SOCIO-ECONOMIC RESEARCH METHODS IN FORESTRY: A...

SOCIO-ECONOMIC RESEARCH METHODS INFORESTRY: A TRAINING MANUAL

Proceedings of an International Training Workshop held in theCollege of Forestry at Leyte State University, Visca, Baybay,

The Philippines, over the period 4-10 February, 2002

Edited bySteve Harrison, John Herbohn,

Ed Mangaoang and Jerry Vanclay

Cooperative Research Centre forTropical Rainforest Ecology andManagement.

ISBN 0 86443 691 2

This work is copyright. The Copyright Act1968 permits fair dealing for study,research, news reporting, criticism orreview. Selected passages, tables ordiagrams may be reproduced for suchpurposes provided acknowledgment ofthe source is included. Major extracts ofthe entire document may not bereproduced by any process withoutwritten permission of the Chief ExecutiveOfficer, CRC for Tropical RainforestEcology and Management.

Published by the Cooperative ResearchCentre for Tropical Rainforest Ecologyand Management. Further copies may berequested from the CooperativeResearch Centre for Tropical RainforestEcology and Management, James CookUniversity, PO Box 6811 Cairns, QLD,Australia 4870.

This publication should be cited as:Harrison, S., Herbohn, J., Mangaoang, E.and Vanclay, J (2002) Socio-economicResearch Methods in Forestry: ATraining Manual. Cooperative ResearchCentre for Tropical Rainforest Ecologyand Management. Rainforest CRC,Cairns.

December 2002

ACKNOWLEDGEMENTS

We would like to thank all those people in the College of Forestry at Leyte State Universitywho contributed to the arrangements and smooth running of the training workshop. Ourthanks also go to the other contributors in this Training Manual, and to the workshopparticipants for their insightful feedback on many issues.

We would also like to thank the residents of Conalum barangay, Inopocan, for theirhospitality and patience on the visit associated with this workshop, and on earlier visits, inproviding us with a practical introduction to smallholder forestry.

We wish to acknowledge the financial support from Australian Centre for InternationalAgricultural Research (ACIAR) which allowed us to hold the workshop and to sponsor theattendance of Australian and Filipino participants. The assistance of the RainforestCooperative Research Centre in the production of this manual is gratefully acknowledged.

Steve Harrison, John Herbohn, Ed Mangaoang and Jerry Vanclay

CONTRIBUTORS

Mr Nicholas Emtage, PhD candidate, School of Natural and Rural SystemsManagement, The University of Queensland, Qld. 4072, Australia.

Dr Steve Harrison, Associate Professor of Economics, School of Economics, TheUniversity of Queeensland, Qld. 4072, Australia.

Dr John Herbohn, Senior Lecturer in Natural Systems Management, School of Naturaland Rural Systems Management, The University of Queeensland, Gatton 4343,Australia.

Dr Eduardo O. Mangaoang, Associate Professor and Head, College of Forestry, LeyteState University, Visca, Baybay, Leyte, The Philippines.

Dr Jungho Suh, Postdoctoral Research Fellow, School of Economics and School ofNatural and Rural Systems Management, The University of Queeensland, Qld. 4072,Australia.

Professor Jerry Vanclay, Professor of Sustainable Forestry, School of EnvironmentalScience and Management, Southern Cross University, Lismore, NSW 2480, Australia.

TABLE OF CONTENTS

1 A Forester’s Perspective of Socio-economic Information Requirements forForestry in LeyteE. Mangaoang

1

2 Socio-economic Research Techniques in Tropical ForestryS. Harrison

5

3 Data Collection Methods in Forestry Socio-economic ResearchJ. Herbohn

15

4 Basic Computer Skills and Statistical Methods for Analysis of Survey DataN. Emtage, J. Herbohn and S. Harrison

29

5 Assessing the Financial Viability of Farm Forestry: A Tutorial ExerciseN. Emtage

45

6 An Introduction to the Statistical Package for the Social SciencesN. Emtage and S. Duthy

53

7 A Personal View of Statistical Packages for Linear RegressionJ. Vanclay

65

8 Simulation Philosophy and MethodsS. Harrison

73

9 An Introduction to SimileJ. Vanclay

85

10 Simile Revisited: The One-minute ModellerJ. Vanclay

95

11 Introduction to Discounted Cash Flow Analysis and Financial Functions inExcelJ. Herbohn and S. Harrison

109

12 Financial Models of Farm and Community Forestry Production SystemsJ. Herbohn, N. Emtage and S. Harrison

119

13 Estimating and Integrating Non-market Benefits of Forests into ProjectAppraisalJ. Suh

133

14 Estimation of Non-market Forest Benefits Using Choice ModellingJ. Suh

149

15 Cost-Benefit Analysis in Forestry ResearchS. Harrison

165

16 Using Financial Analysis Techniques in Forestry ResearchJ. Herbohn

177

17 Forestry Applications of Linear ProgrammingS. Harrison

191

18 Team Building, Grant Seeking and Project AdministrationS. Harrison and J. Herbohn

211

19 Report Writing and Publication StrategyS. Harrison and J. Herbohn

219

20 Experiences Gained and Lessons Learnt from the Training WorkshopS. Harrison, J. Herbohn and E. Mangaoang

235

PARTICIPANTS AT THE LEYTE STATE UNIVERSITY WORKSHOPON SOCIO-ECONOMIC RESEARCH METHODS IN FORESTRY

NAME POSITION/DESIGNATION AGENCY1. Dr Paciencia P. Milan President LSU2. Dr Jose L. Bacusmo Director of Research ODRD, LSU3. Dr Belita A. Vega Director of Center for Social

ResearchCSR, LSU

4. Dr Salome B. Bulay-og Assoc. Professor, Ag. Econ.Dept.

DAEAB, LSU

5. Dr Remberto A. Patindol Assoc. Professor, Stat. Dept. DAEAM, LSU6. Dr Pastor Garcia Assoc. Professor FARMI, LSU7. Dr Eduardo O. Mangaoang Assoc. Professor DOF, LSU8. For. Nestor O. Gregorio Instuctor DOF, LSU9. For. Anatolio N. Polinar Instructor DOF, LSU10. For. Dennis P. Peque Instructor DOF, LSU11. For. Mylene D. Napiza Instructor DOF, LSU12. Prof. Arturo E. Pasa Instructor DOF, LSU13. Dr Ernesto C. Bumatay Professor DOF, LSU DENR Region 814. For. Emma M. Germano Researcher DENR-815. For Edilberto E. Nasayao Researcher DENR-8 PCARRD16. For. Reynaldo Dimla Researcher PCARRD Local Government

Unit/Provincial Government17. Rev. Manuel E. Constatino

Jr.CWMP In-charge LGU, Isabel

18. Rodolph Hernandez Environment Officer ENRO, Leyte (PASAR) NGO/Industry19. Louie O. Pable Executive Director LGM, Isabel20. Engr Amos S Calda Community Relations Officer LIDE, Isabel Research & Development

Organization21. ICRAF Staff (3) Research Officers22. LIP-GTZ Staff (1) Research Officer UQ, Australia23. Dr John Herbohn Senior Lecturer UQ, Australia24. Dr Steve Harrison Associate Professor UQ, Australia25. Prof. Jerry Vanclay Professor Southern Cross

University26. Nicholas Emtage PhD Student UQ, Australia27. Dr Jungho Suh Post-doctoral Research Fellow UQ, Australia28. Darwin Posas MSF Student CF, LSU29. Jusifer Mendoza MSF Student CF, LSU30. Edwin Cedamon Sc. Res. Asst. ACIAR-CF

PREFACE

The training workshop on socio-economic research methods in forestry on which thesepapers are based was a consequence of a long-held ambition to offer such a trainingpackage, arising from research activities in the socio-economics program of the RainforestCooperative Research Centre. This ambition came to fruition at Leyte State University(LSU), Visca, Baybay, The Philippines, over the period 4-10 February, 2002.

The workshop – a hectic but enjoyable experience – was the first run of such a program byour research group within the ACIAR Project ASEM/2000/088 – Redevelopment of a TimberIndustry Following Extensive Land Clearing. It represents part of the capacity buildingactivity planned for this project. Other objectives of the workshop were: to allow an exchangeof views on the forestry research mindset of the Australian and Filipino project members; toadvance the planning of and participation in the ACIAR project; and to document the variousresearch methods we have employed in this and other forestry research projects.

It has been a rewarding experience to be a part of an enthusiastic and active researchgroup, and fortunately several of the members of this group were able to contribute at thetraining workshop. Hence a wide range of skills and interests are represented in the paperscontained here. It is our hope that this training manual will be of assistance to others inplanning training workshops, and to researchers in various disciplines who wish to gain skillsin the particular research areas addressed here.

Steve Harrison

1

1. A Forester’s Perspective of the Socio-economicInformation Requirements for Forestry in Leyte

Eduardo O. Mangaoang

In this module, a number of observations are made about the need for socio-economicresearch in forestry, on the basis of experience in the College of Forestry at Leyte StateUniversity (LSU). These comments draw particularly on teaching and outreach activities ofthe College. They are also influenced by collaboration with international research agencies,and with other research units within LSU. The module is designed to provide a context forthe subsequent presentations on research techniques.

1. BACKGROUND AND RATIONALE

In the recent years, the focus on forestmanagement and conservation hassignificantly shifted from the highly technicalcommercial forestry to a more people-oriented social forestry orientation. Goneare the days when forestry is looked uponas solely management and utilization oftrees by large-scale timber-product-orientedlogging corporations to meet demands forwood and wood-based products. The morerecent scenario is a paradigm shift in theforestry sector to small-scale, multiple-product-based, people-oriented andcommunity-based sustainable forestmanagement. This shift recognizes thatsustainable forest management andconservation can only be achieved throughthe commitment and active participation ofthe local people in communities.

Experiences in Leyte Island, Philippines,reveal that local people in communities –commonly called smallholders – haveimportant knowledge and skills to offer forsustainable forest management andconservation. Studies conducted by theViSCA-Darwin Project of the College ofForestry (CF) at LSU (Lawrence et al.1999), and the collaborative researchundertaking of CF, LSU and theInternational Centre for Research inAgroforestry (ICRAF), Visayas, (CF andICRAF 2002) have provided clear evidenceof the high level of knowledge and skills andactive involvement of smallholder farmers inforest management and conservation.Various development activities on people-oriented community-based forestry hadbeen conducted by government and non-

government organizations, as well aspeople’s organizations and privateinstitutions. Various Local GovernmentUnits (LGUs), for example, have initiatedactivities for the rehabilitation and protectionof forest areas in their respective localities.The rise of effective people’s organizationsin communities with concerns for forestryand the environment is evident.

Despite the apparent shift in focus andsupport for people-oriented community-based forestry, until quite recently not muchhad been achieved in view of theexpectations for sustainable forestmanagement and conservation. This lack ofprogress stimulated researchers in forestryand related fields to identify the majorforestry impediments and measures bywhich they may be overcome. At LSU,much concern had been given to the socio-economic research in forestry. The initiativeundertaken by the ViSCA-Darwin Project onlocal knowledge and biodiversity con-servation and the collaborative undertakingof CF, LSU and ICRAF on localmanagement practices of indigenous trees,are testimonies to the importance of socio-economic research for this present-dayforestry situation. Careful examination ofthe socio-economic factors is a majorrequisite in arriving at appropriate answersto the failures and shortcomings of people-oriented community-based forestry in Leyte.

2. SUGGESTED SOCIO-ECONOMICINFORMATION REQUIREMENTS

Lessons learned from research conductedon smallholder community forestry andexperiences acquired from various

Socio-economic Research Methods in Forestry2

community-based forestry projects indicatethe need to examine carefully the key socio-economic factors that may haveencouraged or discouraged local people incommunities to engage in tree farming andplantation development. For example,observat ions in selected uplandcommunities in Leyte reveal that smallfarmers would engage in tree farming ifseedlings are conveniently available, andthere is a clear market demand andreasonable price for the expected woodproduct. This section identifies the kinds ofsocio-economic information that arenecessary in determining and analyzingfactors that may make a major contributionto the promotion of farm and communityforestry in Leyte.

Adoption and suitability of trees forsmallholder farmers’ needs

Farmers grow trees on their land and in theprocess are selective in the kind of trees tocultivate or maintain. Some are enthusiasticto raise fast-growing exotic trees forimmediate cash income and homeconsumption such as for house constructionpurposes. Others prefer indigenouspremium timber species because ofdurability and finishing quality. Designingmeasures to promote sustainable treecultivation among smallholders thereforerequires a knowledge of the kinds of foresttrees they want to adopt in view of theirrespective needs and objectives.

People’s attitude and perceptions aboutforest and trees

Adoption of tree farming and communityforestry in a given locality depends largelyon how people perceive the value andusefulness of the undertaking in their day-to-day living. Farmers who perceive thatcultivation of trees is beneficial, mostespecially in economic sense, are morelikely to engage in tree farming.

Local knowledge and skills on trees andtree management

Smallholders have important knowledgeand skil ls about trees and theirmanagement that scientists need to know.This arises from passed-down knowledge,years o f exper ience , in fo rma l

experimentation, and discussions with otherfarmers. Promotion of tree cultivation onfarms can be easily attained if forestresearchers and extension agents startdoing things with what the farmers knowand starting from what they initially have. Itis vital to recognize and draw upon thestore of knowledge they have to share onsustainable forest management andconservation.

Smallholders’ awareness andunderstanding of forest policies, rulesand regulations

Smallholders’ desires and eagerness tocultivate trees on farms are often hinderedby their perceptions about inconsistenciesin government policies, rules andregulations. Many believe mistakenly that,for example, harvesting and marketing ofplanted indigenous trees on farms isprohibited by law. This policy, in theirthoughts, is not supportive of thegovernment’s intention to encourage smallfarmers to engage or participate in thena t ion ’s re fo res ta t ion p rogram.Comprehending better the smallholders’level of awareness and understandingabout forest rules and regulations can pavethe way for development agencies todevelop simple and effective mechanismsthat will help farmers to access andunderstand better the relevant policiesgoverning tree farming and communityforestry.

Processing, marketing and pricing oftree farm products

The economic aspect of tree farming andcommunity forestry is an important farmers’consideration as to why they would engagein the venture. Experienced farmers inInopacan, Leyte, for example, areinterested to grow trees on farms becauseof the perceived substantial amount of cashincome that they can realize upon harvest,which may be enough to support their dailyneeds and to send their children to school.The amount of cash income that farmerscan realize in due time, however, is dictatedby the kind and quality of wood productsthat they can produce, together with howthese are valued in the market. Often, theopportunity to improve quality and value ofwood products sold in the market is not

A Forester’s Perspective of the Socio-economic Information Requirements for Forestry 3

accessible to the smallholders.

Local policies and arrangementsregarding tree growing, harvesting andmarketing

In upland communities, a considerableamount of interesting information isavailable about policies and arrangementsformulated and agreed upon by the localpeople in relation to the growing, harvestingand marketing of tree products. Access tothis information may be beneficial not onlyto researchers, but also to developmentworkers and policy makers. These localpolicies and arrangements may haveplayed a key role in areas where the forestenvironment has been sustainablymanaged and protected over time.

Financial requirement of tree-growingversus smallholders’ financial capability

One of the major constraints againstsmallholders engaging in tree farming is thefinancial requirement of establishing andmaintaining the stand during the first aboutfive years. Detailed information about thecost of establishing a hectare of tree farmand how much cost an average smallholdercan shoulder in doing so are criticalinformation. These are relevant to designingprograms that can support smallholders’desires for tree cultivation. For example, ina given community forestry project wherepeople’s participation is indispensable, onecan have a clear idea about appropriatesharing of resource responsibility amongpartners (development organization andlocal people) if they have a soundknowledge about financial requirements oftree farming and what people can afford tooffer.

Alternative source of livelihood

Tree farming is an attractive businessventure but the long wait before the treecrop can be harvested makes smallholdershesitant to commit their scarce land, labourand cash resources. Alternative sources offunds to support day-to-day living needs is

usually the common demand of people incommunities. Since the kind of alternativelivelihood demanded varies from onelocality to another, information is needed onpeople’s desires and preferences foralternative livelihood activities whileengaging in tree farming.

3. DISCUSSION

The Philippines, including Leyte Island, hasundergone large-scale deforestation, suchthat timber harvesting from native forestshas disappeared, and only a fledgling farmand community-based timber industry nowexists. A priority area for forestry researchis therefore the promotion of reforestation,for timber production and environmentalpurposes. The focus of forest managementin recent years has switched to smallholderforestry. This module has identified thekinds of socio-economic information that willassist in the promotion of farm andcommunity forestry in Leyte. Research intoother areas of socio-economic research,timber supply chain analysis, and woodprocessing and marketing, will no doubt beimportant. However, the current priority is toincrease the level of planting, within forestrysystems that integrate with smallholderfarming and meet socio-economic andenvironmental sustainability requirements.

REFERENCES

CF and ICRAF (College of Forestry, LeyteState University and the InternationalCentre for Research in Agroforestry,Visayas) (2002), Local knowledge onindigenous trees in Central Philippines,Summary report submitted to ICRAF TreeDomestication Program, Southeast AsianRegional Office, Bogor, Indonesia.

Lawrence, A.L., Mangaoang, E.O. andBarrow, S. (1999), Foresters, Farmers andBiodiversity: New Issues for the ForestryCurriculum, Proceedings of the NationalWorkshop on Local Knowledge andBiodiversity Conservation in ForestryPractice and Education, held on October19-23, 1998 at ViSCA, Baybay, Leyte.

5

2. Socio-economic Research Techniques in TropicalForestry

Steve Harrison

A wide variety of techniques from the areas of economics, management, sociology andplanning are relevant for carrying out research into non-industrial forestry. This moduleintroduces some of the more widely used of these techniques, which are discussed in moredetail in later modules. On the basis of experience in a forestry research group, commentsare made about relevant applications of these research techniques.

1. INTRODUCTION

To provide input to forest policy, it isnecessary understand the priority thatdifferent groups within the community placeon forestry, and the costs and benefits ofvarious kinds arising from forestry. There isa need to generate information which willassist decision makers, in government andthe private sector, in regard to forestryinvestment and management. Governmentis concerned with strategies to promoteforestry for timber production and widercommunity benefits, and also to regulatethe use of native forests and plantation landfor the public good. Investors andlandholders seek to manage their resourcesin forestry in such a way as to make asatisfactory return on their investment andeffort and promote their livelihood.

Forestry training has tradit ionallyconcentrated on large-scale production andsilvicultural technology, relevant to themaximization of forest biomass production,from monoculture plantations of a few well-known conifer species. A policy interest insmall-scale forestry – particularly farm andcommunity forestry – is relatively new. Inthese areas, financial aspects usuallyremain important, but there is alsoconsiderable emphasis on social andenvironmental objectives. Research in thissituation requires a different mindset to thatof traditional silviculture.

Another point of departure in this module isthat it is concerned with forestry in thetropics. Some broad differences can berecognized between tropical and temperateforestry. In particular, reforestation is inmost cases of relatively recent origin in the

tropics, and the challenge is to induce morelandholders to plant trees. In contrast, insome European countries, forests havebeen in the hands of the same families forhundreds of years, and approach theconcept of a normal or even-aged forest. Insuch cases, the research interest may thenbe in monitoring economic performance offorestry and its relationship with other farmactivities, harvest scheduling and marketingstrategies, and intergenerational transfer offorest resources.

A wide variety of economic and other socialscience research techniques are availablefor tropical forestry research. Theseapproaches are the central theme of thismodule, although it is necessary to placethem in a decision-support perspective.Material in this module on socio-economicresearch techniques in small-scale forestryarises in particular from activities carried outin the Rainforest Cooperative ResearchCentre in Queensland, and draws onHarrison (2001).

This module first outlines the nature androle of socio-economic analysis, withparticular emphasis on natural resource andenvironmental economics. The role ofgovernment in forestry in then examined,including the instruments which govern-ments use to support farm and communityforestry. Next, various research techniquesare outlined briefly, and comments aremade about the relevance of thesetechniques to address particular planningand management issues. Finally, somespeculation is made about research issuesand methods which are likely to figureprominently in the future. A brief discussionsection follows.


2. THE NATURE OF SOCIO-ECONOMICRESEARCH IN RELATION TO SMALL-SCALE FORESTRY IN THE TROPICS

Plantation forestry as a long-terminvestment, usually with the predominantobjective of producing lumber and otherwood products for sale, and providinguncertain financial returns to growers. Italso generating positive externalities for thecommunity. In the case of small-scaleforestry in particular, plantations are oftengrown for multiple objectives. Themanagement of native forests, on the otherhand, is increasingly dominated byenvironmental objectives. For bothplantation and native forests, governmenthas an important role to play, though largelyas a facilitator in the former case and as aregulator in the latter case.

Formulation of appropriate governmentpolicy has high information needs.Economics is one the disciplines which canprovide policy-relevant information. Theorientation of economics is essentiallyanthropocentric, i.e. management policiesare examined in terms of their contributionto satisfying the needs and wants ofhumankind.

A distinction is made between positive andnormative economics. Positive (or ‘what is’economics) attempts to explain howeconomic systems work, by hypothesisingand testing theories on the basis ofempirical evidence. Normative (or ‘whataught to be’) economics attempts toanalyses consequences of alternativecourses of action in terms of specificobjectives, and hence derive prescriptionsfor management. Sometimes the termconditional normative is used, in thatoptimal management policies areprescribed, conditional on specifiedobjectives, constraints and environments.

A distinction is sometimes drawn betweenfinancial and economic analysis. Financialanalysis takes account of costs and returnswhich can be identified through markettransactions. Economic analysis takes abroader perspective, and includes forexample the opportunity cost the unpaidinputs, e.g. of the labour and financialinputs of managers of small firms includingfarms. The term ‘social economic’ is also

used sometimes, e.g. in social cost-benefitanalysis. This does not refer to combinationof economics and sociology, but rather tomaking allowance for impacts beyond theboundaries of the firm, i.e. typically calledspillover effects or externalities. This wouldinclude impacts of environment andcommunities, viewed in terms of their(unpriced) costs and benefits.

Within the discipline of economics, thereare many areas of specialisation. In broadterms, these involve macroeconomics(dealing with national economies) andmicroeconomics (dealing with firms andindustries). At a more specific level, thereare specialisations in international econ-omics, monetary economics, developmenteconomics, transport economics, healtheconomics, econometrics, and so on.During the last 20 years, there has beenrapid development of the fields of naturalresource and environmental economics.These are essentially areas of appliedmicroeconomics, although they also drawon macroeconomics (e.g. in relation togreen national accounts) and developmenteconomics. Within natural resourceeconomics, there is further specialisation,e.g. into land economics, forestryeconomics, fisheries economics, waterresource economics, energy economics,and the economics of outdoor recreation.The areas dealing with agriculture, forestry,fisheries and wildlife (a sub-group of therenewable resources) are sometimesreferred to as bioeconomics. In the lastdecade, ecological economics has alsoemerged as a recognizable economicsspecialisation.

The discipline of natural resourcemanagement has also undergone rapiddevelopment in the last 20 years. This is afield concerned with economics, geography,conservation biology and other areas, butwith a focus on improving the performanceof bioeconomic systems.

Economists and natural resourcemanagement professionals working in thefields of agriculture and forestry require asound understanding of biology. Often theywill work in multidisciplinary teams, whichalso comprise specialists in other fields,including the natural sciences. No onediscipline has a mortgage on all the

Socio-economic Research Techniques in Tropical Forestry 7

knowledge and analysis skills necessary toguide efficient, equitable and sustainablemanagement of natural resources. Ifresearch is carried out from amultidisciplinary (or even interdisciplinary)perspective, then managers can have moreconfidence that the advice provided to themwill be sound. There is a lower risk thatdecisions will be criticised for overlookingimportant perspectives. Sometimes it isargued that what is needed is atransdisciplinary team, in which membersare multi-skilled. While this has obviousbenefits, it is also necessary thatdisciplinary rigour be applied in research.

While early forestry research tended to takea reductionist approach of investigatingspecific relationships, there is now strongemphasis on a systems philosophy, whichtakes an holistic approach to defining andexamining forestry systems.

Any structured research activity can beviewed in the context of the scientificmethod of inquiry, in that a set of implicitsteps can be recognized. In general terms,these include identifying a decision problemand an associated system (e.g. a collectionof resources and their interactions), posingquestions or making hypotheses about thissystem, collecting data, and analysingthese data to draw inferences.

In economics, there has been a tendency torely heavily on quantitative researchtechniques, and mathematical andstatistical models, particularly in NorthAmerica. However, in recent years therehas been increased interest in qualitativetechniques, the usefulness of which hasbeen clearly demonstrated in other socialsciences (e.g. see Patton 1990).

When collecting and analysing social andeconomic data (as indeed in most technicalresearch), it must be recognized that whileobjectivity is sought, nevertheless a gooddeal of subjectivity will be involved. That is,to some extent researchers subjectivelychose their research topics, sites andmethods, and the data they will collect,through to subjectively choosing thesignificance level at which they will declaresignificant differences.

3. THE ROLE OF SOCIO-ECONOMICRESEARCH

Governments play a major role in forestindustries, and particularly non-industrialforestry. This includes support programs toenhance the profitability of forestry and theattitudes of potential tree growers, and toovercome various impediments to treeplanting. It also includes the regulation offorestry, and designed to achieveenvironmental and social sustainability.

In general, the objective of economists andnatural resource management professionalsis to provide information to agencies andindividuals who have the task of managingnatural resources. That is, they producequantitative or qualitative information, withina normative framework, which fulfills adecision-support role. This informationaugments (but does not replace) theinformation, judgments, hunches andtentative decisions of managers.

Socio-economic research is in generaldesigned to support government policy-making, i.e. a social rather than private-producer perspective is taken. The researchoutput augments the information thatdecision-makers already have, confirmingor challenging tentative decisions. The aimis to generate qualitative and quantitativeinformation with will assist in the making ofgood decisions from the viewpoint of thewider community. The spatial focus may beon a community, district, watershed or widerregion.

Socio-economic research provides one ofthe inputs to government policy. It movesbeyond technical considerations andexamines the impact of forestry policies onpeople and on the environment. In the caseof environmental impacts, these may beassessed by scientists, but again the viewof the community towards environmentalissues is critical to policy. For example, ifdeforestation is leading to flooding ofcoastal cities, sedimentation of fisheriesand depletion of groundwater supplies, thenthese issues very much affect humansettlements and place high priority onreforestation.

Desirable policies are not static, but ratherevolve over time, for example as


technology, environmental conditions andcommunity attitudes change. Governmentsdo not automatically support environmentaland social causes, but need to beconvinced or pressured into recognizingtheir worth. Strong environmental lobbiesare now well established, and do notappear to be losing their influence. In somecases, ‘green parties’ have gainedconsiderable power with parliaments. Inlow-income developing countr ies,environmental pressure often comes forexternal loan agencies and donors.

The emphasis of this module is on researchinto small-scale forestry in the tropics andsub-tropics, and this has some differencesfrom small-scale forestry in developedcountries in temperate zones. Typically,there is a history of relatively recentdeforestation, and recognition of the criticalneed for reforestation. Forestry activitiestypically involve reforestation of agriculturalland, rather than management of arelatively ‘normal’ forest. There is greateremphasis on impediments to planting,benefits of reforestation (including carbonsequestration and watershed protectionbenefits) and government assistanceprograms, and less on financial monitoringand harvesting systems.

4. SOCIO-ECONOMIC RESEARCHMETHODS IN SMALL-SCALEFORESTRY

A wide variety or research techniques, bothqualitative and quantitative, have beenapplied to examine socio-economic issuesin small-scale forestry. Some of the moreimportant of these are listed in Table 1.

When a problem in relation to naturalresource management has been defined, itis important to determine who are theaffected parties of stakeholders who avested interest in policy outcomes (Harrisonand Qureshi 2000). It is then necessary tocollect data which will shed light onresource management. If appropriateexperts can be identified, it may be possibleto collect data by discussion with theseexperts and from informal surveys not usinga probabil ity sampling framework.Considerable use was made of informedsources in the study of potential foradoption of Australian tree species in the

Philippines reported by Harrison andHerbohn (2000).

More formal methods can be applied toelicit expert opinion. Delphi surveys seek toobtain group consensus views whileminimizing the interactions between expertsso as to prevent domination on the basis ofpersonality or rank. This was used topredict harvest ages and stand yields to beused in financial modeling of non-traditionaltree species in farm forestry in NorthQueensland, Australia (Herbohn et al. 1999,Herbohn and Harrison 2000). SWOTanalysis (group identification of strengths,weaknesses, opportunities and threats) issometimes used when evaluating a specificprogram or enterprise and exploringimprovement measures, e.g. see Hobbs etal. (2001). Focus group meetings are ameans of generating and testing ideas asan aid to further analysis, such as setting upscenarios for non-market valuation.

Case studies of particular systems (e.g.nurseries, Community Based ForestManagement (CBFM) sites, timberprocessing plants, lumber dealers, furnituremanufacturers) may provide importantinsights into the role and management ofthese systems. There has been increasingrecognition of these qualitative researchtechniques in recent years, particularlywhen a relevant population of substantialsize is not available from which to drawdata (e.g. see Patton 1990).

The various methods of participatory ruralappraisal (PRA) also fit within thisqualitative research sphere. PRA has beenfound useful in designing researchprograms in relation to small-scale forestryin the Philippines (e.g. Singzon et al. 1993).PRA is ‘a systematic, semi-structuredapproach and method of assessing andunderstanding . . . village situations withthe participation of the people and throughthe eyes of the people. It comprises a richmenu of visualisation, interviewing, andgroup work methods that have been provenvaluable for understanding the localfunctional values of resources, for revealingthe complexities of social structures and formobilizing and organizing local people. It istherefore a family of methods andapproaches to enable local people topresent, share, and analyze their


knowledge of life and conditions, to plan, toact, monitor, and evaluate’ (PROCESSFoundation 1996, p. 2)1. Pratt (2001)provided a recent evaluation of PRAapplications in Nepal.

An alternative to these qualitative methodsis to identify a target or reference populationand develop a sampling frame (containingas comprehensive representation of thispopulation as possible), and conduct asample survey. For example, landholdersurveys are used to obtain informationabout attitudes and impediments to small-scale forestry.

While these ‘stated’ opinions may not matchexactly the real decisions of landholders,attitude surveys are much easier to carryout than observing actual behaviour, and inpractice a combination of both may be theoptimal approach. For sample surveys, it isnecessary to choose a sampling design(often a form of stratified or multistagesampling), and to develop and test aquestionnaire.2 Sometimes a semi-structured approach will be useful, wheresome questions are of closed format (e.g.recall a specific fact, respond yes or no,state a degree of belief or preference on aLikert scale), some are open ended (e.g. listreasons for a belief) and some allow free-form discussion (which may be voicerecorded for later qualitative analysis).

Survey results are typically entered onto acomputer package. Electronic spreadsheetshave become extremely popular for dataentry and storage, and for deriving relativelysimple descriptive statistics (frequency 1 PROCESS Foundation (1996) observed that

PRA draws on five traditions – activistparticipatory research, agroecosystemanalysis, applied anthropology, field researchin farming systems, and rapid rural appraisal –in an attempt to achieve ‘peopleempowerment’ and avoid the mistakes of ‘ruraldevelopment tourists’.

2 Some frequently misused terms are ‘survey’,‘questionnaire’, ‘sample’ and ‘observation’.Typically, a single survey is carried out, inwhich there are a number of sample members,to each of which is administered a singlequestionnaire. The sample consists of asubset of members from the population; anumber of observations of each variable underinvestigator make up a single sample to beused in a single survey.

distributions, means and variances) andmaking presentation (line and bar graphs,pie diagrams). Spreadsheets also havesome capability for statistical inferences tobe made concerning the underlyingpopulation. The Statistical Package for theSocial Sciences (SPSS) has proved usefulfor more complex analysis, e.g. clusteranalysis for identifying distinctive groups oflandholders in terms of their attitudes to treeplanting (e.g. Emtage et al. 2001). A varietyof statistical time series analysis techniqueshave been developed, which are powerfulmethods of explaining observed data andmaking forecasts of future values ofvariables. Scenario development isdesigned to describe and understand whatfuture types of situations might arise.

Over about the last 20 years, non-marketvaluation techniques have become widelyapplied in forestry research. The (zonal)travel cost method (TCM) allows demand tobe estimated for recreation sites. Thehedonic price method involves amultivariate analysis of past transactionrecords to determine the relationshipbetween the market value of an asset andits characteristics, including environmentalcharactersitics such as freedom frompollution and good views. In this way, thevalue placed by the market onenvironmental characteristics can beestimated statistically. The contingentvaluation method (CVM) and choicemodeling (environmental choice modelling,choice experiments) are used to estimatetotal economic value (TEV) of naturalassets such as forests, including use andnon-use values. CVM has beencontroversial due to the large number ofpotent ia l b iases and apparent lyunrealistically high values obtained in someapplications. Benefit transfer – inference ofvalues from a source site to target site –provides a time-saving alternative to makingnew estimates for each specific new site. Inpractice, benefit transfer methodology is themost widely used approach to non-marketvaluation, and is being supported bydevelopment of databases of environmentalvalues (e.g. see Morrison 2001).

Agencies concerned wi th forestmanagement have a reporting responsibilityto government in regard to theachievements from spending public funds.


Table 1. Socio-economic research methods in small-scale forestry

Data Stakeholder analysiscollection Elicitation of expert opinion (including consultations with experts, SWOT

analysis, the Delphi method, focus groups)Case studiesParticipatory rural appraisalSample surveys using probability sampling

Data analysis Analysis of survey data – descriptive statisticsMultivariate analysis (including cluster analysis and factor analysis)Price forecasting (time series models)Scenario analysis

Non-market Valueing non-wood forest products and servicesvaluation Evaluation of forest recreation benefits using the travel cost method

Estimation of total economic value – the contingent valuation methodChoice modelling or choice experimentsThe hedonic price methodBenefit tranfer

Reporting Reporting systems for forest enterprises and agencies

Physical and Stand yield modelling (including under sparse data)financial The optimal economic rotation (the Faustmann formula)modelling Discounted cash flow analysis and sensitivity analysis

Development of financial models of forestry investments and overallenterprisesModelling carbon sequestrationCost-benefit analysis (CBA) and cost-effectiveness analysis (CEA)Risk or venture analysis

Watershed Geographical information systems (farm, watershed and regional level)and Interindustry input-output analysisregional Transhipment modelling (locational efficiency and logistical analysis)modelling Multicriteria analysis (and the analytic hierarchy process)

Resource allocation models – linear programmingResource allocation models – goal programmingRegional development models

Policyanalysis

Synthesis of policy directions (transferring research to policy)

While financial outcomes are normallyreported, it is only in recent times thatserious attempts are being made inenvironmental reporting (Herbohn 2000),and these reporting systems are still verymuch at a research stage.

A special type of forestry reporting – that ofcarbon accounting, report ing andmonitoring – is now under developmentand will become critical if carbonsequestration credits from plantation

forestry become available and subject totrade (e.g. see Lamb 2000).

A variety of modelling approaches are usedby researchers in relation to small-scaleforestry. On the physical side, it isnecessary to generate estimates of theyield of woodlots and small borderplantings. This can involve particulardifficulties in the case of small-scaleforestry, where yield observations uponwhich to base modelling are scarce andperformance is generally considerably


below that of research trial and commercialplantation yields. This problem iscompounded when non-traditional speciesare grown. Stand yield prediction in suchsituations using the Chapman-Richardsmodel has been employed by Venn et al.(2000).

The Faustmann optimal economic rotationmodel (Pearse 1990) provides anappropriate economic framework forestimating the returns from forestry andcomparing the economics of alternativeforestry systems. Financial modelling –which requires yield and price estimates,and is usually carried out using aspreadsheet package and applyingdiscounted cash flow functions – allows thepayoff from forestry to be predicted; thismay be restricted to modelling of a forestryenterprise or involve modelling the overallbusiness undertaking (e.g. farm business orcommunity development program). Thisanalysis could be extended to determiningthe financial performance and rent sharingat various stages of the timber productionpipeline, including milling and timbermerchandising.

Economic modelling may be viewed as anextension to financial modelling, in which aneffort is made to include shadow pricesrather than simply market prices, and inwhich non-market values are included,typically in an extended cost-benefitframework. Where a specific objective is tobe achieved (e.g. the planting of 200 ha ofcommunity forestry) or performance can bemeasured in specific units (costs perhectare planted), cost-effective analysis(CEA) rather than CBA may be employed,thus averting the need for estimation ofeconomic benefits.

Since forestry is a long-term enterprise, withuncertain stand yield and future timberprice, some form of risk analysis is normallyattached to the financial analysis, as asensitivity analysis or a risk simulation, sayusing the @RISK simulation add-on toExcel or Lotus 1-2-3 (Harrison et al. 2001).

A variety of approaches have beendeveloped for forestry planning at a regionallevel. Methods of spatial modeling usinggeographical information systems (GIS) arebeing used increasingly in socio-economic

studies, particularly with regard toconservation objectives. The application ofGIS in riparian revegetation studies isreviewed with a case study by Harrison etal. (1998).

Inter-industry input-output analysis isdesigned to estimate the impacts of achange of expenditure (e.g. large on-offinvestment) in an enterprise, and yieldsvarious types of ‘multipliers’ (income, outputand employment) which are indicators ofcommunity benefit from the investment.Multiplier values for a ‘normal forest’ arereasonably well established, but when itcomes to reforestation the evidence issparse. In practice, the upstreaminvestment in tree planting (fromexpenditure on inputs and the consequenteconomic activity driven by thisexpenditure) is usually modest, with a longdelay to harvesting, so that the multiplierstend to be quite small (e.g. Todd et al.1999), raising questions about theusefulness of estimating multipliers.

When dealing with a number of stakeholdergroups with often conflicting objectives, ithas become apparent that the method ofanalysis should take account of multiplegoals. In recent years, there has been muchinterest in multicriteria analysis (MCA) ormulti-objective decision-support systems(MODSS) as an approach to planninglanduse at a catchment level, including forreforestation planning (e.g. RAC 1992;Robinson 2000; Qureshi and Harrison2001). The analytic hierarchy process(Saaty 1995) is sometimes used to elicitstakeholder preference weights in relationto various goals in MCA (e.g. see Harrisonand Herbohn 2002). MCA and MODSSapproaches are able to take into accountthe preferences of the various stakeholdergroups, to utilize both quantitative andqualitative information, and are reasonablyrapid to apply. A criticism can be the highlevel of subjectivity involved.

One form of MCA which can be carried outusing mathematical programming softwareis goal programming. Here a number ofaspiration levels for specific goals are setand the objective is to minimize the sum ofappropriately weighted shortfalls (weightedgoal programming) or to select activitiessuch that the goals are satisfied in priority


order (lexicographic or preemptive goalprogramming). Goal programming is beingapplied by Venn to compare alternativeforest utilization policies by the indigenouscommunity in Cape York in Australia.

Mathematical programming has furtherapplications in forestry, such as harvestscheduling and determining the least-costtransport allocation between supply originsand demand destinations and the optimallocation of processing facilities.

Devising strategies for small-scale forestindustry development in any particularregion is a challenging task. Theoreticalfoundations for this kind of analysis areprovided by Tykkyäinen et al. (1997). TheFLORES model of Vanclay et al. (2000) isan attempt to develop structuredmethodology for examining therequirements for more rapid adoption ofsmall-scale forestry, and is to be trialed inthe Philippines. In contrast, the collection ofpapers of Hyttinen et al. (1996) examine therole of forestry in regional development.

The output of research has to becommunicated to policy makers, and takenup by them, if it is to have practicaloutcomes. This requires assembling theinformation as an integrated package whichcan be comprehended by officers of therelevant administrative staff, and is viewedby them as sensible and politicallyacceptable.

5. DISCUSSION

Socio-economic analysis of forestrysystems has been a neglected researcharea. It is unlike silvicultural research, anddraws on the techniques of the socialscientist, recognizing the community settingand multi-goal nature of small-scaleforestry. Particular issues arise in thetropics and sub-tropics, where reforestationis urgently needed following extensiveforest logging and clearing. There is asevere lack of information about theperformance of non-traditional species andmixed-species plantations. Small-scaleforestry is found to face a large number ofconstraints and to present a wide variety ofpolicy issues. Critical amongst these issuesis how to promote new forest industries, therole of government (if any) in supporting

farm and community forestry investments,and the potential contribution of venturecapital, bank loans, NGO funds, and ethicalor green investments. A number ofinnovative methods appear to offer promisefor enhancing the economic attractivenessof small-scale forestry.

A wide variety of research approaches areavailable in relation to small-scale forestry,with techniques of the economist,sociologist, environmentalist and plannerfound relevant to analysis of forestry issues,to generate information for policy-makers.Economics provides both a conceptualframework for analysis (e.g. the cost-benefitanalysis framework) and specific tools (e.g.discounted cash flow analysis, goalprogramming, transshipment models,interindustry input-output analysis). Someapproaches consist of a diverse collectionof related techniques, e.g. MODSS. Someare tools which allow models to bedeveloped, e.g. the simulation modelingcomputer languages such as Simile (usedin development of FLORES).

Further improvement in – and greateracceptability of – socio-economic researchmethods can be expected in the future.Attention may be paid to refinement ofresearch methodology. It is likely that therewill be greater integration of methods ofanalysis, e.g. of MCA, financial anddiscounted cash flow (DCF) analysis andGIS, to development a more powerfulresearch approach.

The need to have sound information tosupport policy for small-scale forestry canbe expected to intensify, particularly asgovernments appear to be reducing thelevel of support for tree planting. While it iscritical to appreciate the various researchtechniques, any research program needs tobe viewed in a broader context, includingresearch objectives, project design, teambuilding and project management, and thepolicy context in which results will beviewed.

REFERENCES

Emtage, N.F., Harrison, S.R. and Herbohn,J.L. (2001), ‘Landholder Attitudes to andParticipation in Farm Forestry Activities inSub-Tropical and Tropical Eastern


Australia’, in S.R. Harrison and J.L.Herbohn, eds., Sustainable Farm Forestryin the Tropics: Social and EconomicAnalysis and Policy, Edward Elgar,Cheltenham, pp. 195-210.

Harrison, S. (2001), ‘Research Approachesto Environmental-economic Issues in Small-scale forestry’, in A. Niskaanen and J.Vayrynen, eds., Economic Sustainability ofSmall-Scale Forestry, EFI Proceedings No.36, European Forest Institute, Joensuu, pp.201-214.

Harrison, S.R., Herbohn J.L. eds. (2001),Socio-Economic Evaluation of the Potentialfor Australian Tree Species in thePhilippines, Aust ra l ian Centre forInternational Agricultural Research,Canberra.

Harrison, S.R. and Herbohn, J.L. eds.(2002), A Whole-Farm and regionalAgroforestry Decision-Support System.Rura l Indust r ies Research andDevelopment Corporation Research ReportSeries, Canberra.

Harrison, S.R., Herbohn, J.L. and Emtage,N.F. (2001), ‘Estimating investment risk insmall-scale plantations of rainforest cabinetspecies and eucalypts’, in S.R. Harrisonand J.L. Herbohn, eds., Sustainable FarmForestry in the Tropics: Social andEconomic Analysis and Policy, EdwardElgar, Cheltenham, pp. 47-59.

Harrison, S.R. and Qureshi, M.E. (2000),‘Choice of Stakeholder Groups andMembers in Multicriteria Decision Models’,Natural Resources Forum, 24(1): 11-19.

Harrison, S.R., Qureshi, M.E., Sharma P.C.and Tidey, M.E. (1998), ‘Using GIS inMult icr i teria Analysis of RiparianRevegetation Options’, in P. Lai, Y Leungand W. Shi, eds., Proceedings of theInternational Conference on ModellingGeographical and Environmental Systemswith Geographical Information Systems,Chinese University of Hong Kong, HongKong, Vol. 1, pp. 216-222.

Herbohn, J.L. and Harrison, S.R. (2000),‘Assessing Financial Performance of Small-Scale Forestry’, in S.R. Harrison, J.L.Herbohn and K.F. Herbohn, eds.,Sustainable Small-scale Forestry: Socio-

economic Analysis and Policy, EdwardElgar: Cheltenham, pp.37-49.

Herbohn, J.L., Harrison, S.R. and Emtage,N. (1999), ‘Potential Performance ofRainforest and Eucalypt Cabinet TimberSpecies in Plantations in NorthQueensland’, Australian Forestry, 62(1): 79-87.

Herbohn, K.F. (2000), ‘Accounting andReporting of Forest Enterprises’, in S.R.Harrison, J.L. Herbohn and K.F. Herbohn,eds., Sustainable Small-scale Forestry:Socio-economic Analysis and Policy,Edward Elgar, Cheltenham, pp. 104-119.

Hobbs, M, Hytönen, L. and Kangas, J.(2001), Factors Affecting the EconomicSustainability of the Non-industrial PrivateForest Enterprise: A Comparison ofStakeholder Perceptions, paper presentedat the International Symposium onEconomic Sustainability of Small-scaleForestry, IUFRO Group 3.08.00, Joensuu,Finland.

Hyttinen , P., Mononen, A. and Pelli, P. eds.(1996), Regional Development Based onForest Resources – Theories and Practices,EFI Proceedings No. 9, European ForestInstitute, Joensuu, Finland.

Lamb, K. (2000), ‘Carbon-based MarketingOpportunities for Small-scale FarmForestry’, in S.R. Harrison, J.L. Herbohnand K.F. Herbohn, eds., Sustainable Small-scale Forestry: Socio-economic Analysisand Policy, Edward Elgar, Cheltenham, pp.89-103.

Morrison, D. (1991), ‘Non-market ValuationDatabases: How Useful Are They?’,Economic Analysis and Policy, 31(1): 33-56.

Patton, M.G. ed. (1990), Qual i ta t i veEvaluation and Research Methods, 2n d

edn., Sage Publications, Newbury Park.

Pearse, P.H. (1990), Introduction toForestry Economics, University of BritishColumbia Press, Vancouver.

Pratt, G. (2001), Practitioners CriticalReflections on PRA and Participation inNepal, IDS Working Paper 122, Institute of


Development Studies, University of Sussex,Brighton.

PROCESS Foundation (1996), Training onParticipatory Rural Appraisal: A TrainingManual, Tagbilaran City, Bohol.

Qureshi, M.E. and Harrison, S.R. (2001), ‘ADecision Support Process to CompareRiparian Revegetation Options in ScheuCreek Catchment in North Queensland’,Journal of Environmental Management,62(1): 101-112.

RAC (1992), Multi-Criteria Analysis as aResource Assessment Tool, ResourceAssessment Commission, RAC ResearchPaper No. 6, Canberra.

Robinson, J. (2000), ‘Does MODSS offer analternative to traditional approaches tonatural resource management decision-making?’ , Austra l ian Journal ofEnvironmental Management, 7(3): 170-180.

Saaty, T.L. (1995), Decisions making forleaders: The analytic hierarchy process fordecisions in a complex world, RWSPublications, Pittsburgh.

Singzon, S.B., Baliña, F.T., Gabunada, F.M. and Morales, N.O. (1993), TropicalParticipatory Rural Appraisal about Treesand Tree Planting Activities of Farmers inMatalom, Leyte, Philippines, Farm andResource Management Institute, ViSCA,Baybay, Leyte, Philippines.

Tykkyäinen, M., Hyttinen, P. and Mononen,A. (1997), ‘Theories of regionaldevelopment and their relevance to theforest sector’, Silva Fennica, 31(4): 447-459.

Todd, C.R., Loane, I.T. and Ferguson, I.S.(1997), Potential Impact of a Farm ForestryIndustry on the Goulburn RegionalEconomy, Trees for Profit Research Centre,Bulletin No. 10, School of Forestry andResource Conservation, University ofMelbourne.

Vanclay, J.K., Muetzelfeldt, R., Haggith, M.,and Bousquet, F. (2000), FLORES: Helpingpeople to realize sustainable futures, XXIIUFRO World Congress 2000, Forests andSociety: The Role of Research, Sub-plenarySessions, Volume 1, pp. 723-729.

Venn, T.J., Beard R.M. and Harrison S.R.(2000a), ‘Modelling Stand Yield of Non-Traditional Timber Species Under SparseData’, in S.R. Harrison and J.L. Herbohn,eds., Socio-Economic Evaluation of thePotential for Australian Tree Species in theP h i l i p p i n e s , Australian Centre forInternational Agricultural Research,Canberra, pp. 55-77.

Venn, T.J., Harrison, S.R. and Herbohn,J.L. (2000b), ‘Impediments to Adoption ofAustralian Tree Species in The Philippines’,Ch. 14 in Harrison, S.R. and Herbohn, J.L.,Socio-Economic Evaluation of the Potentialfor Australian Tree Species in theP h i l i p p i n e s , Australian Centre forInternational Agricultural Research,Canberra, 167-181.

Vize, S.M. and Creighton, C. (2001),‘Institutional Impediments to Farm Forestry’,in S.R. Harrison and J.L. Herbohn, eds.,Sustainable Farm Forestry in the Tropics:Social and Economic Analysis and Policy,Edward Elgar, Cheltenham, pp. 241-255.

15

3. Data Collection Methods in Forestry Socio-economic Research

John Herbohn

Data collection is an important component of the research process, yielding facts, figuresand views which are subsequently analysed and manipulated in various ways. This moduleoutlines some useful data collection techniques for socio-economic data – including surveysof individuals in a target population, data collection from groups, and scenario development– with reference to the appraisal of forestry projects. The coverage is not exhaustive, butfocuses on some of the techniques that have frequent application in the area of forestry andresource management research.

1. DATA COLLECTION IN THE SOCIALVERSUS PHYSICAL SCIENCES

Data collection methods in the socialsciences differ markedly from those used inthe biological and physical sciences.Typically in the latter, some form of physicalmeasurement of real-life phenomena ispossible under controlled and oftenrepeatable conditions. For instance, theeffects of fertilizer application on seedlinggrowth can be assessed under controlledconditions in a greenhouse by measuringkey parameters such as length of shootsand roots, biomass accumulation andphotosynthetic activity. These types ofexperiments can be controlled, e.g. theamount of nutrients added can be controlledby adding know concentrations in solution,the amount of light can be controlled byusing artificial lighting, and temperature canbe controlled by using a constanttemperature room or a heating or air-conditioning unit linked to a thermostat andtimer. Replication and repeatability inscience is usually achievable. This is notthe case in social science research.

The key difference between collecting datain the social sciences compared with thebiological and physical sciences is ‘people’.Social sciences by their very nature dealwith people – either directly or indirectly –which poses challenges when collectingdata. For example it is not possible to‘measure’ peoples attitudes in the sameway that that the growth rates of seedlingscan be measured. Also, people change inresponse to past experiences and changesin the environment. It is thus extremely

difficult if not impossible to achieve thesame replication and repeatability that is thecase in the physical and biologicalsciences.

2. SURVEYS

Surveys are the most common form of datacollection in the social sciences. Thissection will discuss how surveys can beused to collect information from and aboutindividual people, while the followingsection will deal with the collection ofinformation from groups.

Conducting a survey

The survey process may be thought of asboth the development and administration ofa questionnaire or survey instrument, andthe analysis of the survey data. The majorsteps involved in the survey process are setout in Figure 1. The decisions made in theearly stages will affect the choices at thelater stages – thus the forward links inFigure 1. For instance the informationneeds specified at the start will affect thechoice of sampling design, the way in whichthe questionnaire is structured and theselection of data analysis techniques. Ifthere were only forward links in the processthen the conducting of a survey could bedone one step at a time, completing eachstep before considering the next. Implicit inthis ‘single direction’ approach is theassumption that there are no limiting factorsin later steps. This is seldom, if ever, thecase. For instance there are invariablybudgetary limitations on data collection ordata processing resources. These


limitations restrict the alternatives availableat earlier steps; these backward linkagesare indicated in Figure 1 by dashed linesrunning upwards. Backward linkages runfrom the ‘collect data’ and ‘analyse data’boxes back to the ‘develop questionnaire’and ‘sampling design’ phases. Thisillustrates that major decisions concerningdata collection and analysis should alwaysbe considered before selecting a sampleand designing a questionnaire.

While the amount of information that couldbe collected about a project is almostunlimited, limited time and other resourcesmake it necessary to prioritise theinformation sought. Information needs can

be categorised into three levels ofimportance: (a) absolutely essential,constituting the reason for the survey (in thecase of project appraisals, these data arerequired for the appraisal to be undertaken),(b) highly valuable for making importantdecisions, (c) supporting information whichclarifies the picture but is not essential.Care must be taken to group questionslogically, and to identify the most importantquestions to be put to respondents, and toplace these appropriately within thequestionnaire, e.g. at a point where rapporthas been established with the respondent.More intrusive or personal questions areoften placed near the end of aquestionnaire.

A crucial part of any survey is deciding whatgroup of people is to be surveyed; thisgroup is commonly referred to as thereference population. When seekingestimates for input into project appraisals it

is critical to ask the people who have theexperience, knowledge and skills to be ableto provide reliable estimates. There is nopoint asking people in production to makeestimates about likely sales of a new

Identify information needs

Sampling design

Develop questionnaire

Collect data

Analyse data

Write report

Forward links

Backward links

Figure 1. Major steps in the survey and data analysis process Source: Based on Alreck And Settle (1995, p. 26).

Data Collection Methods in Forestry Socio-economic Research 17

product; this question is best addressed toeither marketing staff or customers. In thecase of gathering judgemental estimatesused in project appraisals, the population islikely to comprise a small number of expertsor semi-experts. In such cases, it may befeasible to distribute questionnaires to allmembers of the population, i.e. to carry outa census. Where the population is of a sizethat does not permit every member to becontacted, within the budget and timeframe,a choice needs to be made regarding thebasic sampling design. Here the typicalchoices are between probability or non-probability sampling. If probability samplingis chosen then further choices need to bemade between sampling designs, thetypical contenders being simple randomsampling, stratified sampling and multistagesampling. As a rule of thumb, the lessexpert or focussed the population withrespect to the parameters being estimated(often corresponding to a large population),the shorter should be the questionnaire.Long questionnaires distributed to groupswith little or no interest in the outcomes ofthe survey will result in a low response rate.Long questionnaires are also moreexpensive to produce and analyse and arethus highly costly when large numbers aredistributed.

Questionnaire development usuallyproceeds through a number of drafts. Aspart of this process the instrument may betested on a small sub-sample in a pilotsurvey; this usually leads to some revisionof questions. It is also critical to ensure thequestionnaire is designed to elicit all theinformation required and that no redundantinformation is sought. This is best done byreference back to the previously identifiedinformation needs.

Implementation of the survey may bethrough personal interview, telephoneinterview, drop-off-and-collect or by mail(‘snail mail’ or email). Personal interviewsare generally expensive and timeconsuming and are suited to the situationwhere the target or reference population issmall and not widely dispersed. Non-response bias is generally not a problembecause of the high participation rategenerally associated with this method. Thisis especially so if respondents are beinginterviewed as part of their employment

duties, which is sometimes the case wheninformation is being collected as input intoproject appraisal. Telephone interviewingcan be effective, especially where theinformation required is straightforward, butis not suitable for the collection ofinformation that is complex and requiresdetailed thought or calculations. Postalsurveys are usually undertaken when largesample numbers are required, from amodest budget. Non-response bias is anissue that needs to be considered no matterwhat method is used; however, it isespecially a concern with postal surveys.

Data analysis is the process through whichthe survey responses are summarised intodescriptive statistics such as tables andgraphs, and perhaps subjected to inferentialstatistical methods such as chi-squaredtests, regression analysis and analysis ofvariance. In a highly structuredquestionnaire, highly specific information issought. Respondents will be required toprovide specific estimates, such asestimates of area of trees planted and thedate when planting took place or thinningand harvesting are likely to occur.Alternatively, respondents may be requiredto choose one option from a discrete set ofoptions or to rank a particular statement ona predetermined scale. In such cases,descriptive statistics such as means,medians and standard errors can be easilycalculated and used in project appraisal.Open-ended quest ions wi th in aquestionnaire allow respondents theopportunity to answer the question in theirown words, and relay their particularperceptions, which can provide insights intospecific issues and problems, but also posechallenges when analysing the responses.

Report generation produces a permanentrecord of the processed data and itsinterpretation. When the information isbeing compiled for internal use – often thecase in project appraisal – the report, ifprepared at all, may be rudimentary andinvolve simple summary tables and briefdiscussion of the data.

A number of texts are available whichprovide more detail on the survey process.An excellent survey research resource isThe Survey Research Handbook by Alreckand Settle (1995).


3. COLLECTING DATA FROM GROUPS

Survey data collected from individuals isoften aggregated to provide insights intogroups beliefs and attitudes or to explainsocial phenomena. In the fields ofeconomics and accounting these data isalso often used to develop economic andfinancial models. However, in sometimesdata are collected directly from groups. Thissection discusses a number of techniquesthat can be used to collect data from groupsas opposed to individuals.

Evidence suggests that forecasts and otherestimates produced by groups have greateraccuracy than those derived fromindividuals. Groups also provide moreinformation, although the marginal increasein information content decreases as groupsize increases. The use of groups alsoprovides an opportunity to gain moreinformation about the range of possibleoutcome values hence giving an insight intothe risk associated with the estimates. Froma behavioural perspective, it is also likelythat a group responsible for a implementinga project will have greater commitment to itif they are involved in providing estimates ofvariables used in the financial analysisleading up to a decision to proceed.

When group members are allowed tointeract, this may be in a structured orunstructured manner. Group processes,particularly those that are unstructured (alsoreferred to as ‘interacting groups’), have anumber of potential shortcomings (Janisand Mann 1977; Lock 1987):

a) Group think. In meetings, one idea isoften pursued for a considerable periodof time and thinking consequentlybecomes narrow or confined. This oftenreflects a common information base ofgroup members and a desire for andencouragement of conformity.

b) Inhibition of contributors. Within groupsthere are often power differentials andmembers of a group may be unwilling tocontradict a superior, or even expressan opinion. Also, dominant personalitiesmay reduce the willingness of others tocontribute.

c) Premature closure. There often is atendency for the group to adopt the firstsatisfactory option or estimate withoutfully exploring other options orpossibilities.

A number of structured group techniqueshave been developed which aim tominimise these social and psychologicaldifficulties, some of which are nowexamined.

Jury of executive opinion

This technique is one of the simplest andmost widely used forecasting approaches.In its most basic form it involves simplyexecutives meeting and deciding on thebest estimate for the item being forecast. Asa precursor to the meeting, it is common toprovide background information toexecutives.

A major drawback of this approach is that itplaces those making the forecasts in directcontact with each other thus allowing adhoc and uncontrolled interaction. Thepotential exists for problems of groupinteractions (e.g. dominance of individuals,group think) to arise. In particular, theweight attached to the opinion of aparticular individual is likely to bedetermined by the rank (and personality) ofthat individual in the organisation. Theviews of executives with the bestinformation or in the best position to makean accurate forecast are not necessarilygiven sufficient weight.

One variation of the this approach involvesthe jury (group) periodically submittingestimates in writing, which are thenreviewed by the president or some othersenior member. The person reviewing theindividual forecasts will sometimes make afinal assessment based on the opinionsexpressed. In doing so, this person will alsooften call on past experience to take intoaccount which executives are biased inwhich direction and then weight eachindividual’s estimates accordingly.Alternatively, the individual estimates canbe averaged to derive an which isconsidered representative of the group.This later approach could almost beconsidered an informal variant of the Delphimethod discussed below


The Delphi method

The Delphi method was originallydeveloped by the Rand Corporation in the1950s to obtain consensus among experts.Since this time it has been refined furtherand applied to gain information in a widerange of fields. These fields are as diverseas regional economic development, healthcare policy, sociology, environmental risk,prediction of fruit prices, tourism andrecreation, forestry and advancedmanufacturing techniques. The Delphitechnique may be particularly useful insituations where objective data are scarce.

The Delphi method is designed to elicitestimates from experts within a group orpanel without allowing interaction betweenindividuals on the panel, thus avoidingproblems with dominant members. Expertsdo however have the ability to revise theirestimates on the basis of group views. Suchan option is not available using thetraditional survey method. This techniqueproceeds through a series of data collectionrounds.

In a classic Delphi survey, the first round isunstructured, allowing panellists to identifyfreely and elaborate on the issues that theyconsider important. These are thenconsolidated into a single set by themonitors, who then produce a structuredquestionnaire designed to elicit the views,opinions and judgements of the panellists ina quantitative form. The consolidated list ofscenarios is presented to the panellists inround two, at which time they placeestimates on key variable such as the timean event will occur. These responses arethen summarised and the summaryinformation is presented to the panellists,who are invited to reassess their originalopinions in light of anonymous individualresponses. In addition, if panellistsassessments fall outside the upper or lowerquartiles, they may be asked to providejustifications as to why they consider theirestimates are more accurate than themedian values.

Further rounds of collection of estimates,compiling summary information and invitingrevisions continues until there is no furtherconvergence of expert opinion. Experience

reveals this usually occurs after two rounds,or at the most four rounds (Janssen 1978).

There are a number of variants on theclassical Delphi method. When the issuesare well defined, a clearly defined scenariocan be developed by the monitoring team.In such circumstances, it is common toreplace the unstructured first round with ahighly structured set of questions throughwhich specific estimates of parameters areobtained. A statistical summary of allresponses is then provided to the panel forthe second round, rather than in the third. Insuch cases, it is common for the Delphimethod to include only one or twoiterations.

The classic Delphi method is conductedthrough a combination of a pollingprocedure and a conference. Commun-ication between conference panellists ishowever restricted and undertaken throughthe monitoring team. Even though panellistsare at the same physical location, there isno face-to-face contact. A variant is the‘paper’ Delphi (sometimes also known as a‘paper and pencil Delphi poll’) that isconducted entirely by mail. Another variantis the ‘real time’ Delphi whereby feedback isprovided by computer and final results areusually available at the end of the session.

The quality of forecasts and other estimatesprovided by Delphi method very muchdepends on how the technique is applied.The following list of suggestions of how tobest apply the Delphi method are primarilythose of Parente et al. (1984) with a fewadditions from other sources:

1. The criteria for the selection of panellists(e.g. education, experience) should becarefully determined and clearlycommunicated.

2. A minimum of 10 panellists after dropoutare recommended although it issometimes suggested that five issufficient

3. Commitment to serve on the panelshould be secured before the first roundof estimates is requested. This willimprove motivation and ensure abalanced sample if dropout is likely.Time should be taken to explain theDelphi technique and the informationprovided.


4. A range of estimation problems may bepresented, although these should beless than 25 in number. Whereappropriate, the main estimates shouldbe broken down into sub-problems.Alternatively, different outcomes mightbe presented and their likelihoodrequested. Either way, the estimates willbe useless unless the right problemsare presented, hence the effort needs tobe put into framing the problem. Someprior testing may be appropriate,especially if the Delphi survey is beingundertaken through the post.

