills/compe-tence on productivity and innovation aregenerally positive and significant. That wealso start to see studies that connect educa-tion and skills with profitability might besomewhat more unexpected. That firmsextract profit from prior education is alsorelated to the ability of firms to capturereturns from general training investments.

(d) Employee development practices, such astraining policies and methods for analysingtraining needs, appear to be importantelements in explaining the provision oftraining and training outcomes. Similarly,innovative or advanced HRM practices aregenerally associated with firm performance.

(e) Innovation and IT not only cause firms toinvest more in training but are also highlydependent on education, skills and training ingenerating profit from these investments.Other findings suggest that training, togetherwith comprehensive HRM practices, is closelyrelated to firms’ innovative capacity.

(f) The lack of studies connecting SMEs, labourmarket characteristics, and social partnerswith training strategies and company perfor-mance measurements such as productivity orprofitability makes it difficult to draw anyconclusions. This research gap provides animportant incentive to investigate these typesof question more thoroughly in the future.

In conclusion, research concerning the effects ofeducation and training on firm performance is gath-ering momentum. Much more research on thissubject will appear in the near future. However, thefindings thus far raise questions and issues that wewill discuss in more detail in the next section.

6.1. Implication of firm financedgeneral human capitalinvestments

That firms invest in general training implies that theremight be a market failure in vocational training.Several authors, (Acemoglu and Pischke, 1999a;Bassi and Ludwig, 2000; Booth et al., 1999), see inrejection of Becker’s theory on company trainingevidence of under-investment in vocational training.In a perfect labour market, individuals pay for theirgeneral training by accepting a lower wage than theirproductivity during the training. The individual thencaptures all benefits from the training by an increasedwage after the training. In this case it is likely that theprovision of training is close to the social optimal levelas the investment decision is made by the individual.Acemoglu and Pischke (1999a) noted that in aperfectly competitive labour market, insufficientinvestments in skills could only arise becauseworkers are severely credit constrained. But in thiscase, the solution may be better loan markets ratherthan direct subsidies to training.

6. Summary and discussion

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 306

When the firm makes the training decision it ismost likely that too little training is provided. Or,as Booth et al. (1999) put it, when training isgeneral to an industry, firms will choose a subop-timal level of such training, since they realise thatworkers would take these skills with them whenthey leave for other firms in the same industry.However, human capital is not lost to society so amarket failure arises. The findings that firms areactive in, and profit from, general human capitalinvestments might thus warrant governmentregulations and subsidies for training, as thesefindings suggest under-investment in vocationaltraining. However, such a definite statement isnot warranted by the present state of research oncompany training. More research is required to becertain that company-based decision-makingconcerning the provision of general training leadsto fewer investments in training.

That firms profit from all types of traininginvestments indicates that we have underesti-mated the benefits from company training.Because most (formal) training is general innature, research on the impact of training hasbeen largely focused on the effects for the indi-vidual (wages); the benefits for employers havebeen considered to a lesser extent. As noted byDearden et al. (2000) by only examining the effectof education and training on wages, economistsmay have underestimated the importance oftraining for modern economies. The authors’conclude that it is time to start casting the netwider than wages in seeking the impact oftraining on corporate and national economicperformance.

It appears that it is not only researchers who haveunderestimated the effects of training investmentsbut, perhaps more severely, also the owners of thecompanies and investors. The findings of Bassi et al.(2001) indicate that firms investing more in traininghave a better stock market performance. This resultsuggests that investors are not aware of the invest-ments in training and that this type of information hasrelevance for investors. The lack of information abouttraining investment and the consequent benefitsleads to a suboptimal allocation of resources totraining in the capital market. It is conceivable thatthe lack of information about company training leads

to under-investment in profitable training projects(training projects with a positive net present value).So another implication of the evidence that trainingpredicts stock market returns is that investorspossibly need more information about these invest-ments in order to make better decisions about whereto allocate their financial resources. The issue aboutinformation to the capital markets will be discussedin more detail in the next section.

The problem of allocating enough resources is,however, complex as information asymmetries isone of the more prominent reasons given for theexistence of firm financed general human capitalinvestments. According to Katz and Ziderman(1990) information asymmetries between the firmcarrying out the training and other firms makefirms more willing to invest in general training.This is because the lack of information about thetraining investment reduces the potential benefitsthat a worker with general training can obtain bymoving to another firm. If Katz and Ziderman areright, providing more information about traininginvestments to capital markets might have anegative effect on the provision of training.

However, another information-based argumentimplies the opposite effect. Acemoglu and Pischke(1999b) argue that firms train their employeesbecause they have sufficient monopsony powerover their employees due to information asymme-tries. While asymmetric information encouragesfirms to invest in training, it reduces the workers’incentive to invest in their skills, as most of thereturns on training will be appropriated by the firm.This means, in contrast to Katz and Zidermann’sargument, that asymmetric information in labourmarkets might undermine the existence of trainingby not giving enough incentives to workers. Moreinformation about the training investment in thiscase leads to more investment in training (18).

6.2. Information on training andintangibles to capitalmarkets

We proposed earlier that investors possibly needmore information about training investments.


(18) The divergent views of the outcome of more information about training investments between Katz and Ziderman (1990) andAcemoglu and Pischke (1999b) appear largely a consequence of different opinions about how inefficient the labour marketreally is.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 307

Johanson (2003) proposes that capital marketactors are hesitant regarding recent knowledgegathered from research on the importance ofhuman capital investments because of thefollowing five reasons.

First, capital market actors might be ambiva-lent because they fail to understand the impor-tance of a certain human capital investment. Theyprobably are not aware of recent research on theprofitability of human capital investments. Theylack the necessary understanding of the potentialof human capital investments in a specific firm.They have little or no appreciation of how humancapital contributes to the value creation process.This inability to comprehend the meaning ofhuman capital could be conceptualised as a‘knowledge’ problem.

Second, even if capital market actors dounderstand the connection between indicatorsand the vision of the firm, they are probably hesi-tant about human capital investments becausethey do not know if they can rely on the indica-tors. Do indicators of human capital transformadequate information? Are they valid? And arethe methods of measurement reliable? Theseissues of validity and reliability could be referredto as the ‘uncertainty’ problem.

Third, this reluctance might be connected tothe lack of ownership of intangibles related topeople. For example, because an organisationcannot own individual competence, the risk ofloosing this competence might be overly exag-gerated. This condition could be known as theproblem of ‘ownership’.

A fourth problem could be that capital marketactors are ultimately hesitant and indecisivebecause they do not know if the measurementsmatter in the management control processes ofthe firm. Is information taken care of? Doesmanagement take the necessary action on data,i.e. a ‘management’ problem?

The final problem suggested by Johanson(2003) concerns the ‘mentality’ of different capitalmarket actors, who are neither used to consid-ering human capital investments as importantfactors that drive firm performance nor encour-aged to do so.

These five barriers are probably relevant notonly for capital market actors but also forcompany management and policy makers.

There is a need to develop a new way of

measuring and reporting internally as well asexternally on training investments with the poten-tial to increase understanding of the financialimpact of education and training. Our proposal atthis point relates to the debate about humanresource costing and accounting, intellectualcapital (IC), balanced scorecard, etc. (herereferred to as the IC-movement).

During the last decade numerous initiatives havebeen taken to encourage the development of a newglobal framework for the measurement, manage-ment and reporting on intangibles. Major initiativeshave been taken by the OECD and the EuropeanCommission. In 1998 the Commission decided tosupport a six-nation (Denmark, Finland, France,Norway, Spain and Sweden) research projectnamed Measuring and reporting intangibles tounderstand and improve innovation management(Meritum, 2002). The Meritum work, which wasperformed between the years 1998 and 2001 wasorganised in four different activities:(a) definition and classification of concepts, e.g.

intangibles and intellectual capital;(b) investigation of how management control of

intangibles was performed at company level;(c) capital market implications of the poor infor-

mation from firms on intangibles;(d) development of guidelines for the reporting

and management of intangibles.The guideline was subject to a Delphi test at

the end of the project.The Meritum work is presently subject to a

follow-up project E*KNOW-NET which is alsofinanced by the European Commission. The aimof the follow-up project is to spread the findingsfrom the Meritum work, to improve guidelines andto propose a research and education agendaregarding intellectual capital.

The Meritum and E*KNOW-NET works are basedupon the belief that firms are facing a major trans-formation in the value creation process. Intangibles,including, more specifically, knowledge, areincreasingly becoming the major driver of firms.These changes pose a great challenge to firmsbecause intangible resources are not easily identi-fied, not measured, and not reported internally orexternally. Another basic assumption is that there isa need to develop a common framework, whichinvolves definitions and classifications of intangi-bles and a guideline for measuring, managing, andreporting intangibles.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 308

The mismeasurement of knowledge may leadto inefficient allocation of financial and humanresources. As the European Commission states inits report Towards a European research area, ‘theEuropean financial market has not yet sufficientlydiscovered the economic value to investment inknowledge’ (European Commission, 2000, p. 7).This is partly due to the fact that the informationprovided by companies to the financial markets isprimarily based on traditional tangible invest-ments, whereas value increasingly relies oninvestment in intangibles.

Efforts are needed both to provide informationon how knowledge is produced and accumu-lated, and on the way knowledge can be trans-formed into profits. The generalisation of goodpractice in the management of intangibles alsoneeds to be encouraged. New common proce-dures, documents, rules, etc., should be providedin order to improve the informative capacity of thefirm’s financial statements. This is precisely themain purpose of the Guidelines for managing andreporting on intangibles (Meritum, 2002), here-after referred to as Guideline.

The Guideline document attempts to supportfirms in the process of developing their ability toidentify, manage and value their intangible assets.To start with, a set of definitions on intangibleresources and intangible activities is provided; itis integrated with a classification used for theproposed intangible management system (humancapital, structural capital and relational capital).Based on the experience of best practice firms, amodel for the measurement and management ofintangibles is suggested, which covers threedifferent phases: identification, measurement andmonitoring of intangibles.

The Guideline also contains information on thestructure and contents of an external documentcalled the Intellectual capital statements. Threedifferent parts are considered for inclusion in thatdocument:(a) vision of the firm,(b) a summary of intangible resources and activi-

ties, and(c) a system of indicators.

To overcome the barriers proposed byJohanson (2003) it is important, as well as chal-lenging, to develop understandable indicators onissues related to training investments. The indica-tors have to be measurable and valid. Because

the very idea behind the development of trainingindicators is to increase the understanding of theimportance of knowledge, the indicators have tobe clearly related to the vision of the firm or thevalue creation process. It is probable that thisnew kind of standardised indicator also needs tobe subject to independent audit.

6.3. Strength and weakness ofdata and methods

The lack of information on training investmentsalso poses a problem for researchers, as the datahas to be gathered from different sources.Depending on how one defines training, the esti-mate of working time spent on training variesconsiderably. The IPF at the Uppsala Universityregularly carries out surveys of companies listedon the Stockholm stock exchange. The humancapital survey of 2002 included questions onwhat firms defined as company training. Somecompanies report only training conducted outsidethe firm (12 %), some report internal and externaltraining sessions with a defined curriculum(39 %), and others report anything from formaltraining sessions to such informal training aslearning by doing and self studies (45 %).

The lack of a coherent definition of training thatis used and reported consistently by companiesis one of the more important issues for researchon company training. The problem of varyingmeasurements of training is not likely to besolved by defining what is regarded as training indifferent surveys. Firms are unlikely to changetheir data collecting methods regarding trainingfor each new survey. It seems likely that whatcompanies report as training is what they havedata for, no matter what is defined as training ineach specific survey. At best one can expect theprovider of the information to make some profes-sional judgement or correction of their data fordifferent surveys. Some straightforward guide-lines or general agreement among researchersand companies on what to define as trainingseems be warranted.

Apart from a common definition and standard oftraining, another problem with the data concernsagreement on what type of costs should beincluded in training measures. The variety ofmeasurements of training costs in different studies


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 309

and in different databases hinders the opportunityto make comparisons across countries and acrossdifferent studies. A comprehensive measurementof training investments in companies will not onlywork as a foundation for comparison and acrosscountries and studies but also facilitate compar-ison with other types of tangible investments. Ifinvestments in human capital can be comparedwith, and have the same credibility as, tangibleinvestments we would advance understanding ofwhat drives firms and, ultimately, what generateswealth for firms and society.

However, varying views of what is to beconsidered as company training do not mean thatinferences drawn from, for example, currentcross-sectional data are not valid. If there is atrue relationship between training and companyperformance, vague definitions of training typi-cally make estimates less precise and less signif-icant. In other words, as long as it cannot beshown that profitable firms are constantly using abroader definition of training, and also accountfor more training investment, than less profitablefirms, the consequence of vaguely definedtraining measures is a downward bias in theimpact of training. Increased variance due tomeasurement problems (errors) leads normally toless significant results or, in the case of severemeasurement problems, insignificant results.Definition problems might partly explain the lowor insignificant impact of training in some of thestudies reviewed in the present paper.

In the case of studies based on panel data thedefinition problem is less important since we arelargely concerned with changes in variables overtime in this type of investigation. If we follow thesame company at different time intervals thismeans that we cancel out all time-invarianteffects that can bias the results. As long as theunit of analysis (e.g. the company) does notchange its definition over time the differencesbetween how companies measure training is ofless importance. The fact that we can drawstronger conclusions from panel data studies alsomitigates problems with vague definitions.

In general, the weakness of data and the weak-ness of methods used in the reviewed studies donot exaggerate the results but, on the contrary,work against finding positive responses to humancapital investments and thus tend to underestimatetheir impact on company performance.

6.4. Policy implications andfuture research

This review of research has highlighted a numberof questions that possibly need more attention inan effort to establish a common understanding ofhow human capital and human capital invest-ments influence growth and performance onfirms. The general impression of the research thathas been reviewed in this study is encouraging inthat the economic effects of education, skills andtraining can be observed in company data. Thatfirms appear to benefit from all types of training(no matter how general) is an important findingthat also generates a number of other trainingrelated questions and implications. We will high-light some of the implications and research ques-tions we feel should be pursued in the future.(a) What are different aggregated measures of

training actually measuring?We have seen in the Cranet survey that twocommon ways of measuring training (theproportion of employees being trained andpercentage of wage bills spent on training)are determined by quite different factors.Whether this is the case in other samples andwhether it has any meaningful interpretationin regard to company performance would beuseful to determine in future research.

(b) What influence has firm performance on theprovision of training?We have also discussed the timing of training.We have seen a number of studies indicatingthat training is more likely to be carried out intimes of weak productivity; we have also seenin the Cranet survey that how much firmsinvest in training is not conditioned by pastperformance. Clarification of any mutualdependence between training and economicconditions is important as the answer wouldfacilitate interpretation of cross-sectionalresults.

(c) When do the effects of training materialise?Some of the studies reviewed suggest that ittakes some time before the effects of trainingare seen in company performance measure-ments. Whether this is the case is an impor-tant research question as it influences inter-pretations made in cross-sectional estimates(are we measuring the effects of past trainingor the effects of current training efforts?) and


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 310

it also influences the time horizon withinwhich we can expect benefits from training.

(d) Are certain types of training or certain ways ofconducting training more efficient?The current state of research typically usesvery coarse and aggregated measurements.A more precise division of human capital andhuman capital investments might give us abetter understanding of the type of educationand training that generates profit and growthfor firms.

(e) Do firms benefit from employing moreeducated and skilled workers?More research is warranted that connectsskills, competence and education withcompany measurements such as profitabilityand stock market return.

(f) How important are supportive HRM practicesfor generating training and performanceeffects?There is a need to understand how and in whatway HRM practices influence training decisionsand to what extent are they increase efficiencyin training and training outcomes.

(g) More research is needed on the impact oftraining for SMEs, and on the influence ofsocial partners and the labour market.While smaller firms undertake less training,more research is needed to explain thereasons for this. Research is also warrantedon the influence that social partners anddifferent labour market systems might have

on the ability of firms to benefit from traininginvestment. We see quite large differencesbetween different countries in companytraining and an explanation for these differ-ences might be found in the effect that thesetwo factors have on the provision of trainingand training outcomes.

(h) Another important issue is a common stan-dard for defining and measuring training andtraining costs.There is a significant difference betweenasking individuals about training and askingfirms about their training. Gathering informa-tion on training in large organisationsconsumes considerable resources andtraining data is not easily changed in accor-dance with different surveys. A common defi-nition of what to regard as training and acommon definition of costs in training invest-ments would be beneficial not only forresearchers but to compare companies,industries, and countries on company training.

(i) More information for capital markets abouttraining investments appears to be warranted.A way of providing information on how muchfirms invest in training appears to be animportant issue in establishing a more effi-cient allocation of resources to firms withgood investment opportunities. Morecompany-based information on training mightalso lead to a better allocation in the humancapital (labour) market.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 311

GMM General method of moments

HPWS High performance work systems

HRM Human resource management

IC Intellectual capital

IT Information technology

LFS Labour force survey

OLS Ordinary least square


SME Small and medium enterprise

STW School-to-work

TQM Total quality management

List of selected abbreviations

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 312

Training related questions

3:1 a) Approximately what proportion of the annual salaries and wages bill is currently spent ontraining?

——————————————— % 1� don’t knowb) Approximately what proportion of employees have been on internal or external training activi-

ties within the last year?

——————————————— % 1� don’t know

3:3 Do you systematically analyse employee training needs?

1� Yes 2� No 3� Don’t know

3:5 Do you monitor the effectiveness of your training?

1� Yes 2� No 3� Don’t know

3:6 Does your organisation have a policy for the following personnel/human resource managementareas:

Yes, Yes, No Don’twritten unwritten know

C. Training and development �1 �2 �3 �4

Performance related questions

7. If you are a private organisation would you say the gross revenue over the past 3 years hasbeen:A. Well in excess of costs �1B. Sufficient to make a small profit �2C. Enough to break even �3D. Insufficient to cover costs �4E. So low as to produce large losses �5

9. Compared to other organisations in your sector, where would you rate the performance of yourorganisation in relation to the following?

Top 10 % Upper half Lower half Not applicableA. Service quality �1 �2 �3 �4B. Level of productivity �1 �2 �3 �4C. Profitability �1 �2 �3 �4E. Rate of innovation �1 �2 �3 �4F. Stock market performance �1 �2 �3 �4

Annex — Selected Cranet questions

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 313

Questions related to internal job market and unionisation

2:5 How are managerial vacancies generally filled? (Please tick as many as applicable for eachmanagement level).

Senior Middle JuniorManagement Management Management

A. Internally �1 �1 �1B. Recruitment by head hunters/consultancies �1 �1 �1C. Advertise in newspapers �1 �1 �1D. Word of mouth �1 �1 �1

5:1 What proportion of the total number of employees in your organisation are members of a tradeunion?1� 0 % 2� 1-10 % 3� 11-25 % 4� 26-50 %5� 51-75 % 6� 76-100 % 7� Don’t know

Employee related questions

6:3 Please provide the following information about your workforce:

A. Annual staff turnover ____________% turnover per year �1 don’t know

B. Age structure ____________% of employees over 45 years �1 don’t know

C. Absenteeism ____________average days per year �1 don’t know

D. Education structure ____________% of graduates �1 don’t know


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 314

Acemoglu, D.; Pischke, J-S. Why do firms train?Theory and evidence. Quarterly Journal ofEconomics. Cambridge: MIT Press Journals,1998, Vol. 113, Issue 1, p. 79-119.

Acemoglu, D.; Pischke, J-S. Beyond Becker:training in imperfect labour markets. TheEconomic Journal. Oxford: BlackwellPublishing, 1999a, Vol. 109, Issue 453,p. 112-142.

Acemoglu, D.; Pischke, J-S. Certification of trainingand training outcomes. Cambridge: MIT –Massachusetts Institute of Technology, 1999b,p. 99-28 (Working paper).

Arthur, J. Effects of human resource systems onmanufacturing performance and turnover.Academy of Management Journal. BriarcliffManor, NY: AOM – Academy of Management,1994, Vol. 37, Issue 3, p. 670-687.

Autor, D. H. Why do temporary help firms providefree general skills training? Quarterly Journal ofEconomics. Cambridge: MIT Press Journals,2001, Vol. 116, Issue 4, p. 1409-1448.

Azariadis, C.; Stiglitz, J. E. Implicit contracts andfixed price equilibria. Quarterly Journal ofEconomics. Cambridge: MIT Press Journals,1983, Vol. 98, Issue 1, p. 1-22.

Baldwin, J. R.; Johnson, J. Business strategies inmore- and less-innovative firms in Canada.Research Policy. Amsterdam: Elsevier ScienceB.V., 1996, Vol. 25, Issue 5, p. 785-804.

Bardeleben, R.; Beicht, U.; Fehler, K. Betrachtungdes Nutzens betrieblicher Ausbildung. In:Beicht, U.; Bardeleben, R.; Fehler, K. (eds)Betriebliche Kosten und Nutzen der Ausbil-dung. Ergebnisse aus Industrie, Handel undHandwerk. Bielefeld: Bertelsmann,1995.

Barnard, M. E.; Rodgers, R. A. How are internallyoriented HRM policies related to high-perfor-mance work practice? Evidence from Singa-pore. International Journal of Human ResourceManagement. New York: Routledge, 2000,Vol. 11, Issue 6, p. 1017-1046.

Barrett, A. et al. Exploring the returns to continuingvocational training in enterprises: a review ofresearch within and outside of the EuropeanUnion. Luxembourg: Office for Official Publica-

tions of the European Communities, 1998(Cedefop Panorama series, 83).

Barrett, A; O’Connell, P. J. Does training generallywork? The returns to in-company training.Bonn: IZA – Institute for the Study of Labour,1999 (Discussion paper, No 51).

Barrett, A. Economic performance of education andtraining: costs and benefits. In: Descy, P.;Tessaring, M. (eds) Training in Europe: secondreport on vocational training research in Europe2000. Background report. Luxembourg: Officefor Official Publications of the EuropeanCommunities, 2001, Vol. 2, p. 383-404(Cedefop Reference series, 3008).

Barron, J. M.; Berger, M. C.; Black, D. A. How welldo we measure training? Journal of LaborEconomics. Chicago: University of ChicagoPress, 1997, Vol. 15, No 3, part 1, p. 507-528.

Barron, J. M.; Berger, M. C.; Black, D. A. Do workerspay for on-the-job training? Journal of HumanResources. Madison, WI: University ofWisconsin Press, 1999, Vol. 34, No 2,p. 235-252.

Bartel, A. P. Productivity gains from the implemen-tation of employee training programs. IndustrialRelations. Oxford: Blackwell Publishing, 1994,Vol. 33, p. 411-425.

Bartel, A. P. Training, wage growth, and job perfor-mance: Evidence from a company database.Journal of Labor Economics. Chicago: Univer-sity of Chicago Press, 1995, Vol. 13, No 3,p. 401-425.

Bartel, A. P. Measuring the employer’s return oninvestment in training: evidence from the litera-ture. Industrial Relations. Oxford: BlackwellPublishing, 2000, Vol. 39, No 3, p. 502-524.

Bassi, L. J.; van Buren, M. E. The 1998 ASTD stateof the industry report. Training and Develop-ment. Alexandria, VA: ASTD – American Societyfor Training and Development, January 1998,p. 23-43.

Bassi, L. J.; Ludwig, J. School-to-work programs inthe United States: a multi-firm case study oftraining, benefits, and costs. Industrial LaborRelations Review. Ithaca, NY: School of ILR,

References

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 315

Cornell University, 2000, Vol. 53, No 2,p. 219-239.

Bassi, L. J. et al. Human capital investments andfirm performance. Washington: Human CapitalDynamics, 2001 (Working Paper).

Bassi, L. J. et al. Profiting from learning: firm leveleffects of training investments and marketimplications. Singapore Management Review.Singapore: Singapore Institute of Management,2002, Vol. 24, No 3, p. 61-76.

Becker, B. E.; Gerhart, B. The impact of humanresource management on organizationalperformance: progress and prospects.Academy of Management Journal. BriarcliffManor, NY: AOM – Academy of Management,1996, Vol. 39, p. 779-801.

Becker, B. E. et al. HR as a source of shareholdervalue: research and recommendations. HumanResource Management. Alexandria, VA:Society for Human Resource Management,1997, Vol. 36, p. 39-47.

Becker, B. E.; Huselid, M. A. High performance worksystems and firm performance: a synthesis ofresearch and managerial implications.Research in Personnel and Human ResourcesManagement. Greenwich: JAI Press, 1997,Vol. 16, p. 53-101.

Becker, G. S. Investment in human capital: a theo-retical analysis. Journal of Political Economy.Chicago: University of Chicago Press, 1962,Vol. 70, p. 9-49

Becker, G. S. Human capital: a theoretical andempirical analysis with special references toeducation (3rd ed.), Chicago: University ofChicago Press, 1993.

Bellmann, L.; Büchel, F. Betrieblich finanzierte Weit-erbildung und Unternehmenserfolg: EineAnalyse für West- und Ostdeutschland unterbesonderer Berücksichtigung von Selektion-seffekten. Nuremberg: IAB – Institute forEmployment Research, 2000 (Working paper).

Bellmann, L. Vocational training research on thebasis of enterprise surveys: an internationalperspective. In: Descy, P.; Tessaring, M. (eds)Training in Europe: second report on vocationaltraining research in Europe 2000. Backgroundreport. Luxembourg: Office for Official Publica-tions of the European Communities, 2001,Vol. 2, p. 279-312 (Cedefop Reference series,3008).

Bewley, T. F. Why not cut pay? European EconomicReview. Amsterdam: Elsevier Science, 1998,Vol. 42, Issues 3-5, p. 459-490.

Bishop, J. H. The incidence of and payoff toemployer training: a review of the literature withrecommendations for policy. Ithaca, NY:CAHRS – Center for Advanced HumanResource Studies, 1994 (Working paper,94-17).

Black, S.; Lynch, L. Human-capital investments andproductivity. American Economic Review.Nashville: American Economic Association,1996, Vol. 86, No 2, p. 263-267.

Black, S.; Lynch, L. How to compete: the impact ofwork practices and information technology onproductivity. Cambridge, MA: NBER – NationalBureau of Economic Research, 1997 (Workingpaper, w6120).

Blandy, R. et al. Does training pay? Evidence fromAustralian enterprises. Adelaide: NCVER –National Centre for Vocational EducationResearch, 2000.

Booth, A. L.; Francesconi, M.; Zoega, G. Training,rent-sharing and unions. Colchester: ILR –University of Essex, 1999 (Working paper, 99-33).

Booth, A. L.; Zoega, G. Is wage compression anecessary condition for firm-financed generaltraining? A comment on Acemoglu andPischke. Colchester: ILR – University of Essex,1999 (Working paper).

Boselie, P.; Paauwe, J.; Jansen, P. Human resourcemanagement and performance: lessons fromthe Netherlands. International Journal ofHuman Resource Management. New York:Routledge, 2001, Vol. 12, Issue 7, p. 1107-1125.

Bosma, N. et al. The value of human and socialcapital investments for the business perfor-mance of start-ups. Amsterdam: TinbergenInstitute, 2002 (Discussion paper, TI 02-027/3).

Brunello, G. Is training more frequent when wagecompression is higher? Evidence from 11 Euro-pean countries. Munich: Center for EconomicStudies and Institute for Economic Research,2002 (CESifo working paper, No 637).

Büchel, F. The effects of overeducation on produc-tivity in Germany: the firms viewpoint. Bonn: IZA– Institute for the Study of Labour, 2000 (Discus-sion paper, No 216).

Carr, C. Productivity and skills in vehicle componentmanufacturers in Britain, Germany, the USAand Japan. National Institute Economic Review.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 316

London: Sage Publications, 1992, Issue 139,p. 79-87.

d’Arcimoles, C-H. Human resource policies andcompany performance: a quantitativeapproach using longitudinal data. OrganisationStudies. Berlin: Walter de Gruyter Publishers1997, Vol. 18, Issue 5, p. 857-874.

Dearden, L.; Reed, H.; van Reenen, J. Who gainswhen workers train? Training and corporateproductivity in a panel of British industries.London: IFS – Institute for Fiscal Studies, 2000(Working paper, w00/04).

Delaney, J. T.; Huselid, M. A. The impact of humanresource management practices on percep-tions of organizational performance. Academyof Management Journal. Briarcliff Manor, NY:AOM – Academy of Management, 1996, Vol. 39,p. 949-969.

Descy, P.; Tessaring, M. (eds) Training and learningfor competence: second report on vocationaltraining research in Europe. Synthesis report.Luxembourg: Office for Official Publications ofthe European Communities, 2001 (CedefopReference series, 6).

Doucouliagos, C.; Sgro, P. Enterprise return on atraining investment. Adelaide: NCVER –National Centre for Vocational EducationResearch, 2000.

European Commission, DG for Education andCulture. Continuing training in enterprises :facts and figures / prepared by the Centre forTraining Policy Studies. Sheffield: University ofSheffield, 1999.

European Commission. Towards a EuropeanResearch Area. Communication from theCommission to the Council, the EuropeanParliament, the Economic and SocialCommittee and the Committee of the Regions.Luxembourg: Office for Official Publications ofthe European Communities, 2000.

