Sample 1 Dissertation

7/28/2019 Sample 1 Dissertation

1/54

THE IMPACT OF SCIENCE AND MATHS

EDUCATION ON ECONOMIC GROWTH

BY

****

0324841

8TH SEPTEMBER, 2004


2/54

ii

ACKNOWLEDGEMENTS

First and foremost, I wish to express my gratitude to my entire family for their constant

support, not only in the write-up of this dissertation, but also for the entire duration of my

MSc studies.


3/54

iii

TABLE OF CONTENTS

ACKNOWLEDGEMENTS ------------------------------------------------------------- ii

LIST OF TABLES------------------------------------------------------------------------- iv

SECTION 1 INTRODUCTION-------------------------------------------------- 1

SECTION 2-THEORETICAL FRAMEWORK AND LITERATURE REVIEW

2.1 THEORETICAL BACKGROUND-------------------------------------------------

2.2 EMPIRICAL EVIDENCE-----------------------------------------------------------

2.3 CONCLUSION------------------------------------------------------------------------

3

3

9

14

SECTION 3- METHODOLOGY-----------------------------------------------------------

3.1 MODEL SPECIFICATION----------------------------------------------------------

3.2 DATA-----------------------------------------------------------------------------------

3.3 EMPIRICAL TECHNIQUES AND METHODOLOGICAL ISSUES--------

3.4 LIMITATION -------------------------------------------------------------------------

15

15

20

23

25

SECTION 4 - RESULTS----------------------------------------------------------------------

4.1 PRESENTATION AND INTERPRETATION OF EMPIRICAL RESULTS--

4.2 THE BIG PICTURE---------------------------------------------------------------

26

26

34

SECTION 5 - CONCLUSION-------------------------------------------------------------- 40

REFERENCES----------------------------------------------------------------------------- 41

APPENDIX A COUNTRIES OF THE OECD---------------------------------------

APPENDIX B ECONOMETRIC TESTS----------------------------------------------

APPENDIX C OECD SCIENCE AND TECHNOLOGY INDICATORS-------

44

45

48


4/54

iv

LIST OF TABLES

TABLE 1 PROXIES FOR MODELS A and B 21

TABLE 2 CROSS COUNTRY GROWTH REGRESSIONS: IMPACT OF

PERFORMANCE IN SCIENCE AND MATHS OF 15 YEAR

OLDS ON ECONOMIC GROWTH- DEPENDENT VARIABLE

DY 2000-2003 27

TABLE 3 CROSS COUNTRY GROWTH ACCOUNTING

REGRESSIONS: IMPACT OF THE PROPORTION OF

SCIENCE GRADUATES ON ECONOMIC GROWTH-

DEPENDENT VARIABLE DY 1999-2003 30

TABLE 4 TIME SERIES REGRESSION RESULTS: IMPACT OF

SCIENCE GRADUATES ON ECONOMIC GROWTH OF THE

USA: 1980-2001. DEPENDENT VARIABLE: Dlny 33


5/54

1

SECTION 1

INTRODUCTION

The link between education and economic growth is an issue that has been investigated

for decades. This is because education is essential for the build-up of human capital,

which is one of the most essential pre-requisites for productivity and growth. Capital can

only be put to productive usage if it is combined with human capital.

There is a vast literature that examines the contribution of education to growth. However,

few studies exist on the benefits of maths and science education to the economy. The aim

of this paper is to fill this gap. In doing so, I present an econometric analysis of the

impact of science and maths education on productivity and growth. The rationale is that

the modern economy of today is driven by technology. Technological breakthroughs

through inventions have revolutionized the manufacturing, industrial and service sectors

thus enhancing economic growth. To sustain these economies, high technology must be

combined with the necessary skilled manpower to produce and operate them. This is

where science and maths education is necessary. There must be a sufficient number of

manpower skilled in these subjects to enable an economy to realize the potential of

technological change for bringing substantial and lasting benefits to it (Greenspan

2001). Not surprisingly, countries such as the U.S.A1

and the UK2

have expressed

concern on the dwindling number and performance of students in science and maths

especially at the secondary level. This is because if fewer people were skilled in science

1 See Greenspan (2001).2 See Green (2004)


6/54

2

and maths, then there could be negative implications on long-term growth, especially if

the economy is heavily influenced by the manufacturing sector.

Nelson and Phelps (1966) stressed that human capital should not be added simply as an

additional input in the production process. Rather, human capital affects growth

indirectly by giving rise to technological progress. This study aims to assess the impact of

science and maths education on economic growth through its effect on technology. To

explore this link, two models are employed. In the first model, cross sectional regressions

are applied along the lines of Benhabib and Spiegel (1994) using data from the OECD

countries. The second model, on the other hand, involves application of time series

regressions along the lines of Loening (2002) using data from the USA as a case study.

The results from both models show that education in science and maths has a positive and

significant impact on economic growth by affecting technological progress.

The main policy implications of these findings are thus self-evident: it is essential for

various countries to promote education of maths and science in order to sustain their

economies.

The remainder of this paper is organized as follows: section 2 presents the theoretical

background and empirical evidence on the growth effects of education. In section 3, the

methodology is explained while the results from my analysis are presented and discussed

in section 4. Finally, section 5 concludes.


7/54

3

SECTION 2

THEORETICAL BACKGROUND AND LITERATURE REVIEW.

This section presents both the theoretical background (2.1) underlying education and

economic growth, as well as empirical evidence (2.2). The theoretical background

focuses on two economic growth theories: the neo-classical growth theory and the new

growth theories. In section 2.2, I focus more on empirical evidence where the new growth

theories are applied since they are of greater relevance to this paper. I begin with those

studies that discuss the direct role of education on economic growth, and end with those

that bring the interplay of education and technology to the forefront. A summary in 2.3

concludes this chapter.

2.1 THEORETICAL BACKGROUND

NEOCLASSICAL THEORY

The basic framework for the neo-classical theory was provided by Robert Solow (1956).

In fact, the Solow model is the benchmark model of this theory. It begins with the

aggregate production function, where aggregate output (Y) is a function of aggregate

capital (K) and efficiency units of labor (AL).

Y = F (K, AL) (1)

A usually captures knowledge or the level of technological knowledge.


8/54

4

Two main assumptions of the production function govern the model: Constant Returns to

Scale (CRS) and Diminishing Returns to the accumulation of output (DR).

CRS enables the model to be worked in ratios. Dividing equation (1) through byAL so as

to work in ratios, we have

y = f (k) (2)

Wherey = Y/AL and refers to output per effective unit of labor, andk = K/AL denoting

capital per effective unit of labor.

A andL are assumed to grow at constant and exogenous rates ofg andn respectively but

capital evolves according to (3).