5. Problem statements should not belonger than 20 words and should usequantitative data (e.g. 50% increase)rather than fuzzy linguistics (e.g.‘considerable increase’).

6. The ‘rules’ for good questionnairedesign should be applied to thepresentation of problems. These includeavoiding compound sentences.

7. If the purpose of the Delphi process is togenerate estimation problems then it issuggested that examples of attractiveand undesirable scenarios bepresented.

8. There is little difference in the manner inwhich the Delphi approach is designedin the sense that the same steps areinvolved in the process, regardless ofwhether it is administered by mail, anetworked computer or face-to-facemeeting. Factors such as cost, the needfor timely information or the availabilityof experts to attend a face-to-facemeeting may determine the appropriatemethod.

9. The principle of anonymity should beensured. The organiser’s opinions onthe est imates should not becommunicated to the panellists.

10. The amount and form of the feedbackwill need to be carefully managed. Thenumber of rounds will depend on thepanellists and the manner in which theDelphi survey is conducted (i.e. at whatstage a highly structured questionnaireis distributed). The general advice isthat more rather than fewer rounds, aswell as descriptive feedback, arepreferable. Medians should be provided.

11. Extreme responses should be screenedfor the panellist’s expertise. If the experthas relatively low expertise, then theresponse might be discounted.

12. If the Delphi survey is directed toresearch applications, a detailed reportof the process should be published toallow replication by other researchers ata later time. The range of responsesshould be published to demonstrateconsensus or panellists’ reasoning.

The nominal group technique

The nominal group technique (NGT) usesthe basic Delphi structure but in face-to-face meetings which allow discussionamong participants. A meeting with NGTstarts without any interaction, withindividuals initially writing down ideas orestimates related to the problem orscenario. Each individual then presentstheir ideas or estimates, with no discussionuntil all participants have spoken. Theneach idea or estimate is discussed. Theprocess is then repeated. For this reason,NGT is sometimes known as the ‘estimate-talk-estimate’ procedure. In practical terms,like Delphi, the framing of the questions orthe scenario is crucial for the success of theprocess. Also, ideally, the leader ormoderator of the discussion should comefrom outside the group.

Other group techniques

A number of other group techniques areavailable. The Devil’s Advocate andDialectical Inquiry involve individuals orsmall groups taking a ‘devil’s advocate’ roleor using the dialectic approach (presentingmultiple views) to explore alternativedifferent options. Both methods areconsidered to be ways of overcoming theproblem of ‘group think’.

Lock (1987) outlined a further approach togroup judgmental forecasting which drawsupon elements of the nominal grouptechnique and Inquiry Systems. Inquirysystems according to Lock are simplyphilosophical systems that underlie differentapproaches to analyzing or investigatingparticular phenomena. This approachconsists of seven phases:

1. Problem or task definition2. Pre-collection of estimates of the

variable of interest and the reasoningbehind the estimate


3. Sharing of the estimates andclarification of the underlying reasoningbehind them

4. Discussion of underlying reasoning5. Encouragement of multiple advocacy

(dialectic inquiry)6. Individual revision of estimates7. Synthesis of estimates.

This approach recognises the benefits ofcommunication between groups.

4. THE DELPHI TECHNIQUE APPLIEDTO APPRAISING FORESTRYPROJECTS

The Delphi technique is now illustrated as ameans of collecting information toundertake a financial analysis of forestryprojects, based on two ‘real-life’ Delphisurveys undertaken in northern Australia.

A simple model of appraising forestryinvestment

A simple model for appraising investment inforestry projects is illustrated in Figure 2.This diagram illustrates the key parameterswhich need to be estimated for evaluationof forestry projects, viz. harvest volumesand stumpage prices for the various typesof timber harvested, and input costs. It isalso critical to have estimates of the timingforestry operations and costs and returnsthroughout the plantation life or ‘rotationlength’. The estimates made at the time ofplanting become forecasts for deriving cashflows for the various years throughout theplantation. This information can then beentered onto a spreadsheet in which annualnet cash flows and financial performancecriteria are derived. Performance estimatesare typically made on a one-hectare basis,and then aggregated up for plantation size.

For traditional exotic conifer plantationssuch as radiata or caribbean pine, it isrelatively easy to obtain estimates of thevarious parameters of the model. Forexample, costs of establishment, continuingmaintenance and non-commercial thinningare easily obtained, for example fromcontractors. Yield estimates, along with finalharvest price, are the two key parameters indetermining final harvest revenue. For pineplantations there are well developed standgrowth models for various site indices

based on many years of past growth datathat can provide accurate projections oflikely yield. The following two examples, onthe other hand, apply to situations wherenon-traditional species are grown, hencelittle stand growth data are available.

Example 1: Appraising Forestry ProjectsInvolving New Planting Systems

In recent years there has been a moveaway from traditional silviculture systemsinvolving monocultures of a small number ofmostly softwood species. In Australia forexample, plantations of native hardwoodsincluding many rainforest species havebeen established. In the case of nativetimber species for which little is knownabout the silviculture, it is extremely difficultto obtain estimates of growth rates that canbe accepted with a high degree ofconfidence.

The Delphi technique is a convenientmethod of obtaining estimates of expectedgrowth and harvest age of native speciesfor which there exist no growth modelsbased on past performance or physiologicalcharacteristics. This was in fact the caserecently in tropical Australia, where it wasnecessary to obtain estimates of growthrates and harvest ages for 31 species(Herbohn et al. 1999). In this case, theDelphi method proved to be an effectivemethod to collect plantation productivitydata necessary for the financial appraisal.

This project used the Delphi method toprovide estimates of (a) mean annualincrement or MAI (m3/ha/yr) and (b) time toharvest (years) of 31 species. Harvest ageand MAI are the key biological parametersneeded to estimate yield and harvestscheduling for use in financial models. Inthis case the species for which informationwas sought had either been widely plantedin the area or had been included in aprevious Delphi survey.

Opinions were sought from 13 individualswith extensive experience in growing ofAustralian tropical and sub-tropicalrainforest species for either timber prod-uction or reforestation. Individuals generallyhad either extensive field experience or hadundertaken research involving nativerainforest and tropical eucalypt species.


Panellists were provided with a table listingthe 31 selected species and asked toprovide estimates of their ‘best guess’ ofoptimal rotation period (years) for eachspecies along with estimates of ‘shortesttime to harvest’ and ‘longest time toharvest’. Estimates were also requested forthe ‘best guess’ for expected yield( m 3/ha/yr) based on the ‘best guess’rotation period along with estimates of

‘highest yield expected’ and ‘lowestexpected yield’. In this section, participantswere asked to assume that the trees wouldbe planted on relatively fertile basaltic soils,that average annual rainfall would bebetween 1500-2000 mm, that initial plantingdensity would be around 660 stems perhectare (sph) and suitable thinning regimeswould be applied.

Figure 2. A simple model for appraising investment in forestry projects

Questionnaires were distributed toparticipants followed by a visit by one of theresearch team. Responses for theestimates of growth rates and harvest agesof the 31 selected species were thencollated and averages calculated. Asummary table including the groupaverages was prepared and distributed toparticipants along with their originalestimates. In this second round of theDelphi survey participants were given theopportunity to review their original estimatesof growth rates and harvest ages in light ofthe group averages and to provide anyappropriate revisions or comments. Fewrevisions were received in this second

round and the Delphi process was thenterminated. An extract of the survey formused in the first round of the Delphi surveyis provided in Figure 3. In the second round,a similar table was compiled with theaverages of estimates provided from allpanellists, along with the estimates fromthat particular panellist.

Outcomes of the Delphi survey

The outcome of this Delphi survey was atable of harvest ages and yields, where foreach variable the group mean and highestand lowest estimates were recorded.

Volume at harvest(m3/ha)

StumpagePrice ($/m3)

Cash inflows fromfinal harvest ($) at tn

EstablishmentCosts ($) at t0

Annual maint-enance costs

($) t1 to tn

Cash inflowsfrom thinnings($) at tx, ty, etc.

Plantation netcashflows from

($) t1 to tn

Financial performancecriterion, e.g. NPV


Species Commonname

Optimal rotation period(years)

Yield based on ‘best guess’rotation period (m3/ha/yr)

Best guessharvest age

(years)

Shortesttime toharvest(years)

Longesttime toharvest(years)

Bestguessyield

Highestexpected

yield

Lowestexpected

yield

Acaciamangium

Mangium

Acaciamelanoxylon

Black-wood

Agathisrobusta

Kauri Pine

Araucariacunninghamii

Hoop pine

Beilschmiedabancroftii

Yellowwalnut

Blepharocaryainvolucrigera

RoseButternut

Cardwelliasublimis

Northernsilky oak

Castano-spermumaustrale

Black Bean

Cedrelaodorata

WestIndiancedar

Ceratopetalumapetalum

Coach-wood

Figure 3. Extract of survey form used in Stage 1 of Delphi survey in Example 1

Example 2: Collecting Data for ForestryProjects Involving New PlantingSystems

In north Queensland it has become acommon practice to plant Flindersiabrayleana with Eucalyptus cloeziana andmany potential investors are interested inthe possible financial returns from such asplantation. There is however a lack ofgrowth models for this mixture. Flindersiabrayleana exhibits marked crown shyness(i.e. stops growing when the leaves in itscrown touch the leaves in a crown ofanother tree). There is also a neatrelationship between crown diameter and

the diameter of the stem. These twocharacteristics make it very easy to developa well-structured plantation scenarioinvolving the two species. A Delphi surveywas carried out to obtain information that fordevelopment of a financial model for thetwo species mixture (see Herbohn andHarrison 2001).

A planting and harvesting scenario wasdeveloped for a 50:50 mixture of Flindersiab ray leyana and Eucalyptus cloeziana(Table 1). Thinning and harvesting regimesare timed to occur just as F. brayleyanacrowns touch, at which time lock-up ofgrowth would be expected to occur.


Table 1. Planting and harvesting scenario for a Flindersia brayleyana and Eucalyptuscloeziana mixture

Stage and activity Density after treatment(stems/ha)

Estimates requested from participantsin Delphi survey of reforestationexperts

1. Plant alternating rows of maple andeucalyptus at 3m x 5m spacing

660 (330 F. brayleyana,330 E. cloezianna)

2. Thin to waste every second tree.,when F. brayleyana reaches 18 cmdbh.

340 (170 F. brayleyana,170 E. cloeziana)

Age at which F. brayleyana expectedto reach 18 cm dbh

3. Thin every second eucalypt towaste or for strainer posts. Thinwhen F. brayleyana reaches a dbhof 32 cm.


Age when F. brayleyana expected toreach 32 cm dbh. Bole dbh, small-enddiameter, bole length of E. cloeziana atthis age

4. Remove every second F.brayleyana for strainer posts orsmall diameter logs


5. Remove remaining eucalyptus forpoles.

85 F. brayleyana Ages when E. cloeziana expected toreach 5 specified pole dimensions

6. Harvest every second F.brayleyana when crowns touch(50cm dbh)

43 F. brayleyana Age at which F. brayleyana expectedto reach 50 cm dbhSmall-end diameter and bole length atthis time

7. Harvest remaining F. brayleyana(81cm dbh)

Nil Age at which F. brayleyana expectedto reach 81 cm dbh. Small-enddiameter and bole length at this time

Personal interviews were conducted withfive north Queensland forestry expertschosen for their familiarity with the speciesbeing modelled. At the commencement ofinterviews the plantation system wasoutlined and the requirements forinformation stated. Panellists were providedwith a table, similar to Table 2 but withcolumns 2 and 3 blank, in which to recordtheir estimates after collation of estimates,outliers were identified and clarification wassought from participants. The finalestimates of the parameters that were usedas inputs to the financial analysis areprovided in Table 2.

In this instance, the Delphi method provedto be a timely and cost-effective meansthrough which to collect the information andforecasts necessary to construct thefinancial model.

While it is difficult to judge the accuracy andquality of the forecast information obtainedin Examples 1 and 2, the Delphi surveysprovided information without which theconstruction of a financial model wouldhave been impossible. The stimulus for

under choosing to undertake a Delphisurvey was the fact that forecasts of treegrowth and harvest age from models basedon quantitative growth data will not beavailable until recent plantings using thisspecies mix reach harvest age, and effortsto develop quantitative models based onphysiological and environmental param-eters had failed to produce suitable models.

5. SCENARIO PROJECTION

What will the transport infrastructurerequirements for forestry in Leyte in 25years? What will electricity demand bewithin the province in 10 years and howmuch electricity will be generated fromrenewable resources such as thehydropower and fuelwood? What demandswill the increasing population place ontimber for housing, furniture and charcoal in20 years time? What is the long-termmarket prospects of a new product such asplywood requiring large-scale research anddevelopment expenditure? All of thesequestions are important for capitalbudgeting purposes.


Table 2. Estimates of model parameters for a Flindersia brayleyana and Eucalyptuscloeziana mixed plantation

Parameter estimate from Delphisurvey

Stage and parameter

Average Range2. Age when F. brayleyana expected to reach 18 cm

dbh8.6 years 7 - 10 years

3. Age when F. brayleyana expected to reach 32 cmdbh

17.6 years 15 - 20 years

At this age, the following are expected for E .cloeziana

- bole dbhsmall end diameter- bole length

41.4 cm26.2 cm16.4 m

35 - 50 cm15 - 35 cm12 - 20 m

5. Age when E. cloeziana expected to reach specifieddimensions for

- Pole 1- Pole 2- Pole 3- Pole 4- Pole 5

17.417.621.024.225.6

10 - 25 years11 - 25 years12 - 30 years14 - 35 years15 - 35 years

6. Age when F. brayleyana expected to reach 50 cmdbhExpected small-end dia. at this ageExpected bole length at this age

34 years

33 cm14.8 m

25 - 40 years

30 - 40 cm6 - 20 m

7. Age when F. brayleyana expected to reach 81 cmdbhExpected small-end dia. at this ageExpected bole length at this age

60 years

57 cm16.6 m

50 - 65 years

45 - 60 cm10 - 20 m

For instance, infrastructure for transport,electricity and plywood require long leadtimes to develop. In deciding on whether toproceed with infrastructure and new productinvestments, it is first necessary to estimatedemand for those services and products inthe future. Scenario projection provides aconvenient technique to do this.

Scenarios have been described as‘descriptions of alternative hypotheticalfutures’. Scenarios can be used to describewhat potential futures we might expect,depending on whether major events will orwon’t come to pass. Generating scenarioscan be a useful tool in capital budgeting.

While the term ‘scenario’ is used withvarious meanings, there are severalfeatures which characterise scenarios, viz.

• A scenario is hypothetical – it describessome possible future.

• A scenario is selective – it represents onpossible state of some complex future.

• A scenario is bound – it consists of alimited number of states, events, actionsand consequences.

• A scenario is connected – its elementsare related, that is each element isconditional on or caused by otherelements.

• A scenario is assessable – it can bejudged with respect to its probability ordesirability.

Ducot and Lubben (1980) suggest thatscenarios can be distinguished in a numberof different ways:

Exploratory vs anticipatory. Exploratoryscenarios start with some known orassumed states or events and explore whatmight result, i.e. they look at what mightresult. Anticipatory scenarios start withsome assumed final state of affairs and lookat what preconditions (events or actions)could produce this state of affairs, i.e. theyare backward looking.


Descriptive vs normative. Descriptivescenarios present a possible futureirrespective of their desirability or otherwise.Normative scenarios take values and goalsexplicitly into account.Trend vs peripheral. A trend scenarioextrapolates the normal, surprise-freecourse of events that might be expected ifnothing out of the normal was to happen orno particular course of action taken. Aperipheral scenario depicts radical, trend-breaking or unprobable developments.

Based on pract ica l exper ience,Schoemaker (1991) provided someguidelines for dealing with scenarios:

1. Develop an understanding of any issuesthought important, especially in terms oftheir history, to get a feel for thedegrees of uncertainty.

2. Identify the major stakeholders whowould be interested in these issues.Both those with power and thoseinfluenced should be noted to clarifytheir roles, interests and exact power.

3. Make a list of current trends that mightaffect the issues. Explain how thesetrends might impact on the issues.

4. Identify key uncertainties and explainhow they matter.

5. Construct two ‘forced’ scenarios byplacing all the positive outcomes in onescenario and all the negative outcomesin another.

6. Assess the plausibility of these ‘forced’scenarios. Eliminate impossiblecombinations. These revised ‘forced’scenarios might be called ‘learning-only’scenarios.

7. Reassess the stakeholders in thelearning scenarios. Identify and studytopics for further consideration

8. Develop outline plans based on whathas been learned so far. Communicatedesired scenarios to responsiblemanagers

Example 3: Using scenarioforecasting to estimate demand

As an example of scenario forecasting,suppose the planning department of acompany in the electricity generation andwholesale sector wishes to forecastelectricity demand in 10 years time, toassist in deciding whether to construct a

new hydropower station. The steps inscenario development might proceed asfollows:

1. Understanding of issues and history.Develop an understanding of any issuesthought important, especially in terms oftheir history, to get a feel for thedegrees of uncertainty. These demandshave been growing, but there have beensome unrealistically high demandforecasts in the past which have at ledto excessive capital outlays, andcriticism by environmentalists.

2. Identifying major stakeholders groups.The major stakeholders are otherelectricity generators, electricityconsumers (industrial, commercial anddomestic) and consumer advocates(sensit ive to price increases),environmental groups (concerned aboutincreasing greenhouse gas emissions).

3. Identifying current trends. Electricitydemand is increasing steadily due toincreased population, rural electrificationand increasing ownership of electricalappliances. A campaign has beenundertaken by government toencourage adoption of electricity ratherthan charcoal for cooking, for air qualityreasons. The net impact is likely to berapidly increased electricity demanduntil adoption of electrical appliancesreaches saturation, then a slow increaseas energy efficiency is pursued.

4. Identify key uncertainties. On the supplyside, other generators may install newplant. On the demand side, it is possiblethat new industrial facilities will beconstructed, and major ruralelectrification projects will proceed, butthe boom in construction of office blockscould come to an end. These eventscould lead to major changes in overallgeneration capacity, and in industrialand commercial demand.

5. Identify two ‘forced’ scenarios. The mostfavourable scenario for the generatorwould be where there is little increase ingeneration capacity, but a largeincrease in demand. The most negativeoutcome would be where othergenerators build new plant, butexpected increase in demand does noteventuate. Demand and supplyquantities in megawatt hours wouldneed to be placed on these scenarios,


and price impacts inferred.6. Experience indicates that both extremes

are possibilities, although installation ofcompeting new plant, moderateincrease in industrial demand and littleincrease in commercial demand overthe forecast period appears moreprobable.

7. Reassessment of the stakeholders inthe learning scenarios. Identify andstudy topics for further consideration.Here, further investigation might beundertaken into how researchdevelopments in the field of renewableenergy might affect householdconsumption and the capacity of exportof energy from co-generation of smallelectricity producers such as cane millsmight change with developingtechnology.

8 . Deve lop out l ine p lans , andc o m m u n i c a t e s c e n a r i o s t omanagement. The ‘most probable’forecast and some comments onalternative outcomes would form thecore information.

Further information on the use of scenariosin forecasting can be found in Jungermannand Thüring (1987). Vlek and Otten (1987)have provided an excellent coverage of theuse of scenarios in the energy industry.

6. CONCLUDING COMMENTS – WHICHTECHNIQUE IS BEST?

In this section a number of data collectiontechniques have been outlined which maybe useful in socio-economic studies forestimating future values of key variable asinput into financial appraisals. No matterwhat technique is selected, it is important torecognise its limitations because these willaffect how the technique is applied and thequality of the estimates generated.Furthermore, with any technique thatinvolves the collection of data it is essentialto proceed in an orderly and well thought-out manner. Sometimes there is a tendencyto collect information first and then worryabout how it is to be used. The startingpoint however should be first to clearlyidentify what information is needed, decideon the most appropriate technique to collectthe data (in context of the resources, time

and other limitations) and only thencommence data collection.REFERENCES

Alreck, P.L. and Settle, R.B (1995), TheSurvey Research Handbook: Guidelinesand Strategies for Conducting a Survey,Irwin Professional Publishing, Chicago.

Ducot, C. and Lubben, G.J. (1980), Atypology of scenarios, Futures, 12: 51-57.

Herbohn, J.L., Harrison, S.R. and Emtage,N. (1999), ‘Potential performance ofrainforest and eucalypt cabinet timberspecies in plantations in North Queensland’,Australian Forestry, 62: 79-87.

Herbohn, J.L. and Harrison, S.R. (2001),‘Financial analysis of a two-species farmforestry mixed stand’ in S.R. Harrison andJ.L. Herbohn, eds., Sustainable FarmForestry in the Tropics: Social andEconomic Analysis and Policy. EdwardElgar, Cheltenham, pp. 39-46.

Janis, I.L. and Mann, L. (1977), DecisionMaking, The Free Press, New York.

Jungermann, H. and Thúring, M. (1987),‘The use of mental models for generatingscenarios’, in G. Wright and P. Ayton, eds.,Judgmental Forecasting, Wiley, Chichester,pp. 245-266.

Lock, A, (1987), ‘Integrating groupjudgments in subjective forecasts’, in G.Wright and P. Ayton, eds., JudgmentalForecasting, Wiley, Chichester, pp. 109-127.

Parente, F.J., Anderson, J.K., Myers, P.and O’Brien, T. (1984), ‘An examination offactors contributing to Delphi accuracy’,Journal of Forecasting, 3: 173-182.

Schoemaker, P.J.H. (1991), ‘When and howto use scenario planning: A heuristicapproach with illustration’ Journal ofForecasting, 10: 549-564.

Vlek, C. and Otten, W. (1987), ‘Judgmentalhandling of energy scenarios’, in G. Wrightand P. Ayton, eds. , JudgmentalForecasting, Wiley, Chichester, pp. 267-289.

29

4. Basic Computer Skills and Statistical Methods forAnalysis of Survey Data

Nick Emtage, John Herbohn and Steve Harrison

This module provides an introduction to the use of spreadsheet software packages, to enter,organise and report data from attitudinal and behavioural surveys. In particular, applicationof the Excel spreadsheet for these purposes is illustrated. The data used for illustrationpurposes drawn from a survey of landholders’ attitudes to forest plantation establishment innorth Queensland, Australia. To ensure comprehensive and accurate reporting of theresponses to a survey, it is necessary to carry out a carefully designed series of procedures.The basic stages are data entry, reduction and transformation, analysis and reporting. Figure1 illustrates the methodology adopted to analyse a survey of landholders attitudes to treeplanting and management.

The specific procedures which are discussed in this module include:

1. data entry (spreadsheet formatting, data encoding, data entry, data categorisationand transformation);

2. data summary (development of descriptive statistics such as means and measure ofvariance, summary tables, error checking);

3. data categorisation and transformation (re-categorising nominal data, transformingdata to fit normal distributions);

4. data analysis (Chi-square analysis, one-way ANOVA’s); and5. data reporting (presentation of results of analyses).

1. DESIGNING OF THE SURVEYINSTRUMENT TO MAXIMISE DATAUTILITY

The steps taken following data entrydepend on the project duration and budgetand on the researchers’ aims, experience,training and skill. There is no ‘right’ way toanalyse data from surveys, although theformats or types of data collected in thesurvey and the way they are recorded doesdetermine the types of statistical analysisthat can be undertaken. Compilingdescriptive statistics of the variables in thedata set is the first step and many surveyreports fail to go beyond this and analysethe relationships between the variables.The depth of data analysis required willdetermine the further actions which must beundertaken. If analysis of the relationshipsbetween variables is planned, some form ofdata reduction and transformation istypically needed. Different data types arereduced and transformed in different ways,as illustrated in Figure 1.

It is critically important that the surveyinstrument (i.e. questionnaire) be designedto provide data in a form that is appropriatefor entry into the computer and analysis.Important decisions about the analysis andintended uses of the survey data need to bemade prior to the design of thequestionnaires. The format of the questionsused affects the types of analysis that canbe later undertaken. Those designing thesurvey instrument need to understand thelimitations of different formats of data. Datatypes include nominal data, ordinal data,scales and interval data. If data arecollected in nominal (i.e. categorical) form,this limits the way that analyses can beundertaken. Data collected in an ordinalform (i.e. ranked observations) allow theuse of more powerful statistical analysistechniques and the data can be collapsedinto categories of the analyst’s choosingshould this be required.


Data description

Data reductionand transformation

Data analysis

Figure 1. Methodology for analysing the responses to a survey of landholders in the FarNorth Queensland region of Australia

Source: Emtage et al., in prep.

The desire to collect data in formats thatallow greater analytical power has to bebalanced against ethical concerns and theneed to maximise responses. In Australiamany university ethics committees will not

approve research that asks for a largeamount of detail about an individual. Forexample, questions about respondents’ ageare often required to be formatted as classintervals rather than specific number of

Data entryExamine the frequencies of

each variable for errors

Calculatemeans for

Likert scalevariables

Categoriseresponses to

open questions

Calculatefrequenciesfor nominal

variables

Examinecorrelations betweencontinuous variables

Chi squareanalyses betweennominal variables

One-way ANOVA’sbetween continuous and

nominal variables

Linear modeling

Clusteranalysis

Calculatemeans forcontinuousvariables

Transformvariables withnon-normaldistribution

Factor analysis ofrelated sets of

variables to createscales

Recategorisenominal data

Basic Computer Skills and Statistical Methods for Analysis of Survey Data 31

years. Such formatting may also be morecomfortable for respondents than askingthem to state their exact age.

It is important to test the survey instrumentto ensure that it is well designed, thatquestions are clear, and the range ofresponses can be accurately assessed. It isalso important to test the data entry andanalysis. This provides the researcher theopportunity to set up the data-entryspreadsheets, and to assess what statisticaltests can be legitimately used given thetypes and formats of data being collected. Italso allows the researchers to assess thenumbers of responses that may be requiredto run various statistical tests if the numberof categories used for nominal variables isknown.

2. DATA ENTRY

A number of factors require consideration atthe time of data entry. They includechoosing which software package to use forthe data analysis, setting up the data entryspreadsheet, and setting up datacategorisation and transformations.

Choosing the software package to use

When entering data from survey responsesthe researcher needs to consider the typesof analysis they wish to undertake and theavailability of software packages. If theresearcher plans to undertake advancedstatistical analysis using multivariateanalysis of variance, multiple regressions,factor analysis, cluster analysis ordiscriminant analyses, then specialiststatistical packages such as SPSS(Statistical Package for the SocialSciences) or SAS (Statistical AnalysisSystem) will probably be required. Unlessthe researcher understands advancedmathematics and statistical theory and canwrite their own formulae, entering datadirectly into these specialist programs cansave time. If the researcher does not plan toundertake sophisticated statistical analysesor does not have access to such specialistpackages then basic analyses can beundertaken using spreadsheet packagessuch as Microsoft’s’ Excel or Lotus 1-2-3.

The SPSS package allows users to importdata directly from Excel if the user laterdecides to undertake more advancedstatistical analyses or the packagebecomes available. It is recommended thatresearchers use the specialist packages forall analyses where possible, even the mostbasic, because of the greater ease ofanalysis and reporting from statisticalprograms relative to spreadsheet programs.It should be remembered that all softwarepackages take time to learn. Basicfamiliarity with large programs such asSPSS and Excel can take months while ahigh level of expertise may take years ofexperience to acquire. For the purposes ofthis module, data entry and analysis isillustrated with reference to Excelspreadsheets, because this package iswidely available (as part of the MicrosoftOffice software) and most researchers havesome familiarity with it.

Setting-up the data entry spreadsheet

Just as it is important to know what types ofanalysis will be attempted when designingthe survey instrument, it is also important tokeep the intended analysis in mind whenentering the data into the software program.As a general principle, a masterspreadsheet (and back-up copies!) shouldbe used to enter the data where attemptsare made to capture the greatest possibledetails in the survey responses. For ease ofanalysis the detail can be summarised orreduced in later copies of the spreadsheet.It is inconvenient to add detail at a laterstage and the data entry has to be finishedbefore analyses can commence so it is bestto start by entering all available information.

In the north Queensland forestry survey, thesurvey instrument was a self-administered(i.e. postal) questionnaire. Respondentssent the questionnaires back to theresearch team using pre-paid and self-addressed envelopes. A master recordingspreadsheet was set-up in Excel with therespondents labeled using an identifyingcode in the first column, and with theirresponses to each question recorded insubsequent columns (Figure 2).


Figure 2. Extract from data entry spreadsheet for north Queensland forestry survey

Data categorisation and transformation

In the example presented in Figure 2, tomaintain confidentiality the respondentshave been labeled using a code (in columnA). Coding is used not only to maintainconfidentiality, as in this case, but also tospeed up data entry. Note that theresponses to some questions are alreadycoded.

For example the responses to the questionabout the ownership type (which included‘partnerships’, ‘sole trader’, ‘business’ andothers) has already been coded into anumerical format rather than writing the fullcategory title for each respondent. This iseasily done when there are a limitednumber of categories. In column ‘I’(croptype) the full text of responses hasbeen entered because this question wasframed as an ‘open’ question. Once all ofthe responses have been entered the rangeof responses can be assessed and adecision made about how to collapse orreduce the data. In SPSS, labels can beapplied to categories which are then shown

in reports of analyses to aid interpretation ofthe data. An important part of pilot testing asurvey instrument is to identify the likelyrange of responses to such a question inorder to determine whether to include adiscrete range of responses in thequestionnaire (plus an ‘other’ category), orframe the question in an open format.

An example of the categorisation ofcontinuous data is provided in columns ‘D’and ‘E’ relating to the size of the propertyoperated. In this case the range ofresponses in column ‘D’ were examinedand size classes were determined andcomputed as a new variable in column ‘E’(i.e. less than 20ha = 1, 20-<50 ha = 2, 50-<100 ha = 3, and >100ha = 4). This is oneexample of transforming variables to createnew variables to assist in summary andanalysis of the survey responses. In othercases, transformation of responses may benecessary because of the assumption ofnormal distribution required for somestatistical tests, including one-way ANOVA,as discussed later.


3. DATA SUMMARY

Part of the advantage of usingspreadsheets to enter and organise datafrom surveys is the potential to calculatequickly descriptive statistics of responses tovarious questions. The specialist statisticalsoftware packages such as SPSS aredesigned for this task and are easier to usethan spreadsheet programs such as Excelfor this purpose although Excel is relativelysimple to use. The development ofdescriptive statistics by writing formulae intocells is illustrated in Figure 3. Note that thedifferent data types or formats requiredifferent summary measures. Thecalculation of means for categoricalvariables such as ‘location’ (column B inFigure 3) is meaningless while the ‘count’ ofthe number of responses in each categoryis valid. It is quicker to type a formula into acell (e.g. cell B228 in Figure 3) then copy itacross the spreadsheet than to enterformulae into each cell individuallydepending on the data type. Users can gothrough the columns and delete theirrelevant statistics if they wish to avoidconfusion. Organisation of the data foranalysis and reporting is necessary. Thiscan be done through categorisation of thesheets in a spreadsheet. Data entry ismade onto a ‘master’ spreadsheet, thencopies of this are and are used to carry outdata transformation and analyses. Theseparate sheets in the workbook can beorganised to summarise data by topics,organised as summaries of the statisticaltests used in analyses, or both can be used.The filing system used to manage thevolumes of data generated by surveys andtheir analysis is up to the researcher.

Some of the summary statistics that can bedeveloped using functions in Excel areillustrated in Figure 4 that shows the ‘PasteFunction’ dialog box. Clicking on the ‘fx’button on the ‘standard’ toolbar at the top ofthe screen when Excel is running (as shownin Figures 2 and 3) accesses thesefunctions. The dialog box then promptsusers to enter the required parameters for afunction. Once the user knows the syntaxfor these functions they can be typed

directly into the formula bar (as shown inFigure 3). An alternative to generatingsummary statistics using the calculationfunctions is to use the ‘Pivot Table Function’that is available under the ‘Data’ menu inExcel. This function is discussed furtherbelow.

The summary statistics serve threefunctions. First, they illustrate the types ofrespondents in terms of their land size,average age, education, land use activitiesand so on. Second, these averages can becompared to regional or national averagesto assess if the respondents to the surveyare representative of the broadercommunity (non-response bias tests shouldalso be used). Third, examining thesummary statistics helps to identify if thereare recording errors in the database. It iseasy to make typographic errors that canseriously affect later statistical tests andexamination of the database prior torunning statistical tests is essential.

Another powerful feature of Excel that canbe used to help analyse and report data isthe ‘macro’. Macros allow users to writetheir own functions in Visual Basiccomputer code for specific applications.Like the use of the Excel program generally,it takes time to become familiar with the useof macros and to set-up new code.

If users only need to undertake anoperation such as categorising an ordinalvariable several times it is probably moreefficient to do these tasks manually. If atask is repetitive and needs to be carriedout many times it can be more efficient torecord and alter a macro to automate thetask. Following data entry, macros can beused to automate virtually any of the tasksinvolved in transforming, analysing andreporting data from a survey. Whetherdeveloping macros is more efficient thanmanually carrying out these tasks dependsupon the size of the database being used,the repetition involved in the tasks, and theskills of the researcher as a programmerusing Visual Basic.


Figure 3. Descriptive statistics developed in Excel

Figure 4. The ‘Paste function’ dialog box

Once the summary statistics have beencomputed they can be entered into tables toaid the interpretation of the data. The tables

can be organised to contain only relatedvariables, i.e. those related to a particularsubject. Two such tables are Tables 1 and


2. It is likely that a reasonable sized surveycovering several topics will require theconstruction of many such tables. Graphsare another way to present data, asdiscussed in a later section.

In some cases it is useful to presentsummaries of data using two categoriessuch as land size classes by location asillustrated in Table 3. The ‘Pivot TableReport’ function in Excel (available underthe ‘Data’ menu) allows users to puttogether quickly tables that summarise oneor more than one variable.

Another Excel function that can be used toconstruct summary tables for numericaldata is the ‘descriptive statistics’ function.This function is located under ‘Dataanalysis’ which is in the ‘Tools’ menu. Thedialog box shown after following the above

steps guides users through the use of thefunction. The table produced is like Table 4.

4. DATA CATEGORISATION ANDTRANSFORMATION

Once the responses to the survey havebeen entered into a database and thedatabase has been examined for errors, thenext step toward data analysis involvescategorising and transforming the data intoformats suitable for analyses. In the case ofnominal data, particularly with questionsthat have been framed in an open format,the researcher often has to re-categorisethe initial responses before analyses arepossible.

A trade-off is usually necessary betweenmaintaining the details of the responsesand being able to analyse and report them.

Table 1. Land uses as a proportion of the total landholding for all respondents (%)

Statistical measure Qualitypasture

Degradedpasture

Cropping Fallow Forest Other

Average 67.29 19.23 68.04 11.85 26.89 12.72Standard deviation 29.88 16.70 33.09 11.38 28.41 17.17Minimum 2 1 1 1 0.3 1Maximum 100 60 100 50 100 100

Table 2. Ratings of importance (on 5-point scale) for various reasons for planting trees by allrespondents

Reason for planting trees Average Standarddeviation

Minimumrating

Maximumrating

n

Other reasons 4.39 0.839 2 5 23Protect land resources 3.98 1.157 1 5 172Protect water resources 3.96 1.193 1 5 170Provide fauna habitat 3.64 1.256 1 5 169Personal reasons 3.44 1.301 1 5 170Aesthetic reasons 3.35 1.327 1 5 168Increase value of land 3.16 1.362 1 5 166Windbreak 3.15 1.483 1 5 168Legacy for children 3.13 1.514 1 5 166To make money 2.66 1.472 1 5 167Diversification of income 2.39 1.492 1 5 163Superannuation 2.16 1.483 1 5 164Fenceposts 1.52 0.975 1 5 161


Table 3. Size classes of respondents by location

Location 10 – 20ha

20 – 50ha

50 – 100ha

>100 ha Missing Total

Atherton 6 13 12 13 5 49Johnstone 1 26 30 44 8 109Eacham 12 11 19 16 3 61Unknown 1 3 1 5Totals 20 53 61 74 16 224

Table 5 presents the results of applying thepivot table function to count responses toan open-ended question that askedlandholders what types of crops they growon their land. It can be seen that a numberof the categories are really the same (e.g.Banana and cane; or Cane, bananas, orcane and bananas), but slight differences inthe way they have been entered means thatthe pivot table function reads them asdifferent categories.

There two steps to addressing this problem.The first is to be consistent when enteringthe responses into the database. A hard(i.e. paper) copy of the questionnaire can

be used to record new responses to eachquestion as they are being entered.

This copy can be consulted when recordingresponses to open-ended questions, orcategorising responses to nominalquestions that have an ‘other’ category thatis effectively open ended. This ensures thatconsistent names are given to the sameresponses. The second step onceresponses have been entered into thedatabase is to define categories based onexamination of a range of responses, likethose presented in Table 5.

Table 4. Descriptive statistics for selected land use variables

Statistic Qualitypasture

Degradedpasture

Cropping

Mean 68.2368 23.6667 68.4741Standard Error 2.69739 3.27375 3.02579Median 80 20 83.5Mode 100 30 100StandardDeviation

28.8003 19.6425 32.5887

Sample Variance 829.457 385.829 1062.03Kurtosis -0.754 -0.1582 -0.9025Skewness -0.6836 0.86976 -0.744Range 98 70 99Minimum 2 1 1Maximum 100 71 100Sum 7779 852 7943Count 114 36 116


Table 5. Initial crop types in the responses database

Crop type Frequency Crop type FrequencyNone 107 Cane, bananas 2Aloe Vera, maize and taro 1 Cane, bananas, nursery 1Avocados 1 fruit trees 1Banana and cane 2 Hay 1banana, pawpaw 1 Maize 2Bananas 15 Maize Peanuts Potatoes 1Bananas, pawpaw 1 Maize, potatoes 1Beans and zucchini 1 Maize, peanuts, vegetables 1Cane 62 Mangoes 1Cane & banana 7 Orchid 1Cane & exotic fruit 1 Pasture seed 1Cane & pawpaws 2 Peanuts, cane 2Cane and pawpaws 1 Sorghum, oats and hay crops 1Cane pawpaw 2 Sorghum, oats, rye and grass

for silage1

cane, bananas 1 Tea, cane 1Total 223

The definition of categories is up to theresearcher and depends upon the numberof responses to the questionnaire and thevariation in the data. Categorical data aremore limiting than ordinal data in terms ofthe statistical analyses that can be used.One question facing researchers that wishto analyse relationships between variablesdefined using categorical data is how toestablish a series of categories thatmaintain the diversity in the data yet stillhave sufficient responses in each categoryto allow the use of statistical analyses likethe chi-squared test and one-way ANOVA.When carrying out chi-square tests, eachcell in the table of expected responsesshould have at least five respondents. Ifmore

than 25% of the cells in the expectedfrequency table do not have five responsesthe test results may be unreliable.

Several new variables could be createdfrom the data in Table 5. The simplestvariable would record the presence orabsence of cropping as shown in Table 6.This variable would have the advantage ofhaving many respondents in each category,and the disadvantage of losing a lot ofinformation about the types of crops thatare grown.

Another way to classify the data couldinclude some more details about the typesand mixtures of crops commonly grown(Table 7).

Table 6. Number of respondents growing crops on their land

If crops grown FrequencyCrops 121No crops 103

Table 7. Number of respondents growing crops on their land

If crops grown FrequencyNo crops 107Cane only 62Cane and other crops 22Crops other than cane 33


The resulting classification scheme has fourcategories and reasonable numbers ofrespondents in each category. Theimplications of different classificationschemes for categorical data will be furtherexamined in the following section.

5. DATA ANALYSIS

The Excel program contains a number ofbasic data analysis functions including chi-square tests for independence. An ‘add-in’can be loaded with additional statisticalfunctions including t-tests, z-tests,correlation, covariance, regression andANOVA. In this section the chi-square testis examined.

The relevant application of the chi-squarefor this discussion is to assess whetherthere is a relationship between two sets ofnominal (categorical) data, known as thechi-squared test of independence.

The null hypothesis for this test is that thereis no relationship between the two datacategories1. To run the test in Excel theuser has to calculate the expectedfrequencies of values under the nullhypothesis in a table and compare thesevalues with the distribution of observedfrequencies. The Pivot Table functionmakes it easy to compile the table of actualvalues. An example is provided in Tables 8and 9. The expected frequencies arecalculated by multiplying the row total bythe column total then dividing the result bythe grand total. Thus the expectedfrequency of those who have primaryschool education and have not planted iscalculated as (33 x 123)/196 = 20.71.

The chi-square test for independence isperformed using the CHITEST function inExcel. The chi-square statistic is calculatedas the sum over the rows and columns of:(observed frequency – the expectedfrequency)2 / expected frequency. Thecalculated statistic is then compared to acritical value for the chi-square statistic forthe relevant number of degrees of freedom

1 Technically, this is a test of whether the joint

probability distribution is the product of theunivariate probability distributions for each of thevariables. Further details can be found in Harrisonand Tamaschke (1993, pp. 222-224).

(the product of number of rows less 1 andnumber of columns less 1). The CHITESTfunction returns the probability for a chi-square statistic for the relevant number ofdegrees of freedom. If the probability of thestatistic is less than the designatedsignificance level (usually set at 0.05), thenthe null hypothesis is rejected and it isconcluded that there is a relationshipbetween the two variables or categories. Inthe above example, with the probability ofthe chi-square statistic of 1.3-5 or 0.000013,it is concluded that there is a difference inplanting behaviour between those withdifferent levels of formal education. In otherwords, those with diplomas and degreesare more likely to plant trees than thosewith primary and secondary education.

As mentioned in the preceding section, thecategorisation scheme used to reclassifydata for analyses has important implicationsfor the types of statistical tests that canlegitimately be carried out.

Difficulties may arise in surveys withrelatively small samples if researchersattempt to test relationships betweenordinal variables with more than a fewcategories each.

Consider the example of the different waysof categorising the types of crops grown bylandholders in Tables 6 and 7. The data setof responses to the survey does not havesufficient information to legitimately test therelationship between the crop types grownby respondents and their level of formaleducation (Tables 10 and 11). More than25% (5/16) of the cells in the table ofexpected values (Table 11) have a value ofless than 5. The probability of obtaining thechi-square statistic in this case is 0.02,which is less than 0.05, but the resultshould not be reported since the test isinvalid.

In the example below there are too manycategories in each variable to carry out achi-square test. The alternative is to reducethe number of categories in one or both ofthe variables. An example of this procedureis illustrated in Tables 12 and 13.


Table 8. Actual frequency of respondents who have planted more than 30 trees by educationclasses

Education category If planted TotalNo Yes

Primary school 23 10 33Secondary school 82 31 113Diploma 12 11 23Degree 6 21 27Total 123 73 196

Table 9. Expected frequency of respondents who have planted more than 30 trees byeducation classes

Education category If planted TotalNo Yes

Primary school 20.71 12.29 33Secondary school 70.91 42.09 113Diploma 14.43 8.57 23Degree 16.94 10.06 27Total 122.99 73.01 196

In the second example (Tables 12 and 13,)the reduction in categories of the croppingvariable means that there is sufficientresponses in each cell to use a chi-squaretest. For this example the probability of thechi-square statistic returned by the test isless than 0.0001. Thus the statisticaldecision can be made to reject the nullhypothesis, with the practical inference thatthere are different proportions of thepopulation growing crops when comparingthose with different levels of formaleducation. Inspection of the observed andexpected frequencies used in the test tellsus that those with lower levels of formaleducation are more likely to grow cropsthan those with higher levels of formaleducation. The combining of categoriesinvolves some loss of information aboutrelationships between the variables andthus diminishes our understanding aboutthe relationships.

It can be seen from Table 10 that norespondent with a degree reported growingonly sugarcane as a crop. If the researcherthinks that this point is important and worthpursuing then it possible to constructanother variable for the types of cropsgrown by respondents, with threecategories.

As the survey has sufficient respondentswho report growing sugarcane only thiscategory can be retained, as can thecategory of respondents who grow nocrops. The third category combines thosewho grow sugarcane and other crops, andthose who grow other crops but nosugarcane. The observed frequency tableof those with different levels of education bydifferent crop growing categories wouldthen appear as in Table 14, and theexpected frequencies would be as in Table15.

Table 10. Actual frequency of cropping categories by education classes

Cropping category Education category TotalPrimary Secondary Diploma Degree

No crops 12 42 15 21 90Cane only 14 39 3 56Cane and … 2 15 1 1 19Other 5 17 4 5 31Total 33 113 23 27 196


Table 11. Expected frequency of cropping categories by education classes

Cropping Education category Totalcategory Primary Secondary Diploma DegreeNo crops 15.2 51.9 10.6 12.4 90Cane only 9.4 32.3 6.6 7.7 56Cane + other 3.2 11.0 2.2 2.6 19Other 5.2 17.9 3.6 4.3 31Total 33.0 113.1 23.0 27.0 196

Table 12. Actual frequency of crop growing categories by education classes

Education category Crops Nocrops

Total

Primary school 21 12 33Secondary school 72 41 113Diploma 8 15 23Degree 6 21 27Total 107 89 196

Table 13. Expected frequency of crop growing categories by education classes

Education category Crops Nocrops

Total

Primary school 18.0 15.0 33Secondary school 61.7 51.3 113Diploma 12.6 10.4 23Degree 14.7 12.3 27Total 107.0 89.0 196

Table 14. Actual frequency of those with different levels of education by different cropgrowing categories

Educationcategory

No crops Cane only Cane andother crops

Total

Primary school 12 14 7 33Secondary school 42 39 32 113Diploma 15 3 5 23Degree 21 6 27Total 90 56 50 196

Table 15. Expected frequency of those with different levels of education by different cropgrowing categories

Educationcategory

No crops Cane only Cane andother crops

Total

Primary school 15.2 9.4 8.4 33Secondary school 51.9 32.3 28.8 113Diploma 10.6 6.6 5.9 23Degree 12.4 7.7 6.9 27Total 90.1 56.0 50.0 196


The probability for the chi-square statisticfor the data in Tables 14 and 15 is 0.010.As this is less than the critical probability of0.05, the decision is made to reject the nullhypothesis, i.e. there is a significantdifference in terms of the types of cropsgrown by respondents with different levelsof formal education. It can thus beconcluded that this type of difference existsin the underlying population Comparison ofthe observed and expected frequenciessuggests that the likely source of thedifference is the lower than expectedfrequency of those with degrees growingonly cane.

6. DATA REPORTING

The preceding section has illustrated someforms of summary tables used to present data.The way in which data are presented dependsupon the type of report being compiled and thetypes of statistical tests performed. Whensurvey data are analysed, the presentationcan occur on a number of levels (asillustrated in Figure 1). Reporting of surveyresponses should cover:• responses to survey questions;• transformation of response data in

preparation for data analysis; and• results of all analyses of relationships

between variables prepared from thesurvey responses.

The first stage of reporting is to summariseresponses to each question used in thesurvey before they are modified. Mostsurvey reports have a section describing onthe types of respondents to the survey;tables summarising the data collected aboutthe socio-economic characteristics of therespondents can be used to describerespondents as well as discuss the potentialof non-response bias. Where the survey islarge – in terms of sample size and numberof questions – the researcher may useappendices to report large amounts of dataand concentrate on those analyses anddescriptions that are most relevant to theresearch questions. In the case of theexamples used in this paper (drawn from asurvey of landholders tree planting andmanagement attitudes and behaviour), theinitial data should include description of thesocio-economic character ist ics ofrespondents. The descriptive sections for areport should be organised to present the

responses by topics covered in the survey.The various topics in this case included thereasons landholders plant trees, restrictionsto tree planting on their land, their past andintended planting behaviour, their attitudesto tree planting on a regional scale, andtheir attitudes to past and potential treeplanting incentive and assistance schemes.

In the initial descriptive reporting of surveyfindings, the responses should be reportedas an average or mean figure for allrespondents. Where the survey hascovered clearly different political orgeographic areas, or clearly different typesof people in socio-economic terms, then thedescriptions of responses may beorganised to illustrate these differences inthe respondents. In the case of the northQueensland survey, three local governmentareas over two distinct bio-geographicregions were included. Two of thegovernment areas are located in an uplandarea, and the third is coastal. Thedifferences in the two types of areas arisefrom differences in their climates,topography and soils, as well as the farmsizes and enterprise types. Initialdescription of the responses to the surveyshowed the average responses to thevarious questions for all respondents andfor respondents from each localgovernment area. The presentation of thesedata also described tests for significantdifferences in characteristics of respondentsin the various local government areas. Anexample of such information is provided inTable 16.

Using graphs is an excellent way to displaydata for descriptive purposes or to illustratethe results of analyses. Note that graphs inExcel are called ‘charts’. The type of graphused varies according to the type of datainvolved and the intentions of theresearcher. The pie chart format can beused to illustrate the average proportion ofland used for different activities as shown inFigure 5.

Where the data are in continuous or ordinalform, line graphs or histograms may beused. Line graphs are particularly useful toaid interpretation of relationships betweenordinal variables and to assess if thedistribution of the variable is ‘normal’ or atleast linear. An example of this is shown in


Figure 6, illustrating the initial distribution ofland sizes before they are standardised,with Figure 7 illustrating the distribution ofthe standardised values.

To obtain the graph shown in Figure 6, theraw data were first copied to a new sheet

and sorted according to property size (landarea). Examination of the maximum valuefor the variable showed that one respondentreported a property size of 6902 ha which isclearly an extreme case given that the nextlargest property size is only 500 ha.

Table 16. Importance placed upon various reasons for planting trees by landholders in theJohnstone, Atherton and Eacham shires

Reason for plantingRating by shire Sign. diffs. Mean rating

(all shires)n Frequency

rated 5J A E LSD Bon. (%)

To protect and restore land 3.9 3.9 4.2 ns ns 4.0 172 42To protect the local watercatchment

3.8 4.0 4.2 ns ns 4.0 170 42

To attract wildlife and birds 3.5 3.7 3.8 ns ns 3.6 169 31Personal interest in trees 3.3 3.4 3.7 ns ns 3.4 170 26To improve the look of theproperty

3.2 3.5 3.6 ns ns 3.3 170 26

To increase the value of thefarm

3.1 3.2 3.2 ns ns 3.2 166 19

To create windbreaks 2.8 3.4 3.4 A. E.> J

ns 3.1 168 25

Legacy for children or grandchildren

3.3 2.7 3.2 J > A ns 3.1 166 26

To make money in the future 2.9 2.5 2.4 ns ns 2.7 167 15To diversify farm business 2.6 2.2 2.2 ns ns 2.4 163 13Superannuation or retirementfund

2.3 2.1 2.1 ns ns 2.2 164 13

To provide fence posts 1.5 1.8 1.4 ns ns 1.5 161 3

Notes: (1 = not important, through to 5 = very important). ‘J’ = Johnstone, ‘A’ = Atherton, ‘E’ =Eacham. Significant differences between means for each shire were tested using least squaredifference (lsd) and Bonferroni tests (P > 0.05). Significant differences between mean ratings forresponses for each question were tested using the Bonferroni test. Overlapping lines indicate meanswhich are not significantly different from each other. The mean rating for all shires includes fiveresponses that could not be classified by shire.


Figure 5. Average proportion of landholding used for various purposes in far northQueensland

Figure 6. Distribution of values for the variable Landsize

The graph used to illustrate the distributionof the variable therefore dropped the largestvalue as the graph scale becomes uselesswhen it is included. The shape of thedistribution is parabolic indicating that itcould be transformed to an approximatelylinear cumulative distribution using theLog10 function (i.e. which calculates

logarithms to the base 10) in Excel. Thedata for the variable were transformed bytaking the Log10 of the initial values and anew variable LogSize was created. Thedistribution of this new variable is illustratedin Figure 7.

High quality pasture

Degraded pasture

Cropping

Fallow

Forest

Other

0

100

200

300

400

500

600

Pro

pert

y si

ze (

ha)

Respondent number


Figure 7. Distribution of values for the variable LogSize

When copying and pasting graphs fromExcel to Word (or PowerPoint), open boththe Excel file from which the graph is to betaken and the Word file into which it is to beplaced. Copy the graph using the ‘copy’function under the ‘Edit’ menu in Excel, thenuse the ‘Paste special’ function under the‘Edit’ menu in Word to select the formatused to save the graph in the Worddocument. Using the ‘picture’ format for thegraphs creates the smallest file size, butdoes not maintain a link with the Excel fileused to create the graph, and is moredifficult to edit than a graph saved as an‘Excel object’.

7. CONCLUDING COMMENTS

Modern statistical packages provide aconvenient means to store survey data andpowerful facilities of descriptive andstatistical analysis. Individual researcherstend to have their favourite data analysis

packages, although Microsoft’s Excelspreadsheet package and SPSS are widelyused. Familiarity with statistical packagesrequires practice in their use, but somesimple steps can be laid down for newusers, as set out in this module. It is criticalto plan the types of analysis intended whendeveloping the questionnaire for a survey.

REFERENCES

Emtage, N. F., J.L. Herbohn, S.R. Harrison,and D.B. Smorfitt (in prep.), ‘Landholdersattitudes to farm forestry in far NorthQueensland: report of a survey oflandholders in Eacham, Atherton andJohnstone shires’, Rainforest CooperativeResearch Centre, James Cook University,Cairns.

Harrison, S.R. and Tamaschke, R.H.U.(1993), Statistics for Business, Economicsand Management, Prentice-Hall, New York.

0

0.5

1

1.5

2

2.5

3

1 101 201 Respondent number

Log

10 o

f pr

oper

ty s

ize

(ha)

45

5. Assessing the Financial Viability of FarmForestry: A Tutorial Exercise

Nick Emtage

This module is designed to aid in the understanding of the concepts and techniques used toassess the financial viability of a farm forestry planting, and to develop basis computer skillsin assessing the financial viability of farm forestry plantings. The vehicle for these purposesis a simple financial model of farm forestry, from which the net present value and internalrate of return can be derived. The data entry and financial analysis procedure is set up to befollowed step by step to derive these financial performance criteria, using the spreadsheetpackage Excel. This module provides background to operation of the Excel-based AustralianCabinet Timbers Financial Model (ACTFM).

1. FINANCIAL MODELLING OFSMALLHOLDER FORESTRY

The Australian Cabinet Timbers FinancialModel (ACTFM) was developed by a teamof researchers from the RainforestCooperative Research Centre, at theUniversity of Queensland and James CookUniversity (see references). It is designed toallow landholders and extension officers toassess the potential viability of single andmixed species cabinet timber plantations infar north Queensland. The model wasdeveloped using the Microsoft Excelspreadsheet package, and requires thissoftware to run. A Visual Basic version ofthe model (the Australian Farm ForestryFinancial Model, AFFFM) has also beendeveloped, under a project funded by theRura l Indust r ies Research andDevelopment Corporation (RIRDC). Thisextends the analysis to take into accountthe whole-farm financial situation (i.e. costsand returns to agriculture, management ofnative forests and plantations, andutilization of farm and loan finance), Bothfinancial models – the ACTFM and AFFFM– are currently available free of chargethrough the Rainforest CRC, althoughfurther development and testing iscontinuing for the latter model.

To help understand how these modelsoperate, increase understanding of theconcepts used, and to assist workshop

participants in developing their own model,an exercise has been developed willinvolves the construction of a basic financialmodel. For this exercise,

• All the information needed to developthe model is listed in Table 1.

• It is recommended that workshopparticipants work in teams of two orthree. Preferably these teams shouldinclude one experienced computeruser and one inexperienced usereach. To facilitate learning, theinexperienced user should operatethe computer.

• The various steps to carry out modeldevelopment outlined below are tobe followed. Data should be enteredin the same row and column locationas in the instructions; otherwise, itwill become difficult to follow the cellreferences in the instructions.

2. THE FORESTRY SCENARIO TO BEEVALUATED

The case study for this exercise is a simplefarm forestry scenario limited to a singlespecies, with only one commercial harvest,and a single price paid for all timber on thestump (stumpage price). The followingcosts and timing of management activitiesof Table 1 are to be used.

Table 1. Plantation costs and revenues and other input data


Establishment and maint. costs ($/ha) Year of cost AmountEstablishment cost 0 $2995

Maintenance costsPost plant weed control 1 $1,310.00Post plant weed control 2 $812.00Post plant weed control 3 $213.00First prune (plus certification) 4 $880.00Second prune (plus certification) 8 $648.60Third prune (plus certification) 12 $864.40Thinning 8 $501.00First harvest marking and inventory 44 $80.00

Annual costsInsurance $30Fencing and road maintenance $25Total $55

Project revenues and other parameters

Mean annual increment (MAI) = 19 cubic metres/ha/yearRevenue = MAI * harvest age * stumpage priceHarvest age = 44 yearsPrice = $85/cubic metre (stumpage)Discount rate = 6%

The exercise will involve setting up aspreadsheet containing these data, derivingannual cash flows, and calculating the netpresent value (NPVC) and internal rate ofreturn (IRR). The model will then be used toexamine the effect on NPV and IRR ofchanging cost and revenue parameters andthe discount rate.

3. STEPS IN SETTING UP THESPREADSHEET MODEL

The following steps are to be followed to setup the spreadsheet model for assessing thepotential financial returns from farmforestry:

1. Start the computer and open the Excelspreadsheet program from the ‘Start’menu at the bottom left of the screen.(Click ‘Start’, hold the left button downand move the mouse pointer up to‘Programs’, then highlight ‘Excel’ beforereleasing the mouse button). A newworkbook will appear, as in Figure 1.

2. Insert the floppy disk supplied into the ‘A’drive. Open the file named ‘Financialexercise’ from the floppy disk. (If thisdisk is not available, the data of Figure 2will need to be typed into the

spreadsheet.) The computer screen willthen appear as in Figure 2. (Use themouse to access the ‘File’ menu, holdingthe left mouse button down highlight‘Open…’ then release the button. Usethe ‘Look in’ drop-down text box tohighlight the ‘A’ drive. Release themouse button, then double-click thename of the file with the left mousebutton and the file should open).

3. Enter the costs for each year in themaintenance costs column (column F).Establishment costs go into year 0 (cellF2). The costs for each year could betyped directly into the cells. However, themake the maintenance costs columnupdate automatically when changes aremade to the cost parameters in columnC, it is necessary to enter cell addressesrather than numerical values. This canbe achieved by typing the entry ‘=C4’(without the inverted commas) in cell F2,‘=C8’ in cell F3, ‘=C9’ in cell F4, and soon. A simpler approach is to first click ona cell in column F (e.g. F2) then type ‘=’.Next click on the cell in column C whichrelates to that year (here C4). Press‘Enter’ and the cell should now show thevalue (e.g. $2,995) (see Figure 3). The

Assessing the Financial Viability of Farm Forestry 47

same procedure can be carried out foreach of the other maintenance costs (foryears 1-4, 8, 12 and 44).

4. Click on cell G2 again and position themouse pointer over the square at thebottom right corner of the cell. Thepointer should change from a doublelined cross shape to a single line cross.When this happens, hold down the leftmouse button and drag the pointerstraight down the column so that thecells are within a shaded frame. Go tocell G46 then release the mouse button.Now the value of annual costs shouldappear in G column for each year.

5. Add the total costs for each year incolumn H. First, click on the top cell inthe ‘total costs’ column (H2). Type ‘=‘,then click on the adjacent cell formaintenance costs (F2). Next type ‘+’,

then click on cell G2 to add the totalannual costs for the year. Press enter.Next click on cell H2 again and positionthe mouse pointer over the square at thebottom right corner of the cell. Thepointer should change from a doublelined cross shape to a single line cross.When this happens, hold down the leftmouse button and drag the pointerstraight down the column so that thecells are within a shaded frame. Go tocell H46 then release the mouse button.Now the total value of costs for eachyear should appear in H column. Notethat omitting the ‘$’ sign in the referenceto a cell allows the program to changethe formula as you copy down thecolumn so that it always refers to theadjacent two cells rather than the cellsreferred to in the original formula.

Figure 1. New workbook in Excel


Figure 2. The financial exercise file

6. Calculate the revenues from the sale oftimber and enter this amount into therelevant cell in the ‘revenue’ column (incell I46). Timber revenue is the productof timber price, mean annual incrementand harvest age, here B30 multiplied byB31 multiplied by B32. Note that whenwe nominate cell addresses formultiplication, we are really referring tocell contents. Also, the asterisk is themultiplication sign adopted (to avoidconfusion with the letter ‘x’). Once thecursor is clicked on cell H46, thecalculation ‘=B30*B31*B32’ could betyped directly into this cell. Alternatively,we could type ‘=‘, click on cell B30, type‘*’, click on cell B31, type ‘*’, click on cellB32, and then press Enter. Since cellreferences rather than actual parametervalues have been entered into cell I46,the revenue wi l l be updatedautomatically if a change is made to anyof the three parameters (timber price,MAI and harvest age).

7. Calculate the net cashflows for eachyear. To do this, click on cell J2. Type‘=‘, then click on cell I2. Next type ‘-’,

then click on cell H2. Press ‘Enter’, andthe cashflow for year 0 will be shown incell J2 (see Figure 5). To copy thisformula to the rest of the cells in thecashflows column, click on cell J2 again.Next place the mouse pointer over thesquare on the bottom right corner of thecell and the pointer should change froma double lined cross shape to a singleline cross. When this happens, holddown the left mouse button and drag thepointer straight down the column so thatthe cells are within a shaded frame. Goto cell J46 then release the mousebutton, and the formula will be copied.The values for the scenario shouldappear as in Figure 5.

8. Calculate the net present value (NPV) ofthe scenario. This could be entered insay cell J47, with a caption typed to thecell to the left saying ‘NPV =’. In cell J47,type ‘=NPV(B33,J3:J46) + J2’. Here B33is the discount rate. (Note that is thepercentage sign is not included, thenumber 0.06 has to be typed into thecell; a 6 would be interpreted as 600%.)The formula calls up the NPV function


already available in the spreadsheet toderive the discounted sum of the cashflows. Note that the range of cells in theargumen t within the NPV functioncorresponds to years 1 to 44, and thatthe cash flow for year zero is placedoutside the brackets for the NPVfunction. This is necessary because theNPV function regards the first cash flowin the specified range as correspondingto year 1, and discounts this value backby one year.

The cursor should be now flashing in the‘rate’ box. Click on the cell containing the

discount rate (B35) or type ‘B35’ and areference to this cell this will appear inthe ‘rate’ box (Figure 7).

Next click the mouse pointer in the‘Value 1’ box. Select the on the first cellin the Cashflows column (J2). It will besurrounded by a flashing dashed line.Put the mouse pointer in the middle ofthe cell, hold down the left button anddrag the pointer to cell J46. Release themouse button and the cell referenceJ2:J46 will appear in the ‘Value 1’ box.Last, click on the ‘OK’ button, and theNPV formula is now active in cell B37.

Figure 3. Entering figures into the ‘Maintenance costs’ column


Figure 4. Revenues calculated for the scenario

Figure 5. Completed cash flows for the scenario


9. Calculate the internal rate of return (IRR)for the investment scenario. This couldagain be obtained by using the formulapaste approach but selecting IRRinstead of NPV. An initial guess of theIRR is required, and a value of ‘0.05’ (for5%) is suggested. More simply, the IRRcan be obtained by simply typing‘=IRR(J2:J46)’ in the chosen cell. Thefinancial model is now complete!