Euwals, R.; Winkelmann, R. Why do firms train?Empirical evidence on the first labour marketoutcomes of graduate apprentices. Bonn: IZA–Institute for the Study of Labour, 2001 (Discus-sion paper, No 319).

Glick, H. A.; Feuer, M. J. Employer-sponsoredtraining and the governance of specific humancapital investments. Quarterly Review ofEconomics and Business. Champaign, IL:BEBR – Bureau of Economic and BusinessResearch, 1984, Vol. 24, Issue 2, p. 91-103.

Goux, D.; Maurin, É. Returns to firm-providedtraining: evidence from French worker-firmmatched data. Labour Economics. Amsterdam:Elsevier Science, 2000, Vol. 7, Issue 1, p. 1-19.

Green, J.; Khan, C. M. Wage-employmentcontracts. Quarterly Journal of Economics.Cambridge: MIT Press Journals, 1983, Vol. 98,p. 173-187.

Green, F. et al. The impact of training on labourmobility: individual and firm-level evidence fromBritain. British Journal of Industrial Relations.Oxford: Blackwell Publishing, 2000. Vol. 38,Issue 2, p. 261-275.

Groot, W. Productivity effects of enterprise-relatedtraining. Applied Economic Letters. New York:Routledge, 1999, Vol. 6, No 6, p. 369-371.

Gunnarsson, G.; Mellander, E.; Savvidou, E. Ishuman capital the key to the IT productivityparadox? Stockholm: IUI – Research Institute ofIndustrial Economics, 2001 (Working paper,No 551).

Hagen, A. et al. Ressursbruk til etter- og videreut-danning i norsk arbeidsliv [Resource allocationin initial and continuous vocational training inthe Norwegian working life] Oslo: FAFO – Insti-tute for Applied Social Science, 2001 (Workingpaper).

Hanchane, S.; Méhaut, P. Training, mobility andregulation of the wage relationship: specific andtransversal forms. In: Descy, P.; Tessaring, M.(eds) Training in Europe: second report on voca-tional training research in Europe 2000. Back-ground report. Luxembourg: Office for OfficialPublications of the European Communities,2001, Vol. 2, p. 45-90 (Cedefop Referenceseries, 3008).

Hansson, B. M. Personnel investments andabnormal return: knowledge based firms andhuman resource accounting. Journal of HumanResource Costing and Accounting. Stockholm:PEI – Personnel Economics Institute, 1997,Vol. 2, No 2, p. 9-29.

Hansson, B. M. Marketable human capital invest-ments: an empirical study of employer spon-sored training. Stockholm: School of Business,2001 (Working paper).

Harel, G. H.; Tzafrir, S. S. The effect of humanresource management practices on theperceptions of organizational and marketperformance of the firm. Human ResourceManagement. Alexandria, VA: SHRM – Society


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 317

for Human Resource Management, 1999,Vol. 38, Issue 3, p. 185-200.

Huselid, M. A. The impact of human resourcemanagement practices on turnover, produc-tivity, and corporate financial performance.Academy of Management Journal. BriarcliffManor, NY: AOM – Academy of Management,1995, Vol. 38, p. 635-672.

Ichniowski, C.; Shaw, K.; Prennushi, G. The effectsof human resource management practices onproductivity. Cambridge, MA: NBER – NationalBureau of Economic Research, 1995 (Workingpaper, w5333).

Johanson, U. Why are capital market actorsambivalent to information about certain indica-tors on intellectual capital? Accounting,Auditing & Accountability Journal. Bradford:Emerald, 2003, Vol. 16, No 1, p. 31-38.

Jovanovic, B. Job matching and the theory ofturnover. Journal of Political Economy.Chicago: University of Chicago Press, 1979,Vol. 87, No 5, p. 972-990.

Kalleberg, A. L.; Moody, J. W. Human resourcemanagement and organizational performance.American Behavioral Scientist. London: SagePublications, 1994, Vol. 7, p. 948-962.

Katz, E.; Ziderman, A. Investment in generaltraining: the role of information and labourmobility. Economic Journal. Oxford: BlackwellPublishing, 1990, Vol. 100, p. 1147-1158.

Kazamaki Ottersten, E.; Lindh, T.; Mellander, E.Cost and productivity effects of firm financedtraining. Uppsala: Industrial Institute forEconomic and Social Research, 1996 (Workingpaper, No 455).

Kazamaki Ottersten, E.; Lindh, T.; Mellander, E.Evaluating firm training, effects on performanceand labour demand. Applied Economic Letters.New York: Routledge, 1999, Vol. 6, No 7,p. 431-427.

Krueger, A.; Rouse, C. The impact of workplaceeducation on earnings, turnover and job perfor-mance. Journal of Labor Economics. Chicago:University of Chicago Press, 1998, Vol. 16, No 1,p. 61-94.

Laursen, K.; Foss, N. J. New HRM practices,complementarities and the impact on innova-tion performance. Copenhagen: CBS – Copen-hagen Business School, 2000 (Working Paper,2000-5).

Leget, J. Personeelbeleid en success van organ-isaties: resultaatgericht human resourcesmanagement in Nederland. Deventer: Kluwer,1997 (Dissertation).

Leiponen, A. Competence, innovation and prof-itability of firms. Helsinki: ETLA – Research Insti-tute of the Finnish Economy, 1996a (Workingpaper, No 563).

Leiponen, A. Education, tenure and innovation inmanufacturing firms. Helsinki: ETLA – ResearchInstitute of the Finnish Economy, 1996b(Working paper, No 561).

Leitner, K. H. Intangible resources and firm perfor-mance: empirical evidence from AustrianSMEs. Paper prepared for the 16th NordicAcademy of Management meeting, Uppsala,16-18 August 2001.

Lengermann, P. A. The benefits and costs oftraining: a comparison of formal companytraining, vendor training, outside seminars andschool based training. Human ResourceManagement. Alexandria, VA: SHRM – Societyfor Human Resource Management, 1996,Vol. 35, Issue 3, p. 361-381.

Leonard, J. S. Bringing the firm back in. LabourEconomics. Amsterdam: Elsevier Science,1999, Vol. 6, Issue 1, p. 43-51.

Loewenstein, G.; Sicherman, N. Do workers preferincreasing wage profiles? Journal of LaborEconomics. Chicago: University of ChicagoPress, 1991, Vol. 9, No 1, p. 67-84.

Loewenstein, M. A.; Spletzer, J. R. Dividing the costand returns to general training. Journal of LaborEconomics. Chicago: University of ChicagoPress, 1998, Vol. 16, No 1, p. 142-171.

Loewenstein, M. A.; Spletzer, J. R. General andspecific training: evidence and implications.Journal of Human Resources. Madison, WI:University of Wisconsin Press, 1999, Vol. 34,No 4, p. 710-733.

MacDuffie, J. Human resource bundles and manu-facturing performance: organizational logic andflexible production systems in the world autoindustry. Industrial and Labor Relations Review.Ithaca, NY: School of ILR, Cornell University,1995, Vol. 48, No 2, p. 197-221.

Maglen, L.; Hopkins, S. Linking VET to productivitydifferences: an evaluation of the Prais programand its implications for Australia. Victoria: CEET– Centre for the Economics of Education and


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 318

Training, Monach University, 1999 (Workingpaper, 18).

Marcus, A. J. Efficient risk sharing, non-marketablelabor income and fixed-wage contracts. Euro-pean Economic Review. Amsterdam: ElsevierScience, 1984, Vol. 25, Issue 3, p. 373-385.

Mason, G.; Prais, S. J.; van Ark, B. Vocationaleducation and productivity in the Netherlandsand Britain. National Institute EconomicReview. London: Sage Publications, 1992,Issue 140, p. 45-63.

Meritum. Guidelines for managing and reporting onintangibles. Cañibano, L. et al. (eds) FundaciónAirtel Móvil, 2002.

Michie, J.; Sheehan, M. HRM practices, R&Dexpenditure and innovative investment:evidence form the UK’s 1990 workplace indus-trial relations survey. Industrial and CorporateChange. Oxford: Oxford University Press, 1999,Vol. 8, Issue 2, p. 211-234.

Neal, D. Industry-Specific human capital: evidencefrom displaced workers. Journal of LaborEconomics. Chicago: University of ChicagoPress, 1995, Vol. 13, No 4, p. 653-677.

Neumark, D.; Taubman, P. Why do wage profilesslope upward? Tests of the general humancapital model. Journal of Labor Economics.Chicago: University of Chicago Press, 1995,Vol. 13, No 4, p. 736-761.

NUTEK. Företag i förändring: Lärandestrategier förökad konkurrenskraft [Enterprises in transfor-mation: learning strategies for improvedcompetitive power]. Stockholm: Närings ochTeknikutvecklingsverket, 2000.

Orrje, H. The incidence of on-the-job training: Anempirical study using Swedish data. Stock-holm: SOFI – Swedish Institute for SocialResearch, 2000 (Working paper 2000-6).

Polachek, S. W.; Robst, J. Employee labor marketinformation: comparing direct world of workmeasures of workers’ knowledge to stochasticfrontier estimates. Labour Economics.Amsterdam: Elsevier Science, 1998, Vol. 5,Issue 2, p. 231-242.

Reilly, K. T. Human capital and information: theemployer size-wage effect. Journal of HumanResources. Madison, WI: University ofWisconsin Press, 1995, Vol. 30, No 1, p. 1-18.

Romijn, H.; Albaladejo, M. Determinants of innova-tion capability in small UK firms: an empiricalanalysis. QEH – Queen Elizabeth House,University Oxford, 2000 (Working paper, No 40).

Salop, J.; Salop, S. Self-selection and turnover inthe labor market. Quarterly Journal ofEconomics. Cambridge: MIT Press Journals,1976, Vol. 90, p. 619-627.

Schlicht, E. Endogenous on-the-job training withmoral hazard. Labour Economics. Amsterdam:Elsevier Science, 1996, Vol. 3, Issue 1, p. 81-92.

Schriver, R.; Giles, S. Real ROI numbers. Trainingand Development. Alexandria, VA: ASTD –American Society for Training and Develop-ment, 1999, Vol. 53, Issue 8, p. 51-55.

Topel, R. Specific capital, mobility and wages:wages rise with job seniority. Journal of PoliticalEconomy. Chicago: University of ChicagoPress, 1991, Vol. 99, No 1, p. 145-176.

Trouvé, P. The employment and training practices ofSMEs: examination of research in five EUMember States. In: Descy, P.; Tessaring, M.(eds). Training in Europe: second report onvocational training research in Europe 2000.Background report. Luxembourg: Office forOfficial Publications of the European Commu-nities, 2001, Vol. 2, p. 91-232 (Cedefop Refer-ence series, 3008).

van Buren, M. E.; Erskine, W. State of the industry:ASTD’s annual review of trends inemployer-provided training in the United States.Beaverton, OR: ASTD – American Society ofTraining and Development, 2002.

Verburg, R. M. Human resource management: opti-male HRM praktijken en configuraties.Amsterdam: Vrije Universiteit, 1998 (Disserta-tion).

Veum, J. R. Training, wages, and the human capitalmodel. Washington, DC: US Bureau of LaborStatistics, 1995 (NLS report, 96-31).


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 319

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 320

The benefits of education, training andskills from an individual life-course

perspective with a particular focus onlife-course and biographical research

Maren Heise, Wolfgang Meyer

AbstractThis paper has been prepared within the framework of the third Cedefop report on vocational educationand training (VET) research in Europe which is dedicated to evaluation and impact research. The objec-tive is to provide an overview of national and cross-national research into the benefits of education andtraining from a life-course perspective. The existing literature and approaches in this field are reviewedand their results discussed from a European perspective. The report tries to develop an integratedperspective on the material and non-material benefits of education and training throughout the life courseand introduces the theoretical approach of life-course and biographical research and its methodologicalimplications. The added value of life-course studies and biographical research for conventional researchon education and training benefits is highlighted through lines of thematic investigation. After a review ofcurrent empirical work carried out at national and European level, a summary of key findings is presentedwhich highlights those results that allow European (or at least cross-national) comparisons to be takeninto consideration. Finally, addressed are the implications of research evidence for policy and practiceand recommendations for further research and on how to improve data comparability, particularly atEuropean level.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 321

1. Introduction 324

2. Individual benefits of education, training and skills 325

3. Approach and focus of investigation 328

3.1. Life-course and biographical research 328

3.2. Relevance of life-course and biographical research for the investigation of education

and training benefits 334

4. Empirical evidence 338

4.1. Life-course and biographical research in Europe 338

4.1.1. Monetary returns on education and training and life-time income 339

4.1.2. Education, training and labour-market participation 340

4.1.3. Education and transitions 340

4.1.4. Generational differences in education and training benefits 341

4.1.5. Social differences in education and training benefits 342

4.1.6. Subjective perception of educational benefits in the biography 342

4.2. Research design and data 343

4.3. Empirical Studies 346

4.3.1. Studies at national level 347

4.3.2. Cross-national studies 352

5. Discussion of results 357

5.1. Individual monetary returns on education and training 357

5.2. Education, training and labour market participation 359

5.3. Education, training and transitions 360

5.4. Generational and cohort differences in education and training benefits 361

5.5. Social differences in education and training benefits 361

5.6. Non-material benefits of education and training and subjective biographical perception 362

5.6.1. Non-material benefits of education and training 362

5.6.2. Subjective biographical perception of educational benefits 363

6. Conclusions and recommendations 365


Annex: list of data and information sources 368

References 371

Table of contents

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 322

Tables

Table 1: Longitudinal data sets for research into the impact of education and training in Europe 346

Table 2: National studies on education and training benefits from a life-course perspective 348

Table 3: Selected cross-national studies on education and training benefits from a life-course perspective 354

Figures

Figure 1: Individual benefits of education and training 326

Figure 2: Cross-sectional design 331

Figure 3: Retrospective design 332

Figure 4: Comparative cross-sectional design with retrospective questions 332

Figure 5: Panel-design 333

Figure 6: Panel-design with retrospective questions (source population only) 333

Figure 7: Panel-design with retrospective questions (additional sampling) 333

Figure 8: Lexis diagram 334

Figure 9: Prospective data designs 343


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 323

This paper has been prepared within the frameworkof the third Cedefop report on vocational educationand training (VET) research in Europe, which is dedi-cated to evaluation and impact research. The paperaims to present an overview of national andcross-national research into the benefits of educa-tion and training from a life-course perspective andwill discuss relevant approaches and methodologiesin the field. Since other contributions focus on VETimpacts on a macro or meso level, our study isconcerned with the individual, or microsocial bene-fits of education. In particular, we look at the impactof education, training and skills from a life-courseperspective. Life-course and biographical researchcomplements other studies by investigating thelong-term educational benefit for individuals and theimpact over time of social and structural change. Theaim of our study is to provide an overview of existingliterature and work in this field and to discuss its find-ings on the benefits of education and training for theindividual in different European countries. In contrastto other approaches, life-course and biographicalresearch is not limited to investigating materialissues such as employment and income differentia-tion, but also focuses on the non-material outcomesof education and training. Our paper will, therefore,consider the material as well as non-material bene-fits of education and training.

As with many other contributions in this report,the basis of this paper is a secondary analysis ofrecent empirical studies. Besides reviewingpublished literature for this investigation, we usesources from internet-based material, nationalexperts and grey literature. A complete list ofsources is attached in the annex. Nevertheless,the search for relevant material turned out to bequite complicated, especially when trying toprovide a European overview. As little work hasbeen found in southern European Member States,any generalisation of results remains limited.Furthermore, it has become obvious that national

research is diverse in both its quantity and quality,which is clearly reflected in the data sources used.Comparisons between different countries aretherefore difficult to reach. These and furtherproblems of investigation will be discussed in thefollowing chapters of the paper.

Following the scientific debate on what benefits ofeducation, training and skills precisely are, thesecond chapter sets out different concepts and triesto develop an integrated perspective of material andnon-material benefits. Furthermore, a classificationof education and training benefits is attempted inorder to provide an analytical tool for the laterarrangement of empirical research in the field. Thethird chapter introduces the approach of life-courseand biographical research and illustrates themethodical implications for empirical research work.The focus of the investigation into the benefits ofeducation and training is explained from theperspective of life-course analysis, and the addedvalue of life-course studies for conventional researchinto education and training is highlighted. Chapter 4provides an overview of existing empirical work onthe benefits of education and training from alife-course and biographical perspective. Also in thischapter the main methods used to investigate theimpact of education and training through life-courseand biographical research are clarified andsupported by studies from national and cross-national empirical work. As most of the studies arebased on different national longitudinal data sets, acomparative and critical approach is adopted.Chapter 5 sums up the key research results. Theevidence for the material and non-material benefitsof education and training throughout the individual’slife course is reviewed and scrutinised critically for itsrelevance for European VET research. The finalchapter tries to derive some implications for policyand practice through focusing on recommendationsfor further research and on how to improve datacomparability at European level in particular.

1. Introduction

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 324

The simple assumption that education andtraining have short-term and long-term effects onlife-course patterns, at least on the individual’scareer and (life) wages, is generally accepted andits correctness seems to be fairly obvious. Butwhen it comes to actual education and trainingbenefits, questions arise: what do we understandby the term ‘benefits’ and – equally contentious –how can these be measured? Furthermore,education and training can take quite differentforms as regards its type, content, degree offormality and resources invested. This chaptertherefore deals with current perceptions of thebenefits of education and training and theirconsequences for empirical investigation. Asconcepts of education, as well as training andskills, differ strongly not only between Europeancountries but also because of different ‘schoolsof tradition’, it is necessary to clarify the defini-tions of education and training as used in thispaper. A research review of empirical work needs,as an introduction, a relatively simple technicalunderstanding of its central terms (education,training and skills), in order to clarify the variousdefinitions of education, training and skills/qualifi-cations in existing empirical surveys:(a) education:

the term education is used to meanprogrammes of learning with general objec-tives relating to the personal development ofthe learner and his/her acquisition of knowl-edge. Formal education takes place in a struc-tured and taught manner normally in schoolsor other educational institutions. Education isalso a property that a person possesses aftergoing through this process, usually confirmedby a formal and generally accepted qualifica-tion. Therefore, education as a concept istangible and is – in comparison to intangibleterms like learning – relatively easy to measure(Tessaring et al., 2003; Desjardin, 2001);

(b) training:in comparison to education, training is moredirectly related to the preparation of individualsfor employment in current or emerging occupa-tions. Training can take place on-the-job as well

as off-the-job, the latter usually being organisedas programmes offering a sequence of courses.Training can include applied learning,problem-solving skills, work attitudes, generalemployability skills, and the occupational-specific skills necessary for economic indepen-dence as a productive and contributing memberof society. The training a person has obtained isusually measured in quantitative terms (duration,frequency) discriminating between types (initial,continuing), degrees of formality and place(Pfeiffer, 2001);

(c) skills:The term skill is defined as the relevant knowl-edge and experience needed to perform aspecific task or job. Skills also constitute theproduct of education, training and job experi-ence together with relevant technicalknow-how. Specific skills can only be measuredthrough elaborate testing procedures which arenormally too costly to perform. As an alternativeto the lack of objective measurements, surveysrely on subjective statements from respondentsregarding the skills they believe they possess.This is unlikely to be reliable, because thesubjective perception of skills can differ stronglybetween individuals (Bjørnåvold andTissot, 2000; Mertens, 1999; ETF, 1998).

Most empirical studies seem to be based onrather implicit definitions of education andtraining and reveal a somewhat unstructuredpicture. As there are many different notions ofcomplex concepts like education, training orskills, there are also many different perceptions ofthe benefits of education and training. In econo-metric studies, for example, individual benefitsare often reduced to educational returns in termsof income or wage development, avoidance costsand other measurable economic benefits.Predominantly based on human capital theory(Section 4.1.1.), these studies have a clearconcept of educational benefits which is asconsistent as it is simplistic.

In contrast, psychological and educationalresearch concentrates on non-monetary or‘wider’ benefits like health, reduction of criminal

2. Individual benefits of education, training and skills

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 325

behaviour or social exclusion, usually withoutproviding any corporate concept of the benefitsof education and training. Theoretical educationaldiscussion does not really contribute any greaterclarity. Lacking more or less any descriptiveconcept, most of this research dissipates intosome kind of philosophical discourse about themeaning and importance of educational benefits.

From a sociological perspective, benefits areperceived as bringing all manner of advantages –material or non-material – concerning the indi-vidual’s place within society. Social institutions (e.g.educational system, labour market) and individualbehaviour (e.g. educational decisions) determinethe chances and risks of attaining, or not attaining,the social position to which an individual aspires.Such a definition allows at least for the integrationof economic and non-economic benefits (e.g.power, prestige, satisfaction) and – as sociologicalresearch shows – points to their strong interrela-tionship. The all-embracing sociological concept issocial status, i.e. the position of a person in thesocial structure of a certain group or society. Status

can be assigned (e.g. through nationality, age) butalso attained actively through individual perfor-mance. It is assumed that education and trainingplay a major role in status attainment. Within thisconcept ‘hard’ education and training outcomes,that might be termed material benefits (employ-ment, occupational position, etc.), are more or lessdirectly linked to monetary aspects while other‘soft’ or non-material outcomes are rather indirectlyinfluenced by education and training and oftenconveyed through these material benefits. Forexample, health or participation in social andcultural life are, at least in part, influenced by indi-vidual earnings or the economic status of a person.We recommend the following differentiationbetween material and non-material benefits as afirst analytical tool in this review of studies dealingwith the benefits of education and training.

Nevertheless, this applied perspective alsobrings with it some difficulties. Although it seemsto be common sense that material benefits aremainly represented by patterns of employment,income and career prospects, concepts of


Figure 1: Individual benefits of education and training

Culture indipendent Culture dependent

Lifelong learning

‘Learning career’

Material benefits

Monetary

current incomefuture incomelifetirne income

Non monetary

career prospectsjob securityjob adequaci

Non material benefits

e. g. health, quality of life, social andcultural participation, personal well-being

age, activity, life expectancy, chancesof marriage, honorary activity

Individually perceived

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 326

non-material benefits are more controversial (1).Moreover, non-material benefits are not primarilyobjective constructs but are merely the result ofsubjective judgements or cultural dispositions.Marriage, for example, has largely been seen as asocial benefit and higher education has increasedthe likelihood of marriage. Today – as a result ofindividualisation – marriage is losing its attractive-ness, especially for those with experience ofhigher education. Nevertheless, some non-mate-rial beneficial aspects can be considered substan-tially agreed upon, such as health and socialparticipation, even if the perception of their bene-fits remains again rather subjective. For analyticalcorrectness we recommend three additionaldimensions of education and training benefits:individually perceived benefits; culture dependentbenefits; and culture-independent benefits.

Most empirical research into the benefits ofeducation and training claims to apply an objectiveperspective as a benchmark for beneficialoutcomes. But this objectivity is rather questionableeven with regard to material benefits, because everybenefit remains subjective at micro level. As a kindof compensation for this problem of attribution (howcan we know then what is an individual benefit andwhat is not?) an implicit transfer of a macrosocial tothe individual perspective is carried out. Whatever islabelled a benefit from a mass perspective is also a

benefit from the individual perspective and viceversa. Understanding individual (micro) benefits is,therefore, largely congruent with social (macro)benefits (for a discussion of the macrosocial bene-fits of education see the contribution of Green et al.to this report). There is also a tendency to stress theintended outcomes of education and training ratherthan the unintentional benefits or harm.

Research from a life-course perspective tendsto concentrate on those education and trainingbenefits which have a longer-term relevance in anindividual’s development and which undergo aform of accumulation over the life course. Amongpossible material benefits are primarily (lifetime)income, employment opportunities, careerprospects and the avoidance of unemployment.Non-material outcomes of education and trainingin which life-course research has an interestrelate mainly to health, marriage and familyformation as well as social participation andstatus attainment. Furthermore, the inequalities,or neutrally expressed differences, in accessingeducation and training and the benefits of educa-tion and training between social groups, are ofspecial interest in life-course research. Thefollowing chapter provides explanations for thisspecial interest by giving a short introduction tothe theoretical approach and focus of empiricalinvestigation of the life-course perspective.

The benefits of education, training and skills from an individual life-course perspective

with a particular focus on life-course and biographical research 327

(1) The separation of material and non-material benefits is, of course, artificial. While, for example, monetary outcomes can becomparatively easily labelled as a material benefit, the occupational position a person achieves is not only a material andsometimes (e.g. compared to a position occupied before) not even a material benefit, but can also be a non-material one –expressed, for example, in achieving higher social prestige.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 327

The individual life course can be observed fromvery different viewpoints and with a great varietyof social theories in mind. For the purpose of thisresearch review, the following short definitions oflife course and biography (according to Meule-mann, 1990a) should be used:(a) the life course is defined as – and mostly

determined by – a series of individual deci-sions between institutionally offered alterna-tives which an individual is forced to make atseveral specific points in history. Each deci-sion influences the path through the institu-tion and the future decision situation. Theopportunities presented by social institutionsopen different pathways of one’s individuallife through society and its social structure.These generalised pathways are divided intotypical patterns and several well-definedsequences. Changes between these path-ways (in form of status passages or socialmobility) are only possible at special historicaltransition points, leading from the exit of onesequence to the entrances of a limitednumber of new sequences. Although thehistorically observable individual life course isgenerally structured, and in main parts deter-mined by social institutions, individual deci-sions between the opportunities offered atthese transition points are equally importantfor its development;

(b) Biography is the subjective interpretation anddigestion of all life-events, from which deci-sions during the life course (and their results)are only one (though important) part. More-over, biography is the self-perception of one’slife history and is recognised by the individualas a single entity (in contrast to the clear-cutsequences of the pathways determined bysocial institutions). While individual biogra-phies are structured by more or less the samepatterns (due to the institutional design of lifecourse), homogenous forms of self-percep-tion and identification with social positionsseem to be obvious. As a result, typicalsociocultural milieus with common interests,experiences, feelings, knowledge, etc., will be

formed and have a major impact on socialstratification, which itself influences the insti-tutionalisation of life courses by offeringopportunities at the transition points. Thesociocultural milieus and their way of thinkingof society are merely reflected in biographicreports on individual lifestyles or personalautobiographies, these being the most impor-tant sources for biographic research. AsRoberts (2002, p. 1) mentioned, ‘biographicalresearch is an exciting, stimulating andfast-moving field which seeks to understandthe changing experiences and outlooks ofindividuals in their daily lives, what they seeas important, and how to provide interpreta-tions of the accounts they give of their past,present and future.’

By using this differentiation, life-course researchconcentrates on the decision situation offered byinstitutions and its well-defined alternatives (whichare mostly also guaranteed by law), while biographyfocuses on the individual (psychological) process ofperceiving, assimilating, understanding and recon-structing reality. It has to be mentioned that otherdefinitions and separations between life course andbiography (if any at all) are available in literature andthe authors of the studies presented here may notnecessarily agree with this characterisation.

3.1. Life-course and biographicalresearch

While talking about life-course and/or biographicalresearch, one has to consider that most disciplineswithin the social sciences have some links to thiskind of research. A number of examples concerningspecific research questions on the relationship ofeducation and individual life course will bepresented in Chapter 4. In this chapter the focus lieson the historical development and its methodolog-ical implications of life-course and biographicalresearch. Only a brief overview is possible and nosystematic investigation of all theoretical perspec-tives should be considered (for further information:

3. Approach and focus of investigation

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 328

Heinz, 1997; Ecarius, 1996; Mayer, 1990;Voges, 1987; Sørensen, 1986; Clausen, 1986;Elder, 1985; Kohli, 1978). However, the strong inputof both economic and sociological thinking on thedevelopment of life-course and biographicalresearch should be mentioned here.

As a result of World War II, social scientificresearch in Europe almost disappeared in the1940s. Many economists and social scientistsescaped to America, finding not only peace but alsoa continuously improving infrastructure for theirresearch within an open-minded society. Stimu-lated by this ‘brain-drain’ from Europe (and othernations), the newer research areas of economicsand sociology developed very quickly. One stronginput for life-course research, the study of socialchange, came out of some of the early studies oncity growth, industrial change, and migration in theUS, which itself formed one of the most importantsociological fields of research interest (for a briefoverview and introduction see Sanderson, 1995). Ingeneral, this research tradition is mainly committedto a macrosociological perspective and stronglyrelated to the macroeconomic view on long-termdevelopments. Depending on the political positionof the researchers as well as on membership ofcompeting scientific schools different mixtures ofscientific disciplines (e.g. political economy,socioeconomic research) and terms to describeand explain the phenomena under observation (e.g.modernisation, imperialism) have been developed.In the 1960s and the early 1970s, many researchstudies were conducted on the socioeconomicdevelopment of countries. Most had a comparativeperspective and they were mainly based on (Neo-)Marxist or structural-functional theories.

Confronted with Karl Marx and his postulationof historical materialism (the first explicitly formu-lated theory of social change with class conflictsas its driving force, dominating theoreticalthinking, at least in Europe, up to the midst of the20th century) functionalism, the North Americanmainstream of sociological theory-building of thistime, was criticised for its static and thereforeconservative implications. As a reaction to thiscriticism, Parson (1966) linked his structural-func-tional theory to biological evolution theory. Whilethe British social theorist Herbert Spencer failedto earn broad respect for his trial to connect soci-ological thinking on biological evolution theoryonly a short time after its revolutionary influence

at the end of the 19th century, Parsons reformu-lation had been recognised as a stimulating inno-vation. This paved the way for several theoreticaland empirical works which were summarisedunder the label of ‘modernisation theory’(although they never built one homogenous theo-retical school; as an overview see Zapf, 1975).