(3)

Thus, capital accumulation is the only factor that drives growth. Households are assumed

to save a constant proportion of their income Y and existing capital depreciates at a

constant rate of. The model is better analyzed per unit of effective labor. Capital per

effective labor grows according to (4):

(4)

K = sY - K

k = sy (n+g+)k(t)


9/54

5

Therefore, the excess of savings per effective labor over breakeven investment gives rise

to the growth rate of capital. The breakeven investment, (n+g+) k (t), is the amount of

investment that must be done just to keep kat its existing level. (Romer 2001, p. 15). To

keep capital at its existing level, enough capital must be produced to allow for the rate of

depreciation and the rate of growth in effective laborn+g. When the economy is in the

state where k grows at a constant rate because sy = (n+g+) k (t), then the economy is

said to be in a steady state. It has attained the long run equilibrium.

According to the Solow model, in the long run, kconverges to k*. This is because of the

crucial role played by DR. The growth rate of k is governed by equation 4. With DR,

output, and therefore savings will not rise as fast as breakeven investment. Therefore, in

the long run, savings is exactly offset by breakeven investment. This implies that savings

only has a temporary or short-run effect on growth but not a long-run effect. The only

factor that affects growth is technological progress because only this offsets the effect of

diminishing returns (Aghion and Howitt 1998, p.14-16). This, however, is where the

Solow model and the neo-classical theory as a whole come to a dead end. The only factor

that affects growth (technological progress) is exogenous and cannot be explained by the

model!

Finally, the Solow model predicts convergence. According to this hypothesis, poorer

economies grow at a faster rate than richer ones so that the poorer economies eventually

catch up with the richer ones. Therefore, convergence refers to the tendency of poor

economies to catch up with the worlds rich economies. (Mankiw 2003, p.220)


10/54

6

NEW GROWTH THEORIES

New growth theories3 arose to combat some of the deficiencies of the neo-classical

theory. In particular, technological progress is modeled endogenously instead of

assuming it to be exogenous. Hence, new growth theories (also referred to as

endogenous growth theories) give rise to endogenous growth models. Innovation,

knowledge accumulation and a more pronounced role for human capital are among the

key factors that are brought to the forefront. Human capital denotes all the skill, ability,

education and productive investments that are embodied in human beings. Many

endogenous growth models have been developed but I will only provide a brief analysis

of four of them- the AK model, the Schumpeterian model, the Lucas model and the

Nelson-Phelps model.

The AK model is the most basic model in the new growth theory. The production

function is:

Y = AK (5)

WhereA is a constant andKis an aggregate of all types of capital- physical, human and

knowledge. There are constant returns to scale to capital accumulation (not DR as in the

Solow model). One reason why we have constant returns rather than DR is that

knowledge generates externalities since it gives rise to ideas, which is transferable.

3 Most of the models in this section are adapted from Aghion and Howitt (1998, chapters 1, 2, 3 and 10)


11/54

7

The implication this has for growth are depicted clearly by Mankiw, et al. (1995, p.296),

which is replicated below.

Consider the capital accumulation equation

(6)

Combine equations 5 and6.

(7)

As long as sA > , then income grows indefinitely. Therefore, unlike the Solow model

where the rate of savings can only affect growth temporarily, here savings always leads

to growth.

Another approach to endogenous models is what Aghion and Howitt (1998) term the

Schumpeterian Approach. The key concept here is that of creative destruction

whereby new inventions render old technologies obsolete. Therefore, there is a negative

relationship between current and future research. The basic setup of the model is as

follows: Aggregate output is produced by an intermediate good. Labor is employed either

to produce intermediate goods (i.e. goods for manufacturing) or for research. Firms are

motivated to innovate so as to reap monopoly profits since any firm that succeeds in

innovation monopolizes the intermediate sector until the next innovative firm displaces it.

This captures the notion of creative destruction or replacement effect. In its most basic

setting, the unique steady state equilibrium occurs when the societys division of labor

between research and manufacturing remains constant over time. Aghion and Howitt

(1998) then describe the extended version of this basic setup. By incorporating multiple

K = sY - K

(Y/Y) = (K/K) = sA -


12/54

8

intermediate sectors and capital accumulation all in a Cobb-Douglas setting, they show

that both capital accumulation and innovation are relevant for growth.

The Lucas model and the Nelson-Phelps model are two models that are singled out by

Aghion and Howitt (1998) as models that are relevant for depicting the relationship

between education an economic growth. In the Lucas approach, a cobb Douglas setting

is used and human capital is simply an additional input in the production function. Here,

the growth rate of output depends on the growth rate of human capital. On the other hand,

in the Nelson-Phelps approach, the growth rate of output depends on the level of human

capital. Human capital has an indirect effect on growth by influencing the growth of

technology (which in turn promotes growth) through two channels: the ability of

individuals to innovate (domestic innovation) and to make use of technology from abroad

(i.e. technological diffusion). For Aghion and Howitt (1998), the two approaches have

very different policy implications. The policy implication of the Lucas approach is the

promotion of basic education while that of the latter approach is the promotion of more

advanced levels of education.

The empirical evidence that is derived from the two approaches above is presented

below, although more emphasis is placed on empirical analysis that is based on

endogenous models. Ultimately, the evidence shows that education is an important

determinant of economic growth.


13/54

9

2.2. EMPIRICAL EVIDENCE

The relevance of education for productivity and growth is an issue that has always caught

the attention of researchers, one way or another. With the neoclassical theories, it was

difficult to single out the impact of human capital itself on growth empirically, let alone

on education in particular. Nevertheless, it was possible to disaggregate the labor force

into labor input and labor quality. Labor quality takes into account factors such as the

educational attainment of workers and demographic factors. See Temple (2001) and

Sianese and Reenen (2002). As a result of the above, researchers were able to assess the

contribution of labor input and labor quality to growth.

With the development of endogenous models, it became possible to isolate the

contribution of human capital (driven by education) either directly on economic growth

or indirectly through its impact on technology, R&D, fertility and other measures. It also

made models more flexible to encompass whatever factors the researcher felt influenced

growth.

An example of such flexibility is Barro (1991) where cross country regressions were run

to explain the growth rate of real per capita GDP of 98 countries over the period 1960-

1985. Specifically, Barro regressed average annual growth rates of per capita GDP on

initial (1960) values of per capita GDP, primary and secondary school enrolment rates,

the ratio of physical investment to GDP, and other variables that served as controls such

as political instability and government expenditure. His analysis led him to two key

results. First, initial GDP levels had a negative, significant relationship with the average


14/54

10

annual growth rates of GDP (if initial levels of human capital were held constant), thus

supporting the convergence hypothesis. Secondly, a positive and significant relationship

existed between initial levels of school enrolment rates and the average annual growth

rates. In addition, school enrolment rates were positively related to investment ratios and

negatively related to fertility rates. Thus, his work showed that apart from the direct

effect education had on growth, it also had indirect effects by encouraging investment

and discouraging high fertility.

Another very influential study is by Mankiw, Romer and Weil (1992)4

. They applied the

Solow model, augmented with human capital to cross country data from 1960-1985.

Their proxy for human capital was secondary school enrolment ratios. They found the

impact of human capital on long run income per capita to be positive and significant.

Specifically, in the non-oil sub sample of their data, they estimated that a rise in the

average percentage of the working-age population in secondary school by 1% would lead

to an increase in long-run income per capita by about 0.66%.