4. ASKING ‘WHAT IF?’ QUESTIONS ANDPERFORMING SENSITIVITYANALYSIS

Once a spreadsheet has been developed, itis a powerful tool for examining the impactof variations in assumptions or parameterlevels on project performance. In particular,we are able to pose questions such as‘What will the NPV be if the discount rate isreduced to 5%?’ Arranging such questionsmay be formalised into sensitivity analysis(exploring the sensitivity of financialperformance to levels of variousparameters) and breakeven analysis(determining the levels of parameter valuessuch that the project just covers costs, i.e.such that the NPV is zero). As an exercise:

1. determine how the NPV changes fordiscount rates between 4% and 7%, insteps of 1%.

2. test how the NPV and IRR change whenthe timber yield is (MAI) is reduced to 15m3/ha and increased to 23 m3/ha.

3. determine what timber price would berequired for the project to break even, ata 6% discount rate.

REFERENCES

Emtage, N.F., Herbohn, J.L. and Harrison,S.R. (2001), Australian Farm ForestryFinancial Model (Visual Basic version), TheUniversity of Queensland, Brisbane.

Herbohn, J.L., Harrison, S.R. and Emtage,N.F. (1999), ‘Potential performance ofrainforest and eucalypt cabinet timber inplantations in North Queensland’, AustralianForestry, 62(1): 79-87.

Herbohn J.L. and Harrison. S.R. (2001),‘Financial Analysis of a Two-Species FarmForestry Mixed Stand’, in S.R. Harrison andJ.L. Herbohn, eds., Sustainable FarmForestry in the Tropics, Edward Elgar,Cheltenham, pp. 37-46.

Herbohn, J.L., Harrison, S.R., Emtage, N.and Smorfitt, D.B. (2001), ‘The AustralianCabinet Timber Financial Model: A modelfor mixed species cabinet timber plantationsin North Queensland’, in A. Grodecki, ed.,Managing and Growing Trees: FarmForestry and Vegetation Management,Kooralbyn, 19-21 October 1998,Department of Natural Resources andGreening Australia, Brisbane, pp. 397-405.

53

6. An Introduction to Statistical Package for theSocial Sciences

Nick Emtage and Stephen Duthy

This module provides an introduction to statistical analysis, particularly in regard to surveydata. Some of the features of the Statistical Package for the Social Sciences (SPSS) arethen explained, with reference to a farm forestry survey. Of necessity, this is a brief overviewto the highly complex and powerful SPSS package.

1. INTRODUCTION

Computer based statistical packages are animportant tool for researchers in the socialsciences. The prospect of using statistics issometimes either repugnant or simplyfrightening for people, yet most researchersrecognise the potential utility of statisticalanalysis to aid them to describe, analyse,interpret and report their data. Themathematics behind statistical analysis canbe daunting for those who have little formaltraining in either mathematics or the use ofstatistics. The development of specialiststatistical analysis packages has greatlyreduced the mathematical challenge ofundertaking many analyses. It should beemphasised, however, that these packageshave not reduced the need for researchersto understand the assumptions behindstatistical analyses, and to be able tointerpret their results. The packages havehowever reduced the need for researchersto be able to undertake many of thecalculations that are required for statisticalanalyses. In this way they allow researchersto concentrate on understanding theassumptions behind the various methods,as well as the potential applications andlimitations of various statistical tests.

Statistical software packages have, likeother software packages, changed greatlysince the advent of the personal computer alittle over 20 years ago. Some of theauthors still remember programmingmainframe computers with paper cards.Holes were punched into the cards andthese were then fed into the computer.Computers in those days were scarce,especially the big ones with four megabytememory! Needless to say that statisticaltests were difficult to perform unless the

user had an advanced understanding of themathematics required. More recentcomputer software packages arereasonably easy to use for people withsome familiarity with computers. Most of thepackages have features such as drop-downmenus, ‘tree’ structure diagrams and on-linehelp systems. This said, it should beremembered that the packages discussedhere are large and highly complex. Whilethey are considerably easier to use todaythan they were even 10 years ago, likeother large software packages, familiarityand ease of use are only developed throughpractice with the package. A user canbecome functionally proficient with apackage such as Excel and Word afterseveral weeks, use but development of ahigh level of expertise can take manymonths or even several years.

When choosing which package to use forstatistical analyses a number of factorsmust be considered. These include theavailability of a package, its cost, thefunctions it can perform, familiarity with thepackage and the availability of an expertstatistician to assist with the analysisprocess. As discussed above, the packagestake time and effort to learn, and manyresearchers prefer to continue using aparticular package once they learned howto use it. Other factors may affect thishowever. Availability of a package is animportant factor in deciding whether to useit or not. If an institution has alreadyobtained the rights to use a particularpackage, it may be the only choiceavailable. Buying copies of the latestversions of the specialist statisticalpackages is expensive, as is the cost ofmaintaining the license to use the package.If an institution already has a package that


can provide the functions required then theresearcher may be forced to use thatpackage despite preferences for othersoftware because of limited funds.

Where expert statisticians are available toassist with data analyses then the preferredpackage of the expert is likely to be the oneused. As discussed in other modules it isimportant to discuss research projects withexpert statisticians during their design toensure that the data collected will be in aformat that allows the use of the desiredanalysis techniques. It is also important atthis stage to discuss the packages availableto the researcher and the time available toaccess a computer for data entry, analysisand reporting. Where access to thecomputers with the statistical software islimited it may be possible for the researcherto enter the data into a spreadsheetprogram like Excel and then transfer thedata set to the statistics package in order tocarry out the analyses. In this case it isimportant to have some understanding ofthe formatting required by the statisticalpackage to be used so as to avoidunnecessary reformatting of the data in thestatistical package. Where possible datashould be entered directly into the statisticalpackage to avoid the potential need toreformat the data.

2. THE STATISTICAL PROGRAM FORSOCIAL SCIENTISTS (SPSS)

The SPSS Corporation first produced theSPSS software package in the early 1980’sand has recently released version 11.0. It ispresently one of the most commonly usedstatistical packages in Australian researchinstitutions and is available at all Australianuniversities. The advantages of thepackage are its relative ease of use, itsfamiliarity to many statistical experts and itsfunctionality. One of SPSS’s majordisadvantages is its cost. The SPSScorporation appears to be progressivelybreaking up the program into differentsections that can be purchased separately.For Australian students an individual users’license (one year) costs approximately$A100 for a ‘base’ student version and$A350 for a ‘graduate pack’ licensed for 5years (as of March 2002). The differentversions have varying analytical functionsand different capacities in terms of the

number of cases and variables that can beused. An institutional license costs evenmore, depending on the number ofexpected users. The different packageshave licenses that also differ. In most caseslicenses are set up to expire automaticallyafter a limited period after which thepackage can no longer be used. Thepackage is developed for a number ofoperating systems including Windows andUnix. Information about SPSS products isavailable on-line at www.SPSS.com.

Organisation of the SPSS package

The set-up of the version 10.0 package(used for illustration here) is organised intotwo main sections, for defining and enteringdata and for output. When defining andentering data, users can move between the‘variable’ and ‘data’ ‘views’ by clicking onthe tabs at the bottom of the screen. Thethird ‘output’ section opens in a separatewindow and displays the results of thestatistical analyses. The ‘output’ data aresaved as a separate file to the data set.

In the ‘variable view’ (Figure 1) the userssets up the data entry and analysis cells bynaming and defining the variables includedin the data set. Users are required to usenames for the variables of eight or fewercharacters. Names must begin with analphabetic character. Longer descriptions ofthe variables can be added using the‘Labels’ dialog box (Figure 2). A quick wayto define the variable format (including thevariable type, the number of charactersused and labels) if a number of variableshave a similar format is to copy theattributes of a variable then paste them intoother variable fields.

Once the variables to be recorded havebeen named and defined the user canaccess the ‘data view’ to enter in the valuesfor each variable. The SPSS data viewlooks similar to a spreadsheet program. Thevariables are organised as columns witheach row as a single ‘case’ in the data setcontaining values for the variables relatingto that case. It is common practice to usecodes to enter data into the package andlabels can be used to describe valueswhere needed. For example, codes may beused to record the types of agriculturepracticed on a landholding, or respondent’s

An Introduction to Statistical Package for the Social Sciences 55

educational levels. The defined labels willappear, by clicking the drop-down list arrow

on the right side of the cell, and the usercan select the relevant value (Figure 3).

Figure 1. Variable view in SPSS 10.0

Figure 2. Defining variable labels using the Value labels dialog box


Figure 3. Entering data into the SPSS

This is handy when there is a large numberof possible responses, and thus codes, fora variable, and the user cannot rememberall of them. The user can choose to havethe codes or the labels displayed in the dataview by selecting the ‘Value labels’ optionunder the ‘View’ menu.

Data analysis using SPSS

The SPSS ‘student pack’ has a wide rangeof analytical functions, from basicdescriptive statistics to advanced generallinear modeling capabilities. Specificfunctions are also included to allow thetransformation of variables as preparationfor different tests (e.g. for creatingstandardised or logarithmic values, or thecalculation of scales from a number ofvariables) (Figure 4). The use of thesefunctions allows researchers to calculatequickly new variables based on the valuesof other variables, test variations incategory schemes used to classifyresponses to ‘open ended’ questions, andcollapse categories where necessary.

Once the data are entered into the SPSSprogram it is important to check thedatabase for typographic errors that mayaffect the results of statistical analyses. Onemeans of achieving this is to examine thefrequencies of categorical (nominal) data,and descriptive statistics of numeric(ordinal, scale or interval) data. All of theanalytical functions available in SPSS canbe accessed using the ‘Analyse’ menu(Figure 6). If the ‘Descriptive statistics’ thenthe ‘Frequencies’ options are selected, thedialog box illustrated in Figure 5 appears.This dialog box enables users to select thevariables for which frequencies arecomputed as well as control the types and,to a limited extent, the formatting of displaysof the analyses.

If calculation of descriptive statistics isrequired, users should select ‘Descriptivestatistics’ and the ‘Descriptives’ optionsunder the Analysis menu to reveal the‘Descriptives’ dialog box (Figure 7).


Figure 4. Data transformation functions in SPSS

Figure 5. Frequencies dialog box


Figure 6. Analysis options available in SPSS

Figure 7. Descriptives function dialog box


Once the ‘Descriptives’ dialog box is shown,the variables to be included in the analysesare selected from the list on the left side ofthe box (Figure 7), and transferred to the liston the right side of the box (labeled‘Variables’ in Figure 7) using the arrow inthe centre of the box. The types ofdescriptive statistics that will be Calculatedusing this function can be selected byclicking on the ‘Options…’ button (Figure 7).This reveals the ‘Options’ dialog box for theDescriptives function (Figure 8).

Other analytical functions included in theSPSS student pack (Version 10) includechi-square tests, correlations, regressions,principal components analyses, ANOVA,cluster analyses, general linear modelingand more.

Whilst this paper does not attempt toprovide the reader with statistical skills, theflowchart in Figure 9 may act as a guide forthe reader to access quickly those functionsin SPSS that will best serve their statisticalanalysis needs.

The analytical functions are adequate for allbut the most advanced researchers orthose requiring highly specific analyses.Most advanced or specific applications canbe met as well, with SPSS open tomanipulation via user compiled ‘Sax Basic’computer code (also known as ‘scripts’ inSPSS). This is similar to the use of theVisual Basic programming language to

develop and execute macros in MicrosoftExcel. The Sax Basic language iscompatible with Visual Basic forApplications.

Figure 8. Options dialog box for the‘Descriptives’ function dialog box

Pu

rpo

se o

fsta

tistical a

na

lysis

Su

mm

arizin

gu

niva

riate

da

ta

Exp

lorin

gre

latio

nsh

ips

be

twe

en

varia

ble

s

Te

sting

sign

ifican

ce o

fd

iffere

nce

s

De

scriptive

statistics

(me

an

,sta

nd

ard

de

viatio

n,

varia

nce

, etc)

Fo

rm o

f da

taN

um

be

r of g

rou

ps

Fre

qu

en

cies

Nu

mb

er o

fva

riab

les

Me

asu

rem

en

ts

Nu

mb

er o

fva

riab

les

On

e: co

mp

are

d to

the

ore

tical

distrib

utio

n

Tw

o: te

sted

for

asso

ciatio

n

Ch

i-squ

are

go

od

ne

ss-of-fit

test

Ch

i-squ

are

test fo

ra

ssocia

tion

Tw

o: d

eg

ree

of

rela

tion

ship

Mu

ltiple

: effe

cto

f 2+

pre

dicto

rso

n a

de

pe

nd

an

tva

riab

le

Mu

ltiple

reg

ressio

nL

eve

l of

me

asu

rem

en

t

Sp

ea

rma

n's

rho

Pe

arso

n's

corre

latio

nco

oe

ficien

t

On

e: m

ea

nco

mp

are

d to

a sp

ecifie

dva

lue

On

e-sa

mp

let-te

st

Tw

o

Ind

ep

en

de

nt

sam

ple

sR

ela

ted

sam

ple

s

Fo

rm o

fd

ata

Fo

rm o

f da

ta

Ord

ina

lIn

terva

l

Ma

nn

-Wh

itne

yU

test

Ind

ep

en

de

nt-

sam

ple

s t-test

Ord

ina

lIn

terva

l

Ord

ina

lIn

terva

l

Wilco

xon

ma

tche

d-

pa

irs test

Pa

ired

-sa

mp

les

t-test

Mu

ltiple

Ind

ep

en

de

nt

sam

ple

sR

ela

ted

sam

ple

s

Re

pe

ate

d-

me

asu

res

AN

OV

A

On

ein

de

pe

nd

en

tva

riab

le

On

e-w

ay

AN

OV

A

Mu

ltiple

ind

ep

en

de

nt

varia

ble

s

Mu

ltifacto

rial

AN

OV

A

CH

OO

SIN

G A

N A

PP

RO

PR

IAT

E S

TA

TIS

TIC

AL

PR

OC

ED

UR

E

So

urce

: Co

rston

an

d C

olm

an

(20

00

)

Figure 9. C

hoosing an appropriate statistical procedure


Using SPSS to Describe Data

Whilst computer-based statistical packagesprovide a high degree of functionality withregard to data analysis, they also provide anumber of highly useful tools for thedescription and presentation of summariesof the dataset.

These functions include Descriptives andFrequencies as explained earlier andCrosstabs, also found under the DescriptiveStatistics menu, and Basic Tables, GeneralTables, Multiple Response Tables andTables of Frequencies all located under theCustom Tables menu item (Figure 10). It isoften useful to undertake one or more ofthese processes before commencing dataanalysis to identify any weaknesses in thedataset such as poorly represented groupswithin the sample that may limit thestatistical validity of some forms of analysis.Crosstabs are also an efficient way of

presenting data summaries in research andproject reports.

The charting functions available in SPSSalso provide a number of techniques for theinitial exploration and the presentation ofdata. Scatter Plots (Figures 11 and 12) canbe used to identify quickly the presence andnature of any correlations betweenvariables while Histograms (Figures 13 and14) can be used to present a graphicalrepresentation of the shape of thedistribution of the data for importantvariables.

There is a reasonable amount of literatureavailable to assist users of the SPSSpackage produced by the SPSSCorporation and by independent authors.The tutorial and help facilities for thepackage are comprehensive, generallyeasy to understand and include the on-lineStatistics Coach and Syntax Guide.

Figure 10. Custom Tables drop-down menu selections


Figure 11. Design options for a Simple Scatterplot

Figure 12. Simple Scatterplot displayed in the Output Viewer


Figure 13. Design options for a histogram

Figure 14. Histogram displayed in the Output Viewer



Researchers frequently collect largequantit ies of data, from surveys,experiments and other forms ofobservation. A statistical computingpackage provides a convenient means tostore these data, and derive descriptive andinferential statistics. The Statistical Packagefor the Social Sciences (SPSS) is a widelyused general-purpose survey analysis

package, and hence a useful one to master.It is necessary to allow some learning timeto become familiar with this package, andannual license fees can be a disincentive.

REFERENCES

Corston, R. and Colman, A. (2000), ACrash Course in SPSS for Windows,Blackwell, Oxford.

65

7. A Personal View of Statistical Packages for LinearRegression

Jerome Vanclay

This module presents some personal observations and views on statistical packages forforestry research, with particular attention on suitability for regression analysis, andimportant activity in development of systems models. There are many packages able toassist statistical analyses of data, designed for different purposes and audiences. Thismodule illustrates some of the features that are particularly important when choosing apackage for linear regression analyses. Among these is the graphical abilities of a package,especially the ease with which data (and residuals) can be plotted, because a model may bejudged by the residuals rather than by the R-squared statistic. However, preferences forparticular features and approaches are personal, so provided that candidates have thefunctionality one needs, a package may be chosen on the basis of personal preference. Ifyou don’t like the one you have, or find that it cannot handle your data or your analysis, lookfor another package. There are many well designed and tested statistical packagesavailable, so it should always be possible to find one that suits your needs and budget.

1. INTRODUCTION

Two-variable and multivariate analysis areimportant steps in fitting relationships foruse in systems models. Many statisticalpackages for use on personal computersare available with regression capabilities,and there is great variation in range ofcapabilities, ease of use and cost. This is apersonal overview of statistical packagesused by the author, emphasizing the utilityof the package for fitting curves to datausing linear regression. It is not acomprehensive review, and does notconsider expensive packages such asSPSS, SAS, and S-plus. Instead, it looksmainly at the free or cheap packages thatdo not require an annual license fee. Somebasic concepts in regression analysis arefirst introduced, and then a number ofpackages with regression capabilities arereviewed – specifically Excel, CurveExpert,GLIM, ARC and ViSta.

2. SOME BASIC CONCEPTS INSTATISTICS

Let’s begin by re-examining the principlesunderlying curve fitting with regressionanalysis. Figure 1 is a graph with six datapoints. I’m going to ask you to draw the

single straight line that best describes thetrend evident in these points. Your lineshould be the free-hand equivalent of theleast-squares approach used in regressionanalysis. Understanding the position of thisline is fundamental to an understanding ofregression analysis. Go ahead, and drawyour line.Examining the data graphically, anddeveloping a mental image of the best fitare important first steps in model-building.As Forest Young (2001) puts it in poetry,

first, you see your datafor what they seem to bethen, you ask them for the truth - are you what you seem to me?

you see with broad expanseyet ask with narrowed poweryou see and ask and seeand ask and see … and ask …

with brush you paint the possibilitieswith pen you scribe the probabilitiesfor in pictures we find insightwhile in numbers we find strength

Forrest W. Young, 2001


Now compare your line with thoseillustrated in Figure 2. Many people drawthe first principal component (dotted line)rather than the least squares fit (solid line).They are minimizing the overall distancebetween the points and their line, but that’snot what linear regression does. Linearregression assumes that the X-values (thenumbers on the horizontal axis) are knownwith certainty, and that any errors areassociated only with the Y-values. Thuslinear regression minimizes the verticaldistances between the points and the lines(Figure 3). In fact, it minimizes the square ofthose distances – that’s why it’s called theleast squares fit, and why one of themeasures of goodness-of-fit is the residualmean square (the average of the verticallines squared). How did your line comparewith those in Figure 2?

The ver t ic al li nes in Fi gur e 3 are known as r es idual s. Ex amining the resi duals is a usef ul way to judge the quali ty of a fi t. And becauset he regr es si on technique is bas ed on thes quar e of these res i dual s , you shoul d tak es peci al note of lar ge res iduals , whi c h mayhav e a str ong inf luenc e on the shape andpos it ion of the fit t ed li ne. Consult anys tandar d tex t on regress i on anal ys is for af ul ler explanat ion. The tex t by Cook andWei sber g ( 1994) i s a hel pful pl ace t o begin.

Many peopl e judge the quali ty of a fi t us ing as tati st i c known as R-s quared. An R-s quaredof zero indi c at es that the fi tt ed li ne is nobet ter than a simpl e aver age, and an R- s quar ed of one reveals a perf ec t fit . When R-s quar ed is cl os e to one, it may indi c at e a

good fi t , but it does not indic ate if the fi t is good enough.

Ans combe (1973) created four dat a set s that r ev eal some limit at i ons of R- squar ed andemphasi s e the need to ex ami ne datagraphic all y (Fi gure 4) . The four dat a set s t hathe formulated hav e exact l y the same linearr egress i on es ti mates (Y = 3.0 + 0. 5 X), exact ly the same res idual mean squar e( 13.75) and exact ly the same R- s quar ed( 0. 667) . However, despit e these si mi l ar v al ues, the quali ty of the fi t var ies great ly

0

1

2

3

4

5

0 1 2 3 4 5

Figure 1. A scatter plot.

0

1

2

3

4

5

0 1 2 3 4 5

Figure 3. Regression analysis chooses thedotted line in such a way as to minimize thesum of the squares of the solid vertical lines

0

1

2

3

4

5

0 1 2 3 4 5

Figure 2. First principal component (dottedline) and least-squares fit (solid line) to thesix data points (redrawn from Vanclay 1994,Figure 6.3)

A Personal View of Statistical Packages for Linear Regression 67

bet ween the four dat a set s (Figure 4) . They r ev eal ‘pure er ror’ (t op left ), an outl ier( bott om left ) , us e of the wrong model (aquadr at i c ter m may be needed; top ri ght ), and a case wher e the est i mated trend reli es ent ir el y on a singl e poi nt wi th hi gh leverage( bott om ri ght ). Wit hout fur ther infor mati on, iti s impos si bl e to judge whic h of thes e model s i s suit abl e, but al l exc ept the fi rs t cas ewar rant caref ul i nv est igati on.

A ful ler dis c us si on of this ill ust rat ion was giv en by Ans combe (1973) and Wei sber g( 1985), but the import anc e of pl ot ti ng botht he dat a and the model r emains obv ious.

Bec ause it is so impor tant to pl ot the dataand examine the poi nts vi sual ly , I am goi ngt o cons i der onl y those st at is ti c al pack ages t hat hav e st r ong gr aphic s capabi li ti es. AsFor rest Young (2001) say s in hi s poem, it is ‘ … in pi c tures we fi nd ins ight , whi le innumbers we f i nd s tr ength’ .

Fiv e pac kages wil l be il l us tr at ed, and us ed tof it a si mple li near regr ess ion (Y = a + b X) toa small set of data. The data ar e 10 pair s of numbers , wit h X = 1, 2, 3, …, 10 and Y = X2. Fit ti ng a st r ai ght line isn’t part ic ularl yappropr i at e, but the R-s quared is cl ose toone (0. 95) . We cont r as t the ins i ghts we gai ni nt o thi s analy si s as we ex amine the fi vepac kages – Ex cel, Cur veEx per t, GLIM, ARCand ViSta.

3. MICROSOFT EXCEL

MicroSoft Excel is part of the MicroSoftOffice suite of programs, and is available onmany computers. It’s not my favouritepackage, but it is handy and sometimesmay be the only statistical softwareavailable.

To perform regression analysis with Excel,click on ‘Tools’ and on ‘Data Analysis’. Usethe slider to find ‘Regression’, select and

0

5

10

15

0 5 10 15 20

0

5

10

15

0 5 10 15 20

0

5

10

15

0 5 10 15 20

0

5

10

15

0 5 10 15 20

Figure 4. Four data sets constructed by Anscombe (1973) in which the R-squared is exactly the same (0.667), but in which the quality of the fit variesgreatly


click ‘OK’, and the computer screen shouldlook like Figure 5. If the Tools menu doesn’tinclude ‘Data Analysis’, it may be necessaryto select the add-in: Click on ‘Tools’ and on‘Add-Ins’, and make sure that the ‘AnalysisTookPak’ is selected. If you cannot find the‘Add-Ins’, ask your computer supportpeople to install the full version of Excel.

When performing a regression analysis withExcel, be sure to plot the raw data with thefitted model, and examine the residuals(Choose both the ‘Residual Plots’ and ‘LineFit Plots’ in the regression dialogue box,Figure 5), and do not rely only on the R-squared and the F-test to judge theadequacy of a model.

Statistical procedures in Excel are notalways reliable. McCullough and Wilson(1999) reported that the StatisticalReference Dataset of the AmericanNational Institute of Standards andTechnology reveals problems with Excel’sunivariate summary statistics, analysis ofvariance, linear regression, non-linearregression and random number generation.The problems appear to be present in Excel4, Excel 95, Excel 97 and Excel 2000. Thus

it may be advisable not to use Excel forcomplex or critical analyses.

4. CURVE EXPERT

CurveExpert has limited capabilities, fittingcurves to (X, Y) pairs of data, but it is aneasy way to fit an equation to data.CurveExpert automatically fits andcompares 35 built-in regression models andup to 15 additional user-defined modelscomprising up to 19 parameters. Theseinclude both linear and non-linearregression as well as splines. There is nolimit on the number of data points. Gooddocumentation is available on the Helpmenu.

Data entry and analysis involves simplytyping pairs of data (X, Y) into the built-inspreadsheet, and pressing control-f, andCurveExpert will establish the line of bestfit, graph it, and display the correspondingequation (Figure 6). Graphs can becustomized and copied onto the clipboardfor use in other Windows applications.

Figure 5. Image of a computer screen during a regression analysis with Excel


The major limitation of CurveExpert is that itcan use only pairs of data. This is adequateif you want to build a simple yield tablebased only on stand age, but means thatyou cannot easily include site index or otherexplanatory variables.

CurveExpert is shareware developed byDaniel Hyams. It is available freely on thei n t e r n e t a thttp://www.ebicom.net/~dhyams/cvxpt.htm.Users are requested to pay a once-onlyregistration fee of US$40.

5. GLIM

GLIM (Generalized Linear InteractiveModelling, release 4, Aitkin et al. 1989) isan old favourite, with which I do most of myanalyses. It’s compact – the whole systemfits on a couple of diskettes. And it ispowerful – many analyses can becompleted with just a few commands (e.g.Figure 7, in which the data are simulatedand analysis completed in one line). But it isnot easy to use, so I rarely recommend it toothers. There are no pull-down menus, andwhen GLIM is started, the user is greetedwith a simple prompt, as it awaits acommand. However, it is flexible and

powerful, and is the only package that I canuse to complete some analyses (e.g.Phylogenic regression, Grafen 1989). Themodel fitting illustrated below is achievedwith the following commands:

$yvariate Y$fit X$display estimates$calculate r=%yv-%fv$plot r %fv

The commands (prefixed with a ‘$’) can beabbreviated, often to one or two characters.The equation r=%yv-%fv calculates theresiduals by taking the difference betweenthe y-variables (%yv) and the fitted values(%fv). The graph is a plot of the residualsand the fitted values, and is a good way tojudge the quality of a fit

GLIM is marketed by the NumericalAlgorithms Group in Oxford, UK (NAG Ltd,[email protected]). When I purchasedmy copy (several years ago), there was aonce-only fee of about £100. Crawley(1993) published a helpful guide to the useof GLIM.

Figure 6. Image of computer screen while fitting a curve with CurveExpert


6. ARC

Arc (Cook and Weisberg 1999) is a revisionof the program R-code (Cook and Weisberg1994). R-code greatly influenced myapproach to regression analysis, and Archas become my favourite regressionanalysis package. It has many innovativetools to allow insightful graphical scrutiny ofdata.

I quickly found Arc intuitive and easy to use.It has pull-down menus, but the user mayneed to consult the text (Cook andWeisberg 1999) to get the most from thepackage. Figure 8 shows the analysis of thedata used in Figure 6 in Arc, but the realstrengths of Arc become apparent only withmore complex sets of data.

ARC is copyrighted software, but may bedistributed and used free of charge. It isava i l ab le f r om the web a twww.stat.umn.edu/arc.

7. VISTA

I’ve recently discovered ViSta, the VisualStatistics System (Young 2001), and amstill learning it, so I’m not well qualified topresent it. But I would like to attention tothis package, because I think that it is auseful system and it has many innovativefeatures. It encourages graphicalexamination of data, and has an innovativeway to document what has been done, andto suggest what can be done with the data(Figure 9).

Figure 7. Image of computer screen while using the statistics package GLIM (release 4)


Figure 8. Image of computer screen while using the statistical package ARC

Figure 9. Image of computer screen while using Vista

Note: Note symbols to show what has been done in the analysis (top left), and automatic calculation anddisplay of residuals and influences (bottom right) for every analysis.


ViSta is free, and can be downloaded fromthe web at www.visualstats.org.


I am not going to recommend any particularone of these packages – they all havestrengths and weaknesses. I like the utilityof Excel for data input, the ease of fittingequations CurveExpert, the power of GLIM,the graphics abilities of Arc, and theinnovation in ViSta. But Excel has bugs,CurveExpert has limited capabilities, GLIMisn’t easy for non-statisticians, Arc doesn’toffer sufficient help (unless the text is alsoobtained), and ViSta takes time to learn.Preferences for particular features andapproaches are personal, so provided thatcandidates have the functionality you need,choose a package that you like. If you don’tlike the one you have, or find that it cannothandle your data or your analysis, look foranother package. There are many goodpackages out there, and I’m sure that you’llfind one that suits your needs and budget.

REFERENCES

Aitkin, M., Anderson, D., Francis, B. andHinde, J. (1989), Statistical Modelling inGLIM, Oxford Statistical Science Series 4,Oxford Science Publications, ClarendonPress, Oxford.

Anscombe, F.J. (1973), ‘Graphs instatistical analysis’, American Statistician27(1): 17-21.

Cook, R.D. and Weisberg, S. (1994), AnIntroduction to Regression Graphics, Wiley,New York.

Cook, R.D. and Weisberg, S. (1999),Applied Regression Including Computingand Graphics, Wiley, New York.

Crawley, M.J. (1993), GLIM for Ecologists,Blackwell, Oxford.

Grafen, A. (1989), ‘The phylogeneticregression’, Philosophical Transactions ofthe Royal Society, London, Series B, 205,581-98.

McCullough, B.D. and Wilson, B. (1999),‘On the accuracy of statistical procedures inMicrosoft Excel 97’, ComputationalStatistics and Data Analysis, 31(1): 27-37.

Vanclay, J.K. (1994), Modelli ng Fores tGr owth and Yiel d: Appli cat ions to Mi xedTr opical For est s, CAB Internati onal, Wall ingfor d.

Weisberg, S. (1985), Appli ed Li nearRegress i on, 2nd. edn., Wi ley, New Yor k.

Young, F.W. (2001), How to use ViSta: theVisual Statistics System, The VisualStatistics Project, Psychometric Laboratory,University of North Carolina.

73

8. Simulation Philosophy and Methods

Steve Harrison

In general, to ‘simulate’ means to mimic or capture the essence of something, withoutattaining reality. For management-oriented applications, the something is an identifiedsystem under the control of management – for example of a bioeconomic system – and theessence is captured by way of a symbolic or algebraic model. Simulation consists of thesteps of developing the model to represent a real system, and then performing experimentsusing this model to predict how the real system would behave under a range of managementpolicies. The objective of simulation may be to increase understanding of the behaviour ofthe system, or to compare various policies for management of the system. While manyquantitative techniques take a well-recognized form, simulation differs in its great flexibility,variety of applications and variations in form. These features, while highly valuable formodelling complex systems, make this a difficult methodology to explain and to comprehend.In fact, simulation has been described as ‘more art than science’. Proficiency with thistechnique cannot be gained easily in the classroom, but rather requires considerablepractical experience. This module presents the basic concepts of simulation and the stepsfundamental to simulation studies. The module first defines the nature of simulation and thephilosophy behind this technique. Modelling concepts and elements of models are thendiscussed. The typical steps when using simulation are next outlined. A simple example ofdeveloping and applying a model is presented to aid discussion. Some comments are madeabout validation of simulation models and about design of simulation experiments.

1. WHAT IS SIMULATION?

Simulation consists of a number ofdisparate concepts and techniques, withdifferent terminologies adopted by differentdisciplines. Hence it is useful to establishthe terminology which will be adopted in thisdiscussion of the application of simulation tobioeconomic systems.

Proponents of simulation usually subscribeto what is called the systems approach.According to Shannon (1975), a system is‘a group of objects united by some form ofinteraction or interdependence to perform aspecified function’. In other words, anysystem consists of a number of interrelatedand interacting parts; further, these partsshould not be studied in isolation but ratherin the context of the overall system and itscomplex interdependencies. The whole ismore than just the sum of the parts. Anychange to one part of the system maycause unexpected changes elsewhere.Scientists, engineers, managers and so onhave increasingly realized that it isnecessary to take this holistic view of thesystems they design and operate.

As the body of scientific knowledge hasincreased, there has been a tendency forgreater specialization of research, with aloss in overall perspective and loss ofcommunication between researchers indifferent disciplines. This ‘spread ofspecialized deafness’ has led to the studyof more narrowly defined systems. This isnot to suggest that reductionist researchinto narrowly defined systems does not playan important role in advancement ofknowledge. However, from a managementpoint of view, the system of interest isusually the level of aggregation at whichplanning and control decisions are made,which is often an overall business firm or aparticular project being undertaken by afirm.

Various terminologies have been adopted inpresentations of simulation methodology. Inparticular, the terms systems analysis,systems research and simulation have beenused interchangeably. The term ‘systemsresearch’ is typically applied to describe allthe steps in the study of an organizedsystem. ‘Systems analysis’ was originallyused in this broad context, but has recentlybecome applied more often to just one stepin systems research, viz. that of identifying


the boundar ies , e lements andinterrelationships of a system prior tomodelling it. The term ‘simulation’ issometimes applied to the overall procedureof modelling and experimentation, andsometimes to the experimentation stageonly. In this module the term ‘simulation’ willbe used in the broader context, as asynonym for systems research.

Under the systems approach, a model isdeveloped which represents as closely aspossible the essential structure andperformance of the real system, in terms ofthe specific behavioural features theresearcher wishes to examine. This modelonce constructed and tested for reliability isthen used to simulate or mimic how theactual system would behave underparticular circumstances, by conductingexperiments on a computer using themodel. In these experiments, measures ofsystem performance are generated whenlevels decision variables are set at anumber of levels.

Simulation tends to be used where asolvable model is not available (i.e. wouldnot represent the complex structure of thereal system adequately). The simulationexperiments may be likened to observinghow the real system would perform ifparticular management policies were to beadopted, except that the real system is notinterfered with, real resources are not used(apart from computing resources), and timeis greatly compressed. Computer simulationexperiments can thus provide a great dealof information about how an actual systemwould behave, under a host of differentpolicies and environments, and this mayprovide a greatly improved understanding ofthe system and how to manage it.1

1 A particular type of simulation is that of various

forms of games. There is a long history of wargames, both using model ships or soldiers ona table, or using actual troops but dummyammunition. In recent years, enormousresources have been put into development ofcomputer games, and some of the computergraphics in these have become incrediblygood simulations of real world scenes andactivities. Another gaming application is that ofmanagement games for training in business,and macroeconomic games for training ineconomics. For these games to be consideredgenuine by participants, a credible model must

Practitioners of the systems approach needto have a good understanding of the overallsystem, and the ability and willingness toconsult with experts on various componentsof the system. In fact, systems research isof ten conducted by groups ormultidisciplinary teams rather thanindividuals, since these can take on board abroader range of expertise on variousaspects of the system.2

2. THE NATURE OF MODELS

The term ‘model’ has wide everyday use.Physical or iconic models are typicallyscaled down versions of real systems.Some physical models are not true-to-scalereplicas, but still convey information about asystem. In a broad sense, any map is amodel of a territory, and any timetable is amodel of the operation of a transportsystem. Hence, any form of model buildingcould be thought of as a form of simulation.However, the term has become associatedwith a particular approach to decisionsupport. From a forestry researchperspective, the models we are interestedin are usually algebraic models ofbioeconomic systems.

Nowadays, people are familiar with buildingalgebraic models, through the widespreaduse of spreadsheets. Any spreadsheet –including one to derive the net presentvalue (NPV) of a project – is in effect analgebraic model. The spreadsheet allowsexperimentation with the model, such asasking ‘what if’ type questions throughvarying a model input and checking theperformance estimates.

be present which mimics the behaviour of theunderlying business or economic system forwhich the training is being conducted.

2 The alternative is to have studies conductedb y transdisciplinary indiv iduals, i .e.researchers with multiple skills across a rangeof disciplines. The obvious difficulty here is forindividuals to have sufficient depth across arange of disciplines to adopt a systemsapproach.

Simulation Philosophy and Methods 75

The need for a model

Sometimes it is possible to gain experienceabout the operation of a system byimposing various managements upon thesystem itself. Agricultural researchers havetraditionally used research stations and on-farm trials to evaluate pasture species,cropping systems, irrigation practices andso on. Many business managers wouldadmit to learning how to operate theirbusiness in the ‘school of hard knocks’.However, learning by one’s mistakes canbe a slow, costly and unpredictable way ofga in ing exper ience. Conduct ingexperiments on a model rather than the realsystem may be preferable on a number ofgrounds:

(i) the real system might not yet exist. Forexample, a company may wish to expand ordiversify its activities, perhaps investing alarge amount of capital. It would be usefulto have an idea of potential outcomes ofthis venture, and their likelihoods, beforecommitting capital to it.

(ii) performing experiments on the realsystem may be unacceptably timeconsuming or expensive. This would be thecase, for example, if management wishedto know the cash flows resulting over thenext 10 years if forestry plantations were tobe located in either of two alternative sitesor either of two different species or mixtureswere adopted. Once a model isconstructed, stand growth and economicperformance can be generated andcompared on the computer for the two sitesor species, in a matter of seconds of realtime.

(iii) performing experiments may be inimicalto the real system. For example, supposethe directors of a forestry company wishedto know whether the company could surviveunder various levels of borrowed finance. Itwould certainly be a risky proposition for thecompany to take out a very large loan andsee if the debt could be serviced. However,no harm would be done to the company if amodel were used to experiment withdifferent gearing ratios, and to determinewhat debt level leads to an acceptably lowprobability of ‘simular’ bankrupcy ortakeover.

(iv) The exercise of developing a model canbe in itself a valuable learning experienceabout a system. A model provides aframework for systematically organizingexisting information, and for assembling the(often subjective) knowledge of experts.The explicit nature of models makes ideasand assumptions more transparent, andplaces these under the scrutiny of morepeople. This in part explains why modelstend to be built through a series ofprototypes, evolving over time asquestionable assumptions and gaps inknowledge are exposed, more questionsare asked and more information isassembled.

Types of models

Typically, a bioeconomic system is modeledas a set of equations or relationshipsbetween variables, although often thealgebra of the model is implied rather thanmade explicit. Various classifications ofsymbolic models have been advanced. Forexample, models may be grouped as

(i) biological, economic or bioeconomic(ii) discrete or continuous(iii) single-period or multiperiod (or static or

dynamic)(iv) deterministic or stochastic.

A great deal of modelling has been carriedout in the natural sciences, includingforestry. For example, models have beendeveloped of physiological processes withintrees, and of the growth of individual treesand stands of trees. A biological model of aforestry stand could take into account soilmoisture, nutrient uptake, biomass growth,partitioning of nutrients between plant parts,and so on. The time step would normally bea period less than a year, perhaps as shortas a day. The performance measure mightbe total biomass or bole biomass produced.A financial or economic model of a forestrystand would typically take account ofexpenditure and of product sales, on ayearly basis, and seek to estimate the netpresent value of the forestry investment.3

Components of a biological and aneconomic model may be combined to

3 Some differences exist between financial and

economic models, as discussed in Modules 15and 16.


produce a bioeconomic model, whichpredicts profitability from the forestryinvestment, but taking into account theproduction processes rather than simplytaking final yield estimates.

Models in which time changes in acontinuous fashion are used extensively inmathematics and engineering, frequentlytaking the form of sets of simultaneousdifferential equations. Most management-oriented models are of discrete form, withvariables taking time steps of say one weekor one year.

Static models make no allowance forchanges in values of variables over time.They are thus not suited to examining long-term production processes such as forestrysystems. Multiperiod or dynamic modelsinclude specific representation of changesin levels of variables over time, such asgrowth, partition and decay of biologicalcomponents and accumulation of funds orassets.

In a deterministic model, all variables takesingle point or best-estimate values, and allrelationships between variables areassumed to be known with certainty.Stochastic models on the other handincorporate probability distributions for oneor more variables, or uncertain error termsin relationships between variables. Thus aforestry bioeconomic model could takeaccount of variable rainfall, crop hazards(e.g. pests, wildfires) and timber prices.With a stochastic model, the measure ofperformance could include both expectedpayoff and a measure of uncertainty. It isgenerally considered that a refinement inmodelling introduces a need to incorporateuncertainty. That is, as the modeling iscarried out in increasing detail, it becomesnecessary to introduce of stochasticvariables.

3. ELEMENTS OF MODELS

It is useful to introduce some furtherterminology to assist in the discussion ofmodels. A convenient classification(drawing on Naylor et al. 1966) is to dividemodels into the following elements:

(i) components or building blocks. Forexample, in a forestry model these could

include trees, soil and weather, or at a moreaggregate level nursery, plantation stands,workforce, and machinery and equipment.By definition, management is outside butexerting planning and control functions overthe system (and hence not a component ofthe model).

(ii) variables, the levels of which may bedetermined outside the system (exogenous)or within it (endogenous), or which maydescribe the state of the system (statusvariables). Exogenous variables may beunder the control of management (decisionvariables) or not controllable (environmentalvar iab les ). For a forestry operation,silvicultural treatments and associatedlabour and material inputs typically arecontrollable, while prices of labour andother inputs and of products generally arenon-controllable. In simulation experiments,exogenous variables form the inputs whichdrive the model. Those variables under thecontrol of management form the decision orpolicy instruments, the levels of which areadjusted in simulation experiments. Non-controllable exogenous variables may bemodelled as either fixed values (orpredetermined time series) or as probabilitydistributions from which random values aregenerated. The desirability of particularmanagement policies is assessed in termsof one or more model outputs or levels ofendogenous performance variables. In aforestry model, these might include, meanannual increment (MAI), and NPV from aplantation. Status variables record the stateof the system at each period in time.Examples for a plantation include mean treeheight and girth at various ages, and alsofinancial variables such as cash and debts.The growth rate of trees will depend ongrowth in previous time periods, sometimesreferred to as a ‘feedback loop’.

(iii) functional relationships. These indicatehow variables are interrelated, e.g. by linearor non-linear equations, and with or withouttime lags. Included are identities, which aretrue by definition (e.g. value = price xquantity sold) and operating characteristicswhich have to be estimated statistically orsubjectively. The latter might include therelationship between tree height and age.Functional relationships give the system itsunique behaviour, and needless to say thereliability of any systems model depends


vitally on how accurately the relationshipsare identified and estimated.

(iv) parameters: these are the coefficients ofthe operating characteristics, values ofwhich can only by estimated to within givenconfidence levels.

Any systems model may be summarized bythe following symbolic relationship:

Z = f(X,Y,S,A)

where Z is a set of performance variablesX is a set of policy variablesY is a set of environmental variablesS is a set of initial levels or statusvariables (including initial resource levels)A is a set of parameter values, andf signifies that a functional relationshipexists between the variables in thevarious sets (i.e. f represents the model).

4. THE STEPS IN A SIMULATION STUDY

Carrying out a simulation study is likecarrying out any other quantitative study(though sometimes a bit more complicated).The so-called scientif ic method isemployed, which means a series of stepsare performed in a logical fashion toachieve the overall task. The terminologyfor simulation steps varies between experts,but the following is a workableclassification.

(i) identification of the problem(ii) analysis of the system(iii) synthesis of the model(iv) programming the model to a computer(v) testing the model(vi) experimentation with the model(vii) interpretation of results, and reporting

to the relevant authority.

These steps are performed in the sequencelisted, except that there is usually somecycling between them, e.g. testing themodel may reveal the need to modify thestructure then revise the computer program.Each of the individual steps will now bediscussed briefly.

Identification of the problem

It is most important to identify clearly thestudy objectives in terms of the research or

managerial problem which is to beexamined. The nature of the model to bedeveloped will depend on the problem to beanalysed. Is the objective to understand thesystem or to prescribe managementpolicies? If the latter, who is responsible forthe system, and what are their goals, andwhat is wrong with present policies?

Analysis of the system

Once the problem is identified, the ‘systemsanalysis’ stage can be performed, in whichthe boundaries of the system, the relevantvariables and their interrelationships areidentified. This may involve drawing variouscharts or diagrams of the system.

Systems synthesis

The next step, of ‘systems synthesis’,consists of expressing the relationshipsbetween variables in symbolic form, andestimating the parameters of theserelationships. Because of the high degree offlexibility possible, it is difficult to lay downrules for this major step, though someguidelines can be given. Where possible,statistical techniques should be used toestimate relationships between variables.Distributions or random variables can beobtained by testing the goodness-of-fit ofalternative probability models using say thechi-squared test. Where historical data arescarce or are not considered relevant tofuture behaviour of the system, subjectiveestimation by people regarded as havingexpert knowledge about the system may bepreferable. It is usually recommended tostart with a relatively simple model, andgradually extend and refine it. To the extentthat sub-systems are suff icient lyindependent, the model should beconstructed in the form of a number ofrelatively self-contained modules, allowingthese to be programmed and testedseparately. Existing models of similarsystems should be examined for relevance,since it may be possible to obtain ideas oreven adapt modules from them.

Programming to a computer

Once a prototype version of the model isconstructed, computer programming cancommence, using a spreadsheet packageor – if the model is to complex for this –


using a computer programming languagesuch as Visual Basic, FORTRAN or C, or aspecialist simulation modelling languagesuch as Simile.

Testing the model

Having constructed a working model, it isnecessary to test whether this model is anadequate representation of the real system.There is far from general agreement on howsystems models should be tested, but aworkable approach is to divide testing intoveri f icat ion , val idat ion and sensi t iv i tyanalysis. Verification is the process oftesting whether the model takes its intendedstructure, i.e. whether the model is free oflogical errors and whether the computerprogram performs as intended. Validationexamines the broader question of whetherthe intended structure truely represents thereal system. Validation efforts often lead tofurther refinement of the model.

Once a model has been validated as far aspracticable, the effect of remaining errorson parameter estimates may be assessedthrough sensitivity analysis. If the purposeof the model is to identify optimalmanagement policies, errors in estimates ofperformance due to inaccurate parametervalues may not be of concern unless theylead to identification of inferior policies asoptimal. That is, we are not concerned withthe predictive ability of the model inabsolute terms so much as the model’sability to correctly rank alternativemanagement policies. For this reason, it isdesirable to include a sensitivity analysiswith respect to optimal values of decisionvariables. Sensitivity analysis usuallyinvolves adjusting parameter values bysmall amounts, and calculating varioussensitivity criteria. Sensitivity may beexpressed quantitatively in terms of‘elasticity’ of performance with respect toparameter levels, or elasticity of optimalmanagement policies with respect toparameter values. High sensitivities(elasticities) give cause for concern aboutthe reliability of a model.

Performing experiments

Once sufficient confidence has been gainedin a model, a variety of simulationexperiments may be conducted. Various

designs can be used for these experiments,as discussed later in the module. Many ofthe principles of experimental design asapplied in the physical sciences arerelevant to simulation experiments.Management policies may be specified interms of a single policy variable or anumber of variables. In practice, optimal (orat least desirable) levels of a number ofpolicy variables often have to bedetermined simultaneously. For example, aforest manager may wish to know whatfertilizer strategy, pruning and thinningregime and harvest schedule maximizesreturns from a plantation. In the language ofexperimental design, the decision variableswhich are identified are known asexperimental factors, and any combinationof levels of these factors (i.e. anymanagement policy) is known as atreatment. Measures of performance of thesystem are known as response variables. Acomputer run in which a number oftreatments are evaluated is known as asimulation experiment. If random variabilityis built into the model (i.e. if the model isstochastic), then it is necessary to evaluateeach treatment or policy under a number ofdifferent environments, i.e. to includereplication treatments in the simulationexperiment.

Experiments conducted on a computer alsohave important differences from real-worldexperiments. Three main sources ofdifference arise:

(i) compression of time. Because of thespeed of computing and the low cost ofcomputer time, it is usually possible toinclude a larger number of treatments and agreater degree of replication.

(ii) sequential processing. Traditionally,each of the treatments in a real-systemexperiment is evaluated at the same time.For example, in a plantation fertilizerexperiment, the complete experimentaldesign is decided, with all plots planted atthe same time, and fertilizer applied to eachon the same day, and girth and heightmeansurements made at the same times.On the other hand, because a computer is asequential processor, treatments areevaluated sequentially in a computersimulation experiment. This means that theperformance level for the first treatment is


known before the second treatment isevaluated, and the performance for the firstand second treatments are known beforeevaluating the third, and so on. Sequentialprocessing opens the opportunity to useinformation gained from earlier treatmentsto direct factor levels in later treatments,within the same experiment. ‘Optimum-seeking’ experimental designs arediscussed later in the module.

(iii) control over experimental variability. In acomputer simulation experiment, thevariability in the ‘environment’ is under thecontrol of the researcher. A random numbergenerator is used to produce numbersbetween zero and one, and these aretransformed to random observations fromthe distributions specified for the randomvariables. If the random number generatoris given the same seed for each treatmentthen the treatments are evaluated under thesame sequences of random numbers, i.e.under the same environments. This reducesrandom variability in response levelsbetween treatments relative to independentseeding (i.e. not re-seeding the randomnumber generator). The result is greaterpower to detect differences betweentreatments for a given sample size (numberof replicates).

Analysis and interpretation of computeroutput

Simulation experiments often generatereams of computer printout, and this outputmust be distilled and interpreted to a formuseable by managers in a decision-supportrole. Since experiments conducted usingstochastic simulation models do not provideexact results, estimated performance levelsshould be thought of in a confidenceinterval context.

5. EXAMPLE OF AN INVENTORYSIMULATION

Some of the concepts introduced above willnow be demonstrated with reference to asimple inventory model.

Example 1

A seedling nursery faces uncertain demandfor Mahogany seedlings. Recent experiencesuggests that demand can be approximated

by a uniform distribution with a range of5,000 to 10,000 seedlings in autumn, and3,000 to 5,000 seedlings in each of winter,spring and summer. Seedling productioncost is $300/1000 seedlings and the saleprice is 60c/seedling. Seedlings are grownto be ready for planting in autumn, but maybe retained until the following summer (afterwhich they must be discarded), the holdingcost being $120/1000 seedlings for eachquarter.

Simulate quarterly marketing of Mahoganyseedlings over four years, for a policy ofgrowing 20,000 seedlings per year.

The spreadsheet model

The simulation model may be developed asan Excel spreadsheet, as in Table 1. Thelevel of the decision variable (autumninventory level) and parameter values(prices, costs, demand limits) are listed atthe top of this sheet. For convenience, theyear runs from autumn (the preferredplanting time) through to summer. Eachquarter is represented by a row in thesimulation, and an estimate of net revenueis obtained for each quarter. Quarterlydemand is obtained by the formula

lower demand level + random number x(demand range)

where the random number is from a uniformdistribution in the range zero to one,obtained by the spreadsheet functionRAND() (see Appendix A). Quarterly salesare obtained as

MIN(inventory level, demand level)

and any quantity left unsold is transferred toinventory in the following quarter, exceptthat no inventory is carried forward fromsummer. Quarterly sales revenue isobtained as sales quantity multiplied byprice net of production cost. Quarterly netrevenue (or loss) is obtained as salesrevenue less holding costs. Average annualrevenue is obtained by summing quarterlynet revenues over the 16 quarters anddividing by four.

In this example, the mean annual netrevenue is about $3,980. Various inventoryproduction policies (treatments) could be


compared, by changing the ‘Annualproduction’ level, and re-running thesimulation to determine the mean annualnet revenue. In this way, the optimalproduction level could be determined.Because of the random element in demand,it would be preferable to increase thenumber of years over which sales aresimulated, relative to the four years in thisillustration.

While While separate random numberswould be generated for each inventorypolicy simulated, leading of confounding ofeffects of inventory policy and demandenvironment, this can be overcome by‘freezing’ the sequence of random numbers.

Table 1. Spreadsheet model for inventory simulation example

Parameters of the simulation model

Production cost ($/1000) 300Sale price ($/seedling) 0.6Holding cost ($/1000 seedlings/quarter) 80Autumn demand - lower limit (1000) 5Autumn demand - upper limit (1000) 10Off-season demand - lower limit (1000) 3Off-season demand - upper limit (1000) 5

Simulation of quarterly sales and net revenue

Year Quarter Inventory Random Demand Sales Holding Residue Sales Net (1000) number (1000) (1000) cost ($) (1000) rev. ($) rev ($)

1 Autumn 20.00 0.8501 9.25 9.25 0 10.75 2775 27751 Winter 10.75 0.4579 3.92 3.92 859.94 6.83 1175 3151 Spring 6.83 0.6244 4.25 4.25 546.67 2.58 1275 7281 Summer 2.58 0.3745 3.75 2.58 206.77 0.00 775 5692 Autumn 20.00 0.3107 6.55 6.55 0 13.45 1966 19662 Winter 13.45 0.5514 4.10 4.10 1075.72 9.34 1231 1552 Spring 9.34 0.5205 4.04 4.04 747.50 5.30 1212 4652 Summer 5.30 0.4510 3.90 3.90 424.21 1.40 1171 7463 Autumn 20.00 0.2979 6.49 6.49 0 13.51 1947 19473 Winter 13.51 0.6859 4.37 4.37 1080.85 9.14 1312 2313 Spring 9.14 0.6435 4.29 4.29 731.10 4.85 1286 5553 Summer 4.85 0.8500 4.70 4.70 388.14 0.15 1410 10224 Autumn 20.00 0.7927 8.96 8.96 0 11.04 2689 26894 Winter 11.04 0.8044 4.61 4.61 882.90 6.43 1383 5004 Spring 6.43 0.7076 4.42 4.42 514.19 2.01 1325 8104 Summer 2.01 0.3933 3.79 2.01 160.97 0.00 604 443

Mean annual net revenue 3979

6. VALIDATION OF SIMULATIONMODELS

As noted above, validation is a critical taskin development of a simulation model.While validation may be applied to theassumptions of the model, and to thevarious sub-models, in practice tests areusually concerned with the reasonablenessof outputs or predictions of the overallmodel. In this regard, it is often

recommended that statistical tests be usedto compare output of the model with that ofthe real system, where both have beengenerated under the same managementpolicies and environments. These testsexamine hypotheses of the general form

H0: outputs of the real system conform tothose of the modelH1: outputs of the real system differ fromthose of the model.


The output of complex (dynamic,stochastic) simulation models typicallyconsist of time series for performance,which can include various statisticalcontributions (e.g. trend, seasonal effects).Since the two series of outputs mayconform with respect to some propertiesand differ with respect to others, acomparison needs to be made of thevarious parameters (e.g. means, variances,autocorrelations and trends), and of overalldistributions. On the face of it, these testsshould allow a thorough check of all thestatistical properties of the two outputseries. However, in practice a number ofproblems arise with statistical validation.

Often, little output is available from the realsystem with which to compare output from amodel. All available data tend to be used inmodel construction, yet it is desirable for thedata for testing purposes to be independentof that used when constructing the model.

Important differences arise betweentraditional applications of hypothesis testsand their role in validation of systemsmodels. The motive behind traditionalstatistical testing is usually to demonstratethat a particular null hypothesis is false; thenull hypothesis is simply set up as a ‘strawman’. For example, when comparing twosample means the null hypothesis maystate that the two underlying populationmeans are equal, while the alternativehypothesis may state that they are unequal,i.e.

H0: µ1 = µ2 (mean predicted output = meanreal system output)

H1: µ1 ≠ µ2

and it may be possible to ‘prove’ that H0 isfalse in the sense of demonstrating that theprobability of µ1 equalling µ2 is negligible.When applying a validation test, the motiveis usually to accredit the model, i.e. proveH0 is true. Comparison of the t-statistic withcritical values provides an indication of theprobability that H0 is false. If this probabilityis low, the model can be declared invalid.But if the probability is high (i.e. above thechosen significance level) then all that maybe concluded is that the model has notbeen demonstrated to be invalid, beyondreasonable doubt. To state that the model

must therefore be valid is at best a tenuousstatistical inference.

Traditional hypothesis testing is designed tolimit the frequency of type 1 errors. The costof a type 1 error in terms of declaring a validmodel as invalid may not be great;unnecessary marginal refinements may bemade to the model. On the other hand, thecost of accepting as valid an invalid modelmay be substantial, both in terms of loss ofcredibility of the modeller if his or hercreation is later found to be misleading, andin dollar terms for users from makingincorrect management decisions based oninformation generated by the model. For agiven sample size, the lower thesignificance level the greater the incidenceof type 2 errors (accepting as valid modelswhich are in fact invalid). In other words, tomake the test procedure more stringent interms of ability to reject invalid models, it isnecessary to increase the significance level,say to 20%.

A further problem with statistical validationis that the assumptions underlying the testsmay not be appropriate. For example,consider the t-test on mean model and real-system outputs. A paired-sample t-test islikely to be more appropriate than a test ofindependent samples, when the outputshave been generated under the sameenvironments and managements. Thispaired-sample test assumes thatsuccessive differences are independent; inreality output series are often positivelyautocorrelated over time, with the result thatthe variance of differences between modeland real-system outputs may be seriouslyunderestimated. A modified form of t-testmay be used which takes account ofautocorrelations between paired differencesfor various time lags. Even this test requiresthe (relatively weak) assumption of second-order stationarity in differences, i.e. thecorrelation between differences dependsonly on the number of time periods betweenthem, and is constant over time.

As illustrated by the above discussion, theapplication of traditional statistical tests tomodel validation is not a simple matter. It isno wonder that most model testing hasbeen subjective in nature, such as graphicalcomparison of time series of model andreal-system output, and assessment of the


reasonableness of model output by peoplejudged as experts with respect to thesystem.

Another reason for concern about statisticalvalidation procedures is that they are beingapplied to a moving target. Simulationmodels typically evolve over time. A needfor a model is identified, and a prototypemodel is developed and implemented tofulfil this need. Often, limitations of themodel are recognized, and the model isfurther refined. As well as applying themodel for its initial design purposes, otheruses are often discovered, and extensionsof adaptions of the model are undertaken tomeet these other purposes. As aconsequence, it is generally conceded thatthere is no clear finish to the testing of asystems model. Rather, confidence isgradually built up in a model over time astests are performed, new information isobtained, and new versions of the modelare produced.

7. THE DESIGN OF SIMULATIONEXPERIMENTS

Suppose a simulation experiment involvestwo decision variables or experimentalfactors. Combining each level of one factorwith each level of the other leads to fullfactorial design. The full factorial design isconceptually simple and well suited forsimulation experiments in which there areonly two or three factors, each taking only afew levels. In general, if there are m factors,each taking n levels, then there are mndistinct treatments. If m is more than two orthree, the number of treatments canincrease dramatically, and the experimentmay become unmanageably large, evenwhen the number of replicates is small.Alternatives to the full factorial design, suchas the fractional factorial and centralc o m p o s i t e designs, allow for somereduction in treatment numbers for given mand n.

An alternative to designs which arespecified in advance of the simulationexperiment, is to include a set of rules in thecomputer program to select factor levelsduring the experiment on the basis ofinformation gained from earlier treatments.The simplest of these designs is steepestascent (or steepest descent). This is an

iterative procedure in which the slope of theresponse surface is estimated with respectto each experimental factor, then all factorlevels are adjusted so as to achieve themost rapid increase in performance. Moresophisticated hill climbing designs may beadopted. These will involve programmingthe procedure for allocation of treatmentson the basis of progressive performanceduring the experiment, or using sub-routines which have been developed byothers for this purpose, which are availablefrom a variety of sources.

Yet another alternative is random search, inwhich treatments are chosen simply byselecting the level of each factor at random.It is necessary to place upper and lowerbounds on factor levels in order to confinethe search area, based on prior knowledgeof the system and exploratory computerruns. Factor levels are then sampled fromuniform distributions over these ranges. Ifthe number of treatments evaluated issufficiently large, then there is a highprobability that near-optimal policies will beidentified. Often, computer output isobtained only for those treatments for whichperformance exceeds a specified thresholdlevel. Random search is relatively easy toprogram on a computer. This experimentaldesign procedure is particularly useful whenthe levels of policy variables are discrete,and cannot be approximated satisfactorilyby continuous variables. This design is alsoused where activities form complementaryor mutually exclusive sets. Moresophisticated random search routinesinclude provision for adjustment ofprobabilities or heuristic learning during theexperiment; this may involve narrowing thesearch ranges or departing from uniformdistributions.

8. SIMULATION IN PERSPECTIVE

Computer simulation relies on a overallsystems philosophy, which asserts thatcomponents of a system should not beviewed in isolation and in a reductionistmanner, but rather in terms of the complexinterrelationships and interactions betweenvariables. An attempt is made to analysecarefully all the components of the systemand their interrelationships, and torepresent these in a simplified and abstractmodel. This model is used to conduct


experiments in which the behaviour of thesystem is simulated for variousenvironments and management policies.

Simulation has proved to be a powerful andversatile approach to improving theunderstanding of systems and determiningnear-optimal management policies. Anumber of advantages are afforded overanalytical techniques for problem solving,especially with regard to the handling ofuncertainty, multiple goals, non-linearrelationships and other real-worldcomplexities. On the other hand, somedifficulties arise in simulation studies. Thesuccessful application of this techniquerequires considerable experience withmodelling, an understanding of statisticaltechniques and their limitations, andsubstantial human and computingresources. It is not usually possible to makeuse of a recognized model structure orexisting computer program, and particularattention must be directed to validation ofsimulation models. A variety of designs maybe exploited in simulation experiments,including optimum-seeking designs whichtake account of the special features ofcomputer-based experimentation.

REFERENCES

Dayananda, D., Irons, R., Harrison, S.,Herbohn, J. and Rowland, P. (2002),Capital Budgeting: Financial Appraisal ofInvestment Projects, Cambridge UniversityPress, Cambridge.

Shannon, R.E. (1975), Systems Simulation:the Art and the Science, Prentice-Hall,Englewood Cliffs.

APPENDIX A: GENERATION OFRANDOM VARIATES

A critical step in any stochastic simulationmodel is to generate values of random(stochastic) variables or variates. A widevariety of forms of probability distributionshas been recognized as reflecting thebehaviour of particular random variables.The starting point for producing (orgenerating) values from any form ofprobability distribution is to use a randomnumber from a known distribution. Inpractice, this is usually a number from auniform distribution over the range zero to

one, for example as provided in the Excelfunction ‘= RAND( )’.4

Generating uniform variates over aspecified range

Perhaps the simplest form of continuousprobability distribution to use in simulationmodels is the uniform distribution. Here a isthe smallest value and b the largest valuethat the variable is expected to take. Sincethe area under any probability curve offunction must be unity, and the range ofvalues of the variable is b-a, the height ofthe curve must be 1/(b-a).