What is common to these studies of socialchange (as well as for dependency theory or trans-formation theory) is the attempt to explain socialdevelopment primarily with macrosociological vari-ables (for an overview on these theories see Box 1).This is also true for macroeconomic modelling,especially those studies concerned with labourmarket development (including neoclassicaltheory, as well as Keynesianism; for an overview onlabour market economy see Franz, 1991; Ashen-felter and Layard, 1986; Holler, 1986). Althoughmicro theories of social action are not generallyrejected, there are only weak ties to them withinthese studies. While most modernisation theorists(economists as well as sociologists) in functional-istic tradition argue in terms of social systems(sometimes using individuals only as puppets onthe strings of norms and sanctions, which perfectlydetermine their behaviour), the majority of macroe-conomists refer to the homo oeconomicusconcept, describing human decisions as perfectlyrational in terms of economic considerations. Infurther theoretical development, both kinds of over-simplification proved to be too weak as a satisfyingexplanation of social change.

Contrary to this, the second major line ofresearch associated with life-course and biograph-ical research focuses on individual developmentand has its roots in microsociological, ethnograph-ical, psychological, and pedagogical theories (foran overview see Elder, 1991). Early works on familypatterns and migration in Europe and the USconsidered the importance of personal history toexplain individual decisions. To analyse this, alongitudinal approach to life history (Volkart, 1951,p. 593), including for example continuous qualita-tive life records, had been first developed byW. I. Thomas early after World War II. Althoughsome research (especially on child welfare inthe US) used these new methods, the popularity ofsuch (qualitative) longitudinal studies stayed verypoor until the mid-1960s. Other examples of theorigins of life-course and biographical research atmicroanalytical level can be found in the theoretical



01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 329

reflections on generations (especially the work ofMannheim, 1952) and in the development of thecohort concept in demographic research (e.g.Ryder, 1965). The analysis of family-cycles(Glick, 1947) and of passage between different agegroups (Gennep, 1960), and those parts of soci-ology which had been interested in differentlife-periods (e.g. the sociology of ageing, the soci-ology of youth) are further examples.

To connect the research lines on social changeand on individual development, a third theoreticalperspective was of great importance. Starting withthe discussion on the ‘open society’(Popper, 1966) and its opportunities for intergen-erational mobility in the 1950s, inequality researchreached another stage of investigation by askinghow people get into their social positions (e.g.classes). Important steps were made through theanalyses of status attainment processes within

societies (Blau and Duncan, 1967), the discussionon objective placement in classes and themeaning of collective judgements like prestige(Svalastoga, 1959), the differentiated analyses ofoccupational mobility (Rogoff, 1953) and theempirical definition of position structures as unitsfor mobility research (Goldthorpe, 1980).

Starting with an understanding of intergenera-tional mobility through the use of mobility tables incross-sectional surveys, the focus moved more andmore to a consideration of intragenerational mobilityand the influence of social institutions like school,family, the economic system and so to structuringthe individual life course (see Mayer andMüller, 1986; Mayer et al., 1991). As a result, statuspassages, transitions and critical life-events duringthe occupational career came into the spotlight ofanalyses. Within this research tradition, individualqualification and the development of the educa-


The modernisation approach had been primarily evolved to explain underdevelopment by using the

historical observable development patterns (and its causes) of advanced (western) societies as an

analytical model. By using the label ‘modernisation theory’, several, sometimes competing, theories on

sociocultural evolution have been summarised. Basically these theories are united by two assumptions

(Sanderson, 1995; p. 212 et seq.).

First, development is postulated as a standardised, general transition from underdevelopment to

modern society by several steps. Rostow (1960), for example, distinguished five stages of economic

development: traditional society, the preconditions for take-off, take-off, the drive to maturity and mass

consumption. As important influences for transition from one stage to another he assumed social

patterns, political structures and value systems as preconditions for economic development. These

variables had been measured on a national level (not mentioning regional or individual differences) and

used for international comparisons.

Second, modernisation theories recognise development as an endogenous process of national societies,

only determined by internal deficiencies or capacities. International networks and dependences had not

been mentioned by this approach in its early formulations. Therefore it had been criticised by dependency

theorists in the early 1970s, which put (economic) interconnections between developed and underdevel-

oped countries as the focus of its argument (Caporaso, 1980; Palma, 1981; Valenzuela and Valenzuela,

1978). Common to modernisation theories, dependency theory argues purely with macrosociological (e.g.

migration balances of elites) or macroeconomic (e.g. terms-of-trade) variables.

More recent versions of modernisation (e.g. the concept of reflexive modernisation of Beck, 1994) try to

avoid the disadvantages of those early approaches by integrating other concepts of social change (like

dependency theory) and by using micro theoretical explanations to support their macrosociological (or

macroeconomic) considerations. For actual development of modernisation theory and critical comments on

the historical background see for example Berger (2000), Engerman et al. (2003), Tiryakian (1991).

More recent versions of modernisation (e.g. the concept of reflexive modernisation of Beck, 1994) try to

avoid the disadvantages of those early approaches by integrating other concepts of social change (like

dependency theory) and by using micro theoretical explanations to support their macrosociological (or

macroeconomic) considerations. For actual development of modernisation theory and critical comments on

the historical background see for example Berger (2000), Engerman et al. (2003), Tiryakian (1991).

Box 1: Modernisation theories

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 330



Figure 2: Cross-sectional design

Informationavailable

Left-censored

N

Righ-censored

(no information) (no information)

Observationwindow

t

t0

tional system has been considered as one of themost important determinants of these processes.Hence, the individual development perspective hasbeen strongly linked with the macroprocesses ofsocial change as well as with the quickly developingmicroeconomic research on labour market develop-ment, which has increased following the ongoinglabour-market crisis in Europe since the 1970s.

The first steps towards the institutionalisation oflife-course and biographical research as a researchfield in its own right (in the US as well as in someEuropean countries like Germany, the UnitedKingdom, the Netherlands and the Scandinavianstates) can be found in the mid 1970s. For the firsttime, conferences were held and handbook articlesand readers on this topic were published(Elder, 1975; Kohli, 1978). However, life-course andbiographical research still remained stronglyrelated to the sociology of age and family – studieson the transition within the educational and/or theoccupational system continued to be exceptions tothe norm. An increasing interest in topics like statuspassages during life course (for an overview seeLevy, 1991) and the institutionalisation of standardbiographies (Kohli, 1988) began to appear only atthe beginning of the 1980s.

One important aspect of the growing popularity oflife-course analysis within the context of educa-tional and labour-market research in the early 1980swas the improvement in data quality and analysistechniques. As presented later, the major datasources for longitudinal research in Europe hadbeen implemented no earlier than the beginning ofthe 1980s. In Germany, for instance, the interdisci-

plinary Special Research Unit 3, set up in 1980 at theUniversities of Frankfurt/Main and Mannheim, initi-ated and developed over the next 12 years some ofthe most important data sources for life-courseresearch to date. Examples of this work are thewelfare surveys, the German socioeconomic panel(GSOEP), and the German life history study (GLHS)(Hauser et al., 1994). Today in Germany, not onlyscientific surveys but also process-produced dataare available for life-course research onlabour-market development. The Federal Institutefor Employment Research (IAB) offers a data ware-house, which includes longitudinal data from thelabour offices, the social statistics and severalregular surveys (some of them in panel design) andtries to match this data to increase the informationalbasic on every case (Schwarzfärber, 2002). Similardevelopments can be found in other Europeancountries (especially in the Netherlands, Scandi-navia and the UK). Even in the US only a few compa-rable data sets can be found earlier than 1980 (e.g.the Michigan Panel Study of Income Dynamicsstarted at the end of the 1960s).

Compared to a single representativecross-sectional survey (Figure 2), the claims putforward for the quality of the data of life-courseresearch are obviously very high. A perfect data setshould contain a complete sequence of informationfrom the beginning to the end (which in many casesmeans from birth to death of a person).Cross-sectional surveys (as for example used inopinion polls) are only able to represent exactly theinformation valid at the time of data collection. Itmight tell us whether one person is employed or notand whether he (or she) has a university degree. Ingeneral, this kind of survey will not tell us anythingabout the individual historical development (pastand future) and the circumstances under which thisdevelopment has taken place. The data set isincomplete due to right and left censoring.

A data set is called left censored if informationneeded from the past of a single person’s life courseis missing. One simple solution to this problem is toask for this kind of historical information during theinterview, e.g. using questions like: ‘when did youleave school?’ or ‘how long have you beenemployed?’ (Figure 3). Although some elaboratetechniques have been developed, the collection ofretrospective information is strongly restricted bythe respondent’s ability to recall the facts required(Papastefanou, 1980). Due to the results of method-ological studies, the use of retrospective data

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 331


Figure 3: Retrospective design


N

Righ-censored

(no information) (no information)

Observationwindow

t

t0

Figure 4: Comparative cross-sectional design

with retrospective questions

InformationavailableN

Observationwindows

t

t2t1t0 t3

collection methods concerning central events inone’s life (e.g. marriage, birth of child) seems to bepossible (Blossfeld, 1985b). Questions regardingeducational and employment history proved to be ofreasonable validity too (Brückner, 1994; Dex, 1991;Mayer and Brückner, 1989). Nevertheless, the use ofretrospective data collection is limited: subjectivejudgements and opinions especially seem to bestrongly biased by an individual’s attempt to‘streamline’ their own behaviour and decisions(Schwarz et al., 1994; Schwarz and Sudman, 1994).

By using a single cross-sectional designanother serious problem occurs related to thehistorical time of data collection. The informationavailable at the time of observation is alwaysright-censored (lack of future information) andmay become obsolete at a future date due tosocial change (e.g. political decisions, new laws,and economic development). To measure theseeffects one needs comparative representativesurveys at regular time intervals, which also try tocapture the individual life history by additionalretrospective questions (Figure 4). However, suchanalyses only allow population comparisons andare not able to follow individual development overtime. Retrospective questions are necessary, butgive no information on opinions or options facingthe individual in the past. Therefore, in order alsoto investigate the reasons for individual decisions,panels should also include prospective questions,e.g. concerning future plans, problems, alterna-

tives, preferences, etc., which afterwards formthe basis for explaining decisions taken.

To analyse the individual life course over time,a panel design is needed, for which the samepersons are interviewed again at regular timeintervals (Figure 5). Obviously, the technicalrequirements (and therefore the expenses)increase. In addition to the regular costs of repre-sentative surveys one has to control addressesand invest some extra time to find people whomoved between the two dates of investigation.Additionally, some special problems associatedwith this kind of research design occur (for a briefoverview see Blossfeld and Rohwer, 1995; p. 11et seq.; detailed information can be found inHsiao, 1986). From the viewpoint of life-courseand biographical research, pure panel designswithout any retrospective questions are limitedbecause they only offer information for a series ofdiscrete points on the time scale.

Only a panel-design with retrospective infor-mation included (Figure 6) will give the opportu-nity to investigate individual life courses and theircontinuous development. Even then, there is stillone problem remaining: the attrition of thesample. While those previously sampled in paneldesign will be repeatedly interviewed, the numberof participants declines as time goes by (panelmortality). Moreover, the data set stays represen-tative for the original sampling process, ignoringthe population change (due to migration andfertility) in the society under observation. There-

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 332



Figure 5: Panel-design


N

Observationwindow

t

t0 t1 t2 t3

Figure 6: Panel-design with retrospective

questions (source population only)


N

Observationwindows

t

t0 t1 t2 t3

Figure 7: Panel-design with retrospective

questions (additional sampling)


N

Observationwindows

t

t0 t1 t2 t3

fore, additional sampling at each point of obser-vation (or at least at some reasonable time inter-vals) is necessary to keep the information of thepanel data representative (Figure 7).

Although some of the problems mentioned hereare almost the same for life-course and biograph-ical research, there is at least one important differ-ence that makes it more difficult to collect adequatedata for analyses in life-course than in biographicalresearch. According to its interpretative approach,which tries to understand the individuals’ percep-

tions of their own biographies and to analyse themin detail, biographical research uses qualitative,hermeneutic methods of analysis. The oral orwritten biographical material is generally inter-preted by the researcher applying interpretationtechniques of the so-called life-history method: ‘Abiographical study is the study of an individual andher or his experiences as told to the researcher orfound in documents and archival material’(Creswell, 1998). In contrast to this, life-courseresearch is interested in explanations of populationdevelopment by using representative datasampling and quantitative analysis methods todraw conclusions. Obviously, most data collectionproblems listed above are closely related to theneed for representative data sampling, which isseldom used in biographical research for itspurpose of understanding instead of explaining.

Moreover, compared to common quantitativeapproaches, life-course research introduced timeas the most important new variable. Time can beobserved from three different points of view, whichare strongly interrelated. From an individual percep-tion, time means the process of ageing, while froma collective point of view time is primarily historicaldevelopment. Both aspects are connected in thecontext of cohorts, which are defined by means ofa common historical starting point for individuals(e.g. those people born in the same year form birthcohorts, staying at the same time in specific socialinstitutions such as school). This interrelationship

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 333

was discussed and presented in diagram form(Figure 8) by the demographer Lexis (1875) at theend of the 19th century. Nevertheless, most socialscientific research concentrates only on one ofthese aspects. Life-course research offers thepossibility of separating age, cohort and periodeffects by using specific statistical measures oncethese had been developed (Hagenaars andCobben, 1978; Rogers, 1982; Mayer andHuinink, 1990, 1994) (2). Moreover, the computerequipment required for handling longitudinal datasets of adequate size did not become availablebefore the beginning of the 1980s (for an early intro-duction see Tuma and Hannan, 1984).

The central and commonly applied statisticalmethod of life-course research is ‘event historyanalyses’ (for introduction see Yamaguchi, 1991;Blossfeld and Rohwer, 1995; Giele andElder, 1998). Using these techniques requires aspecial data structure. First, as dependent vari-able (the event), a transition between discretestates has to be defined. In general, two clearlydivided states (one initial and one destinationstate) are used. Secondly, the time axis isassumed to be continuous and – as analyticalunit– the time span an individual spends on thisaxis until a change of states occurs (the episodeor spell) will be used for analyses. Hence, aprecise and clearly distinguishing definition of thetwo states and a permanent observation of dura-


Figure 8: Lexis diagram

Age

100

80

60

40

20

1900 1920 1940 1960 1980 2000 Historicaltime

Birthcohorts

People

born

betwee

n 19

00 an

d 192

0

tion time are needed. Thirdly, the transition musthappen at an explicitly measurable point on thetime axis and not as a gradual change from originto destination state.

Right censoring can be handled adequately byusing (non-parametric) survival analyses or agroup of related parametric transition ratemodels, which additionally need the definition ofa specific shape of time dependence for the tran-sition rate. In contrast, left censoring is a seriousproblem in that transition rates depend on theduration in the original state (which commonlyhas to be assumed). Moreover, some other theo-retical and methodological problems are experi-enced using such advanced statistical analyses(e.g. Bretagnolle and Huber-Carol, 1985; Hamerleand Tutz, 1989; Galler and Pötter, 1992).

With respect to these restrictions, event-historyanalysis can be done by using standard statisticalsoftware packages like SAS or SPSS. However,such procedures were not available in theseprogrammes before the end of the 1980s and thecapacity of personal computers for this kind ofcalculation was not sufficient before the mid-1990s.Hence, the improvement of life-course research washighly dependent on the technical development instatistical hard- and software. We have to considerlife-course and biographical research as still a veryyoung and fast developing new discipline, influ-encing (and being influenced by) a great variety ofsocial sciences. Amongst these, economics andsociology seem to be the most important.

3.2. Relevance of life-course andbiographical research for theinvestigation of educationand training benefits

The objective of the last section was to describebriefly the development of life-course andbiographical research and to give a shortoverview of the methodological implicationsdirectly related to this specific perspective. Asoutlined above, life-course and biographicalresearch was not primarily developed to investi-gate the benefits of education or labour-market

(2) A simultaneous consideration of all three effects within a single model is not even possible with advanced theoretical or statis-tical methods because of the tautological relationship between the three effects (see for more detail Descy andTessaring, 2001; p. 322)

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 334



processes. Nevertheless, these research topicshave made at least strong contributions to thedevelopment of life-course and biographicalresearch as a specific scientific research field. Inthe following sections, the input of life-courseand biographical research on the development ofeducational and labour market research isdiscussed. The main question is to examine thecentral contribution of life-course and biograph-ical perspective for these research areas. Again,no complete overview or systematic introductioninto these studies should be expected.

In the second half of the 20th century, educa-tion was one of the most important and some-times most controversially discussed topicswithin European societies and a main source ofnew social movements. The most recent devel-opments have emerged from the results of thePISA study, ranking the efficiency and competi-tiveness of national educational systems withinOECD countries by measuring the performancesof schoolchildren on comparable levels andwithin different subjects (OECD, 2001a, 2002aand b, 2003). While some European countries(especially Finland) did very well, others such asGermany and Luxembourg achieved only poorresults, which initiated a political debate in thesecountries (3) (for the German results and reactionssee Deutsches PISA-Konsortium, 2001, 2002;Adam, 2002; Terhart, 2002). Previous historicalevents (especially the ‘Sputnik-Shock’ 1957 andthe Student-Movements at the end of the 1960s)also stimulated political discussion inside as wellas outside of the parliaments in several countriesall over Europe and finally led, to a greater orlesser degree, to deeply influential reforms and arestructuring of national educational systems.

Unsurprisingly, political controversies alsoencourage and support scientific research andtherefore many different research institutionsacross the whole of Europe have producedstudies and publications on educational topics.Among the most important issues investigated bythis research are:(a) the world wide expansion of education since

the 1950s. Not only in Europe but all over theworld, an increasing number of young peopleare enrolled in education, with the greatestmovement towards higher and higher levelsof education. Research enquiries try to under-stand the reasons, as well as the results ofthis process for the world community, interna-tional relationships, nation states, andregional development (e.g. Boudon, 1974;Dore, 1976; Müller, 1998);

(b) equal access to education. One importantpoint associated with the expansion ofeducation is the question of whether there areequal opportunities for entering differentlevels of education and what are the criteria(e.g. ability, money, social status) used forselection. In Europe, the ideal of equal oppor-tunities and selection focused strongly onability is widespread, although the reality inmost countries varies (e.g. Shavit and Bloss-feld, 1993);

(c) right to follow individual aspirations. Particu-larly in contrast to the communist states ofeastern Europe, which postulated the state’spreferences on the use of individual educa-tion, the western world followed the principleof free decision-making within the educa-tional system (excluding from this right theduty to attend schools). Therefore, not onlypersonal abilities and performances but alsoindividual aspirations and wishes were putinto the focus of analyses (e.g. Duncanet al., 1971).

Undoubtedly, studies from the life-course andbiographical perspective made some very impor-tant contributions to these (and of course someother) debates. By putting cohorts into the centreof analyses, the expansion of education provedto be one important source of inequality amongstgenerations and the increasing importance ofeducation for status-attainment can be demon-strated (Mayer and Blossfeld, 1990). The questionof equal opportunities cannot be addressed atany one particular point of the educational career,because the circumstances at its beginning and

(3) Besides the great differences of the PISA results between the Member States, one has to consider important differences in thepublic and in the scientific reactions. By using the OECD database on press reaction (OECD, 2002a) and the number ofpublished press reports in each country as an indicator, the range within the EU ranges from one article (Greece and Luxem-bourg) to about 600 in Germany. More than half of all articles collected for EU countries are from Germany. Belgium andAustria, with around 100 articles, are next. Therefore, the German reaction on PISA is not representative by far for the EU.Compatible data for the scientific debates on PISA are not yet available in Europe.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 335

the process itself are important influences whichhave to be considered (Meulemann, 1990a). Andfinally, individual aspirations, as well as abilitiesand performance, are not stable parametersacross time – they change because of personaland societal development.

As dependent variables, some very importanttransitions into, within and out of the educationalsystem were analysed. The transition from schoolto university and from the educational to theoccupational system was particularly examined.For independent variables – besides the opera-tional definition of education (which in practiseoffers many difficulties, as discussed later) –some personal characteristics (e.g. gender, age,social status, occupation of parents) and struc-tural aspects (e.g. school structure or system,region, class performances) were used. Ingeneral, a very broad spectrum of analyses – verysimilar to the one presented here – was used forlife-course and biographical design in respect ofeducation since the beginning of the 1980s andthis changed at all events the scientific discus-sion on educational topics.

Compared with the tremendous amount ofresearch on education, literature on vocationaltraining and occupational careers seems to berather small. While work is one of the centralthemes in sociology and economics, andlabour-market processes attracted a great scien-tific attention following the labour-market crisis inWestern Europe shortly after the first oil crisis inthe 1970s, the benefits of education and trainingfor the occupational life course are more or lessmarginal topics inside this field. Other contribu-tors will accentuate this kind of research from amacro perspective in the third Cedefop report onVET-research. Therefore, only some shortremarks on the development of social andeconomic research on work, employment, occu-pation, and the labour market with respect to theimpact of life-course and biographical researchare necessary here.

Again, three specific topics should be high-lighted:(a) entrance to the labour market. The increasing

extent of youth unemployment has focusedattention on the circumstances under which thetransition from the educational to the employ-ment system happens (e.g. Farvaque andSalais, 2002; Jahnukainen, 2001; Kortteinen

and Tuomikoski, 1998; Blossfeld, 1985a). Theadaptability of the VET-system to economicneeds, which underwent a great deal of investi-gation from an international comparativeperspective, became an important researchtopic;

(b) unemployment and the chances of re-enteringthe labour market. The increasing number ofunemployed people in Europe also led to thequestion of how to get them into work again (forthe evaluation of active labour-market policyfrom a macroeconomic point of view see thecontribution of Hujer et al. to this report). More-over, to avoid long-term unemployment with itssocio-psychological consequences (asdemonstrated earlier by Jahoda et al., 1933;and re-confirmed by other analyses e.g.Jahoda, 1982; Kelvin and Jarrett, 1985) thistopic became an important political issue. Todevelop adequate measures, scientists (espe-cially economists) became involved in the polit-ical process (e.g. for the EU, the EuropeanEmployment Strategy and the National ActionPlans (NAP) for Employment: EuropeanCommission, 2000, 2003; for Germany also:Senatverwaltung, 1997);

(c) career-mobility and lifelong learning. One of themost important resources for the economy indeveloped countries is the skill level of theemployed. Due to increasing worldwide compe-tition, the pressure to increase individual compe-tences and to adapt them to accelerating innova-tion cycles is growing both for individuals andcompanies. From an individual perspective, life-long learning is increasingly becoming a premisefor job security. However, the flexibility needed toacquire new skills also requires the individual tochange jobs more often than before and to acceptperiodic phases of unemployment. According tothe individualisation thesis of Beck (1994),‘patch-work’ career patterns like this will take theplace of family-like types of company member-ships (for empirical research on the risks of thiskind of job career see for example Andreß, 1989;Büchel, 1992; Felstead et al., 1997;Tuominen, 2000; van de Werfhorst, 2002).

The contribution of life-course and biograph-ical research to these three topics (as for someothers) is again very straightforward. As entrance,or re-entrance, to the labour market is atime-dependent transition (according to the dura-


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 336



tion in the educational system or whether or notemployed), the use of longitudinal data and eventhistory analyses proved to be the only adequateway of analysing these processes. Moreover,life-course and biographical research assumestime outside the employment system not onlydependent on specific historical situations (e.g.labour-market imbalances due to businesscycles) but also on individual experiences andskills assembled through the life course. Botheffects can be separated and controlled for theanalysis of cohort differentiation, which seems tobe the right way to test the ‘patch-work’ careerassumption of individualisation theory as aneffect of structural social change.

In conclusion, both educational andlabour-market research was productively influ-enced by the life-course and biographical perspec-tive and its techniques of analysis. However,life-course and biographical research neverbecame the leading force within this research field.Additionally, the mainstream of life-course andbiographical research followed broader researchinterests other than the benefits of VET for work life.Nevertheless, a small but slowly growing group ofEuropean social scientists uses the increasingamount of appropriate data for this kind of analysis.An overview of the different theoretical assump-tions, scientific perspectives and results of thiswork is presented in the following chapter.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 337

4.1. Life-course and biographicalresearch in Europe

Originating from national social research move-ments in the US and northern and central Euro-pean countries (amongst them France, Germany,Norway, and the UK) empirical life-courseresearch is increasingly starting to locate itsinsights in international contexts. Besides theprogress made in several European countries todevelop an internal professional framework oflongitudinal empirical research, cross-national andinternational cooperation enables researchers tostudy country comparative issues. Current issuesunder investigation are societal trends, compar-isons of different political systems and nationaldifferences in access to public goods, such aseducation. In contrast, the exploration of educa-tional benefits in the life course has received lessattention in European social research, not only incomparative but also in national contexts. More-over, earlier research in countries with a compara-tively long tradition of life-course analysis and ofinvestigation into educational issues like Germany,the Netherlands and the UK, seems to be under-going a shift in emphasis away from research intoeducational impact to a broader investigation ofindividual life chances under certain social condi-tions and developments – where education playsonly a minor role. The reasons for this develop-ment may lie in difficulties regarding the definitionand measurement of education (or, no lessdemanding, training, skills, abilities, qualifica-tions, etc.) which make hypotheses on educa-tional outcomes less attractive than the meredescription of social situations and developments,using education as one explanatory factor amongothers. And indeed, specific educational benefitsare hardly determinable by quantitative measure-ments, because cause(s) and effects are interre-lated, leaving almost no possibility to arrange theeffects in a testable causal order. Keeping theabove in mind, this paper will attempt to providean overview of the limited existing knowledge onthe benefits of education, training and skills in an

individual life course. To be able to do so it isnecessary to include several empirical workswhich give information on the benefits of educa-tion and training in the life course but which stemnot primarily from life-course research as a scien-tific discipline. However, even these studies do atleast take up a lifetime perspective and use longi-tudinal data for analysis. This section will initiallygive an introduction to the thematic interests,approaches, variables and methods utilisedpredominantly in the investigation of educationand training benefits from a life-course perspec-tive. Subsequently, the data employed for empir-ical analysis will be presented. Finally, variousstudies are listed and described along with thecriteria developed in Chapter 2, Sections 4.1.1. to4.1.6. and Section 4.2.

Before proceeding, one important factor needsto be addressed in order to understand themeaning of the term life-course perspectivewithin the framework of research into the impactof education and training. As shown in Chapter 3,life-course research itself is not historicallywithout precedence but has developed from theevolution and convergence of different theoreticalapproaches, faculties and empirical researchareas. The range of theoretical approachesinvolved are, for example, human capital,segmentation and status allocation theories,gender and ageing theories, and theories ofgenerational change. Faculties linked areeconomics, sociology, social psychology, devel-opmental psychology and social demography.Empirical impacts come from mobility research,family cycle observations, qualification and careerresearch, etc. Despite all their inherent differ-ences this mix has one main thing in common:the investigation of dynamic processes.Life-course research is to be understood as aninstrument to catch these dynamics, rather moreas a perspective than an independent and unitedtheory. Therefore, it is more an enlargement(mainly in an empirical sense) of other theoreticalapproaches than an alternative. In general,life-course and biographical research haverecourse to one or other of these theoretical

4. Empirical evidence

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 338



approaches. In the following illustration of themain lines of investigation of life-course researchinto the benefits of education and training, thelife-course perspective should be understood aslying transverse to theories of educational impactand outcome.

4.1.1. Monetary returns on education and

training and life-time income

One of the most investigated benefits of educa-tion and training in social research is ‘earnings asdependent variable’ of education and training (4).As in all outcome oriented research into educationand training, a good deal of the investigation ofeducation and training benefits from a life-courseperspective is dedicated to the measurement ofindividual monetary returns of education, trainingor skills. The term ‘individual returns’ refers to theprivate rate of return in terms of higher incomeresulting from prior individual investment ineducation and training (5). Empirical research,therefore, investigates the impact that education,training and skills have on income as one of themost important material benefits. In doing so,most of the outcome oriented education andtraining research, including life-course research,refers to human capital assumptions. In brief,human capital theory assumes a direct relationbetween the accumulation of education in termsof years or levels of education and the rise ofcurrent or life-time income (Mincer, 1997, 1964;Sweetland, 1996; Becker G., 1993, 1964).According to this theory, employees are paidsolely for their productivity which on its partdepends entirely on their qualifications achievedthrough education and training. Therefore, thebasic hypothesis is that the more a person investsin his or her education the higher his or herproductivity and the higher the individual rates ofreturn in terms of higher income (Becker andSchömann, 1996).

The OECD defines human capital as follows:human capital refers to ‘the knowledge, skills,competences and attributes embodied in individ-uals that facilitate the creation of personal, socialand economic well-being.’ (OECD, 2001b, p. 18).