Gemmel (1996) built on the above model by Mankiw, Romer and Weil by using a more

superior measure of human capital constructed by him using data from school enrolment

rates and the economically active population. He disaggregated his human capital

measure into primary, secondary and tertiary educational levels and found that primary

education was significant for growth in poor countries, secondary education was the key-

4 See Siamese and Reenen (2002, p.75)


15/54

11

determining factor for growth in intermediate countries, and tertiary education was the

most important factor for OECD countries.

Some researchers employed or constructed qualitative measures of education rather than

quantitative measures such as those used in the studies above. Examples include

Hanushek and Kimko (2000), and Hanushek and Kim (1995). In both papers, labor force

quality was estimated using international test scores in maths and science across

countries. Cross-country regressions were applied in order to explain growth rates

between 1960 and 1990. They found that a positive, significant relationship existed

between labor quality and growth rates of per capita income, and that this was robust to

adding quantitative measures and population growth. Moreover, the quality measures

where more important in explaining growth than the quantity measures as measured by

years of schooling.

Other studies such as Bassanini and Scarpetta (2001) focused on the impact of education

in OECD countries so as to analyze more homogenous datasets. Bassanini and Scarpetta

(2001) applied the Pooled Mean Group estimator (PMG) to estimate their growth

equations for 21 OECD countries over the period 1971-1998 using pooled cross country

time series data. They found that both the accumulation of physical capital and human

capital had very significant impacts on the growth of output per capita over the period.

Specifically, their results showed that an extra year of average schooling increased GDP

per capita by about 6%. Moreover, they identified several factors that affected the


16/54

12

standard of living in OECD countries including research and development and well

developed financial markets.

The last group of studies that I present is those that emphasize either the study of the

impact of science related education on growth or the impact of education on growth

through its link with science and / or technology.

Murphy, Shleifer and Vishny (1991) showed that the allocation of the talents in a country

is essential for growth. Their idea was that there were two kinds of activities that the

countrys talents get involved in: either entrepreneurial activities or rent seeking

activities. Entrepreneural activities are growth enhancing whereas rent-seeking activities

are detrimental to growth. They then presented a theoretical model based on Lucas (1978)

to prove their point. Finally, they tested their hypothesis empirically. They extended

Barro regressions to include ratio of college enrolments in law in 1970 to represent

allocation of talent to rent seeking activities; and the ratio of college enrolments in

engineering in 1970 to represent the allocation of talent to entrepreneurial activities for a

total sample of 91 countries. Their results were as follows: Engineering had a large direct

and indirect effect on the growth rates of real per capita GDP for the period 1970-1985.

The indirect effect worked through increased investment. On the other hand, law had a

negative, significant direct effect on growth. Thus, the educational choice of students

matters for growth.


17/54

13

Two studies that emphasized the role of education on economic growth through

technological diffusion are Engelbrecht (2003) and Benhabib and Spiegel (1994).

Benhabib and Spiegel (1994) based their analysis on the Nelson and Phelps (1966) model

which explicitly models the role of education in enhancing technological diffusion.

Benhabib and Spiegel used cross-country regressions to show that human capital affected

growth by influencing the rate of total factor productivity. First, they regressed a simple

human capital augmented Cobb Douglas function and found that human capital had an

insignificant impact on growth. They then constructed models that took into account the

ideas of Nelson and Phelps (1966) and Romer (1990a) and these yielded the following

results: human capital impacts economic growth by employing domestic technology for

innovation purposes and adopting technology from abroad so as to catch-up with the

leading country in technology but that human capital is more significant in its catch-up

role. When data was considered for OECD countries, however, the role of human capital

in domestic innovation came to the forefront. They also showed that human capital was

crucial in attracting physical capital, which is essential for growth.

Engelbrecht (2003) on the other hand combined the Nelson-Phelps (1966) model with the

Lucas (1988) model. The Lucas model specifies that human capital directly influences

growth. It uses the cobb-douglas function augmented with human capital. Engelbrecht

used datasets from Hanushek and Kimko (2000) and De la Fuentes and Domenech (2000)

with his hybrid model to account for economic growth in the OECD countries. He

concluded that human capital was important for growth both directly and indirectly by

encouraging technological diffusion.


18/54

14

In summary, the empirical evidence implies that education is an important ingredient for

economic growth.

2.3 CONCLUSION

This chapter has presented both the theoretical framework and the empirical evidence of

the impact of education on economic growth. The evidence shows that education has a

positive and significant impact on economic growth, either directly or indirectly through

different channels. The gap my paper aims to fill is to test whether science and maths

education specifically impacts economic growth through its effect on technology. This

could have important policy implications in the sense that although education in general

is necessary for growth, there might be need to emphasize (or at least not neglect)

specific subjects or fields in education.


19/54

15

SECTION 3

METHODOLOGY

This section discusses three broad issues: the model specification (3.1), data (3.2) and the

econometric techniques and important methodological issues (3.3). Finally, the

limitations to my analysis are briefly mentioned in 3.4

3.1 MODEL SPECIFICATION

I employ two models that are consistent with endogenous growth theory to carry out my

analysis. The aim of both models is the same: to examine the impact of maths and science

education on economic growth via its influence on total factor productivity (TFP).

Mankiw (2003, p.233) defines TFP as a measure of the current level of technology.

Specifying the influence of human capital in this manner rather than as an additional

input in the production function is in conformity with Nelson and Phelps (1966) who

argued that the latter specification of human capital was a misspecification of the role of

human capital.

MODEL A

This is adopted from Benhabib and Spiegel (1994). Specifically, I make use of his more

structural specification. It begins with a Cobb- Douglas production function.

Yt=At(Ht).Kt.Lt

(1)

Where Yt is per capita income, At is total factor productivity, which depends on human

capital (Ht), Kt is physical capital accumulation and Lt is the labor force.


20/54

16

Logs are taken in order to derive an expression for economic growth from time 0 to T.

[logYT-logY0]= [logAt(Ht) - logAt(Ht)] + .[logKT - logK0)

+.[logLT- logL0] + [logT- log0] (2)

In this study, I estimate economic growth using changes in income per person employed

rather than changes in income per capita because the former gave more robust estimates.

Both are valid indicators of economic growth according to Bassanini and Scarpetta

(2001).

The relationship between growth in TFP (technological progress (TP)) and human capital

for a representative country i is expressed in (3).

[logAt(Ht) - logAt(Ht)]i = c + g.Hi + m.Hi[(Ymax - Yi) / Yi] (3)

Where c is exogenous TP, Hi denotes the average level of human capital (if regressing

over long periods of time) or initial level of human capital (if regressing over short

periods of time)5

andg.Hi represents domestic innovation.Hi (Ymax-Yi /Yi) is an interactive

term between H and a productivity catch-up term, where Ymax is the leading countrys

initial income per person employed6, andmHi.[(Ymax - Yi) / Yi] represents the diffusion of

technology from abroad.

5 See Engelbrecht (2003), p. S436 Again, I make use of income per person employed rather than income per capita, which was used inBenhabib and Spiegel (1994).