A value y from a uniform distribution can begenerated by taking a random number rthen applying the expression

y = a + r(b-a),

where the random number is be obtainedusing the spreadsheet function ‘=RAND( )’,i.e. the lower limit plus the random numbertimes the distance between the upper andlower numbers. For the first demandestimate in Table 1, this becomes

y = 5 + 0.8501 (10-5) = 5 + 4.2505 = 9.25

To obtain a series of values on the randomvariable, simply repeat this process anumber of times (on a computer).

Generating values from a discretedistribution

In essence, this involves expressing thedistribution in cumulative form, and thenassociating random numbers withcumulative probability ranges. For example,suppose timber price (in $/m3) can be

4 The procedure used on the computer is

typically to divide a very large number byanother large divisor, and to take theremainder as a fraction of the divisor (henceyielding a number between zero and one).As well, the remainder is multiplied by a thirdlarge number, and the product divided by theoriginal divisor. By repeating this process, aseries of random numbers can be generated.The numbers are sometimes called‘pseudorandom’ because given the sameinitial three large numbers (and samecomputer accuracy), the same series willalways be produced.


represented by the discrete values 30, 40and 55, with probabilities 0.3, 0.5 and 0.2.Cumulative probabilities are then 0.3, 0.8and 1.0, and ranges of random numbers asassigned 0 to 0.3000, 0.3000 to 0.8000 and0.8000 to 1.0000.

Now suppose the sequence of randomnumbers 0.2488, 0.8324, and so on isobtained. The first of these numbers falls inthe first cumulative probability range hencea timber price of $30/m3 is generated. Thesecond number falls in the range 0.8000 to1.0000 hence a timber price of $55/m3 isgenerated. Proceeding in this way, asequence of price ‘observations’ can begenerated for which the relative frequenciesapproximate the discrete probabilities,providing the sample is sufficiently large.

Generating normal variates

The normal distribution is widely recognizedin statistical methods as a commonlyoccurring or approximated distribution.There are several methods for generatingrandom normal variates. The simplest is totake the sum of 12 random numbers from auniform 0-1 distribution (e.g. a computerrandom number generator), subtract six,then multiply by the target standarddeviation σ and add the target mean µ. Thereason this works will not be explained

here, but is associated with the CentralLimit Theorem.

Generating values from a triangulardistribution

A convenient distribution for fittingsubjectively to random variables is thetriangular distribution, defined in terms ofthe most pessimistic, most likely (modal)and most optimistic values. For example,for timber price these might be 30, 40 and55 (in $/m3). If these points are called a, band c, and a distance parameter is definedas

d = (b-a)/(c-a),

then using a random number r a value fromthis distribution y can be generated as:

if r ≤ d then y = a + √r (c-a)(b-a)

if r > d then y = c - √(1-r)(c-a)(c-b).

Procedures or ‘recipes’ can be developedalong similar lines for sampling from anumber of other continuous or discreteforms of probability distributions. Somecomputer packages in fact have thesesampling procedures built in, such as therisk simulation package @RISK.

85

9. Introduction to Simile

Jerry Vanclay1

This module provides an introduction to Simile, a powerful modelling language with someinnovative features. Simile is being developed by Dr Robert Muetzelfeldt, at the University ofEdinburgh, with suggestions and testing by a number of users. The package is still underdevelopment, so it is getting better all the time, but it is already a sophisticated and stablepackage. Documentation is available at http://www.ierm.ed.ac.uk/simile, including some thatwas prepared when Simile was known by its previous name, AME (Agroforestry ModellingEnvironment). The latest version of the software can also be downloaded without chargefrom this Web site, for PCs with Windows 9x or computers running Linux. Dr Muetzelfeldthas long had an interest in efficient representation of ecological information, and hisdevelopment of Simile was due in part because of his concern that with many models, thedocumentation, diagrams and computer implementation diverge. So he set out to build amodelling platform where the diagram was the model and the documentation, and he’s closeto achieving this goal. He’s also provided the possibility to create a model that runs on acomputer without the need for computer code or mathematical equations. However, there’sno free lunch, and it is necessary to learn some of the standard notation used in systemsdynamics to be able to use this package. This module provides introduces the systemsdynamics notation, and then provides a worked example of developing a Simile model torepresent a personal bank account. Some of the unique features of simile are then examinedwith reference to a simple forestry model. Finally, two other examples of forestry models arepresented briefly.

1. SYSTEMS DYNAMICS NOTATION

Although an effort will be made to keep the jargon to a minimum, the user does need to learnsome basic systems dynamics vocabulary to get started. In particular, a commonunderstanding is needed of the following terms: compartment, flow, cloud, variable, influenceand equation.

A compartment• is represented by a rectangle• represents amount of some substance (or occasionally non-substance things like height

or position)• should be labelled as object:substance, e.g. tree volume with units cu_m/ha (or no./ha,

kgs, etc)• unlike real compartments, can go negative, and has infinite capacity• cannot receive an influence arrow – it changes only as a result of flows• rate of change is the net effect of all inflows minus all outflows• two connected compartments must have same substance, same units• can only contain one substance.

A flow• usually corresponds to a process, and must flow into or out of a compartment• has units that correspond to compartment units per unit time• can be negative (flow in reverse direction)• may be one of several flows between two or more compartments, in either direction• ideally represents a single process, so it is preferable to have one flow for each

separately-analysable process.

1 This module is a modification of lecture material prepared for students at Southern Cross University

in March 2001.


A cloud• is like a compartment (at start or end of a flow), but when its value is irrelevant, unknown,

or unspecified• corresponds to ‘the outside world’, and thus cannot receive or be the source of an

influence arrow• in Simile, clouds are created automatically when a flow starts nowhere or goes nowhere.

A variable can be• a parameter that is ‘constant’ during a simulation (e.g. a coefficient in an equation)• an intermediate variable, which can be both the source and recipient of influence arrows• an output variable used only for reporting on model behaviour• an exogenous variable (i.e. an ‘external variable’, just a function of time) that influences

the model, but is not influenced by it (e.g. climatic factors).

An influence• usually corresponds to a link in an ‘influence diagram’• captures the idea that something affects something else• formally, represents the fact that one term is used in the calculation of another• can start from a compartment, flow or variable, and go to a flow or a variable (but not to a

compartment).

Finally, it should be recognised that an equation may appear within a compartment, a flow oran influence. The equation• says how a value for a variable is to be calculated• often represents the relationship between one quantity (Y), and one or more influencing

quantities (X1, X2, etc)• uses standard algebraic expressions and conditional elements• may include a sketched graph function or a tabulated function

In Simile, a single ‘equation window’ is used for all quantities (incl. parameters and initialcompartment values).

2. GETTING STARTED

When the new version of Simile isopened, a window like the oneopposite is obtained. One of thenice features of this version isthat if you’re not sure what abutton does, hold the mouse overit, and an explanation will appear.

3. BUILDING A SIMPLE MODEL OF A BANK ACCOUNT

To build a simple simile model which will simulate changes in the level of savings over timein a bank account, follow the steps listed below:

1. Click on the COMPARTMENT tool (left-most rectangle) in the top toolbar.2. Click in the centre of the ‘Desktop’ window to deposit a compartment symbol there.

Note that it is automatically labelled ‘comp1’.3. Click on the FLOW tool (straight thick arrow) in the toolbar.4. Drag a flow into the compartment symbol in the Desktop window. ‘Drag’ means:move the mouse onto the Desktop window, depress the mouse button

where the flow should begin, move the mouse to touch the compartment symbol (itshould turn green), and release the mouse button. Note the flow is automatically labelled‘flow1’

Introduction to Simile 87

5. Click on the INFLUENCE ARROW tool (curved thin arrow) in the toolbar.6. Drag an influence arrow from the

compartment symbol to the flowsymbol in the Desktop window.This indicates that the interestpayments depend on (are influencedby) the bank balance.

Your diagram should now look like theone on the previous page.

7. Click on the POINTER tool (arrowinside a rectangle) in the toolbar.

8. Click on the compartment symbol inthe Desktop window.

9. Use the Delete and/or backspace keys to remove the label ‘comp1’.10. Type in a name for the compartment, e.g. balance.11.Click on the flow symbol in the Desktop window.12. Use the Delete and/or the backspace keys to remove the label ‘flow1’.13. Type in a name for the flow, e.g. interest. We’ve finished defining the structure of the model, but have not initialized it; that’s why it

is in red.14. Double-click on the compartment symbol.15. Enter the value 100 in the ‘Equation’ box. This states that the initial bank balance is $100.16. Click on OK. Notice the compartment is now black, indicating that it has been initialized.17. Double-click on the flow symbol (the double-triangle bit).18. Enter the expression 0.1*bank_balance in the ‘Equation’ box. This expresses the fact that the annual

interest is 10% of the balance. Noticethat Simile offers help with functions ifyou place the mouse over them. Alsonote that you must type names exactlythe same way as they appear in theequation box – it is often convenient todouble-click on variable names ratherthan typing them.

19. Click on OK. The model is now completely set up,

and ready to run. It should look like thisthe window opposite.

4. RUNNING THE MODEL

This comprises the following steps:

1. Click on the word ‘File’ in the Desktopmenu bar.

2. Select the ‘Run’ option in the File menu,then select the ‘In tcl’ sub-option.Note that a run control dialogue windowappears.

3. Re-arrange the windows to avoidoverlap (drag by top title bar).

4. Click in the ‘Update every’ edit field.5. Delete the ‘0.1’ and replace it by ‘1’ (without the quotes).6. Change the value 100 to 10 in the ‘Execute for’ edit field.7. Open the ‘I/O tools’ menu in the Desktop window.


8. Select the ‘Add tool’ option, then the ‘Plot value against time’ sub-option.Note the little window that pops up telling you what to do now.

9. Click on the compartment in the Desktop window. This sequence specifies that you want to plot values for your bank balance against time.10. Again, re-arrange the windows if necessary.11. Click on ‘Start’ in the Run Control dialogue window. The model should now simulate according to the values in the run control window: 10

years, displaying the results every year, and calculating interest once per year. Observethe results in the window titled ‘Current value for balance’, and note that the actual valueof balance is displayed in the box in the bottom-left corner.

12. Try re-sizing the graph window; scrolling up and down; and clicking on the ‘Var’ and‘Time’ buttons.

5. CHANGING THE MODEL

This may be achieved in a number of ways.

By changing the equation

1. Click in the Desktop window to return to the model-design system.2. Change the interest rate from 10% to 15%. You should realise that this involves selecting the pointer tool, double-clicking on the flow

symbol in the Desktop window, and editing the interest equation, replacing 0.1 by 0.15.3. Run the model again, and compare the results with the previous results.

By changing a compartment value during a run

1. With the graph still on the screen, drag the slider near the y-axis down to 50.This simulates the effect of removing some money from the bank account, leaving just$50.

2. Click on ‘Start’ to continue the simulation. Note how the graph drops down the climbs back up again. Why do you think the line

drops down at an angle rather than dropping down vertically?

With an explicit symbol on the model diagram

1. Click on the VARIABLE tool (circle with a cross inside it).2. Click near the flow symbol in your model diagram to place a ‘variable’ symbol there.3. Re-label this symbol as ‘rate’.4. Double-click on this symbol, enter the value 0.15 in the ‘Equation’ box, and click on OK.5. Draw an influence arrow from the variable symbol to the flow symbol.6. Replace the expression for interest (0.15*interest) by rate*balance7. Run the model again. The results should be the same as before. You have now just changed the diagrammatic

appearance of the model, not its underlying mathematical structure. In general, it’s up toyou whether you show model parameters explicitly or hide them in expressions

By adding a slider to the model

1. Click on the POINTER tool.2. Double-click on the VARIABLE symbol

‘rate’ to open the equation window.3. Click the button ‘Input parameter’ and

supply the Min and Max.4. Click on the ‘Run’ tool to re-run your model. Notice the appearance of a slider that enables you to vary the interest rate at any time

during simulation.


By adding more variables to the model

As an exercise, add features to your model to represent regular deposits (e.g. your salary)and withdrawals (e.g. loan repayments) to your model.

6. SAVING THE MODEL

To save the model, take the following steps:

1. Click on the SAVE tool or select Save under File in the Desktop menu bar.2. Find the required drive (normally A, C or your home drive) and select the required

directory, as you would do in any Windows program.3. Type the name for the file (e.g. model01). Simile will automatically add the extension ‘.sml’

to the file name. Older versions may add ‘.ame’ or ‘.sim’.4. Click on ‘Save’.

So far, the standard systems dynamics notation has been demonstrated. Now some moreadvanced features unique to Simile will be introduced.

7. SOME UNIQUE FEATURES OF SIMILE

Simile offers some special features that make it uniquely amenable for modelling andteaching. One of these is that the diagram is the model and the documentation, and offers acomprehensive and clear overview of a model, even to the uninitiated. Another uniqueaspect is the concept of multiple-instance submodels, including several special instances ofsubmodels, both of which are now discussed.

A forestry example of diagram-is-model-is-documentation

Take a quick look at the following four illustrations of the same model implemented indifferent ways (BASIC, Excel, System dynamics, Simile). You don’t need to understand allfour approaches; these are presented to emphasise some of the strengths of Simile.

1) Denis Alder’s size class model in Basic

From Alder, D. (1995), Growth Modelling for Mixed Tropical Forests, p. 80.2) The same model in Excel

Diameter Class

SUB stproj(st())‘Updates a stand table of tree diameters by species using simple stand projection‘Get array bounds. Lower bound should be 1.nsp% = UBOUND(st, 1)ndc% = UBOUND(st, 2)DIM st0(ndc%)FOR j% = 1 TO nsp%

‘Keep old stand table as st0 to avoid overlap while updatingFOR k% = 1 TO ndc%

sto(k%) = st(j%, k%)NEXT k%‘Get first class update using ingrowth functionst(j%, k%) = ingrowth(j%) + st0(k%) * (1 – growth(j%, k%) – death(j%, k%))‘Get updates to other classesFOR k% = 2 TO ndc%

st(j%, k%) = st0(K%-1)*growth(j%, f%-1) + st0(k%)*(1 – growth(j%, k%) – death(k%, j%))NEXT k%

NEXT j%END SUB


From (cm) 5 10 20 40 60 80 100 120

To (cm) =C3 =D3 =E3 =F3 =G3 =H3 =I3 +

Incrementcm/yr

0.686 0.77 0.8 0.74 0.625 0.485 0.327 0.24

Outgrowth%/5yr

=5*B5/(B4-B3)

=5*C5/(C4-C3)

=5*D5/(D4-D3)

=5*E5/(E4-E3) =5*F5/(F4-F3) =5*G5/(G4-G3)

=5*H5/(H4-H3)

0

Mortality%/5yr

0.14 0.049 0.049 0.049 0.049 0.049 0.049 0.049

Harvest % 0.5

Year --------------------------------------------------- Trees/km2 ------------------------------------------------------------------

0 500 321 232 146 164 136 64 14

=A12+5 =B12*(1-B$6-B$7)+87

=B12*B$6+C12* (1-C$6-C$7)

=C12*C$6+D12* (1-D$6-D$7)

=D12*D$6+E12* (1-E$6-E$7)

=E12*E$6+F12* (1-F$6-F$7)

=F12*F$6+G12* (1-G$6-G$7)

=G12*G$6+H12* (1-H$6-H$7)

=H12*H$6+I12* (1-I$6-I$7)*I$8

=A13+5 =B13*(1-B$6-B$7)+87

=B13*B$6+C13* (1-C$6-C$7)

=C13*C$6+D13* (1-D$6-D$7)

=D13*D$6+E13* (1-E$6-E$7)

=E13*E$6+F13* (1-F$6-F$7)

=F13*F$6+G13* (1-G$6-G$7)

=G13*G$6+H13* (1-H$6-H$7)

=H13*H$6+I13* (1-I$6-I$7)*I$8

=A14+5 =B14*(1-B$6-B$7)+87

=B14*B$6+C14* (1-C$6-C$7)

=C14*C$6+D14* (1-D$6-D$7)

=D14*D$6+E14* (1-E$6-E$7)

=E14*E$6+F14* (1-F$6-F$7)

=F14*F$6+G14* (1-G$6-G$7)

=G14*G$6+H14* (1-H$6-H$7)

=H14*H$6+I14* (1-I$6-I$7)*I$8

From Alder, D. (1995) Growth Modelling for Mixed Tropical Forests, p. 78.

3) The same model in a systems dynamic representation

This is an exact re-implementation of the spreadsheet model given above. It is implementedin Simile, but uses only standard system dynamic notation used in industry-standardpackages like Stella (from Muetzelfeldt’s submodel library).

Initial values:n1...n8 = 500, 321, 232, 146, 164, 136, 64, 14Equations:recruit = 87mort1...mort8 = 0.14*n1, 0.049*n2...0.049*n8grow1...grow8 = 0.686*n1, 0.385*n2, 0.2*n3, 0.185*n4, 0.156*n5, 0.121*n6, 0.082*n7, 0*n8harvest = 0.5*n8

4) The same model in Simile, using a multiple-instance submodel


This is the same model, but using a multiple-instance submodel to represent a generic size-class. It has eight instances, since the published model has eight size-classes.

Initial values:recruitment = 87N_trees =element([500,321,232,146,164,136,64,14],index(1))Equations:mortality = if index(1)==1 then0.14*N_trees else 0.049*N_treesoutgrowth =element([0.686,0.385,0.2,0.185,0.156,0.121,0.082,0],index(1))*N_treesharvesting = if index(1)==8 then0.5*N_trees else 0upgrowth = [outgrowth]ingrowth = if index(1)>1 thenelement([upgrowth],index(1)-1)else 0

Which representation is clearer and easier to follow? Why?

Submodels

One of the powerful features you’ve justseen is the ability to create a submodel.

Submodels can be used to• visually separate parts of the model• control the appearance of a complex

model• enable separate saving and loading• move parts of a model around• create multiple instances• specify relations between objects• allow parts of the model to exist

conditionally• s p e c i f y d y n a m i c a l l y - v a r y i n g

populations.

Using submodels to model individual trees

Let’s build an individual tree growth modelwith Simile, starting with a simple model ofthe growth of a single tree. It’s much likethe bank-balance model illustrated inSection 3, but here the compartment iscalled diameter, and the flow is calledgrow. Suppose there is no establishedequation for tree growth, and that themodel is to be constructed using a graphto represent growth. Double-click on theflow and click on the ‘Sketch graph'button. The Sketch graph dialoguewindow appears, with its default straightline and 0,0 to 100,100 coordinates. Sincegrowth is to be predicted from diameter (in


cm dbh), 0–100 is probably OK on the horizontal axis, but a more appropriate maximum isneeded on the vertical axis (let’s chose 2, a reasonable maximum when growth is expressedin cm/year). Next, ‘draw’ the response curve with the mouse, either by dragging, or byclicking on the grids. When this is finished, press Enter to return to the equation dialogue.Notice that "graph()" has appeared in the equation box; this must be completed by double-clicking on the local name diameter to complete the equation, or Simile won’t know how thegraph should be used.

Let’s also compute the mean annualincrement (MAI) and compare the current(CAI or growth) and MAI curves. Create avariable called MAI, and drag an influencearrow from diameter to MAI. Double-clickon the variable MAI and enter the formulafor MAI: "if time(1)==0 then 0 elsediameter/time(1)". The function time(1) is asystem function that returns the currentsimulated time. Notice that Simile uses thenotation "==" to indicate a test of equalityto discriminate it from the common usageof "=" indicating "takes the value of".

Remember how to run the model? (If not,see Section 4 above). An "I/O tool" isneeded to illustrate the CAI and MAIcurves. The screen opposite illustrates theplotter. To use this, select "I/O tools",""Add tool" and "Plotter".

Now use the submodel tool to include allthese details in a submodel. Suppose it isdesired to model a plantation with 1111trees (3 x 3 metre spacing). Click on thepointer tool and double click on the whitespace inside the model. A submodeldialogue box should appear, and you canspecify the number of instances (1111),choose a colour if you wish, and addsome comments. When this is done, thesubmodel diagram will have changed,now appearing like a pack of cards, torepresent multiple instances.

If the screen does not appear like this,use the move tool and the zoomfit tool totidy it up a bit.

Now try to compute the basal area. Usethe variable tool to create a variable calledBA outside the submodel, and drag aninfluence arrow from the diameter compartment to the BA variable. Double-click to open theequation dialogue. Notice that the influence arrow makes an array of diameters available tothis variable. Double-click the sum function, double-click the local name [diameter], and type^2 to square it. Next, move the cursor to the end and divide by 12732, which is equivalent to40000/π. π is not yet available as a function in Simile. The figure 40000 converts diametersquared into radius squared, and cm2 into m2. Try running the model, and explore some ofthe I/O tools.


Let’s adapt this model for uneven-aged forests, and regeneration. Double-click on thesubmodel again, and chose the Population button. When this is done, you’ll notice that the‘pack of cards’ looks ‘slightly untidy’, signifying that the number of instances can vary. Apopulation submodel also activates several new symbols:

• c r e a t i o n tells Simile how manyinstances exist at the start of asimulation.

• reproduction creates ‘offspring’ for eachexisting instance; good for modellingregeneration in species with short-livedseed, and for modelling animalpopulations (the symbol is an egg).

• immigration creates new instanceswithout regard to the existing instances;good for modelling regeneration fromseed banks, of species with widely-dispersed seed, and animal migration.

• loss provides a way to model mortality,emigration, etc. (the symbol is a swordor a cross).

Alder’s model will now be represented as apopulation model, so it can be comparedwith the illustrations above. A few numberswill need to be changed, to adopt an areaof one hectare (not a square kilometre),and to use annual time steps instead of 5-year steps. Here’s the model. Does itmake sense to you? One thing that itportrays clearly is that growth and deathdepend on tree size (and tree size only),and everything else remains constantthroughout the simulation.

One further embellishment will be made,namely adding x- and y- coordinates foreach tree so that a ‘lollipop’ diagram canbe created. Module 10 demonstrates how to use these Simile functions, developing aforestry example model. Also, some tutorials on Simile can be accessed on the Web site:http://www.ierm.ed.ac.uk/simile.


Simile is a powerful modeling and programming medium, which has proven valuable inmodeling forestry systems. As with any computer programming language, or softwarepackage, there are a number of overheads involved in gaining familiarity with simile. Thepackage is available as freeware, so anyone can obtain a copy and start using it. It isobviously necessary to learn the systems dynamics notation, and to become familiar with themenu items of the simile screen. After that, working through example applications such asthe bank account model presented here provides initial confidence for the user. To becomea competent user, it is essential to experiment with the Simile package, including the menuand sub-menus, and find out by trial and error what operations can be performed. Handyhelp facilities are available to assist this experimentation.

95

10. Simile Revisited: The One-Minute Modeller

Jerry Vanclay

The Simile programming language provides a powerful and relatively easy to use mediumfor developing models and simulating the behaviour of forestry systems. This is a highlyvisual approach to modelling, in that the flow diagram is in effect the computer program. Thismodule provides a simple introduction to use of the Simile programming language forpotential users, which has been developed to provide an initial understanding of theprogramming features and steps in classes and for workshops.

1. INTRODUCTION

A variety of computer programminglanguages have been used for developingmodels to simulate forestry biologicalprocesses, stand growth, processing,marketing, human involvement and other‘systems’ in forestry. Over time, thesemodel development media are becomingmore powerful and easy to use. The Simileprogramming language, originally known asAME (Agroforestry Modelling Environment),has been developed by researchers at theUniversity of Edinburgh and elsewhere,during the last five years, with a specificfocus on forestry modelling. This softwarecan be downloaded free of charge from theweb site http://www.simile.co.uk. Otherlanguages such as Vensim and Stella offersimilar capabilities.

A best-selling book was recently writtencalled ‘The One-minute Manager’. In thesame spirit, this exposition of Simile can bethought of as ‘The One-minute Modeller’.That is, it is a tutorial presentation designedto ‘break the ice’ for researchers interested

in using Simile, by demonstrating some ofthe main features of the language in aneasy-to-understand manner.

The example used for this expositionconcerns growth of a forestry stand.However, the principles apply equally wellto other application areas. A number ofcomputer screen images will be presented,to indicate menu options and appearance ofthe visual model throughout thedevelopment steps. While there is muchtalk about models that are big andexpensive, this paper demonstrates thatmodelling need not be slow, expensive, orexcessively demanding of data.

2. DEVELOPING THE MODEL

When Simile is first opened, the screenappears as below. To save space, allscreen images have been reduced slightlyin size in this paper, and unfortunately onlyblack and white images can be providedhere. The purposes of the various icons atthe top of the screen will be explainedduring presentation of this application.


A starting point is to introduce an individualtree, and to set out what is known abouttrees as a submodel. Select the submodeltool by clicking on the icon showing a boxwith rounded corners, and then click within

the modelling space at the desired locationfor this submodel. The submodel has beenrenamed ‘tree’, using the pointer tool in thetoolbar.

For this example, the only property of treesof interest is their diameter. Thus acompartment is added to the modeldiagram to represent tree diameter. Click onthe compartment tool (the rectangle icon at

top left) and move the mouse to the middleof the modeling space. This compartmenthas been renamed ‘diameter’, again usingthe pointer tool.

Simile Revisited: The One-Minute Modeller 97

Double-click on the compartment symbol toobtain the following dialogue box. Enter anexpression of say 1 in the Equation box.

Here the diameter is specified incentimeters.

Trees grow, so it is necessary to allow thediameter to increase. This is achieved bycreating a flow into the compartment calleddiameter. The flow tool is the icon showing

an arrow with an engineering symbol for atap or regulator. The flow has beenrenamed as ‘grow’.


Tree growth may depend on many things,but here only size dependent growth will bemodelled. This calls for drawing aninfluence arrow from ‘diameter’ to ‘grow’ in

the model diagram. Click the influence toolwith a curved thin arrow in the toolbar anddraw an influence arrow from thecompartment symbol to the flow symbol.

Next, a growth relationship must besupplied. By double-clicking on the flow

icon called ‘grow’, the following dialoguebox is obtained:


Not everyone is comfortable with equations.To avoid the maths, the relationship may besupplied as a hand-drawn sketch. Byclicking on the ‘Sketch graph’ button, agraph window is obtained. A graph of any

shape can be sketched with the mouse onthe graph window. The default is tointerpolate on 20 line segments, but thiscan be varied with buttons for "Less" and"More" X-axis resolution.

Notice that start and end values for both thex and y axes can be provided. Therelationship can now be accepted bypressing the ‘Enter’ key. Simile knows thatthe y-axis represents the growth rate, but it

is necessary to specify in the ‘Equation’ boxthat the x-axis represents the diameter of atree. Having done that, press the ‘OK’button. A model has been built for the firsttree for the moment.


Now, the objective is to model not just onetree, but a forest. To do this, double-click onthe boundary of the tree submodel. A newwindow for the tree submodel appears. Inthis window, select ‘Properties’ under ‘Edit’on the menu bar. Next, specify the numberof instances of the tree submodel in the

‘Dimensions’ box. If the aim was to model aplantation, a set of say 1000 trees may begenerated. However, here the interest is ina natural forest, in which the number oftrees varies continually. Thus, just checkthat the population option has been chosen,with a dot in the ‘population’ circle.

When the ‘Done’ button is pressed, it will benoticed that the tree submodel is no longerrepresented as a single instance, but hasdouble lines around the bottom-right andtop-left corners like an untidy pack of cards.This is the Simile notation for a population.Now, in order to draw some diagrams of theforest, it is necessary to add variables

representing the x- and y-coordinates. Clickon the variable tool (a circle icon with across in it), and then click in the modeldiagram to deposit a variable symbol.Repeat clicking in the model diagram to addadditional variables. Two variables havebeen renamed as x and y.


By double-clicking on each variable symbol,the following dialogue box is obtained. It isnecessary to specify the initial value foreach tree. For simplicity, random numbers

between 0 and 100 will be chosen for eachtree. Enter the expression rand(0,100) inthe Equation box. The location, size andgrowth of trees have now been specified.

Since these trees are not immortal, it isnecessary to specify a mortality function.The mortality tool is used to specify thedestruction of instances of a populationsubmodel. Click on the mortality tool (theicon with a sword in the toolbar), and thenmove the mouse to within the envelope ofthe population submodel to deposit the

mortality symbol in the submodel. Noticethat the icon with a cross instead of a swordturns up. The next version of Simile will usea hatchet for both the tool bar and the icon,but for the moment, Simile users just haveto accept both the sword and the cross andrecognise that they mean the same thing.


By double-clicking the mortality symbolrenamed as ‘die’, the following equationdialogue window appears. Suppose a 1%

mortality rate is assumed, irrespective ofsize, competition or any other factor. Toexpress this, enter 0.01 in the Equation box.

Next, to introduce new trees into ourpopulation as seedlings, while making themdensity-dependent, it is necessary tointroduce the notion of crowdedness. Placea variable symbol outside the population

submodel and rename it ‘crowdedness’.The level of crowdedness will depend onthe number and sizes of the trees in ourpopulation. Thus, ‘crowdedness’ receivesan influence arrow from ‘diameter’.


The forester’s concept of stand basal areais adopted here, which is the sum of thesectional areas of the tree stems, becauseit is easy to compute, and because it is wellcorrelated with many interesting propertiesof a forest, such as biomass, biodiversityand evapo-transpiration. Stand basal areais simply the sum of the squares of thediameters, converted to sectional area insquare metres per hectare. Double-click onthe variable symbol for ‘crowdedness’ toobtain the following dialogue box. Double-

click the sum function from the ‘Availablefunctions’ box at left in this window. Next,double-click the local name {diameter} andtype ^2/12732 to square and divide by12732 the diameter. The number 12732,which is equivalent to 40000/π, is usedbecause π is not yet available as a functionin Simile. The 40000 converts diametersquared into radius squared, and cm2 intom2.


Recruitment of seedlings into the populationmay be represented as ‘migration’ (the birdsymbol) rather than ‘birth’ (the egg symbol),to recognise that seeds may be transported

by birds, wind, and other factors. Thisprocess has been renamed as ‘migration’.An inf luence arrow goes f rom‘crowdedness’ to ‘migration’.

Double-click on the bird symbol to obtain adialogue box. Again to avoid equations,click on the ‘Sketch graph’ button. A graphmay be sketched illustrating the negativerelationship between stand density and thenumber of seedlings that survive to the endof the first year. The graph predicts

numbers of recruits (stems/ha/year) fromstand basal area (m2/ha). It seemsreasonable to assume say 10 recruits at lowstand density, with an asymptotic declineapproaching zero as stand basal areareaches 30 m2/ha. The relationship can nowbe accepted by pressing the ‘Enter’ key.


Simile knows that the y-axis represents‘migration’, but it is necessary to specify in

the ‘Equation’ box that the x-axis represents‘crowdedness’.

3. RUNNING THE MODEL

Now that the model has been developed, itis time to run it and inspect the output. Inthis demonstration, the model will be run

with the TCL interpreter, rather than cross-compiling it as C-code allowing run time tobe reduced. The two options are availablein the Run command under the File menu.


When ‘In Tcl’ under File/Run is selected, theSimile Run Environment widow appears.Some displays may be chosen to improve

the output display. One of the standardSimile displays will be used here, namely the‘lollipop diagram’.

When the lollipop diagram is selected, thex- and y-coordinates must be chosen. Also,

the object to be displayed, which is diameterin this forestry example, must be indicated.


Simile will draw a picture of a forest in thelollipop window, with no trees because it

has not been initialized. Maximise thelollipop window.

When the ‘Start’ button in the Run controlbox within the Simile Run Environmentwindow is pressed, trees start to grow. Inthe mean time, some trees die, and newinstances are created. When the ‘Stop’button is pressed, Simile will now display apicture of the forest with the size of thetrees at a given time. This forest simulationis an impressive example of what can beachieved. It could, however, be extended to

have more than one tree species, to makemortality size dependent, to make thegrowth rate dependent on competition aswell as tree size, and so on. But thisdemonstration has been sufficient toformalize and communicate some ideasabout tree growth, and to test whether theideas are reasonable, in both the logicaland empirical sense.



Simile is a powerful medium for modeldevelopment and implementation as acomputer program. The example modeldeveloped here has demonstrated variousfeatures on Simile, including submodels,flows, populations, and birth and death.

This model took just one minute to build, alittle less to run, and a little more to

document. Hopefully, this example hasillustrates some of the capabilities of Simile,and demonstrated that, using this powerfulmodel and program development medium,models with considerable functionality donot necessarily need to be big, complicated,expensive or excessively demanding ondata.

109

11. Introduction to Discounted Cash Flow Analysisand Financial Functions in Excel

John Herbohn and Steve Harrison

The financial and economic analysis of investment projects is typically carried out using thetechnique of discounted cash flow (DCF) analysis. This module introduces concepts ofdiscounting and DCF analysis for the derivation of project performance criteria such as netpresent value (NPV), internal rate of return (IRR) and benefit to cost (B/C) ratios. Theseconcepts and criteria are introduced with respect to a simple example, for which calculationsusing MicroSoft Excel are demonstrated.

1. CASH FLOWS, COMPOUNDING ANDDISCOUNTING

Discounted Cash Flow (DCF) analysis isthe technique used to derive economic andfinancial performance criteria for investmentprojects. It is important to review some ofthe basic concepts of DCF analysis beforeproceeding to topics such as cost0benefitanalysis (CBA), financial analysis (FA),linear programming and the estimation ofnon-market benefits.

Cash flow analysis is simply the process ofidentifying and categories of cash flowsassociated with a project or proposedcourse of action, and making estimates oftheir values. For example, whenconsidering establishment of a forestryplantation, this would involve identifying andmaking estimates of the cash outflowsassociated with establishing the trees (e.g.the cost of buying or leasing the land,purchasing seedlings, and planting theseedlings), maintaining the plantation (suchas cost of fertilizer, labour, pruning andthinning) and harvesting. As well, it wouldbe necessary to estimate the cash inflowsfrom the plantation through sales ofthinnings and timber at final harvest.

Discounted cash flow analysis is anextension of simple cash flow analysis andtakes into account the time value of moneyand the risks of investing in a project. Anumber of criteria are used in DCF toestimate project performance including NetPresent Value (NPV), Internal Rate ofReturn (IRR) and Benefit-Cost (B/C) ratios.

Before discussing criteria to measureproject performance, it is necessary to

introduce some concepts and procedureswith respect to compounding anddiscounting. Let us begin with the conceptsof simple and compound interest. For themoment, consider the interest rate as thecost of capital for the project.

Suppose a person has to choose betweenreceiving $1000 now, or a guaranteed$1000 in 12 months time. A rational personwill naturally choose the former, becauseduring the intervening period he or shecould use the $1000 for profitableinvestment (e.g. earning interest in thebank) or desired consumption. If the $1000were invested at an annual interest rate of8%, then over the year it would earn $80 ininterest. That is, a principal of $1000invested for one year at an interest rate of8% would have a future value of $1000(1.08) or $1080.

The $1000 may be invested for a secondyear, in which case it will earn furtherinterest. If the interest again accrues on theprincipal of $1000 only, it is known assimple interest. In this case the future valueafter two years will be $1160. On the otherhand, if interest in the second year accrueson the whole $1080, known as compoundinterest, the future value will be $1080(1.08) or $1166.40. Most investment andborrowing situations involve compoundinterest, although the timing of interestpayments may be such that all interest ispaid before further interest accrues.

The future value of the $1000 after twoyears may also be derived as

$1000 (1.08)2 = $1166.40


In general, the future value of an amount$a, invested for n years at an interest rateof i, is $a (1+i)n, where it it to be noted thatthe interest rate i is expressed as a decimal(e.g. 0.08 and not 8 for an 8% rate).

The reverse of compounding – finding thepresent-day equivalent to a future sum – isknown as discounting. Because $1000invested for one year at an interest rate of8% would have a value of $1080 in oneyear, the present value of $1080 one yearfrom now, when the interest rate is 8%, is$1080/1.08 = $1000. Similarly, the presentvalue of $1000 to be received one yearfrom now, when the interest rate is 8%, is

$1000/1.08 = $925.93

In general, if an amount $a is to be receivedafter n years, and the annual interest rate isi, then the present value is

$a / (1+i)n

The above discussion has been in terms ofamounts in a single year. Investmentsusually incur costs and generate income ineach of a number of years. Suppose theamount of $1000 is to be received at theend of each of the next four years. If notdiscounted, the sum of these amountswould be $4000. But suppose the interestrate is 8%. What is the present value of thisstream of amounts? This is obtained bydiscounting the amount at the end of eachyear by the appropriate discount factor thensumming:

$1000/1.08 + $1000/1.082 + $1000/1.083 +$1000/1.084

= $1000/1.08 + $1000/1.1664 +$1000/1.2597 + $1000/1.3605

= $925.93 + $857.34 + $793.83 + $735.03

= $3312.13

The discount factors – 1/1.08t for t = 1 to 4– may be calculated for each year or readfrom published tables. It is to be noted thatthe present value of the annual amounts isprogressively reduced for each year furtherinto the future (from $925.93 after one yearto $735.03 after four years), and the sum isapproximately $700 less than if no

discounting (a zero discount rate) had beenapplied.

2. DEFINITION OF ANNUAL NET CASHFLOWS

DCF analysis is applied to the evaluation ofinvestment projects. Such a project mayinvolve creation of a terminating asset (suchas a forestry plantation), infrastructure(such as a road or plywood plant) orresearch (including scientific and socio-economic research). Any project may beregarded as generating cash flows. Theterm cash flow refers to any movement ofmoney to or away from an investor (anindividual, firm, industry or government).Projects require payments in the form ofcapital outlays and annual operating costs,referred to as cash outflows. They give riseto receipts or revenues, referred to asproject benefits or cash inflows. For eachyear, the difference between projectbenefits and capital plus operating costs isknown as the net cash flow for that year.The net cash flow in any year may bedefined as

at = bt - (kt + ct)

where bt are project benefits in year t kt are capital outlays in year t ct are operating costs in year t.

It is to be noted that when determiningthese net cash flows, expenditure items andincome items are timed for the point atwhich the transactions takes place, ratherthan the time at which they are used. Thusfor example expenditure on purchase of anitem of machinery rather than annualallowances for depreciation would enter thecash flows. It is to be further noted thatcash flows should not include interestpayments. The discounting procedure in asense simulates interest payments, so toinclude these in the operating costs wouldbe to double-count them.

Example 1

A project involves an immediate outlay of$25,000, with annual expenditures in eachof three years of $4000, and generatesrevenue in each of three years of $15,000.These cash flows data may be set out, andannual net cash flows derived, as in Table1.

Introduction to Discounted Cash Flow Analysis and Financial Functions in Excel 111

Table 1. Annual cash flows for a hypothetical project

Year Project Capital Operating Net cashbenefits outlays costs flow

($) ($) ($) ($)0 0 25000 2000 -270001 15000 0 4000 110002 15000 0 4000 110003 15000 0 2000 13000

Two points may be noted about these cashflows. First, the capital outlay is timed forYear 0. By convention this is the beginningof the first year (i.e. right now). On the otherhand, only half of the first year’s operatingcosts are scheduled for the Year 0 (thebeginning of the first year). The remaininghalf of the first year’s operating costs plusthe first half of the second year’s operatingcosts are scheduled for the end of the firstyear (or, equivalently, the beginning of thesecond year). In this way, operating costsare spread equally between the beginningand the end of each year. (The final half ofthe third year’s operating costs arescheduled for the end of Year 3.) In thecase of project benefits, these are assumedto accrue at the end of each year, whichwould be consistent with lags in productionor payments. These within-year timingissues are unlikely to make a largedifference to overall project profitability, butit is useful to make these timingassumptions clear.

A second point to note about Table 1 is thatnet cash flows (second column less thirdplus fourth column) are at first negative, butthen become positive and increase overtime. This is a typical pattern of well-behaved cash flows, for which performancecriteria can usually be derived withoutcomputational difficulties.

3. PROJECT PERFORMANCE CRITERIA

Let us now consider a number of projectperformance criteria which can be obtainedby discounted cash flow analysis. Thesecriteria will be defined, then derived for thecash flow data of Example 1.

Net present value

The net present value (NPV) is the sum of

the discounted annual cash flows. For theexample, taking an interest rate of 8%, thisis

NPV = a0 + a1/(1+i) + a2/(1+i)2 + a3/(1+i)3

A project is regarded as economicallydesirable if the NPV is positive. The projectcan then bear the cost of capital (theinterest rate) and still leave a surplus orprofit. For the example,

NPV = -27000 + 11000/(1.08) +11000/(1.08)2 + 13000/(1.08)3

= -27000 + 11000/1.1664 +11000/1.2597 + 13000/1.3605

= $2935.73

The interpretation of this figure is that theproject can suppport an 8% interest rateand still generate a surplus of benefits overcosts, after allowing for timing differences inthese, of approximately $3000.

Net future value

An alternative to the net present value is thenet future value (NFV), for which annualcash flows are compounded forward to theirvalue at the end of the project’s life. Oncethe NPV is known, the NFV may beobtained indirectly by compounding forwardthe NPV by the number of years of theproject life. For Example 1, the net futurevalue is

NFV = NPV (1.08)3 = $2935.73 x 1.3605 =3994.06

Internal rate of return

The internal rate of return (IRR) is theinterest rate such that the discounted sum


of net cash flows is zero. If the interest ratewere equal to the IRR, the net presentvalue would be exactly zero. The IRRcannot be determined by an algebraicformula, but rather has to be approximatedby trial and error methods. For the aboveexample, we know that the IRR issomewhere above 8%. Deriving the NPVwith a range of discount rates would revealthat the IRR falls between 13% and 14%,but closer to the latter. In practice, afinancial function can be called up toperform the trial-and-error calculations. Itwould be found in this case that the IRR isabout 13.8%.1

The IRR is the highest interest rate whichthe project can support and still break even.A project is judged to be worthwhile ineconomic terms if the internal rate of returnis greater than the cost of capital. If this isthe case, the project could have supporteda higher rate of interest than was actuallyexperienced, and still made a positivepayoff. In the above case, the project wouldbe profitable provided the cost of capitalwas less than 13.8%.

The IRR as a criterion of project profitabilitysuffers from a number of theoretical andpractical limitations. On the theoretical side,it assumes that the same rate of return isappropriate when the project is in surplusand when it is in deficit. However, the costof borrowed funds may be quite different tothe earning rate of the firm. It could be moreappropriate to use two rates whendetermining the IRR. The actual cost ofcapital could be used when the project is indeficit, and the earning rate (unknown, to bedetermined by trial-and-error) could beapplied when the project is in surplus. Thiswould give a better indication of the earningrate of the project to the firm or government.

From a practical viewpoint, the IRR may notexist or it may not be unique. This problemmay be examined in terms of the N P Vprofile, a graph of NPV versus the rate ofinterest. When the IRR is well behaved, thisprofile takes the form as in Figure 1. As the

1 Calculation of internal rate of return in fact

involves solving a polynomial equation, andefficient solution methods such as Newton’sapproximation are used in computerpackages.

interest rate increases the NPV falls, beingzero where the NPV curve crosses theinterest rate axis; the IRR corresponds withthis discount rate.

Consider a project for which the net cashflow in each year (including Year 0) ispositive. Regardless of the interest rate, theNPV will never be zero, so it will not bepossible to determine an IRR. Similarly, aproject with a large initial capital outlay andfor which future benefits are relatively smallor negative may not have a positive NPVregardless of the interest rate, so again thecurve for the NPV profile may never crossthe interest rate axis.

If a project generates runs of positive andnegative net cash flows, the NPV profilemay take the form of a roller-coaster curve,crossing the interest rate axis in severalplaces.2 This indicates multiple internalrates of return, one at each interest ratewhere NPV is zero. It is then by no meansclear which if any of the rates we shouldchoose to call the IRR. Further, for somesections of the NPV profile (those that areupward sloping), the NPV is increasing asthe interest rate increases. This implies thatthe greater the cost of capital the moreprofitable the project. Clearly, multipleinternal rates of return and perverserelationships between the NPV and thediscount rate are not very satisfactory.

Benefit-to-cost ratios

A number of benefit-cost ratio conceptshave been developed. For simplicity, we willconsider only two concepts, referred to asthe gross and net B/C ratio and definedrespectively as

Gross B/C ratio = PV of benefits/(PV ofcapital costs + PV of operating costs)

Net B/C ratio = (PV of benefits – PV ofoperating costs)/PV of capital costs

2 Mathematically, the polynomial equation defin-

ing the NPV can have up as many ‘roots’ orsolutions as there are turning points in NPVvalues (changes from positive to negative ornegative to positive). A project which hasalternating runs of positive and negative cashflows is a candidate for problems withestimation of the IRR.


For the above project, the present value ofcapital outlays is $25,000, since outlays aremade immediately and as a single amount.The present values of project benefits andoperating costs are:

PV of benefits = $15000/1.08 +$15000/1.082 + $15000/1.083 = $38656.45

PV of operating costs = $2000 +$4000/1.08 + $4000/1.082 + $2000/1.083 =$9418.38

Hence the benefit-to-cost ratios are

38,656.45 gross B/C ratio = = 1.12 25,000 + 9418.38

38,656.45 - 9418.38 net B/C ratio = = 1.45 25,000

A project is judged to be worthwhile ineconomic terms if it has a B/C ratio isgreater than unity, i.e. if the present value ofbenefits exceeds the present value of costs(in gross or net terms). If one of the aboveratios is greater than unity, then the otherwill be greater than unity also. In the aboveexample, the ratios are greater than unity,indicating that the project is worthwhile oneconomic grounds. It is not clear on logicalgrounds which of the ratios is the mostuseful.

Net present value

Internal rate of return

0 Interest rate

Figure 1. The NPV profile for a project

The payback period

The payback period (PP) is the number ofyears for the projects to break even, i.e. thenumber of years for which discountedannual net cash flows must be summedbefore the sum becomes positive (andremains positive for the remainder of theproject’s life). The payback period for aproject with the above net cash flows canbe determined as in the following table.From this table, it is apparent that the sumof discounted net cash flows does notbecome positive until Year 3, so thepayback period is three years.

The payback period indicates the number ofyears until the investment in a project is

recovered. It is a useful criterion for a firmwith a short planning horizon, but does nottake account of all the information available,i.e. the net cash flows for years beyond thepayback period.

The peak deficit

This is a measure of the greatest amountthat the project ‘owes’ the firm orgovernment, i.e. the furtherest ‘in the red’ itgoes. In the above table, the largestnegative value is -$27,000, so this is thepeak deficit. Peak deficit is a usefulmeasure in terms of financing a project,since it indicates the total amount of financethat will be required.


Table 2. Derivation of ‘project balances’ and payback period

Year Net cash PV of net Cumulative PV of net cashflow cash flow flow (or project balance)($) ($) ($)

0 -27000 -27000.00 -270001 11000 10185.19 -27000 + 10185.19 = -16814.812 11000 9430.73 -16814.81 + 9340.73 = -7384.093 13000 10319.82 -7384.09 + 10319.82 = 2935.73

Review of DCF performance criteria

The most commonly used discounted cashflow performance criteria – NPV, IRR, B/Cratios and payback period – may besummarised as in Table 3. The variouscriteria are closely related, but measureslightly different things. In this respect, theytend to complement one another, so that itis common to estimate and report morethan one of the measures.

Perhaps the most useful measure, and theone most often reported, is the net presentvalue. This tells the total payoff from aproject. A limitation of the NPV is that it isnot related to the size of the project. If one

project has a slightly lower NPV thananother, but the capital outlays required aremuch lower, then the second project willprobably be the preferred one. In thissense, a rate of return measure such as theIRR is also useful.

The payback period and peak deficit areuseful supplementary project information fordecision-makers. They have greaterrelevance for private sector investments, forfirms which cannot afford long delays inrecouping expenditure, and where carefulattention must be paid to the total amount offunds that will need to be committed to theproject to remain solvent.

Table 3. Summary of definitions of main DFC performance criteria

Net present value (NPV) p∑ at / (1+i)t, where t is time, ct is the annual net cash flow,t=1 i is the discount rate p is the planning horizon

Internal rate of return (IRR) p

The value of r such that ∑ at / (1+r)t = 0 t=1

Benefit to cost ratio (B/C) Present value of project benefits / present value of projectcosts

Payback period Number of periods until NPV becomes (and remains)positive

4. USING EXCEL FINANCIALFUNCTIONS

It is common to use a spreadsheet programsuch as Excel to undertake financial andcost benefit analyses. Excel has a range offinancial functions that can be useful (aslisted in Apppendix 1). Two of the mostuseful and commonly used financialfunctions in Excel are those to calculate NetPresent Value (‘NPV’) and Internal Rate of

Return (‘IRR’). This section illustrates theuse of two of the NPV and IRR functions,applied to the data in Example 1, in anExcel worksheet:

Using the ‘NPV’ and ‘IRR’ functions

The ‘NPV’ function calculates the netpresent value of an investment (i.e. seriesof cash flows) by using a discount rate anda series of future payments (negative


values) and income (positive values). Thesyntax for this function is:

NPV(discount rate, value1, value2 ….)

It is important to remember that the NPVcalculation is for cashflows starting at theend of period 1. If the first cash flow occursat the beginning of period 1 (i.e. period 0),the first value must be added to the NPVresult, not included in the value ‘arguments’of the function.

Thus to value the NPV for example 1, thefollowing formula would be inserted into aspreadsheet cell:

=NPV(8%, 11000, 11000, 13000) - 27000

Note that the cashflow sequence used inthe NPV calculation is for years 1 to 3. Thevalue for year 0 (-$27000) is added to theNPV result.

An alternative means of calculating NPVand IRR in the above example would be toreplace the string of values with cellreferences to where the values are locatedwithin the workbook. This is illustrated in theworksheet below. This approach has the

advantage of allowing changes to the madeto the cashflows (in Column E) which thenautomatically flow through to the calculationof the NPV. This is particularly useful whenit is desired to investigate how changes incash flows affect NPV. Similarly, thediscount rate given in the formula (in thiscase 8%) could be replaced with a cellreference. In this case the effects of achange in discount rate on NPV could beobserved simply changing the cell valuerather than the formula. In the worksheetbelow, the ‘8%’ in the formula for NPV incell B8 would be replaced by a reference tocell B12. The new formula would be‘=NPV(B12, E3:E5)+E2’.

Another approach to calculating NPV is tosimply calculate the present value of cashflows for each year using the discountingformula given in Equation 1 and then simplyadd these individual figures. This approachhas a number of advantages in morecomplex applications – particularly thoseinvolved complex financial models withvariable periods of cash flows and wherediscount rates change over the life of aproject. This is illustrated in the workbookbelow:


The formula in Cells F2, F3, F4 and F5calculate the present value of the net cashflow in each of the years 0 through to 3using the formula given in Equation 1.These present values for each of the yearsare then summed in cell F6 to give the NetPresent Value of the series of cash flowsassociated with the project. The resultingNPV in cell F6 is identical to the NPVcalculated in cell B8. Note that an absolutecell address for the discount rate has beenused ($B$12) using ‘$’. This allows theformula in cell F2 to be simply copied intocells F3, F4 and F5.

APPENDIX 1: FINANCIAL FUNCTIONS INEXCEL

ACCRINT Returns the accrued interest fora security that pays periodic interestACCRINTM Returns the accrued interestfor a security that pays interest at maturityAMORDEGRC Returns the depreciationfor each accounting periodAMORLINC Returns the depreciation foreach accounting period COUPDAYBS Returns the number of daysfrom the beginning of the coupon period tothe settlement dateCOUPDAYS Returns the number of daysin the coupon period that contains thesettlement dateCOUPDAYSNC Returns the number ofdays from the settlement date to the nextcoupon dateCOUPNCD Returns the next coupon dateafter the settlement dateCOUPNUM Returns the number ofcoupons payable between the settlementdate and maturity dateCOUPPCD Returns the previous coupondate before the settlement dateCUMIPMT Returns the cumulative interestpaid between two periodsCUMPRINC Returns the cumulativeprincipal paid on a loan between twoperiodsDB Returns the depreciation of an assetfor a specified period using the fixed-declining balance methodDDB Returns the depreciation of an assetfor a specified period using the double-declining balance method or some otherdepreciation methodDISC Returns the discount rate for asecurity

DOLLARDE Converts a dollar price,expressed as a fraction, into a dollar price,expressed as a decimal numberDOLLARFR Converts a dollar price,expressed as a decimal number, into adollar price, expressed as a fractionDURATION Returns the annual duration ofa security with periodic interest paymentsEFFECT Returns the effective annualinterest rateFV Returns the future value of aninvestmentFVSCHEDULE Returns the future value ofan initial principal after applying a series ofcompound interest ratesINTRATE Returns the interest rate for afully invested securityIPMT Returns the interest payment for aninvestment for a given periodIRR Returns the internal rate of return fora series of cash flowsISPMT Calculates the interest paid duringa specific period of an investment.MDURATION Returns the Macauleymodified duration for a security with anassumed par value of $100MIRR Returns the internal rate of returnwhere positive and negative cash flows arefinanced at different ratesNOMINAL Returns the annual nominalinterest rateNPER Returns the number of periods foran investmentNPV Returns the net present value of aninvestment based on a series of periodiccash flows and a discount rateODDFPRICE Returns the price per $100face value of a security with an odd firstperiodODDFYIELD Returns the yield of asecurity with an odd first periodODDLPRICE Returns the price per $100face value of a security with an odd lastperiodODDLYIELD Returns the yield of asecurity with an odd last periodPMT Returns the periodic payment for anannuityPPMT Returns the payment on theprincipal for an investment for a givenperiodPRICE Returns the price per $100 facevalue of a security that pays periodicinterestPRICEDISC Returns the price per $100face value of a discounted security


PRICEMAT Returns the price per $100face value of a security that pays interest atmaturityPV Returns the present value of aninvestmentRATE Returns the interest rate per periodof an annuityRECEIVED Returns the amount receivedat maturity for a fully invested securitySLN Returns the straight-line depreciationof an asset for one periodSYD Returns the sum-of-years' digitsdepreciation of an asset for a specifiedperiodTBILLEQ Returns the bond-equivalentyield for a Treasury billTBILLPRICE Returns the price per $100face value for a Treasury billTBILLYIELD Returns the yield for aTreasury bill

VDB Returns the depreciation of an assetfor a specified or partial period using adeclining balance methodXIRR Returns the internal rate of return fora schedule of cash flows that is notnecessarily periodicXNPV Returns the net present value for aschedule of cash flows that is notnecessarily periodicYIELD Returns the yield on a security thatpays periodic interestYIELDDISC Returns the annual yield for adiscounted security. For example, aTreasury billYIELDMAT Returns the annual yield of asecurity that pays interest at maturity

119

12. Financial Models of Farm and CommunityForestry Production Systems

John Herbohn, Nick Emtage and Steve Harrison

An integral component of investment in forestry is the need for financial information aboutthe likely cash flows associated with the establishment, management and final harvest of aplantation. This paper uses the development of a financial model for forestry investment as acase study of financial modeling. Financial evaluation of forestry projects poses manychallenges and this chapter includes discussion of the methods employed in modeldevelopment. The techniques of model development are illustrated by case studies of thedevelopment of the Australian Cabinet Timber Financial Model (ACTFM) and the AustralianFarm Forestry Financial Model (AFFFM).

1. FORESTRY EVALUATION MODELS:USES AND USER GROUPS

The traditional use of financial models is toassess simply the financial viability of aproposed investment. As well as derivingsingle point estimates of financialperformance criteria, sensitivity analysisand risk simulation may be performed toexplore the impact of variations from thecore assumptions. Models can also be usedto explore investment alternatives. Forinstance forestry models applied to jointventures can be used to explore the impacton cash flows of different equityparticipation arrangements. Large pastoralcompanies can also use financial models toexplore the impact that forestry may haveon their overall business risk. The way inwhich a model is to be used is a crucialconsideration in design of the model.

Forestry financial models can potentially beused by a number of different groups andfor a number of different purposes. Forinstance, a company which frequentlypromotes forestry schemes to investorswould require a model that could be easilyadapted to new areas and different treespecies. Similarly, a government depart-ment promoting joint venture arrangementsinvolving only one or two species but whichinvolve d i f ferent equi ty shar ingarrangements with private landholderswould need a model that was flexible indifferent way. In both cases, it would beinefficient to develop a new spreadsheet-based model for each new situation.

Globally, forestry is a major economicactivity at the farm level and can be asignificant contributor to gross domesticproduct (GDP), particularly in manyEuropean countries. Farm operations arehowever highly diverse, in terms of areaplanted (1000s of hectares down to verysmall plots), species chosen, managementregime, harvest age, and other variables.Forestry is typically only part of the overallbusiness undertaking of the landowner, andmust be considered in this context.

Farmers often lack the skills to developfinancial models capable of modelling theiractivities. The provision of models to beused as a tool to undertake this type ofanalysis is often done as a form ofassistance or extension service bygovernments. In such cases, models needto be highly flexible and take into accountthe impact that forestry has on the overalloperations of the farm, particularly cashflows. The impact that forestry has on thefarm financial structure and cash flows canbe critical. In such cases, a project with apositive NPV may be rejected because itcannot be accommodated within thefinancial structure of the business (i.e. thequestion of how, if at all, the project can befinanced becomes critical).

Forestry financial models are also useful forland valuers, real estate agents and lendingagencies, all of which often have a need toplace a value on immature forests whenestimating market values or evaluating loanproposals for rural properties. They furtherhave a role in training foresters and as a


teaching device.

2. KEY PARAMETERS FOR FORESTRYMODELS

The most common cash outflowsassociated with forestry operations include:establishment costs (e.g. land and sitepreparation such as clearing, fencing andripping; purchase of planting stock; planting,mulching and watering); weed control;fertilization; pruning and thinning to waste,and commercial harvesting. The major cashinflows come from sale of commercialthinnings and final harvest, and perhapstree residues such as firewood. In somesituations revenues are also derived fromnon-wood forest products such as wildberries, mushrooms, honey and hunting, allof which are made possible by theenvironment created by the forest.

Cash flow estimates of growing nativetimbers as a business enterprise will varydepending on a variety of factors, some ofthe more important of which are listed inTable 1. In principle, a spreadsheet couldbe devised which incorporates each ofthese factors. One factor that can have amajor impact on eventual cash flows istimber yield from plantations. Yield iscommonly expressed as ‘mean annualincrement’ (MAI) which is the aggregatevolume of harvestable timber produced in ayear from the growth of trees in aplantation. It is usually expressed as cubicmetres of timber produced per ha per year.Other factors, such as site productivity,species mixture, harvest ages and timberprices (usually expressed as $ per cubicmetre) may also strongly affect eventualcashflows.

Table 1. Key factors in financial performance

Site characteristics (e.g. climate, soil, aspect)Species mixtureSilvicultural system (e.g. planting density, weed control and pruning)Harvest age and harvest schedulingFinal yield or MAI of individual speciesInteractions between species in mixed-species plantationsStumpage price (affected by a number of supply and demand factors)Costs (land preparation, planting and establishment, maintenance, harvesting)Amount of government assistanceHarvest rights (buffer zones, harvest on steep land, roading permission)Taxation regime (deductions allowable, treatment of harvest revenue)Allowance for non-wood benefitsDiscount rate

3. SOURCES OF VARIABILITY INFORESTRY INVESTMENTPERFORMANCE

Plantation forestry is not a risk-freeinvestment. A variety of factors cancontribute to the uncertainly in revenuegenerated. Most cash inflows fromplantations occur with the final harvest oftrees when they reach maturity, althoughsome additional cash inflows may begenerated from the sale of trees that arethinned as part of the normal managementpractices. Typically it takes 30 or moreyears until plantations are ready for finalharvest trees, hence long-term predictionsof physical and financial performance haveto be made. The main sources of variability

in financial performance of forestry aresummarised in Table 2.

4. DEVELOPING A FINANCIAL MODEL –A STEP BY STEP APPROACH

As with any project appraisal, undertaking afinancial analysis of a forestry project canbe broken down into a number of individualand relatively simple steps:

1. identify the forestry system to beadopted – for example, the type of treesto be planted and at what density, whenthey will be pruned, the types of productintended (e.g. sawlogs, veneer logs,pulpwood, poles, or some combination

Financial Models of Farm and Community Forestry Production Systems 121

of these) and when harvesting is likelyto occur.

2. estimate the likely cash outflows.3. estimate the cash inflows from harvest.4. develop the financial model and

est imate f inancial performancemeasures.

5. evaluate investment risk.

Table 2. Sources of risk in farm forestry

Risk category Major sourcesRisk of poor establishment Dry weather, poor weed controlProduction (timber yield) risk Storm or cyclone

FirePest (insect, disease)Unsuitable species or mixture for the siteCollateral damage at harvest

Timber quality and product Inappropriate pruning and thinning regimetype risk Insect damage

Product type and fashion changes in demandSovereign risk Regulatory changes re machinery use and roading

Changes in taxation arrangementsUncertain harvest rights and compensation

Market risk Uncertain future timber prices

5. REVIEW OF MODEL DEVELOPMENTAND DESIGN OPTIONS

Models may be developed on a one offbasis – that is a unique model is developedfor each investment being appraised.Alternatively a generic model can bedeveloped which is flexible enough toevaluate a different investment scenarios.This is basically a question of cost-effectiveness. If a model is to be usedrepeatedly, then greater effort in modeldevelopment is warranted. For thesegeneric models to be used by a variety ofstakeholders, most of whom have littleunderstanding of the mechanics of themodel, the user interface or series ofscreens for data input and output hasproved to be a critical design factor. Also,ease of navigation around the model iscritical, and a convenient way to providethis is through button bars which are‘clickable’ with the mouse.

What are the options?

It may be that a generic model is alreadyavailable, which has been developedelsewhere. Regardless of whether this is‘freeware’ or moderately expensive toacquire, it may be in the best interests of afirm or individual to gain access to such apackage rather than commit resources to

developing their own model. Some of theimportant modeling options for forestry areoutlined in Table 3.

Well-designed generic models can copewith a wide range of investmentcircumstances. However, it is not possibleto predict all of the situations in whichpeople may wish to use a model. In thedesign of a generic model, it is critical togive consideration to the degree of flexibilitythat is required for changes to be made,and the availability of skilled modelers tomake changes in the model toaccommodate novel or unforeseenapplications, and choice of computingplatform to suit user needs. This can alsobe a major influence on the software usedto develop the model.


Table 3. Modelling options for forestry investments

Characteristic of model OptionsModel complexity and resource demands One-off vs genericModel property rights Developed by investor, off-the-shelfLevel of business activity Forestry enterprise or whole-of-businessLevel of flexibility Narrow to wide range of applicationsModeling platform Spreadsheet programs (e.g.Excel), programming

languages (e.g. C++, Visual Basic) or simulationlanguages (e.g. Stella, Simile)

Type of yield modeling Growth curves, growth simulations usingpackages such as Plantgro, discrete harvest ageand MAI

Source of input data Default (trial-based, subjectively estimated,simulated) vs user provided

User interface Menu system, button barsChoice of project benefit categories Wood only vs wider benefits (financial vs

economic)Representation of investment risk RADR, sensitivity analysis, risk analysisChoice of project performance variables NPV vs LEV, IRR, not payback periodUser support None, access to technical assistance, package

maintenance

What type of platform to use?