The main critical point of the human capitalapproach is that it implies a labour marketwithout any restrictions or discrimination and arational individual who is perfectly informed abouthis or her options (Alewell, 1993; Franz, 1991;Katz and Ziderman, 1990; Sengenberger, 1987).On the other hand human capital research hasproved to be (at least partly) successful in esti-mating income effects of education and trainingand in explaining variations in individual rates ofreturn, e.g. regarding country-specific conditions(Asplund and Pereira, 1999). Moreover, recentdevelopments show that vigorous efforts arebeing made to improve the original approach byconsidering several interceding factors that mighthave an impact on the relationship betweeneducation and training and income.

As early as the 1980s, Tuma (1985) developedan advanced human capital model, which hadbeen modified from the life-course perspective. Inher analysis of occupational careers she consid-ered career courses based on ill-informed deci-sions and unbalanced labour-market conditions.She estimated a resulting risk of misallocationthat could be corrected or compensated for byjob-specific training received by the individual,which would again offer returns from investmentin education, training or skills.

Analogue enhancements of the human capitalapproach can be viewed in the so-called filterand signal theories which assume individuals actunder uncertainty and impaired transparency inthe labour market. According to job-market filterand signal theories, educational credentials actas mere signals or filters in the job-matchprocess (Spence, 1973; Arrow, 1973), separatingindividuals with higher ability (= higher educa-tional level) from the rest. However, these signalscan turn out to be wrong, because a highereducational level does not mean necessarily leadto higher productivity.

In summary, despite several attempts toadvance human capital research in order to bringin the life-course perspective, only a small numberof empirical investigations from this economicscientific school utilised longitudinal life-course

(4) We speak of earnings instead of income, because earnings describes the payments someone receives from his or heremployment (and which is influenced also by education and training), whereas income refers to the total money at the disposalof an individual, that is earnings plus money gained otherwise, e.g. rents, dividends, interest, etc.

(5) Generally, the rate of return to education is defined as the extra income earned as a result of attaining one additional year orlevel of education (Harmon et al., 2001) and discounted to the time of entrance into the labour market.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 339

data for the estimation of long-term educationalreturns. In contrast – maybe surprisingly – empir-ical studies which argue from a life-courseperspective (see theoretical assumptions inChapter 3) examining education and training bene-fits are based to a respectable degree on humancapital assumptions of educational returns.

4.1.2. Education, training and labour-market

participation

Apart from these income-oriented estimations ofindividual monetary returns from education andtraining, life-course research looks at a range ofdifferent facets of an individual’s participation inthe labour market which are viewed as beneficialoutcomes of education and training. Amongthese are transitions from education and trainingto work and mobility between and within jobs,employment prospects, avoiding unemploymentand reacquiring employment, obtaining qualifica-tion-adequate jobs, as well as participation incontinuing education and training programmes.To sum up, any kind of labour-market events canbe used (and have been used to a certain extent)as dependent variables in life-course andbiographical research.

From a life-course perspective success orfailure at different times in an individual’s workinghistory have an impact on present and latersuccess or failure, not only regarding the furtherworking career but also far beyond it. An indi-vidual’s position in the labour market, for example,has an impact on that person’s social status, notonly in the present but also – and maybe moreimportantly – at the time of later status passages.Life-course research therefore deals with directpossible outcomes or benefits from education andtraining such as successful job entry after finishingeducation and training as well as with indirect, orlater outcomes, which are mediated through thesedirect benefits. Access to continuing educationand training for example is viewed as an indirect,later benefit of initial education and training,because it is assumed that education and traininghave a major impact on labour-market entry. Thisin itself determines the later occupational careerand therefore the access to continuing educationand training. The same could be said for otheraspects of an individual’s career. Of course therelationships are not as deterministic as purportedhere, many different factors of a person’s educa-tional and occupational career interfere with each

other. But the central principle of coherencesshould have become clear.

As the concrete subjects of investigationrelated to labour-market participation are mani-fold so are the applied approaches. Therefore,life-course research utilises a number of differentlabour-market related theories which range fromstatus attainment approaches through segmenta-tion theories and human capital to mere descrip-tive research. Applying the life-course perspec-tive to educational outcomes or benefits relatedto the labour market only means that differenttime points in an individual’s working history mustbe associated with important differences in thequantity and quality of these benefits. Therefore,despite all the differences in the theoreticalapproaches, life-course research points to thefact that neither individual behaviour nor socialsystems are static by nature. From this it followsthat statements on a society or its subsystemscan be made only with reference to certain pointsin historical time.

4.1.3. Education and transitions

From the life-course perspective, transitions havea special relevance for the individual working lifehistory and are therefore highlighted here as aspecial case for bringing possible benefits toeducation and training. Successful transitionsper se do not yield benefits to education andtraining, but the other way around: successfultransitions follow (as a benefit) from educationand training. The determination of life histories byinstitutional structures in modern societiesproduces a cycle of certain phases and transi-tions, which are of a very different character.These institutional structures are mainly definedby the educational and occupational system of acountry which, through their organisation,account for the existence of sensitive and lesssensitive phases in an individual’s educationaland occupational career (Blossfeld, 1989). Transi-tions (from school to work, to further education,etc.) in this sense are particularly sensitive orfixed phases, because they are more susceptibleto internal and external influences (recessions,educational expansion, etc.). Periods betweentransitions, for example after the decision topursue a certain job or training, are correspond-ingly less susceptible to change. As different lifephases constitute an interrelated course whereearly phases can be seen as relatively formative


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 340



for later periods it is assumed that early transi-tions are of high importance for later states andtransitions and are open only to limited correc-tion, for example by further/continuing training.

Particularly in countries with a strong emphasison formal qualifications, highly differentiatedprofessions and a low permeability of occupa-tional and educational systems, successful orunsuccessful transitions from education totraining and/or the first job are believed to beessential for the whole further educational andoccupational career of an individual (Mayer, 1996).An effective first placement on the labour marketcan therefore determine an accumulation ofeducational benefits such as high wages, a quali-fication-adequate job, good career prospects, etc.An ineffective job placement is likely to lead to anaccumulation of disadvantages in the further lifecourse. Blossfeld (1989) first formulated the thesisthat once a disadvantage is experienced in thetransition from education to training and/or towork it cannot be compensated for over the wholeoccupational career. Therefore, from the perspec-tive of life-course research, transitions – especiallythose that occur in the early phases of the lifecourse and the individual working career – arecrucial for the analysis of educational benefits.The associated material and non-material returnsare not only believed to constitute benefits inthemselves but also to account for a goodproportion of chances and benefits in the subse-quent life course. Besides the outcomes of earlytransitions from education to training and/or work,the consequences of later transitions, upwardmobility and generally all forms of mobility leadingto the avoidance of unemployment are seen asimportant possible benefits of education andtraining in the life course. On the one hand,mobility in later career is likely to be determinedby early transitions and therefore in terms ofupward mobility it is as a rule an indirect beneficialoutcome of education and training. On the otherhand, different forms of training or continuingeducation on or off the job can lead to upwardmobility as a direct benefit.

4.1.4. Generational differences in education

and training benefits

Besides the impact of education and training onso-called ‘intra-generational’ mobility, asdescribed in the previous section, life-courseresearch has a special interest in inter-genera-

tional mobility. The social (upward) mobilitybetween generations can be determined bycomparing the occupational status of childrenwith that of their parents (Hradil, 1997;Goldthorpe, 1980; Featherman andHauser, 1978). It is assumed that there are largedifferences between generations regarding theallocation of life chances and risks and in gainingbenefits from education and training. Through thecomparison of successive generations or birthcohorts it becomes possible to follow socialchange, specifically the change of educationaland occupational trajectories and how they aredetermined by institutional changes over histor-ical time. A birth cohort consists of a group ofpeople who are of homogenous age and there-fore share the same significant life event(s) atalmost the same historical date.

Therefore, the life-course perspective arguesthat different generations are confronted withdifferent social and historical conditions and evenbirth cohorts which are only a few years apartfrom each other face these conditions in different(sensitive or less sensitive) phases of their life.According to this, people belonging to differentgenerations or birth cohorts are exposed to verydisparate chances and risks regarding theireducation, training and working career and there-fore receive a very different quantity and qualityof education and training benefits. Analytically,these differences can be assigned to threeeffects: the age of a person within a cohort, thecohort to which a person belongs and the frame-work conditions a person faces in a certainperiod of time (war, high unemployment, etc.) asa consequence of his or her belonging to acertain birth cohort (for a detailed description ofage, cohort and period effects see Descy andTessaring, 2001; Part 5, Sections 1.3.2.1. and2.3.1.). Taking up again the thesis that a disad-vantage once experienced during a particular lifephase can hardly be compensated for over awhole (working) life, it becomes obvious that notonly the individual’s fate is influenced – to what-ever extent – by this, but also the fortune ofwhole generations or birth cohorts. Earlylife-course research confirmed the ‘entry-job’thesis by demonstrating how the different educa-tion and training benefits for older and youngercohorts were determined by having been facedwith quite different historical conditions such as

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 341

the post-war period, the economic boom in the1950s and the beginning of educational expan-sion in the 1970s (Blossfeld, 1989). In the presentday, studies into the effects of cohort and gener-ational differences on the benefits of educationand training constitute an important research lineof life-course analysis. The increased attempts atcross-national comparisons of education andtraining benefits make it particularly essential toconsider generational change. A separation ofcross-national from historical comparison would,in any case, lead to misinterpretations of coher-ences in the analysis of educational and occupa-tional pathways.

4.1.5. Social differences in education and

training benefits

Besides cohort and generation specific differ-ences regarding the gaining and utilisation ofeducation and training benefits, social determi-nants of education and training benefits inrespect of ascriptive attributes like family back-ground, gender, ethnicity, etc., are importantinterest fields for life-course research – as forevery empirical social science discipline. Genderand ethnicity have been recognised asoutstanding factors in the access to andoutcomes of education and training. Althoughthere has been a certain equalisation of life-chances for men and women in many respectsover the last 30 years, gender still matters in theutilisation of human resources like education andtraining. From a life-course perspective, genderdifferences in education and training benefits notonly remain but increase and accumulate over thelife courses of men and women. Family back-ground is a further important factor in the analysisof differences in the attainment of education andtraining benefits. Generally, it is acknowledgedthat family background determines the chancesof educational participation and attainment in avariety of different ways. From a static position,we see that children from families with a compar-atively high income and social status usually havebetter chances to achieve higher levels of qualifi-cation and are over-represented in tertiary educa-tion and under-represented in the group of earlyschool leavers or in lower vocational education.From a life-course perspective, it is assumed thatthese inequalities are carried further not onlythrough the whole individual life course but also

by the reproduction of inequality via successivegenerations. Therefore, studies of family back-ground and gender differences are oftencombined with studies of generational and cohortdevelopments.

4.1.6. Subjective perception of educational

benefits in the biography

Research into life-course perspective investigatingeducation and training benefits as described showonly the objective aspects of the impact of educa-tion, training or skills on several outcome vari-ables. This view has been criticised because theindividual perspective, and therefore the subjectivemeaning of learning and occupational biographies,is more or less ignored. As judgements on andsubjective interpretations of the individual educa-tional biography are likely to have importantimpacts on the working and life careers of individ-uals, they cannot be ignored from a life-courseperspective. Compared to the demographiclife-course perspective which focuses on the lifecourse as a unit, the biographical perspective issubject oriented and concentrates on the differ-ences in and specialities of individual learning andoccupational pathways. Therefore, studies ofsubjectively perceived benefits in individualbiographies are becoming increasingly interestingand help supplement rather objective life-coursenotions. Furthermore, in consequence of thissubject oriented perspective, biographicalresearch which is concerned with individualeducation and training benefits often includesresultant ‘soft’ or non-material beneficial aspectsand thereby complements life-course research on‘hard’ material education and training benefits.

The research topics presented here are viewedas the main lines of investigation of educationand training benefit measurement from alife-course perspective in Europe. The nextsection outlines the empirical research conductedfrom one or more of these lines. Since studiesfollow their own logic according to their subject(s)of investigation, the classification chosen tosubdivide this research is rather artificial. In orderto give an overview of results from life-courseresearch it seemed nevertheless appropriate touse such means. The description of studies asclassified in Section 4.3. is therefore not to beunderstood as a scientifically derived categorisa-tion but rather as a review tool.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 342



4.2. Research design and data

The perspective of life-course and biographicalresearch has some important implications forempirical methodology and data requirements(Chapter 3). As the focus of investigation is on thewhole life course of individuals rather than on asingle point in time, cross-sectional measurementsdo not serve the required purpose so a longitudinaldesign is needed. The term ‘longitudinal’ can benarrowed down to research, which allows adiachronic analysis of the incidence of conditionsand events by collecting data for each item from atleast two distinct time periods. Longitudinalresearch involves analysis of the same individuals,or more generally cases, over different time periods;and involves some comparison of data betweenand among periods (Menard, 1991; Ruspini, 1999;Mayer, 1987). In contrast to comparative-staticcross-sections, longitudinal analysis can thereforecontrol for cohort-, period- and age-effects thatmight have a substantial impact on the relationshipbetween education and training and assumedoutcomes. Furthermore, longitudinal data normallycontains a broader spectrum of individual charac-teristics than cross-sections and therefore allowsfor a more reliable analysis of interactions.

There are a number of different designs inlongitudinal analysis from which, for the purposeof life-course research, only three are of interest:prospective designs; retrospective designs andfollow up designs.

While in surveys with a prospective design thesame individuals are interviewed repeatedly indefined periods from one common starting point,retrospective studies collect data via life-historystudies that cover the past life course of therespondents. Retrospective studies have theadvantage that they provide very detailed andprecise information (Blossfeld and Rohwer, 1995),

but are less reliable in comparison to prospectivestudies data on punctual events and do not offerthe same strength for research on causalprocesses. Furthermore, they cannot includeopinions or plans for the future. Therefore, retro-spective surveys are often addressed as‘quasi-longitudinal’ (Hakim, 1987; Ruspini, 1999).Follow-up studies can be considered as a specificform of prospective surveys concentrating on thepursuit of selected offset groups, in our casemainly school, training or programme graduates.A further central point for life-course research is todivide between panel data and cohort data.

Panel surveys trace individuals of different age atfixed points in time and thereby offer the possibilityof exploring individual change, for example ineducational attainment, job mobility, etc. Surveyswith a cohort design can be considered as a specificversion of panels that aim to give information ongenerational differences or the process of genera-tional change. A cohort is defined as a group ofpeople within a population who share the samesignificant life event(s) in a given period of time.Thus a cohort typically consists of people of thesame age group, meaning that they share the samedate of birth, year or period of years (e.g. 1920-25).Nevertheless, physical age is not the only criterionfor constructing cohorts. School or job entry andother shared events can be used to form coherentcohorts as well. Whereas panel surveys trace indi-viduals over time, the observation unit of cohortsurveys is a group of people sharing the same lifeevent(s). It should be emphasised that the distinc-tion of panel and cohort surveys lies transverse tothe other three categories which leads to fourdifferent longitudinal designs: prospective panel(PP) surveys, retrospective surveys based on indi-vidual data (R), prospective cohort (PC) surveys,retrospective cohort (RC) surveys. The design offollow-up studies does not essentially differ fromother prospective data collections (Figure 9).

Figure 9: Prospective data designs

Observation unit/reference group

Longitudinal design

Prospective

Retrospective

Individual

PP

P RC

PC

Cohort

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 343

Well-known PP are household panel surveyswhich are well established in most Europeancountries. Among them the most reliable arepresumably the British household panel survey(BHPS), the Dutch and the German socioeco-nomic panel (SEP; GSOEP). However, thecomparability of these data sets is limitedbecause of variations of variable construction. Amajor step towards overcoming these problemswas the invention of the European Communityhousehold panel (ECHP) in 1994. It has alreadybecome a unique basis for cross-sectional andperiodical comparison of education and trainingbenefits in 12 European countries.

An important example of retrospective surveysbased on individual data (R) is the GLHS that alsoserves to exemplify retrospective cohort studies,because data has been collected for individualsbelonging to subsequent birth cohorts. In thesurvey, about 8 000 persons belonging to cohortsof around 10 years apart (1919-21, 1929-31,1939-41 up to 1971) were interviewed in greatdetail about their life courses. A further example ofRCs is the Norway Survey (NORS) in which acohort of people born between 1956 and 1958was interviewed retrospectively. PC designs arebest represented by the British National ChildDevelopment Study (NCDS) and the 1970 BritishCohort Study (BCS70). In their detailed format andsample size the cohort studies in the UK provide aunique database for single cohort research inEurope. However, the great disadvantage of thesesurveys is that each of them covers only onecohort of individuals. Apart from the comparison ofthese two different data sets, no comparisons canbe made regarding other cohorts.

In contrast to cross-sectional analysis ofeducation and training benefits, which can

already draw on a substantial selection ofharmonised databases, longitudinal research hasno similar means of comparing education andtraining benefits in Europe. Apart from the ECHP,which contains only limited information on educa-tion, training and skill-related issues, no longitu-dinal survey on a European level exists. Asalready pointed out, the comparability of countryspecific data remains quite limited and resultsneed to be treated very carefully. Table 1 containsthe most important individual data sets utilisedfor life-course research in European countries.

Beyond this roughly presented differentiationof longitudinal data, a further more sophisticatedclassification is generally needed when dealingwith empirical social research. When comparingresults from studies of varying concept andresearch design we should if possible considerthe following dimensions of alternative empiricalmethodologies:(a) data sources: survey/questionnaire – obser-

vation – written material – mix – register;(b) data analysis: quantitative – qualitative;(c) data collection: self-collected – secondary

analyses;(d) sample: representative sample/non represen-

tative sample;(e) level of aggregate: regional – national –

cross-national.In the next section studies are presented which

apply a life-course perspective in the analysis ofeducation and training benefits. Due to the limitedscope of this study, it is not possible to give acomplete overview of empirical investigations.Nevertheless, a good number of national as well ascross-national and European studies are listedaccording to the dimensions proposed. In addition,selected studies are explained in more detail.


The creation of the ECHP was a major step towards cross-European data comparability. 1994: launch by Eurostat (Statistical office of the European Union): parallel surveys were taken in12 Member StatesFollow up: every yearRestricted time frame: so far only three waves availableAs compensation the EPAG (European Panel Analysis Group) has produced a longer data set fromthree household panel surveys (Germany, the Netherlands, the UK)EPAG data set has been available since 1991, accessible through ISER (Institute of Social andEconomic Research, UK)

Box 2: Eurostat’s ECHP and EPAG data set

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 344



Survey Country Type Aims Sample Time ofmeasurement

British household panel UK PP C Representative sample of British annually, survey (BHPS) households since 1991

National child development UK PC B Cohort of people born from 1965, 1969,study (NCDS) 3 to 9 March 1958 1974, 1981,

1985, 1999

Youth cohort study (YCS) UK PC B Representative sample regularly, of 10 cohorts aged 16+ since 1985

1970 British cohort study UK PC B Cohort of people born from 1975, 1980, (BCS70) 5 to 11 April 1970 1986, 1996,

1999/2000

Irish panel IRL PP B Representative sample of Irish annually, households 1987-1995

Dutch socio-economic NL PP C Representative sample of annually/panel (SEP) Dutch households biannually

since 1984

Brabant cohort study NL PC A Representative sample of a cohort 1952, 1983, of pupils who in 1952 were in the 1993sixth grade

OSA (Organisation for NL PP B Representative sample of individuals regularly,Strategic Labour Market aged 16 to 64 1985-1996Research) panel

Explanation of NL PC B Cohort study, N= 18 973 regularlysocio-economic inequalities in health (GLOBE)

German life history study D RC B Representative sample of cohorts retrospective(GLHS) born in 1929-1931, 1949-1951,

1954-1956, 1959-1961, 1971

German socio-economic D PP B Representative sample of the annually, panel (GSOEP) German population since 1984

Life chances, careers and D PC A Representative sample of 1989 1989-2001delinquency of low level school leavers from German school leavers Haupt- and Sonderschule

Malmo study (MS) S PC B Original sample size 1 542 children regularly, at the age of 10 1937-1988

Malmo longitudinal study S PP C Sample of all households receiving 1995-1999of social assistance (MLS) social assistance in Malmo

Evaluation through follow-up S PP A Original sample size 10 000 children regularly, at the age of 10 and 13 since 1961

Swedish level of living S PP B Sample of about 6 000 individuals 4 waves,survey (LNU) between ages 16 and 75 1968-1991

Norway survey (NORS) NO RC B Cohort of people born from 1956-1958 retrospective

Norway level of living NO PP B Sample of Norwegian adult 5 waves, survey (NLLS) population 1980-1995

Danish longitudinal DK PP B Representative sample of the annually, database (DLDB) population in the age group 16-75 1976-1990

drawn from admin. registers

Zuercher longitudinal study: CH PC A 394 school leavers in Switzerland regularly,from school to middle since 1978adulthood (ZLSE)

Transitions from education CH PC A National PISA 2000 sample leaving 2001-2003to employment annually

2004-2010 biannually

Table 1: Longitudinal data sets for research into the impact of education and training in Europe

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 345

4.3. Empirical Studies

Most of the existing studies referring to educationand training benefits from a life-course perspec-tive are naturally based on national data setswhich have been carried out by statistical officesor large research institutes. The reason for thisconcentration on secondary analysis lies in theamount of capacity and financial resourcesneeded for representative longitudinal investiga-tion. Indeed, only a few European countries areeven able to provide data material, which servesthe demands of life-course analysis, amongstthem primarily Germany, the Netherlands and theUK. About two thirds of relevant life-coursestudies on education and training benefits, there-fore, draw on data from these countries (6).

Among the data sets used most for secondaryanalysis of education and training benefits are theBritish, Dutch and German representative Panels(BHPS, SEP, SOEP), the British cohort studies(NCDS, BCS70, YSC) and the GLHS. The Dutchcohort studies (Brabant, GLOBE) and Swedish

and Norwegian panel and cohort data (MS, MLS,NORS) are also often used for country specificanalyses. In total, there are around 60 differentusable studies dealing in the broadest sense witheducation and training benefits from a life-courseperspective. Compared to cross-section analysisof individual data this might be a small numberbut even given this limitation the sheer amount ofvarying conceptual approaches, specific investi-gation interests, definitions of explanatory anddepending variables, sample sizes and above allresults makes it very hard to provide a compre-hensive overview.

Given the heterogeneity of study concepts,selected national and cross-national works areclassified according to the dimensions proposedin the previous parts of this report. Generally,education and training research from a life-courseperspective is prevalent in the northern andwestern Member States while there is only littleevidence from southern or eastern Europeancountries. On this basis, any generalisation orattempt to compare results at European levelmust be handled with care (7). Furthermore, there


Panel study of Belgian B PP B Representative sample of Belgian annually, households (PSBH) households since 1992

Luxembourg panel L PP B Representative sample of annually,Luxembourgish households 1985-1995

Survey on class structure, E R C Sample of 6 629 individuals, retrospective,consciousness and not fully representative regarding 1991biography (ECBC) region and educational level

Greek household panel EL PP B 2 952 Greek households every two years, 1988-

Hungarian household HU PP B 2 000 Hungarian households annually,panel (HHP) 1988-1997

European Community EU PP B Representative sample of annually, household panel (ECHP) households in 12 Member States since 1994

Types: PP = Prospective panel data;PC = Prospective cohort data;R = Retrospective individual data;RC = Retrospective cohort data.

Aims: A = Main objectives are the impact of education and/or training;B = Main focus is on other issues, but these are related to the impact of education and training;C = Focus is on other issues, but some education and training variables are included.

(6) Even the oldest longitudinal panels (British household panel, German socioeconomic panel, and Dutch socioeconomic panel) allhad their starting point in the mid-1980s. Therefore, the period of measurement is limited to at least 20 years and covers only apart of an individual’s life course. Although the panels sometimes enclose records on educational and occupational histories, theamount of information collected in this retrospective section is limited and therefore concentrates on main life events.

(7) The limited study material detected in the southern European countries might be also in part due to the fact that most of themstill lack a working publication and communication system of research groups, communities, journals, databases and otherdispersive media.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 346



is a clear concentration on material benefits(primarily income and employment patterns) andthere are few studies on non-material benefits.

4.3.1. Studies at national level

Life-course studies, which are concerned withindividual education and training benefits on anational level, are of a very diverse character andcannot really be compared. The reason for thislies not only in the difficulty of comparing resultsfrom studies with different definitions of explana-tory and dependent variables but in the entirelydiverse research designs of these studies.Depending on the particular research lines andtraditions existing in different Member States thefocus of investigation varies to an extent thatactually allows for no serious comparison ofresults at all. Table 2 shows selected nationalstudies and areas of current research related toeducation and training benefits.

Studies related to human capital approachesmeasuring longer term rates of return from years ineducation, level of education or type of training canbe found in most European countries (Steiner andLauer, 2000; Brunello and Miniaci, 1999; Uusi-talo, 1999; Chevalier, 1999; Hægeland, et al., 1999;Denny and Harmon, 1998; Curti, 1998; Hartoget al., 1993; Goux and Maurin, 1994; Pedersenet al., 1990). In contrast, other studies are not aswidespread. Current research on individual workhistories or labour-market participation over the lifecourse is concentrated in Germany, Scandinavia andthe UK. The explanatory factors related to educationand training in this research are normally limited tothe level and/or the number of years spent in educa-

tion and/or training (Müller and Shavit, 2000;Wolbers, 2000; Hujer and Wellner, 2000; Bryson andLissenburgh, 1996; Hammarström, 1996; Beckerand Schömann, 1996; Kettunen, 1994).

Few studies analyse dynamic factors such asincome development or changes regarding educa-tional returns over the life course. An example is thestudy of Steiner and Lauer (2000) who examinedreturns on education in Germany from 1984 to 1997on the basis of the GSOEP and showed that ratesof return are not constant over the life cycle, espe-cially for women who experienced a substantiallyincreased rate of return until 1994 and subse-quently a continuing decline. An analysis of longitu-dinal data from France and Spain confirms thisresult, but also displays national differences. Whilein France educational returns already start todecline after five years of job tenure, in Spain returnsdo not decrease until 20 years of tenure. Unfortu-nately the authors of the article do not give any infor-mation on the data sets used for their analysis (inprogress) (Barceinas-Paredes et al., 2001). Beckerand Schömann (1999, 1996) estimate the long-termeducational returns of further vocational educationon the basis of the GLHS and the GSOEP anddetect a tendency to lowering rates of furthereducation returns in Germany but also continuingincome growth resulting from participation infurther education, especially for men. A verydetailed analysis of the impact of further on-the-joband off-the-job training on the working career isprovided by Pannenberg (1995). Besides themeasurement of income development, he looks atindividual job mobility as a possible benefit ofeducation and training and finds that short (two to

Study Data Data Sample Longitudinal Outcome Explanatory BenefitSource analyses size* design variables** variables*** (result)****

aggregate

Pannenberg (1995) GSOEP (D) secondary/ 1965 PP national OC, ID FVET materialquantitative

Hillmert (2002) GLHS (D) secondary/ > 1 000/C RC national OC LE, TQ material quantitative (+/0)

Diewald and GLHS (D) secondary/ 2 323 RC regional OC LE material (+)Sørensen (1996) quantitative SQ 45 % P2 PC

Blossfeld and GSOEP (D) secondary/ n.n. PP national M LE, YE, FB non-Timm (1997) quantitative P11 material

Born (2000) SFB 186 (D) self-collected/ 2 130 RC national OC TQ materialquantitative

Table 2: National studies on education and training benefits from a life-course perspective

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 347


Becker (1998) SOEP/GLHS secondary/ > 14 000/1 781 PP/R national LEx YE non-material(D) quantitative (+)

Becker and SOEP/GLHS secondary/ 3 075/2 171 PP/R national ID FVET material (+)Schömann (1999) (D) quantitative

Becker and GLHS (D) secondary/ > 1 000 RC national ID FVET material (+/-)Schömann (1996) quantitative

Bender and Social secondary and 2 674 RC national OC, UR LE material (+)Dietrich (2001) insurance self-collected/

data quantitative

Bynner and NCDS/BCS70 secondary/ 11 400/9 000 PC/PC UR YE/S material/Parsons (2001) (UK) quantitative (10 % last) national non-material

P37/P21

Bynner et al. NCDS/BCS70 secondary/ 11 400/9 000 PC/PC ID, OC, H S material/(2001) (UK) quantitative (10 % last) national non-material

P37/P21

Dale and Egerton NCDS (UK) secondary/ 11 193 PC national OC, FF, ID LE/TQ/FB material (+/-)(1997) quantitative P33

Payne (1995) YCS (UK) secondary/ n.n. PC national EC LE, TQ materialquantitative

Schallberger and ZLSE (CH) self-collected/ 1 706 PP (follow-up) OC, PD LE/VET/TQ material/Spiess Huldi (2001) quantitative SQ 23 % P21 national non-material

Unwin and MA (UK) self-collected/ 120 PP (follow-up) non VET/TQ/S material/Wellington (2001) quantitative national non-material

and qualitative (subjective)

Meyer TREE/PISA secondary and 6 500 PP (follow-up) EC, OC LE material(not concluded) (CH) self collected national

Wolbers (2000) OSA (NL) secondary/ 10 514 PP national UR LE material (+)quantitative P14

Skov (1998) (DK) self-collected/ 5 100 PP (follow-up) EC LE/VET materialquantitative national

Koivuluhta (FIN) self-collected/ 1 cohort of PP (follow-up) EC, OC LE/VET/TQ material/(1999) quantitative training leavers regional non-material

P16

Nummenmaa OSLM (FIN) secondary 1 732 PP (follow-up) EC, OC, PD LE/VET material/(1996) P10 national non-material

(+/0)

Gash and (IRL) self-collected/ n.n. PP (follow-up) EC, OC, ID LE material (+)O’Connell (2000) quantitative national

Keogh and individual self-collected/ 114 R national non FVET material/Downes (1998) autobiographical qualitative non-material

summaries (IR)

* Sample size SQ: survival quota of respondents (where available);P: period of measurement in years.