21/54

17

Equation (3) can be rewritten as:

[logAt(Ht) - logAt(Ht)]i = c + (g-m)Hi + mHi[(Ymax/ Yi] (4)

Inserting (4) into (2) yields

[ logYT- logY0]= c + (g-m).Hi + mHi .[(Ymax/ Y) +

(logKT- logK0) + (logLT- logL0) + (logT-log0) (5)

Equation (5) is the model to be estimated. According to (5), the LEVEL of human capital

affects economic growth (by affecting the growth of TFP) in two ways:

Ability of a country to innovate (domestic innovation) as reflected by the

coefficient (g-m). I define domestic innovation as the creation of new ideas,

products and technologies.

Ability of a country to absorb technology from abroad (technological diffusion)

and so catch up with the technological leader as reflected by the coefficientmHi

I apply cross sectional regression analysis to equation (5) using data from the OECD7

countries. The proxies that I used for the above variables and the sources of my data are

reported in the next section (3.2). Three sets of regressions are run where I respectively

make use of international science literacy scores, international maths literacy scores and

the proportion of bachelors degrees awarded in science. The first two regressions test the

impact of competency in science and maths at the secondary level on economic growth

7 Appendix A lists the OECD countries


22/54

18

for the period 2000-2003 whereas the third regression measures the impact of the number

of science graduates8 on economic growth between 1999 and 2003. The empirical

techniques are discussed in greater detail in 3.3

Model A can be viewed as examining the impact of science and maths education on

economic growth in the very short run. This motivated Model B where I examine the

impact of science and maths education on economic growth for a longer period (1980-

2001) using USA as a case study. One reason for my choice of USA is to examine

whether the recent concern of the American government on the dwindling performance

and numbers of its students in the science (and maths) field is justifiable.

MODEL B

Given the concerns of unit roots in time series, I adopted the model of Loening (2002)

where an error correction mechanism was applied to a model similar to that of Benhabib

and Spiegel (2001).

Again, this model begins with the Cobb-Douglas production function:

Yt= A.KtLt

(1-) (6)

where A denotes TFP, Yt is aggregate output, Kt is physical stock of capital andLt is the

labor force. Dividing through by labor and taking logs

8 This comprises the natural sciences, mathematics, computer science and engineering.


23/54

19

ln yt= ln A + .ln kt+ ut (7)

whereytandktare output and capital per worker respectively.

An error correction mechanism (ECM) is applied to the above model. The ECM is

needed to combat the problem of unit roots that occurs in time series data. It captures the

behavior of variables in the short-run and the long run. The error correction form of

equation (7) is

lnyt= 1. lnkt- 2.(ln yt-1 - lnkt-1 lnA) (8)

Total factor productivity is specified as a function of education and foreign inputs. Here,

foreign inputs, rather than the interactive term in Benhabib and Spiegel (1994) captures

the notion of technological diffusion

lnA = c + 4 . lnht+ 5 . (IMt/It) + 6 . DUMMYt (9)

Where c is exogenous technological progress, h is the average level of human capital

(which I proxy with the average number of science graduates), (IMt/It) is the ratio of total

imports to total gross domestic investment andDUMMY are factors that affect the

country regressed. (For USA, oil price shocks that affected the economy where used as

dummies)


24/54

20

Combining equations (8) and(9) yields the one step error correction model. The model

must, however, be re-parametrized in order to apply OLS.

lnyt = c +1. lnkt + 2.ln yt-1 + 3 . lnkt-1 + 4 . lnht +5 . (IMt/It) + 6 . DUMMYt (10)

Equation 10 is the model that is estimated. My proxies for the above variables together

with my sources of data are discussed in 3.2, and the empirical techniques are discussed

in 3.3.

3.2 DATA

This study employed the use of secondary data from four main sources: the (OECD)

Corporate Data Environment (CDE), World Development Indicators 2004(WDI), the

Groningen Growth and Development Centre (GGDC) and the OECD Programme for

International Student Assessment (PISA). PISA is a body that conducts standardized

assessments of reading, mathematical and scientific literacy of fifteen year olds. The first

assessment was carried out in 2000. Other sources include the Digest for Education

Statistics, 2002 (DES) and the Extended Penn World Tables (EPWT). Details on the

proxies of the variables used in models A and B are presented in table 1.


25/54

21

TABLE 1

PROXIES FOR MODELS A and B

Model Variable Proxy Source Of Data

A Dependent Variable

(Y)

GDP per person

employed expressed in

constant 1999 US

dollars.

GGDC Total

Economy Database

(2004)

H9 Mean score in

scientific literacy of

15-year olds (2000),

Mean score in

mathematical literacy

of 15 year olds (2000)

and Percentage of

Bachelors degrees

awarded in science

(1999)

PISA 2000

DES 2002

K Stock of physical

capital

EPWT

L Total Labor force CDE

9 Data for all my proxies of H in model A are in Appendix C, table I


26/54

22

B Dependent variable

y

GDP per worker

(obtained by dividing

GDP at constant 1995

US$ by total labor

force)

WDI 2004

k Physical capital per

worker (obtained by

dividing physical

capital by total labor.

EPWT for physical

capital

WDI 2004 for labor

force

h Average human capital

(obtained by dividing

total science and

engineering graduates

by total labor force)

DES for graduates

WDI 2004 for labor

force

IM Imports of Goods and

Services (constant

1995 US$)

WDI 2004

I Gross Capital

Formation

(constant 1995 US$)

WDI 2004


27/54

23

3.3 EMPIRICAL TECHNIQUES AND METHODOLOGICAL ISSUES

The first model was estimated using Stata 7.0. Regressions were run with the Robust

command which makes use of Huber and Whites heteroscedasticity correction to obtain

robust variance (and therefore standard error) estimates. The following tests were carried

out:

Tests of significance and goodness of fit using the t-test, the F-test and the

multiple coefficient of determination (R2 ).

Tests for multicollinearity using the Variance Inflation factor (VIF) indicator.

Normality tests for each of the variables using the Shapiro-Wilk test.

o Bootstrapping procedures were used to estimate the standard errors of

those variables that were found not to be normal. Bootstrapping is a way

of estimating standard errors that is based, not on the assumption of

normality but on repeated sampling of the data. The greater the number of

times of repeated sampling, the more accurate the estimates.

The second model was estimated usingEviews 4.1. The following tests were carried out.

Tests of significance and goodness of fit using the t-test, the F-test and the

multiple coefficient of determination (R2 ).

Tests for unit roots using the Augmented Dickey fuller test.

Tests for structural stability using Chows forecast test.

Test for parameter stability using the Recursive estimation technique

(the CUSUM test).

Test for autocorrelation using the Durbin-Watson test


28/54

24

Residual tests using the correlogram and Q-statistics.

There are certain methodological issues that are associated with growth theory. This

paper was able to address some of these issues.

MEASUREMENT OF HUMAN CAPITAL

There is usually concern on how best to proxy for human capital. This issue does not

arise in this paper since it is concerned specifically with science and maths education.