For a large and complex generic model, aspreadsheet platform may not be suitable.In such cases, it may become necessary toresort to programming languages such asVisual Basic or C++. In other cases modelingenvironments such as Simile, which havebeen specifically developed for use indeveloping models for complex resourcemanagement decisions (as is often involvedwith forestry projects) may be appropriate.

Spreadsheets are a very convenientmodeling environment in which to developfinancial models. Excel for instance has arange of features that are extremely usefulin model building including button bars,financial functions, the ability to easilyrecord macros, and input forms. Manypeople are now reasonably proficient atusing spreadsheets. As such, one of themajor benefits of using spreadsheet modelsis that these models need far less technicalsupport provided to users and can be easilymodified. In contrast, models developedusing a programming language such as

Visual Basic or C++ can require significantongoing technical support and requireexpert programmers for modifications to bemade.

6. THE AUSTRALIAN CABINETTIMBERS FINANCIAL MODEL(ACTFM)

The Australian Cabinet Timber FinancialModel has been developed as a flexible anduser-friendly computer-based modelsuitable for assessing the financial viabilityof timber plantations using Australian nativeeucalypt and cabinet species in NorthQueensland. A schematic representation ofthe structure of the model is provided inFigure 1. Little published information existson growth rates of many of these native treespecies. To provide information for themodel about the growth rates and optimalharvest ages for the 31 species included inthe model, it was necessary to undertake aDelphi survey of experts on NorthQueensland forestry. Details of this studyare provided in later sections.


Figure 1. Schematic representation of the structure of the ACTFM

The model contains a number of linkedsheets in an Excel workbook format. Sheetsare used to store or display data orinformation about the program, and toperform calculations of the value of theplantation scenario established by the user.The model was developed on the Microsoft

Excel (version 7.0) software and contains acollection of Visual Basic macros. Itrequires the Excel program to operate. Tonavigate between various data input andmodel output screens, a system of ‘buttonbars’ is provided. Notably, the model was

Plantation output – values

1.Speciesselection

2.Growth

parameters

Common names

Species

(Up to 5 species

selected out of 31)

Site Performance Index

Harvest ageMAI

5.Costs

CostsYear of cost

Costs 6. Calculatesharvestvalues

Harvest values7.Generates

cashflows foreach year

Cashflows Cashflows

8.CalculatesNPV foreach year

11. ReturnsIRR forhighest

harvest age

10. Sums upand returns

NPV forhighest

harvest age

9.Calculates

IRR

NPV

IRR

NPV IRR

3.Economicparameters

Timber Price

Discount rateTimber price change index

12.Output /recomm-endations

4.Plantation output –

summary sheet


developed by researchers and notprofessional computer programmers. The model has been set up with anattractive opening sheet, which indicatesthe model name, names of the developersand version of the model, andacknowledges the agencies providingfinancial support for model development.From the title page access is gained to the‘Plantation Output’ summary sheet (Figure2) by clicking on a ‘Start’ button. This in turnis linked to the various other sheets in theworkbook which provide default data andinstructions and other information to theuser, to allow calculation of net presentvalue and the internal rate of return. The

menu system to allow the user to movefreely between sheets in the workbooksconsists of a series of ‘buttons’, on the usersimply clicks the computer mouse. Thevarious sheets provide:

• information about the program;• data relating to MAI’s, stumpage

prices, and harvest ages of 30different cabinet timber species;

• data relating to establishment andmaintenance costs;

• tables for the entry of alternativeMAI, price, costs and harvest agedata; and

• pages containing visual basic code.

Figure 2. Example of Plantation Output Sheet

Users can see and access all theinformation about a plantation scenario withup to five species from one sheet (Figure2). Users can also choose to select defaultdata to run the model, or may enter theirown growth, cost and price data. The main outputs of the model aresummarised in the ‘Plantation Output’ sheetin order to make it easy to see what ishappening in terms of the volumes of timber

produced, the value of the harvest, andtiming of returns. This sheet also assists inunderstanding the effects of changing eachvariable on the overall f inancialperformance of the plantation. Moredetailed output from the financial analysis isdisplayed in the ‘Economic Summary’ sheetwhich can be accessed from the ‘PlantationOutput’ sheet by a button bar.


Default values of the model

The model contains embedded estimatesfor a number of parameters used in thecalculation of NPV and IRR, some of whichare presented in Table 4. Default values forMAI and harvest ages were derived from aDelphi survey of experts on NorthQueensland forestry undertaken for 31cabinet timber species in 1996-97, and areprobably the best projections currentlyavailable of future growth of many of the

species. They are however for restrictedgrowth conditions (relatively fertile basaltderived soils in areas of 1,600-2,000 mmrainfall) and so users may wish to entersite-specific values. This need is to someextent overcome by the inclusion of a siteindex variable, which allows the user to varythe MAI (growth) data for all species in ascenario according to the conditions forwhich the model is being used relative tothose specified for the estimates.

Table 4. Estimated harvest ages, timber yields and timber prices for eucalypt and cabinet

timber species in North Queensland

Species Harvest age(yrs)

Yield (MAI inm3/ha)

Timber price($/m3)

Acacia mangium 23.9 20.6 Acacia melanoxylon 29.5 10.8 299 Agathis robusta 46.2 15.3 Araucaria cunninghamii 43.8 19.4 Beilschmieda bancroftii 102.0 2.9 Blepharocarya involucrigera 47.9 8.7 Cardwellia sublimis 58.1 7.5 317 Castanospermum australe 66.7 5.9 280 Cedrela odorata 37.5 10.3 242 Ceratopetalum apetalum 113.3 5.0 355 Elaeocarpus angustifolius 33.8 16.9 317 Endiandra palmerstonii 156.0 4.5 653 Eucalyptus camalduensis 28.1 17.0 Eucalyptus citriodora 32.0 16.1 Eucalyptus cloeziana 35.5 17.2 47 Eucalyptus cloeziana (poles) 25.5 15.8 Eucalyptus drepanophylla 44.4 10.8 Eucalyptus grandis 31.4 20.6 Eucalyptus microcorys 30.6 15.8 Eucalyptus pellita (resinifera) 30.6 16.3 75 Eucalyptus tetreticornis 31.7 16.0 Flindersia australis 57.5 - 355 Flindersia. bourjotiana 52.5 10.3 373 Flindersia. brayleana 43.0 13.5 205 Flindersia iffliana 56.7 8.0 243 Flindersia pimenteliana 55.0 9.0 317 Flindersia schotiana 48.3 8.5 467 Gmelina fasciculiflora 54.2 8.6 280 Grevillia robusta 35.0 8.0 149 Melia azedarach 31.1 13.7 243 Paraserianthes toona 55.0 7.1 Toona cilata (australis) 48.8 8.9 336

Source: Herbohn et al. (1999). Notes: Assumed biophysical conditions: basalt-derived soils, annual rainfall 1,600-2,000 mm per year.Price estimates were not available for some uncommon species.


If, for example, the soils are less fertile thanbasalt derived soils or the site underconsideration receives less rainfall, then thesite productivity index can be set at lessthan 100%, reducing the estimated volumeof timber produced in the scenario.However, if the rainfall regime in the areafor which the plantation scenario isestablished is greater than 2,000 mm peryear and on fertile soil, or irrigation of theplantation is planned, then the siteproductivity index may be set at greaterthan 100%. The maximum allowable valuefor the index is 150%, and the minimum50%. The default value for the index is100%.

The default values for the costs ofestablishing and maintaining a plantationthat are supplied with the model are thosefor Community Rainforest ReafforestationProgram (CRRP) plantings in the AthertonTablelands region. Cost items include landpreparation and planting, pruning andcertification of pruning, weeding, fertilisersand fencing. Also, allowance is made foropportunity cost for financial returns to theland foregone with plantation establishment,and insurance and road maintenance. It isarguable whether a land rental amountshould be included as part of the cost of theplantation. particularly in the case ofdegraded or otherwise unproductive land. Timber prices and price changes overtime The generic stumpage price of $100/m3 hasbeen used as a default value, being theapproximate royalty received by the Stateforest service for plantation Hoop Pine atthe time of model development. The modelallows users the option to select their owngeneric stumpage price. Stumpage priceshave a major influence on estimates of NPVand IRR, hence sensitivity analysis isreported in the ‘Economic Summary’ sheet. Given that the most of the species beingplanted are premium cabinet timbers it islikely that the average stumpage receivedfor these will be considerably higher thanthat Hoop Pine. The only availableestimates of future stumpage price for many

of the species in the model have beenderived by a commercial consultant and arereported in Russell et al. (1993). No defaultprice is supplied for species not examinedby Russell et al. (1993). Users can set theirown timber prices by selecting the ‘Settimber prices’ button on the front sheet, andselecting the ‘Set own timber prices’checkbox in the dialog box that issubsequently displayed. This will result in anew table being displayed, showing thespecies and their prices that can be alteredindividually by users. Figures 3, 4 and 5 report default costs(CRRP costs) from the Australian CabinetTimbers Financial Model. Additionalprescriptive costs can be added to theprescriptive costs of Figure 3 by enteringthe cost type and amount under section 2,to be added to the sub-total. The rate at which timber prices change peryear can be set in the ACTFM model byselecting the button ‘Set timber pricechange rate’. Russell et al. (1993)suggested a possible increase of up to1.3% per annum in timber prices, which isconsidered conservative by the authorscompared to a number of long-termforecasts. The default value for the model is0% although the option is available to set areal price change of up to +5% or -5%. Output values generated by the ACTFM The ACTFM calculates the financial returnsthat may be expected from a hypotheticaleucalypt or cabinet timber plantation,expressed in the form of net present value(NPV) and internal rate of return (IRR) perhectare of the plantation. The NPV of aplantation in the model is the financialreturns that can be expected fromharvesting the trees for timber in today’svalues. The model adopts a 7% discountrate, but also provides NPV estimates withthe discount rate varied by +1% and -1%.Outputs from the model are displayed onscreen, but a ‘print button’ is available togenerate a hard-copy version.


Figure 3. Prescriptive costs sheet

Figure 4. Costs during plantation sheet

Figure 5. Annual cost sheet


ACTFM takes into account only financialbenefits, and does not include non-woodbenefits to the firm or agency establishingthe plantations. If for example carboncredits were available from growing thetrees, then this additional income could beincluded in the model.The model assumes fixed harvest ages(default values or entered by the user) anddoes not directly allow for a trade-offbetween harvest age and timber revenue.However, the user could enter their ownestimates of harvest age and timber price,and could re-run the model with differentcombinations. The model does not allowdirectly for any interaction between speciesin mixed species plantings, although theuser could adjust timber yields if they wereconfident about their estimates ofinteractions.

Evolutionary development of the ACTFM

A stochastic version of this model has beendeveloped (Harrison et al. 2001), whileother versions have had other featuresadded such as the calculation of LEV andthe ability to add new species for use inspecific financial analyses. It has been usedby a variety of forestry stakeholders,including forestry investors, consultants,farm forestry advisers, researchers andstudents; corporate users; valuers and realestate agents.

Development of ACTFM commenced with avery simple financial model that had beenproduced in a national forestry inquiry. TheAFFFM model utilized components of twoprevious models (ACTFM and the CAREPty. Ltd. model). Even though the ACTFMrepresents the culmination of several yearsof model development, it is still recognizedto have limitations in terms of its suitabilityto evaluate a wide range of forestryinvestment projects. Experience indeveloping and trialling the ACTFM led torecognition of a need of some users tomodel the impact of forestry on the widerbusiness activities along with a more robustmodelling framework not dependant onExcel. In response to this, the whole-farmplanning components of the CARE Pty. Ltd.financial model and the ACTFM financialmodel have been integrated in a user-fr iendly stand-alone Visual Basicprogrammed package, known as the

Australian Farm Forestry Financial Model(AFFFM). This model is designed toevaluate forestry investment proposals onmixed farming properties in easternAustralia.

7. THE AUSTRALIAN FARM FORESTRYFINANCIAL MODEL

The Australian Farm Forestry FinancialModel (AFFFM) was developed to assessthe potential effects of farm forestryactivities on an individual farms’ financialposition. The key user groups for theAFFFM are anticipated to be farm financialadvisors, farm forestry extension officers,companies and corporations involved inplantation establishment and management,and individual landholders. The model wasdesigned to account for the financial effectsof livestock and cropping activities, timberplantation development, forestry activities innative forests, and other basic financingcosts and revenues for farm enterprises(Figure 6). By adding the details of thefinancial performance of farm enterprisesand financing details the AFFFM allows fornot only the assessment of the financialviability of forestry enterprises, but alsoassessment of whether a farm enterprisecan finance new forestry enterprises.

The model was developed as a series of‘modules’ relevant to the different activitieson a landholding (e.g. grazing, cropping,farm financing and forestry) (Figure 6).Users decide which activities are relevantfor each financial analysis. Users of themodel specify ‘scenarios’ for the differentactivities on a hypothetical landholding. Thecashflows included in each scenario aredetermined by the activities selected to beincluded in the scenario on the ‘FarmStructure’ form (Figure 7). The data enteredby users to create scenarios can be savedas text files for later retrieval. The settingsfor the main three activities covered by theAFFFM (i.e. plantations, native forestry andagriculture) can be saved as separatescenarios for each activity. Users can alsosave and retrieve ‘whole-farm’ scenariosthat include the settings of the mainactivities and those for farm finances.

Data availability is a key limitation formodeling any forestry system using nativetree species in Australia. There is little


published data about the potential growthrates, harvest ages and products for mostAustralian tree species that can potentiallybe grown in plantations. The modelcontains and is linked to data about treegrowth and potential products from harvestsin several Eastern Australian regions thatcan be accessed by users. Data relating tocabinet timber species suitable forplantations in far north Queensland is thesame as that supplied with the ACTFM(Herbohn et al. 1999). Data relating to treespecies suitable for plantations in the NewEngland Tablelands and Darling Downsregions was compiled using the Plantgro

software package. The Plantgro softwarewas developed by researchers from CSIROto estimate plant growth in circumstanceswhere there is limited field trial data(Hackett 1991). Data about individualspecies is stored as text files that can belinked to the AFFFM. When resourcesbecome available to expand the regionscovered by the model, the technique ofusing linked text files allows data about newtree species and regions to be added withonly small modifications of the codecontrolling the AFFFM.

Figure 6. Schematic diagram of the structure of the AFFFM

Notes: Forms in black font are reporting forms whilst those in green font are data entry forms. Linesbetween forms indicate that users can navigate directly between them.

Farmstructure form

Plantationsform

Activity optionsform

Reporting textfiles

Plantation costsform

Plantationproducts form

Suitabilityratings form

Native forestsform

Native forestcosts form

Native forestproducts form

Agricultureform

Under plantationsgrazing form

Reportinggraphs

Farm financeform

Sensitivityanalyses


Figure 7. The main reporting form from the Australian Farm Forestry Financial Model

Functions of the model

The financial viability of each activity can beassessed separately or in conjunction withany or all of the other activities. The modelcalculates the cashflows for each relevantactivity in each year of a scenario, and thencalculates the overall cashflow for thelandholding in each year. The costs andreturns for each year in the future are thendiscounted back to their present value atthe discount rate set by the user. TheAFFFM lets users assess the relativeinfluence of the various costs and revenueson the farms’ financial position by allowingadjustment of these figures for the differentactivities. Users can also assess the effectof changes in timber prices and plantationgrowth rates on the NPV of a scenario.

The potential financial impacts of forestrydevelopments are reported using a numberof commonly used measures of long terminvestments including the net present value(NPV), internal rate of return (IRR), landexpectation value (LEV) and business cashposition of a scenario. Most measures offinancial performance are reported to the

main form of the program, the ‘FarmStructure’ form (Figure 7). The model canbe used to assess the cashflows fromdifferent activities and overall farmcashflows from a combination of activities.When forestry activities are included inconjunction with agriculture the modelreports the IRR as the difference betweenthe ‘with’ and ‘without’ forestry situations.Other reporting features of the AFFFMinclude graphic display of yearly cashflowsand the business cash position of the farmenterprise; reports of the settings andfinancial measures of scenarios; and writingof text files with yearly values of keyvariables.

Data requirements of the model

The data required to operate variousmodules of the model are listed in Table 5.Those parameters listed as ‘basic’ arerequired to operate the various activities ofthe model while those listed as ‘advanced’may be used when the user has sufficientinformation.


Table 5. Parameters used in the Australian farm forestry financial model

Farm activity Basic parameters Advanced parametersPlantations Species used – name

Plantation size(s) – hectaresStems – per hectare ORPercent of hectare per speciesMean annual increment(s) – m3/ha/yr-

Harvest age(s) – yearsStumpage price(s) – dollarsEstablishment costs – dollarsMaintenance costs – dollarsHarvest costs – dollarsAnnual costs – dollars

Harvest cycles (three harvestages per species)

Plantation product mix perharvest cycle (as % of standingvolume) and product prices

Series of plantationsTime between plantationestablishments

Number of plantations to beestablished

Plantation design(woodlots/shelterbelts)

Native forestry Forest area(s) – hectaresMean annual increment – m3/ha/yr-

Maintenance costs – dollarsStumpage price(s) – dollars

Non-commercial thinning costsand timing

Timber products and recoveryrates

Livestock Carrying capacity – DSE/haStock numbers – No. of headDSE ratings – per headGross margins – $/DSE

% DSE improvement with shelter

Crops Crop type(s) – nameCrop area(s) – hectaresGross margin(s) – $ per ha

Farm finances Overhead costs – $Capital costs – $Off-farm income – $Other farm income – $

Loan(s) – $Interest rates – loans, savings,overdrafts

The model was developed to cater fordifferent levels of complexity of data withinthe modules. For example, in the plantationscenario the prices received for timber canbe average prices for a single harvest ordifferent prices can be set for differentmixtures of products for up to three harvestcycles. In the case of native forestryactivities users can again set an averageprice for all the forest products for variousareas or else they can set the proportion ofthe harvests that will be allocated to variousproducts with individual prices for eachproduct.

8. DISCUSSION

The use of the Visual Basic (VB)programming language has a number ofadvantages over the use of Excel linked toVisual Basic macros. In editions of Excelsince 1997 have used the Visual Basiclanguage to construct macros. Macros arecomputer code written in Visual Basic

language. Macros can be used to automatedata processing and reporting, navigatethrough spreadsheets, carryout dataanalysis, and provide input and reportingforms. Macros used in Excel require Excelsoftware to operate. Using the Visual Studiosoftware package allows users to utiliseelements of the Windows operating systemand functions from Microsoft software suchas Word, Access and Excel to create theirown software programs. Programmers canaccess virtually all the functions of Excel,Access and other popular Microsoftprograms with computer code plus createspecific data input, analysis and reportingforms, as well as help systems, all similar tothose in standard commercial software.Once the code for the data input, analysisand reporting has been written it can becompiled into software that can be copied,distributed and installed onto mostcomputers with Windows operatingsystems.


Using computer code to develop theAFFFM has other advantages over the useof macros linked to Excel workbooks. Theuse of a programming language allows thedevelopment of more complex models thatbetter represent real-life situations. Oneexample is the plantation module of theAFFFM. Whereas the ACTFM allowed onlyone harvest age and timber price perspecies in a plantation, the AFFFM allowsusers to use up to three harvest ages perspecies and up to 8 product categories perharvest. The processing speed of programswritten in Visual Basic relative to that ofExcel has allowed this improvement in thesophistication of the model. It has alsoallowed the incorporation of the financialdetails of other farm enterprises, and formore options for users in each part of themodel. These functions were not possible inthe Excel based ACTFM because of limitsto the file size that could be handled bymost computers. Other features of theAFFFM such as the links to text files andthe incorporation of an on-line help systemare not possible in Excel.

The expansion of the ACTFM to a whole-farm model allows it to be used to answer aquestion basic to financial appraisals thatcould not be handled by the ACTFM, canthe new enterprise be financed? TheAFFFM still allows for financial appraisals ofindividual enterprises like the ACTFM aswell. By looking at the cashflows andfinancing costs and revenues of forestryenterprises in conjunction with thosespecific to the individuals other farm andnon-farm enterprises it is possible to assess

the effects of different rates and scales ofplantation establishment, plus the impactsof finance costs and revenues.

Like all models, the ACTFM and AFFFMcannot perfectly simulate what will occur interms of the risk factors listed in Table 2.While the AFFFM does have some featuresto calculate the interaction between forestryand agricultural activities, these are limitedand no default data is supplied for theseparameters. Agricultural returns remainstatic in terms of the gross margins used tocalculate returns ignoring the likelihood ofclimatic and market factors changing overthe course of a scenario. Likewise timberprices are assumed to remain static. Someform of sensitivity or risk analysis should beused for key variables to determine the riskprofile of the scenario.

REFERENCES

Hackett, C. (1991), Plantgro: A SoftwarePackage for Coarse Prediction of plantGrowth, CSIRO, Melbourne.

Harrison, S.R. and Herbohn, J.L. eds. (inpress), A Whole-Farm and RegionalAgroforestry Decision-Support System: Areport for the RIRDC/Land and WaterAustra l ia /FWPRDC Joint VentureAgroforestry Program, Canberra.

Herbohn, J.L, Harrison, S.R. and Emtage,N.F. (1999), ‘Mixed species plantations ofrainforest and eucalyptus cabinet species inNorth Queensland’, Australian Forestry,62(1): 79-87.

133

13. Estimating and Integrating Non-market Benefitsof Forests in Project Appraisal

Jungho Suh

Forests generate a variety of environmental effects, including watershed protection, airpollution reduction, flooding mitigation, wildlife habitat and public recreation. Theseenvironmental benefits are not marketed and not site specific. Thus, they need to beconsidered from a social perspective. In recent years, a number of applications have beenmade to estimate non-market environmental values in monetary terms and incorporate theseestimates into cost-benefit analyses in forestry research projects. This module provides anintroduction to the concepts and methods of non-market valuation by which to estimate theaggregate benefits to society of environmental improvement. In the next section, thecategorical framework of economic value is introduced. Non-market valuation methods arethen discussed, including the hedonic price method, the travel cost method, the contingentvaluation method and conjoint analysis. A contingent valuation survey of the KakaduConservation Zone in Australia is examined in particular. A few alternative valuationtechniques are next briefly reviewed. Finally, ways of integrating non-market values inproject appraisal are discussed.

1. THE ECONOMIC VALUE OF AFORESTLAND

As long as attempts are made to quantifythe value of non-market environmentalgoods and services, the term ‘value’ isconfined to economic value. In the resourceeconomics literature, the economic value ofa forestland is usually broken into use value

and non-use value, the latter also beingreferred to as passive use value. The useclass of economic value consists of directand indirect use value, and option value.The non-use class of benefits falls into twosub-categories, namely bequest value andexistence value. Figure 1 illustrates the useand non-use values that a forestlandprovides.

Economic value

Use value Non-use value

Direct use Value

Indirect usevalue

Option valueQuasi-optionvalue

Bequest value Existence Value

Timber, harvesting, recreation, medicine, education

Nutrient cycling, air pollution reduction

Biodiversity, preserved habitats

Habitats, prevention of irreversible change

Habitats, species, genetic, ecosystem

Figure 1. The economic value of a woodland

Source: Adapted from Bateman and Turner (1993) and Barbier (1994).


Direct use value is made up of consumptiveand productive use value. Indirect use valuerefers to values conferred by forests suchas carbon sequestration and waterpurification. Option value is defined as thepotential use benefit, opposed to presentuse value, of an environmental good. Thevalue is viewed as the willingness-to-pay(WTP) for preservation of a naturalresource that will be made use of at a laterdate by the present generation. Bishop(1982) provided an excellent review of theevolution along with an extension of theconcept of option value. It is known thatWeisbrod (1964) originated this concept byproposing that many individuals expect theymay possibly visit a national park forexample and are willing to pay for an optionthat would guarantee their future access.

Bishop (1982) extended the concept ofoption value with supply side uncertainty. Ifa risk averse consumer is certain ofdemanding the services of anenvironmental asset in the future anduncertain about its future availability, thereexists a positive option value. That is, themaximum WTP to avoid the risk to thesupply of the environmental resource islarger than the expected loss. This conceptis grounded on the fact that an individual inthe real world will be willing to pay morethan the expected consumer surplus inorder to ensure that he or she can makeuse of the environmental resource later on.Edwards (1988) reported empiricalevidence of positive option value from astudy of households’ WTP to preventuncertain future nitrate contamination ofgroundwater in Cape Cod, Massachusetts.Bishop noted that option value ceases toexist in the case of supply side certainty.

Quasi-option value is present when there isuncertainty about future availability of anatural resource given some expectation ofthe growth of knowledge. For example,there are uncertain benefits for scientific orcommercial purposes from the preservationof a tropical forest, which could becomemore certain through time as information isaccumulated about the uses. Arrow andFisher (1974) originally introduced theconcept of quasi-option value in the contextof an irreversible development decision.Quasi-option value is always positive if theexpected growth of information is

independent of a proposed development ofa natural resource. In contrast, as Freeman(1984) argued, if the uncertainty is primarilyabout the benefits of development, thisstrengthens the case for development. Thatis, quasi-option value is negative. Whateverthe case is, the presence of quasi-optionvalue implies that the value of additionalinformation is likely to be of greatestimportance in valuing goods subjectirreversible changes (Mitchell and Carson1989).

Krutilla (1967) argued that many personsmay be willing to pay for the satisfactionderived from knowledge of the bequest ofunique environmental resources to futuregenerations. Thereafter, bequest value isoften defined as the benefit accruing tocurrent generations from knowing thatfuture generations will benefit from theresources. This concept takes a strongstance for intergenerational moral duty soas to prevent future sufferings from theshortage or degradation of the environment.

Pearce and Turner (1990) definedexistence value as a value placed on anenvironmental good that is unrelated to anyactual or potential use of the good. Theconcept of existence value becomescomplicated because it is sometimes mixedwith that of intrinsic value of a resource.Existence value is often recognised on thebasis of ecocentric value orientation thatnature has the right to exist for its ownsake, and destruction of species andwilderness is intrinsically wrong. From thisviewpoint, Aldred (1994) insisted thatexistence value captures a WTP topreserve the environment for the continuedexistence of particular species or wholeecosystems. He argued that existencevalue may include intrinsic value, but theyare not the same thing. Solow (1993)supported the view that particularlandscapes or species have to bepreserved for their own sake because theyare intrinsically important to preserve.Pearce and Turner (1990) stated thatintrinsic value equals existence value aslong as the former is defined as somethingunrelated to human use but captured bypeople through their preferences in the formof non-use value. On the other hand,Mitchell and Carson (1989) saw the term‘intrinsic’ being contradictory to the term

Estimating and Integrating Non-market Benefits of Forests into Project Appraisal 135

‘economic’ so that the intrinsic value cannotbe part of economic value. A confusing useof the term ‘existence value’ can be avoidedby supposing that existence valueembodies the welfare of particular speciesand whole ecosystems, but ultimatelyappeals to human welfare. In other words,existence value may stem from intrinsicmotives such as sympathy, responsibilityand a concern about the state of the worldthat some people may feel towards non-human beings, but the value is stillanthropocentric and does not reveal theintrinsic value to non-human beings(Bateman and Turner 1993). In this context,Bateman and Langford (1997) clarified thatexistence value is a human value whereasintrinsic value is a non-human value, whichcannot be estimated.

2. CLASSIFICATION OF THEBEHAVIOURAL LINKAGE APPROACHTO NON-MARKET VALUATION

Smith and Krutilla (1982) dividedmeasurement techniques of environmentalbenefits into the physical linkage approachand the behavioural linkage approach.Under the category of physical linkageapproach, a researcher can specify a modelof the relationship between levels of anenvironmental contaminant and some typeof observed damage such as reducedagricultural crop yields or impaired humanhealth. Linked with physical data, thebenefit of the reduction in the contaminantcan be estimated in dollar values. When

there is no such physical link to beobserved, an alternative is the behaviourallinkage approach. This approach uses aconceptual linkage between the servicesprovided by environmental resources andsome directly or indirectly observableconsumer responses. The behaviourallinkage approach for valuing environmentalamenities is traditionally divided into fourcategories of economic valuation methodsas presented in Table 1.

With the indirect and revealed preferenceapproach, the information from revealedmarkets – that is, actual choices made byconsumers – is used to develop models ofchoice. One problem is that data on markettransactions and product characteristics areoften unavailable or incomplete forenvironmental goods.

When revealed market data are not reliableor are unavailable, economists have usedthe stated preference approach that relieson hypothetical market situations. As atypical example, the economic value withrespect to a change in the level ofenvironmental service flows of an unpricednatural resource can be estimated byeliciting consumers WTP or willingness-to-accept compensation (WTA) amounts forthe proposed change. WTP measures givewelfare estimates for quality-improvingchanges in resource use, whereas WTAmeasures provide information about welfaredecreases resulted from quality-decreasingenvironmental moves.

Table 1. Behaviour-based valuation methods of environmental public goods

Type of valuationapproach

Direct Indirect

Revealed (observed)market behaviour

DIRECT and REVEALEDReferenda

Simulated markets

INDIRECT and REVEALED Travel cost method Hedonic price method

Stated (hypothetical)markets

DIRECT and STATED Contingent valuation methodAllocation games with tax Refund

INDIRECT and STATED Contingent ranking Contingent rating Choice modelling

Source: Adapted from Mitchell and Carson (1989, p. 75).

3. THE HEDONIC PRICE METHOD

The hedonic price method (HPM) has beenwidely used to estimate the externalities of

environmental characteristics, particularlyair quality and visual amenity, in thehousing market. A botanical garden thatgives pleasure to passers-by is an example


of a positive externality. By contrast, anegative externality occurs in the situationin which air pollutants such as sulfur dioxidecan make life miserable for peoplebreathing the air and cause healthproblems. Since the effects associated withproduction of botanical gardens orconsumption of air pollutants extendsoutside the market to some third party, theyare termed ‘externalities’. In HPMapplications, it is assumed that externalitiesare capitalised into the value of realproperties. Differentials in property pricesare then examined to find the marginalvalue of an environmental good underinvestigation. One problem with thistechnique is that it is often difficult to obtainan adequate sample of property transactionrecords.

Garrod and Willis (1992) investigatedwhether the amenity benefits of ForestryCommission estate in the UK are reflectedin the values of nearby properties. Anumber of independent variables in thehedonic price model were tested, includingthe forest characteristics such as theproportions of Forestry Commission estateareas covered by broadleaved trees. Theauthors found that selling price increasedby about A$111 from 1% increase in therelative proportion of forested area in agiven 1km2 to broadleaf woodland, with allother independent variables held at theirmean values. The aggregate amenitybenefit, which Forestry Commission estateprovides to those households that live inclose proximity, was found to be aboutA$913,000.

Tyrvainen and Miettinen (2000) examinedthe relationship between property pricesand urban forest amenities of Salo, which islocated 110km to the northwest of Helsinki.The town of Salo had approximately1,100ha of forest, 74% of which was foundin the urban fringe. Detached houses orterraced houses predominated in mosthousing areas.A range of explanatory variables was testedincluding housing attributes, locationalattributes and four different variablesmeasuring forest amenities. Forest amenityvariables included distance to nearestwooded recreation area, direct distance tothe nearest forested area, relative amountof forested areas in the housing district and

the view from the dwelling window. Thisstudy found that a 1km increase in distancefrom a forest park reduced terrace houseprice by 5.9%, and a forest view increasedterrace house price by 4.9%.

4. THE TRAVEL COST METHOD

TCM was the first technique ever used tomeasure the demand for natural resourcesfor recreation purposes in terms of whatpeople spend in travel costs to visit them.As outlined by Bateman (1993), Hanley andSpash (1993) and Bennett et al. (1996), theTCM was conceptually first suggested byHotelling in a letter to the director of the USPark Service in 1947. The letter suggestedthat “the price for visiting a park or othernon-marketable recreational area (even onefor which entry is free) would vary accordingto the travel costs of visitors coming fromdifferent places” (reported by Portney,1994, p. 4). This suggestion was formallyintroduced to the literature by Wood andTrice (1958). Clawson (1959) modified theidea and first developed empirical models.The TCM was further developed by Knetsch(1963) and Clawson and Knetsch (1966).

In practice, two types of TCM are used. Thezonal method (ZTCM) divides the entirearea from which visitors originate into a setof approximately concentric zonesemanating from the recreation site,representing increasing levels of visit cost,and then defines the dependent variable asthe ‘visitor rate’ – the number of visits madefrom a particular zone in a period divided bythe population of that zone. The individualmethod (ITCM), in contrast, simply definesthe dependent variable as the individual’sannual number of visits to a site, and theindependent variable as the cost oftravelling to the site. Then, the recreationaldemand curve can be produced. With theITCM, it is possible to involve socio-economic variables under the hypothesisthat they are separate factors influencingtravel behaviour. The ITCM does notrequire data about zonal visitor rates.Cooper and Loomis (1990) mentioned thatthe ZTCM model has several drawbacksrelative to the ITCM model, includingstatistical inefficiencies from grouping dataand a less direct link to consumer demandtheory. Hence, in these aspects, the ITCMseems to be preferable. On the other hand,


Fletcher et al. (1990) came to the view thatthe zonal approach more adequatelyaccounts for declining participation rateproportioned to increasing travel distances.Uncorrected estimates of the ITCM maylead to biased results in the circumstancewhere visitation rate to a site declines asdistance to the site increases. For thesereasons, many travel cost practitionershave adopted Clawson and Knetsch’s(1966) ZTCM.

Suppose there is a recreation areaattracting people from three zones havingpopulations of 1,000, 4,000 and 10,000people, respectively as illustrated in Table2. Visitors from Zone 1 who have a cost of$1 participate at the rate of 300 per 1,000. Ifthe cost to the visitors from Zone 1increased to $2, the new visitation ratewould be the same as that of Zone 2, i.e.200 per 1,000.

Table 2. Demand schedule for recreation experience for a hypothetical area

Zone Population Cost per visit Number of visits Visits per 1,000

1 1,000 $1 300 300

2 4,000 $2 800 200

3 10,000 $3 1,000 100

Using data presented in Table 2, one canmap the recreation demand curve for thehypothetical area by plotting visit ratesacross all zones against visit costs. Thetotal consumer surplus for recreation is thenobtained by integration of the area underthis demand curve. In addition, using thisdemand curve, one can estimate theresulting change in consumer surplus with achange in environmental quality of the site.If the demand for the recreational use of anenvironmental resource such as a nationalpark increases as its quality improves, thedemand curve shifts outwards under theassumption that the quality of a naturalresource is positively correlated with itsrecreational use value. Then, the monetisedincremental benefits to visitors of improvingthe quality of the park can be estimated.

It should be noted that the demand curverelating visitation rate and cost per visitindicates demand for the ‘whole recreationexperience’ rather than the recreational usevalue of the resource alone. The wholeexperience includes travel to andexperience on the site, travel back andrecollection. This distinction is importantwhen the researcher intends to estimate therecreation value attached to a particular

natural resource. In order to isolate the on-site recreation experience from the wholeexperience, Clawson and Knetsch (1996)suggested constructing the new demandschedule for the resource by plotting addedentry fee to the site as a proxy for the pricevariable and total visits as the quantityvariable.

Suppose that the entry fee on visits to thesite has increased. One can next determinethe effect of a rise in the entry fee of say $1.The visitors from Zone 1, for example, whoused to spend $1 per visit, are now facedwith the situation of spending $2 due to therise in overall cost as much as $1. People inthis region would participate in the samerecreation activities on the site at the rate of200 per thousand, as indicated in the Table2, when faced with costs of $2 per visit.One can then estimate the number of visitsunder varying added travel cost using thisresult. The number of recreatists attendingfrom Zone 1 would be 200 – i.e. 200 perthousand multiplied by 1,000, the basepopulation of Zone 1. The numbers of visitsto the site from other zones can be found inthe same manner. Table 3 reveals theestimated total number of visits to the siteunder varying added entrance fee from $1

to $3. It can be seen from the table that a$3 rise would result in choking off all visits.The regression of the estimated number oftotal visits on added travel cost willeventually generate the new demand curve,from which on-site recreation benefits can

be estimated.

A number of TCM applications (e.g. Everitt1983; Willis and Garrod 1991; Bennett1995) have demonstrated that forestlandscan have substantial recreational use value


and TCM is a powerful tool to estimaterecreational use value of a particularforestland. For example, Bennett (1995)estimated economic value associated withrecreational visits to two National Parks inthe north-east of New South Wales,Australia. Dorrigo National Park forms twoamphitheatres of 3,600ha from thetablelands to the coastal plain, and is the

habitat for a wide variety of flora and fauna.Gibraltar Range National Park of 17,300hais noted for its high number of rare andthreatened plants. The park is essentially ahigh plateau with granite occurring overmuch of its area. Bennett (1995) estimatedrecreational value of A$34 per visit toDorrigo National Park and A$19 per visit toGibraltar Range National Park.

Table 3. Effect of increases in travel cost on visits to a hypothetical area – deriving ademand schedule

Zone Number of visitors at added entrance fee

$0 $1 $2 $3

1 300 200 100 02 800 400 0 03 1,000 0 0 0Total visitors 2,100 600 100 0

There are many continuing controversiesover TCM. The practices of time valuationin TCM are the most controversial. Fletcheret al. (1990, p. 125) pointed out that “manyof the most vexing problems encountered inTCM literature, both theoretical andempirical, relate to the appropriate inclusionof time constraints and time values”.Likewise, Randall (1994) argued that thelevel of money-valued welfare measures inTCM is inherently subjective because itdepends on arbitrary and simplisticspecifications of household productiontechnology and particular accounting oranalytical conventions for the household’simplicit cost of time. Another fundamentalproblem with TCM is that the valuationtechnique cannot be applied to someforestlands that people do not visit forrecreation purposes. Harrison (2001) drewattention to a number of other complexitiesarising in application of TCM. First, whenpeople visit multiple sites during a singlerecreation trip, it can be difficult to allocate aproportion of their travel cost to a specificsite, which is the target for valuation.Second, since recreation demand typicallyis highly seasonal, and has peak visitationduring school and public holidays, it isnormally necessary to carry out surveys ona year basis. Third, where there is a groupvisit, with members of varying ages, theissue arises of which members to includeas recreationists and how to allocate costsbetween party members.

5. THE CONTINGENT VALUATIONMETHOD

The CVM question involves askingindividuals how much they would be willingto pay or whether they would be willing topay a given amount to prevent a specifieddecrease in the quality or quantity of aparticular good, or how much they would bewilling to pay to obtain improvements. WTPvalues elicited from CVM surveys are‘contingent’ upon a hypothetical marketdescribed to the respondents, hence thisapproach came to be called the contingentvaluation method. Mitchell and Carson(1989) provided the full history of earlydevelopment of CVM, which first came intouse in the early 1960s by Davis (1963).

CVM has been widely applied in the lastthree decades. Mitchell and Carson (1989)listed more than 120 CVM studies, most ofwhich were undertaken in the USA andEurope. Carson et al. (1995 cited in Bennett1996, p. 190f) listed 2,131 CVM studies andpapers applied to health, education andtransportation as well as the environment.

Major bidding methods are continuous anddiscrete ones. The continuous methodusually refers to as the open-endedelicitation method, where respondents aresimply asked to state their maximumwillingness to pay for the good being


valued. The discrete method refers to asthe dichotomous choice question, whererespondents determine whether their WTPis larger or smaller than a set dollar value.The dichotomous choice format is alsoknown as a take-it-or-leave-it or closed-ended format, or a referendum format if thequestion is structured so that it is similar toa referendum.

Some examples of open-ended CVMapplications include Mattsson and Li (1994)and Hadker et al. (1997). Mattsson and Li(1994) estimated non-timber use values offorests in the county of Vasterbotten innorthern Sweden. Forests in the county aredominated by pine and spruce. This CVMstudy examined a variety of on-site humanactivities such as camping, berry andmushroom picking, hiking or simply takingwalks, as well as the off-site visualexperience of the forests. Respondentswere asked the amount that they would bewilling to pay annually for using orexperiencing the non-timber commodities tothe level that they currently use orexperience them.

Hadker et al. (1997) estimated thepreservation value of the Borivli NationalPark in Bombay. Notably, this study is oneof the first reported uses of CVM in India(cited from www.epa.nsw.gov.au/envalue).The Borivli National Park of 104km2

accounts for one-fifth of the BombayMetropolitan Region. The national parkincludes the water bodies that supplydrinking water to Bombay and is home for alarge number of endangered mammals,birds and reptiles. The park being valued isthe most popular park in India. According tothe authors, about 2.5 M people visit thenational park each year. The park has beenfacing financial constraints and its potentialas a recreational spot and habitat for wildlifehas been deteriorating. In this study,Bombay residents were asked how muchthey would willing to pay to maintain andpreserve the Borivli National Park byfunding an autonomous management body.The mean household WTP of Bombayresidents was found to be 7.5 INR permonth for 5 years. The authors noted thatdespite India being a developing countrywith medium to low income levels, theempirical evidence suggested people are

willing to pay for preserving environmentalamenities.

The open-ended CVM question istheoretically vulnerable to strategic bias.This bias occurs in either one of twodifferent ways. Respondents mayunderstate their WTP for a welfare-improving change if they believe that otherswill bid sufficiently high to provide thedesired quality of an environmental good.By contrast, respondents may overstateWTP for an environmental benefit, whenthey believe that the possibility of theimprovement going ahead will increase ifthey bid high. For this reason, open-endedformats had largely been superceded by thedichotomous formats in CVM applications(Blamey 1996).

The dichotomous choice format is the mostcommonly employed format in CVM studiesnowadays. Using the dichotomous CVM,researchers may ask people as to whetherthey would pay a specific amount of dollars($X) if conditions of a natural resource were

changed from { 002

01 k,z,,zz ⋅⋅⋅ } to { 11

211 ,,, kzzz ⋅⋅⋅ },

where 0kz represents the kth attribute under

the current status and 1kz represents the kth

attribute under an alternative. It is notnecessary to have all attribute levels in thenew option distinct from the current status.

Respondents are expected to answer ‘yes’– i.e. to willing to pay $X – if the utility withthe new attribute levels and $X decrease inincome is greater or at least as much as thestatus quo utility. From the response,researchers can elicit directly eachrespondent’s WTP for the specifiedenvironmental changes. The basic modelfor an individual who is willing to pay $X forsome changes in environmental quality isthus:

V (E1, M0 – X) ≥ V (E0, M0) (1)

where X is the individual bidding price and Edenotes the quality of the natural resource.V represents utility or satisfaction, which isa function of the individual’s income M andthe quality of the resource. With an amountof money M0, the individual can purchaseenvironmental or non-environmental goodsand services. The maximum that a person


is willing to pay for promotion ofimprovements in the quality of the particularpublic good from E0 to E1, will be such thatboth sides of the above expression becomeequal.

Imber et al. (1991), as a typical example ofthe close-ended CVM application, used theCVM technique to estimate the dollar valueAustral ians place on the KakaduConservation Zone. The zone is located inthe boundaries of the 20,000km2 KakaduNational Park, but was not part of thenational park. Most of the Kakadu NationalPark is registered on the World HeritageList. The park is annually visited by about230,000 people, and renowned for itsgeographic features, wetlands, wildlife andscenic vistas. Mining was proposed on theconservation zone. The extent ofenvironmental damage from the proposedmine was in dispute. Some people arguedthat the conservation zone should be madepart of the Kakadu National Park, whereany mining activities are not allowed. HPMis not applicable to estimate the value of theconservation zone simply because anadequate sample of property transactionrecords cannot be obtained. Nor is TCM anappropriate method due to the occurrenceof part-whole bias. This bias occurs wherethe good being valued by the respondentdiffers from the commodity that is intendedto be valued. In other words, respondentscan experience difficulty in distinguishingbetween the environmental value of theconservation zone and the whole nationalpark. This study found that Australians arewilling to pay about A$124 per person peryear for 10 years to avoid the effects of themajor impact scenario, and A$53 to avoidthe effects of the minor impact scenario.The mining proposal was finally withdrawndue to a variety of factors, including thehigh conservation values.

CVM mimics real market transactions.Suppose that a person comes to ashopping mall to buy a T-shirt. Beforepurchasing a shirt, the person needs tochoose the desirable size, colour anddesign of the shirt, and most importantly hisincome level. Likewise, the CVMquestionnaire must be framed in such away that respondents are made awarethere are an array of substitute andcomplementary goods for the environmental

good under consideration, and also budgetconstraints. Otherwise, framing bias mayoccur. For this reason, Imber et al. (1991)asked several preliminary questions for theframing purpose, gradually bringingrespondents to the WTP question. Thepreliminary questions included: “Do youthink Australia needs to concentrate moreon protecting the environment, or more ondeveloping the economy, or would you saywe currently have a reasonable balance?”;and “What do you think the two or threeenvironmental matters most important toAustralia right now?” To give sufficientinformation to respondents, Imber et al.(1991) also showed respondents a set ofcolour photographs of the site and itssurroundings after respondents completedthe preliminary section. If there isinsuff icient information about thecommodities being valued, respondents’WTP may not be equivalent to their actualWTP. In giving information at this stage,CVM interviewers must be in the neutralposition so as to prevent potentialinterviewer bias.

Respondents were then asked whether theywould be willing to have their incomereduced by A$2 a week for the next 10years to add the Kakadu ConservationZone to the Kakadu National Park ratherthan use it for mining. If a respondentanswered yes about whether he or shewould pay A$2 a week, another WTPquestion was asked using a higher price ofA$5. If the answer was no, lower price ofA$1 was then proposed. Respondents weresplit into a few sub-samples, each sub-sample being asked if they are willing topay the distinct combination of dollaramounts. This is necessary to ensure thatthere is enough variation in the WTPvariable to carry out statistical analysis on it.The range of maximum WTP amounts waspredetermined through focus groupmeetings.

It is strongly recommended to make use offollow-up questions. Socio-economic datasuch as age and income need to beobtained to test whether these variablesinfluence the WTP responses. One or morefollow-up questions are also required tocheck whether respondents understand thechoice they are being asked to make and todiscover the reasons for their answer. Imber


et al. (1991) included an interesting follow-up question: ‘Do you recycle things such aspaper or glass?’ This question wasdesigned to determine whetherenvironmentalists predominated within therespondents and whether the attitudes hadan impact on the WTP bids.

Apart from the possible occurrence ofbiases that are inherent to non-marketvaluation or generated by the questionnairedesign and survey processes, CVM hascome under rather philosophical criticism.Most critiques are concerned about thecategorical mistake. According to Sagoff(1988), CVM has its theoretical grounding inmeasuring consumer preferences, but thismethod is inappropriate for measuring non-use value that is related to citizenpreferences. In other words, if respondentsare motivated to bid WTP by the role ofcitizen in a value survey, they areconcerned with the public interest, ratherthan their own self-interest. Thus, whenanswers to CVM questions do not arisefrom an expression of underlying consumerpreferences, CVM estimates are not asuitable source of information about valuesfor CBA (Diamond and Hausman 1993). Bythe same token, Knetsch (1994) stated thatCVM responses are more likely to be anindication of an attitude or good feeling ofmoral satisfaction than of economic value,and therefore CVM results can provide littleor no guide to resource allocation policies.

6. CONJOINT ANALYSIS

Conjoint analysis is a variant of CVM in thesense that both are used to measurepreservation value of non-marketenvironmental goods based on hypotheticalbehaviour and they require surveyrespondents to trade off dollars forattributes. In conjoint analysis applications,respondents are asked to rank or ratemultiple profiles, or choose only one optionfrom the given number of options. In theenvironmental economics literature, thesequestion formats are named ‘contingentranking’, ‘contingent rating’ and ‘choicemodelling’, respectively. Compared to CVM,a single conjoint analysis exercise canseparately and simultaneously estimate thecoefficients of all factors involved in choicesets.

7. SOME FURTHER VALUATIONMETHODS

In the economic valuation literature,alternative valuation methods that have notbeen discussed in this module so farinclude the production function approach(Barbier 1994), composite methods of TCMand CVM, the contingent activity method(Heyes and Heyes 1999) and benefittransfer.

The production function approach is uniquein the sense that it views the environmentas an input into production process tocapture the value of ecological functions.For example, Narain and Fisher (1995)used this approach to model thecontribution of the Anolis lizard to pestcontrol. Despite provision of a useful way inwhich to value elements of environmentalfunctions, the approach requires detailedknowledge of the physical effects onproduction of changes in an environmentalresource (Barbier 1994), and often entails anumber of assumptions due to the paucityof adequate data on how an environmentalfunction is linked to the production of othergoods (Acharya 1998).

Hoagland et al. (1995 cited in Davis andTisdell 1996) suggested the use of multiplevaluation techniques to help correct for anypotential biases attached to a specifictechnique. For instance, the ‘hedonic TCM’has been in place as a class of methods forvaluing non-market public goods. Themethod has been applied to value thecharacteristics of a destination such asproportion of the forest as open space,presence of water features and diversity ofspecies, using revealed preference data(e.g. Englin and Mendelsohn 1991; Hanleyand Ruffell 1993).

Davis and Tisdell (1996) used a compositeof TCM and CVM to value the consumersurplus of scuba divers at Julian RocksAquatic Reserve, a small marine protectedarea in New South Wales, Australia. Threepoints were estimated: (1) the cost thatconsumers were currently confronting intraveling to and accessing the attractionand the annual number of visits theyundertook each year at that price, as inTCM; (2) the upper limit to the cost ofvisitation beyond which the consumer would


cease to access the site, as in CVM; and(3) the frequency with which they wouldwilling to access the site each year if thecost was half way between their actualcosts and upper limit, data of which werealso obtained by a CVM question.

Heyes and Heyes (1999) developed‘hypothetical TCM’. Instead of being askeddirectly to place a dollar value on ahypothetical recreational site, respondentsunder the hypothetical TCM are asked themaximum additional distance they wouldhave been willing to travel in order toaccess such a site. By converting these intomonetary equivalents, alternative measuresof consumer surplus can be derived. Heyesand Heyes (1999) drew attention to theirfinding that the estimates derived in thisway lay somewhere between those takenfrom the TCM and CVM analyses. Theystated that the principle advantage of thismethod is that it can be expected to reducethe extent of protest-motivated bidding.That is, respondents are in effect beingasked how much they would be willing topay for access to the site, but the paymentis expressed in a non-monetary currency,i.e. travel distance.

Another possible approach is to transfer thebenefit or cost assessment estimated fromprevious studies to a current valuationproblem. The practice of benefit transferhas been developed because the high timeand other resource requirements oftenprevent new valuation studies. Willis andGarrod (1995) suggested constructing adatabase of all environmental benefitestimates with details of the modellingprocedures and all relevant assumptions inorder to make further progress of valuationand benefit transfer studies. Morrison(2001) provided a review of three currentlyavailable databases, viz. ENVALUE (NewSouth Wales EPA Environmental ValuationDatabase), EVRI (Environmental ValuationResource Inventory) and NZDB (NewZealand Non-Market Valuation Database).1

1 These are found at the wetsites: www.epa.nsw.gov.au/envalue;

www.evri.ec.gc.ca/evri/; and learn.lincoln.ac.nz/markval/. The Department of the Environment, Transport

and Regions of the UK provides anenvironmental valuation source-list for the UKat: www.detr.gov.uk/environment/evslist/.

8. INTEGRATING NON-WOOD VALUESOF FORESTRY PROJECTSAPPRAISAL

A question often arises as to how toevaluate a new land-use option for aforestland. Conceptually, there are severalsolutions to the problem of integrating non-wood values in projects appraisal.

Cost-benefit analysis

Assume there are only two options for aparticular forestland area. For example, agiven habitat can be either preserved ordeveloped. The preservation option is notavailable. A conventional evaluation methodfor dealing with discrete development-preservation options is social cost-benefitanalysis (CBA). The basic CBA rule forthese two options is to compare the netbenefit of each option. Whether to preservea forestland or develop it from society’sstandpoint is dependent upon the differencebetween the net present value (NPV) ofthese exclusive choices. The application ofthe traditional CBA rule as to whether aproject can be accepted can be expressedas:

0)(1

NB-NBNPV >

+=∑ t

PtDt

r (2)

where r is the discount rate and NBDt andNBP t are net development benefits(development benefits minus costs of thedevelopment option, i.e. input costs) andnet preservation benefits (preservationbenefits minus costs of the preservationoption such as policing, maintenance andmonitoring costs) in year t, respectively.

Safe minimum standard

Pearce and Turner (1990) and Tisdell(1991) discussed the ‘safe minimumstandard’ (SMS) as an alternative to CBA ofdiscrete options on an irreplaceable naturalresource. The SMS approach stresses thatone should avoid irreversible environmentaldamage unless the social cost of doing sois unacceptably high. The rule is intendedfor minimising the maximum losses, which


is called a minimax strategy. Assume thereis no compromise between preservationand development, that is, there are onlydichotomous choices. For example, if aforestland is developed, the maximum lossis the preservation benefit (BP).Alternatively, if the given habitat ispreserved, the maximum loss is thedevelopment benefit (BD).

In effect, the SMS amounts to comparingthe net present values of BD and BP. Thelower BD is, the more the minimax solutionis likely to be the preservation option. Inother words, the SMS approach makespreservation the preferred option unless thesocial cost of preservation (i.e. the forgonedevelopment benefit) is unacceptably large.This rule appears precise except that itrequires value judgement to determine whatis unacceptably ‘large’. Pearce and Turner(1990) pointed out that the requirement ofthe SMS approach may be a deliberateattempt with the aim of not relying on asingle criterion for making discrete choices.Tisdell (1991) pointed out that the SMSapproach is in fact more supportive ofpreservation than the CBA rule, consideringthat the size of BP is likely to be though it isnormally believed to be greater than BD.

Ecological approach

Ecological approaches to project evaluationstress that ecological consequences andsustainability must be taken into accountseriously before making irreversibledevelopment decisions. This notion ofsustainable development was an importantagenda item at the 1992 Earth Summit inRio de Janeiro, Brazil. Economistsconcerned about sustainable developmenttend to argue that the conventional CBA isa mere monetary analysis. From theviewpoint of the extremely strongsustainability conditions (Daly 1980; 1996),however, sustainable development is alsodangerous, and the term is an oxymoron,and that mankind must be saved from thisdangerous path.

Multi-criteria analysis

Multi-criteria analysis (MCA) – also knownas multiple objective decision-support

systems (MODSS) – is a structuredframework for evaluation of various policyoptions across multiple objectives, whichare often conflicting with one another. Inthis technique, the various criteria thatreflect the multiple objectives underconsideration, and the decision options, arelisted in an effect table. This is an m × nmatrix with m criteria and n options. Thedata used are either qualitative orquantitative. In order to determine therelative value of each decision option,importance weights are attached to eachcriteria. The qualitative data are ranked orput on a scale, and then combined with thequantitative data. The preferred option iscomputed by a series of mathematicaltechniques (Cameron 1992; Hajkowicz etal . 2000). Bennett (2000) noted whatdifferentiates MCA from CBA is that in CBAthe weights used are the per unit valuesderived for each of the impacts, whereasthe weights used in MCA are driven by theanalyst.

In comparison to CBA, MCA does not haveto involve the conversion of all costs andbenefits associated with a project intomonetary terms. Cameron (1992) pointedout that MCA has the advantage in avoidingthe risk of spurious quantification of non-market values: that is, the difficulties ofconverting to dollar value can be avoided byleaving the results of qualitativeassessments of environmental values in aqualitative form. Gurocak and Whittlesey(1998) came to the parallel view that publicprojects often have a variety of economic,ecological, social and political objectives,many of which cannot or perhaps shouldnot be converted to monetary terms. It isnot necessary, however, to leave asideenvironmental value in a qualitative form.Values which can be quantified by a suitedvaluation method can be included in theMCA process (Cameron 1992).

The common disadvantage of MCA is thatthis method requires decision makers toidentify preference weights for the criteriainvolved in the decision process (Cameron1992). In this context, MCA tends to bemuch dependent on the subjectivity ofpreference weight inputs from decisionmakers (Gurocak and Whittlesey 1998).Subjectivity itself is not necessarilyundesirable but may cause an inconsistent


framework in making the unavoidable hardchoices, diminishing the effectiveness ofMCA.

9. DISCUSSION

Several techniques have been developingto estimate non-market values. Fairlyobviously, it is necessary to seek the mostappropriate technique. Harrison (2000)came to the point that although choice ofvaluation technique becomes complex inreality, simple statements can be made as arough guide. When the task is to value therecreation benefits at a recreation site, TCMis likely to be appropriate. Nevertheless, it isdoubtful whether TCM is appropriate forcapturing the value of a specificcharacteristic of the recreation site. CVMneeds to be considered when social welfarechanges associated with an environmentalchange within a recreation site is to beestimated.

There are various categories of non-timbervalues associated with forestlands, fallingunder the headings of indirect use value,option value and existence value. Evidencefrom current Philippines experiences is thatextreme deforestation results in not only atimber shortage, but also environmentalproblems in watersheds. Other problemsincluding loss of human life due toincreased flood severity, damage to cropsand fisheries, reduction in the life ofhydropower generators, which are causedby flooding, are also attributed in part todeforestation. Further, some may concernfor the fauna and flora issues ofdeforestation, for example, loss of nativeanimals. The damage function method orproduction function approach can be usedto estimate damage to crops caused bydeforestation. CVM or choice modelling isan appropriate method to value wildlifehabitat in farm forestry. Harrison (2000)provided further insights into methods ofestimating various non-wood benefits offarm forestry.

While the importance of non-market valuesis being increasingly realized, accuracy ofvaluation methods reviewed in this moduleremains a lingering problem. Reliability ofvalue estimates might be the top criterion,from the viewpoint of policy makers, tojudge with whether to include them into

project appraisal. Thus, valuationresearchers must continue to strive to refineexisting valuation methods, or developingnew ones, in a way of enhancing thereliability. However, as Harrison (1999)stressed, the ability to estimate non-marketvalues precisely should be treated as aseparate issue to the importance ofintegrating them in project appraisal.

Economists have devised various projectevaluation methods. The conventionalcost–benefit approach originated from theutilitarian philosophy of ‘the greatesthappiness of the greatest number’. Thephilosophical notion was incorporated intothe Kaldor–Hicks compensation principle. Anumber of decisions on environmentalprojects have often been made according tothe principle. The justification of thecompensation principle is based on apotential Pareto improvement, in whicheconomic welfare of a society is higher ifthe monetary value of gains exceeds thelosses. Under the principle, the identity ofthe gainers and losers does not matter. Onthe other hand, the multi-criteria approachevaluates several policy options at thesame time, using the idea of weightingvarious criteria. Further, ecologicalapproaches stress that researchers mustalso take into account scientific evidence ofthe adverse impact of development overnative flora and fauna, water quality andother ecological considerations. Ecologicalapproaches are applicable particularly tothe discrete land-use options. Among theseevaluation methodologies, each of themworks well in some cases and not well inother cases.

REFERENCES

Acharya, G. (1998), ‘Valuing theEnvironment as an Input: The ProductionFunction Approach’, in M. Acutt and P.Mason, eds., Environmental Valuation,Economic Policy and Sustainability: RecentAdvances in Environmental Economics,Edward Elgar, Cheltenham, pp. 63–78.

Aldred, J. (1994), ‘Existence Value, Welfareand Altruism’, Environmental Value, 3:381–402.

Arrow, K.J. and Fisher, A.C. (1974),‘P reserva t ion , Uncer ta in ty , and


Irreversibility’, Quarterly Journal ofEconomics, 88(2): 312–319.

Barb ier , E .B. (1994) , ‘Va lu ingEnvironmental Functions: TropicalWetlands’, Land Economics, 70(2):155–173.

Bateman, I. (1993), ‘Valuation of theEnvironment, Methods and Techniques:Revealed Preference Methods’, in R.K.Turner, ed., Sustainable EnvironmentalEconomics and Management: Principle andPractice, Wiley, Chichester, pp. 192–265.

Bateman, I.J. and Langford, I.H. (1997),‘Non-users’ Willingness to Pay for aNational Park: An Application and Critiqueof the Contingent Valuation Method’,Regional Studies, 30(6): 571–582.

Bateman, I.J. and Turner, R.K. (1993),‘Valuation of the Environment, Methods andTechniques: The Contingent ValuationMethod’, in R.K. Turner, ed., SustainableEnv i r onmen ta l Econom ics andManagement: Principle and Practice, Wiley,Chichester, pp. 120–191.

Bennett, J. (1995), Economic Value ofRecreational Use: Gibraltar Range andDorrigo National Parks, A report preparedfor the Environmental Economics PolicyUnit, Environmental Policy Division, NSWNational Parks and Wildlife Service.

Bennett, J. (1996), ‘The ContingentValuation Method: A Post-KakaduAssessment’, Agenda, 3(2): 185–194.

Bennett, J. (2000), ‘On values and theirEstimation’, International Journal of SocialEconomics, 27(7–10): 980–993.

Bennett, J., Gillespie, R., Powel, R. andChalmers, L. (1996), ‘Economic Value andRegional Economic Impact of NationalParks, Australian Journal of EnvironmentalManagement, 3(4): 229–239.

Bishop, R.C. (1982), ‘Option Value: AnExposi t ion and Extension’ , LandEconomics, 58(1): 1–15.Blamey, R.K. (1996), ‘Citizens, Consumersand Contingent Valuation: Clarification andthe Expression of Citizen Values andIssue–Opinions’, in W.C. Adamowicz, W.

White and W.E. Phillips, eds., Forestry,Economics and the Environment, CABInternational, Wallingford, pp. 103–133.

Cameron, J.I. (1992), ‘Valuing theEnvironment: A Social EcologicalPerspective’, in M. Lockwood and T. deLacy, eds., Valuing Natural Areas:Applications and Problems of theContingent Valuation Method , TheJohnstone Centre of Parks, Recreation andHeritage, Charles Sturt University, Albury,Australia, pp. 155–174.

Carson, R., Wright, J., Carson, N., Alberini,A. and Flores, N. (1995), The Bibliographyof Contingent Valuation Studies andPapers, Natural Resource DamageAssessment Inc., La Jolla.

Clawson, M. (1959), Methods of MeasuringDemand for and Value of OutdoorRecreation, Reprint No. 10, Resource forthe Future, Washington, D.C.

Clawson, M. and Knetsch, J.L. (1966),Economics of Outdoor Recreation, TheJohns Hopkins University Press, Baltimore.

Cooper, J. and Loomis, J. (1990), ‘PooledTime-Series Cross-Section Travel CostModels: Testing Whether RecreationBehavior is Stable over Time’, LeisureScience, 12(2): 161–171.

Daly, H.E. (1980), ‘The Steady-StateEconomy: Toward a Political Economy ofBiophysical Equilibrium and Moral Growth’,in H.E. Daly, ed., Economics, Ecology,Ethics: Essays Toward a Steady-StateEconomy, W.H. Freeman and Company,San Francisco, pp. 324–356.