** Outcome variables: EC: educational career; M: marriage;FF: family formation; OC: occupational career;H: health; PD: personal development;ID: income development; R: rate of educational return;LEx: life expectancy; UR: unemployment risk.

*** Explanatory variables: E: general education; S: skills;FB: family background; TQ: type of qualification (subject);FE: further education; VET: vocational education or training;FVET: further vocational education/training; YE: years of education or training.LE: level of education or qualification;

**** if possible

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 348



seven days) as well as longer (one week to onemonth) training on-the-job increases the possibilityof upward mobility significantly. Shorter trainingperiods tend to have a more positive impact onin-company mobility whereas longer trainingprogrammes rather increase inter-companymobility. The medium-term income development of‘job-stayers’ (1986-91) is positively affected byon-the-job training but this is not true for‘job-changers’. This might be due to the fact thaton-the-job training is normally closely related to therequirements of the specific job. Off-the-job trainingafter having become unemployed significantlyreduces the risk of staying unemployed but only ifprogrammes are medium-term (6 to 12 months).Long-term programmes (over 12 months) incontrast increase the risk of staying unemployed. Astudy by Fitzenberger and Prey (1999) shows thattraining-on-the-job also increases job stability(GSOEP 1984-97). Wolbers (2000) analyses theeffects of educational level on unemployment riskand the chances of reemployment on the basis ofthe OSA Panel and finds a continuous decreasingunemployment risk with higher levels of educationfor the Netherlands (1980-94). Also the years of jobexperience have a decreasing impact on the proba-bility of becoming unemployed. Furthermore, heconcludes that higher qualified individuals havebetter chances of regaining employment than thoseunemployed without higher qualifications.

However, examined in detail, studies on educa-tional returns and individual working careersreveal no sign of convergence of educationalbenefits across the European countries. Forexample, while Scandinavian countries especiallyhave rather low rates of educational returnsIreland and the UK show high rates over time(Harmon et al., 2001; Asplund and Pereira, 1999).Central European countries are located some-where in-between these two extremes. Some ofthem show a downward trend in rates of return(like Germany and Austria) whilst other remainstable or even show slight upward trends (Italy,Portugal) (Chevalier, 1999; Brunello andMiniaci, 1999; Goux and Maurin, 1994). Othereducation and training benefits however need notbe distributed in the same way. The correlationbetween education and training and employment

for example is very strong in Belgium, Germany,the Netherlands and Scandinavian countries whileit is moderate in Spain and the UK and muchlower in Portugal (Hannan and Werquin, 2001).However, apart from these national differencesthere remain some European-wide general educa-tion and training benefits. These are:(a) the nearly universal material benefits of more

education in terms of higher entry wages,higher wages throughout the working life,increased job opportunities and a reducedrisk of becoming unemployed;

(b) an indication of potentially high returns ofmore education for socially disadvantagedindividuals who reached only a low level ofeducation (8).

Beyond the investigation of educational returnrates and other working career related educationand training benefits, there is a continuing boom ofparticular transition and mobility research in manyEuropean countries. This can presumably berelated to the worsening condition of the labourmarket for young education and training graduatesand the unemployed in particular. National longitu-dinal data sets have been utilised widely to studythe processes of transitions of young people fromeducation and training to working life. See the first(Tessaring, 1998) and second (Descy andTessaring, 2001) research reports of Cedefop fordetails. In the absence of a longitudinal survey atEuropean level, national research nevertheless indi-cates some differences concerning the job-entry ofeducation and training leavers (Liefbroer, 2002;Gash and O’Connell, 2000; Fabrizio, 2000; Koivu-luhta, 1999; Bratberg and Nielsen, 1998;Breen, 1995; Arum and Shavit, 1995; Bynner, 1995).Transitions seem to be easier in Denmark, Germany,the Netherlands, and Austria than in southern Euro-pean countries like Greece, Spain or Italy. InBelgium, France, Ireland and the UK young peoplealso face serious difficulties in their transitions buttake up an average position in the EU regardingyouth unemployment rates. There is some evidencethat countries with very well coordinated educa-tional and occupational systems (well-definedoccupational labour markets and a high standardi-sation of education and training systems) have thebest transition rates. Nevertheless, only a small part

(8) One should remember at this point the limitation of comparative research based on national surveys. The comparability isgreatly impaired by use of different data sets, definition of variables and differences in estimated model specifications.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 349


of transition research has been carried out from alife-course perspective. Most of the studies look attransition points separately, especially those fromeducation and training to work or to further educa-tion and training, ignoring the fact that that transi-tions are embedded in the framework of an indi-vidual’s life history and can only be interpretedproperly when considered independently. Further-more, from a life-course perspective transitions, assensitive phases (Blossfeld, 1989) in the life course,are of special relevance for the entire working careerof an individual (Section 4.1.3). As most studiescover only the single transition points(cross-sections) or at most the first years of jobtenure (follow-up), there is little to extract from themfor further life-course analysis. One exception is thestudy of Schaeper et al. (2000) who have examinedcareer development for up to eight years after thetransition from VET to work in Germany. They findthat successful career development dependsdirectly on the occupation in which an individual isfirst trained and indirectly on the educational back-ground of the individual. Bender and Dietrich (2001)confirm the long-term incidence of successful orless successful transitions from education andtraining to work. Furthermore, cohort differenceshave an important impact on the stratification ofeducation and training benefits.

Studies that have examined the effect of cohortand generational differences on education andtraining benefits constitute an important researchline in life-course analysis. One might say thatlife-course research has almost rediscovered therelevance of individual affiliation to a generation orage cohort as a social factor in determining one’sopportunities in life (Mayer, 1987). Studiescomparing the different benefits accrued througheducation and training resulting from differentgenerational or cohort patterns focus mostly on agegroup differences of material (educational returns,labour-market participation, working life) andnon-material (family formation, health) benefits.Generational differences have been intensivelyinvestigated in Germany, the Netherlands, Norway,Sweden, and the UK thanks to existing cohort datasources. Scandinavian cohort studies have empha-sised health related benefits and often stem frommedical research (Gridley et al., 1999; Veierød

et al., 1997; Hall et al., 1993). Life-course researchin Germany, the Netherlands and the UK hasconcentrated more on material and othernon-material benefits (Konietzka, 2002, 1999;Hillmert, 2002; Schrijvers et al., 2001; Sackmann, 2001; Born, 2000; Boockmann andSteiner, 2000; Becker, 1998).

Becker (1998), on the basis of GSOEP andGLHS data, looks at the education and lifeexpectancy of successive birth cohorts whilecontrolling for other social determinants, forexample stratum. As well as confirming existingresearch that tends to show a causal linkbetween higher educational attainment inGermany and an increase in average life span,Becker finds that the persistence of educationalinequality throughout generations has contributedto the continuing variance in life expectancyregarding social stratum and class. Hillmert(2002), in an analysis of GLHS, found that overfour birth cohorts from 1974 to 1992 the educa-tional level of an individual became slightly lessimportant as a factor in getting a first job. He alsofound that vocational qualifications continued tobe more advantageous than general education ingetting work in Germany. Based on the samedata set, Mayer and Brückner (1995) and Mayer(1996), in a comparison of 1930 and 1960 birthcohorts, found that education and training levelswere an increasingly important factor in getting afirst job. They found an even closer link betweenthe education and training level and job status inthe later years (5, 10, 15) of the working career.Konietzka (2002) analysed differences in the tran-sition from initial training to work of successivegraduate cohorts and concluded that patterns ofjob entry have changed in mainly qualitativeterms over time. Born (2000) used self-collecteddata (9) to examine how the choice of initial voca-tional training affected the life course of threesubsequent female cohorts. She found that thesechoices determined women’s working career andfamily life in all three cohorts. Different types ofvocational qualifications of equal level are notequivalent in terms of educational benefits.According to Born, women tend to have VETqualifications that typically yield lower educationand training benefits. Although she noted a slight

(9) Data is collected in the framework of research projects (1988-1996) of the German special research unit Sfb 186 (Sonder-forschungsbereich).

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 350



enhancement of former job concentration acrossthe cohorts, the most frequently occupied jobsstayed the same. Bynner and Parsons (2001)compared the relevance of qualifications andbasic skills of two birth cohorts (1958 and 1970)as protection against unemployment. Theyconcluded that qualifications and skills arebecoming increasingly important factors in deter-mining the levels of unemployment in the UK.Unemployment rates at different ages wereconsistently higher amongst the more lowly qual-ified in the younger cohort than in the oldercohort. Boockmann and Steiner (2000) showeddeclining educational returns from the 1925 to1974 cohorts in Germany, especially for women.

There are also a considerable number of studiesexploring how gender and family background affectthe benefits gained from education and training(Lauer, 2002; Antoninis and Tsakloglou, 2001;Ermisch and Francesconi, 2000; Ganzach, 2000;Dale and Egerton, 1997). These studies are oftencombined with a cohort approach, searching fordifferences in the influence of background orgender on the connection between education andtraining and its material and non-material benefitsand the development of this connection over timeor over generations. A study by Antoninis andTsakloglou (2001) in Greece indicates that ‘themost well-off segments of the population’ (p. 216)have the highest educational opportunities andbenefits. As longitudinal data was not available tothem Antoninis and Tsakloglou (2001) – based ontwo cross-sections from the Greek householdsurvey (HBS 1993-94) – can only speculate onwhat effects this might have for the whole lifecourse. However, they estimate that these couldbe even greater than their static results indicate(taking into account the different circumstancesexperienced by the offspring of well-to-do fami-lies and the less well-off throughout their workinglife).

Lauer (2002) examined the educational impact offamily background, gender and cohort in France forthe cohorts born between 1929 and 1968. Besidesa general improvement in educational standardsthroughout the cohorts, she found that the parents’own educational performance had a significanteffect on the school outcome and performance oftheir children. Also the impact of the father’s occu-

pation was significant: children of senior managershad the best educational prospects whereasworkers’ offspring experienced the worst educa-tional outcomes. The educational impact of familybackground, however, does not differ acrossgender. In France, women achieve better than menin secondary education. In post-secondary educa-tion, however, women have higher thresholds thanmen, which means that they are at a disadvantage.Payne (1997), using the YCS (UK) as her basis,investigated the consequences of the choice ofdifferent post 16 routes in the UK. She found thatyoung women who leave full-time education at 16are more likely than young men with similar charac-teristics to be without a full-time job or training. Onthe other hand, young women with above averageGCSE (10) results for this group have a better chanceof securing a job than similarly qualified males. Thestudy of Dale and Egerton (1997) is concerned withthe benefits of education and training for highlyeducated women, based on the NCDS (UK).Women in this cohort (born in 1958) are less likely tohave a degree level qualification and are more likelythan men to be under-achieving in occupationalterms. It is evident that both men and women showa considerable return on qualifications gained.However, women earn less than men at the samelevel of qualification.

With very few exceptions (e.g. Born, 2000;Dale and Egerton, 1997), quantitative studiestreat education and training in the same way asthey treat any other explanatory variable (e.g.work experience, family background, gender) inorder to explain differences in the economic,social and private wellbeing of individuals. There-fore, measurements of education and trainingtend mainly to be limited to the level of gradua-tion and the number of years in schooling ortraining. This creates a considerable gap in thefield of life-course research, because too fewcross-European comparisons can be made aboutdifferences in the type and quality of educationand training and of qualifications and skills.

The analysis of individual learning biographiescomplements, therefore, the studies of the bene-fits of education and training described so far.The biographical perspective, which is concernedwith the individual’s perception of education andtraining benefits, is in general less often used

(10) General Certificate of Secondary Education

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 351

than studies from a solely objective life-courseperspective. However, there have been some veryilluminating attempts to include the subjectiveperceptions of the benefits to be gained fromeducation and training (Sauer-Schiffer, 2000;Rabe-Kleberg, 1994). Possibly the mostoutstanding are those carried out in Ireland andthe UK (Unwin and Wellington, 2001; Bloomer andHodkinson, 2000; Stone et al., 2000; Keogh andDownes, 1998). In contrast to the informationcollected quantitatively on material and non-mate-rial benefits, it is not possible to quantify or tomake generalisations about how an individualmight perceive the advantages to be gainedthrough education. Therefore, results frombiographical research are difficult to summarisefrom any general perspective. Moreover, thiswould be contrary to the entire intention behindthe analysis of individual biographies, because itis precisely their individual differences that makesthem worthwhile. Keogh and Downes (1998), forexample, asked 114 participants of the Irish voca-tional training opportunities scheme (VTOS) towrite down their learning biography and experi-ences during and after having taken part in theVTOS. Besides giving a record of their educationalachievements, their stories document the transi-tion between social exclusion and inclusion and114 different ways of coping with this situation.While the document overall gives a very positiveimpression of the benefits gained from the VTOSprogramme in Ireland, quantitative research thatmeasures the impact of compensation orientedvocational training schemes does, however,present a rather critical picture. This reveals theoccasionally important difference between thesubjectively perceived and objectively measuredbenefits of education and training. For this reasonSchaeper et al. (2000) complement their quantita-tive analysis of working careers after VET (contin-uous and discontinuous) through biographicalinterpretations of the individuals concerned. Theyconclude that different subjective interpretationsof discontinuous careers correlate with the profes-sion and occupation a person has attainedthrough VET. Amongst those with a lower levelVET and profession, negative interpretations(discontinuity as threat) prevail, whilst positiveinterpretations (discontinuity as chance) figuremore often amongst higher level occupations.

In this section various national life-course and

biographical studies on education and trainingbenefits have been presented to give an impres-sion of the numerical and thematic scope ofpossible lines of investigation. It should havebecome clear that any comparison of results onthis level is nearly impossible and moreover inap-propriate, not only because of differences in thequantity and quality of data material but also –and more importantly – due to the variety ofdifferent research questions and theoreticalassumptions subsumed under the term‘life-course research’. In order to compare the lifecourse benefits gained from education, trainingand skills where different country specific condi-tions apply, a study would be needed whichdisplays a corporate set of research questions orhypothesis based on a particular theory or set oftheories. This would also need to be based on aprospective or retrospective survey conducted ata European level.

Such a study, however, does not exist atpresent. The few, limited studies that deal withcross-national and European comparisons arepresented in the next section.

4.3.2. Cross-national studies

The lack of sufficient European longitudinal datalimits cross-national life-course research into thebenefits of education and training to a smallnumber of studies, which are either based on:(a) European panel data;(b) corrected or harmonised national data;(c) data collected or reproduced within the

framework of single cross-national projects.A selection of relevant studies is shown in

Table 3.Studies which draw on data from the only

available European longitudinal data set, theECHP, cover a comparatively short period of timesince the survey was first conducted in 1994(Deding and Dall Schmidt, 2002; Pedersen andDall Schmidt, 2002; Brunello, 2001). The limitedperiod of coverage is also a reason for the factthat research has been limited only to measure-ments of short-term educational returns. Itfollows that the potential of the ECHP forcomparative research into the benefits of educa-tion and training has not been fully exploited. Aswith most national panels it contains only limitedinformation, focusing mainly on the time spent ineducation and training and on levels achieved.Employment prospects and the potential for


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 352

Th

e b

en

efits

of e

du

ca

tion

, train

ing

an

d s

kills

from

an

ind

ivid

ua

l life-c

ou

rse

pe

rsp

ec

tive

with

a p

artic

ula

r foc

us o

n life

-co

urs

e a

nd

bio

gra

ph

ica

l rese

arc

h353

Study Data Countries Data Sample Longitudinal Outcome Explanatory BenefitSource covered analyses size* design variables** variables*** (result)****

aggregate

Brunello (2001) ECHP A, DK, FIN, F, secondary/ ≈130 000 PPD, I, P, UK, NL, quantitative cross-national VET, R YE material (+)B, IRL, EL, E

Brunello and 11 national A, CH, DK, secondary/ n.n. PP***** Ex, R YE material (+)Comi (2000) data sets FIN, F, D, I, N, quantitative

(G-SOEP, OSA, P, UK, NLLFS, etc.)

Kieselbach et al. Qualitative I, F, E, FIN, secondary and qual.: 300 R non E, VET, material,(2000) interviews; UK, I, D, S self-collected/ FVET non-material

official quantitative and statistics qualitative

Hillmert (2001) BHPS, GLHS UK, D secondary/ ≈ 7 000 (BHPS) PP/RC OC E, VET materialquantitative Ø 800/C

(GLHS)

Schömann (1994) GLHS, PSSM D, PL secondary/ 465/n.n. RC/RC R E, VET material (+)quantitative

* Sample size SQ: survival quote of respondents (where available);P: period of measurement in years.

** Outcome variable: Ex: experience;OC: occupational career;R: rate of return;VET: vocational education or training.

*** Explanatory variable: E: general education; S: skills;FE: further education; TQ: type of qualification (subject);FVET: further vocational education/training; VET: vocational education or training;LE: level of education or qualification; YE: years of education or training.

**** if possible***** different panel designs

Table 3: Selected cross-national studies on education and training benefits from a life-course perspective

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 353

further education and training, mobility patternsand income growth are all possible areas ofbenefit that could be explored. Brunello (2001), forexample, used the ECHP to investigate the rela-tionship between educational attainment andtraining incidence (11) and finds that acrossEurope training incidence is higher among individ-uals with more experience of education. He alsodiscovered that this relationship varied signifi-cantly across countries and birth cohorts. Individ-uals had a higher training incidence in countrieswith a more educated labour force and a lessstratified schooling system. The most significantfinding from a life-course perspective was thestrong disparity observed between the effect ofprevious and of current training on the growth ofindividual earnings. While current training hasmedium to strong positive effects on educationalreturns, previous training has very low positiveand often negative effects. This result points tothe possibility that much of the private returns totraining is rather temporary. Furthermore – asmight be expected – there is evidence that indi-viduals with more education and limitedlabour-market experience enjoy higher privatereturns from recent training than those with thesame experience and less education.

Most of the studies which use harmonised oradjusted national data look at the impact of theoccupational and educational systems of two ormore central European Member States in order tocompare education and training (life-course)benefits (Hillmert, 2001; Sackmann, 2001;Brunello et al., 2001a; Kaiser and Siedler, 2000;Brunello and Comi, 2000; Müller and Shavit, 1998;Brauns et al., 1997). Brunello and Comi (2000) andBrunello et al. (2001a) investigated the relationshipbetween education and earnings growth in11 countries. They found evidence in all countriesthat education not only gave an initial advantagefor entry into the job market but that this advan-tage was permanent and increased over time.Analysing this relationship in more detail, theyshowed that earnings growth generated througheducation was lower in countries with a higherlevel of corporatism (e.g. Denmark, Germany andNorway) and higher in countries which experi-enced both relatively fast labour productivitygrowth and relatively low educational attainment

(e.g. Italy and Portugal). When comparing educa-tional systems, they found that countries with amore stratified system of secondary education (asin Germany and the Netherlands) had smallerdifferences in earnings growth through educationthan other countries. However, whether causalrelationships exist is not explained.

Hillmert (2001) provides a more specific analysisof the effects of different educational and trainingsystems. Using the BHPS and the GLHS, he investi-gated differences of education and training benefitsin the life course through comparison of twocompletely different VET systems: the British‘training-on-the-job’ version and the German dualsystem. He compared the life courses and workingcareers of different generations (both 1930-1970) bylooking at the transition from education to employ-ment position in the labour market at the time of jobentry and the occupational career in the years afterfirst employment. Contrary to general belief, hefound that the two systems had many similar educa-tion and training outcomes. With regard to the tran-sition into work the differentiation of successiveeducation and training levels had increased. Thismeant that it became more and more difficult forpeople with low levels of education to enter thelabour market successfully. Nevertheless, in bothcountries intermediate vocational – non-academic –qualifications were the most relevant for gainingdirect entry into stable employment. In addition, thestability of the first employment showed a constantdecline in both countries over subsequent cohorts.However, there were also expected differences: thecorrelation between the level of education andtraining and successful entry into the job market wasgenerally higher in Germany than in the UK and thisdid not change over the cohorts. This seems toconfirm the assumption that despite attempts toimprove the traditional ‘training-on-the-job-learning’in the UK through national VET programmes (NVQand modern apprenticeship) there is still a lack ofappropriate vocational qualifications for youngpeople. On the other hand, there seems to be greaterflexibility in the British system to get a job withouthaving the exact qualifications needed for it.

Results from other studies also confirm the rele-vance of the education and training and occupa-tional systems for the distribution of education andtraining benefits over the life cycle. Schömann


(11) Training incidence refers to the probability of taking part in training schemes and its frequency.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 354



(1994), for example, compared the development oflabour earnings over the life course in Germany andPoland and concluded that, despite very differentpolitical, occupational and educational systems,the impact of education and training on wageattainment throughout the life course was quitesimilar. Higher levels of education and training hada significant, positive impact both on initial wagelevels and after subsequent changes of employer.Nevertheless, a deeper analysis showed someimportant differences between the two systems. Incontrast to Germany, the level of education andtraining in Poland did not account for wage growthwithin the same job and after job changes within thesame company. Also, the differences in startingwages based on qualifications are only half as highin Poland as in Germany. However, the study suffersfrom some comparability problems in respect of thedesign and collection of data material. To sum up, itis clear that comparative studies from a life-courseperspective are still rare in Europe and that thosewhich do exist are limited in their explanatory powerand comparability due to the differences in thenational surveys they depend on – even if correctedor harmonised.

This gap was partly closed by increased efforts todevelop cross-national and international researchprojects. Within these projects additional data setswere produced or existing national data related toeducation and training and other social issues wereintegrated (Kieselbach, 2000; Sofer, 2000;Hannan, 2000; Blossfeld, 1999; McIntosh andSteedman, 1999; Furlong and Hammer, 2000).Although few European social research projectsadopt a life-course perspective in particular, somerecent projects do relate education and training todifferent aspects of current and future life success.The analysis of such projects does give more of aqualitative insight into the mechanisms of educationand training and its benefits than by simplymeasuring the impact of education and training onthe basis of qualifications achieved and the numberof years spent in education and on quantitativeoutcomes such as higher income or job stability. TheTSER project Yuseder (Youth unemployment andsocial exclusion: dimensions, subjective experienceand institutional responses in six European coun-tries), for example, looks at the risk of and experi-ence of unemployment and its relationship to theprocess of social exclusion. Besides the collectionof empirical evidence on youth unemployment and

its determinants in each country, a qualitative anal-ysis of 300 (50 per country) long-term unemployedpeople was conducted. As one important factor inthe reduction of unemployment risk they identify theeducation and training undertaken by the individual.Research shows that youth unemployment is likelyto have severe consequences for personal wellbeing and health and that suicidal behaviour is moreprevalent amongst the long-term unemployed.Therefore, investment in education and training indi-rectly safeguards young people from thesenon-material risks of social exclusion. Yusederconfirms that, depending on the labour-market situ-ation and trends as well as on the educationalsystem and its connection to the labour market, thepossibility of young people reducing the risk ofbecoming unemployed through their participation ingeneral or vocational education varies from countryto country. Other projects which examine life courseeducation and training opportunities and benefitsare the European Commission project FAME (Voca-tional identity, flexibility and mobility in the Europeanlabour market), the TSER project Newskills (New jobskill needs and the low skilled), theLeonardo da Vinci project ‘Dual Qualifications andVocational Mobility’, VTLMT (Education, VocationalTraining and Labour-Market Transitions), PURE(Public funding and private returns to education),SEdHA (Socioeconomic determinants of healthyageing) and the Globalife project funded by theGerman Volkswagen Foundation. As some of theseprojects have already been dealt with in the secondCedefop research report (Descy andTessaring, 2001), only three comparatively unknownprojects – in the context of research into the impactof education and training – are briefly discussedhere. These are PURE, Globalife and SEdHA.

PURE, a recent European Commission projectexamined the national educational returns of15 Member States not only with regard to wagesbut also to higher job opportunities. Its findingsshowed that the status and the trends of returnrates were totally different for each Member State.Though the research was not entirely from alife-course perspective, results from (in part) longi-tudinal measurement point to several life-courserelevant aspects (Section 5.1). The descriptionand main conclusions of the project can beviewed in Box 3.

The Globalife project analyses the influence ofglobalisation on different life-course dimensions

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 355

in OECD-type societies. Central to the focus ofthis investigation is the increasing tensionbetween the growing uncertainty of futurelife-course situations and peoples’ natural need tomake self-binding decisions and how they copewith this. Therefore, the analysis concentrates onthe transition from youth to adulthood, onchanges in career mobility, on forms of employ-ment and unemployment over the life course andon the transition from employment to retirement.Due to its broad research approach it drawsmainly on the secondary analysis of existinglongitudinal data sets rather than on collectingnew data. As there are numerous, different inves-tigations of education and training benefits in thelife course within the framework of Globalife, it isnot possible to give a summary here. Neverthe-less, important results are discussed in the next

chapter along with findings from the other studiesand projects presented.

A project presently running on a European levelwhich is concerned with the non-material benefits ofeducation and training in the life course is SEdHA. Theproject aims to provide an analysis of socioeconomicdifferences in health expectancy among the elderly indifferent European countries, and to contribute to theexplanation of these differences by looking at riskfactors and the accumulation of health problems overthe life course (Mackenbach et al., 1999). Since theexplanatory variable ‘socioeconomic status’ is oper-ationalised as educational level, it is mainly the impactof education on health that is measured. Besidesseveral national surveys and longitudinal studies, theECHP is used to improve the comparability of infor-mation. At present, no results have been published,but a first report is due soon.


Objective:

Study the impact of different systems of public funding and school differentiation on observableoutcomes in the labour market.

Outcome:

Main outcome variables are level and dispersion of private educational returns and education-relatedinequality in earnings.

Method:

(a) examine broad national data sets which contain individual-level information on wages and education;

(b) link the observed patterns and trends to national educational systems and policies. Besidesrepeated cross-sections, a few longitudinal studies are also utilised.

Countries:

Austria, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland and the UK.

www.etla.fi/PURE; www.cordis.lu

Box 3: PURE – Public funding and private returns to education

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 356

So far, studies which deal with the benefits ofeducation, training and skills or which at leastcontain some information on education andtraining benefits in the life course, have beenpresented at national, cross-national and Euro-pean level. Some results have already been indi-cated, but in a rather fragmentary way. Thischapter, therefore, arranges and discusses themost significant results according to the researchlines outlined in Section 4.1. However, the centralresult of this study already limits this intentionsince, up to this point, there is not a single study(or even survey) which would allow for a seriouscomparison of education and training benefits inthe life course of individuals or cohorts on a Euro-pean level. The establishment of the ECHP hasbeen a major step in this direction but as yet itcovers too limited a period of time and onlyprovides information on the number of years spentin education, training and skills acquisition, onlevels achieved and on formal degrees acquired.

5.1. Individual monetary returnson education and training

As already indicated at several points in thisreport, the studies reviewed are of very differentcharacter in terms of concepts, specific researchinterests, definitions of explanatory and dependentvariables, sample sizes, etc. Many of the studies,especially those concerned with educational returnrates, do not display a genuine life-course conceptas it was described in Chapter 3. Instead, theytake up other approaches first (e.g. human capital)and then add a longitudinal lifetime perspective(e.g. through the investigation of lifetime earnings)as an enhancement of this approach. For thepurpose of gathering information on impactresearch into education and training, it was never-theless unavoidable that all these ‘borderland’studies be included in our review, because other-wise there would have been almost nothing torevise. The life-course approach is itself – asexplained earlier – a perspective that enhances

other theories on the impact of education andtraining. Studies of this kind (even if they do notdisplay the life-course approach in detail) thereforecontribute a great deal to the understanding of theimpact and benefits of education and training inthe individual life course.

As pointed out in Section 4.1.1, educationalbenefits are often defined and measured in termsof individual monetary returns in the life course. Inthe tradition of the human capital approach,studies assume a direct relationship between theaccumulation of education, training and skills andthe rise of current and lifetime income (Steinerand Lauer, 2000; Brunello and Miniaci, 1999;Goux and Maurin, 1994). The general relevanceof skills and qualifications acquired througheducation and training for individual incomedevelopment is confirmed by several studies onincome development and lifetime income. Mostof the national and cross-national studies alsoshow a direct correlation between education andtraining, continuing education and training andthe gaining of higher rates of monetary returnsthroughout the life course (Becker and Schö-mann, 1999; Asplund and Pereira, 1996; Pannen-berg, 1995; Schömann, 1994).