ENDOGENEITY

There is always concern as to how to determine whether human capital influences growth

or vice-versa. To my knowledge, this issue has never really been fully solved. However,

one way of dealing with it, according to Sianesi and Reenen (2002), is to relate growth to

the initial value of an explanatory variable, and this is precisely what is done in the first

model.

PARAMETER HETEROGENEITY

Another problem occurs when data is obtained from both developed and developing

countries because of the impact of differences in the development of the countries. In this

paper, data is obtained from only OECD countries. There is therefore some degree of

homogeneity.

TIME SERIES CONCERNS

The ADF was used to test for unit roots. Results show the presence of unit roots

reflecting non-stationarity (Appendix B). The issue of non-stationarity was dealt with

using the error-correction mechanism.


29/54

25

3.4 LIMITATION

The two major limitations of this study are on data coverage and the length of my period

of analysis. It was considerably difficult to get data on all the OECD countries, especially

for the education measures. The most complete set of data on science and maths

performance was obtained for 27 out of 30 countries in 2000 from PISA, and the largest

set of data on science graduates was obtained for 24 countries in 1999.

This leads to the second limitation concerning the period of analysis. The time period for

both models is comparatively short. An attempt was made to extend the period of the first

model to 1995 using data from the Trends in International Maths and Science study

(TIMSS) but this had data for only 21 countries and thus led to less robust estimates. For

the second model, I began my analysis from 1980 since the computer revolution began

in 1981 due to the production of the IBM PCs. This computer revolution spurred rapid

technological change especially in the US and a rise in the demand for skilled workers, as

explained by Card and Dinardo (2002).


30/54

26

SECTION 4

RESULTS

This section is divided into two broad sections, 4.1 and 4.2. In 4.1, the results of the

regressions for both models are discussed and interpreted. Comments are also made on

the results of the econometric tests that were carried out. In 4.2, the results of this study

are discussed with respect to economic theory and present reality.

4.1 PRESENTATION AND INTERPRETATION OF EMPIRICAL RESULTS

MODEL A

The results, in general, show that the levels of science and maths education have a

positive and significant impact on economic growth by acting through the channel of

domestic innovation rather than by technological diffusion from abroad (TDA).

The results of the regressions for model 1 are presented in tables 2 and 3. Table 2 shows

the regression results of the impact of science and maths performance of 15 year olds on

economic growth for the period 2000-2003. H is proxied by science literacy scores in

models 1 and 2, and by mathematics literacy scores in models 3 and 4. The catch-up term

reflecting TDA is proxied byH(Ymax / Y).


31/54

27

Table 2

Cross-country growth accounting regressionsa: Impact of performance in science

and maths of 15 year olds on economic growth- dependent variable DY 2000 - 2003

Scienceb Mathsc

Model 1 Model 2 Model 3 Model 4

Constant -0.08059*

(0.03900)

-0.08883*

(0.04630)

-0.0496

(0.03461)

-0.06283

(0.04089)

H 0.00023***

(0.00007)

0.00028***

(0.00008)

0.00017**

(0.00007)

0.00023***

(0.00007)

H (Ymax / Y) -0.00006*

(0.00004)d

-0.00006

(0.00005)d

-0.00007*

(0.00004)d

-0.00007

(0.00005) d

DK 0.28352

(0.16670)

0.35130

(0.20801)

0.28120

(0.16556)

0.33820

(0.20789)

DL -0.56410 -0.63131**(0.29393)

-0.55473*

(0.26839)-0.60323

*

(0.29569)

Expsec -0.00402

(0.00321)

-0.00383

(0.00316)

Stuteach -0.00048

(0.00182)d

-0.00033*

(0.00162)d

Obs 27 23 27 23

R-squared 0.4994 0.5759 0.4676 0.5525

F-Stat 11.09 6.75 7.78 5.60

a Regressions with Ordinary least squares with robust variance estimates. Robust standard errors are in parenthesis.b H represents performance of 15 year olds in science,c H represents performance of 15 year olds in mathsdBootstrapping procedures were used to obtain standard errors*Statistically significant at 10% level, ** Statistically significant at 5% level , *** Statistically significant at 1% level


32/54


33/54

29

Table 3 presents the result of the regressions of the impact of the proportion of science

10graduates on economic growth for the period 1999-2003. Here, H denotes the

proportion of science graduates. The results mirror that of table 2. Hhas a positive and

significant impact on economic growth only when it operates through the channel of

domestic innovation (model 1). Although the statistical significance is 10%, it is very

nearly significant at the 5% level by a difference of 0.01 units. The catch-up term is,

however, negative and significant at the 5% level. This means that the effect of having

more science graduates leads to a significant reduction in TDA, and an increase in

domestic innovation. Thus, countries become less dependent on foreign technology and

become more productive and innovative. The results are robust to the addition of average

expected years of tertiary education (Expter- model 2). Interestingly, with Expteradded,

the significance of theHcoefficient rises to the 1% level. In model 3, the proportion of

graduates in the social sciences, business and law was used instead of science graduates.

The results show that graduates in this field fail to have a significant impact on growth

through either channel. These results, therefore, underscore that the subjects or fields that

the youths are educated in are very important to the economic growth process.

10 Science graduates includes the natural sciences, mathematics and computer science and engineering.


34/54

30

Table 3

Cross-country growth accounting regressionsa: Impact of the proportion of

science graduates on economic growth- dependent variable DY 1999 - 2003

Model 1 Model 2 Model 3c

Constant 0.01212

(0.01862)

-0.00444

(0.02108)

0.03381**

(0.01419)

H 0.00091*

(0.00044)

0.00101***

(0.00033)

-0.00015

(0.00078)

H (Ymax / Y) -0.00158**

(0.00069)

-0.00177**

(0.00066)

-0.00107

(0.00068)

DK 0.39824*

(0.20595)

0.41191*

(0.20742)

0.46943*

(0.22908)

DL -0.42957

(0.27625)b

-0.54637

(0.41252)b

-0.39013

(0.27683)b

Expter 0.00667

(0.00413)

Obs 25 24 25

R-squared 0.4436 0.4845 0.5004

F-Stat 9.04 15.88 10.92

aRegressions with Ordinary least squares with robust variance estimates. Robust standard errors are in parenthesis.

b Bootstrapping procedures were used to obtain standard errorsc. This model uses graduates from the social sciences, business and law as a proxy for H*Statistically significant at 10% level, ** Statistically significant at 5% level, *** Statistically significant at 1% level


35/54

31

Another feature indicated by the results is that the growth of capital and labor play little

significant roles in determining economic growth. In table 2, capital growth enters with a

positive but insignificant coefficient but enters with a 10% level of significance in table

3. On the other hand labor growth is insignificant in both tables. Surprisingly, it enters

with a negative coefficient. This, however, is not unique to this paper. Similar results are

obtained by Gemmel (1996).

Finally, from an econometric point of view, the goodness of fit of the models is fair,

about 0.50. Normality tests of the variables were conducted using the Shapiro-Wilk test.

Results (see Appendix B) indicated not all the variables were normally distributed.