Daly, H.E. (1996), ‘Sustainable Growth? NoThank You’, in J. Mander and E. Goldsmith,eds., The Case Against the GlobalEconomy: And for a Turn Toward the Local,Sierra Club Books, San Francisco, pp.192–196.

Davis, D. and Tisdell, C.A. (1996), ValuingOutdoor Recreational Sites: A NewApproach, Discussion Paper No. 206,Department of Economics, The Universityof Queensland, Brisbane.

Davis, R.K. (1963), The Value of Outdoor


Recreation: An Economic Study of theMaine Woods, unpublished Ph.D. thesis,Harvard University.

Diamond , P.A. and Hausman, J.A. (1993),‘On Contingent Valuation Measurement ofNonuse Values’, in J.A. Hausman, ed.,Cont ingent Valuat ion: A Cri t icalAssessment, North Holland, Amsterdam,pp. 3–38.

Edwards, S.F. (1988), ‘Option Prices forGround Water Protection’, Journal ofEnv i r onmen ta l Econom ics andManagement, 15(4): 475–487.

Englin, J. and Mendelsohn, R. (1991), ‘AHedonic Travel Cost Analysis for Valuationof Multiple Components of Site Quality: TheRecreation Value of Forest Management’,Journal of Environmental Economics andManagement, 21(3): 275–290.

Everitt, A. (1983), ‘A Valuation ofRecreational Benefits’, New ZealandJournal of Forestry, 28(2): 176–183.

Fletcher, J.J., Adamowicz, W.L. andGraham-Tomasi, T. (1990), ‘The TravelCost Model of Recreation Demand:Theoretical and Empirical Issues’, LeisureScience, 12(1): 119–147.

Freeman, A.M III. (1984), ‘The Quasi-Option Value of Irreversible Development’,Journal of Environmental Economics andManagement, 11(3): 292–295.

Garrod, G. and Willis, K. (1992), ‘TheAmenity Value of Woodland in GreatBritain: A Comparison of EconomicEstimates’, Environmental and ResourceEconomics 2(4): 415–434.

Gurocak, E.R. and Whittlesey, N.K. (1998),‘Multiple Criteria Decision Making: A CaseStudy of the Columbia River SalmonRecovery Plan’, Environmental andResource Economics, 12(4): 479–495.

Hadker, N., Sharma, S., David, A. andMuraleedharan, T. (1997), ‘Willingness toPay for Borivli National Park: Evidence froma Contingent Valuation’. EcologicalEconomics, 21: 105–122.

Hajkowicz, S.A., McDonald, G.T. and

Smith, P.N. (2000), ‘An Evaluation ofMultiple Objective Decision SupportWeighting Techniques in Natural ResourceManagement’, Journal of EnvironmentalPlanning and Management, 43(4):505–518.

Hanley, N. and Ruffell, R. (1993), ‘TheValuation of Forest Characteristics’, in W.C.Adamowicz, W. White and W.E. Phillips,eds., Forestry and the Environment:Economic Perspective, CAB International,Wallingford, pp. 171–197.

Hanley, N. and Spash, C.L. (1993),Cost–Benefit Analysis and the Environment,Edward Elgar, Cheltenham.

Harrison, S.R. (1999), ‘Progress inEstimation of Intractable Non-marketVlaues’, in S.B. Dahiya, ed., The CurrentState of Economic Science, SpellboundPublications, Model Town, Rohtak, India,Vol. 5, pp. 2719–2737.

Harrison, S.R. (2000), ‘Total EconomicValue of Multiple-use Farm Forestry andNon-market Valuation Techniques’, in S.R.Harrison, J.L. Herbohn and K.F. Herbohn,eds., Sustainable Small-scale Forestry:Socio-economic Analysis and Policy,Edward Elgar, Cheltenham, pp. 50–64.

Harrison, S.R. (2001), Valuation of theEnvironment, unpublished paper, School ofEconomics, The University of Queensland,Brisbane.

Heyes, C. and Heyes, A. (1999),‘Willingness to Pay Versus Willingness toTravel: Assessing the Recreational Benefitsfrom Dartmoor National Park’, Journal ofAgricultural Economics, 50(1): 124–139.

Hoagland, P., Kaoru, Y. and Broadus, J.M.(1995), ‘A Methodological Review of NetBenefit Evaluation for Marine Reserves,Paper No. 27, Environmental EconomicsSeries, The World Bank, Washington, D.C.

Imber, D., Stevenson, G. and Wilks, L.(1991), A Contingent Valuation Survey ofthe Kakadu Conservation Zone, RACResearch Paper No. 3, Vol. 1, ResourceAssessment Commission, Canberra.

Knetsch, J.L. (1963), ‘Outdoor Recreation


Demands and Benefits’, Land Economics,39(4): 387–394.

Knetsch, J.L. (1994), ‘EnvironmentalValuation: Some Problems of WrongQuestions and Misleading Answers’,Environmental Values, 3(4): 351–368.

Krutilla, J.V. (1967), ‘ConservationReconsidered’, American EconomicReview, 57(4): 777–786.

Mattsson, L. and Li, C.Z. (1994), ‘How DoDifferent Forest Management PracticesAffect the Non-timber Value of Forests? AnEconomic Analys is ’ , Journal ofEnvironmental Management, 41: 79–88.

Mitchell, R.C. and Carson, R.T. (1989),Using Surveys to Value Public Goods: TheContingent Valuation Method, Resourcesfor the Future, Washington, D.C.

Morrison, M. (2001), ‘Non-market ValuationDatabases: How Useful Are They?’,Economic Analysis and Policy: 31(1):33–56.

Narain, U. and Fisher, A. (1995), ‘Modellingthe Value of Biodiversity Using a ProductionFunction Approach: The Case of the AnolisLizard in the Lesser and Greater Antilles’, inC.A. Perrings, K.G. Mäler, C. Folke, C.S.Holling and B.O. Jansson, eds., BiodiversityConservat ion : Problems and Policies,Dordrecht, Kluwer Academic Publishers,pp. 115–125.

Pearce, D.W. and Turner, R.K. (1990),Economics of Natural Resources and theEnvironment, Harvester Wheatsheaf,London.

Portney, P.R. (1994), ‘The ContingentValuation Debate: Why Economists ShouldCare’, The Journal of EconomicPerspectives, 8(4): 3–17.

Randall, A. (1994), ‘A Difficulty with theTravel Cost Method’, Land Economics,70(1): 88–96.

Sagoff, M. (1988), The Economy of theEarth: Philosophy, Law and Environment,Cambridge University Press, Cambridge.

Smith, V.K. and Krutilla, J.V. (1982),‘Toward Reformulating the Role of NaturalResources in Economic Model’, in V.K.Smith and J.V. Krutilla, eds., Explorations inNatural Resource Economics, The JohnsHopkins University Press, Baltimore, pp.3–43.

Solow, R.M. (1993), ‘Sustainability: AnEconomist’s Perspective’, in R. Dorfmanand N.S. Dorfman, eds., Economics of theEnvironment: Selected Readings, 3r d edn.,Norton, New York, pp. 179–187.

Tisdell, C.A. (1991), Economics ofEnvironmental Conservation: Economics forE n v i r o n m e n t a l a n d E c o l o g i c a lManagement, Elsevier, Amsterdam.

Tyrvainen, L. and Miettinen, A. (2000),‘Property Prices and Urban ForestAmenities’, Journal of EnvironmentalEconomics and Management, 39: 205–223.

Weisbrod, B.A. (1964), ‘Collective-Consumption Services of Individual-Consumption Goods’, Quarterly Journal ofEconomics, 78(3): 471–477.

Willis, K.G. and Garrod, G. (1995),‘Transferability of Benefit Estimates’, in K.G.Wil l is and J.T. Corkindale, eds.,Environmental Valuat ion : N e wPerspectives, C A B I n t e r n a t i o n a l ,Wallingford, pp. 191–212.

Wood, S. and Trice, A. (1958),‘Measurement of Recreation Benefits’, LandEconomics, 34: 195–207.

149

14. Estimation of Non-market Forest Benefits UsingChoice Modelling

Jungho Suh

This module is concerned with some fundamental features and conditions of choicemodelling applications to non-market valuation. Choice modelling is an advance on thecontingent valuation method (CVM), and is creating strong interest among researchers, andhas much potential for non-market valuation in multiple use forestry. To undertake a choicemodelling application for estimating non-market values, potential practitioners need tounderstand the theoretical issues and practicalities involved in applying the technique. Theseinclude the statistical foundation of choice modelling, strict rules for the experimental designand ways of utilising the estimates. This module first outlines the characteristic features ofchoice modelling in terms of the method of evaluating resource use alternatives, comparedwith contingent ranking, contingent rating and CVM. Some basic assumptions andconsiderations needed in designing choice sets are then examined. Roles and rules of focusgroups are next introduced. Some choice modelling applications made in forestry researchare then briefly reviewed. Finally, ways of extrapolating welfare measures from a choicemodelling application are reviewed. More complex statistical issues are explained in threeappendices.

1. THE CHOICE SET FORMAT OFCHOICE MODELLING

Choice modelling originated from conjointanalysis, and is also a variation oncontingent valuation. In comparison toCVM, conjoint analysis describes options bydecomposing them into a number ofattributes, and presents respondents with achoice between j available options (j = 1,2,⋅⋅⋅, J ). This situation can be made quiterealistic by mirroring actual market choicethat may depend upon a number ofattributes.

If the number of available options is toolarge, the full options are divided intoseveral sets. Then, respondents are askedto rank, rate or choose their preferredcombination from each set. Moreover, thenumber of sets can be increased to asmany as each respondent can answerwithin a limited time. For this reason, it issaid that one of the major advantages ofconjoint analysis compared to CVM is thatmany options provide a large number ofobservations so that fewer respondents arerequired to yield results within acceptableconfidence limits.

Conjoint techniques have been widely usedin marketing studies dealing with market

goods rather than non-market goods.Conjoint analysis is founded on the theoryof consumer preference in an attempt todescribe how consumers choose betweensimilar products, for example, beers,coffees and soft drinks. Respondents areasked to rank or rate or choose from a setof multiple product profiles. Setting pricesfor the products was not necessarily theprimary concern of conjoint analysis inmarketing studies. In this sense, the factthat conjoint analysis was eventuallydeveloped to value non-market publicgoods can be dubbed a ‘paradigm change’in the field of economic valuation.

The rationale of conjoint analysisapplications for estimating environmentalnon-market values is that it is possible toestimate the amount that people are willingto pay to achieve a greater amount of oneor more environmental attribute, given thatthe dollar cost is treated as one of thecharacteristics for a non-market good. Infact, the price factor does not represent aninherent attribute of a commodity underconsideration. Rather, the price presentsdollar costs that are traded off for proposedchanges in attribute levels. This is whyMitchell and Carson (1989) classifiedconjoint analysis as a ‘hypothetical andindirect’ approach.


In contingent ranking, respondents rankthree or more options from most to leastpreferred. In the contingent ratingapplication, respondents are asked to rateeach option separately on a given ratingscale instead of ranking the options. Forexample, consider the case of a protectedforest area as illustrated in Table 1, where

jkz represents the kth attribute of the jth

option and jpz is the price factor. The

protected area is here defined as acombination of attributes, each of whichmay take various levels. If a respondent

prefers the jth option { jk

jj ,z,,zz ⋅⋅⋅21 } to the

other options, a higher ranking or rating isassigned to the jth option. Compared tocontingent ranking, contingent ratingcontains cardinal information. In choice-based conjoint analysis, respondents areasked only to choose their highestpreference from among several options –for example the set of choices presented inTable 1. Carson et al. (1994) called thismethod ‘choice modelling’ to distinguish itfrom contingent ranking or rating. In someliterature, the term ‘environmental choiceexperiments’ is used rather than choicemodelling, especially by the Canadiangroup of practitioners.

It is notable that a dichotomous CVMquestion is the same as a binary choicemodelling one except for the pricing format(Bennett and Carter 1993; Roe et al. 1996;Stevens et al. 2000). Consider a choicemodelling question with only one alternative(two options), as in Table 2, whererespondents are asked whether to acceptthe new option, comparing to the currentstatus option. Note that one of zk representsthe WTP amount (zp). It can be seen thatthe question is virtually identical to that of adichotomous CVM where respondents areasked whether they would be willing to payzp for the same change in zk.

Choice modelling has an advantage overcontingent rating in the sense that theformer is free of metric bias with which thelatter is plagued. Metric bias occurs when arespondent values an amenity according toa different metric or scale than the oneintended by the researcher (Mitchell andCarson 1989). This bias also relates to theproblems of interpersonal comparison of

cardinal measurement of utility (Morrison etal. 1996). Since individual rating scales incontingent rating applications reveal onlyrelative value between the respondents, it isnecessary to assume that rating scalesbeing used are consistent acrossindividuals (Rolfe and Bennett 1996).Similarly, contingent ranking suffersinconsistent ordinal measurement of utilityacross individuals.

2. ASSUMPTIONS ANDCONSIDERATIONS IN THEEXPERIMENTAL DESIGN

Any type of market or non-market good canbe described by a range of characteristics.In applications of choice modelling, anumber of hypothetical profiles are createdby combining distinct levels of attributes,which must represent a wide range ofcharacteristics of the object being valued.The number of attributes and their levelsdetermines the total number of distinctprofiles. A full factorial design includes allcombinations of the attribute levels, whereevery level of a given attribute is combinedwith all levels of every other attribute. Ingeneral, if there are m factors and n levelsof each, nm unique combinations can bemade. If factor space S has k factors, S =(z1, z2,···,zk), and each factor zk has Lk

possible levels, then S has L1 × L2 ×··· × Lk

possible combinations.

Question formats based on completefactorial design quickly become impracticaldue to the cost of administering the survey,not to speak of the respondents’ confusionand fatigue, as the number of either factorsor levels of the individual factors increases.Indeed, in many cases, a choice modellingresearcher is simply unable to conduct asurvey using a large number of profiles.Hence, the researcher is forced to adopt afractional factorial design, where only someof the combinations of factor levels areincluded. In choice modelling practice, aselected fractional factorial design is againbroken into a number of separate choicesets. Rolfe and Bennett (1996) noted thatthe number of choice sets should not be tooonerous for a single respondent. Theysuggested that choice sets be divided intomanageable blocks, with each blockallocated to a sub-sample of respondents.

Estimation of Non-market Forest Benefits Using Choice Modelling 151

Table 1. A choice format with several scenarios with various levels of attributes

Option ( j ) Attribute (zk)

z1 z2 … zk zp

1 11z 1

2z … 1kz 1

pz

2 21z 2

2z … 2kz 2

pz

: : : : : :J Jz1

Jz2… J

kz Jpz

Table 2. A binary choice modelling question format

Option Attribute (zk)

z1 z2 … zk zp

Current situation 01z 0

2z … 0kz

0pz

New option 11z 1

2z … 1kz

1pz

Louviere (1988) warned that one must becautious of fractional designs because astrictly additive utility function, known as theorthogonality assumption, underlies choicemodelling. The orthogonality assumptionmeans that choice modelling estimates onlythe main effect of each attribute on theoverall utility, assuming that all interactioneffects between attributes are zero. Thus,choice modelling questions should bedesigned to comply with the orthogonalityassumption. Further explanation on theorthogonal experimental design is providedin Appendix A.

For designing a choice modellingquestionnaire, a few other considerationsare required. First, the number of choicesavailable in a choice set should bemanageable from the viewpoint ofrespondents. Hanley and Spash (1993)warned that a choice set with five or morechoices for the contingent ranking methodwould make respondents confused andthreaten the reliability of the informationgained from the contingent ranking study. Inchoice modelling applications, the confusionwould be less evident because respondentsare not asked to rank the options in order oftheir preferences. Nevertheless, a largenumber of options in a choice set maycause the same problems. It is notable that

the minimum number of options that shouldappear in each choice set is three.

Second, extreme care is called forregarding the levels and range of thepayment variable. Lareau and Rae (1989,pp. 729–730) in an empirical study of thecontingent ranking technique warned that “ifprices are too low, respondents orderoptions by focusing mainly on theenvironmental attributes, while if prices aretoo high, respondents order optionsaccording to the price attribute.” Thiswarning is applicable to choice modellingstudies. Respondents would choose anoption by focusing mainly on theenvironmental attributes if prices are toolow and by focusing on the financialattribute if prices are too high. By the sametoken, Rae (1983) called attention to theprice range. Too small a range may result inunderestimating the values of otherattributes when a large number ofrespondents, who are willing to pay morethan the maximum, exist. Similarly, too widea price range with only a small number ofprice levels may effectively force a trade-off,but lessen the chance of obtaining preciseestimates within the range.

Third, too many attributes causerespondents to state their preferences by


‘indicator attributes’. The indicator attributeeffect occurs when respondents facecomplicated combinations of attributes, andthen use a single attribute to indicate whatwould happen to the other attributes( M o r r i s o n et al. 1997), rather thanevaluating each option by weighing up thelevels of each attribute. On the other hand,too few non-monetary attributes maygenerate payment vehicle bias andsubstantially offset the effects of de-emphasising the payment mode (Morrisonet al. 1996). Payment vehicle bias occurswhen the payment vehicle employedinfluences how an individual responds toquestions about WTP.

Bennett (1999) stressed that the paymentvehicle must be compulsory and havebroad coverage so as to be relevant to allrespondents, as used in Rolfe et al. (2000).The rationale for presenting the compulsorypayment vehicle is that the use of adonation as a payment vehicle encouragesstrategic responses. Therefore, it is arguedthat respondents must be convinced thatthey will be called upon to actually pay theamount they agree to pay when theychoose options from choice sets.

The compulsory payment vehicle is notalways desirable in the sense that it mayalso cause strategic responses. Morrison(1999) stressed that if the payment scenariois not acceptable, people may simply refuseto be interviewed in protest against thepayment vehicle, take the survey lessseriously or ignore the amount of payment.That is, people may reject the notion thatthey should have to pay for hypotheticalnew improvements in the quality of naturalresources because they have paid taxes forthe environmental improvement, eventhough respondents obviously derive utilityfrom the environmental improvement.Flatley and Bennett (1994) reported thatLockwood et al. (1993), in a pilot study of aCVM survey to value areas of Gippslandforest, found a strong anti-governmentsentiment emanating from publicity over thetroubles being experienced by variousstate-controlled financial institutions.Consequently, the forest valuation studyhad to use contributions to an independenttrust as the payment vehicle in order toavoid the public’s sentiment against the

imposition of new taxes influencing WTPbids.

3. ROLES AND RULES OF FOCUSGROUP SESSIONS

Focus group sessions are a commonlyused tool in psychology and are oftenregarded as a crucial step in shaping themarket strategy for products. Krueger(1988) defined a focus group meeting as acarefully planned discussion designed toobtain perceptions on a specific area ofinterest in a permissive, non-judgmentaland non-threatening atmosphere. Themeeting with a small group of people isinitiated and guided by preferably anexperienced facilitator. Leading a focusgroup may require the combined skills of anethnographer, a survey researcher and atherapist. Morgan (1988) pointed out thatthe main advantage that focus groups offeris the opportunity to observe a large amountof interaction on a topic in a limited periodof time. In comparison with directiveinterview that is dominated by theinterviewer, members of a focus group areallowed to talk without setting boundaries orproviding cues for potential responsecategories. The moderator is not supposedto take the lead, but is expected to allow theadvantages of non-directive interviews thatuse open-ended questions. The facilitator’sskilled control over focus groups is,therefore, a key to successful meetings.

Krueger (1988) and Bernard (1995)suggested that a focus group becharacterised by homogeneity, but ideallycomposed of strangers who do not knoweach other and will likely not ever see eachother again. Homogeneity is sought interms of occupation, social class,educational level, age and family character-istics, so as to avoid some mixes ofparticipants that do not work well becauseof limited understanding of other lifestylesand situations. For example, participantswill be inhibited by those whom theyperceive more experienced, knowledgeableor better educated. Krueger (1988),however, pointed out that sufficient variationamong participants to allow for contrastingopinions is helpful to obtain the contrast andvariation that spark lively discussions.

Rolfe and Bennett (1995) and Blamey


(1998) pointed out that focus groupsessions are an integral part of statedpreference methods, in which underlyingtheory of consumer behaviour is linked withpsychology. Focus groups are routinelyused in the developmental phase of CVMstudies. Rolfe and Bennett (1995, p. 3)summarised the major roles of focus groupsin an environmental valuation exercise.These are:

1. establishing the overall frameworkand characteristics of the good inquestion including the relationshipto other goods and applicableinstitutional settings;

2 . ascertaining the extent ofknowledge that people have aboutparticular goods, and the ways inwhich they value those goods;

3. identifying and describing the majorattributes that people considerwhen valuing particular goods; and

4. establishing appropriate trade-offsor WTP amounts associated withchanges in the particular goods.

Running focus groups is even more vital inchoice modelling studies, given that achoice modelling questionnaire tends to bemuch more complex than a CVMquestionnaire. Through focus groups,choice modelling practitioners candetermine which attributes are relevant toparticipants and test whether theinformation presented is appropriate;whether the main issues are communicatedeffectively and whether participantsunderstand the choice modelling processsufficiently to handle choice sets; andwhether the upper and lower bounds for thelevels of the financial attribute are adequate(Morrison et al. 1997).

Focus group participants are expected toreveal the beliefs and attitudes that theyactual ly consider when makingenvironmental decisions. For this reason,Rolfe and Bennett (1995) argued that focusgroup sessions have to be involved fromthe questionnaire design stage rather thanfrom the testing stage. In practice, focusgroups are often used for testing a draftquestionnaire and designing the finalquestionnaire.

One and a half to two hours is reasonable

to run a focus group meeting for statedpreference studies (Morrison et al. 1997).Stewart and Shamdasani (1990) noted thattaking part in a focus group is a timeconsuming activity for participants inpractice. Spending two or more hourstalking to a group of strangers is not likely tobe viewed as an appealing prospect. This ismore likely the case if one has worked allday. They noted that a variety of incentivesmay be used to encourage participation,and monetary incentives are commonlyused to induce individuals to spend time ina focus group.

It is recommended that a focus group havefive to 10 members plus a moderator (Rolfeand Bennett 1995; Morrison et al. 1997).The group size is determined by twofactors: it must be small enough foreveryone to have opportunity to shareinsights yet large enough to providediversity of perceptions (Krueger 1988).Bernard (1995) pointed out that if a group istoo small, the group can be dominated byone or two ‘loudmouths’, and if the numberof group members is beyond 10, it becomesdifficult to manage the group. Krueger(1988) mentioned that a focus group istypically composed of seven to 10 people,but the size can range from as few as fourto as many as 12. He noted that small focusgroups or mini-focus groups with four to sixparticipants are becoming increasinglypopular because the smaller groups areeasier to recruit and host, and morecomfortable for participants.

4. CHOICE MODELLING APPLICATIONSIN THE NATURAL ENVIRONMENTINCLUDING FORESTRY

There have been many choice modellingapplications to recreation studies orenvironmental conservation research.Adamowicz et al. (1994) analysed thewelfare impacts of changes in a set ofattributes of recreational fishing sites.Hanley et al. (1998) focused on theidentification and valuation of key attributesaffecting forest choice by conservation-oriented recreational users and non-users.Recent choice modelling applications dealtwith urban tourists’ portfolio choices ofdestination and transportation components(Dellaert et al. 1997), the environmentalvalues of water supply options of a river


(Blamey et al. 1999), improved wetlandquality (Morrison et al. 1999) and choice ofrecreational theme parks (Kemperman et al.2000). Examples of choice modellingapplications to forestry studies includevaluing international rainforests (Rolfe andBennett 1996; Rolfe et al. 1997) andevaluating a tree clearing policy (Rolfe et al.2000).

Rolfe and Bennett (1996) used choicemodelling to estimate demand byAustralians for rainforest conservation inoverseas countries. This choice modellingstudy was undertaken in Brisbane in 1995.The key attributes that were used in thesurvey were location, rarity, effect on localpeople, potential for future visits, size andpossession of special features. The locationattribute was varied across six locations,while each of the other attributes variedacross three levels. There were six optionsin each choice set. Each respondent waspresented with nine choice sets. This studydemonstrated how it is possible to deriveestimates of value for overseas rainforests.

Rolfe et al. (2000) conducted a choicemodelling experiment to estimate the valuesheld by Brisbane residents for bothenvironmental and social factors associatedwith tree clearing in the Desert Uplandsregion of central-western Queensland. Theimplication of changes to tree clearingregulations were described in terms of sixattributes as presented in Table 3.

Respondents were presented with a statusquo option (Option A) and two options forincreased preservation. A series of eightchoice sets were presented to eachrespondent. This study found that Brisbanehouseholds hold substantial protectionvalues for native vegetation in the DesertUplands.

The choices made of interview are typicallyentered into a spreadsheet for subsequentprocessing. Suppose that a respondentchose Option 3 from the list in Table 3. Withthis choice observation, three records canbe entered as presented in Table 4 on atypical spreadsheet such as MicrosoftExcel. Choice modelling requires that thedataset be arranged with a line of data foreach option. If choice modelling data arecollected from 10 respondents, and eachrespondent is asked to tick eight differentchoice sets, 240 records are generated. Inchoice modelling applications, sample sizesare generally about 300. Staying with eightchoice sets per respondent, this means that7,200 records are obtained from a choicemodelling survey.

The dataset can be analysed with astatistical package such as Limdep 7.0, aspecialised program for the estimation ofqualitative response models and limiteddependent variable models.

Table 3. A practical example of choice set

Implications Option A(current

guidelines)

Option B Option C

Levy on your income tax None $60 $20

Income lost to the region ($ M) None 5 10

Jobs lost in the region None 15 40

Number of endangered species lost to region 18 8 4

Reduction in population size of non-threatened species

80% 75% 45%

Loss in area of unique ecosystems 40% 15% 28%

Source: Rolfe et al. (2000, p. 12).


5. WELFARE MEASURESEXTRAPOLATED FROM CHOICEMODELLING ESTIMATES

Discrete choice models have historicallybeen the main models used in choicemodelling studies. What is meant bydiscrete choice models is that those modelsin which the dependent variable takes

discrete values. The simplest of thesemodels is the binary choice model in whichthe dependent variable y takes the value of0 and 1 (Maddala 1983). An example of thisis presented in Table 4: y is defined as 1 ifthe respondent chooses Option 1, y isdefined as 0 if the respondent does notchoose the option, and so forth.

Table 4. An example of choice modelling data entered on a spreadsheet

VariableRecordno.

Levy Regionalincome

Jobs Endangeredspecies

Populationsize

Ecosystems

Choice(y)

1 0 0 0 18 80 40 0

2 60 5 15 8 75 15 0

3 20 10 40 4 45 28 1

For a choice set with J alternatives, anestimated discrete choice model can beexpressed as:

jKK

jjk

j'

jj zbzb)=bV(z)

P

P(L +⋅⋅⋅++== 110ln

j = 1,2,⋅⋅⋅,J (1)

where jkz refers to the kth attribute of the jth

option, one of zk represents the price, andbk is the weight or coefficient associatedwith an independent variable kz . Equation 1

indicates that the logarithm of probabilitiesthat a particular choice will be made can berepresented by the systematic componentof utility of the jth option. This is assumedbecause choice modelling aims to yield theunbiased estimate of the main effects ofthose attributes on utility only (Adamowiczet al. 1994). The independence propertythat all the attributes must be independentof one another is the most important featurepertaining not only to choice modelling butto all types of conjoint analysis. Furtherdetails of the discrete choice model arepresented in Appendix B.

The independence of irrelevant alternatives(IIA) is inherently assumed for logit models.The IIA assumption implies that the odds ofchoosing the jth option in relation to one ofthe other options must be constantregardless of whatever other options arepresent (Louviere and Woodworth 1983).

Violations of the IIA property may occur inapplications where options are closesubstitutes for one another. The property isalso implausible when there existsheterogeneous tastes (Morrison et al.1999). When the IIA property is invalid,parameter estimates from the relevantdiscrete choice model will be biased. TheIIA assumption is a substantive restrictionon the generality of the logit models. Thus,testing for presence of the IIA property is astandard empirical procedure in obtaining avalid empirical logit model. Appendix Cprovides commonly used methods ofdetecting the IIA violations.

The estimated parameters of a discretechoice model provide a basis to computethe trade-offs between dollars andenvironmental quality. Compared to CVM, asingle choice modelling exercise canseparately and simultaneously estimate thecoefficients of all factors involved in choicesets. For example, the WTP to avoid each1% reduction in non-threatened species canbe obtained from the choice modellinganalysis using the Rolfe et al. (2000)dataset. Consequently, compensatingsurpluses for numerous hypotheticaloptions can be extrapolated. Compensatingsurplus estimates are put to use asparamount welfare measures associatedwith characteristics of an environmentalresource. In contrast, CVM generally valuesa composite of non-monetary factors persurvey. It is, however, not correct to argue


that the contribution of various attributes tothe value of an environmental good cannotbe estimated separately with CVM (Scarpa2000). It can be achieved with adequatelydesigned interviews. In particular, CVMpractitioners can prepare a number ofdistinct profiles and present one profile toone respondent at a time, though this iscostly in obtaining as many observations asin choice modelling.

Implicit prices for unit changes withinattributes

The implicit price means the marginal rateof substitution between a non-monetaryattribute and the monetary factor. Theimplicit price is also referred to as ‘part-worth’ or the point estimate of WTP. Theimplicit price is derived holding constant allother parameters except for the parameterof the attribute for which the implicit price isbeing computed. Mathematically, let one ofzk represent the price factor zp. Holding ∆Vj

= 0 yields:

∆Vj = ∆∑bk j

kz + ∆ bp zp = 0 (2)

Assume that the level of an attributechanges from a base level to another interms of environmental quality and levels ofk−1 other attributes remain unchanged. Theratio of the particular attribute coefficient tothe price coefficient measures the WTP forthe hypothetical change of the specificattribute, because utility of the non-monetary attributes is indirectly related tothe price factor. The point estimate of WTPwith respect to zk is obtained by:

p

k

k

p

b

b

dz

dz−=− (3)

For an improvement in environmentalquality, bk is greater than zero. Thus, theratio is expected to be positive because thesign on bp, the price parameter, will benegative. Positive ratio values representattributes that increase utility, whereasnegative ratio values (i.e. bk < 0) representattributes that reduce utility. It is importantto note that implicit prices are notappropriate for use as measures of overallwelfare created by an environmentalchange (Morrison et al. 1999). The reasonis that an implicit price is computed under

the assumption that change in utility isequal to zero, as expressed in Equation 2.Compensating surpluses need to becalculated, in order to understandcomprehensively the environmentalattitudes of stakeholder groups towards aspecific environmental option. Thecalculation of the compensating surplus foran environmental change involves implicitprices of model attributes multiplied byproposed units of changes within eachattribute, and the alternative specificconstant estimated for the specific option.

Deriving the compensating surplus forenvironmental improvement

Two types of compensation measures ofconsumer welfare may be distinguished.‘Compensating variation’ is defined as theamount of income that, if given to anindividual, would make the individualindifferent in utility terms between the initialand subsequent combinations of the moneyincome and the consumption level of goods.Instead of ‘compensating variation’, theterm ‘compensating surplus’ is used todescribe the welfare change in terms ofincome, in the situation where individualsare not free to choose the consumptionlevel of goods – for example, public goodssuch as air quality. In short, compensatingsurplus measures the difference in utilitybetween two options — namely, the changein income that would make the utility level ofan individual indifferent between the initialand subsequent options.

Using ind i rect u t i l i ty funct ions,compensating surplus (CS) for beneficialenvironmental changes can be defined asin the following equation:

VC (E0, M0) = VC (E1, M1) = VC (E1, M0 − CS)

= VN (E1, M0) − CS (4)

where E indicates a part icularenvironmental good in consideration and Mdenotes an individual’s total money income.The individual can purchase anycombination of market goods and servicesincluding other environmental goods andservices with total money income M, whereM 0 > M1. The person’s other option isassumed to be an environmentalimprovement denoted by E, where E1 > E0.


VC and VN represent the utility levels of thecurrent situation and alternative state,respectively. From Equation 4, M 1 = M0 −CS, and therefore CS = M0 − M1. Unlikewhen deriving implicit prices, the utility levelis not required to remain the same. Hence,compensating surplus represents marginalWTP for a change from the current bundleof commodity (E0, M0) to (E1, M0).

The welfare estimates are obtained by thedifference in utility between two options,which is scaled by the marginal utility ofincome to determine compensating surplus:

∑ ∑−−=

−−=

) (1

)(1

CS

11N

kC

kp

NCp

zbzbb

VVb

(5)

where 1kz indicates the kth attribute (except

for the price factor). Equation 5 has oftenbeen used for estimating compensatingsurplus for a new level of environmentalquality attached to a single site (e.g. Boxallet al., 1996; Blamey et al., 1998; Morrisonet al., 1999).

6. SUMMARY

Choice modelling is grounded on thediscrete choice model, which can produce astrictly additive utility function. From theestimated utility function, one can ultimatelyextrapolate consumer surplus for proposedhypothetical environmental changes. Forthis reason, discrete choice models haverecently been applied to value non-marketenvironmental goods such as forests andwetlands around the world in a number ofbenefit measurement studies.

In designing a choice modell ingquest ionnaire, keeping val id theorthogonality assumption is vital. In themodelling phase, testing for the IIA propertyis a standard empirical procedure. As well,given the diversity of beliefs, attitudes,interests, knowledge and other factors thatmight exist in populations, running focusgroups sessions is clearly a beneficial stepto designing and testing a non-marketvaluation questionnaire. Focus groupsessions are essential particularly forchoice modelling practices, considering that

the questionnaire format with a number ofchoice sets is relatively new and complex topotential respondents. Using focus groupsessions, a choice modelling practitionercan select design attributes, finalise choiceformats, frame the hypothetical marketcontext and identify attitudes that influenceparticipants’ responses.

In choice modelling applications toenvironmental resources, environmentalquality as well as monetary factors areincluded as attributes of the options in achoice set. Thus, choice modelling allowsone to obtain compensating surplusestimates so that one can account for thewelfare change generated by a bundle ofchanges in environmental attributes. It isalso possible to determine the relativeimportance of these attributes to people inmaking their choices.

In choice modelling practice, respondentsare offered hypothetical combinations ofattributes associated with a set ofscenarios. By downplaying the paymentfactor, choice modelling is much more aptlyable to estimate preferences with the effectof reducing the payment vehicle biasrelative to CVM. The mechanical frameworkof a discrete choice model allowsresearchers to disaggregate the utilityfunction attached to a particular good,which is a compelling advantage of choicemodelling over CVM. A choice modellingexercise can produce point estimates ofWTP for changes in the attribute levelsvaried in the choice sets as well ascompensating surplus estimates for aconceptually unlimited number of underlyingscenarios. Choice modelling can avoidproblems posed with contingent rankingand contingent rating. These latter methodsface many critics because of theoreticalproblems such as lack of error theory, andmeasurement bias that occurs because ofinterpersonal comparison of utility.

REFERENCES

Adamowicz, W and Louviere, J. andWilliams, M. (1994), ‘Combining Revealedand Stated Preference Methods for ValuingEnvironmental Amenities’, Journal ofEnv i r onmen ta l Econom ics andManagement, 26(3): 271–292.


Aldrich, J.H. and Nelson, F.D. (1984),Linear Probability, Logit and Probit Models,Sage University Paper No. 45, SagePublications, Beverly Hills, California.

Ben-Akiva, M. and Lerman, S.R. (1985),Discrete Choice Analysis: Theory andApplication to Travel Demand, the MITPress, London.

Bennett, J. (1999), Some Fundamentals ofEnvironmental Choice Modelling, ChoiceModelling Research Report No. 11, Schoolof Economics and Management, TheUniversity of New South Wales, Canberra.

Bennett, J.W. and Carter, M. (1993),‘Prospects for Contingent Valuation:Lessons from the South-East Forests’,Austral ian Journal of AgriculturalEconomics, 37(2): 79–93.

Bernard, H.R. (1995), Research Methods inAnthropology: Quantitative and QuantitativeApproaches, Altamira Press, Walnut Creek,Canada.

Blamey, R. (1998), ‘Contingent Valuationand the Activation of Environmental Norms’,Ecological Economics, 24(1): 47–72.

Blamey, R.K., Bennett, J.W., Morrison,M.D., Louviere, J.J. and Rolfe, J.C. (1998),Attribute Selection in Environmental ChoiceModelling Studies: The Effect of CausallyPrior Attributes, Choice Modelling ResearchReport No. 7, School of Economics andManagement, The University of New SouthWales, Canberra.

Blamey, R., Gordon, J. and Chapman, R.(1999), ‘Choice Modelling: Assessing theEnvironmental Values of Water SupplyOptions’, The Australian Journal ofAgricultural and Resource Economics,43(3): 337–357.

Boxall, P.C., Adamowicz, W.L., Swait, J.,Williams, M. and Louviere, M. (1996), ‘AComparison of Stated Preference Methodsfor Environmental Valuation’, EcologicalEconomics, 18(3): 243–253.

Carson R.T., Louviere, J.J., Anderson, D.A.,Arabie, P., Bunch, D.S., Hensher, D.A.,Johnson, R.M., Kuhfield, W.F., Steinberg,D., Swait, J., Timmermans, H. and Wiley,

J.B (1994), ‘Experimental Analysis ofChoice’, Marketing Letters, 5(4): 351–367.

Dellaert, B.G., Borgers, A.W. andTimmermans, H.J. (1997), ‘Conjoint Modelsof Tourist Portfolio Choice: Theory andIllustration’, Leisure Sciences, 19(1): 31–58.

Flatley, G and Bennett, J. (1994), ‘TheValue of Vanuatu Forest Protection toAustralian Tourists’, Vanuatu ForestConservation Research Report No. 6,Depar tment o f Economics andManagement, The University of New SouthWales, Canberra.

Green, P.E. (1974), ‘On the Design ofChoice Experiments Involving MultifactorAlternatives’, The Journal of ConsumerResearch, 1(2): 61–68.

Greene, W.H. (1997), EconometricAnalysis, 3rd edn., Prentice-Hall, London.

Hanley, N. and Spash, C.L. (1993),Cost–Benefit Analysis and the Environment,Edward Elgar, Cheltenham.

Hanley, N., Wright, R. E. and Adamowicz,V. (1998), ‘Using Choice Experiments toValue the Environment’, Environmental andResource Economics, 11(3–4): 413–428.

Hausman, J. and McFadden, D. (1984),‘Specification Tests for the Multinomial LogitModel’, Econometrica, 52(5): 1219–1240.

Kemperman, A.D., Borgers, A.W., Oppewal,H. and Timmermans, H.J. (2000),Consumer Choice of Theme Parks: AConjoint Choice Model of SeasonalityEffects and Variety Seeking Behavior’,Leisure Science, 22(1): 1–18.

Krueger, R.A. (1988), Focus Groups: APractical Guide for Applied Research, SagePublications, Newbury Park, California.

Lareau, T. and Rae, D. A. (1989), ‘ValuingWTP for Diesel Odor Reductions: AnApplication for Contingent RankingTechnique’, Southern Economics Journal,55(3): 728–742.

Lockwood, M., Loomis, J. and DeLacy, T.(1993), ‘A Contingent Valuation Survey andBenef i t -Cost Analysis of Forest


Preservation in East Gippsland, Australia’,Journal of Environmental Management,38(3): 233–243.

Louviere, J.J. (1988), Analysing DecisionMaking: Metric Conjoint Analysis, SagePublications, Newbury Park, California.

Louviere, J.J. and Woodworth, G. (1983),‘Design and Analysis of SimulatedConsumer Choice or Al locat ionExperiments: An Approach Based onAggregate Data’, Journal of MarketingResearch, 20(November): 350–367.

McFadden, D. (1974), ‘The Measurement ofUrban Travel Demand’, Journal of PublicEconomics, 3: 303–328.

McFadden, D., Tye, W. and Train, K.(1977), ‘An Application of Diagnostic Testsfor the Irrelevant Alternatives Properties ofthe Multinomial Logit Model’, TransportationResearch Record, 637: 39–46.

Maddala, G.S. (1983), Limited-dependentand Qualitative Variables in Econometrics,Cambridge University Press, Cambridge.

Mitchell, R.C. and Carson, R.T. (1989),Using Surveys to Value Public Goods: TheContingent Valuation Method, Resourcesfor the Future, Washington, D.C.

Morgan, D.L. (1988), Focus Groups asQualitative Research, Sage Publications,Newbury Park, California.

Morrison, M. (1999), Rethinking ContingentValuation: Ethics and Defensibility,unpublished paper presented at theAustralian and New Zealand Society forEcological Economics Conference, GriffithUniversity, Brisbane.

Morrison, M.D., Bennett, J.W. and Blamey,R.K. (1997), Designing Choice ModellingSurveys Using Focus Groups: Results Fromthe Macquarie Marshes and GwydirWetlands Case Study, Choice ModellingResearch Report No. 5, School ofEconomics and Management, TheUniversity of New South Wales, Canberra.

Morrison, M.D., Bennett, J.W. and Blamey,R.K. (1999), ‘Valuing Improved Wetland

Quality Using Choice Modelling’, WaterResources Research, 35(9): 2805–2814.

Morrison, M.D., Blamey, R.K., Bennett, J.W.and Louviere, J.J. (1996), A Comparison ofStated Preference Techniques forEstimating Environmental Values, ChoiceModelling Research Report No. 1, School ofEconomics and Management, TheUniversity of New South Wales, Canberra.Pindyck, R.S. and Rubinfeld, D.L. (1991),Econometric Models and EconomicForecasts, 3rd edn., McGraw-Hill, NewYork.

Rae, D. (1983), ‘The Value to Visitors ofImproving Visibility at Mesa Verde andGreat Smoky National Parks’, in R.D. Roweand L.G. Chestnut, eds., Managing AirQuality and Scenic Resources at NationalParks and Wilderness Areas, WestviewPress, Boulder, Colorado, pp. 217–234.

Roe, B. and Boyle, K. and Teisl, M.F.(1996), ‘Using Conjoint Analysis to DeriveEstimates of Compensating Variation’,Journal of Environmental Economics andManagement, 31(2): 145–159.

Rolfe, J. and Bennett, J. (1995),International Demand for Rainforests: UsingFocus Groups to Establish ValuationFrameworks, Vanuatu Forest ConservationResearch Report No. 9, School ofEconomics and Management, TheUniversity of New South Wales, Canberra.

Rolfe, J. and Bennett, J. (1996), ValuingInternational Rainforests: A ChoiceModelling Approach, Vanuatu ForestConservation Research Report No. 12,School of Economics and Management,The University of New South Wales,Canberra.

Rolfe, J., Bennett, J., Blamey, J. (2000), AnEconomic Evaluation of Broadscale TreeClearing in the Desert Uplands Region ofQueensland, Choice Modelling ResearchReport No. 12, School of Economics andManagement, The University of New SouthWales, Canberra.

Rolfe, J., Bennett, J., Louviere, J. (1997),Choice Modelling – An Exercise inEnvironmental Valuation, ConferencePaper, the Australian Agricultural and


Resource Economics Society, Gold Cost,Australia.

Scarpa, R. (2000), ‘Contingent ValuationVersus Choice Experiments: Estimating theBenefits of Environmentally Sensitive Areasin Scotland: Comment’, Journal ofAgricultural Economics, 51(1): 122–128.

Stevens, J.H., Belkner, R., Dennis, D.,Kittredge, D. and Willis, C. (2000),‘Comparison of Contingent Valuation andConjoint Analysis in EcosystemManagement’, Ecological Economics, 32(1):63–74.

Stewart, D.W. and Shamdasani, P.N.(1990), Focus Group: Theory and Practice,Sage Publications, Newbury Park,California.

Zhang, J. and Hoffman, S.D. (1993),‘Discrete-Choice Logit Models: Testing theIIA Property’, Sociological Methods andResearch, 22(2): 193–213.

APPENDIX A: THE ORTHOGONALFRACTIONAL FACTORIAL DESIGN

The orthogonal fractional factorial designcan be obtained by the systematic drawfrom the factorial of possible combinations.The requirement is satisfied when eachlevel of one factor occurs with each level ofanother factor with proportional frequencies(Green 1974). Rolfe and Bennett (1996)recommended al locat ing consistentnumbers of levels across attributes ifpossible for the simplicity of a symmetricalorthogonal array. However, nothing istechnically or practically wrong with varyingnumbers of levels between attributes, whichleads to an asymmetrical orthogonal array,as long as each level will occur an equalnumber of times within each attribute, ineach individual choice set and throughoutthe entire set of alternatives.

Louviere (1988), Adamowicz et al. (1994)and Bennett (1999) emphasised thatseparate choice sets of alternatives shouldbe treated independently. They suggestedthat this requirement be achieved by addingthe ‘no-change’ option – i.e. the ‘currentsituation’ option – to each and every choiceset. Each choice set is then evaluatedwithout regard to previously presented

choice sets. The provision of the ‘currentsituation’ option allows respondents to statethat they would prefer not to purchase anyof hypothetical alternatives presented in thechoice set. Carson et al. (1994) called forcaution about the possibil ity thatrespondents may use the ‘no-purchase’option as a means to avoid making difficultdecisions. In that case, adding the ‘no-change’ option to each choice set canadversely influence on the value estimates.

APPENDIX B: MATHEMATICALFORMULATION OF CHOICE MODELLING– THE DISCRETE CHOICE MODEL

The discrete choice model is usually calledthe conditional logit model, whichMcFadden (1974) derived from randomutility. The conditional logit model concernsthe effects of choice-specific attributes onthe determinants of choice probabilities. Toformulate the conditional logit model,consider the case of J mutually exclusiveand collectively exhaustive options labelledarbitrarily, where the numberings cannot betaken to indicate order or any magnitude.Respondents will be making comparisonsbetween options with the utility of the jthoption. The indirect utility of the jth optioncan be represented by:

Vj = V( jkz ) + εj (6)

where V( jkz ) is the systematic component

of utility and εj is a random unobservablecomponent. The systematic component isassumed to be the same for allobservations while the random componentis unique to each consumer. Assuming E(εj)= 0, the probability Pj that the jth outcome isobserved is:

Prob [V( jkz ) > V( 'j

kz )] j = 1,2,⋅⋅⋅,J for all j' ≠ j

(7)

The systematic component V( jkz ) can be

expressed as the sum of combinations ofattributes given by:

bZ zb

zb zbzb)=bV(zK

k

jkk

jKK

jjjk

≡∑=

+⋅⋅⋅+++

=1

22110

(8)


where jkz refers to the kth attribute of the jth

option, and one of zk represents the price,bk is the weight or coefficient associatedwith an independent variable kz , and bZ is

the linear combination of attributes with asingle parameter vector b. It is notable thatthe same attributes appear in the utilityfunction for every choice with varying levelswithin each attribute. This is not arequirement of conditional logit models, buta common feature of most choice modellingapplications.

The probability of an outcome in Equation 7is a linear function of independent variables.The problem with the linear probabilitymodel specification is that bZ is used toapproximate a probability Pj that is limitedbetween from 0 to 1, whereas bZ is not soconstrained. That is, P j is non-linearlyrelated to zk as well as to the b k . Thismeans that ordinary least squareprocedures cannot be used to estimate theparameters. To solve this problem, Aldrichand Nelson (1984) have demonstratedderivation of non-linear probabilityspecifications. To begin with, it is necessaryto take the logistic function as a non-lineartransformation function of Equation 7:

Pj = Prob [yj = 1 | V( jkz )]

= ∑=

J

j 1)bZexp()bZexp( (9)

where yj is the index of the choice madegiven that the random utility of the choice is

V( jkz ). Next, the upper bound P j can be

estimated by taking the ratio Pj /Pj' . The

ratio must be positive, and since Pj and Pj'

are constrained between 0 to 1, the ratiohas no upper bound. The lower boundary ofzero can be eliminated by taking the naturallogarithm, ln (Pj/ Pj'), the value of which canbe any real number from negative topositive infinity:

jKK

jjk

j'

jj zbzb)=bV(z)

P

P(L +⋅⋅⋅++== 110ln

(10)

The ultimate goal of applying choicemodelling is to estimate the coefficients (i.e.bk) from the logit model. The logit modelrepresented by Equation 10 defines the

logarithm of probabilities that a particularchoice will be made. Yet the coefficientsestimated also directly relate to the utility(Rolfe and Bennett 1996). The higher theutility of a particular attribute level in Optionj, the higher the odds ratio – that is, thehigher the probability that a respondent willchoose the particular option. Again, eachcoefficient estimated represents themarginal contribution of an attribute tooverall utility.

Logit models are estimated by themaximum likelihood technique. This can beachieved by using a non-l inearmaximisation program such as Limdep. Themaximum likelihood estimation procedurehas a number of desirable statisticalproperties. Among those, two aspects arenotable (Pindyck and Rubinfeld 1991). First,parameter estimators are asymptoticallyconsistent and efficient. Second, allparameter estimators are known to benormally distributed, so that the analog ofthe regression t-test can be applied. Onecan then apply the likelihood ratio test todetermine whether a logit model providesan appropriate model specification.

APPENDIX C: TESTING FOR‘INDEPENDENCE OF IRRELEVANTALTERNATIVES’

When an independence of irrelevantalternatives (IIA) violation is found for aconditional logit model, Morrison et al.(1999) suggested including socio-economiccharacteristics, attitudinal variables orquestionnaire evaluation variables in themodel. For the conditional logit model,these variables can be included in one orboth of two ways. The first is throughinteractions with the alternative specificconstants. These interactions reflect theeffect of these variables on the choiceprobability that a respondent will chooseeither Option 2 or Option 3. The second isby interactions with the attributes in thechoice sets. These interactions indicate thatthe non-attribute variables can modify theeffects of attributes on the choice probabilityof an option. The interaction effectsbetween non-attributes (e.g. socio-economic characteristics of individuals) andchoice specific dummy variables orbetween non-attributes and attributes arenot restricted by the orthogonal design.


If a conditional logit model containinginteraction terms of non-attribute variablesis found to violate the IIA property, itbecomes necessary to estimate a morecomplex model that relaxes part of the IIAassumption. An example of this type ofmodel is the nested logit model. The IIAassumption in the nested logit model isrelaxed because the options are dividedinto subgroups across which the variance isallowed to differ while maintaininghomoscedasticity within the groups (Greene1997). However, it is still necessary to test anested logit model for the IIA propertybecause the IIA violations can occur due tothe presence of close substitutes within thesame branch in a choice set.

In non-nested logit models, all elementalnodes are connected to the root of the treeas in Figure 1. In nested logit models,construct nodes are introduced. Assumethere are three choices in each choice set.Option 1 is the ‘no-change’ option andappears in every choice set. Combinationsof attribute levels for Options 2 and 3 varyacross choice sets. Option 1 is labelled asthe ‘current situation’, whereas Options 2and 3 are labelled as ‘new options’. Thenesting occurs when an individual choosesfirst, for example between Option 1 (currentsituation) and Option 2 or 3 (new options),and then, conditional on not choosingOption 1, chooses between Option 2 andOption 3. A tree relevant to this situationcan be specified as in Figure 2.

Various methods of detecting violations ofthe IIA property have been proposed (Ben-Akiva and Lerman 1985; Zhang andHoffman 1993). A Hausman and McFadden(1984) test is often used. To carry out thetest, a specific option throughout choicesets is excluded. A logit model with therestricted set of options but with the samemodel specification as the model with allchoices included is then estimated. Onecan next check whether there is any changebetween the two models, and evaluatewhether the new coefficients are sufficientlysimilar to the original ones, so as to satisfythe IIA property. The test statistic forchecking the IIA assumption is distributedas a chi-square variable. This type of testcan be carried out using the Limdepstatistical package.

Root

1 2 3

Figure 1. Tree diagram for a non-nestedlogit model (Model 1)

Root

1 (Current situation) New options

2 3

Figure 2. Tree diagram for a two-levelnested logit model (Model 2)

Another test proposed by Hausman andMcFadden (1984) is applicable to nestedlogi t models, which al lows theindependence property to be tested directly.For example, IIA violations occur if Options2 and 3 in Figure 2 have correlatedunobservable factors.

In Figure 2, ‘new options’ is a constructnode with parameter θ. This nested logitmodel assumes that the correlationbetween the errors of Options 2 and 3 isgiven by (1 – θ2). If the hypothesis that θ = 1cannot be rejected, which can be testedusing a likelihood ratio test comparingModels 1 and 2, then it can be concludedthat Model 2 does not violate the IIAassumption. That is, testing of the IIAproperty is equivalent to determiningwhether the estimated value of θ issignificantly different from 1.

There are other methods to test for the IIAassumption. Rolfe and Bennett (1996)suggested checking if a change in oneattribute produces shifts in choice greaterthan expected. Adamowicz et al. (1994)noted that the cross effects should equalzero if the attributes are strictlyindependent, apart from the belief that theattributes of one option could influence theutility of another option. Morrison et al.(1999) introduced the universal logit model,which was also proposed by McFadden etal. (1977), to detect violations of the IIAassumption. In estimating a universal logit


model, called a ‘mother logit model’, theattributes of one option are entered into theutility function of a second option, and thenthe attributes of the second option into theutility function of the first option. If the model

is found to be the more accurate model, theutility of one option can be considered todepend on the utility of other options, thusviolating the IIA property.

165

15. Cost-benefit Analysis in Forestry Research

Steve Harrison

Cost-benefit analysis (CBA) is both a technique of analysis and a broad framework foreconomic evaluation of investments, for example, in development and infrastructureprojects, and in research. It is applied to projects which run of a number of years, such thatthat discounting is used to bring costs and benefits to a comparable (present value) basis.The methodology is particularly relevant to forestry, which is a long-term investment, andforms one of the main socio-economic techniques in forestry research. As well, CBA isrelevant to judging the desirability of research projects. In fact, some research fundingagencies now require that a CBA of the research project be prepared as part of a grantapplication. This module first introduces the economic rationale for CBA. The procedures ofDCF analysis and the various project performance criteria are then explained. The followingsection examines various practical complexities in applying CBA. Some remarks are thenmade on integration of non-market values in CBA. Concluding comments follow.

1. THE NATURE OF CBA

Cost-benefit analysis is used to examinethe desirability of investment projects froman economic perspective. It is also relevantto estimating the desirabi l i ty ofinfrastructure development, the economicimpacts of new regulations, the benefits ofconservation programs, and the expectedpayoff from research projects.

When CBA was first applied widely (it wasmandated for watershed projects in theUSA in the 1960s), the emphasis was onmarket or financial transactions, with notingbut not valuation of ‘intangibles’. Nowadays,an attempt is made to include all relevantcosts and benefits of a project, in what iscalled extended or social CBA. The basicidea of social CBA is to determine whetherinvestment projects are worthwhile from asocial or taxpayer viewpoint, taking intoaccount all the costs and revenues andpositive and negative externalities theyincur or generate.

In the field of forestry, social CBA isrelevant to evaluation of public-sectorfunded projects such as community forestryprojects, and government investment inindustrial plantations and timber processingplants as well as forestry researchprograms. Each application of CBA is tosome extent unique or one-off, but anumber of common concepts andprocedures are involved in applications.Prominent in these concepts are

considerations of the value of money overtime, identification of ‘incremental cashflows’ and definitions and estimationmethods for cost and benefit flows.Discounted cash flow (DCF) analysisprovides the computational method forderiving project performance criteria suchas benefit-to-cost ratios, net present valueand internal rate of return. Whenconsidering forestry plantations, theFaustmann formula for optimal economicrotation provides a framework for evaluationof plantation costs, revenues andexternalities.

Cost-effectiveness analysis (CEA) issometimes used as a simpler alternative toCBA. Here only the costs of a project aretaken into account. Costs may be comparedon a unit of physical benefit basis, e.g.dollars per amount of timber produced. CEAis also appropriate when alternative waysare being considered to achieve a constantoutput, e.g. to meet timber demands adecade from now. Many infrastructureprojects have clearly defined outputs, whichwill be more-or-less constant regardless ofhow they are brought about provided designspecifications are met, and hence arecandidates for CEA. Public health and pestcontrol projects could fall into this class. Forexample, CEA could be used to comparealternative strategies for control of a say aforest disease in a region or country. Oncethe decision has been made to eradicate apest and disease, alternative methods couldbe compared by CEA. Note however that


CBA would be the relevant technique to usewhen determining whether it is in thenational interest to eradicate the pest ordisease in the first place.

Economists can play an important role byexamining the economic implications ofvarious alternative course of action, andpointing these out to policy-makers andadministrators. Policy-makers (that is, thegovernment) can combine this economicinformation with other information (e.g.about ecological values and communityattitudes) and judgment and intuition toarrive at a decision. In other words, theinput of the economics is information toaugment other information already held orbeing gathered by policy- or decision-makers. Economists provide decision-support information. Usually, the economistdoes not make the decision, or take theconsequences of that decision. But bypointing out the economic tradeoffsinvolved, they can assist managers ingovernment or private enterprise to makebetter decisions.

Particularly where large public investmentor conservation decisions are beingconsidered, economists often work as partof a multidisciplinary team. To communicateeffectively with other specialists, they needto have some appreciation of biology,sociology, planning, law and other areas.

2. THE ECONOMIC BASIS OF CBA

CBA is a widely used conceptual frameworkfor applied microeconomic analysis and apractical methodology for projectevaluation. Some form of CBA is invariablyused to evaluate investments, althoughoften in implicit rather than explicit form.

Economists typically take what is known asan anthropocentric or human-centredapproach. Goods and services are valuedin terms of what people are prepared to payfor them. The principle of c o n s u m e rsovereignty is adopted, in which people areregarded as the best judge of what is goodfor them. This approach implies forexample, that the preservation of species isimportant only insomuch as people place avalue on species. An alternative approachwould be a biocentr ic approach whichwould imply that species have value in their

own right, regardless of whether thecommunity wishes to spend money toprotect them.

According to the Pareto principle, a projectis desirable if it makes one or more peoplebetter off, and no-one worse off. In practice,projects invariable affect the incomedistribution in their area of influence, andtypically some people become better offand some worse off. Hence a modifiedcriterion – known as the Kaldor-Hickscriterion – was devised. This states that aproject is desirable if it makes one or morepeople better off, and out of their gains theycould potentially compensate any loserssuch that no-one is worse off. Note use ofthe term ‘potentially’ in this definition – thecompensation may not actually take place.In general, it is the role of government tomake adjustments to income distribution(such as taxes and welfare payments) toadjust to changes in an economy.

From a human-centred orientation, thedesirability of a project may be assessed interms of the changes in human welfare orsatisfaction it creates. While welfare is oftenconceptualized in terms of what economistscall ‘utility’, in practice this is normallymeasured in economic terms. Economicimpacts of a project can be illustrated usingmarket diagrams. Figure 1 is such adiagram, depicting the supply and demandcurves (S and D) for a particular economicgood (i.e. commodity or service). When qunits of the good are produced and theprice is p (in dollars per unit), the market issaid to be in ‘equilibrium’. That is, at thisprice the market clears, and there is noshortage or deficit to drive the price up ordown.

With reference to Figure 1, all consumerscan purchase the good at price p, whereassome are prepared to pay higher prices(indicated by the demand curve forquantities less that q ). The area abcrepresents the aggregate gain toconsumers in terms what they would beprepared to pay but do not have to pay.This is called the consumer surplus.Similarly, some producers would beprepared to place goods on the market at aprice of less than P , as indicated by thesupply curve for quantities of less than q.The area cbe represents the aggregate gain

Cost-benefit Analysis in Forestry Research 167

to producers, in terms of income theyreceive above what they would have beenprepared to accept, and is known as theproducer surplus. The sum of consumer

and producer surplus, known as economicsurplus for the market, represents the totalcommunity benefit from the market.

S S1

Price

b

p1

D

q q1 Quantity

Figure 1. Market diagram illustrating producer and consumer surplus

Now suppose a project is carried out, whichreduces the costs for producers. Supposethis shifts the supply curve (which accordingto economic theory is the short-run marginalcost curve) moves to the right, from S to S1.As indicated in Figure 1, there is a newequilibrium price p1 and quantity q1. In thisnew situation, both the producer surplusand the consumer surplus increase. It isnotable that not all the gains from lowerproduction costs are appropriated byproducers. Rather, the benefits are dividedbetween producers and consumers, therelative shares depending on theproportionate changes in areas of theproducer and consumer surplus, which inturn depend on the shapes (elasticities) ofthe supply and demand curves.

Although the economic surplus concept isconfined to a single time period, it doesshed light on how benefits from projectsshould be estimated. Producer surplus canbe approximated by net revenue or profitsto producers. Gains to consumers or‘consumer profit’ can be estimated by whatconsumers are required to pay relative towhat they would be willing (and able) topay. Willingness-to-pay (WTP) is a criticalindicator of value in economics

2. DISCOUNTED CASH FLOWANALYSIS AND PROJECTPERFORMANCE CRITERIA

The technique known as discounted cash(DCF) analysis, which provides thecomputational mechanism for CBA, wasdescribed in Module 11. As indicated in thatmodule, the economic performance criterionof net present value (NPV) is often used asa measure of project performance. Alimitation of NPV is that it does not indicatea rate of return on investment, and in thiscontext B/C ratios and the internal rate ofreturn (if it exists and is unique) can beused to provide additional information. Avariety of definitions of benefit-to-cost ratiomay be devised, depending what items areincluded in the numerator (some measureof gross or net benefits) and thedenominator (some measure of overheadand perhaps operating costs).

In this module, the emphasis is on thebroader issues of defining the costs andbenefits and other parameters relevant toCBA, so as to derive annual cash flowsestimates for a project (from which DCFperformance criteria can be calculated).

pc

a

e


3. COMPLEXITIES IN COST-BENEFITANALYSIS

Cost-benefit analysis is used by a widevariety of people for a wide variety of tasks.The availability of powerful computerspreadsheets with build-in financialfunctions has greatly facilitated theapplication of CBA. At the same time, theease with which the computational side ofthe technique can be applied has frequentlyled to a rather mechanistic approach beingadopted. No two projects will have exactlythe same characteristics, and it is unlikelythat a standardised approach to CBA canever be relied upon. Blind use of thetechnique can lead to results which do nottruly represent the investment situation, andto poor information for decision-makers.The practitioner needs both to be aware ofa number of complexities which often ariseand the best ways of dealing with them. Tosome extent, successful application of CBAis an art rather than a science, and there isno substitute for experience in carrying outreal-world applications. However, there area number of complexities of which potentialusers can be made aware.

The concept of a project

The term ‘project’ will be used in thebroadest sense throughout this module tomean any type of investment which affectscosts and benefits over time, for anindividual firm or industry or for the publicsector. To be more precise, the terms‘proposal’, ‘project’ and ‘program’ can betaken to mean different things. A proposal isa planned investment, prior to acceptance.A project is a specific type of investment,carried out by an individual or team usuallyfrom a single firm or government agency. Aprogram consists of a number of relatedprojects, such that various projects in aprogram may proceed in relative isolationfrom each other, to achieve related goals.For example, a forest industry developmentprogram may contain various individualtree-planting projects.