As regards the level of private returns oneducation and training, there seem to be differ-ences among European countries. Scandinaviancountries, for example, display rather low averagerates of educational returns, whereas Ireland andthe UK are the opposite and central Europeancountries are located somewhere in-between.Some of them, such as Austria, Germany andSwitzerland, show a downward trend in rates ofreturn; others remain stable or even show slightupward trends, such as Italy or Portugal (Harmonet al., 2001; Asplund and Pereira, 1999). Theconclusion of Asplund and Pereira (1999) from aEuropean comparison of educational returns isthat there are few general trends and that theseare rather confusing to interpret given the diffi-culty of identifying any general tendencies. More-over, due to the lack of sufficient cross-nationalsurveys, the results from national studies need tobe interpreted very carefully. Data, measurement

5. Discussion of results

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 357

and estimation differ to the extent that compara-bility can be seriously questioned. The generalproblem of comparability is shown very clearly ifwe consider that estimations based on the samekind of data differ significantly between studiesdue to their different model specifications, theaddition of control variables, etc.

There is evidence that education and trainingregarding earnings complement each other. Theincidence of training is higher amongst individ-uals with more education (Brunello, 2001). Botheducation and training affect the earnings of indi-viduals in the short term and the long term(Pischke, 2000; Fougère et al., 2001; Arulam-palam et al., 1995). However, the direct effect oftraining on earnings growth seems to weakenafter a few years and can even show negativegrowth without an investment in further training.This suggests that training has primarily a tempo-rary impact on monetary returns and that qualifi-cations and skills become outdated within a fewyears (Brunello, 2001). However, the monetaryimpact of participation in continuing trainingprogrammes seems to be hard to determine.Depending on the definition of training, measure-ments and data material, national return ratesdiffer strongly between the studies and results aredifficult to interpret. There is some evidence thatin countries with a high standardisation of initialtraining systems, the monetary benefits of contin-uing VET seem to be lower than in countries withno such system (Hujer and Wellner, 2000).However, further research utilising a uniquedatabase is needed to throw light on this rathertentative conclusion.

Rates of educational returns do not seem to beconstant over the life cycle, especially for women.Though returns clearly rise in the early years of anindividual’s working career, the increase slackensin mid career and then stagnate and declinecompletely in the years leading to retirement.Assuming that productivity decreases with age,the results are compatible with the human capitalapproach which claims a direct link between indi-vidual productivity and rates of return (Steinerand Lauer, 2000).

There is a correlation between higher levels ofeducation and training and higher incomes, but alsoa higher variance in income. At this level, job expe-rience is a further factor influencing the rates of indi-vidual returns, because it acts as an additional

selectivity mechanism and therefore reduces vari-ance (Asplund and Pereira, 1999; Lauer andSteiner, 1999).

Studies that investigate differences of educa-tional returns with regard to types of education,qualification and skills are still rare and exist onlyat a national level. The impression is that returnson different types of qualification depend heavilyon the inherent structure of the educational andoccupational system of a country. The returnrates for VET qualifications, which particularlydepend on traditional structures and values, differfrom those for general educational qualifications.In countries with a traditionally high degree ofinstitutional standardisation of their trainingsystem and a clear differentiation of general andvocational education, individual return rates toVET seem to be much higher than in countrieswhich show a rather low degree of standardisa-tion and differentiation in education and training.For the UK, as an example of the latter, there isevidence that individuals who participate in YouthTraining Schemes (YTS) after compulsory schoolage receive significantly lower wages than thosewho leave school at the age of 16 with no furthertraining. These differences cannot be explainedby a simple delay in returns as a result of theextra years training and the opportunity costsincurred. Rather, they seem to be stable over time(Green et al., 1994). Therefore, from an individual(life-course) perspective it can be more beneficialnot to invest in further formal qualifications even ifthe opposite is true from a macrosocial perspec-tive. This leads us back to the weaknesses ofhuman capital theory as described inSection 4.1.1, because this approach onlyexplains differences in structural circumstances inrational terms and completely ignores traditionaland cultural explanations for these differences.Furthermore, this approach sees the benefits ofeducation and training simply in terms of mone-tary returns. It disregards other beneficial aspectsthat are not as easily quantifiable, such as thedifferent patterns of individual career prospectsor non-material benefits, e.g. the gaining ofhigher social prestige.

In summary, it has to be pointed out that toexplain life course benefits just in terms of humancapital rates of monetary return is ultimately verylimited. Despite several refinements of estimationtechniques, too many dynamic and unpredictable


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 358



factors cannot be anticipated or estimated. Humancapital rates of return ignore structural differencesresulting from traditional factors that produceinequalities in the access to and outcomes ofeducation and training, both within and betweenEuropean countries. Last but not least, it can bequestioned whether these studies have a valueapart from pure scientific interest. As they assumea balanced situation, which cannot be found inreality, any recommendations for policy and prac-tice must, by their very nature, be limited.

5.2. Education, training andlabour market participation

Another group of studies deals with the differentdimensions of an individual’s occupational career.These, generally speaking, look at his or herparticipation in the labour market and are eitherlargely descriptive or based on one or morelabour-market theories (Section 4.1.2.). The diver-sity of topics covered and of theoreticalapproaches and methods makes it very hard tocompare results from these studies, especiallyfrom a European perspective. Nevertheless,some key generalities are summarised here.

There is strong evidence that higher qualifica-tions and skills significantly reduce the danger ofspending a considerable part of one’s working lifein unemployment. Evidence also shows that thevalue of education and training and continuingeducation and training affects not only one’scurrent career but also that it has accumulativeeffect over the whole working life (Bukodi andRobert, 2002; Noguera et al., 2002; Becker andSchömann, 1999). As basic education andtraining determine access to higher and contin-uing education and training to a considerableextent, individuals with higher educational andvocational degrees have much better workingcareer opportunities.

However, research also indicates nationaldifferences regarding the link between educationand training and career prospects over the lifecourse. In countries with a high standardisation oftraining systems and an emphasis on well-definedvocational qualifications, adequate employmentand successful working careers are stronglydependent on specific vocational qualifications.This link is not as strong in countries with less

well-integrated education and employmentsystems and less vocational specialisation (Lief-broer, 2002; Klijzing, 2000; Blossfeld andMayer, 1998). In systems fostering high standardsof vocational qualification, the differentiationbetween the unqualified and vocationally qualifiedconstantly increases over the working life, withthere being hardly any opportunity to compensatefor this (Antoninis and Tsakloglou, 2001). On theother hand, the material benefits of education andtraining in terms of employability, the avoidance ofunemployment and career development aregenerally higher in countries with specific voca-tional education (Shavit and Müller, 2000).

Upward mobility and individual career develop-ment seem to be increasingly bound to the partic-ipation in continuing VET and lifelong learningopportunities (Becker and Schömann, 1999), asdo increased job stability and labour market flexi-bility. In both cases this leads to higher incomeand improved career prospects (Fitzenberger andPrey, 1999). For Germany in particular, completedand certificated participation in further VETsupports upward mobility and income growth butthis might not be equally true for other countrieswith less formalised and structured education andtraining systems (Becker R., 1993). As access tofurther VET is strongly determined by educationallevels, the selective mechanisms which operate toqualify for further training tend to lead to a cumu-lative increase of educational inequality over thelife course (Schömann and Becker, 1995;Becker R., 1993).

Earlier studies (Meulemann, 1990b) sometimesfound that additional qualifications had a negativeoverall effect on subsequent job status becauseemployers possibly considered the acquisition ofadditional qualifications in other fields of study tobe inconsistent or because they were uncertainabout the value of additional skills and flexibility.However, more recent research has rejected thisfinding. This might indicate a constantly growingneed for individuals to adapt to new skill require-ments of the labour market and to the necessityfor lifelong learning. Double qualifications, in thesense of acquiring two entirely different voca-tional or university degrees, do not result in easieraccess to jobs or the achievement of a higher jobstatus at a later date.

The importance of initial and continuing educa-tion and training for participation in the labour

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 359

market is also clearly shown by looking at theother side of the coin: the continuing and cumu-lative disadvantages of unskilled people or thosewith no formal education and training in the lifecourse (Solga, 2002; Stone et al., 2000;Bynner, 1995; Dietz and Matt, 1994). Neverthe-less, research shows that a direct linear relation-ship between disadvantages experienced ineducation, training and labour-market participa-tion on the one hand and problems of individualdevelopment (e.g. delinquent behaviour) on theother hand, contrary to popular belief, does notexist (Schumann and Mariak, 1995;Böttger, 1998; Matt et al., 1998).

5.3. Education, training andtransitions

As pointed out in Section 4.1.3 transition periods(between education and training,education/training and employment, etc.) are ofspecial interest for life-course research, becausethe life course – in contrast to an individual biog-raphy – is in principle merely made up of succes-sive transitions. In particular, the degree ofsuccess achieved in the transition to the first jobis considered to be a formative element for thewhole later career and life path.

In general, research shows the beneficial impactof education and training on the success of the tran-sition from education and training to work. The verylow skilled in particular have major disadvantagescompared to other young people who at least havesome kind of formal education or training(Bynner, 1994; Heinz, 1999; Starrin et al., 2000).

However, clear national differences are alsoapparent (Starrin et al., 2000). Transitions seem tobe comparatively easy in Denmark, Germany, theNetherlands and Austria but in southern Euro-pean countries like Greece, Italy or Spain, directentry in a stable position after completing educa-tion or training seems to be extremely difficult. InBelgium and Sweden, where the systems ofeducation and training are decoupled and wherethere is an emphasis on the general educationsystem, one can see a trend towards the polari-sation of extremely highly and very poorly quali-fied people. In addition, young people experiencethe trap of companies demanding a high level ofqualification but at the same time reducing their

offers of on-the-job training. Due to the lack ofwork experience and job related skills, it hasbecome more and more difficult to gain a first job(Rantakeisu et al., 2000). In contrast, Germanyhas always been known as a country that easesthe transition from education to work by the dualsystem, which combines theory (in schools) andpractice (workplace) in a differentiated VETsystem. However, due to a worseninglabour-market situation and job cutbacks the dualsystem has experienced a serious crisis. Compa-nies have reduced their offers of training or donot take on apprentices (Kieselbach, 2000).

‘The Swedish education system is almostdiametrically opposed to its German counterpart: inSweden, education is not so tailored towards therequirements of the labour market, whereas inGermany the system is too adjusted to short-termdemands by the industry and thus compels youngpeople to decide on a particular vocational trainingtoo early.’ (Kieselbach, 2000, p. 37).

In southern European countries such asGreece and Spain the lack of coordinationbetween the educational system and the labourmarket accounts for many of the problems ofyouth unemployment. In both countries the youthunemployment rate is three times higher than theoverall rate of unemployment. Despite theirlimited job prospects, university degrees inGreece are overvalued, because they are tradi-tionally connected with high social esteem for thefamily. In contrast, vocational training is regardedas being of minor value, which leads to a mass ofunskilled university graduates in a labour marketdemanding specific vocational qualifications.Spain currently tries to challenge its high youthunemployment by a system reform that aims toimprove the adjustment of demand and supply.The decision either to go to university or tochoose a vocational strand is still mainly deter-mined by financial opportunities and social class.However, people with a university degree aremore likely to become unemployed than thosewith intermediate level qualifications because ofan early and continuing decrease in jobs requiringhigh qualifications. A high level of qualificationalone, therefore, does not improve the individual’sopportunities on the labour market (Lemkowet al., 2000; Sokou et al., 2000).

From a life-course perspective, these resultsare of relevance in several different ways. As


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 360



already pointed out, it is assumed that individuallife trajectories are entirely determined by theeducational and occupational system. As hasbeen shown, the education and training systemand its relationship to the labour market has amajor impact on the success or failure of youngpeople’s transition to their first (stable) job. But asfar as is known from studies of working careers,the impact of education and training, and conse-quently of its system, goes far beyond this.Through the link between occupational factors,unemployment and employment and socialin-/exclusion, the educational system contributesto the quality and quantity of educational benefitsthroughout the entire life course (Mülleret al., 2002). This is also supported by the find-ings of the Yuseder project (Starrin et al., 2000),which show that countries with a poorly coordi-nated education and training system not onlydisplay low transition rates but also often providefew possibilities for further education or training.In addition, people with lower qualifications andlow-income jobs typically have fewer chances ofgetting access to further education. This is likelyto result in an accumulation of disadvantages andbenefits in equal measure throughout the lifecourse not only in, but also between, Europeancountries.

5.4. Generational and cohortdifferences in education andtraining benefits

From a life-course perspective differencesbetween successive generations and birthcohorts in the benefits they gain from educationand training are of special interest, becausesocial change over historical time becomestangible when using a life-course approach(Section 4.1.4).

Throughout Europe there are clear generationand cohort effects in the attainment of educationand training benefits. Subsequent birth cohortsdisplay different education and qualification struc-tures and are each faced with different opportuni-ties in entering the job market due to economic,social and political developments (Brunelloet al., 2001b; Bynner and Parsons, 2001; Benderand Dietrich, 2001; Bynner, 1998; Hægelandet al., 1999).

Generally older cohorts possess lower averagelevels of qualification and also had less difficultyin entering the labour market (Bynner, 1998). Dueto the international economic turndown from theearly 1980s onwards, the labour-market situationhas changed and increasing job competition hasbecome prevalent. The pressure of generaleducational expansion has made it more andmore difficult for young people to position them-selves successfully on the labour market. Aheightened significance of qualifications andskills as protection against unemployment isfound in most of the studies reviewed. Theyounger the cohorts the more important higherlevels of education and training become (Bynnerand Parsons, 2001; Mayer, 1996).

The results of research into the monetaryreturns on education and training are not uniform.Whilst in Germany a significant decline in returnrates can be observed across generations andcohorts, especially for women (Steiner andLauer, 2000; Boockmann and Steiner, 2000),research from Norway indicates a trend towardshigher return rates of education and training(Hægeland et al., 1999). An important question ishow these differences can be explained. Theusual explanation that the increased participationof women in the labour force and in their educa-tional attainment is responsible for these differ-ences seems questionable given that these wereEuropean-wide developments, which obviouslyhad not led to a decline in educational returnsacross successive cohorts in all countries.

5.5. Social differences ineducation and trainingbenefits

Besides generational aspects, there are twofurther central factors which help explain differ-ences in education and training benefits in the lifecourse: family background and gender. Childrenfrom low-income households or with less well-educated parents tend to be disadvantaged intheir access to education, training and furthereducation and training (Lauer, 2002). In mostcountries their disadvantages accumulate ratherthan decline over the life course (Bynner, 1995).

Despite numerous training and reintegrationprogrammes for the low skilled and unemployed,

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 361

current educational systems still seem to fostereducational inequality. However, there are alsodifferences between countries as regards the depthand extent of this discrimination. Educationalinequality related to family background seems to bestronger in those countries where traditional classstructures survive, and where hierarchical educa-tion and training systems prevail and counterbal-ancing programmes are rare (Schnabel, R. andSchnabel, I.; 2002; Antoninis and Tsakloglou, 2001;Tsakloglou, 1993, 1997). On the other hand, finan-cial transfers in public education, the promotion ofreemployment programmes and other forms ofintervention have high distributional effects foreducational equality in these countries (Magoulaand Psacharopoulos, 1997).

In a review of studies concerned with theconditions and consequences of being ‘not ineducation, employment or training’ (NEET) at age16-18, Coles et al. (2002) find that NEET isstrongly related to poor family background. Notbeing in education, employment or training,together with other related factors, is very likelyto lead to material as well as non-material prob-lems throughout the later life course: experienceof unemployment, involvement in drug or alcoholmisuse, poor health, parenting at an early ageand involvement in crime. There is also evidenceof lower earnings throughout the life course evenwhen the person is in work. Many of those unem-ployed at the age of 18 have few or no qualifica-tions and this significantly impacts on any laterearnings if employment is obtained. Persistentoffending amongst 18-30 year olds correlatesvery strongly with having been excluded fromschool, having no or low qualifications andregular drug and alcohol misuse. Evidence onpersistent offenders confirms that an accumula-tion of risk factors leads to social exclusion.

In comparison to other socially disadvantagedgroups, women have been able to catch up withmen in education and training both in respect oflevels as well as of types of qualification in thelast 20 years. However, they have not been ableto convert their education and training achieve-ments into higher and better-paid jobs assuccessfully as men. Highly qualified women arestill more likely then men to be under-achieving interms of their occupation. Also women earn lessthan men at the same level of qualification (Daleand Egerton, 1997).

Research shows that there is not only amale-female segregation according to the alloca-tion of occupational levels (vertical) but also fordifferent types of vocations located on the sameformal level (horizontal). Men typically hold occu-pations with higher educational benefits thanwomen (income and job status). Moreover, asthese vocations are typically attributed to malecharacteristics women have fewer opportunities toaccess these jobs. Occupational fields related tomale characteristics (accountancy, law, computing,etc.) lead to higher material educational benefitsthan fields associated with female characteristics(humanities, education, social sciences, etc.)(Born, 2000; Dale and Egerton, 1997)

Women also do not often have the samechances of participating in continuing educationand training programmes due to family commit-ments or because they are employed in parts ofthe (segmented) labour market where there arefewer possibilities to access continuing educationand training and higher career options.

5.6. Non-material benefits ofeducation and training andsubjective biographicalperception

Findings regarding non-material benefits ofeducation and individual learning biographiescomplete the rather ‘objective’ social researchpresented so far. The biographical researchsummarised in Section 5.6.2 is often concernedwith the non-material benefits of education asdiscussed in Section 5.6.1.

5.6.1. Non-material benefits of education and

training

As explained earlier, non-material benefits arecomparatively hard to define and even harder tomeasure. Therefore, life-course research mostlyconcentrates on the ‘hard facts’, the materialbenefits of education. Nevertheless, there arealso some beneficial non-material aspects thatare more or less agreed upon, such as health andthe quality of life. However, with the exception ofa few health studies, empirical measurementshave to rely mostly on subjective judgementsabout personal wellbeing, quality of life and otherrather ‘soft’ issues.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 362



A number of life-course research studiesdisplay a positive correlation (not necessarilycausality) between education, training and skillson the one hand and, for example, health, qualityof life, family formation, reduction of criminalbehaviour and avoidance of social exclusion onthe other (Blackwell and Bynner, 2002; Schulleret al., 2002; Hammond, 2002; Preston andHammond, 2002; Bynner et al., 2001; Lasheraset al., 2001; Hullen, 1998, 2000; Becker, 1998;Dale and Egerton, 1997; Frenzel, 1995; Blossfeldand Huinink, 1990). Education is in general one ofthe most important determinants of healthinequalities. Becker (1998), for example, confirmsthe enduring impact of education on lifeexpectancy on the basis of GSOEP and GLHSdata. As education structures the life course, it isresponsible for the distribution of socioeconomicopportunities. On the one hand the trend tohigher qualifications has led to an increase in theaverage life span of European populations, on theother hand the persistence of educationalinequalities in society contribute to the persis-tence of variance in life expectancy between thesocial classes. The educational level of the indi-vidual and of the family determines to a largeextent their standard of living and quality of life.

There is also an indirect connection betweeneducation and behaviour for health and mortalityrisks (Lasheras et al., 2001; Marmot andWilkinson, 1999; Nehru and Dhareshwar, 1993).As the increase of (lifetime) earnings, theenhancement of social status, etc., are directeffects of education and training, these may alsolead to the reduction of indirect effects such ashealth risks and criminal behaviour. Education isalso seen as an important condition for the forma-tion of cultural capital like personal development,social participation and dealing with institutionaldemands (Becker, 1998). This in turn is central forthe anticipation and processing of critical lifeevents and coping with stressors which can affectthe quality of life, self-consciousness and health.

There is evidence that men and women withhigher qualifications are the most distinctive interms of their marital arrangements. In compar-ison to lower vocationally qualified groups theyare the least likely to have formed partnershipsand to have had children by the age of 33. Alsothere is a strong relationship between the level ofeducational qualification and age at birth of the

first child for both men and women, with the mosthighly qualified delaying parenthood the longest.There is some indication that the most highlyqualified women may be more likely to choose notto have a child (Huinink, 2000; Dale andEgerton, 1997). However, for Germany there issome evidence that the increasing educationalattainment of women has led them to delaygetting married but has not caused a generaldecrease in marriage overall. Contrary to generalbelief, the likelihood of women getting marriedseems not to have been affected by theincreasing participation of women in the educa-tional system (Blossfeld and Huinink, 1990).Life-course research needs to be consideredalongside an individual’s own subjectivebiographical interpretation before these resultscan be interpreted as advantages or disadvan-tages of education and training. Nevertheless, thewell-discussed processes of macro- demographicchange (birth decline, ageing societies) (will) alsohave a revertive impact on the individual level.

5.6.2. Subjective biographical perception of

educational benefits

Biographical research studies complement theobjective, rather quantitative life-course perspec-tive through subjective, qualitative insights ofindividual biographical interpretation(Section 4.1.6). The individual perception ofeducation and training benefits is entirely relatedto the persons’ educational and occupationalbiography and is therefore more dependent onindividual interpretations and self-constructionthan on objective circumstances. However,biographical range is influenced by the structurallimitations of the life course. The acquisition ofnon-material benefits is often placed withinbiographical contexts, because it is more difficultto find rational objective explanations for theirdefinition as ‘benefits’. Non-material benefits arenot primarily objective constructs but are merelythe result of subjective judgements.

Compared to life-course research, there arefew studies concerned with the benefits ofeducation and training from a biographicalperspective. In addition, since the analysis ofindividual biographies aims to identify the special,unique characteristics of each single person, it is– strictly speaking – not possible to give anysummary of results. The aim of this kind ofresearch is rather to come to a reasonable under-

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 363

standing or interpretation of life courses and theirdifferences. In comparison to the mere quantita-tive, objective life-course perspective, the centralquestion is ‘why’ not ‘if’. Therefore, individualinterpretations are hard to generalise. However,the most outstanding outcome of biographical

research is presumably the discrepancy betweenobjective events and their subjective interpreta-tion. The same kind of experience, for example ainterruption in the working career, is perceived inquite different ways depending on the individualbiographical background (Schaeper et al., 2000).


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 364

The purpose of this report was to provide anoverview of life-course research as regardseducation and training benefits in Europe. Afteroutlining the main problems involved in definingand measuring the benefits of education andtraining, the theoretical background of thelife-course perspective was comprehensivelyexplained. The synthesis of the two fieldscovered, namely research into the benefits ofeducation and training and the life-courseperspective, was done by explaining the specialinterest life-course research has in the analysis ofeducation and training benefits for the individualand by showing how the life-course perspectiveenriches research into the impact of conventionaleducation and training. After having set out themain lines of investigation, selected national andcross-national life-course studies, which provideempirical evidence on individual education andtraining benefits, were presented and theirdifferent approaches, use of data and conclu-sions were reviewed. Finally, the main results ofthe studies were summarised and an attempt wasmade to compare them from a European or atleast a cross-national perspective. However, thedifferent studies offer different definitions ofeducation and training, a wide range of educationand training benefits, heterogeneity in data quan-tity and quality and above all an enormous diver-sity of research questions. A serious comparisonof results, which would allow for conclusions andrecommendations to be made about the impactof country specific conditions, is therefore simplynot possible. Our conclusion will offer a cautiousattempt to address some general trends, but itsfocus will be on recommendations for furtherresearch rather than on implications for policyand practice.

As regards the individual’s monetary returns oneducation and training throughout the life course,the studies reviewed indicate differences betweencountries in average return rates as well as inupward and downward trends. The consequencesof these differences are hard to estimate. Somestudies claim that there is higher mobility acrossnational borders, particularly of highly educated

people. They suggest that countries with a contin-uing downward trend in educational return ratesought to increase their investment in educationand training. Hence, these studies indicatesubstantial individual and social benefits to begained from policies aimed at lowering thenumber of early school leavers and from providingsocially deprived groups with the opportunitiesand incentives to continue education and training(Asplund, 2003). In comparison withcross-sections, longitudinal investigations high-light different context dependencies of humancapital configuration. Human capital allocation isnot a static characteristic of a society but is aprocess of acquisition and allocation in the indi-vidual life course and through historical change.However, as estimations of individual rates ofreturns are based on simple human capitalassumptions of ‘input’ and ‘output’, results mustbe treated carefully. In addition, return estimationsstem from national data sets in which the quantityand quality of data material is very different. Acomparison on a European level would actuallyneed a unique database. Results from studiesdealing with a broader concept of education andtraining benefits, including different aspects of anindividual’s changes of status over the life course,are even harder to review for their general trendsand recommendations, because they display verydifferent approaches, focal points, empiricalmethods and measurements.

Moreover, since the majority of these studiesdo not aim to evaluate the education and trainingsystem of one or more countries but rathergenerally try to reveal how the individual lifecourse is determined by education and training,or education and training systems, policy andpractice implications are hard to derive. Never-theless, these studies should not be underesti-mated within the framework of research into theimpact of education and training, because theydemonstrate the relevance that education andtraining has for the whole individual life course byshowing that participation in education andtraining have an increasing and cumulative effecton occupational career and personal develop-

6. Conclusions and recommendations

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 365

ment (Bukodi and Robert, 2002; Nogueraet al., 2002; Becker and Schömann, 1999). Thesestudies confirm what has been identified already inearly life-course research (e.g. Featherman andShavit, 1990): educational policy and practice havea major impact on the individual life course. Theconnection between the education and trainingsystem and the labour market, for instance, is amain factor in determining the entry into the jobmarket and, as a consequence, later career devel-opment. Empirical life-course research also showshow the still existing selectivity of continuingeducation and training, as regards educationallevel, gender and family background and othersocial factors, and how this contributes to thepersistence and accumulation of social discrimina-tion throughout the life course. This means thatsimply increasing the opportunities for continuingeducation would not necessarily improve an indi-vidual’s opportunities for participating in the labourmarket. On the contrary, this might lead to quiteundesirable side effects such as the further polari-sation of educational and occupational opportuni-ties. Our recommendations would be to increasetarget group orientation and to provide permanentevaluation and quality controls. Furthermore,subsidised continuing education and training mustbe more than a mere social sop or a simple cure-all for problem groups, but must, as a matter ofcourse, be integrated into a system of economicand labour-market policy (Allmendinger, 1994).Life-course research also shows the ‘wider’ ornon-material benefits of education and training,especially for health, life expectancy, personaldevelopment and family formation, all of whichalso accumulate throughout the life course. Asregards this – already repeatedly mentioned –accumulation effect of education and training andfurther education and training throughout the lifecourse, the central policy recommendation fromlife course and biographical research would be toimprove the participation in education and trainingof socially deprived groups. This would ease tran-sitions between education and training, initialeducation and training and higher education,education and training and work. It would alsofacilitate vertical and horizontal mobility andactively foster the participation of the lower quali-fied in continuing education programmes. Thelife-course perspective also points to the necessityof devising specific, national solutions to policy

development in education and training, becausedifferent education and training systems do notsimply exist as a matter of fact, but have devel-oped over time due to particular historical andstructural realities. However, life-course researchas a rule does not provide evidence on the impactof particular education and training programs butrestricts its education and training variables tolevels achieved and to the number of years spentin education (or, at most, to types of qualifications).Therefore, suggestions as to how improvementscould be achieved cannot be offered but thisneeds to be demanded of evaluative research.