Therefore, the standard errors of those variables that were not normally distributed at the

5% level were obtained by bootstrapping techniques with 500 repetitions. The VIF

indicator for multicollinearity was generally low, not exceeding 4.12 for any variable.

According to Gujarati(1995), VIF values greater than 10 for any regressor reflect

excessively high colinearity. Finally, heteroscedasticity is dealt with by making use of the

robust variance estimator of Stata 7.0. This uses the Huber and Whites

Heteroscedasticity correction techniques.

MODEL B

The results indicate that the level of science graduates have a positive and significant

impact on the economic growth in the USA over the period 1980-2001. Table 4 presents

the regression results. Model 1 makes use of the full sample of science graduates. The

average level of science graduates (h), have a positive and significant impact on US


36/54

32

growth at the 5% level. Results show that physical capital accumulation is also essential

for US growth. Foreign inputs (IM/I), which reflect TDA, have a positive but

insignificant impact on growth. Finally, the dummy variable has an expected negative

coefficient (at the 10% significance level) given that high oil prices dampen economic

activity in the US.

The goodness of fit of the model is moderate with an Adjusted R2

of about 0.53. The

Durbin-Watson statistic denotes an absence of autocorrelation. Other econometric tests11

were carried out for structural stability, parameter stability, serial correlation and unit

roots. The results show that there are both structural and parameter stability. As for the

unit roots test, as expected output and capital per worker have unit roots and are therefore

non-stationary (hence the use of the error correction mechanism in the model). The

human capital variable has no unit roots. The correlogram and Q statistics show that the

residuals exhibit white noise (i.e. they have zero mean and a constant variance) denoting

stationarity. Finally, a cointegration test between the average level12 of science graduates

(lnh) and economic growth rates (Dlny) was not conducted because both lnh andlny were

not integrated of the same order. While lny was integrated of order 1, lnh was integrated

of order 0. Thus, it was impossible to perform a formal econometric test of long-run

relationships between the two variables.

11 See Appendix B for the results of these tests.12 Note that this study studies the impact of the levels of science and maths (lnh) education rather than thegrowth of science and maths education (Dlnh) on economic growth.


37/54

33

The full sample of science graduates is split into graduates in mathematics (model 2), in

engineering (model 3) and in the physical and computer sciences (model 4). In all cases,

the coefficient ofln h is positive but the levels of significance vary across the models.

Table 4

Time series regression resultsa: Impact of science graduates on economic growth of

the USA: 1980-2001. Dependent Variable: DlnYt

Model 1 Model 2b Model 3c Model 4d

Constant 1.08221**

(0.49000)

1.84174***

(0.59143)

0.93219*

(0.49946)

0.84628

(0.54508)

Dlnkt 0.37773***

(0.11550)

0.34127**

(0.12063)

0.38704***

(0.11527)

0.45487*

(0.12445)

lnyt(-1) -0.48465*

(0.23790)

-0.58940**

(0.23678)

-0.46580*

(0.24003)

-0.55051**

(0.24717)

lnkt (-1) 0.40029*

(0.21679)

0.36265

(0.22364)

0.41250*

(0.21642)

0.49905**

(0.22821)

lnht 0.08767**

(0.03851)

0.06655*

(0.03122)

0.09022**

(0.03978)

0.03767*

(0.02106)

(IMt /It) 5.74E-13

(4.29E-13)

1.03E-12*

(5.56E-13)

4.56E-13

(4.14E13)

-3.64E-14*

(4.48E-13)

DUMMYe -0.00633*

(0.00350)

-0.00644*

(0.00356)

-0.00628*

(0.00350)

-0.006504

(0.00370)


38/54

34

Adjusted R2 0.55822 0.54298 0.55732 0.50726

F-statistic 5.21205 4.96026 5.19656 4.43152

S.E. of

Regression

0.00547 0.00557 0.00548 0.00578

Durbin-Watson 2.46020 2.36131 2.46867 2.39710

N 21 21 21 21

a Regressions with Ordinary least squares with robust variance estimates. Robust standard errors are in parenthesis.b Model uses proportion of graduates in Mathsc. Model uses proportion of graduates in engineeringd. Model uses proportion of graduates in the physical sciences and computer sciencese. Dummy for oil price shocks for the years 1980, 1990-1991, 1999-2001. Labonte and Makinen (2002)*Statistically significant at 10% level, ** Statistically significant at 5% level, *** Statistically significant at 1% level

4.2 THE BIG PICTURE

The results from models A and B suggest that the levels of science and maths education

have a positive and significant impact on economic growth at least in the short run by

increasing TP through the channel of domestic innovation rather than TDA. Whereas

model A indicates that TDA is reduced, model B shows that it is insignificant. Benhabib

and Spiegel (1994) also favored the domestic innovation channel rather than the TDA

channel for the wealthier third of their sample. The differing factor between their results

and mine is that while the coefficient of the TDA in theirs is positive, mine is negative.

The results of model B, with respect to signs, are in agreement with Loening (2002) .

In summary, the transmission mechanism is as follows: the levels of science and maths

education directly influence the ability of nations to innovate. This leads to technological

progress (TP), i.e. growth in TFP, which impacts economic growth.


39/54

35

Given the above results, two important questions emerge which are addressed below:Do

the results conform to economic theory?, and are they of present day relevance? Do they

actually apply to the OECD countries?

CONFORMITY WITH ECONOMIC THEORY

The model and results of this study conform to the endogenous growth theory (EGT) in

general and to the contribution of Nelson and Phelps (1966) in particular. Some of the

basic arguments of EGT are as follows: First, TP is determined endogenously. Second,

human capital is a crucial determinant of growth. Third, ideas, knowledge accumulation

and innovation all play central roles in economic growth and finally, human capital

generates externalities (which prevents diminishing returns to capital).

Comparing the results to the above arguments of EGT, the following can be deduced:

TP is endogenously determined by levels of science and maths education

The levels of education in these fields make the role of human capital extremely

relevant to growth as shown in tables 2 to 4. In particular, table 3 shows that

graduates in an alternative field are much less important.

Knowledge in science and maths stimulates domestic innovation that propels

economic growth.

Domestic innovation produces externalities since it is impossible for any

individual or firm to capture all the benefits of innovation. Education in science

and maths not only play a huge role in the ability of individuals to create new

technology, it also equips non-inventors with the know-how to make productive


40/54

36

usage of technology. In fact, according to Nelson and Phelps (1966, p. 75), if

innovation produces externalities because they show the way to imitators, then

(science and maths)13 education - by its stimulation of innovation - yields

externalities.

As mentioned above, the results of this study concur with Nelson and Phelps (1966).

Their idea, expressed by Aghion and Howitt (1998, p.338), is that education increases

the capacity of individuals to innovateand to adapt to new technology by speeding up

technological diffusion throughout the economy.

The results favor the fact that education in science and maths increases domestic

innovation. These benefits of domestic innovation must be diffused throughout the

economy because innovation generates externalities. The results reject diffusion of

foreign technology as a channel through which science and maths influences economic

growth in OECD countries. It should be noted that Nelson and Phelps stressed new

technology (which is created by domestic innovation), not foreign technology.