Public sector versus private sectororientation

Investment projects are carried out by boththe private and the public sector. In theprivate sector, a landowner may wish to

establish a forestry plantation. This willinvolve substantial initial outlays, with themain payoff taking place many years intothe future. Similarly, a government (local,provincial or national) may wish to promotecommunity forestry projects, providingfunding for the establishment phase andperhaps taking an equity in the project toshare revenue at a later time. Governmentsalso support infrastructure projects, such asroads and schools, which indirectly enablerevenue to be earned. They may alsospend on hospitals, libraries and othercommunity service obligations (CSOs),legitimately funded from taxation revenue.CBA is essentially a technique to be usedfor evaluation of public sector (government)projects. Discounted cash flow analysistechniques are also relevant to privatesector investments, although a number ofimportant differences arise in theirapplication.

Project analysis from a private firm’sviewpoint is known as financial analysis,while social CBA is referred to as economicanalysis. In this context, sometimes boththe financial internal rate of return (FIRR)and the economic internal rate of return(EIRR) are estimated for the same project.

Individual agency versus multipleagency evaluation

Often projects are financed by more thanone public agency. For example,community forestry projects may besupported by national and provincial or localgovernment and by international loanagencies. A CBA may be carried out froman individual agency viewpoint (e.g. takinginto account only costs and returns for thenational government’s perspective) or fromthe viewpoint of all investors in a project.

Identifying the project and its variablesand bounds

Experienced CBA practitioners sometimescomment that one of the most difficult tasksin carrying out any project evaluation is todetermine the bounds to the project. Anychange to a component of an economicsystem is likely to have an impact onvarious other components of that system.The further removed from the project thesecomponents are, the smaller the impact is


likely to be. As an example, consider thecase of community rainforest reforestation.This is likely to have benefits in terms ofcreating a timber resource, and may helprestore degraded farmland, improve wildlifehabitat, protect streambanks and waterquality, and provide employment in treeplanting and maintenance. These may inturn have further impacts, e.g. reducedstream erosion may reduce loss ofagricultural land, stream and water storagesiltation and road maintenance, andimprove fishing and enhance tourism. Someof these impacts are felt a considerabledistance from the tree planting area, e.g.further down a watershed. Obviously, it isnot possible to analyse the impacts of alocal investment project on the ‘wholeworld’. The task is to determine reasonablelimits to the impacts. In essence, thismeans defining the technical scope of theproject, and listing all of the variables whichare likely to be affected, in terms of benefitsreceived and costs imposed. As well, anattempt needs to be made to determine thenature of the relationships between relevantvariables. At some point, it will benecessary to make the explicit assumptionthat some impacts which could be includedare not being considered in the analysis.

Defining the ‘with project’ and ‘withoutproject’ cases

When evaluating a project, it is necessaryto define clearly the situations both with theproject and in the absence of the project;the incremental cash flows arising from theproject can then be identified. In thiscontext, the present situation may notcorrespond to the ‘without project’ case. Asan example, suppose an investment is to bemade to establish a community-basedforestry plantation. The ‘with project’ casewill relate to the new plantation, while the‘without project’ case will relate to thealternative use of the land. If the plantationwere not established, the land may continueto be used as it is currently, or some otheruse may be intended. Similarly, otherresources could be used differently,depending on whether the project wentahead. Thus the labour in site preparationand tree planting may be to some extentdiverted from growing food crops, andhence implementing the project could resultin reduced food crop production. When

determining cash flows, it is not the costsand benefits of the plantation, but rather thedifferences in cost and benefits if theplantation is developed relative to thesituation were it not developed, whichshould be taken into account in projectevaluation.

Project designs, management optionsand environment scenarios

When setting up a CBA, it is necessary todefine what alternative forms the projectcould take, to be compared in the analysis.A number of scenarios with respect to thephysical, economic and legal andinstitutional environment may also need tobe considered. This is all part of projectdefinition.

Time-phased investment projects

Projects sometimes consist of severalstages, some of which may not take placefor a number of years into the future. Thedecision then arises as to whether toevaluate the overall project or only one or asmall number of initial stages. A particularproblem arises where infrastructure is‘oversized’ in the first stage so that it isadequate for later states. For example, aparticle board plant may be constructed inconjunction with a plantation forestryproject, but may be designed to processmore logs than are likely to be available inthe near future, with a view to takingadvantage of new plantings and greaterparticle board demand in the future. It isappropriate to attribute only that proportionof the overhead cost needed in the currentstage of the project to the evaluation of thisstage.

Relevant costs and benefits

When evaluating a project, it is necessaryto include as far as possible all those costsand benefits to society associated with theproject. It is a far from simple task todetermine what cost and benefit variablesshould be included, and in fact, a majorstep in the analysis will be to identify whichcosts and which benefits are relevant.However, some general principles can belaid down. Project benefits includeadditional revenue generated, cost savingsand positive social and environmental


externalities. Project costs include overheadand (discussed further below) operatingcosts. Operating costs include explicit andimplicit costs, and negative social andenvironmental externalities. Explicit costsare those out-of-pocket expenses such aspayments for labour (i.e. wages), fertilizerand seedlings. In contrast, implicit costsarise from use of a firm’s own resources,such as labour, financial capital (money inthe bank) and buildings, for which no actualpayments are made. Environmental andsocial externalities of projects arise withrespect to people other than those carryingout or having a financial interest in theproject. For example, if plantationestab l ishment leads to carbonsequestration or reduced sedimentation ofwater storages (a positive environmentalexternality) or reduces the number of liveslost in coastal towns from floods (a positivesocial externality), then these should betaken into account in the economicanalysis. The valuation of environmentalimpacts presents many challenges. A low-cost approach which is intuitively appealingis to determine standard values forenvironmental conservation, which can beapplied in rapid project appraisal. Forexample, a value could be placed on eachhectare of native forest conserved, or eachindividual of a threatened speciesprotected. A number of environmentaldatabases have been developed to assist inbenefit transfer methods. More advancedvaluation approaches are discussed inanother module.

Treatment of capital items in cash flows

Capital items may be divided into projectcapital outlays, working capital and salvagevalues at the end of the project’s lie.

Capital outlays. Initial outlays and notdepreciation allowances should be includedin the cash flows. That is, the ‘cash’ flows offinancial transactions should be recorded atthe time they are made, rather than makingperiodic allowances for the services yieldedby capital items.

Working capital. Often it is appropriate tomake an allowance for working capitalwhich is tied up during a project. Thisallowance (e.g. 2% of the capital outlays)can be treated as an outlay at the beginning

of the project, with the full amount treatedas an inflow at the end of the project’s life.

Salvage value. Where plant and equipmentare purchased for a project, an allowancefor items on hand may be made as a capitalinflow for the final year of project life, e.g. ascrap value of 10% of initial outlays. Thevalue of standing trees at the end of theplanning period could also be treated as acapital inflow.

Some special cases of project costs andbenefits

Several cash flow items arise which warrantfurrther comment.

Transfer payments. Payments from onegroup of stakeholders to another – such astaxes and subsidies – do not constitute anynet gain or loss to society and hence shouldnot be included in project cash flows insocial cost-benefit analysis. (Theseamounts would however be relevant whenevaluating a project from a private sectorperspective.)

Interest payments. As mentioned earlier,the discounting procedure simulatesinterest payments. In any case, interestpayments are transfers between borrowersand lenders, and so should not be includedin social cost-benefit analysis.

Unpriced or non-market externalitiesspillover effects. It is desirable to include, asfar as possible, social and environmentalexternalities in social CBA. (Externalitieswould not normally be considered whenevaluating a project from a private sectorperspective.) Typically, most of the socialand environmental impacts of a project arenot reflected in market transactions, andhence placing values on them becomesdifficult. Some relevant valuation techniqueswere introduced in Modules 13 and 14. Theterm ‘cash flow’ seems somewhat at oddswith the concept of unpriced values, but isused out of tradition; the term ‘economicflows’ may be more appropriate.

Standard lists of checklists of costs andbenefits

It would seem desirable to develop lists ofcost and benefit categories which can be


referred to when carrying out a CBA.However, projects dealing with differentnatural resources usually have differentcategories of costs and benefits, and evenprojects dealing with the same resourcetypically have a number of unique features.Hence checklists are a useful starting point,but should not be relied upon too heavily.

Determining the planning period

The number of years for which cash flowsare estimated is referred to as the project’splanning period or planning horizon. Thisdepends on the planning horizon of thedecision-maker, and the realistic life of theproject. In the case of a forestry plantation,the planning horizon would normally be atleast at least one rotation to final clearfell,but could be more than one forestryrotation. If there is no clear end point to theproject – such as in the case of a limitedproduction forest – then a period of theorder of 20 to 30 years would normally beadopted. After about 30 years, the cashflows have litt le impact on DCFperformance criteria. The AustralianDepartment of Foreign Affairs and Trade,which commissions financial and economicevaluations of irrigation, water supply andother overseas projects for which Australiangrant aid is provided, suggests ‘a minimumof a 20 year planning horizon is generallyconsidered appropriate’ (AIDAB 1993). It issuggested here that, as a starting point, aperiod of 20 years be adopted, and thatchecks be carried out to determine whetherthis can be reduced or needs to beextended. It is to be noted that controversialissues can arise with respect to sustainableresource use and intergenerational equityrelative to planning horizon and thediscount rate.

In the case of research projects, it isnecessary to predict how long the benefitswill continue before they become replacedby new technology created by subsequentR&D. Also, it is necessary to make ajudgment as to whether subsequentresearch will enhance or replace those fromthe current project; in the former case, thebenefits could be assumed to continuewhile in the latter case they wouldterminate.

Choice of discount rate

The discount rate adopted in CBA has amajor impact on estimated values of theperformance criteria.1 There is considerabledebate in the CBA literature over theappropriate discount rate concept to adopt.The Department of Finance (1991, Ch. 5)noted the concepts of social t imepreference rate (STPR) corresponding tosociety’s preference for present as againstfuture consumption, and social opportunitycost of capital (SOC) corresponding to therate of return on investment elsewhere inthe economy. They recommended adoptionof the project-specific cost of capital. Thisinvolves using the cost of borrowing, whichis in most cases the long-term bond rate. Arate of 8% was recommended, comprising arisk-free rate of 5% and a risk margin of 3%.

Real and nominal interest rates

The interest or discount rate for social CBAis usually taken as the real rate, net of therate of inflation. Consistent with this, theannual cash flow variables are measured inconstant dollars, that is, they are notincreased over time to take account ofinflation. This of course does not mean thatbenefits and costs cannot increase overtime due to other factors. An obviousdifficulty arises from working with constantprices, in that in practice the rate of inflationmay be greater for some project variablesthan others. If timber becomes scarce, theprice increase may relatively rapid, suchthat a real price increase (i.e. an increasegreater than the overall inflation rate) takesplace. This could be incorporated in theanalysis by either working in current pricesthroughout, or by allowing annual increasesin timber prices equal to the difference inthe rate of timber price increase less theinflation rate (i.e. the real price increase).

To determine the real rate of interest forCBA, the nominal rate must be adjusted toremove the inflation component. The waythis is done depends on the relationshipassumed between the real rate and the rateof inflation. In this context, two different 1 As an illustration of this, a sum of $1M in 20

years time has a present value of $377,000 ifthe discount rate is 5%, but only $177,000 ifthe discount rate is 10% and only $26,000 ifthe discount rate is 20%.


models are possible. First, these may beassumed to be additive:

1 + n = 1 + i + f

where n is the nominal rate of interest (e.g.market or long-term bond rate), f is theinflation rate, and i is the real rate ofinterest. Here all rates are expressed on anannual basis, and in decimal form. Forexample, if the nominal interest rate is 11%and the inflation rate is 3% then

1 + 0.11 = 1 + i + 0.03

from which

i = 1.11 - 1.03 = 0.08 or 8%

More often, a multiplicative model isadopted, of the form

1 + n = (1 + i)(1 + f)

The solution to this interest rate formula, interms of the real rate, is

1 + n i = - 1 1 + f

For example, if the nominal rate is 11% andthe inflation rate is 4% then the real rate is

1 + 0.11 i = - 1 = 0.0777 or 7.77% 1 + 0.04

It is to be noted that the real rate is slightlyless under the multiplicative model thanunder the additive relationship. Use of theformer is recommended in the recommend-ations for cost-benefit analysis laid down bythe Department of Finance (1991).

For discounting purposes, a constant rate ofinterest is normally applied throughout theplanning horizon. While the rate will nodoubt vary over time, the current rate isusually adopted. However, if interest ratesare expected to change in a particulardirection in the short term, then this may betaken into account in the choice of rate. It isto be noted that because cash flows furtherremoved in the future have less impact onpresent values than cash flows in the shortterm, predicting an appropriate rate over the

first five to 10 years is more important thanlonger-term predictions.

Dealing with uncertainty in CBA

Typically, the annual cash flows of a long-term project are subject to a high level ofuncertainty, particularly the future projectbenefits, and various procedures areavailable for taking this into account in CBA.Uncertainty can arise because of physicaland financial factors, and also legal andinstitutional change. For some projects,changes in exchange rates with overseascountries will contribute to financial risk, andcosts will be incurred with hedging orinsurance against exchange ratedepreciation.

Typically, a component for risk is includedin the discount rate, on the grounds that ahigher rate of return is needed to justify thegreater project r isk. Somet imesconservative estimates of project benefitsare used; a systematic way of doing this isto compute certainty equivalents (the risk-free amounts which the decision-makerregards as equally acceptable to the higherbut uncertain amounts). Normally,sensitivity analysis is carried out todetermine what effect changes in parameterestimates would have on performancecriterion such as NPV. An alternative whichprovides greater information is to apply riskanalysis, in which probability distributionsare attached to key variables orparameters, and the probability distributionof project performance (say NPV) isderived.

Allowing for slippage in economicperformance of research projects

When applying CBA to research projects, itis usual to make allowance for theprobability of research success (in terms ofachieving the planned technical outputs anddeveloping them into a usable technologypackage) and the level and timing of uptakeof this technology by potential adopters,e.g. see Harrison et al. (1991). Notably, thecost of the development stage, andtechnology transfer (extension) also need tobe included in project outlays.


Shadow pricing

In general, the prices of inputs and outputsused in CBA are those prevailing in marketsat the time of the analysis. However,sometimes these may be regarded asunsuitable for the analysis, and may bereplaced by shadow prices. ‘A shadow priceis an imputed valuation placed on a projectinput or output when a market price doesnot exist or is significantly distorted’ (AIDAB1993). Since natural resources are oftenunderpriced (EPAC 1991; Harrison andTisdell 1992), there is sometimes scope forshadow pricing in projects related to naturalresources. When evaluating projects, alarge premium (20% or more) is sometimesplaced on wages earned (where there ishigh unemployment) or on export revenue(where there is a shortage of foreignexchange). These loadings may be justifiedon the basis that governments spend largeamounts of money on job creation andexport promotion activities.

Multiplier effects

It is not unusual for proponents of particularprojects to argue for them on the basis ofindirect of flow-on benefits, e.g. creation oflocal employment, more local spendinghence improved local business activity, andimprovements to local real-estate prices.Investment in a plywood or pulp mill couldcreate a number of jobs and provide asubstantial stimulus to a regional economy.These are the kinds of benefits which canbe examined with inter-industry input-outputanalysis, in which employment, income andoutput multipliers are derived. While thesebenefits are real enough from a localviewpoint, the case for them is not so strongfrom an overall social perspective. Thequestion needs to be asked: ‘Wouldinvestment in one particular area creategreater flow-on benefits than investmentelsewhere?’. If not, then there may not besufficient justification for estimating flow-onbenefits for the particular area. As indicatedby the Industry Commission in its study ofwater management (IC 1992), the ‘widerbeneficiaries’ argument is not a strong one,and the main use of multiplier effects is toexamine the adjustment problems whenresource uses are curtailed in an area.

Effect on income distribution

Public sector projects will often have aneffect on income distribution in thecommunity. Some people are likely to bebetter off and others worse off. Intuitively, itwould be appealing if a project improvedthe lot of the poor, as is likely with acommunity forestry project. A greaterweight could be placed on incomegenerated by a project which helped thepoor than one which benefited mainly therich. However, this would lead to difficultestimation problems, including interpersonalcomparisons of utility. The economist has arole in identifying the likely impacts onincome distribution. However, judgmentsabout desirability of changes to incomedistributions ‘are almost always mostappropriately made by Government at thepolitical level’ (Department of Finance1991).

Project evaluation versus projectjustification

This module would not be complete withoutsounding a warning about misuses of CBA.Given the complexities discussed above, itis easy inadvertently to use the techniqueincorrectly. More seriously, given that CBAis often used to demonstrate that aparticular investment is worthwhile and toassist in obtaining funding, there can be astrong temptation to seek optimistic benefitestimates and hence ensure a positiveNPV. That is, CBA can be used to generatemisinformation rather than goodinformation. Where agency goals to obtainapproval for new projects are strong,considerable suasion may be placed oneconomists to come up with ‘good figures’.How this is handled may become a difficultquestion of personal integrity versus loyaltyto and peer acceptance in the organization.If the economist disagrees with the figuresthat are being proposed, then he or she hasa responsibility to communicate clearly thebasis of this disagreement.

4. INTEGRATION OF NON-MARKETENVIRONMENTAL VALUES INSOCIAL CBA

In general, if non-market values can beestimated, these can be factored into aCBA, alongside market values. They


become part of the annual incremental cashflows (or ‘economic flows’), to be placedalongside financial flows, and traded off ona dollar for dollar basis. Thus, degradationof the environment and depletion of naturalassets are included as part of the coststream for a project, or protection of thesevalues as part of the benefit stream. This isconsistent with the economic concept ofvalue as community willingness-to-pay forboth market and environmental values.

Acceptability of non-market values and‘funny money’ arguments

The inclusion of, and failure to include, non-market values in CBA are both often thesubject of controversy. This raises thequestion of whether non-market valuesshould be given lower weights than marketvalues in CBA. In general, there does notseem justification for this discriminatoryanalysis, although there can be some caseswhere it might be acceptable.

Consumer versus producer surplus. Often amajor component of environmental valuesare consumer surplus values, e.g.recreation values, landscape amenity, airquality. On the other hand, often the marketvalues are producer values, e.g. fromtimber production. In that governmentssubsidize industry, it would seem that theshadow price of producer surplus issometimes accepted as being higher thanthat of consumer surplus.

Negative signals to investors. Thereservation of areas containing valuableexploitable natural resources, forconservation and recreation purposes, canprovide negative signals to industry. Itwould appear that this has in some casesled companies to invest overseas ratherthan in the domestic economy, with loss ofdomestic production and employment.Again, this suggests there may be anargument for placing a higher weight onproducer surplus than on consumer surplus.

Exports versus domestic products.Sometimes greater weight is placed onexports than domestic products andservices. This is relected in tax concessionsand other incentives to exporters, and issometimes motivated by trade deficits andservicing of foreign debt. This suggests

placing a higher weight or shadow price onexported commodities as against domesticgoods and services.

Uncertainty and bias in value estimates.Precision of estimation of non-marketvalues can be low. Generally, marketvalues can be estimated more precisely,although this certainly is not always thecase. The problem of uncertainty isexacerbated by a number of potentialsources of bias which are recognized forsome non-market valuation techniques, e.g.CVM. If there is a suspicion that non-marketvalue estimates are upwardly biased, thiscould be a reason for downwardadjustment.

It is to be noted that these are possiblereasons for placing a lower weight onconsumer surplus than producer surplus,but not for ignoring the latter (leaving it outof the analysis). Also, changes in levels ofenvironmental services have can affectprices of market goods, e.g. higheragricultural costs, higher expenditures onenvironmental protection, higher costs ofhealth services.


Cost-benefit analysis is both a conceptualframework which is used widely ineconomics, and a technique for projectappraisal. As a framework, it providesguidelines on what cost, revenue and non-market variables to take into considerationin project appraisal and how to define andvalue these variables. As a technique, it haswide application in the evaluation ofproposed public sector projects. CBA isrelevant to a variety of economic issues inforestry, such as evaluation of governmentassistance programs to landholders,communities and industry, and evaluation offorestry research projects.

REFERENCES

AIDAB (Aus t ra l ian In te rna t iona lDevelopment Assistance Bureau) (1993),Economic and Financial Evaluation of DIFFProjects, Canberra.

Department of Finance (1991), Handbookof Cost-Benefit Analysis, AGPS, Canberra.


EPAC (Economic Planning AdvisoryCouncil) (1991), Issues in the Pricing andManagement of Natural Resources,Background Paper No. 16, AGPS,Canberra.

Harrison, S.R., Dent, J.B. and Thornton,P.K. (1991), ‘Assessment of Benefits andCosts of Risk-Related Research in CropProduction’, in R.C. Muchow and J.A.Bellamy, eds., Climatic Risk in CropProduction: Models and Management forthe Semiarid Tropics and Subtropics,

C.A.B. International, Wallingford, pp. 491-510.

Harrison, S.R. and Tisdell, J.G. (1992),‘Reform of Natural Resource Pricing’,Micro-Economic Policy and Reform forInternational Competitiveness, EconomicSociety, Gold Coast, pp. 241-253.

IC (Industry Commission) (1992), WaterResources and Waste Water Disposal,Report No. 26, AGPS, Canberra.

177

16. Using Financial Analysis Techniques in ForestryResearch

John Herbohn

Previous modules introduced the principles and methods of Discounted Cash Flow (DCF)and Cost Benefit Analysis (CBA) in relation to forestry projects. The current module extendsthe discussion of DCF to include its use in financial analyses. The purpose of CBA is todetermine whether investment projects (used in the widest sense) are worthwhile from asocial perspective, taking into account all costs and benefits including positive and negativeexternalities. In contrast, the focal point of financial analysis is assessing the impact of theproject of the wealth of the individual or firm considering the project. In this module, conceptsand methods of financial analysis (FA) will be introduced, and then applied to a case study offinancial evaluation of a forestry project.

1. INTRODUCTION

Cost-benefit analysis (CBA) and financialanalysis (FA) have many elements incommon, the key distinction being that theformer is carried out from the perspective ofthe society or community while the latter iscarried out from the perspective of theprivate interests of the investor. CBA istypically used in making decisions aboutpublic sector projects. Financial analysistypically is used as a technique in thecapital budgeting decisions of firms andindividuals. CBA would include all theimpacts (both financial and non-financial) ofa project to the organization or governmentundertaking the project and to society as awhole. Financial analyses is more narrowlyconcerned about the financial or direct cashflow impact of a project to an individual orbusiness, with environmental externaliltiessuch as pollution only included in a financialanalyses to the extent that they imposecash flows such as fines or clean-up costson the firm or individual. Similarly, socialbenefits such as increased employment orstandard of living by local communities fromthe establishment of a timber mill (relevantin CBA) would not be included in a financialanalysis because there are no direct cashflow consequences to the firm.

In many ways financial analysis can beconsidered as a sub-set of CBA. The directcashflow consequences of a project to theorganization undertaking the project (asidentified in a financial analysis) must be

identified as part of a CBA, although thedistinction is seldom explicitly drawnbetween cashflows directly impacting on theorganization and those impacting on otherparties affected by the project.

Capital budgeting is a multi-faceted activity.There are several sequential stages in theprocess. For typical investment proposals ofa large corporation, the distinctive stages inthe capital budgeting process are depicted,in the form of a highly simplified flow chart,in Figure 1. The evaluation is carried outwithin the strategic planning of thecompany, and the screening of investmentalternatives. Information from the financialanalysis is integrated with other informationand objectives of the company to arrive at adecision to accept or reject each project.Accepted projects must undergoimplementation and monitoring procedures.

2. SOME KEY CONCEPTS IN FINANCIALANALYSIS

Incremental cash flows

The concept of an incremental cash flowwas introduced in the previous module.Incremental cash flows are the cash inflowsand outf lows which will arise from theimplementation of a particular project, andare estimated by comparing the cashinflows and outflows of the firm ‘with’ theproject, and those ‘without’ the project. It isa ‘marginal’ or ‘incremental’ analysiscomparing two situations.


Corporate goal

Strategic planning

Investment opportunities

Preliminary screening

Financial appraisal, Quantitative analysis,Project evaluation or Project analysis

Qualitative factors, judgements and gut feelings

Accept / reject decisions of the projects

Accept Reject

Implementation

Facilitation, monitoring, control and review

Continue, expand or abandon project

Post-implementation audit

Figure 1. The capital budgeting process

Using Financial Analysis Techniques in Forestry Research 179

For analytical purposes project cash flowsmay be separated into two categories:capital cash flows and operating cash flows.Capital cash flows may be disaggregatedinto three groups: (1) the initial investment,(2) additional ‘middle-way’ investmentssuch as upgrades and increases in workingcapital investments, and (3) terminal flows.These are all cash flows and the distinctionbetween them are only to facilitate theconvenient identification of the differentcategories.

Synergistic effects and opportunity cost

All the indirect or synergistic effects of aproject should be included in the cash flowcalculation. Synergistic effects can benegative or positive. For example, supposea land owner has 100 ha of plantationswhich are at a stage of requiring thinning.There is no current market for the thinningsand he will have to pay local people to thinhis plantation. He is also considering buyinga plant to make charcoal. If he buys theplant he could then use the thinnings fromhis plantations to make charcoal. In thiscase there is a positive ‘synergistic effect’ ofinvesting in the charcoal plant i.e. theinvestment creates a use for a previouswaste product.

The rationale for the incorporation of theseindirect effects has its base in the‘opportunity cost’ principle. When a firmundertakes a project, various resources willbe used up and not available for otherprojects. The cost to the firm of not beingable to use these resources for otherprojects is referred to as an ‘opportunitycost’. The value of these resources shouldbe measured in terms of their opportunitycost. The opportunity cost, in the context ofcapital budgeting, is the value of the mostvaluable alternative that is given up if theproposed investment project is undertaken.This opportunity cost should be included inthe project’s cash flows.

To consider another example, suppose alandholder is considering establishing 5 haof Gmelina. The plantation will beestablished on land that he does notcurrently use but which he could rent out for20,000 pesos per year. In this case thefarmer will lose the opportunity to rent theland at 20,000 pesos per year if he

establishes the plantation. This ‘opportunitycost’ should be included in the cashoutflows of the project even though therehas not been any actual cash payment. Thereason is that opportunity cost of the spacemeasures an extra cash flow that would begenerated (for the firm) ‘without’ the project.On the other hand,suppose that the landhas not been rented in the past and there isno intention to rent, sell or use for any otherpurpose in the future. In this case, there isno opportunity cost if the resource is usedfor the proposed project. Therefore, in thissituation, the 20,000 pesos will not beincluded as a cash outflow.

3. TREATMENT OF KEY DCFVARIABLES

The basic principles of DCF analysis havebeen discussed previously. It is howeveruseful to consider some of the key variablesoften found in DCF analysis in a little moredetail.

Treatment of taxation payments

Tax is a cash payment to a governmentauthority. How taxation payments areviewed differs depending on whether a CBAor FA is being undertaken. This is due tothe different perspectives of each of thesetypes of analysis. In CBA, in the perspectiveis that of the community, taxation paymentsare viewed simply as ‘transfer payments’between one section of the community andanother. They do not represent any net gainor loss to society, and hence are excludedfrom incremental cash flows.

With a financial analysis the focus is on thecashflow impacts on the individual or firm,hence tax payments need to be included inthe analysis. If the project generates taxliabilities, then the tax payable is relevant tothe project, and must be accounted for as acash outflow. Corporate tax is a cashoutflow. If the tax were levied on net cashinflow and paid at the same time cash wasreceived, then the after-tax net cash flowwould be easily calculated. However, tax isnot based on net cash flow, but on taxableincome.

Taxable income is defined by the relevanttax act and does not necessarily mean thesame thing as net cash flow or even


accounting income or accounting profit.Taxable income is generally calculated bysubtracting allowable deductions fromassessable income. These terms arespecific to particular tax acts, and are noteasily dealt with in a general context.However, project evaluation needs to beable to accord some treatment to thiscalculation to determine after tax cashflows.

Treatment of depreciation

The tax definition of ‘deductions’ treatssome non-cash items as allowableexpenses. One such item frequentlyencountered in project analysis is assetdepreciation. Depreciation is not a cashflow. It is an allocation of the initial cost ofan asset over a number of accountingperiods. Asset costs are allocated withinaccrual accounting systems so that they arematched over time against the incomegenerated by the assets. That is, the initialcost of an asset is expected to benefit thefirm over several years, hence the totalinitial cost is ‘spread’ over those futurebenefit years.

The actual per annum dollar amount ofdepreciation is only a notional amount. Itdoes not represent the annual decline invalue of the asset, it does not measure thevalue of the asset used up, and it does notmeasure the actual unit costs of the asset’sservices. A number of accounting methodscan be used to calculate depreciation i.e.straight-line, reducing balance, ‘sum ofyears digits’ and units of productionmethods.

In project evaluation, what is relevant is notthe accounting depreciation but the tax-allowable depreciation. The methods ofcalculating tax-allowable depreciation areoften prescribed by the taxation legislationwithin a country. Sometimes the firm willhave a choice between alternativeprescribed methods, and in these cases thefirm usually selects the method which willreduce the overall tax bill. The tax bill will bereduced if higher depreciation is claimed inthe earlier years, thus delaying the paymentof tax. The reducing balance method hasthis effect. Many national tax acts permitaccelerated depreciation of equipment byallowing depreciation methods (defined in

the tax act) which allow higher taxdeductions in early years and lowerdeductions later. The Modified AcceleratedCost Recovery System (MACRS) in theUSA. is an example.

It is important to understand that theinterest in depreciation in financial analysislies only in its tax effect, i.e. thedepreciation tax shield or the reduction intaxes attributable to the depreciationallowance. If depreciation were not a taxdeduction, it would not be considered in anNPV evaluation.

Treatment of financing flows

Treatment of financing flows is an area ofconfusion and sometimes a source of errorin project analysis. It is important todistinguish between project cash flows andfinancing cash flows. For the purpose ofidentifying the relevant cash flows forproject evaluation, the investment decision(project) must be separated from itsf inancing decision. The f inancinginvestment decision involves deciding uponwhat proportion of the cash flows needed tofund the project are provided by debtholders and what proportion are providedby equity holders (i.e. from cash alreadyheld by the firm). The decision about theparticular mixture of debt and equity used infinancing the project is a managementdecision concerning the trade-off betweenfinancial risk and the cost of capital. Theinvestment evaluation decision determineswhether the project’s discounted cash flowsexceed the init ial capital outlay(investment), and so adds (net present)value to the firm.

Generally, interest charges or any otherfinancing costs such as dividends or loanrepayments are not deducted in arriving atcash flows, because the interest is in thecash flow generated by the assets of theproject. Interest is return to providers ofdebt capital. It is an expense against theincome generated to equity holders, and assuch is deducted in the determination ofaccounting profit. However, it is notincluded in project cash flow analysis,because the discount rate employed in theNPV calculation accommodates therequired returns to both equity and debtproviders. Therefore, inclusion of interest


charges in cash flow calculations essentiallywould result in a double counting of theinterest cost.

Interest is tax deductible, and thereforeprovides a tax shield for any investment.This benefit is also accounted for in thediscount rate, as the rate employed inproject analysis is an ‘after-tax’ rate.Accordingly, tax savings on interestexpenses are not included in project cashflow analysis.

Almost always there are exceptions togeneral rules or practices and this is thecase for the general rule for the treatment offinancing flows. There are situations whereinterest charges are explicitly incorporatedinto the cash flows. The question here is notwhether it is right or wrong to incorporatefinancing flows into the cash flow analysis,but whether that incorporation is donecorrectly or incorrectly. If it is necessary toshow the financing flows explicitly in thecash flows, as is often preferred by non-financial mangers and chief executives, it isquite possible to include them explicitly in acorrect and consistent manner withoutdistorting the final results.

In property investment analysis, ‘property’cash flows are distinguished from ‘equity’cash flows. Property cash flows are theequivalent of ‘project’ cash flows discussedin this module. Property cash flowcalculations do not include loan repaymentsand interest charges as deductions. Thisapproach is consistent with the generalcash flow definition in this module. ‘Equity’cash flow calculations deduct loanrepayments and interest expenses.

One of the objectives of propertyinvestment analysis is to evaluate the returnto the investor under various debt and taxsituations. A mortgage applies specificallyto one particular property, rather than beingan unidentified part of the capital structureof a corporation. Some investment inproperty is to gain tax advantages frominterest deductions associated with debtfinancing. For these reasons, in propertyinvestment analysis, equity cash flows arecalculated after deducting loan repayments(principal plus interest) from other cashinflows to enable the effects of borrowingand taxation to be evaluated.

Inflation and consistent treatment ofcash flows and discount rates

Inflation will have an effect on the expectedcash flows of the project. Both cash inflowsand outflows could be affected by inflation.Market rates, such as interest rates andequity returns, in general will also rise whenthe expected inflation rate is high. As themarket rate rises the required rate of returnby investors will also rise. To deal withinflation appropriately, the project analysismust recognize expected inflation in theforecast of future cash flows and use adiscount rate that reflects investors’expectations of future inflation.

If all cash flows as well as the discount rateare equally affected by expected inflation,the net present value is the same whetherthe inflation is include or excluded.However, most projects will consist of amultitude of cash flow items over a numberof years and it would be erroneous toassume that all of the cash flows wouldincrease at exactly the same rate eachyear, or to assume the same effect on thediscount rate. Some cash flows areunaffected by inflation while other cashflows are affected to varying degrees byinflation.

An outstanding example of the differentialimpact of inflation on a project’s cash flow isthe depreciation tax shield. Tax-allowabledepreciation is totally unaffected byinflation. Depreciation tax shields arecalculated on the basis of historical costs ofthe assets (cost at the time of theiracquisition). Similarly, a long-term raw-material contract or the purchase of acommodity in the forward or futures marketsmay lock in the present prices therebyinsulating the cash flow from inflationaryeffects.

Given the differential impact of inflation ondifferent cash flow components, cash flowforecasts in nominal terms – incorporatingthe inflationary effect – have an advantageover cash flow forecasts in real terms –excluding the inflationary effects. That is,nominal cash flow forecasts can incorporatepotentially different inflationary trends inproduct price, labour and material costs,and so on, into cash flow estimates by


applying different inflation rates for differentcomponents of the cash flow.

The required rate of return used fordiscounting cash flows is normally derivedfrom observed market rates such as interestrates and return on equity. These observedmarket rates usually have the expectedannual inflation rates built in and are usuallyquoted in nominal terms (as opposed to realterms).

Observed market rates expressed innominal terms can, if necessary, beconverted into their real terms using thealgebraic relationship expressed in theFisher effect. The Fisher equation is:

(1+ n) = (1 + r)(1+ p) (1)

where, n = the annual nominal interest rate(expressed as a decimal value)

r = the annual real interest rate(expressed as a decimal value)

p = the expected annual inflationrate

From the above equation the real interestrate can be easily derived as

1)1(

)1(−

+

+=

p

nr (2)

Consistency in the discounted cash flowanalysis requires that if the project’s cashflows are in nominal terms then they shouldbe discounted by nominal discount rates,and if the project’s cash flows are in realterms they must be discounted by realdiscount rates. Real and nominal quantitiescannot be mixed and matched.

The interest rate used for discounting cashflows is normally derived from observedmarket rates. In an efficient financialmarket, investors’ required rate of return willinclude a component, (1+p), to compensatefor expected inflation. The use of observedmarket-required rates then implies thatinflation should incorporated into cashflows, to be consistent.

On rare occasions, real cash flows areappropriate in the analysis. In suchsituations, the real discount rate can be

calculated from market (nominal) ratesusing the Fisher equation presented above.

4. THE DISCOUNT RATE

The discount rate used in an analysisgreatly affects the NPV estimate. Thechoice of an appropriate discount rate thusbecomes critical in the appraisal of anyproject. Differences in the focus of CBA andFA once again tend to mean differentapproaches are used in determining theappropriate discount rate. The focus on theindividual or firm in financial analysis meansthat the discount rate is firm specific. Inessence, it is the risk-adjusted rate that thefirm or individual considers appropriate forthe project being evaluated. Often thisequates to the ‘cost of capital’ for theproject. If a project were to be financedentirely by a loan, the cost of capital wouldbe the rate at which the firm could borrowthe money for the project. If this rate wassay 9% per year, then this is the cost ofcapital and the appropriate discount ratewould be 9%. This is a rather simplisticexample and in real life things are muchmore complicated. As in CBA, a sensitivityanalysis involving the application of anumber of discount rates is usuallyundertaken, e.g. using the preferred rateplus or minus two percentage points.

What are the components of a discountrate?

In most standard financial analyses a riskadjusted discount rate (RADR) is used.Conceptually, the RADR has threecomponents:

• Risk free rate to account for the timevalue of money (r).

• A n average risk premium tocompensate investors for the fact thatthe company’s assets (or investments)are risky. This is a risk premium toaccount for the business risk of thefirm’s existing activities, being simplythe average risk premium for the firm.This is denoted as ‘u’.

• An additional risk factor (which couldbe zero, negative, or positive) toaccount for the difference in the riskbetween the firm’s existing business


and the proposed project, which isdenoted as ‘a’.

Thus, conceptually, the RADR (which willbe referred to as k) may be expressed in analgebraic form as:

k = r + u + a. (3)

If the risk of the proposed project is sameas the average risk of the firm’s existingprojects, the additional risk factor ‘a’ is zero.The required rate of return (or the RADR) tobe employed as the discount rate forprojects of average risk, is then ‘r+u’.

Estimating the RADR

In estimating the RADR to be used inproject evaluation, in practice, the threedifferent components may not always becalculated separately and added. Forexample, if cost of capital is used toestimate the RADR, then a value of ‘r + u’ isobtained, to which can be added a value toaccount for ‘a’ (which could be negative).On the other hand, if capital asset pricingmodel (CAPM) is used to estimate theRADR, then separate values for r, u, and awill be obtained.

The risk-free discount rate, ‘r’ may bearrived at by examining government bondyields or insured banks’ term deposit rates.By examining data published in the financialpress, a suitable figure for ‘r’ can easily bearrived at. When doing so, it is important tothat use a bond, guaranteed loan or otherfinancial asset with a similar term to that ofthe project. For example, if a projectstretched over 5 years, then the rate on a 5-year government bond or a 5-year insuredfixed deposit would be adopted.

The Average Risk Premium for the Firm, ‘u’may be estimated using the firm’s cost ofcapital. The weighted average cost ofcapital (WACC) is normally adopted forcalculating this. Another approach toestimate ‘u’ is to use capital asset pricingmodel (CAPM). Illustration and applicationof these involve relatively lengthycalculations.

The Additional Risk factor ‘a’ is thedifference between the average risk facedby the firm and the perceived risk of theproposed project.

5. ANALYSING PROJECT RISK:SENSITIVITY AND BREAKEVENANALYSIS

There are numerous ways to analyseprojects for risk. Two of the most widelyused are sensitivity analysis and breakevenanalysis.

Sensitivity analysis

One of these is to evaluate the projectunder various scenarios in which selectedparameters or variables are steppedthrough their pessimistic, most likely andoptimistic values. In this analysis, often onlyone parameters at a time is changed. Theresult ing set of NPVs reveal tomanagement which variables have amaterial impact on financial performance ofthe project. Management can then decide toeither invest more time and effort inestablishing more reliable forecasts forthese parameters, or to abandon the projectbecause of excessive risk.

The steps in sensitivity analysis are:

1. Calculate the project’s net presentvalue using the most likely estimatefor each parameters.

2. Select from the set of parametersthose which management feels mayhave an important bearing on theproject.

3. Forecast pessimistic, most likely andoptimistic values for each of theseparameters over the life of theproject.

4. Recalculate the project’s net presentvalue for each of the three levels ofeach parameters. Whilst eachparticular parameters is steppedthrough each of its three values, allother parameters are held at theirmost likely values.

5. Calculate the change in net presentvalue for the pessimistic to theoptimistic range of each parameters.

6. Identify parameters for which projectfinancial performance is sensitive.


Break-even analysis

Break-even analysis is a special applicationof sensitivity analysis, which endeavours tofind the levels of individual parameters atwhich the project’s NPV is zero. Forexample, management may wish to knowhow low can the unit selling price can fallbefore the project becomes unsuccessful. Ifmanagers know that this ‘cut-off’ price islikely to eventuate, then they may decidenot to proceed with the project.

In common with sensitivity analysis,parameters selected for break-evenanalysis are typically tested only one at atime. Within an Excel spreadsheet, thecalculations can be carried out in threeways:

1. by creating a data table for a range ofvalues and reading off the appropriatevalue at a zero NPV.

2. by trial-and-error substitution ofvariable values within a spreadsheet

3. by using the Excel ‘Tools – GoalSeek’ function.

Management can use the break-evenresults in two ways. They may decide toabandon the project prior to implementationif forecasts show that near break-evenparameter levels can be expected to occur.Once a project is implemented,management can react to a worst-casescenario involving the investigatedparameters, e.g. suspend production, try tomake production more efficient, or adjustthe unit selling price.

6. DETAILED EXAMPLE OF AFINANCIAL ANALYSIS OF AFORESTRY PROJECT

A case study of a forestry investment is nowpresented to illustrate the steps of financialanalysis. The example concerns a company(lets call it Flores Venture Capital Ltd, orFVC Ltd) which is considering diversifyingits operations into forestry. The companyhas obtained a consultant’s report on theproposal, which involves the establishmentof a plantation estate of 1000 ha in an areawith suitably high rainfall and soil quality. Toencourage investment in the region, a localgovernment has offered the required landrent-free for the period of the project on the

condition that a native species rather thanan exotic pine is grown. The company hasdecided to establish a mixed species(eucalypt and rainforest species) plantation.The finance department of FVC Ltd hasindicated that a rate of return of 7% isrequired for the project.

Step 1: Identifying the forestry system

In Module 3, it was demonstrated how theDelphi method could be used to developestimates of key parameters. The exampleused was for a forestry project involving theuse of two species for which littlequantitative growth and harvest age dataexisted. In this module, it is demonstratedhow a forestry financial evaluation can beundertaken using similar data. The examplein Module 3 was based on a real-life Delphisurvey and the associated financial analysiswas highly complicated; the basicinformation from that example will be usedhere, but will be simplified in a number ofways.

Because there was no past experience ingrowing native species in the region, theconsultant used the Delphi method todevelop an appropriate silviculture (treegrowing) system. Based on the results ofthis investigation, it is recommended thattrees be planted at a density of 660 stemsper hectare. It is expected that extensiveweed control will need to be undertaken inthe first year, with further weed controlrequired in the second and third year.Pruning of the trees to ensure good formwill be required in years 2, 4 and 6. Anumber of experts involved in the Delphisurvey indicated that the amount of pruningrequired is difficult to estimate because noone knows how much branching will occuron the sites and this could range fromminimal to requiring high labour inputs. Theconsultant has explained that pruning iscrucial because it ensures that the treesproduce a straight, knot-free log for whichhigh prices can be obtained. It was alsorecommended that each pruning eventshould be certificated by an external partybecause this will increase the likelihood ofbeing able to obtain a premium price forknot-free wood. A non-commercial thin isrequired at year 8, at which time 320 treeswill be removed.


The first revenue from the plantation ispredicted to come when a commercialthinning occurs at the end of year 17, atwhich time about 85 trees will be harvested.At year 26, 85 further trees will be cut andsold for telephone and electricity poles. Thebest 85 trees will be left to grow until year34, when about half (42) will be cut forsawlogs. The remaining trees will be

allowed to grow to year 60 when they willbe harvested and sold as high qualityveneer logs.

From the above information about theplantation system, the major cash outflowand inflow categories have been identifiedas listed in Table 1.

Table 1. Main cash categories and predicted timing

Cashflow category Nature of cashflow Timing (year)

Establishment (capital) Planning and design 0costs Incidental clearing 0

Site preparation and cultivation 0Cover crop establishment 0Pre-plant weed control 0Cost of plants 0Planting and refilling 0Post plant weed control 0Fertilizer 0Fencing 0

Maintenance costs Post plant weed control (1) 1Post plant weed control (2) 2Post plant weed control (3) 3First prune(plus certification) 2Second prune (plus certification) 4Third prune (plus certification) 6Thinning – non commercial 8

Annual costs Protection and managementLand rental (if applicable)

Cash Inflows Thinning revenue 18Revenue from poles 26Revenue from 1st harvest 34Revenue from 2nd harvest 60

Step 2: Estimating cash outflows

Estimates are now made of the likelyquantum of the cash outflows associatedwith the categories of Table 1. This hasdrawn on information from a number ofsources, e.g. quotes sought for the cost of

establishing the plantation and for pruningcosts based on past experience. Table 2provides the financial estimates of each ofthese activities provided by the consultant.For convenience, all estimates areexpressed on a per hectare basis.


Table 2. Cash outflows and timing associated with a two-species plantation

Cost group Nature of cash outflow Timing (yr) Cost ($/ha)

Establishment costs Planning and design 0 74Incidental clearing 0 158Site preparation and cultivation 0 265Cover crop establishment 0 88Pre-plant weed control 0 92Cost of plants 0 450Planting and refilling 0 645Post plant weed control 0 540Fertilizer 0 83Fencing 0 560Sub-total 2955

Maintenance costs Post plant weed control (1) 1 1300Post plant weed control (2) 2 800Post plant weed control (3) 3 200First prune(plus certification) 2 800Second prune (plus certification) 4 800Third prune (plus certification) 6 800Thinning – non-commercial 8 500

Annual costs Protection and management 40Land rental 0

Step 3: Estimating cash inflows

Cash inflows arise from the harvest of thetrees. Harvest revenue is determined by thevolume of timber (in m3) produced multipliedby the stumpage price paid ($/m3). Forexample, if a commercial thinning occurs atyear 17 (as in the current example) andyields 170 m3 of timber with an estimatedsale (stumpage) price of $30/m3, this willresult in estimated cash inflows of$5100/ha.

Estimates of cash inflows are particularlydifficult to make for forestry investments.The long production cycle means that it isdifficult to estimate what stumpage priceswill be many years into the future. In rarecases, long-term supply contracts may besigned with a guaranteed sale price. Evenin these circumstances the uncertaintyassociated with harvest volumes meansthere remains considerable uncertaintywhen estimating cash inflows fromharvests. Some revenues may be obtainedfrom commercial thinnings part way throughthe production cycle, though stumpageprice is usually low due to small diametertrees and low timber quality. Typically thereis no market for thinnings of a very young

age, in which case the thinning processresults in a net cash outflow. This is thecase in year 8 of the current example wherea non-commercial thin is required costingan estimated $500.

The largest cash inflows from plantationswill come at the end of the production cycle.In the current example, final harvestrevenue takes place after 60 years,although another significant harvest occursat 34 years. For this case study, estimatesof harvest volumes and timing werecollected as part of the Delphi surveyundertaken by the consultants. Theseestimates, combined with estimates offuture timber prices (in nominal dollars andalso collected as part of the survey) can beused to estimate cash inflows (Table 3).The expert panel used in the Delphi surveythought high quality sawlogs of a nativehardwood produced at year 34 shouldachieve a price of $200/m3. Furthermore, itwas thought that it was likely that logsharvested at 60 years would be suitable forthe production of veneer, and attract apremium of 50% above the price of highquality sawlogs.


Table 3. Estimated cash inflows for 1000 ha plantation

Activity resulting in cash inflowYear ofharvest

No.stems/ha

Yield(m3/ha)

Stumpage($/m3)

Revenue($1000)

First thinning 17 170 170 $30 $5,100Second thinning (poles) 26 85 -- $148 per pole $12,580First harvest (sawlogs) 34 42 100 $200 $20,000Second harvest (sawlogs/veneerlogs) 60 43 270 $300 $81,000

Step 4: Developing the financial model

For simplicity, the tax component of theanalysis has been simplified. It is assumedthat all cash outflows are fully allowable asdeductions in the year that they are paid. Inmost years there is no revenue from theplantation against which to offset these taxlosses. It is however assumed these lossescan be claimed against income generatedfrom other company operations, producinga tax benefit equivalent to the amount of thenet cash outflows multiplied by theprevailing tax rate (30%).

A number of other assumptions have beenmade in the analysis that are worthmentioning. A nominal rate of interest of 7%has been used which includes a riskpremium of 3%. Prices of all cash-flowitems are assumed to increase over time ata rate equal to the underlying inflation rate,so that no adjustment of either cash inflowsor outflows are needed.

The computer spreadsheet package Excelprovides a convenient platform for financialanalysis, to calculate key financialparameters such as NPV and IRR. Such aspreadsheet is presented as Table 4.

Step 5: Undertake a sensitivity analysis

Once the financial model has been set up, itis a simple task to examine at the effect ofchanges in parameter levels on the projectperformance criteria. This includes analysiswith respect to required rate of return andwith respect to parameters which are notunder the control of the company.

The Excel Table function has been used toderive NPV values for a range of discountrates, and these have been plotted inFigure 2. Note that the IRR is the discountrate for which NPV is zero, i.e. the pointwhere the curve in Figure 2 crosses the x-axis. This analysis suggests that the projectis marginal in terms of financialacceptability. At the real rate of return of 7%required by management, the NPV for theproject is -$58,214, and the IRR is justbelow 7% (6.96%). Figure 2 also presentsthe land expectation value (LEV) or sitevalue for the project. This is the NPV for aninfinite sequence of identical rotations, andis useful to compare forestry projects ofunequal duration. The LEV will beconsiderably higher than the NPV for shortrotations, but for long rotations (such as inthe case study project) LEV will differ littlefrom NPV.

Parameters outside the control of thecompany which are likely to have mosteffect on NPV have been identified, andpessimistic, most likely and optimistic levelsidentified for each, as in Table 5.

The spreadsheet used in the calculation ofNPV and IRR for the most likely values inStep 4 has been used to recalculate thesevalues for the optimistic and pessimisticvalues for each of the parameters of Table6. Only one variable is changed at a time,while all the other variables are held at theirmost likely values. The ‘Scenario’ functionin Excel allows multiple scenarios to bedeveloped and the results reported in atable in a separate spreadsheet. Thisfunction has been used to undertake thesensitivity analysis, results of which arereported in Table 6.

Table 4. N

PV

calculations for forestry investment, w

ith 7% discount rate ($1000s)

Annual protection and m

anagement

-40-40

-40-40

-40-40

-40-40

-40-40

-40-40

-40-40

-40-40

TO

TA

L OP

ER

AT

ING

EX

PE

NS

ES

-1340-1440

-240-640

-40-640

-40-540

-40-40

-40-40

-40-40

-40-40

OP

ER

AT

ING

RE

VE

NU

E

Thinning revenue 1 (year 17)

5100

Thinning revenue 2 (year 26)

12580

Harvest R

evenue 1 (year 34)20000

Harvest R

evenue 2 (year 60)81000

Total O

PE

RA

TIN

G R

EV

EN

UE

00

00

00

00

00

51000

125800

200000

81000

TA

X P

AID

OR

TA

X B

EN

EF

IT (30%

)886.5

402432

72192

12192

12162

12-1518

12-3762

12-5988

12-24288

NE

T C

AS

HF

LOW

S (O

perating revenue - capitaloutlays - operating costs)

-2068. 5-938

-1008-168

-448-28

-448-28

-378-28

3452-28

8778-28

13972-28

56672

NE

T P

RE

SE

NT

VA

LUE

OF

CA

SH

OU

TF

LOW

S A

TD

ISC

OU

NT

RA

TE

SP

EC

IFIE

D-58.214

INT

ER

NA

L RA

TE

OF

RE

TU

RN

6.96%

CA

SH

FLO

W IT

EM

Tim

ing of cash flow (end of year)

01

23

45

67

89-16

1718-25

2627-33

3435-59

60

ES

TA

BLIS

HM

EN

T C

OS

TS

(Capital C

osts)-2955

OP

ER

AT

ING

CO

ST

S

Post plant w

eed control-1300

-800-200

First prune(plus certification)

-600

Second prune (plus certification)

-600

Third prune (plus certification)

-600

Thinning (non-com

mercial)

-500

Land rental (capitalised @ 4%

of $2000/ha)


-5000.0

0.0

5000.0

10000.0

15000.0

20000.0

3% 4% 5% 6% 7% 8% 9% 10%

NPV

LEV

Figure 2. NPV Profile of FVC Ltd. forestry investment ($1000)

Table 5. Parameters selected for sensitivity analysis

Uncontrollable variable Pessimistic($1000)

Most likely($1000)

Optimistic($1000)

Stumpage price, first thinning ($/m3) 25 30 35Stumpage price, poles ($/pole) 110 148 200Stumpage price, first harvest ($/m3) 100 200 300Stumpage price, second harvest ($/m3) 150 300 450Yield, first thinning (m3) 120 170 190Yield, poles (number) 70 85 85Yield, first harvest) (m3) 80 100 150Yield, second harvest (m3) 220 270 350Establishment costs ($) 3455 2955 2655Post plant weed control, year 1 ($) 1800 1300 1000Post plant weed control, year ($) 1100 800 600Post plant weed control, year 3 ($) 400 200 0Pruning, years 2, 4, 6 ($) 1000 600 500Thinning costs, year 8 ($) 800 500 400

Table 6. NPVs for forestry investment for pessimistic and optimistic parameter levels ($1000)

Parameter Pessimisticestimate

Optimisticestimate

Range Rank

Stumpage price, first thinning ($/m3) -247 130 377 10Stumpage price, poles ($/pole) -448 475 922 4Stumpage price, first harvest ($/m3) -760 643 1403 1Stumpage price, second harvest ($/m3) -547 431 978 3Yield, first thinning (m3) -391 75 465 9Yield, poles (number) -326 -58 268 12Yield, first harvest) (m3) -339 643 982 2Yield, second harvest (m3) -239 232 471 8Establishment costs ($) -408 152 560 6Post plant weed control, year 1 ($) -385 138 523 7Post plant weed control, year ($) -242 64 306 11Post plant weed control, year 3 ($) -172 56 229 13Pruning, years 2, 4, 6 ($) -703 103 806 5Thinning costs, year 8 ($) -180 -17 163 14


The results of the sensitivity analysis couldbe now used by the company as a guide towhich parameters need to be investigatedfurther. From the sensitivity analysis it isclear that the stumpage prices for the firstand second harvests and for poles, alongwith the yield for the first harvest, are theparameters that have the greatest effect onproject NPV. The company may investigateways to reduce the impact of uncertaintywith respect to these parameters, such asthrough investing further resources intodeveloping more accurate yield predictionsor investigating further projections of futuretimber prices. It may also use the existingdata to undertake an investigation of thethree most critical parameters at a greaternumber of values within the range frompessimistic to optimistic.


This module has shown that the approachrequired for financial analysis differssomewhat from that of cost-benefit analysis.Key differences arise in treatment of socialand environmental externalities and transferpayments such as taxation. In corporateproject evaluation, often real discount ratesare used, with allowance for differinginflation rates with respect to cost andrevenue items. The discount rate may alsodiffer from that employed in CBA.

The case study illustrates a stepwiseapplication of a spreadsheet package toperform financial analysis includingdetermining performance criteria such asNPV and IRR and carrying out sensitivityanalysis is relatively straightforward,providing reliable cost and revenue datacan be obtained for a forestry project.

191

17. Forestry Applications of Linear Programming

Steve Harrison

Linear programming (LP) is a highly versatile mathematical optimization technique which hasfound wide use in management and economics. It is used both as a research technique andas a planning tool, particularly at the individual firm and industry level. In general, LP isdesigned to maximize or minimize a linear objective function subject to a set of linearconstraints. Linear programming is one of a group of techniques which may be referred to asmathematical programming. Other related techniques are goal programming, mixed integerprogramming and quadratic programming. Some typical applications of linear programminginclude:

1) determining the most profitable combination of enterprise or activity levels for abusiness firm with limited supplies of various resources

2) determining the most profitable investment portfolio, given the amount ofinvestment capital available, rates of return on various stocks, bonds and other‘paper assets’, and limits on high-risk investments.

3) formulating mixtures to combine ingredients such that a required overallcomposition of the mix is satisfied at least cost. Important applications are fueland fertilizer blending and determination of livestock rations or supplementaryfeeds.

4) scheduling the various tasks in a construction project so as to complete theoverall project in minimal time or at minimal cost

5) determining the location and size of storage facilities and processing plantstogether with the distribution pattern, so as to minimize the total of transport,storage and processing costs.

As illustrated by this list, the range of applications of linear programming is indeed wide, andthis is one of the most widely used operations research techniques. In this module, thealgebraic formulation of LP models in explained, and a simplistic decision problem isformulated in a linear programming framework, and is then solved graphically to illustrate thebasic principles of the technique. Computer solution of this problem using the simplex methodis then demonstrated, with reference to setting up the model on a spreadsheet andinterpreting the output. Further applications to transshipment modeling, capital budgeting andgoal programming are then illustrated.

1. ALGEBRAIC FORMULATION OFLINEAR PROGRAMMING PROBLEMS

In mathematical terms, the purpose of linearprogramming is to optimize a linear objectivefunction subject to a set of linearconstraints. The objective function may, forexample, express the profit from acombination of enterprises or activities, thecost of a combination of ingredients, or thetime required to complete a series of tasks.In these contexts, optimization may meaneither maximization of profits orminimization of costs or time. The nature oflinear programming will be illustrated with

reference to a profit maximization problem.A highly simplified example has beenchosen deliberately, so as to illustrate thetechnique without undue computationaldistraction. More realistic applications will bedeveloped later in this module.

Example 1

A cabinet-maker produces dining roomsuites and grandfather clocks out ofAustralian red cedar timber. He can obtainannual supplies of up to 600 linear metres(lm) of red cedar (one linear metre isequivalent to one metre in length by one


metre in width by one centimetre in depth).Up to 4000 hours of labour a year areavailable for operations such as sawing,joining and polishing. Each dining room suiterequires 12 lm of timber, and eachgrandfather clock requires 7.5 lm of timber.One hundred hours of labour are required toproduce a dining room suite, and 40 hours toproduce a grandfather clock. The profit fromeach dining room suite is $900, and thatfrom each grandfather clock is $450. Thecabinet-maker wishes to know how manyunits of each furniture line to produce inorder to maximize profits. Formulate thisdecision problem as a linear programmingmodel.

Formulation of the model

Before this problem can be solved bygraphical or other means, it must beexpressed in algebraic form, i.e. as a model.The first step is to introduce an ‘x’ notationfor the decision variables, here numbers ofsuites and clocks to be produced. Thus w elet x1 = number of dining room suitesproduced, and x2 = number of grandfatherclocks produced.

It is now possible to formulate an objectivefunction which in this case is an equationdefining total profit. The term ‘profit’ is usedloosely here, in that the figures of $900 and$450 are more correctly called ‘grossmargins’ for the two activities, i.e. they arereturns net of variable but not fixed costs. Inobtaining these figures, allowance is madefor allocatable costs such as materials,labour and marketing costs, but not foroverheads such as rent on premises orrates, depreciation of equipment, andaccountancy. In this module the term netrevenues will be used and the objectivefunction will be referred to as a revenuefunction. If each dining room suite has a netrevenue of $900, then the total of netrevenues from producing (and selling) x1

dining room suites will be 900 x1 dollars.Similarly, the total net revenue fromproducing x2 grandfather clocks will be 450x2. If the symbol Z is used to represent totalnet revenue, the objective function may bewritten as

Z = 900 x1 + 450 x2

The objective can now be identified moreprecisely as finding those values of x1 andx2 for which Z is a maximum, bearing in mindthe restrictions on production imposed bylimited supplies of timber and labour.

Resource restrictions also can beexpressed in algebraic form. If x1 diningroom suites are produced, each requiring 12lm of red cedar timber, then dining roomsuites will consume a total of 12 x1 lm oftimber. Similarly, if x2 grandfather clocks areproduced these will consume 7.5 x2 lm oftimber. The total amount of timber consumedcannot exceed the supply, so the productionplan is constrained by the inequalityexpression

12 x1 + 7.5 x2 ≤ 600

The left-hand-side of this expressionindicates the amount of timber which will beused for any production policy (combinationof x1 and x2 levels); the right-hand-sideindicates timber supply. This timberconstraint ensures that the demand fortimber cannot exceed the supply; anyproduction plan consuming more timberwould violates this constraint and wouldtherefore be infeasible.

Similar reasoning can be applied to derive alabour constraint. Since suites and clocksrequire 100 and 40 manhours of labourrespectively, and since the labour supply is4000 manhours, feasible levels of x1 and x2

are bounded by

100 x1 + 40 x2 ≤ 4000

Two further constraints are necessary todefine the decision problem fully. These arethat the numbers of suites and clocksproduced cannot be negative

i.e. x1 ≥ 0 and x2 ≥ 0

Non-negativity constraints may at firstappear unnecessary in a practical sense;after all, it is not possible to producenegative numbers of suites or clocks.However, they must be included formathematical completeness, to delineatefully the feasible region of production. The

Forestry Applications of Linear Programming 193

cabinet-maker's decision problem may nowbe summarized as a linear programmingmodel with an objective function, twoactivities (dining room suites andgrandfather clocks), two resourceconstraints (timber and labour) and twonon-negativity constraints, as follows:

maximize revenue Z = 900 x1 + 450 x2

subject to the resource constraints

12 x1 + 7.5 x2 ≤ 600

100 x1 + 40 x2 ≤ 4000

and non-negativity constraints

x1 ≥ 0, x2 ≥ 0

General algebraic formulation

For the more mathematically inclined, theabove problem formulation may begeneralized in abstract terms for greaternumbers of activities and constraints.Suppose levels are to be determined for nactivities (x1, x2, ... xn) yielding individual netrevenues c1, c2, ... cn. Limited supplies areavailable of m resources, the supply levelsbeing b1, b2, ... bm. Each activity uses fixedamounts of resources; in particular, therequirement of resource i by one unit ofactivity j is aij. The elements of the matrix ofaij values (for i=1 to m and j=1 to n) areknown as technical or input-outputcoefficients. The linear programming modelmay now be written as

maximize Z = c1 x1 + c2 x2 + ... + cn xn

subject to the linear resource constraints

a11 x1 + a12 x2 + ... + a1n xn ≤ b1

a21 x1 + a22 x2 + ... + a2n xn ≤ b :

am1 x1 + am2 x2 + ... + amn xn ≤ bm

and the non-negativity constraints

x1 ≥ 0, x2 ≥ 0, ... xn ≥ 0.

For readers familiar with matrix algebra, thedecision problem may be expressed morecompactly in matrix notation, as

maximize Z = CT X

subject to A X ≤ B and X ≥ 0

where C is a vector of activity net revenues(and CT is the transpose of C) ;

X is a vector of activity levels; A is a matrix of input-output

coefficients; B is a vector of resource supplies;

and 0 is the null vector.