However, the suggestions made above shouldbe treated carefully. From studies, which draw onlong-term prospective panels or retrospectiveanalyses, we have learned that it becomesincreasingly difficult over time to ascribe aspecific effect to initial education and training orto a particular education and training programme.A variety of other factors, such as life experience,personal attitudes, family and social background,can all interfere with education and representeither conditions or impacts. Furthermore, moststudies are based on national surveys that aredifficult to compare because their definitions andthe quality and quantity of their empirical workvary so much. In Europe longitudinal research hasbeen concentrated in Germany, the Netherlands,the UK and a few Nordic Member States, whilstthere is only a weak database for life-courseinvestigations in southern and eastern Europeancountries. The rapid economic advancement ofthese countries over the last few years would,however, make these the most interesting ones toassess for the impact of education and trainingover time. Increased efforts are therefore neededto produce comparable longitudinal data sets inEurope and to include southern and easternEuropean countries in particular. The ECHP willpresumably be the most appropriate and mostused data source for research into education andtraining benefits from a life-course perspectiveover the coming years. However, as the panelcontains only limited information on education,training and skills a European cohort study, whichfocuses on aspects related to education andtraining, would be an essential addition long-termEuropean research into the impact of educationand training.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 366

BCS70 1970 British cohort study

BHPS British household panel survey

ECHP European Community household panel

EPAG European panel analysis group

GLHS German life history study

GSOEP German socio-economic Panel

IAB Federal Institute for Employment Research

LNU Swedish level of living survey

MLS Malmo longitudinal study of social assistance

MS Malmo study

NCDS National child development study

NORS Norway survey

PC Prospective cohort

PP Prospective panel

PURE Public funding and private returns to education

R Retrospective

RC Retrospective cohort

SEdHA Socioeconomic determinants of healthy ageing

SEP Dutch socio-economic panel

SOEP Socio-economic panel


VTLMT Education, vocational training and labour-market transitions


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 367

National and cross-national electronic catalogues

Acronym Resource Available from Internet[cited 15.07.2003]

BIBSYS Online Library System of Norwegian University Libraries www.bibsys.no/english

BVB Bibliotheksverbund Bayern www-opac.bib-bvb.de

COPAC Union Catalogue of UK and Ireland www.copac.ac.uk/copac

FoDokAB Forschungsdokumentation zur Arbeitsmarkt- ——-und Berufsforschung

FORIS Forschungsinformationssystem Sozialwissenschaften Access limited

GBV Gemeinsamer Bibliotheksverbund www.gbv.de

IBSS International Bibliography of the Social Sciences ——-

IBZ Internationale Bibliographie der Zeitschriftenliteratur www.asu.edu/lib/resources/ db/ibz.htm

KOBV Kooperativer Bibliotheksverbund Berlin- Brandenburg www.kobv.de

KVK Karlsruhe Virtual Catalog www.ubka.uni-karlsruhe.de/kvk

LIBRIS The union catalogue of Swedish libraries http://www.libris.kb.se

LitDokAB Literaturdokumentation zur Arbeitsmarkt- ——-und Berufsforschung

OLC Online Contents www.gbv.de

SOLIS Sozialwissenschaftliches Literaturinformations-system Access limited

WISO Literaturnachweise für Wirtschafts- und http://www.wiso-net.deSozialwissenschaften

Cross-national and national databases


ACROSS Educational issues across Europe www.b.shuttle.de/wifo/ across/=start.htm

CESSDA European social science data archives www.nsd.uib.no/Cessda

CORDIS Community research development information service www.cordis.lu

EHRD Human resource development in Europe www.b.shuttle.de/wifo/ ehrd/=portal.htm

IDEAS http://ideas.repec.org

LLL Base Lifelong learning base www.b.shuttle.de/wifo/ lll/=base.htm

OECD Ed OECD education database www.oecd.org/scripts/ cde/fbase

SSRN Social Science Research Network http://www.ssrn.com/

VET Bib Cedefop VET bibliographical database http://libserver.cedefop.eu.int: 4505/ALEPH

Annex — List of data and information sources

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 368



Online-bibliographies of professional organisations/institutes


CSL Centre for Longitudinal Studies, UK http://www.cls.ioe.ac.uk

ISER Institute for Social and Economic Research, UK http://www.iser.essex.ac.uk

ECASS European centre for the analysis in social science http://www.iser.essex.ac.uk/ ecass/

EPAG European panel analysis group http://www.iser.essex.ac.uk/ epag

Eurydice Information network of education in Europe http://www.eurydice.org

FORUM Forum for research in education and training http://www.theknownet.com

MZES Mannheim centre for social research in Europe http://www.mzes. uni-mannheim.de

MPIB Max-Planck-Institute for human development www.mpib-berlin.mpg.de

DIPF German institute for international educational research www.dipf.de

FiBS Institute of education and socio-economic research www.fibs-koeln.deand consulting

ZEW Centre for European economical research www.zew.de

CRWBL Centre for research on the wider benefits of learning www.learningbenefits.net

ECSR European consortium for sociological research ———

CREST Center for research in economics and statistics www.crest.fr/presentation/ Apresentation.html

CLS Centre for labour market and social research, Denmark http://www.cls.dk

CIM Centre for research in social integration and marginalisation, http://www.hha.dk/eok/CIM/Denmark

IAB Federal institute for employment research www.iab.de

IZA Institute for the study of labour www.iza.org

ROA Research centre for education and the labour market www.fdewb.unimaas.nl/roa

WZB Social science research center berlin www.wz-berlin.de/

Cedefop European centre for the development of vocational training www.cedefop.eu.int

DFES Department for education and skills www.dfes.gov.uk

National and cross-national electronic data archives and statistical offices


ADPSS archivio dati e programmi per le scienze sociali, www.ifdo.org/org_archives/ Milano, Italia arc_bfr.htm

BASS Belgium archive for social science http://logi.rspo.ucl.ac.be

BDSP Banque de données socio politiques http://195.221.54.130/bdsp/

CIS Centro de investigaciones sociológicas www.cis.es

DDA Danish data archive www.dda.dk

ESRC UK data archive www.data-archive.ac.uk/

ESSDA Estonian social science data archive http://sys130.psych.ut.ee/esta

Eurostat Statistical office of the European Communities http://europa.eu.int/comm/ eurostat

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 369

FSD Finish social science data archive www.fsd.uta.fi/english/ index.html

IFDO International federation of data archives www.ifdo.org

NSD Norwegian social science data services http://www.nsd.uib.no

SDB Social data bank, Greece http://www.ekke.gr/databank/

SIDO Swiss information and data archive service http://www.sidos.chfor the social sciences

SSD Swedish social science data service http://www.ssd.gu.se

ZA Zentralarchiv für Europäische Sozialforschung http://www.gesis.org/ZA

Programme- and project bibliographies


CATEWE Comparative analysis of transitions from education www.mzes.uni-mannheim.de/to work in Europe projekte

FAME Vocational identity, flexibility and mobility www.itb.uni-bremen.de/in the European labour market projekte/fame

Globalife Life course in the globalisation process http://alia.soziologie.uni-bielefeld.de/~globalife

ISSP International social survey programme http://www.issp.org/public.htm

Newskills New job skill needs and the low skilled http://158.143.98.51/ homepage/tser

SEdHA Socioeconomic determinants of healthy ageing http://www.eur.nl/fgg/mgz/

STT Schooling, training and transitions: an economic www.b.shuttle.de/wifo/ perspective across/p-st.htm

TSER Targeted socio-economic research programme http://www.cordis.lu/tser/home

Youth Youth unemployment and processes of marginalisation Marginalisation on the Northern European periphery www.b.shuttle.de/wifo/

across/p-yo.htm

Yuseder Youth unemployment and social exclusion www.ipg.uni-bremen.de/ YUSEDER


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 370

Adam, K. Die deutsche Bildungsmisere: PISA unddie Folgen. München: Propyläen, 2002.

Alewell, D. Interne Arbeitsmärkte. Hamburg:Steuer- und Wirtschaftsverlag, 1993.

Allmendinger, J. Lebensverlauf und Sozialpolitik:die Ungleichheit von Mann und Frau und ihröffentlicher Ertrag. Frankfurt/Main: Campus,1994.

Andreß, H. J. Instabile Erwerbskarrieren undMehrfacharbeitslosigkeit. Ein Vergleich mit derProblemgruppe der Langzeitarbeitslosen.Mitteilungen aus der Arbeitsmarkt- und Berufs-forschung, 1989, Vol. 22, No 1, p. 17-32.

Antoninis, M.; Tsakloglou, P. Who benefits frompublic education in Greece? Evidence andpolicy implications. Education Economics,2001, Vol. 9, No 2, p. 197-222.

Arrow, K. J. Social choice and individual values.Second edition. New Haven: Yale UniversityPress, 1973.

Arulampalam, W.; Booth, A. L.; Elias, P.Work-related training and earnings growth foryoung men in Britain. Colchester: ISER, 1995(Working papers, No 18).

Arum, R.; Shavit, Y. Secondary vocational educationand the transition from school to work. Soci-ology of Education, 1995, Vol. 68, p. 187-204.

Ashenfelter, O.; Layard, R. (eds) Handbook of laboreconomics (3 Vol.) Amsterdam: North Holland,1986.

Asplund, R. Public funding and returns on educa-tion. Ely, UK: PJB Associates, 2003 (Newperspectives on learning.– Briefing paper 29).

Asplund, R.; Pereira, P. T. Returns to human capitalin Europe: a literature review. Helsinki:Taloustieto Oy, 1999.

Barceinas–Paredes, F. et al. The screening versushuman capital hypotheses: evidence fromFrance and Spain. In: Asplund, R. et al. (eds)Public funding and private returns to education.PURE Final Report, 2001, Part II, p. 96-99.

Baumert, J. Bildungsverläufe und psychosozialeEntwicklung. Berlin: MPI – Max Planck Institutefor Human Development, 2002.

Beck, U. Risk society: towards a new modernity.London: SAGE, 1994.

Becker, G. Human capital: a theoretical and empir-ical analysis with special reference to educa-tion. New York: National Bureau of EconomicResearch, 1964.

Becker, G. Human capital. Third edition. Chicago,IL: University of Chicago Press, 1993.

Becker, R. Berufliche Weiterbildung und Berufsver-lauf. In: Meier, A.; Rabe-Kleberg, U. (eds) Weit-erbildung, Lebenslauf, sozialer Wandel. Berlin:Luchterhand, 1993.

Becker, R. Bildung und Lebenserwartung inDeutschland: Eine empirischeLängsschnitt-untersuchung aus der Lebensver-laufsperspektive. Zeitschrift für Soziologie,1998, Vol. 27, No 2, p. 133-150.

Becker, R.; Schömann, K. Berufliche Weiterbildungund Einkommensdynamik: Eine Längss-chnittstudie mit besonderer Berücksichtigungvon Selektionsprozessen. Kölner Zeitschrift fürSoziologie und Sozialpsychologie, 1996,Vol. 48, No 3, p. 426-461.

Becker, R., Schömann, K. Berufliche Weiterbildungund Einkommenschancen im Lebensverlauf:Empirische Befunde für Frauen und Männer inWest- und Ostdeutschland. In Beer, D. et al.(eds) Die wirtschaftlichen Folgen der Aus- undWeiterbildung. Munich: Hampp, 1999.

Bender, S.; Dietrich, H. Unterschiedliche Startbe-dingungen haben langfristige Folgen: derEinmündungsverlauf der Geburtskohorten1964 und 1971 in Ausbildung und Beschäfti-gung – Befunde aus einem IAB Projekt. Nurem-berg: IAB, 2001 (Serie Werkstattbericht No 11).

Berger, J. Modernization theory and economicgrowth. In: Schelke, W. et al. (eds) Paradigms ofsocial change: modernization, development,transformation, evolution. Frankfurt: Campus,2000, p. 31-47.

Blaney, D. L.; Inayatullah, N. Neo-modernization?IR and the inner life of modernization theory.European Journal of International Relations,2002, Vol. 8, Issue 1, p. 103-138.

Blossfeld, H. P.; Timm, A. Das Bildungssystem alsHeiratsmarkt. Eine Längsschnittanalyse derWahl von Heiratspartnern im Lebenslauf.Bremen: Sonderforschungsbereich 186 der

References

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 371

Universität Bremen, 1997 (Sfb-Arbeitspapier,No 43).

Bjørnåvold, J.; Tissot, P. Glossary. In: Bjørnavold, J.Making learning visible: identification, assess-ment and recognition of non-formal learning inEurope. Luxembourg: Office for Official Publi-cations of the European Communities, 2000,p. 199-221 (Cedefop Reference Series).

Blackwell, L.; Bynner, J. Learning, family formationand dissolution. London: Centre for Researchon the Wider Benefits of Learning, Institute ofEducation, 2002.

Blau, P. M.; Duncan, O. D. The American occup-tional structure. New York: WiIley, 1967.

Bloomer, M.; Hodkinson, P. Moving into furthereducation: the voice of the learner. London:Further Education Development Agency, 1997.

Bloomer, M.; Hodkinson, P. Learning careers: conti-nuity and change in young people’s disposi-tions to learning. British Educational ResearchJournal, 2000, Vol. 26, No 5, p. 583-596.

Blossfeld, H.-P. Berufseintritt und Berufsverlauf.Eine Kohortenanalyse über die Bedeutung desersten Berufs in der Erwerbsbiographie. Frank-furt/Mannheim: 1985a (Sfb3-Arbeitspapier,No 161).

Blossfeld, H.-P. Zur Repräsentativität derSfb-3-Lebensverlaufsstudie. Ein Vergleich mitDaten aus der amtlichen Statistik.Frankfurt/Mannheim: 1985b (Sfb3-Arbeitspa-pier, No 163).

Blossfeld, H.-P. Kohortendifferenzierung im Karri-ereprozess. Eine Längsschnittstudie über dieVeränderung der Bildungs- und Berufschancenim Lebenslauf. Frankfurt am Main: Campus,1989.

Blossfeld, H.-P. Bildung, Arbeit und soziale Ungle-ichheit im Globalisierungsprozess. Bielefeld:Faculty of Sociology, 1999 (Globalife WorkingPaper Series No 1).

Blossfeld, H.-P.; Hannan, M.T.; Schömann, K. DieBedeutung der Bildung im Arbeitsein-kommensverlauf. Theoretische Ansätze imLichte der Ergebnisse einer Längsschnittunter-suchung. In: Lüdeke, R. (eds) Bildung, Bildungs-finanzierung und Einkommensverteilung I.Berlin: Duncker und Humblot, 1993.

Blossfeld, H.-P.; Huinink, J. Bildung, Karriere unddas Alter bei der Eheschließung von Frauen. In:Felderer, B. (ed.) Bevölkerung und Wirtschaft.Berlin: Duncker & Humblot, 1990, p. 539-553.

Blossfeld, H.-P.; Mayer, K.U. BerufsstrukturellerWandel und soziale Ungleichheit. KölnerZeitschrift für Soziologie und Sozialpsy-chologie, 1998, Vol. 43, No 4, p. 671-696.

Blossfeld, H.-P.; Rohwer, G. Techniques of eventhistory modeling. New approaches to causalanalysis. Hillsdale, NJ: Lawrence ErlbaumAssociates, 1995.

Böttger, A. Gewalt und Biographie. Eine qualitativeAnalyse rekonstruierter Lebensgeschichtenvon 100 Jugendlichen. Baden-Baden: Nomos1998.

Boockmann, B.; Steiner, V. Cohort effects and thereturns to education in West Germany.Mannheim: ZEW – Centre for EuropeanEconomic Research, 2000 (Discussion PaperNo 00-05).

Born, C. Erstausbildung und weiblicher Lebenslauf.Zeitschrift für Soziologie der Erziehung undSozialisation, 2000, No 3 (Beiheft), p. 50-65.

Boudon, R. Education, opportunity, and socialinequality. New York: John Wiley, 1974.

Bratberg, E.; Nilsen, O. A. Transition from school towork: search time and job duration. Bonn: IZA –Institute for the Study of Labor, 1998 (Discus-sion Paper No 27).

Brauns, H.; Müller, W.; Steinmann, S. Educationalexpansion and returns to education: a compar-ative study on Germany, France, the UK andHungary. Mannheim: MZES – MannheimerZentrum für Europäische Sozialforschung,1997 (Working Papers I, No 23).

Breen, R. The persistence of class origin inequalitiesamong school leavers in the republic of Ireland,1984-1993. Paper presented to the ESF work-shop: Transitions in youth: comparisons overtime and across countries. Oostvoorne, TheNetherlands, 22-25 September, 1995.

Bretagnolle, J.; Huber-Carol, C. Sous-estimationdes contrastes due à l’oublie de variables perti-nentes dans le modèle de Cox pour les duréesde survie avec censure. Comptes Rendus del’Academie des Sciences, 1985, Serie I,Vol. 300, p. 359-362.

Brückner, E. Erhebung ereignisorientierter Lebens-verläufe als retrospektive Längsschnitt-rekonstruktion. In: Hauser,R.; Ott,N.; Wagner,G.(eds) Mikroanalytische Grundlagen derGesellschaftspolitik. Bd. 2: Erhebungsverfahren,Analysemethoden und Mikrosimulation. Berlin:Akademie Verlag, 1994, p. 38-69.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 372



Brunello, G. On the complementary of educationand training in Europe. Bonn: IZA – Institute forthe Study of Labor, 2001 (Discussion PaperNo 309).

Brunello, G.; Comi, S. Education and earningsgrowth. Evidence from 11 European countries.PURE project, 2000.

Brunello, G.; Comi, S.; Lucifora, C. The returns toeducation in Italy. A new look at the evidence.Bonn: IZA – Institute for the Study of Labor,2001a (Discussion Paper No 130).

Brunello, G.; Comi, S.; Lucifora, C. Education, earn-ings growth and cohorts. In: Asplund, R. et al.(eds) Public funding and private returns toeducation PURE – Final report, 2001b, p. 22-34.

Brunello, G.; Miniaci, R. The economic returns toschooling for Italian men. An evaluation basedon instrumental variables. Labour Economics,1999, No 6, p. 509-519.

Bryson,A.; Lissenburgh,S. The returns to graduation.Department for Education and Employment,Research Studies RS15. London: HMSO, 1996.

Büchel, F. Die Qualität der Wiederbeschäftigungnach ununterbrochener und nach ‘perforierter’Langzeitarbeitslosigkeit. Nürnberg: IAB, 1992(BeitrAB 162).

Bukodi, E.; Robert, P. The effect of the globalizationprocess on the transition to adulthood inHungary. Bamberg: Department of Sociology,Bamberg University, 2002 (Globalife workingpaper Series, No 27).

Bynner, J. Basic skills and labour market entry.Paper presented to the ESF workshop: Deter-minants of individual success in transitions tothe labour market. Seelisberg, Switzerland, 16-19 September 1994.

Bynner, J. Skills and disadvantage in labour-marketentry. In: Haubrich, K.; Reubel, F. (eds) Beruf-sstart benachteiligter Jugendlicher in Europa.Konzepte zur beruflichen Integration imregionalen Kontext. München: DJI, 1995(Arbeitspapier No 2).

Bynner, J. Education for what? Education andTraining, 1998, Vol. 40, No 1, p. 4-5.

Bynner, J.; Parsons, S. Qualifications, basic skillsand accelerating social exclusion. Journal ofEducation and Work, 2001, Vol. 14, No 3,p. 279-291.

Bynner, J. et al. Improving adult basic skills: bene-fits to the individual and to society. London:HMSO, 2001 (DfEE Research Report RR251).

Caporaso, J. A. Dependency theory: continuitiesand discontinuities in development studies.International Organization, 1980, Vol. 34,Issue 4, p. 605-634.

Chevalier, A. et al. The returns to education in theUK. In: Asplund, R.; Pereira, P. T. (eds) Returnsto human capital in Europe: a literature review.Helsinki: Taloustieto Oy, 1999.

Clausen, J. The life course. A sociological perspec-tive. Englewood Cliffs: Prentice Hall, 1986.

Creswell, J. W. Qualitative inquiry and researchdesign. Choosing among five traditions.London: Sage, 1998.

Coles, B. et al. Literature review of the costs of being‘not in education, employment or training’ atage 16-18. London: DfEE – Departement forEmployment and Education, 2002. (ResearchBrief, No 347)

Curti, M. Does economic stagnation effect unem-ployed workers, even when re-employed?International Journal of Manpower, 1998,Vol. 19, No 7, p. 410-423.

Dale, A.; Egerton, M. Highly educated women:evidence from the National Child DevelopmentStudy. London: DfEE – Departement forEmployment and Education, 1997. (DfEEResearch Studies, RS25).

Deding, M.; Dall Schmidt, T. Differences in incomeinequality across Europe – Market driven or …?Colchester: EPAG – European Panel AnalysisGroup, 2002 (Working Paper, No 37).

Denny, K.; Harmon, C. Measuring the economicreturn to education: evidence from the Irishschool leavers’ survey 1990-1995. Dublin,Department of Economics, University College,1998.

Descy, P.; Tessaring, M. (eds) Training and learningfor competence: second report on vocationaltraining research in Europe: synthesis report.Luxembourg: Office for Official Publications ofthe European Communities, 2001 (CedefopReference series, 6).

Desjardin, R. The effects of learning on economicand social well-being: a comparative analysis.Peabody Journal of Education, 2001, Vol. 76,No 3 and N 4, p. 222-246.

Deutsches PISA-Konsortium (ed.) PISA 2000 –Basiskompetenz von Schülern und Schü-lerinnen im internationalen Vergleich. Opladen:Leske + Budrich, 2001.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 373

Deutsches PISA-Konsortium (ed.) PISA 2000 – DieLänder der Bundesrepublik im Vergleich.Opladen: Leske + Budrich, 2002.

Dex, S. (ed.) Life and work history analyses. Quali-tative and quantitative developments. London:Routledge, 1991.

Dietz, G.-H.; Matt, E. Begrenzte Handlungsspiel-räume bei der Berufsfindung. Der Übergangvon Haupt- und Sonderschülern in das Berufs-bildungssystem. Zeitschrift für Berufs- undWirtschaftspädagogik, 1994, No 5, p. 510-524.

Diewald, M.; Sørensen, A. Erwerbsverläufe undsoziale Mobilität von Frauen und Männern inWest- und Ostdeutschland: makrostrukturelleUmbrüche und Kontinuitäten im Lebensverlauf.In: Diewald, M.; Mayer, K.U. (eds) Zwischenbi-lanz der Wiedervereinigung. Leverkusen: Leskeund Budrich, 1996.

Dore, R. The Diploma disease: education, qualifica-tion, and development. Berkeley: University ofCalifornia Press, 1976.

Duncan, O. D.; Haller, A. O.; Portes, A. Peer influ-ences on aspirations. A reinterpretation. In:Blalock, H. M. Jr. (ed.) Causal models in socialsciences. Chicago: Aldine-Atherton, 1971,p. 219-224.

Ecarius, J. Individualisierung und soziale Reproduk-tion im Lebensverlauf. Konzepte derLebenslaufforschung. Opladen: Leske +Budrich, 1996.

Elder, G. H. Age differentiation and the life course.Annual Review of Sociology, 1975, Vol. 1,p. 165-190.

Elder, G. H. (ed.) Life course dynamics. Trajectoriesand transitions 1968 – 1980. Ithaca UA: CornellUniversity Press, 1985.

Elder, G. H. Lives and social change. In: Heinz, W. R.(ed.) Theoretical advances in life courseresearch. Weinheim: Deutscher Studien Verlag,1991, p. 58-86.

Engerman, D. C. et al. (eds) Staging growth:modernization, development and the globalCold War. Amherst: University ofMassachusetts Press, 2003.

Ermisch, J.; Francesconi, M. Educational choices,families and young people’s earnings. Journalof Human Resources. 2000, Vol. 35, No 1,p. 143-176.

European Commission. The Lisbon EuropeanCouncil – An agenda of economic and socialrenewal for Europe. Contribution of the Euro-

pean Commission to the special EuropeanCouncil in Lisbon, 23-24 March 2000, Brussels:European Commission, 2000 (Doc 00/7).

European Commission. NAP: National Action Planson Employment 2002. Actual Versions for eachMember State can be found on thisEU-Website:http://europa.eu.int/comm/employment_social/news/2002/may/naps2002_en.html [cited11.07.2003]

ETF – European Training Foundation. Glossary oflabour-market terms and standard andcurriculum development terms. Turin: ETF 1998.

Fabrizio, B. Educational performance and educa-tional returns at entry into the Italian labourmarket. Bamberg: Department of Sociology,2000 (Globalife working paper Series, No 10).

Farvaque, N.; Salais, R. Implementing allowancesfor young people in difficulty in France:enhancing capabilities or increasing selectivity?Ninth International Congress, Geneva 12-14September 2002. Geneva: BIEN – BasicIncome European Network, 2002Available from Internet:http://www.etes.ucl.ac.be/bien/Files/ Papers/2002FarvaqueSalais.pdf[cited 11.07.2003]

Featherman, D. L.; Hauser, R. M. Opportunity andchange. New York: Academic Press, 1978.

Featherman, D. L.; Shavit, Y. Human developmentand society: a transactional relationship. HumanDevelopment, 1990, Vol. 33, p. 171-178.

Felstead, A.; Krahn, H.; Powell, M. Contrastingfortunes across the life course: non-standardwork among women and men in Canada and theUnited Kingdom. Leicester: CLMS – Center forLabour Market Studies, 1997 (Working PaperNo 17) Avalaible from Internet: http://www.clms.le.ac.uk/publications/wkpapers/working_paperNo17.pdf[cited 11.07.2003].

Fitzenberger, B.; Prey, H. Berufliche Weiterbildungund die Stabilität der Beschäftigung. Eine Eval-uation auf Basis des SozioökonomischenPanels 1999. In: Beer, D. et al. (eds) Diewirtschaftlichen Folgen der Aus- und Weiterbil-dung. Munich: Hampp, 1999.

Fougère, D.; Goux, D.; Maurin, E. Formationcontinue et carrières salariales. Une évaluationsur données individuelles. Annales d’Économieet de Statistique, 2001, No 62, p. 49-70.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 374



Franz, W. Arbeitsmarktökonomik. Berlin: Springer1991.

Frenzel, H. J. Bildung und Partnerwahl. ZUMANachrichten, 1995, Vol. 19, No 36, p. 61-87.

Furlong, A.; Hammer, T. Youth unemployment andmarginalisation in northern Europe. Oslo: Nova,2000.

Galler, H. P.; Pötter, U. Zur Robustheit von Schätz-modellen für Ereignisdaten. In: Hujer, R.;Schneider, H.; Zapf, W. (eds) Heraus-forderungen an den Wohlfahrtsstaat im struk-turellen Wandel. Frankfurt am Main: Campus,1992, p. 379-405.

Ganzach, Y. Parents’ education, cognitive ability,educational expectations and educationalattainment: Interactive effects. British Journalof Educational Psychology, 2000, Vol. 70,p. 419-441.

Gash, V.; O’Connell, P. J. The Irish graduate labourmarket: a six year follow-up survey of third levelgraduates from 1992. Dublin: ESRI – Economicand Social Research Institute, 2000.

Gennep, A. V. The rites of passage. Chicago:University of Chicago Press, 1960.

Giele, J. Z.; Elder, G. H. Jr. (eds) Methods of lifecourse research: qualitative and quantitativeapproaches. London: Sage 1998.

Glick, P. C. The family cycle. American Journal ofSociology, 1947, No 12, p. 164-174.

Goldthorpe, J. H. Social mobility and class structurein modern Britain. Oxford, Clarendon Press:1980.

Goux, D.; Maurin, E. Education, expérience etsalaire: tendences récentes et évolution de longterme. Economie et Prévision, 1994, Vol. 116,p. 155-178.

Green, F.; Hoskins, M.; Montgomery, S. The effectsof training, further education and YTS on theearnings of young employees. Leicester:Faculty of Social Science, 1994. (DiscussionPaper, No 10).

Gridley, G et al. Is there a healthy worker effect forcancer incidence among women in Sweden?American Journal Industrial Medicine, 1999,Vol. 36, p. 193-99.

Hagenaars, J. A.; Cobben, N. P. Age, cohort andperiod: a general model for the analysis of socialchange. Netherlands Journal of Sociology,1978, No 14, p. 59-91.

Hægeland, T.; Klette, T. J.; Salvanes, K. G. Decliningreturns to education in Norway? Comparing

esitmates across cohorts, sectors and overtime. Scandinavian Journal of Economics,1999, Vol. 101, No 4, p. 555-576.

Hakim, C. Research design. Strategies and choicesin the design of social research. London: Allenand Unwin, 1987.

Hall, E. M.; Johnson, J. V.; Tsou, T.-S. Women, occu-pation and risk of cardiovascular morbidity andmortality. Occupational Medicine: State of theArt Reviews, 1993, Vol. 8, p. 709-719.

Hamerle, A.; Tutz, G. Diskrete Modelle zur Analysevon Verweildauer und Lebenszeiten. Frankfurt:Campus, 1989.

Hammond, C. Learning to be healthy. London:Centre for Research on the Wider Benefits ofLearning, 2002 (Research Brief, No RCB07).

Hammarström, M. Varför inte högskola?: En longi-tudinell studie av olika faktorers betydelse förstudiebegåvade ungdomars utbildningskarriär.Göteborgs: Institutionen för pedagogik, 1996.(Göteborg studies in Educational sciences,No 107)

Hannan, D. F. Education, vocational training andlabour market transition (VTLMT) amongstlower level school leavers in 4 European coun-tries. Dublin: ESRI, 2001 (ESRI report).

Hannan, D. F.; Werquin, P. Education and labourmarket change: the dynamics of education towork transitions in Europe. A review of the TSERprogramme. In: Descy, P.; Tessaring, M. (eds)Training in Europe: second report on vocationaltraining research in Europe 2000: backgroundreport. Luxembourg: Office for Official Publica-tions of the European Communities, 2001,Vol. 3, Part 5, p. 91-135 (Cedefop Referenceseries, 3008).

Harmon, C.; Walker, I.; Westergaard-Nielsen, N.Returns to education in Europe. In: Asplund, R.et al. (eds) Public funding and private returns toeducation. PURE Final Report, 2001, Part I,p. 13-21.

Hartog, J.; Pereira, P. T.; Vieira, J. A. C. Changingreturns to education in Portugal during the1980s and early 1990s : OLS and quantileregression estimators. Mimeo, 1993.

Hauser, R.; Ott, N.; Wagner, G. (eds), Mikroana-lytische Grundlagen der Gesellschaftspolitik.Berlin: Akademie Verlag, 1994.

Heinz, W. R. (ed.) Theoretical advances in life courseresearch. Weinheim: Dt. Studien-Verlag, 1997.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 375

Heinz, W. R. From education to work: cross-nationalperspectives. Revision of papers delivered at aconference at the University of Toronto, April1996. Cambridge: University Press, 1999.

Hillmert, S. Ausbildungssysteme und Arbeitsmarkt.Lebensverläufe in Deutschland and Großbritan-nien im Kohortenvergleich. Wiesbaden: West-deutscher Verlag, 2001.

Hillmert, S. Stabilität und Wandel des ‘deutschenModells’: Lebensverläufe im Übergang zwis-chen Schule und Beruf. In: Wingens, M.; Sack-mann, R. (eds) Bildung und Beruf. Ausbildungund struktureller Wandel in der Wissensge-sellschaft. Weinheim, München: Juventa, 2002.