Aghion and Howitt (1998) opine that the Nelson and Phelps (1966) model favor higher

levels of education as relevant for innovation. This finds some empirical support from

Engelbrecht (2003) who attain positive and significant coefficients when human capital is

proxied by average years of higher schooling. Table 3, however, shows that the impact of

13 Bolded text is my addition


41/54

37

the proportion of science graduates on growth through the domestic innovation channel is

robust to the inclusion of average expected years of tertiary education.

Finally, model B, apart from concurring with Nelson and Phelps (1966), also agrees with

the Schumpeterian endogenous model. According to Aghion and Howitt (1998), the most

important finding of this model is that both innovation and capital accumulation are

important for economic growth. This is supported by the fact that both the coefficients of

Dlnkt and lnht have positive and statistically significant coefficients (table 4) especially

in model 1 where the full sample of science graduates are used.

ECONOMIC REALITY

Do the results conform to reality? This will be discussed with respect to the tables and

figures in appendix C.

Table I presents average maths and science scores across the countries of the OECD. It

also indicates the percentage of BSc degrees awarded in Science across these countries. It

shows that countries with the highest maths and science education include: Japan, Korea,

Finland, Germany and Sweden whereas those at the bottom of the scale include Greece,

Mexico, Luxemburg and Portugal.

Fig.1 indicates countries with the highest number of patents in 1999. Patents are very

useful indicators of domestic innovation. From fig.1, we see that the US and Japan have


42/54

38

the highest shares of patents. In general, the OECD countries hold the dominant share of

patents.

Table II gives ratings of foreign direct investment and technology transfer (TT) for some

OECD countries. TT measures the extent to which foreign direct investment is an

important source of new technology. Countries are rated on a scale of 1 to 7 where

1 reflects FDI as a relatively unimportant source of new technology and 7reflects

FDI as the most important source of new technology. Table II shows that the rates of TT

reduce for most of the countries. This is in stark contrast to the increased rates of patents.

Table III indicates changes in the technology balance of payments in the OECD between

1990 and 2000 and fig 2 plots the data from table III. Both fig.2 and table III show that

the net technology balance of payments as a proportion of GDP rose sharply in the

OECD as whole. According to the OECD STI Scoreboard (2003, p.128), in the 1990s,

the OECD area maintained its position as the net exporter of technology compared with

the rest of the world. Therefore, the tables and figures discussed above in general reveal

that domestic innovation plays a more prominent role than TT.

Apart from the tabular evidence, reports by the US government, the Uk government and

the OECD support the crucial importance of science and maths to growth. The American

government has stressed the need to increase the scientifically educated domestic

workforce because of the greater demand for these kinds of workers given the high tech


43/54

39

economy of the US.14 Education in science and maths is stressed as a factor necessary for

the country to reap the full benefits of technology. The results for model B show how

significant education in science and maths is to economic growth in the US. Therefore,

the desire of the US government to increase the skill of students in this field is well

founded.

The realization of the importance of science and maths education in the US is echoed by

the UK where there is increasing concern of the recent dwindling number of students that

study the sciences due to the low enrolment rates at the graduate level, according to the

article by Green (2004) in the Financial Times. Likewise, comments by the OECD

Committee for Scientific and Policy Meeting (2004, p.5), are worthy of note: Ministers

expressed concern that the recent decline in the number of science and engineering

graduates could hamper the long term growth prospects for OECD countries. The same

report tied education in science with innovation, and proposed ways of increasing the

supply of personnel trained in science, and encourage science-based innovation within

the OECD.

14 See Greenspan (2003)


44/54

40

CHAPTER FIVE

CONCLUSION

This paper explored the impact of science and maths education on economic growth in

the OECD countries using two models based on the endogenous growth theory. The first

model, adapted from Benhabib and Spiegel (1994) used cross-sectional regressions to

analyse the impact of science and maths education across the OECD countries. The

second model, adapted from Loening (2002) examined the impact of science education

on the economic growth of the USA over the period 1980-2001.

The fundamental result of this study is that the levels of science and maths education

have a positive and significant impact on economic growth at least in the short-run, and

that the channel through which this works is by enhancing domestic innovation. This

stimulates technological progress which gives rise to growth.

The key policy implication of this study is that it is crucial to enhance education in these

fields in order to sustain economic growth in the OECD.

Finally, I would like to stress that the results should be interpreted by considering the

OECD countries as a unit rather than as separate entities. Countries in the OECD have

different high-tech capabilities. A useful follow-up to this study would be to test whether

the basic result holds for all the countries in the OECD regardless of the structure of the

economy or technological potential.


45/54


46/54

42

Hanushek, E. A. and D. Kim (1995). "Schooling, Labour Force Quality and EconomicGrowth." National Bureau of Economic Research Working Paper5399.

Hanushek, E. A. and D. Kimko (2000). "Schooling, Labour Force Quality and theGrowth of Nations." American Economic Review 90: 1184-1208.

Labonte, M. and G. Makinen (2002). Energy Independence: Would it Free the UnitedStates From Oil Price Shocks?, CRS Report for Congress.

Loening, L. J. (2002). The Impact of Education on Economic Growth in Guatemala: ATime-Series Analysis Applying an Error-Correction Methodology. Discussion Papers,Ibero-American Institute for Economic Research.

Lucas, R. E. (1978). "On the Size Distribution of Business Firms." Rand Journal ofEconomics IX: 508-23.

Lucas, R. E. (1988). "On the Mechanics of Economic Development." Journal ofMonetary Economics 22(1): 3-42.

Mankiw, G., N. (2003). Macroeconomics. New York, Worth Publishers.

Mankiw, G., N., E. Phelps, et al. (1995). "The Growth of Nations." Brookings Papers onEconomic Activity(1): 275-326.

Mankiw, G., N., D. Romer, et al. (1992). "A Contribution to the Empirics of EconomicGrowth." Quarterly Journal of Economics May(407-438).

Murphy, Schleifer, et al. (1991). "The Allocation of Talent: Implications for Growth."Quarterly Journal of Economics: 407-437.

Nelson, R. and E. Phelps (1966). "Investment in Humans, Technological Diffusion andEconomic Growth." American Economic Review 61: 69-75.

OECD Corporate Data Environment. http://www1.oecd.org/scripts/cde/default.asp.Retrieved August 2004.

OECD (2001). "Science, Technology and Industry Scoreboard."

OECD (2003). "Compendium of Patent Statistics."

OECD (2003). Literacy Skills for the World of Tomorrow- Further Results from PISA2000. http://www.pisa.oecd.org/Publicatn/Literacy.htm. Retrieved August, 2004.

OECD (2004). Science, Technology and Innovation for the 21st Century. Meeting for theCommittee for Scientific and Technological Policy at the Ministerial Level, 29-30January 2004- Final Communique.


47/54

43

Romer, D. (1990a). "Endogenous technological change." Journal of Political Economy

98: S71-S102.

Romer, D. (2001). Advanced Macroeconomics. New York, McGraw-Hill HigherEducation.

Sianesi, B. and J. V. Reenen (2002). The Returns to Education: A Review of theEmpirical Macroeconomic Literature, The Institute for Fiscal Studies.

Solow, R. M. (1956). "A Contribution to the Theory of Economic Growth." QuarterlyJournal of Economics 70 (February): 65-94.

Temple, J. (2001). "Growth Effects of Education and Social Capital in the OECDCountries." OECD Economic Studies(No. 33): 57-101.


48/54

44

APPENDIX A

COUNTRIES OF THE ORGANISATION FOR ECONOMIC CO-OPERATION

AND DEVELOPMENT (OECD)

Australia Austria Belgium Canada

Czech Republic Denmark Finland France

Germany Greece Hungary Iceland

Ireland Italy Japan Korea

Luxembourg Mexico The Netherlands New Zealand

Norway Poland Portugal Slovak Republic

Spain Sweden Switzerland Turkey

United Kingdom United States.

Source

http://www.oecd.org/document/58/0,2340,en_2649_201185_1889402_1_1_1_1,00.html


49/54

45

APPENDIX B: ECONOMETRIC TESTS

MODEL A : NORMALITY TESTS


50/54

46

MODEL B

TESTS FOR UNIT ROOTS

Variables ADF TEST STATISTICS RESULT

lny 1.3 Non-stationary

lnk -1.4 Non-stationary

lnh -5.5*** Stationary

IM/I 1.9 Non-stationary

dlny -3.9** Stationary

dlnk -2.7* Stationary

*,**,*** Rejects the null hypothesis of unit roots at 10%, 5% and 1% respectively, assuming a constant in the test

equation.

TEST FOR PARAMETER STABILITY

-10

-5

0

5

10

91 92 93 94 95 96 97 98 99 00 01

CUSUM 5% Significance


51/54

47

TEST FOR STRUCTURAL STABILITY

Chow Forecast Test: Forecast from 1991 to 2001F-statistic 2.594446 Probability 0.234316Log likelihood ratio 49.40481 Probability 0.000001

CORRELOGRAM AND Q STATISTICS OF THE RESIDUAL

Date: 09/03/04 Time: 16:39

Sample: 1981 2001Included observations: 21

Autocorrelation Partial Correlation AC PAC Q-Stat Prob

.**| . | .**| . | 1 -0.301 -0.301 2.1823 0.140

.**| . | ***| . | 2 -0.301 -0.431 4.4903 0.106. | . | ***| . | 3 0.013 -0.332 4.4951 0.213. |* . | . *| . | 4 0.156 -0.165 5.1867 0.269. |* . | . | . | 5 0.109 0.059 5.5446 0.353. *| . | . |* . | 6 -0.096 0.087 5.8443 0.441. *| . | . |* . | 7 -0.079 0.072 6.0588 0.533. *| . | .**| . | 8 -0.140 -0.233 6.7909 0.559. |**. | . *| . | 9 0.216 -0.077 8.6665 0.469

. *| . | .**| . | 10 -0.090 -0.300 9.0253 0.530. | . | . *| . | 11 0.027 -0.094 9.0596 0.616

. *| . | .**| . | 12 -0.132 -0.253 10.002 0.616


52/54

48

APPENDIX C

OECD SCIENCE AND TECHNOLOGY INDICATORS

TABLE I: SCIENCE AND MATHS EDUCATION INDICATORS

Science literacy Maths literacy % of science graduates

Austral ia 528 533 17.9Austria 519 515 38.4Belgium 496 520 17.6

Canada 529 533 23Czech Republic 511 498 21.3Denmark 481 514Finland 538 536 31.1France 500 517 21Germany 487 490 38.9Greece 461 447

Hungary 496 488 13.5Iceland 496 514 30.7Ireland 513 503 24.8Italy 478 457 13.1Japan 550 557 42.4Korea 552 547 48.3Luxembourg 443 446

Mexico 422 387 17.6Netherlands 17.6New Zealand 528 537 24.4Norway 500 499 21Poland 483 470 3.1Portugal 459 454

Slovak Rep 40.1Spain 491 476 40.1Sweden 512 510 41.5Switzerland 496 529 41.5Turkey 29.8

UK 532 529 21.8USA 499 493 13.7

OECD Average 500 500

Notes: Science and literacy scores refer to average scores of 15 year olds in the PISA assessment (2000).

Source: OECD (2003)

Science graduates refer to the proportion of BSc degrees awarded in science (1999). Source: DES 2002

Bolded figures refer to scores above the OECD average whereas scores in italics are those lowest among the countries


53/54

49

TABLE II: FOREIGN DIRECT INVESTMENT AND TECHNOLOGY

TRANSFER

99 2002/2003 growth rate

Austral ia 5.24 5.1 -2.671755725Austria 5.06 4.8 -5.138339921Belgium 5.33 4.7 -11.81988743Canada 5.48 5.3 -3.284671533Czech Republic 5.46 5.9 8.058608059Denmark 4.81 4.9 1.871101871France 5.04 4.9 -2.777777778

Germany 4.7Greece 4.87 -100Hungary 5.72 6.1 6.643356643Ireland 6.28 6.2 -1.27388535Korea, Dem. Rep. 5.22 4.7 -9.961685824Mexico 5.66 5.1 -9.893992933New Zealand 5.32 4.8 -9.77443609Norway 5.1 4.8 -5.882352941Portugal 5.11 5.3 3.718199609Slovak Republic 5.12 5 -2.34375United Kingdom 5.33 4.7 -11.81988743

Source of 99 and 2002/2003 figures: The Global Competitiveness Report (1999, 2000-2003)

FIG.1 ; LEADING PATENTING COUNTRIES (1999)

0

5

10

15

20

25

30

35

Unite

dStates

Europe

anUnio

n

Japa

n

Germ

any

Fran

ce

Unite

dKing

dom

Swed

en

Netherlan

ds

Switz

erlan

dItaly

Cana

da

Finla

nd

Belgi

umIsr

ael

Korea

Austr

alia

%

0

5

10

15

20

25

30

35%

1991

Note:

Graph shows the share of patents in the triadic patent families (European Patent Office (EPO), US Trademark and

Patent Office (USPTO) and Japanese Patent Office (JPO). Source: OECD (2003, p.13)


54/54

50

TABLE III: CHANGES IN THE BALANCE OF PAYMENTS AS A

PERCENTAGE OF GDP, 1990 AND 2000.

1990 2000

USA 0.23 0.24

JAPAN -0.01 0.12

EU -0.04 -0.06

OECD 0.06 0.09

Source: OECD (2001)

FIG II: CHANGES IN THE BALANCE OF PAYMENTS AS A PERCENTAGE

OF GDP, 1990 AND 2000.

-0.10

-0.05

0.00

0.05

0.10

0.15

0.20

0.25

0.30

United States J apan EU (2) OECD (2,3)

%

1990 2000

Source: OECD (2001)

Sample 1 Dissertation

Documents

Transcript of Sample 1 Dissertation