Holler, M. J. Ökonomische Theorie des Arbeits-marktes. Darmstadt: WissenschaftlicheBuchgesellschaft, 1986.

Hradil, S.; Immerfall, S. (eds) Die westeuropäischenGesellschaften im Vergleich. Opladen: Leske +Budrich, 1997.

Hsiao, C. Analysis of panel data. Cambridge:Cambridge University Press, 1986.

Huinink, J. Bildung und Familienentwicklung imLebenslauf. Zeitschrift für Erziehungs-wissenschaft, 2000, Vol. 3, No 2, p. 209-227.

Hujer, R.; Caliendo, M.; Zeiss, Ch. Macroecono-metric evaluation of active labour-market policy– a case study (in this volume).

Hujer, R.; Wellner, M. Berufliche Weiterbildung undindividuelle Arbeitslosigkeitsdauer in West- undOstdeutschland: eine mikroökonometrischeAnalyse. Mitteilungen für Arbeitsmarkt- undBerufsforschung, 2000, No 3, p. 405-417.

Hullen, G. Lebensverläufe in West- undOstdeutschland. Längsschnittanalysen desdeutschen Family and Fertility Surveys.Leverkusen: Leske + Budrich, 1998.

Hullen, G. The effects of education and employmenton marriage and first birth. Paper in the frame-work of the FFS Flagship conference ‘Partner-ship and fertility – a revolution’, Brussels, 29-31May 2000.

Jahoda, M. Employment and unemployment. Asocial-psychological analysis. Cambridge:Cambridge University Press, 1982.

Jahoda, M.; Lazarsfeld, P. F.; Zeisel, H. Die Arbeit-slosen von Marienthal. Ein sozio-graphischerVersuch über die Wirkungen langandauernderArbeitslosigkeit. Frankfurt am Main: Suhrkamp,1933.

Jahnukainen, M. Social Exclusion and dropping outof education. In: Visser, J; Daniels, H; Cole, T.(eds) Emotional and behavioural difficulties inmainstream schools. Amsterdam: Elsevier,2001, Vol. 1, p. 1–12.

Kaiser, L.; Siedler, T. Exits from unemploymentspells in Germany and the United Kingdom.Colchester: EPAG – European Panel AnalysisGroup, 2000 (Working paper, No 7).

Katz, E., Ziderman, A. Investment in generaltraining: the role of information and labourmobility. Economic Journal, 1990, Vol. 100,p. 1147-1158.

Kelvin, P.; Jarrett, J. E. Unemployment. Its socialpsychological effects. Cambridge: CambridgeUniversity Press, 1985.

Keogh, H.; Downes, T. VTOSpells Success. Dublin:Department of Education and Science, 1998.

Kettunen, J. The effects of education on the dura-tion of unemployment. Review of LabourEconomics and Industrial Relations, 1994,Vol. 8, No 2, p. 331-352.

Kieselbach, T. (ed.) Youth unemployment and socialexclusion: a comparison of six European coun-tries. Bremen: University of Bremen, 2000(Psychology of Social Inequality Series, Vol. 9,Yuseder Publication, No 1).

Klijzing, E. Globalization and the changing role ofeducation in the process of entry into the labourmarket: a cross-national comparison of fivecountries. Bielefeld: Faculty of Sociology, 2000(Globalife Working paper, No 12).

Kohli, M. Erwartungen an eine Soziologie desLebenslaufs. In: Kohli, M. (ed.) Soziologie desLebenslaufs. Darmstadt/Neuwied: Luchter-hand, 1978, p. 9-31.

Kohli, M. Normalbiographie und Individualität. Zurinstitutionellen Dynamik des gegenwärtigenLebenslaufregimes. In: Brose, H. G.; Hilden-brand, B. (eds) Vom Ende des Individuums zurIndividualität ohne Ende. Opladen: Leske +Budrich, 1988, p. 33-53.

Koivuluhta, M. Ammatti – intressit ja ura: pohjoiskar-jalaisen ammattikoulutetun ikäluokan seurantavuosina 1975-1991. Joensuu: Joensuunyliopisto, 1999.

Konietzka, D. Die soziale Differenzierung der Über-gangsmuster in den Beruf. Kölner Zeitschrift fürSoziologie und Sozialpsychologie, 2002,Vol. 45, No 4, p. 645-673.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 376



Konietzka, D. Berufliche Aus- und Fortbildung inder Lebensverlaufsperspektive. Ein Vergleichdes Ausbildungsverhaltens sechs west-deutscher Geburtskohorten. Zeitschrift fürPädagogik, 1999, Vol. 45, No 6, p. 807-831.

Kortteinen, M.; Tuomikoski, H. Miten työttömätselviytyvät? [How do the unemployed survive?]Yhteiskuntapolitiikka, 1998, Vol. 63/1, p. 5–13.

Lasheras, C. et al. Effects of education on thequality of life, diet, and cardiovascular riskfactors in an elderly Spanish community popu-lation. Experimental Aging Research, 2001,Vol. 27, p. 257-270.

Lauer, C.; Steiner, V. Returns to human capital inGermany: review of the empirical literature. In:Asplund, R.; Pereira, P. T. (eds) Returns tohuman capital in Europe: a literature review.Helsinki: Taloustieto Oy, 1999.

Lauer, C. Family background, cohort and education.A French-German comparison., Mannheim:Centre for European Economic Research, 2002(ZEW Discussion Paper, No 02-12).

Lemkow, L. et al. Youth unemployment and socialexclusion in Spain. In: Kieselbach, T. (ed.) Youthunemployment and social exclusion: a compar-ison of six European countries. Bremen: Univer-sity of Bremen, 2000 (Psychology of SocialInequality Series, Vol. 9, Yuseder Publication,No 1).

Levy, R. Status passages as critical life course tran-sitions. A theoretical sketch. In: Heinz, W. R.(ed.) Theoretical advances in life courseresearch. Weinheim: Deutscher Studien Verlag,1991.

Lexis, W. Einleitung in die Theorie der Bevölkerungs-statistik. Strassburg: Trübner, 1875.

Liefbroer, A. Transition from youth to adulthood inthe Netherlands. Bamberg: Department ofSociology, 2002 (Globalife working paper,No 22).

Mackenbach, J. P. et al. Socioeconomic determi-nants of healthy aging. From description toexplanation (SEdHA). Rotterdam: ErasmusUniversity, Department of Public Health, 1999.(Unpublished proposal submission form).

Magoula, T.; Psacharopoulos, G. Schooling andmonetary rewards in Greece: contributions to adebate. Athens: University of Economics andBusiness, 1997 (Discussion Paper, No 90).

Mannheim, K. The problem of generations. In:Kecskemeti, P. (ed.) Essays on the sociology of

knowledge. London: Routledge and KeganPaul, 1952 (German original from 1928).

Marmot, M; Wilkinson, J. (eds) The social determi-nants of health. Oxford: Oxford UniversityPress, 1999.

Matt, E.; Prein, G.; Schumann, K. F. Delinquenz undBerufsbildung. Illegale Drogen als Risiko beimÜbergang in das Erwerbsleben. In: Heinz, W. R.et al. (eds) Was prägt Berufsverläufe? –Lebenslaufdynamik und Institutionenpolitik.Beiträge zur Arbeitsmarkt- und Berufs-forschung. Nürnberg: IAB (215), 1998,p. 169-215.

Mayer, K. U. Lebenslaufforschung. In: Voges, W.(ed.) Methoden der Biographie- undLebenslaufforschung. Opladen: Leske undBudrich, 1987.

Mayer, K. U. Lebensverläufe und sozialer Wandel.Opladen: Westdeutscher Verlag, 1990.

Mayer, K. U. Ausbildungswege und Berufskarri-eren. In: Bundesinstitut für Berufsbildung (ed.)Forschung im Dienst von Praxis und Politik.Bielefeld: Bertelsmann, 1996.

Mayer, K. U.; Blossfeld, H. P. Die gesellschaftlicheKonstruktion sozialer Ungleichheit imLebensverlauf. In: Berger, P. A.; Hradil, S. (eds)Ungleichheit und Lebensverlauf. Göttingen:1990 (Sozialen Welt, No 7).

Mayer, K. U.; Brückner, H. Lebensverläufe undgesellschaftlicher Wandel. Konzeption, Designund Methodik der Erhebung von Lebensver-läufen der Geburtsjahrgänge 1954-1956 und1959-1961. Berlin: Max-Planck-Institut fürBildungsforschung, 1995 (Materialien aus derBildungsforschung, No 48).

Mayer, K. U.; Henz, U.; Maas, I. Social mobilitybetween generations and across the workinglife: biographical contingency, time dependencyand cohort differentiation – results from theGerman life history study. Berlin: Max-Planck-Institut für Bildungsforschung, 1991.

Mayer, K. U.; Huinink, J. Age, period, and cohort inthe study of the life course: a comparison ofclassical A-P-C analysis with event history anal-ysis or farewell to LEXIS? In: Magnusson, D.(ed.) Data quality in longitudinal research.Cambridge: Cambridge University Press, 1990.

Mayer, K. U.; Huinink, J. Lebensverläufe undgesellschaftlicher Wandel: von der Kohorten-analyse zur Lebensverlaufsanalyse. In:Hauser, R. et al. (eds) Mikro-analytische Grund-

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 377

lagen der Gesellschaftspolitik Bd. 1:Ausgewählte Probleme und Lösungsansätze.Frankfurt: Campus 1994, p. 92-111.

Mayer, K. U.; Müller, W. The state and the structureof the life course. In: Sørensen, A. B.; Weinert, F.;Sherrod, L. R. (eds) Human development andthe life course. Hillsdale: Erlbaum 1986,p. 217-245.

McIntosh, S.; Steedman, H. Low skills: a problemfor Europe. Luxembourg: Office for OfficialPublications of the European Communities,1999. (Project Final Report Series - TSERArea 2).

Menard, S. Longitudinal research. Newbury Park:Sage Publications, 1991.

Mertens, L. Labour competence: emergence,analytical framework and institutional models.ILO (Cinterfore/ILO), 1999.

Meulemann, H. Schullaufbahnen, Ausbildungskar-rieren und die Folgen im Lebensverlauf. DerBeitrag der Lebenslaufsforschung zurBildungssoziologie. In: Mayer, K. U. (ed.)Lebensverläufe und sozialer Wandel. Opladen:Westdeutscher Verlag, 1990a.

Meulemann, H. Studium, Beruf und der Lohn vonAusbildungszeiten. Zeitschrift für Soziologie,1990b, Vol. 19, No 4, p. 248-264.

Meulemann, H. Stichwort Lebenslauf, Biographieund Bildung. Zeitschrift für Erziehungswis-senschaft, 1999, Vol. 3, No 2.

Meyer, T. Forschungsprojekt TREE: Transitionenzwischen Erstausbildung und Erwerbsleben inder Schweiz. Bern: ABF – Amt für Bildungs-forschung, 2002.

Meyer, W. Individuelle Erwerbschancen inOstdeutschland. Auswirkungen des wirtschafts-strukturellen Wandels. Wiesbaden: DeutscherUniversitätsverlag, 1997.

Mincer, T. The production of human capital in the lifecycle of earnings. Journal of Labor Economics,1997, Vol. 15, No 1, p. 27-46.

Mincer, J. Schooling, experience and earnings. NewYork: National Bureau of Economic Research,1974.

Müller, W. Mobilitätsforschung und Arbeitsmarkt-theorie. In: Knepel, H.; Hujer, R. (eds) Mobilität-sprozesse auf dem Arbeitsmarkt. Frankfurt:Campus, 1985, p. 17-40.

Müller, W. Erwartete und unerwartete Folgen der Bildungsexpansion. In: Friedrichs, J.;Lepsius, M. R.; Mayer, K. U. (eds) Die Diag-

nosefähigkeit der Soziologie. Opladen: West-deutscher Verlag, 1998, p. 81-112.

Müller, W.; Gangl, M.; Scherer, S. Übergangsstruk-turen zwischen Bildung und Beschäftigung. In:Wingens, M.; Sackmann, R. (eds) Bildung undBeruf. Weinheim: Deutscher UniversitätsVerlag, 2002.

Müller, W.; Shavit, Y. Bildung und Beruf im institu-tionellen Kontext. Eine vergleichende Studie in13 Ländern. Zeitschrift für Erziehungswis-senschaft, 1998, No 1, p. 501-533.

Müller, W.; Shavit, Y. Vocational secondary educa-tion. London: Routledge, 2000.

Nehru, V.; Dhareshwar, A. A new database on phys-ical capital stock: sources, methodology andresults. Revista de Analysis Economico, 1993,Vol. 8, No 1, p. 37-59.

Noguera, C.; Golsch, K.; Bonmati, A. Globalizationand occupational mobility in adult male jobcareers in Spain. Bamberg: Department ofSociology, 2002 (Globalife working paper,No 35).

Nummenmaa, A.-R. Koulutus, sukupuoli jaelämänkulku. Nuoruudesta aikuisuuteenyhteiskunnallisessa muutoksessa. Helsinki:Työministeriö, 1996 (Työpoliittisia tutkimuksia,149).

OECD. Knowledge and skills for life. First resultsfrom PISA 2000, Paris: OECD, 2001a.

OECD. The well-being of nations: the role of humanand social capital. Paris: OECD, 2001b.

OECD. Knowledge and skills for life. First resultsfrom PISA 2000. PISA in the news in … Dec.2001 – Jan 2002. Paris: OECD 2002a (availablefor the following EU-countries: Belgium,Denmark, Germany, Greece, Spain, France,Ireland, Italy, Luxembourg, the Netherlands,Austria, Portugal, Finland, Sweden, the UK)

OECD. Programme for international studentassessment (PISA). PISA 2000 Technical report,Paris: OECD, 2002b.

OECD. Reading for change. Performance andengagement across countries. Results fromPISA 2000. Paris: OECD, 2003.

Palma, G. Dependency and development: a criticaloverview. In: Seers, D. (ed.) DependencyTheory. London: Pinter, 1981, p. 20-89.

Pannenberg, M. Weiterbildungsaktivitäten undErwerbsbiographie. Eine empirische Analysefür Deutschland. Studien zur Arbeitsmarkt-forschung Vol. 8. Frankfurt/M.: Campus, 1995.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 378



Papastefanou, G. Zur Güte von retrospektivenDaten. Eine Anwendung gedächtsnis-psychol-ogischer Theorien und Ergebnisse einer Nach-befragung. Frankfurt: Sfb 3-, 1980. (Workingpaper No 29).

Parson, Talcott Societies: evolutionary andcomparative perspectives. Englewood Cliffs:Prentice-Hall,1966.

Payne, J. England and Wales Youth Cohort Study:options at 16 and outcomes at 24: a comparisonof academic and vocational education andtraining routes. Sheffield: DfEE, 1995.

Payne, J. Options at 16 and outcomes at 24. Acomparison of academic and vocational educa-tion and training routes. Sheffield: DfEE, 1997(Youth Cohort Report, No 35).

Payne, J. Routes at sixteen: trend and choices in thenineties. London: DfEE – Departement forEmployement and Education, 1998 (ResearchBrief, No 55).

Payne, J. Young people not in education, employ-ment or training: data from the England andWales Youth Cohort Study. Sheffield: DfEE,2000.

Payne, J. Work-based training for young people:data from the England and Wales Youth CohortStudy. Sheffield: DfEE, 2001 (Research Report)

Pedersen, P.; Dall Schmidt, T. Search activity andre-employment – a European perspective.Colchester: EPAG – European Panel AnalysisGroup, 2002 (Working paper, No 34).

Pedersen, P. et al. Wage differentials between thepublic and the private sectors. Journal of PublicEconomics, 1990, Vol. 41, p. 125-145.

Pfeiffer, F. Training and individual performance:evidence from microeconometric studies. In:Descy, P.; Tessaring, M. (eds) Training in Europe:second report on vocational training research inEurope 2000: background report. Luxembourg:Office for Official Publications of the EuropeanCommunities, 2001, Vol. 3, Part 5, p. 7-41(Cedefop Reference series, 3008).

Pischke, J. et al. Changes in wage structure, familyincome, and children’s education. Chicago:JCPR – Joint Center for Poverty Research, 2000(Working Papers, No 208).

Popper, K. R. The open society and its enemies.Princeton: Princeton University Press, 1966.

Preston, J.; Hammond, C. The wider benefits offurther education. Practitioner views. London:Institute of Education, 2002.

Rabe-Kleberg, U. Frauen auf dem Weg zurBildungsbiographie? Frauenforschung, 1994,Vol. 11, No 4, p. 5-16.

Rantakeisu, U. et al. Youth unemployment andsocial exclusion in Sweden. In: Kieselbach, T.(ed.) Youth unemployment and social exclusion:a comparison of six European countries.Bremen: University of Bremen, 2000(Psychology of Social Inequality Series, Vol. 9,Yuseder Publication, No 1).

Roberts, B. Biographical research. Buckingham:Open University Press, 2002.

Rogers, W. L. Estimable functions of age, period,and cohort effects. American SociologicalReview, 1982, Vol. 47, p. 774-787.

Rogoff, N. Recent trends in occupational mobility.Glencoe: The Free Press, 1953.

Rostow, W. W. The stages of economic growth: annon-communist manifesto. New York:Cambridge University Press, 1960.

Ruspini, E. Longitudinal analysis: a bridge betweenquantitative and qualitative social research.Quality and Quantity, 1999, Vol. 33, No 3(Special issue).

Ryder, N. B. The cohort as a concept in the study ofsocial change. American Sociological Review,1965, Vol. 30, p. 843-861.

Sackmann, R. Generationsspezifische Arbeits-marktchancen im internationalen Vergleich. In:Berger, P. (ed.) Die Erwerbsgesellschaft.Opladen: Leske + Budrich, 2001.

Sanderson, S. K. Macrosociology. An introductionto human societies. New York: Harper Collins,1995 (Third edition).

Sauer-Schiffer, U. Biographie und Management.Eine qualitative Studie zum Leitungshandelnvon Frauen in der Erwachsenenbildung.Münster: Waxmann, 2000.

Schaeper, H.; Kühn, T.; Witzel, A. DiskontinuierlicheErwerbskarrieren und Berufswechsel in den1990ern: Strukturmuster und biographischeUmgangsweisen betrieblich ausgebildeterFachkräfte. Mitteilungen für Arbeitsmarkt- undBerufsforschung, 2000, No 1, p. 80-99.

Schallberger, U.; Spiess Huldi, C. Die ZürcherLängsschnittstudie von der Schulzeit bis zummittleren Erwachsenenalter. Zeitschrift für Sozi-ologie der Erziehung und Sozialisation, 2001,Vol. 21, No 1, p. 80-89.

Schnabel, R.; Schnabel, I. Family and gender stillmatter. The heterogeneity of returns to educa-

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 379

tion in Germany. Mannheim: ZEW, 2002(Discussion Paper, No. 02-67).

Schömann, K. The dynamics of labour earningsover the life course. A comparative and longitu-dinal analysis of Germany and Poland. Berlin:Ed. Sigma, 1994.

Schömann, K.; Becker, R. Participation in furthereducation over the life course: a longitudinalstudy of three birth cohorts in the federalrepublic of Germany. European SociologicalReview, 1995, Vol. 11, No 2, p. 187-208.

Schrijvers, C. T. M. et al. De achtergronden vansociaal-economische gezondheidsverschillen:resultaten uit het GLOBE onderzoek. In:Stronks, K.; Hulshof, J. (eds) De kloofverkleinen. Assen: Koninklijke Van Gorcum,2001.

Schuller, T. et al. Learning, continuity and change inadult life. London: Centre for Research on theWider Benefits of Learning, 2002.

Schumann, K. F; Mariak, V. BenachteiligungJugendlicher im Bildungssystem und auf demArbeitsmark als Weichenstellung für eine krim-inelle Karriere – ein Mythos? In: Das Jugend-kriminalrecht als Erfüllungshilfegesellschaftlicher Erwartungen? Dokumenta-tion des Symposiums an der kriminologischenForschungsstelle der Universität zu Köln, 1995,p. 178-189.

Schwarz, N. et al. Cognitive and communicativeaspects of survey measurement. In: Borg, I.;Mohler, P. (eds) Trends and perspectives inempirical social research. Berlin: de Gruyter,1994.

Schwarz, N.; Sudman, S. Autobiographicalmemory and the validity of retrospectivereports. New York: Springer, 1994.

Schwarzfärber, H. P. Das Data Warehouse der BA.Information im Rahmen des Workshop zur Eval-uation der Arbeitsförderung. In: Deeke, A.;Wiedemann, E. (eds) Evaluierung aktiverArbeitsmarktpolitik und Datengrundlagen.Report of a workshop at the Federal Institute ofLabour 9 November 2001. Nürnberg: IAB, 2002(Werkstattbericht No 2).

Senatsverwaltung für Arbeit, Berufliche Bildungund Frauen. Innovation, Beschäftigung, Wach-stum und Wettbewerb. Strategien zurHalbierung der Arbeitslosigkeit. Berlin:Senatsverwaltung, 1997 (Kurzfassung desBerliner Memorandums).

Sengenberger, W. Struktur und Funktionsweise vonArbeitsmärkten. Die Bundesrepublik Deutsch-land im internationalen Vergleich. Frankfurt:Campus, 1987.

Shavit, Y.; Blossfeld, P. Persistent inequality:changing educational attainment in thirteencountries. Boulder: Westview Press 1993.

Skov, P. Unges fremtid – meget afgøres tidligt.Copenhague: Danmarks Pædagogiske Institut– DPI, 1998.

Sørensen, A. B. (ed.) Human development and thelife course. Multidisciplinary perspectives Hills-dale: Erlbaum, 1986.

Sofer, C. Schooling, training and transitions: aneconomic perspective (STT). University ofOrleans, 2000 (Final report).

Solga, H. ‘Low education’ as a social stigma. Asociological explanation for the decreasingemployment opportunities of the low educated.Kölner Zeitschrift für Soziologie und Sozialpsy-chologie, 2002, Vol. 54, p. 476-505.

Sokou, K. et al. Youth unemployment and socialexclusion in Greece. In: Kieselbach, T. (ed.)Youth unemployment and social exclusion: acomparison of six European countries. Bremen:University of Bremen, 2000 (Psychology ofSocial Inequality Series, Vol. 9, Yuseder Publi-cation, No 1).

Spence, M. Job market signalling. Quarterly Journalof Economics, 1973, Vol. 87, p. 355-374.

Starrin, B. et al. International debate on social exclu-sion. In: Kieselbach, T. (ed.) Youth unemploy-ment and social exclusion: a comparison of sixEuropean countries. Bremen: University ofBremen, 2000 (Psychology of Social InequalitySeries, Vol. 9, Yuseder Publication, No 1).

Steiner, V.; Lauer, C. Private Returns to Education inGermany. Beiheft der Konjunkturpolitik(Applied Economics Quarterly), 2000, Vol. 51,p. 71-101.

Stone, V.; Cotton, D.; Thomas, A. Mapping troubledlives: young people not in education, employ-ment or training. London: DfEE – Departementfor Employment and Education, 2000.(Research Brief, No 181).

Svalastoga, K. Prestige, class and mobility.Gyldendal: Svenska Bokfoerlag, 1959.

Sweetland, S. R. Human capital theory. Founda-tions of a field of inquiry. Review of educationalresearch, 1996, Vol. 66, No 3, p. 341-360.


01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 380



Terhart, E. Nach PISA. Bildungsqualität entwickeln.Hamburg: Europäische Verlagsanstalt, 2002.

Tessaring, M. Training for a changing society: areport on current vocational education andtraining research in Europe. Second edition.Office for Official Publications of the EuropeanCommunities, 1998. (Cedefop Reference).

Tessaring, M.; Tissot, P.; Descy, P. Glossary of tech-nical terms in education, training and labourmarket research in Europe. Thessaloniki:Cedefop, 2003 (Working document).

Tiryakian, E. A. Modernisation: exhumetur in pace.Rethinking macrosociology in the 1990’s. Inter-national Sociology, 1991, Vol.6, No2, p.165-180

Tsakloglou, P. Aspects of inequality in Greece:measurement, decomposition andinter-temporal change: 1974, 1982. Journal ofDevelopment Economics, 1993, Vol. 40,p. 53-74.

Tsakloglou, P. Changes in inequality in Greece in the1970s and the 1980s. In: Gottschalk, P.;Gustafsson, B.; Palmer, E. (eds) Changingpatterns in the distribution of economic welfare:what happened during the 1980s? Cambridge:Cambridge University Press, 1997.

Tuma, N. B.; Hannan, M. T. Social dynamics. Modelsand methods. Orlando: Academic Press, 1984.

Tuma, N. B. Effects of labour market structure onjob-shift patterns. In: Heckman, J. J.; Singer, S.(eds) Longitudinal analysis of labour marketdata. Cambridge, MA: Cambridge UniversityPress, 1985.

Tuominen, E. The active years of the life course andpension provision in transition. Changes in theduration of employment during the life span inrecent decades. The Finnish example. The Year2000 International Research Conference onSocial Security. Helsinki, 25-27 September2000. Helsinki: ISSA – International Social Secu-rity Association, 2000. Available from Internet:

http://www.issa.int/pdf/helsinki2000/topic2/2Tuominen.PDF [cited 16.07.2003].

Unwin, L.; Wellington, J. Young people’s perspec-tives on education, training and employment.London: Kogan Page, 2001.

Uusitalo, R. Essays in economics of education.Helsinki: Department of Economics, 1999(Research reports, No 71).

Valenzuela, J.; Valenzuela, A. Modernization anddependency: alternative perspectives in thestudy of Latin American underdevelopment.Comparative Politics, 1978, Vol. 10, No 4,p. 535-557.

van de Werfhorst, H. G. Systems of educationalspecialization and labour market outcomes inNorway, Australia, and the Netherlands. Oxford:Nutfield College, 2002. Available from Internet:http://www.nuff.ox.ac.uk/users/werfhorst/ED-LM_NOAUSNL.pdf [cited 16.07.2003]

Veierød, M. B.; Thelle, D. S.; Laake, P. Diet and riskof cutaneous maligant melanoma: a prospec-tive study of 50 757 Norwegian men andwomen. International Journal of Cancer, 1997,Vol. 71, p. 600-604.

Voges, W. (ed.) Methoden der Biographie- undLebenslaufforschung. Opladen: West-deutscher Verlag, 1987.

Volkart, E. H. Social behavior and personality:contributions of W. I. Thomas to theory andsocial research. New York: Social ScienceResearch Council, 1951.

Wolbers, M. H. The effects of level of education onmobility between employment and unemploy-ment in the Netherlands. European SociologicalReview, 2000, Vol. 16, No 2, p. 185-200.

Yamaguchi, K. Event history analysis. In:Bickman, L.; Rog, D. J. (eds) Applied socialresearch methods series. London: Sage, 1991.

Zapf, W. (ed.) Theorien des sozialen Wandels.Königstein/Ts: Athenäum et al., 1975.

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 381

Rob A. Wilson

Institute for employment researchUniversity of WarwickUK – Coventry CV4 7ALE-mail: [email protected]

Geoff Briscoe

School of Science and the EnvironmentCoventry UniversityUK – Coventry CV1 5FBE-mail: [email protected]

Hiro Izushi

Centre for local economic development CoventryBusiness SchoolCoventry UniversityUK – Coventry CV1 5FBE-mail: [email protected]

Robert Huggins

Robert Huggins AssociatesUK – Cardiff CF10 5AFE-mail: [email protected]

Andy Green

John PrestonLars MalmbergThe Center for Research on the Wider Benefitsof LearningInstitute of EducationUniversity of LondonUK- London WC1A 0ALE-mail: [email protected]

Reinhard Hujer

Marco CaliendoChristopher ZeissJohann Wolfgang Goethe-Universität Frankfurtam MainD – 60325 Frankfurt am MainE-mail: [email protected]

Kenneth Walsh

David J ParsonsHOST Policy ResearchUK – Horsham, West Sussex RH12 1YS;E-mail: [email protected]

Bo Hansson

Ulf JohansonIPF – Uppsala Science ParkS - 75183 UppsalaE-mail: [email protected]

Karl-Heinz Leitner

Austrian Research Centers GmbH – ARCA - 2444 SeibersdorfE-mail: [email protected]

Maren Heise

Wolfgang MeyerCentre for Evaluation (CEval)Saarland UniversityD-66041 SaarbrückenE-mail: [email protected]

List of authors

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 382

Cedefop (European Centre for the Development of Vocational Training)

Impact of education and trainingThird report on vocational training research in Europe: background report

Pascaline Descy, Manfred Tessaring (eds)

Luxembourg: Office for Official Publications of the European Communities

2004 – 382 pp. – 21 x 29.7 cm

(Cedefop Reference series; 54 – ISSN 1608-7089)

ISBN 92-896-0277-5

Price (excluding VAT) in Luxembourg: EUR 25

No of publication: 3036 EN

01_2003_2162_1_txt_EN 28-02-2005 10:06 Pagina 383

01 2003 2162 1 txt EN 28-02-2005 10:05 Pagina 4 · 01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 8....

Documents

Transcript of 01 2003 2162 1 txt EN 28-02-2005 10:05 Pagina 4 · 01_2003_2162_1_txt_EN 28-02-2005 10:05 Pagina 8....