2. GRAPHICAL SOLUTION OF RESOURCEALLOCATION MODELS

Having expressed the production planningproblem as an algebraic model, it is now timeto derive optimal levels of the decision orpolicy variables. For simple problems suchas this (with two activities), solution bygraphical means is possible. Graphicalpresentation provides a number of practicalinsights into the nature of the decisionproblem and its solution. For the aboveexample, a graph is set up as in Figure 1, inwhich the number of dining room suites (x1)is measured on the horizontal axis and thenumber of grandfather clocks (x2) isrecorded on the vertical axis. The resourceand non-negativity constraints are enteredas straight lines on this graph, delineatingthe feasible region.

To fix the position of the timber constraintline, note that the maximal number of suiteswhich could be produced if all timber weredevoted to suites is the total timber supplydivided by the per unit demand, i.e. x1 =600/12 or 50. No timber would then beavailable for production of clocks, i.e. x2 = 0.On the other hand, if no timber weredevoted to suites (x1 = 0) then up to x2 =600/7.5 or 80 clocks could be producedbefore the supply of timber becameexhausted. The two end points of the timberconstraint – (0,80) and (50,0) – are drawnin Figure 1. If half the timber were used forsuites and half for clocks then up to 25suites and 40 clocks could be produced,corresponding to midway on the straight linejoining the above end points. In fact, anypoint on the straight line between (0,80) and(50,0) corresponds to a production plan inwhich exactly 600 lm of timber are used.


Any combination of x1 and x2 levels on or tothe left of (below) this line is feasible interms of the timber constraint 12x1 + 7.5x2 ≤600. Any point to the right would use morethan 600 lm of timber and violate thisconstraint.

Applying similar reasoning to that above, thelabour constraint 100x1 + 40x2 ≤ 4000 canbe drawn as a straight line connecting x1 =4000/100 = 40 and x2 = 0 on the horizontalaxis to x2 = 4000/40 = 100 and x1 = 0 on thevertical axis, as in Figure 2. Any productionplan represented by an (x1,x2) pair on or tothe left of this line is feasible in terms oflabour use, while any production planrepresented by a point to the right isinfeasible. The combined effect of the timberand labour constraints is to restrict thefeasible region to on or to the left of the lineBCD in Figure 2.

It was noted earlier that in addition toresource constraints, non-negativity con-straints must be imposed on the feasibleregion, these being of the form x1 ≥ 0 and x2

≥ 0. In Figure 2, the constraint x1 ≥ 0confines solutions to on or to the right of thevertical axis, while x2 ≥ 0 means any pointon or above the horizontal axis. The feasibleregion of production is now defined fully; itis the irregular area ABCD. Any numbers ofsuites and clocks corresponding to an(x1,x2) pair within or on the boundary of thisregion is feasible in that it will not use moretimber or labour than is available. If a pointcorresponding to an (x1,x2) pair is not on theproduction frontier BCD then some timberand labour remain unused.

Having defined the feasible region it is nowpossible to determine the most profitableproduction policy, i.e. the policy for whichthe objective function Z takes as large avalue as possible. Revenue considerationsare introduced by drawing lines on thegraph joining equally profitable (x1,x2)combinations, called isorevenue lines. Thelocation of the initial isorevenue line isarbitrary. Suppose, for example, the policyof producing 30 dining room suites and noclocks were chosen; the total net revenuewould then be

Z = 900(30) + 450(0) = 27,000 dollars

This same level of revenue could beachieved by producing no suites and 27,000/ 450 or 60 clocks. In fact, any combinationof (x1, x2) values along the line from (30,0)to (0,60) corresponds to a total net revenueof $27,000. The $27,000 isorevenue line isdrawn as a broken line in Figure 3.

Various other isorevenue lines could bedrawn, as illustrated in Figure 3. Each ofthese other lines is parallel to the $27,000isorevenue line. The higher the revenue thefurther the line from the origin. For example,the $31,500 isorevenue line corresponds tothe production of 35 suites or 70 clocks orany combination along the straight linejoining these points. The most profitable planwould be indicated by the (x1,x2) pairtouching the highest possible isorevenueline. A parallel shift to the right for theisorevenue line reveals that this must be atpoint C. Reading from the graph, theco-ordinates for point C correspond to anoptimal plan of producing approximately 22dining room suites and 44 grandfatherclocks. The total net revenue from this planis approximately 900(22) + 450(44) or39,600 dollars.

It is to be noted that the above solution isapproximate only, having been readimprecisely from a graph. Procedures areavailable for finding a more precise solution;this is to produce 44.44 clocks and 22.22suites. However, in practice only wholenumbers of suites and clocks can beproduced, so these numbers could berounded downwards to 44 and 22respectively.


80

Number of clocks

(x2)

50 Number of suites (x1)

Figure1. Graphical representation of timber constraint

100 Labour constraint 80 B

Number of clocks C (x2) Feasible region Timber constraint

A D 40 50 Number of suites ( x1)

Figure 2. Feasible region of production (ABCD)



60 Number C of clocks (x2) $27,000 isorevenue Timber constraint line A D 30 40 50 Number of suites (x1)

Figure 3. Isorevenue lines and optimal plan

Sensitivity or stability analysis

The graphical solution procedure illustratedabove has identified the most profitablecombination of activity levels for given netrevenues, resource supplies, and so on.However, in practice the concept of a singlebest policy is of limited value; it is desirableto know under what circumstances thispolicy remains optimal, and how it wouldchange if any of the parameters werevaried. Decision makers frequently requireinformation about the sensitivity or stabilityof the optimal plan with respect to changesin net revenues, resource supplies orinput-output coefficients. Two relativelysimple but highly useful forms of sensitivityanalysis are net revenue ranging andcalculation of shadow prices.

Net revenue ranging

The fact that point C is optimal is due to therelative net revenues of clocks and suites. Ifthe relativity between net revenues were tochange then the slope of the isorevenue linewould change, and a different plan couldbecome optimal. For example, suppose thenet revenue of suites remains at $900 perunit, but that for clocks rises to $600. The$27,000 isorevenue line will now run from(30,0) on the horizontal axis to (0,45) on thevertical axis, as in Figure 4. A parallel shift

outwards in the isorevenue line reveals thatthe optimal plan is now at point B, i.e.production of 80 clocks and no suites, witha total net revenue of 80 times $600 or$48,000.

By examining different slopes of isorevenuelines on a graph such as Figure 4, it ispossible to determine within what range thenet revenue of one activity can vary for afixed net revenue of the other activity, whilethe current production plan remains optimal.The slope of any line on the (x1,x2)co-ordinate axis system is defined as thechange in level of the variable on the verticalaxis divided by the corresponding change inthe variable on the horizontal axis. For thelabour constraint, there is a decline in x2 of100 for an increase in x1 of 40, and hence aslope of -100/40 or -2.5. Similarly, the slopeof the timber constraint is -80/50 or -1.6. Theslope of the isorevenue line is the negativeof the ratio of net revenue of the activity onthe horizontal axis to net revenue of theactivity on the vertical axis, i.e. slope ofisorevenue line net revenue of suites = - (or - c1/c2) net revenue of clocks

= - $900 / $450 = -2



Number of clocks C 45 (x2) $27,000 isorevenue Timber constraint line A D 30 40 50 Number of suites (x1)

Figure 4. Determination of the optimal plan when the net revenue of clocks isincreased to $600

Provided the slope of the isorevenue lineremains between -2.5 and -1.6, the pointfurtherest from the origin touched by anisorevenue line will be point C. Nowsuppose the net revenue per suite is fixedby a sales contract at $900, but that forclocks is uncertain, depending on demandconditions. For point C to remain optimal, thenet revenue of clocks must be such that thenet revenue ratio remains between -2.5 and-1.6, i.e. - 2.5 ≤ -c1 / c2 ≤ - 1.6

i.e. 2.5 ≥ -c1 / c2 ≥ 1.6

Solving the two parts of this inequalityexpression for c2, it is found that

c2 ≥ c1 / 2.5 = $900/2.5 = $360, and

c2 ≤ c1 / 1.6 = $900/1.6 = $562.50.

That is, the net revenue for clocks can varyin the range $360 to $562.50 without optimalactivity levels changing.

A special case exists where the slope ofthe isorevenue line is exactly equal to that ofa linear segment on the boundary of thefeasible region. For example, if the net

revenues of suites and clocks were $900and $360 respectively then the slope of theisorevenue line would be -2.5, exactlymatching that of segment CD on theboundary of the feasible region. In this case,any point along CD would be equallyprofitable; there is no unique optimal policy.Further, small variations in net revenuesmay result in large changes to the optimalpolicy. This, of course, is a an unusualsituation in that it would only be bycoincidence that the slopes of theisorevenue line and one of the constraintlines were exactly equal.

Shadow prices

The computer solution method for LPproblems involves selecting activities to bebrought into the basis (solution) only if theirnet revenue exceeds their opportunity cost.When an LP problem is solved, a number ofactivities may be absent from the optimalsolution. These include real or productionactivities, and also what are called disposalactivities or resource non-use activities.The latter are a device to convertinequalities into equations to provide an initialfeasible solution for the simplex solutionmethod. In this context, the constraints forthe decision solved graphically above can


be written as

12 x1 + 7.5 x2 + 1 x3 = 600

100 x1 + 40 x2 + 1 x4 = 4000

where x3 and x4 are non-use or ‘disposal’ oftimber and labour respectively.

In the case of a maximization problem, for areal activity, the shadow price is the amountby which the net revenue would be reducedif one unit of this activity were forced intothe solution. Equivalently, it is the amount bywhich the net revenue would have toincrease before this activity entered theoptimal solution. For a disposal activity, theshadow price is the cost of forcing one unitof a resource into disposal or non-use, orequivalently the amount by which overallrevenue would increase if another unit ofthe resource were available. The calculationof shadow prices will not be illustrated here,but the concept will be discussed furtherwith respect to computer output.

3. APPLYING THE SIMPLEX METHOD TORESOURCE ALLOCATION MODELS

The above method of solving linearprogramming problems is quick and simplewhen there are only two activities, even ifthere are many constraints. But if there arethree or more activities, then thetwo-dimensional graphical approach breaksdown. However, a mathematical procedureknown as the simplex method has beendevised which can be used to solve LPproblems regardless of the numbers ofactivities and constraints (even when thereare 1000 or more of each). Themathematical basis of the simplex solutionalgorithm involves advanced matrix algebrawhich will not be explained here. While it isdesirable to have an understanding of theeconomic logic of the solution procedure, touse the technique it is only necessary to beable prepare the data (objective function,activities and constraints) for input to acomputer package and to interpret thecomputer output.Invariably, linear programming problems aresolved using a computer package. A varietyof computer programs are available forsolving LP problems. Provided the problem is

not too large – not more than a couple ofhundred activities and constraints – ExcelSolver may be used to obtain the optimalsolution and shadow prices. For illustrationpurposes, an extension of the cabinet-maker's decision problem in which there arethree activities and three constraints ispresented below, and the method of usingExcel Solver is demonstrated.

Example 2

Suppose the cabinet-maker of Example 1can produce roll-top desks as well as diningroom suites and grandfather clocks. Also,he wishes to distinguish between generallabour (for sawing, joining and polishing)and specialized woodcarving labour used indecorating his furniture. Supplies of timberand general labour are as in Example 1,while 700 hrs per year of woodcarvinglabour are available. The net revenue fromeach desk is $600. Suites and clocksrequire timber and general labour as inExample 1, plus five and eight hours ofcarving labour respectively per unit. Eachdesk requires 8 lm of timber, 60 hrs ofcarpentry labour and 7 hrs of carvinglabour. Assuming that the cabinet maker'sobjective is to maximize net revenue,determine the optimal production plan, netrevenue ranges and shadow prices.

Solution using Excel Solver

Table 1 presents the linear programmingformulation for this problem, set up on aspreadsheet ready for solution using Solver.The activities are represented across thecolumns of the tableau (columns B to D),and the resources down the rows (rows 5to 7), with the objective function as row 8.The technical coefficients form the body ofthe tableau (cell B5 through to D7). ColumnG lists the initial resource supplies orconstraint right-hand-sides.


Table 1. Extended cabinet maker’s problem in spreadsheet format

A B C D E F G1 Extended cabinet maker’s decision problem2

3 Constraint or objective Dining roomsuites

Grandfatherclocks

Roll-topdesks

Resourceuse

Sign Resourcesupply

4 Activity level 0 0 0

5 Timber 12 7.5 8 0 ≤ 600

6 General labour (hrs) 100 40 60 0 ≤ 4000

7 Carving labour (hrs) 5 8 7 0 ≤ 700

8 Net revenue 900 450 600 0

Prior to using Excel Solver, it is necessary tointroduce a row for activity levels (row 4);these activity levels are initially set at zero. Itis also necessary to introduce a column forresource use (column E). As well, a columnfor signs of the constraints is introduced(column F), in this case containing only ‘≤’signs.1

The most complex step in setting up theinitial tableau is to enter formulae in the‘Resource use’ column, i.e. column E:

1. the resource use for the timberconstraint (cell E5) is entered as theformula‘=SUMPRODUCT(B$4:D$4,B5:D5)’. Thecell value initially takes a level of zero,because the activity levels in row 4 arezero. Note that absolute cell referencesare required for row 4 (represented bythe dollar sign before the row number).

2. the contents of cell E5 are then copied tocell E6 through to cell E8. Thecoefficients in cell E5 to E7 represent‘resource use’ with respect to theresource constraints, while the value incell E8 is the level of the objectivefunction. Initial values in these cells areagain zero.

Once these data have been entered onto

1 If not available in the Excel version, thesesymbols may be copied and pasted from awordprocessor file. These signs are includedfor readability of the tableau only, and couldbe replaced by text, e.g. ‘le’ for ‘less than orequal to’.

the spreadsheet, Solver can be called up tofurther set up the problem for solution.Solver is to be found under the Tools menuof Excel. (If it is not currently available, seekassistance on how to access it.) Note thatthe general form of the Solver window is:

Set Target Cell:Equal to:By Changing Cells:Subject to the Constraints:

First declare the target cell. This is theobjective function, and is chosen by clickingin the total net revenue cell, here E8. Nextcheck that the maximization option of the‘Equal to’ row has been chosen, with a dotin the ‘Max’ circle. Cells B4 to D4 areselected as the changing cells, i.e. the levelsof the decision or x variables.

It is next necessary to add constraints. Firstclick on the ‘Add’ button. When addingconstraints, select corresponding cells incolumns E and G. In the ‘Add constraints’dialogue box, under ‘Cell reference’, click onthe cell range E5 to E7, then under‘Constraint’ click on cells G5 to G7. The sign‘≤’ is automatically selected as the default.Click on the Add or OK button to confirm thatthis constraint is to be added.

It is also necessary to add non-negativityconstraints. Hence click on the Add buttonagain, then select cells B4 to D4. In thiscase, the sign must be changed to ‘≥’, andthe value of zero must be entered on theright-hand-side (under ‘Constraint’). The OK


button is now pressed to confirm addition ofthis constraint. The constraints will nowappear in the dialogue box. The Changebutton may be used to modify theseconstraints if there are any apparent errors.

The Solve option may now be used to obtainthe optimal product mix. When clicking on theSolve button, the spreadsheet shouldchange to that of Table 2. (If this is not thecase, then there are errors in the tableau orSolver specifications such as errors inconstraints which need to be trackeddown.) The optimal solution is to produce30.76 clocks and 46.15 desks (these levelscould be truncated to whole numbers). Thetotal net revenue is $41,538. All of the timberand general labour is used, but about 130hours of carving labour are unused (or indisposal).

A little further information can be gleanedfrom through sensitivity or post-optimalityanalysis. This is not generated automatically– the ‘Sensitivity’ option must be chosenunder the ‘Reports’ options in the ‘SolverResults’ dialogue box. It is possible to movebetween the spreadsheet and the reportsby clicking on buttons near the bottom of thecomputer screen. Table 3 presents‘Sensitivity Report 1’ for this decisionproblem. In the first section of this table, inthe Reduced Gradient column, it is foundthat the net revenue of dining room suiteswould have to increase by a little over $69for this activity to enter the basis or solution.In the second section, under LagrangeMultiplier, it is found that the reduction in totalnet revenue would be $23.08 if one unit oftimber were removed and $6.92 if one hourof general labour were removed. These arethe amounts the cabinet-maker could affordto pay for additional units of theseresources and still be able to use themprofitably.

As indicated by Example 2, inclusion ofadditional activities and constraints pavesthe way to more realistic formulation ofdecision problems. Of course, if there arelarge numbers of activities and constraintsthen there will be large numbers of columnsand rows in the simplex tableau, andchecking of the tableau becomes morecritical. Even small linear programming

problems are usually solved on a computer,and commercial computer packages ofvarying levels of sophistication areavailable. Some of these packages performa more detailed post-optimality analysis thanthat presented above.

4. PRACTICAL DIFFICULTIES IN LINEARPROGRAMMING ANALYSIS

The linear programming technique asoutlined above is mathematically elegant,computer packages to derive optimal policiesare readily available, and the technique iswidely used. However, the user needs tobe aware of a few theoretical limitations andoccasional computational difficulties.

Correct perception of the role of linearprogramming

Linear programming is capable ofsimultaneously evaluating the profitability oflarge numbers of activities in the presenceof numerous constraints. It can solvedecision problems that are beyond thepower of intuition, pencil-and-papermethods and formal budgeting. However,the role of this powerful decision aid is oftenpoorly understood. The analyst carrying outa linear programming study in not usually theperson responsible for making policydecisions and taking the consequences ofthese decisions. It should not be expectedthat the optimal plan derived from a singlelinear programming analysis will beimplemented in precise detail bymanagement. Decision-makers are likely tohave particular hunches andpreconceptions as to the best course ofaction. Results generated by a linearprogramming study typically are used toreinforce or challenge existing views andtentative plans, i.e. to provide decisionsupport. Further, decision-makers oftenwish to ask a number of ‘what if’ typequestions, e.g. ‘What if suites can beproduced with 90 rather than 100 hours oflabour?’ or ‘What if the net revenue ofclocks is $500 rather than $450?’.


Table 2. LP spreadsheet after solution by Solver

A B C D E F G1 Extended cabinet-maker’s decision problem2

3 Constraint or objective Dining roomsuites

Grandfatherclocks

Roll-topdesks

Resourceuse

Sign Resourcesupply

4 Activity level 0 30.769231 46.15385 Timber 12 7.5 8 600 ≤ 600

6 General labour (hrs) 100 40 60 4000 ≤ 4000

7 Carving labour (hrs) 5 8 7 569.231 ≤ 700

8 Net revenue 900 450 600 41538.5

Table 3. Sensitivity report for cabinet-maker’s decision problem, as generated by Excel Solver

Final ReducedCell Name Value Gradient$B$4 Activity level Dining room suites 0 -69.23083027$C$4 Activity level Grandfather clocks 30.76923077 0$D$4 Activity level Roll-top desks 46.15384615 0

Final LagrangeCell Name Value Multiplier$E$5 Timber Resource use 600 23.07692308$E$6 General labour (hrs) Resource use 4000 6.923076923$E$7 Carving labour (hrs) Resource use 569.2307692 0

Results of post-optimality analysis andfurther computer solution runs will help toshed light on these questions. LP tends tobe used interactively to explore a number ofvariations to a basic model in a singlesession on the computer, and to produce agood deal of information about optimalactivity levels and sensitivity of profits orcosts to variations in parameter levels (dataestimates or assumptions). The broadpicture which is built up of the alternativesand consequences can be of considerablehelp in guiding decision making.

Assumptions of the technique

The linear programming model presentedabove has implicitly relied upon a number ofassumptions, the acceptability of which maybe questioned in particular applications.Planning horizon. In production planningapplications of linear programming, net

revenues make no allowance for overheadcosts. That is, the firm is assumed topossess a fixed asset structure and theanalysis concentrates on how currentresources should be deployed to maximizeprofits in the short term. The exampleswhich have been presented all rely onsingle-period or static models. In the longerterm the firm may acquire more land,factories, machines and so on. Planningthese capital acquisitions over time is aproblem of a different order to maximizinggross margins for a single productionperiod.

Single-valued expectations. It has beenassumed that gross margins, resourcesupplies and input-output coefficients areknown with certainty. In practice these areoften estimated with a good deal ofguesswork. Further, single point estimatesare made when in reality it would be more


meaningful to specify a probabilitydistribution of values, particularly in the caseof net revenues.

Linear objective function. The objectivefunction of the decision maker is assumed tobe a linear function, and to depend on netrevenues or input costs alone. Particularly inenterprise planning applications, thedecision-maker may be concerned aboutboth level of income and variability of incomeover time. Activities which have relativelylow expected net revenues may be chosenbecause of low risk associated with thisrevenue. Suppose in Example 2 that suiteshave an assured market outlet at apre-arranged price but clocks are solddirectly by the manufacturer, at priceswhich can vary widely depending on thewhim of the market. He may then producemore suites than indicated in the ‘optimal’plan because of the lower uncertainty aboutthe net revenue of suites.

Additivity within activities. The amount ofeach resource used per unit of each activityis assumed to be constant regardless of thelevel at which the activity is conducted. Ifone suite requires 12 lm of timber and 100hrs of labour then 20 suites require 240 lmof timber and 2000 hrs of labour. In practice,increasing or decreasing returns to variablefactors are common, e.g. as more suites areproduced the labour input per suite may bereduced due to more streamlined production.

Fixed resource proportions. Theinput-output coefficients are constant withineach activity vector, e.g. fixed proportionsof timber and labour are used to produceeach suite. It may be possible to use timbermore efficiently, including using offcuts, ifgreater time is taken in carpentry and joining.

Independence of activities. It is assumedthat the level at which any activity isconducted has no effect on input levels orrevenue of any other activity, i.e. there is nocomplementarity between activities. Ifoffcuts from production of suites could beused in production of clocks, then the timberrequirement of the two activities would beless than that of either activity in isolation.

Divisibility. Resource inputs and activity

levels are assumed to take a continuousrange of fractional units. In Example 1, the4000 hours of labour may be supplied bytwo full-time tradesmen, and any laboursupply level which is not a multiple of 2000hrs may not be practicable. Also, thesolution obtained for this decision problemincludes 30.77 clocks and 46.15 desks.However, clocks and desks can only beproduced in whole numbers. Anapproximate integer solution can be obtainedby truncating the final values of the decisionvariables, which here would yield 30 clocksand 46 desks. But we cannot be certain thatthis is the optimal integer plan.

In practice, procedures have been devisedto overcome most if not all of the limitationsimplied by these assumptions, though at thecost of greater tableau complexity.Multi-period models may be constructed toallow for expansion in the fixed asset baseover time. Variations in factor proportionsmay be readily accommodated, e.g. if labourand timber can be combined in differentproportions to produce suites then two ormore suite activities, each having differentinput-output coefficient vectors, may bedefined. If complementarity exists betweentwo activities then these may be combinedinto a single activity producing twoproducts. Optimal integer levels may beobtained using the mixed integerprogramming technique. The point to bemade here is that one should not bediscouraged from using the linearprogramming by what appear superficially tolimitations of the technique.

Some other potential problems

Sometimes a linear programming tableau isconstructed which possesses either nosolution or no unique optimal solution. Forexample, consider the following models:


Model 1

maximize Z = x1 + x2

subject to 2 x1 + x2 ≥ 7 2 x1 + x2 ≤ 6 and x1 ≥ 0, x2 ≥ 0

Model 2

maximize Z = x1 + x2

subject to 2 x1 + 3 x2 ≥ 7 4 x1 + x2 ≥ 6 and x1 ≥ 0, x2 ≥ 0

The constraints of Model 1 are contradictorysince 2 x1 + x2 cannot be at the same timegreater than or equal to seven, and lessthan or equal to six, so this problem has nofeasible solution. The objective function ofModel 2 can be increased without limit, sinceno upper bounds are imposed on the valuesthat x1 and x2 can take; this problem is saidto be unbounded. The usual cause of theseunsolvable problems is an error inspecification of the model. For example, it issimply not possible to have unlimitedresource supplies as implied by anunbounded problem.

Other problems can arise during solutionsuch as rounding errors and ‘degeneracy’.Specific procedures are available toovercome these. They will not be discussedfurther here.

5. OPTIMAL TRANSPORTATION ANDLOCATIONAL EFFICIENCY MODELS

The purpose of a transportation model isusually to determine the transport or‘shipping’ allocation or a commodity suchthat quantities are moved from variousorigins to various destinations at minimumcost.

Example 3

Suppose log timber is to be transported fromvarious production regions to variousmarkets in a small country, as in Table 4.The two domestic supply regions – centraland islands – can produce an annual turnoffof 100,000 and 120,000 cubic metres (m3)respectively, and up to 200,000 m3 can be

obtained annually as imports. Demands are200,000 m3 per year in the northernindustrial region, 100,000 m3/year in thewestern region and 70,000 m3/year in thesouth and east region.

Solution

An initial transportation table is set up as inTable 4. This summarises the timber suppliesfrom the various origins and demands at thevarious destinations. Transport costs fromorigin to destination in dollars per cubicmetre are as indicated in the body of thetable, e.g. $40/m3 for ‘shipping’ from thecentral region to the northern industrialregion.

The optimal transport allocation can beobtained by linear programming. If we let

Z = total transport costxij = number of units transported fromorigin i to destination j (1000m3)cij = cost of transport from origin i todestination j ($/m3)

then this decision problem may berepresented algebraically as:

minimize overall transport cost

Z= 40x11 + 25x12 + 20x13 + 30x21 +…+ 60x33

subject to supply and demand constraints

x11 + x12 + x13 ≤ 100x21 + x22 + x23 ≤ 120x31 + x32 + x33 ≤ 200x11 + x21 + x31 ≥ 200x12 + x22 + x32 ≥ 100x13 + x23 + x33 ≥ 70

plus non-negativity constraints on all ninetransport quantities (x11 ≥ 0, … x33 ≥ 0).

Since the quantities are in terms ofthousands of cubic metres, the objectivefunction will be expressed in terms of$1000s.


Table 4. Log supplies, demands and transport costs

Source Destination Supply (1000 m3)Northern

industrial regionWesternregion

South andeast

Central region 40 25 20 100Islands 30 20 35 120Imports 100 100 60 200Demand (1000m3) 200 100 70

The initial spreadsheet for this model is setup as in Table 5. The transport quantities areinitially set to zero (cells B6 to D8). A columnfor sums of supply allocations (column E) isadded, with elements set as the supply sumfor each origin, e.g. cell E6 is set equal tothe sums of the values of cells B6 to D6.Similarly, a row for demand allocations (row9) is added, with elements equal to the sumsallocated for the various destinations, e.g.the value in cell B9 is set equal to the sum ofvalues in cells B6 to B8. Transport costs areentered as a separate block (cells B12 toD14). A total transport cost (objectivefunction) cell is set up) with the formula

‘=SUMPRODUCT(B6:D8,B12:D14)’.

Solver is now called up under the Toolsmenu, and the various entries are made inthe dialog box. The target cell is set as cellC16, and the ‘Min’ option is chosen. Thechanging cells are selected as the rangeB6:D8. The constraints can now be entered.The column vector E6:E8 is set less than orequal to F6:F8 for supply constraints, therow vector B9:D9 is set greater thanB10:D10 for the demand constraints, andthe block of changing cells B6:D8 is setgreater than or equal to zero.

On solving this LP problem, Table 6 isobtained. No timber is to be transported fromthe central region to the northern region(allocation in cell B6 = x11 = 0), 100 units(thousand cubic metres) are transportedfrom the central region to the westernregion (allocation in cell C6 = x12 = 100) andso on. Although up to 200 units of timbermay be imported, the demands are satisfiedby imports of 150 units (cell E8).

A sensitivity table could be generated, andthis would indicate for example the amount

by which transport costs for each non-usedshipment path would have to fall beforetransport through this path would becomewarranted (would reduce overall cost).

Transhipment models

The structure of transhipment models issimilar to that of transportation models,except that intermediate destinations areadded where product is stored orprocessed. For example, timber could beproduced in a number of areas, processedinto plywood at a specific location, and thentransported to other areas for marketing.Changes in volume may take place withstorage or processing, in which case it isnecessary to standardize the volume units.In the case of processing, since the optimalshipment path will depend on processingcost at the various processing locations, butprocessing cost will vary with volume ofthroughput, it is usually necessary to adopta trial-and-error solution approach.

6. CAPITAL BUDGETING OR PORTFOLIOSELECTION MODELS

Another useful application of linearprogramming is to assist in the selection ofinvestment projects. In this application, theinvestment projects are treated as separateactivities, and the objective function isdefined in terms of net present values.


Table 5. Initial spreadsheet for timber transportation model

A B C D E F1 Timber transportation model23 Source Destination4 Northern industrial

regionWesternregion

South andeast

Sum ofallocations

Supply

56 Central region 0 0 0 0 1007 Islands 0 0 0 0 1208 Imports 0 0 0 0 2009 Sum of allocations 0 0 010 Demand 200 100 701112 Central region 40 25 2013 Islands 30 20 3514 Imports 100 100 601516 Total transport cost = 0

Table 6. Optimal log transport allocation

A B C D E F1 Timber transportation model23 Source Destination4 Northern

industrialregion

Westernregion

South andeast

Sum ofallocations Supply

56 Central region 0 100 0 100 1007 Islands 120 0 0 120 1208 Imports 80 0 70 150 2009 Sum of allocations 200 100 7010 Demand 200 100 701112 40 25 2013 30 20 3514 100 100 601516 Total transport cost = 18300


Example 4

Three plantation type options are availablefor 200 ha of community land, viz. fast-growing ‘softwood’ species (e.g. Gmelina)2,hardwoods (e.g. Mahogany) or mixedspecies planting. Resource constraintsinclude land, labour, and capital in the firsttwo years of plantation establishment. (Thetechnical coefficients are provided in Table7.) Over a 20-year period, two rotations ofsoftwoods or one rotation of the otherplantation types could be grown. $400,000in capital is available for use in the first twoyears, and may be supplemented byborrowing of up to $50,000 in each of theseyears. The natural resource managementagency wishes to determine whatcombination of the plantation types wouldprovide the greatest aggregate NPV.

Solution

The initial spreadsheet for this decisionproblem is provided as Table 7. This is in asimilar format to the earlier resourceallocation model, except that a few featureshave been added:

1. Provision is made to transfer unusedcapital from the first year to the second.This takes up or uses capital in the firstyear, but makes capital available in thesecond year, which is achieved byhaving a +1 coefficient for the first yearand a –1 coefficient for the second year.

2. Borrowing activities are introduced, withunits of $1000. These are ‘supply’activities, with negative coefficients (-1s)in the relevant capital rows.

3. The objective function coefficients for theborrowing activities are the loan interestand redemption costs, in year 20currency equivalents. As anapproximation, these are taken asnegative NPVs equal to the amountsborrowed. 3

2 Gmelina is not technically a softwood, but

has similar wood properties.3 Calculation of objective function coefficientfor the borrowing activity is rather complex.This is the present value of the interest plusredemption payments at the end of 20

4. Constraint rows are introduced for theborrowing activities, with +1 coefficientsin the columns of the borrowingactivities, so as to limit the amountsborrowed.

When this problem is solved, thespreadsheet of Table 8 is obtained. Theoptimal investment is to choose softwoodsonly, plant all 200 ha, and borrow $50,000 ineach year. Not all of the available labour isused. An aggregate NPV of $5.9M ispredicted.

Extensions to this portfolio selection modelinclude specifying dependent or mutuallyexclusive projects, and specifying that if aproject is to be introduced then the levelmust be sufficiently high to warrant thetooling up, i.e. imposing a minimum thresholdlevel. For example, it might be decided that atmost two of the three plantation type optionsconsidered above can be adopted, and thatthe minimum area for any plantation type is50 ha. These requirements can easily bebuilt into the model if mixed-integerprogramming – available within Excel Solver– is used.

Portfolio selection models can also bedesigned to take risk on asset returns intoaccount. For example, the variance as wellas the expected payoff for each investmentmay be estimated, and quadraticprogramming used to determine the optimalinvestment portfolio given the decision-maker’s degree of risk aversion.

years, and will depend on when therepayment is made and what is the inflationrate. If it is assumed that the loan interestrate is equal to the inflation rate, then asimple approximation may be made. If a loanwere taken out at the beginning of a year, atsay 15%, and paid at the end of the yearwith interest (i.e. $1.15 repaid for each dollarvalued), the present value of the repaymentwould be $1.15/1.15 or $1. Similarly, if aloan were repaid after 20 years, and interestwere compounded, the repayment wouldhave a present value of $1.


Table 7. Initial tableau for forestry investment portfolio decision problem

A B C D E G H H I J1 Portfolio selection model23

Constraint or objectiveSoft-

woodsNatives

Mixedspecies

Transfercapital,yr 1->2

Borrowcapital,year 1

($1000)


($1000)

Res-ourceuse

SignResource

supply

45 Activity level 0 0 0 0 0 06 Land (ha) 1 1 1 0 ≤ 2007 Labour (person days) 45 30 35 0 ≤ 100008 Capital, year 1 ($1000) 2 2.5 2.3 1 -1 0 ≤ 4009 Capital, year 2 ($1000) 0.5 0.5 0.7 -1 -1 0 ≤ 0

10 Max. loan, year 1 1 0 ≤ 5011 Max. loan, year 2 1 0 ≤ 5012 NPV ($1000) 30 28 24 -1 -1 0

Table 8. Final tableau for forestry investment portfolio decision problem

A B C D E G H H I J1 Portfolio selection model23

Constraint or objective Soft-woods

Natives Mixedspecies

Transfercapital,yr 1->2


($1000)


($1000)

Res-ourceuse

Sign Resourcesupply

45 Activity level 200 0 0 50 50 506 Land (ha) 1 1 1 200 ≤ 2007 Labour (person days) 45 30 35 9000 ≤ 100008 Capital, year 1 ($1000) 2 2.5 2.3 1 -1 400 ≤ 4009 Capital, year 2 ($1000) 0.5 0.5 0.7 -1 -1 0 ≤ 010 Max. loan, year 1 1 50 ≤ 5011 Max. loan, year 2 1 50 ≤ 5012 NPV ($1000) 30 28 24 -1 -1 5900

7. GOAL PROGRAMMING

A more systematic way of dealing withmultiple goals is through goal programming(GP). Here, a number of goals are identified,and a target or aspiration level is specifiedfor each. Deviational activities are thenintroduced in the tableau to allow forunderachievement or overachievement ofgoals relative to the aspiration levels. In

particular, an underachievement activity isintroduced for each aspiration floor, and anoverachievement activity is introduced foreach aspiration ceiling. A series ofconstraints is then set up for the goals. Inthe objective function, the coefficients arenot the activity payoffs. Rather, they are thecosts of underachievement andoverachievement of goals relative to theaspiration levels. The objective function now


states that the sum of shortfalls and over-runs, with appropriate coefficients for each,be minimized.

Because of the presence of deviationalvariables in goal programming, theconstraints are referred to as softconstraints, c.f. the hard constraints forresource limits. Even if some or all theaspiration levels cannot be met, the problemwill not be found to be infeasible (with anerror message when an attempt is made tofind the optimal solution). Rather, levels ofthe deviational variables in the solution willindicate the extent to which it is not possibleto meet the goal targets.

The question arises as to what coefficientsor weights to place on the deviationalactivities. Shortfalls can be expressed in avariety of different units, such as dollars,rate of return as a percentage, jobs andhectares. One approach would be to placea weight on each deviational variable torepresent the cost of goal under-achievement or overachievement. Applyinga system of weights to these is known asweighted goal programming. In practice,expressing the deviations in a common unit– say dollars – may be difficult andunnecessary.

Often a priority ordering can be establishedbetween goal dimensions, referred to aslexicographic or preemptive goalprogramming. For example, the requirementmay be to achieve the target dividend ratefirst, then NPV payoff, then job creation,then the environmental requirement. Thiscould be achieved by placing weights ofdifferent orders of magnitude on theshortfall activities. However, software hasbeen developed in which the priorityordering is stated as input data and thesolution algorithm enforces this priorityordering of goals. Often, it is possible tosolve lexicographic goal programmingproblems by using an appropriate system ofweighting on deviational variables (e.g. $1Mper unit shortfall for top priority goals, $0.1Mfor second priority goals, and so on.)

Example 5

A government wishes to support local

development projects with employment andenvironmental benefits. Three projects areidentified which will contributed to thesegoals, viz. road construction, tree plantingand public housing. Each kilometre of roadconstruction provides 200 jobs (personmonths), has net carbon release of 2tonnes, and requires expenditure of$100,000. Each hectare of tree plantinggenerates 3 jobs, sequesters 5 tonnes ofcarbon per year and requires $2000 ofcapital. Each house constructed provides 20jobs and requires expenditure of $30,000.Determine the government program whichmeets these goals most closely, if equalweights are attached to the three goals.

Solution

The initial tableau is set out as Table 9. Notethat shortfall activities are included for jobsand carbon sequestration (with coefficientsof +1 in the relevant constraint rows), andan expenditure over-run activity is included(with coefficient of +1 in the expenditurerow). The objective is to minimize the sum ofdeviations. Note that the constraints are ininequality form, rather than forcing exactequalities. The constraints are set up in theusual way, i.e. the formula

‘=SUMPRODUCT(B$4:G$4,B5:G5)

is entered in cell H5, and copied into cells H6to H8. For the objective function, this ineffect means summing the numbers of unitsfor the three deviation variables (values incells E4 to G4). The objective function is oneof minimization.

The solution to this problem is reported in thespreadsheet of Table 10. The recommendedportfolio includes about 3.5 km of roadconstruction and a little over 100 ha of treeplanting. The jobs and carbon sequestrationgoals are met, but there is an expenditureover-run of a little over $50,000.

f any goal were considered more importantthan another, this could be reflected byweights of different orders of magnitude inthe objective function. For example, if theexpenditure control was a critical goal, aweight of say 1000 could be entered in cellG8.


Table 9. Initial tableau of local government jobs and environment program

A B C D E F G H I J1 Goal programming example23 Constraint or activity

levelRoadconstn.(km)

Treeplanting(ha)

Houseconstn.(houses)

Jobshort-fall

Carbonseqn.shortfall

Expenditureover-run

Res.use

Sign Resourcesupply

4 Activity level 0 0 0 0 0 05 Jobs (person months) 200 3 20 1 0 _ 10006 Carbon sequestration

(tonnes) -2 5 0 1 0 _ 5007 Expenditure ($1000) 100 2 30 -1 0 ≤ 5008 Deviation 1 1 1 0

Table 10. Final tableau for local government jobs and environment program

A B C D E F G H I J1 Goal programming example23 Constraint or activity

levelRoadconstn.(km)

Treeplanting(ha)

Houseconstn.(houses)

Jobshort-fall

Carbonseqn.shortfall

Expenditureover-run

Res.use

Sign Resourcesupply

4 Activity level 3.48 101.39 0 0 0 50.705 Jobs (person months) 200 3 20 1 1000 _ 10006 Carbon sequestration

(tonnes) -2 5 0 1 500 _ 5007 Expenditure ($1000) 100 2 30 -1 500 ≤ 5008 Deviation 1 1 1 50.70


In this module, the highly versatiletechnique of linear programming has beenintroduced with reference to a simple profitmaximization problem involving twoproduction activities. This decision problemcan be expressed as an algebraic model,and solved graphically, providing insightsinto the nature of the resource allocationproblem. The simplex method, which isquite general in that it can solve linearprogramming problems containing anynumbers of activities and constraints, andis readily available in computer packages.Sensitivity analysis has been shown toprovide further information about solutionstability and opportunities for short-run

expansion of production.

Extensions to the linear programmingtechnique to handle more complexapplications include transportationmodeling, portfolio selection and goalprogramming. The technique has greatpotential for modeling decision situations inforest management and policy.

FURTHER READING

Dayananda, D., Irons, R., Harrison, S.,Herbohn, J. and Rowland, P. (in press),Capital Budgeting: Financial Appraisal ofInvestment Projects, Cambridge UniversityPress, Cambridge.


Dent, J.B., Harrison, S.R. and Woodford,K.B. (1986), Farm Planning with LinearProgramming: Concept and Practice,Butterworth, Sydney.

Pannell, D. (1997), Introduction to PracticalLinear Programming, Wiley, New York.

211

18. Team Building, Grant Seeking and ProjectAdministration

Steve Harrison and John Herbohn

The various research techniques introduced in earlier modules need to be viewed in thebroader context of how they will be supported and applied. Doing research is a little likerunning a business. No matter how good the ideas are and how technically proficient onemight be at doing the job, the success of the business depends largely on the structures thatare put into place to get the job done. To draw an analogy, doing research in isolation is thebusiness equivalent of running a small manufacturing business as a sole trader. To expandthe business beyond this one needs to work with or employ other people (i.e. build or join ateam) and capital is needed to do this (i.e. research grants), and this capital needs to beused efficiently and effectively and in accordance with the guidelines of those people (c.f.banks) that provided it (i.e. project management). The researcher then needs to produce aproduct and sell this product into the market place to generate a return to the business (i.e.completing research and publishing it). This module discusses the issues of research teambuilding, obtaining funds and subsequent project management. In the following module,issues associated with documenting and disseminating the research findings are discussed.

1. WORKING IN RESEARCH TEAMS

Nowadays, it is increasingly difficult for anindividual working largely on their own tocarry out high level, high impact originalresearch. Much of the successful researcheffort arises from highly focussed andcohesive research teams. Hence it isappropriate to consider how teams areformed and maintained, and the strengthsand weaknesses of team research.

Team building and maintenance

Research teams are often formally broughttogether for particular tasks, e.g. publicinquiries. The interest in this module is withmore informal teams, driven by mutualinterest. Teams sometimes emergespontaneously, where common interestsbecome apparent. Research fundingarrangements also favour team building.For example, the cooperative researchcentre (CRC) funding arrangements of theFederal government in Australia have beenresponsible for providing long-term fundingto bring together teams of up to about 200researchers, in different institutions anddifferent states, to focus research in relatedthemes. An example is the Rainforest CRCin Queensland, which has programs onrainforest ecology, entomology, threatenedand threatening species, water quality,

tourism, reforestation and other areas. TheACIAR model is also a valuable contributionto team building. Large grants whichcontinue for a number of years, and providefunding for travel of researchers tomeetings and field sites, and for researchofficers, are critical for teamwork. Successin obtaining large grants is critical tosuccess in team formation and continuation.

In that the research grant scene is difficultto break into, and favours researchers withan established track record, it is often anadvisable strategy for more junior facultymembers to team up with experiencedresearchers.

While a university research team willnormally be coordinated by establishedacademics, research officers or researchassistants1 and postgraduate students(which can be overlapping roles) play animportant role in these teams. They canbring a large amount of energy, initiativeand dedication, and in practical terms are acost-effective way of adding value toresearch funding. The involvement of high-calibre postgraduate students can make amajor difference to research output. Oftenpotential postgraduates have litt le

1 Higher education workers or HEWs in the

Australian University vernacular.


experience in designing research projectsand are looking for guidance on what typesof projects are feasible. One strategy forobtaining postgraduate students to work inareas of interest in our Rainforest CRCprograms has been to develop a portfolio orpotential research projects, and to providethis to people who make inquiries aboutpostgraduate study. Joining an establishedteam also provides advantages for a post-graduate student. Importantly it gives themaccess to team members with a range ofskills and backgrounds. If reseachers haveinteresting projects available, and somefunds available to support the research,then this can be attractive to postgraduatestudents looking for a suitable thesisproject. If the postgraduate students have ascholarship, and some top-up funding canbe provided in the form of researchassistanct pay, this further improves thearrangement and hence the ability to attractcapable people.

Whether teams continue to exist dependscritically on effort put in by individualmembers, continuation of a mutual researchinterest, and success in obtaining grantsand securing publications.

Advantages of working in teams

Working in a research team can have anumber of advantages:

Specialisation. A team benefits from havingmembers with specialist skills, e.g. teammembers can specialise in technical areas(such as silviculture), in developingresearch ideas, in grant seeking and budgetpreparation, survey methods and statisticalanalysis, computing skills, text generationand editing, report writing ability, promotionof research activities and findings and soon. It is unlikely that a single individual willbe sufficiently multi-talented to cover allrelevant areas at a high level ofcompetency.

Critical mass. Some advantages exist fromhaving a number of people available to takepart in research activities. This allowsprogress to continue when other workpressures create unavoidable diversions forsome of the team. The variety of skillsbecomes apparent to funding bodies and toagencies seeking consultancy services.

There is a greater probability that someonein the team (as distinct from an individualworking alone) will be aware of highlyrelevant literature and of experts whoshould be contacted. There is a greaterprobability that calls for grant applicationswill be detected and that proposals can beput together to meet deadlines. A criticalmass of people can also lead to a criticalmass of funding, whereby one or more full-time research officer can be employed, andthis can have major benefits in terms ofresearch continuity, record keeping,communication with client groups, andoutreach (extension) activities.

Synergies. Perhaps the greatest advantageof research teams is the synergies whichresult between members. Members can‘bounce research ideas’ off each other,comment on each other’s work and writings,challenge unsound thinking (an importantvalidation of research), and provoke deeperthinking on a topic. These synergies canlead to more thorough analysis and writing-up, and to identification of further promisingareas of research.

Strategic alliances. Some members of ateam may be members of otherprofessional organisations and researchunits, expanding the range of researchcontacts. Some may be members of journaleditorial boards, or have good contacts withjournal editors, or be members ofdepartmental or external committees whichjudge research fund applications, and soon.

Grant success. The combined researchcapability and track record of a researchgroup tends to more impressive than that ofindividuals, which can be important forattracting interesting work and researchfunding.

Publication success. Working as a teamwhich includes researchers with a strongpublication ‘track record’ can lead to highersuccess rates in publication of journalarticles and book chapters.

Disadvantages of working in teams

There are some disadvantages in workingin teams. Teams are made up of individualswith a range of personal characteristics and

Team Building, Grant Seeking and Project 213

idiosyncrasies. Successful teams that workwell together are invariably composed ofpeople who get on well together and enjoyworking collaboratively. However, even inthe most successful of teams there will bepersonal conflict and tension at times. Tomake teams work, group members must bewilling to compromise and be able to put inplace mechanisms to deal with negativecharacteristics of other team members, i.e.be tolerant. If this is not done, many of thebenefits of teams can be lost and muchtime and energy is wasted in conflictsituations.

Mentoring of junior team members

In research teams, there is a requirement ofmore established researchers to provideencouragement, support and training formore junior members. This can be viewedas in part a social responsibility inacademia. However, it is also importantfrom a self-interest viewpoint, of havingcapable and productive colleagues with afavourable attitude to project work who arelikely to remain with a project rather thanseek alternative work. Some practicalmentoring steps include:

• making research resourcesavailable wherever possible, e.g.funds for fieldwork.

• being reasonably accessible todiscuss research activities.

• providing a sympathetic listener andcounselor when personal problemsarise.

• committing team members tos e m i n a r o r c o n f e r e n c epresentations.

• providing referee support and reviewcomments on grant (andscholarship) applications.

• including team members in jointapplications for research funding.

• directing members to usefulcontacts or relevant reading.

• providing comments on draft papers– within days not weeks – and

making suggestions for publicationoutlets

• including members as authors inresearch papers. More establishedresearchers often have betteraccess to publication outlets.Developing confidence to prepareand submit papers can take sometime and training.

• providing positive reinforcementwherever possible, e.g. making apoint of commenting on successfulachievements, such as paperspublished. (In some cases, it will bepossible to provide financialincentives.)

• should the opportunity arise,providing funding support for keyevents, such as an internationalconference.

• providing positive confidentialreferee reports when these areneeded.

Where younger project workers arepostgraduate students, the supervision andmentoring roles tend to overlap.Supervisors and mentors can leave alasting impression on those that they workwith. However, there is a wide range ofperformance by higher degree supervisors,from totally inert to totally ‘switched on’. Inteam situations it is thus critical to ensurethat strong supervision and ready help isprovided to students. Postgraduates shareinformation readily with other existing andpotential postgraduates. If soundsupervision is provided, then the word ofmouth advertising by existing and paststudents can be a highly effective means ofencouraging new postgraduates to join theteam. The opposite also applies.

2. STRATEGIES FOR OBTAININGRESEARCH GRANTS

Adequate funding is a critical ingredient toachieving research objectives. Even withsufficient time, skills and commitment, it isdifficult to carry out sound research withoutsufficient money for fieldwork travel andequipment, research assistance, computerequipment, postage and printing, purchase


of research books, and so on. Attendingconferences can be critical for keepingabreast with what is happening in aresearch field, and conferences areinvariably expensive to attend. While one’semployment agency generally has someaccessible research funds, the extent ofpaper work for a modest grant can bediscouraging, and there may be limitedflexibility in how the funds can be used.These are best thought of as ‘seed funds’,to assist in developing research conceptsand proposals. All this leads to theconclusion that obtaining external researchgrants is critical to a healthy researchprogram.

Grant seeking involves identifyingopportunities for funding, developingresearch proposals, and frequently makingmodifications or preparing rejoinders inresponse to feedback from the funding bodyor appraisers of the application.In some cases, preparing researchproposals can take several weeks of work –for literature search, drafting the proposal,collecting cost data and preparing thebudget, and so on - possibly as much timeas it takes to write a research paper. Hencegrant seeking needs to be approachedstrategically, and not entered into light-heartedly. It is wize to make an estimate ofthe probability of success before launchinginto development of a research proposal,but bearing in mind that a proposal rejectedby one agency may be reworked foranother. Given the amount of effort that cango into developing a convincing researchproposal, it has been commented that onehas to do the research before obtaining thegrant. Following on from this (tongue incheek) view, when a grant is received, it isused to carry out the research needed toapply for the next grant!

Choosing the research area

It is obviously preferable to choose aresearch area with which the team hasexpertise and is comfortable, to takeadvantage of their comparative advantage.Sometimes, a tradeoff arises betweenchoosing areas of greatest interest andareas for which funding is most readilyobtainable. Sometimes, involvement inprojects that are well funded but not quite inthe area of interest of the researchers can

be advantageous; a researcher or teamwith small amounts of funding has limitedflexibility in what projects then can choose.Many grants have some degree of flexibilityin what can be done with the funds oncethey are awarded, which may allowopportunity to explore issues of particularinterest. Sometimes, modifications toproject objectives will be agreed upon,depending on what lines of research turnout to be feasible.

Identifying funding sources andprospects for success

A surprisingly large number of potentialresearch funding sources exist – includinglocal, national and international bodies –and new ones continue to arise. Thissuggests that it is necessary to be vigilantabout identifying what grant are potentiallyaccessible, and to establish a database ofgrant sources and application deadlines.Most funding bodies have one or two calls ayear, with specific submission dates.

A key strategy is to get to know the morepromising agencies which are sources inrelation to one’s research interests, and tounderstand their priority funding areas.(This is sometimes disclosed on a Website.) It is also helpful to know otheridiosyncrasies of the funding body, e.g. thetype of emphasis they like in projectproposals, and what they like and dislike.Some funding agencies are particularlykeen to see any related researchhighlighted in the proposal; some like inter-agency grant arrangements; some placegreat emphasis on having a technologytransfer component in the research; somehave a highly applied focus (e.g. ‘gettingtrees in the ground’). Some do not like tosee involvement of postgraduate studentsin the proposed research activities, takingthe view that such a proposal is a disguisedform of application for a studentscholarship.

Choosing which grants to apply for is aneconomic problem, of cost of developingthe proposal versus expected payoff inresearch dollars. It is worth bearing in mindthat large grants are not necessarily moredifficult to obtain than small ones, and thatthe amount of effort involved in obtainingsmall grants is sometimes a questionable


use of time.

It is helpful to know what sizes of grants aresupported. Sometimes successfulapplicants will be given the full amount theyrequest, and sometimes amounts will bescaled down, which creates problems whena specific set of objectives and activities inproposed. For some of the more prestigiousgrants, the probability of success isdiscouragingly low. For example, only about20% of applications for Australian ResearchCouncil grants are successful.

Being known to the agency can be a criticalfactor in grant success. Some agencieshave a rather narrow researcher base, andlike to continue funding people who have arecord of achieving their researchobjectives. Successful researchers oftenreceive invitations to apply for grants.

It would appear that funding agencies arehighly risk averse with respect to whomthey grant funds. In one case, a fundingagency commented that if they received aparticularly interesting application from anunknown researcher, they would invitesomeone well known by the agency to carryout the proposed research! This reinforcesthe view that it is difficult to break into theresearch funding circle, and that success inobtaining grants favours further success.One way to break into this circle is toprepare joint applications with moreestablished researchers. The fact that acredible research team is being put up inthe proposal is also important in having theagency look favourably on the request.

The probability of grant success can beincreased greatly by including establishedresearchers in the application. Usually,grant success is related (directly orindirectly) to track record. For example, theAustralian Research Council (ARC) councilrecently placed 50% weight in scoringapplications on the proposal and 50% onthe track record of the applicants.

Preparing the proposal

Funding applications tend to take arelatively standard layout. Some of thethings which are often included are:background and literature review, researchquestions, aims and objectives, research

method, significance of the research,technical outputs and practical outcomes,technology transfer method, timetable,Gantt chart indicating predecessor activitiesand milestones, budget, justification forbudget items, references, ethical clearancestatement, statement of employer’s support,and applicants’ track records.

The typical weak spots in applications are:lack of a theoretical or conceptual modelunderlying the proposed research; theproposed research method is too vagueand lacking in detail; and the justification forbudget is weak (e.g. it says how the moneywill be spent rather than why it is required).Once a proposal has been prepared, it isuseful to review the text in these areas ofoften-encountered weaknesses.

Some expenditure areas tend to be wellaccepted while other raise suspicions of theappraisers. The more acceptableexpenditure areas: research assistance,fieldwork equipment and travel, interviewsurveys, purchase of research books,maintenance (phone, fax, photocopying,postage), and project workshops. Agenciesfunding international projects are generallysupportive of project workshops (planning,training, end-of-project). Requests forfunding of conferences (particularlyoverseas ones) and notebook computersare likely to be viewed critically. Sometimesthe travel budget is considered excessive.Some agencies give applicant theopportunity to place priorities on the variousbudget items, as A, B, etc. In such cases, itis wize to place the lower priorities on theitems which are generally less favourablyviewed. Where expenditure items have tobe prioritised, amounts granted may be lessthan the total amount applied for.

A useful recent development is that manyfunding bodies have a staged applicationprocedure. An initial expression of researchinterest of not more than about two pages inlength may be called. It the proposed workis judged competent and fits within thepriority areas of the funding bodies, then theapplicant is invited to submit an expandedapplication. In some cases, there may bethree or four stages to the application, thedownside of which is that it may take abouttwo years for the project to finally obtainapproval.


It often happens that grant applications are‘recycled’. That is, if an application isrejected by one funding body, it is reworkedand submitted in response to the call forapplications by another agency. This meansthat a further payback can be obtained fromthe sunk costs of development of anunsuccessful research proposal, and helpsjustify the large effort expended in chasingprestigious but low probability grants. In thecase of large grants (e.g. ACIAR grants)several weeks may be required on theproposal and even the budget may take anumber of weeks to prepare. This should beconsidered as part of the research process,not as dead time. That is, the projectproposal should be considered as a workingdocument, which is part of the researchprocess, and which can if necessary berecycled for another funding opportunity, orform the basis of a publication. Persistencein grant seeking often pays off.

A mistake which is sometimes made inpreparing the project budget is to ask fortoo little funds, in the hope that this willincrease the probability of securing a grant.This means that if the application issuccessful, the researcher is stuck withtrying to achieve promised outputs withinadequate resources. We were caught inthis way in a recent study of tribal villages inIndia, but fortunately managed to obtainsupplementary funding from anothersource.

It is a valuable step to obtain a peer reviewof a grant proposal. Some organisationsarrange this as a matter of course.Someone who has not been involved in thepreparation of the proposal is more likely todetect omissions, unclear statements andstatements which could be misunderstoodor cause a negative reaction. It is notuncommon for people promise too much,i.e. to propose to undertake more than theycould reasonably achieve with the fundsand time available.

Ethical clearance is increasingly beingrequired for grant applications, even toundertake surveys (e.g. of households orfarmers).

Responding to requests for more

information or modifications

Funding bodies vary in their decisionprocesses. In some cases, a detailedproposal is prepared, and an accept/rejectdecision is made, with no interaction withthe applicant. At the other extreme, anofficer from the funding body interactsrepeatedly with the applicant to steer theproject document and budget through thevarious approval stages. ACIAR is excellentin providing support for applicants in thisregard. In most cases, some amount ofquestioning of the proposal arise, withopportunity for revisions or clarifications.When proposals are sent to externalreviewers, there may be an opportunity forthe applicants to prepare a rejoinder to theappraisers’ comments.

In the case of projects involving researchersfrom more than one agency, of from twocountries, there may be negotiations on theresearch scope to satisfy the needs of eachagency or country.

Generally, when a project and grant isapproved in principle, it is necessary toobtain statements of agreement from thevarious parties. For example, the office ofresearch of a university normally has to signoff for a project which involves universitystaff and resources. Concerns can ariseabout ethical considerations (clearancefrom an ethics committee may be required)and any restrictions which are imposed bythe funding body on publication of researchfindings.

3. RESEARCH ADMINISTRATION

In any research project, there is aconsiderable amount of administration andreporting (in addition disseminatinginformation about the research and itsachievements). These tasks involveplanning of research act iv i t ies,management of people and funds,monitoring project progress, and reportingto officers of the host organisation andfunding bodies.

Planning research activities

Poor planning can result in both wastedfunds and wasted time – both of which areimportant commodities for a researcher.


The need to plan data collection andanalysis activities has been discussedelsewhere. There is also a similar need toplan the expenditure of funds. Where travelis involved, particularly international travel,a great deal of time can be spent in makingtravel arrangements. If field visits andresearch trips are not planned well they caninvolve a considerable waste of time. Forinstance, a visit to a field site to talk tocommunity members can be a completewaste of time if the key people being soughtare away.

Personnel management

In general, team research is a cooperativeand collegial activity, rather than amanagement-and-line-worker process. Thisis particularly the case when the teammembers have confidence in each other’sabilities and dedication. Team managementthen involves taking a close interest incolleagues’ activities, and providingencouragement and support wherepossible, rather than attempting to makedirectives. Research tends to be mostproductive when there is an open andinquiring atmosphere and people have theopportunity to explore ideas. Coordinationthen takes place through planning meetingsand informal discussion.

Managing the budget

This can be one of the most difficult areasin management of research projects. In onesense, the funds for a project are fullyallocated at the time of funding approval,hence the task is to ensure that the fundsare used in the way contracted between theresearchers and the funding body.Fortunately, more flexibility than this usuallyexists, and modifications to projectobjectives as well as unforeseencircumstances alter the way in which fundsare used.

Where requests for supplementary fundingby team participants are made, it isdesirable to meet these to the extentpossible given the project commitments andfinance constraint.

Making claims for research expenses, andacquittal for funds advanced or disbursedcan be an onerous task for researchers, but

is a necessary part of public accountability.

Project reporting

Invariably, there is a requirement for anend-of-project report to the funding body.For projects with a duration of more thanone year, there is usually annual reportingof expenditure and of performance inrelation to milestones. When field trips aremade, there may be a requirement forindividual trip reports. (This is a requirementof ACIAR, for example.) If projectworkshops are conducted, there will be anexpectation of some form of workshopproceedings or report. There may be anexpectation of production of a technicalmonograph at the end of the project. Ingeneral, the funding body will want toensure that the money is well spent, andproduces useful findings or ‘makes adifference’ to a target group ofstakeholders, and provides favourablepublicity for the activities of the fundingbody.

Project monitoring and staying on trackin terms of budget, outputs andtimetable

Experience indicates that there is often‘slippage’ in projects relative to researchintentions. While some delays andexpenditure over-runs can be tolerated, theproject objectives and milestones need tobe kept in sight. There will be adverseimpacts on researchers if a project does notachieve its objectives, such as: questionsasked by funding bodies about annualreports; imposition of a project review by anindependent expert group; early terminationof a project; more difficulty in obtainingfurther funding.

Dealing with project reviews

Successful performance in reviews isimportant for securing further funding.Reviews sometimes arise because thefunding body is not happy about the way aproject is progressing, so need to behandled with care. There is a need toprovide details of all achievements of theproject (e.g. research technical outputs andpublications, technology transfer, capacitybuilding) of difficulties encountered, and ofmodifications of objectives.



There are no guaranteed strategies forresearch success, but experience indicatesa number of lessons on what measures arelikely to assist. Success is a matter of bothopportunities and effort. It has been saidthat ‘genius is 10% inspiration and 90%perspiration’. Success in research requirescommitment and persistence. In that muchinteresting research findings and ideas fail toappear in print, if one critical tip can begiven, it is to become a habitual writer.

5. DISCUSSION QUESTIONS

1. What are the promising sources ofresearch funds for your organisation?

2. Which journals would you target forresearch outputs?

3. To what extent are you involved in

mentoring colleagues, and through whatmeasures?

4. Do you undertake research in ateamwork situation? If so, what specialskills do individuals in your teampossess?

5. Consider the following text passage.Suppose you are given the task ofproviding editorial comments to theauthor. What aspects of the passagewould you suggest be reworded?

ESTABLISHMENT

‘Brown (1988) notes that plantationestablishment was not good; there weresignificant seedling losses, due to thehot temperatures.

Therefore, although the seedlings werevery cheap, the plantation was notexpected to be profitable.’

219

19. Report Writing and Publication Strategy

Steve Harrison and John Herbohn

In an academic world, it is critical to develop a research ‘track record’, a key indicator ofwhich is the publications list. This can influence job satisfaction, promotion opportunities, andsuccess in obtaining grants and consultancies. Publications are probably the single mostimportant means by which researchers in universities, colleges and research institutions areevaluated. Publishing research results sends a signal to potential research funders andemployers that the researcher is capable of bringing a research project to a conclusion andcan produce a tangible outcome. Peer review provides a critical validation of the researchmethods and findings. Publications are also a major element in the transfer of technology –such as new processes and new understanding of system behaviour and managementtechniques – to potential users. Particularly when grant funding has been obtained,publication is expected, and is the requirement for further funding. Writing papers is a criticaltask for researchers. Many people do excellent conceptualisation, literature reviews,statistical analysis and other research activities, but fail to document their findings in a non-perishable and widely available form. Some people are ‘writerholics’ who write throughcompulsion; others are strongly disinclined to ‘put pen to paper’ and will always findsomething else pressing to do rather than write up their research. Writing is to some extent amatter of habit. However, some tips can be given to improve writing skills. While success inresearch is to a large extent a matter of commitment and perseverance, a number ofstrategies may be employed to increase effectiveness. Different strategies work best fordifferent people, but a number of observations may be made as to how to be more effectivein this endeavor. This module examines various aspects writing up and publishing researchfindings. The observations made here represent to some extent the personal views andexperiences of the authors, and are designed to provoke interest and discussion on researchpublication strategies.

1. WRITING GENRE

The researcher must bear in mind the styleof writing which is most appropriate for theparticular type or output. For example,when writing a thesis, there is little pressureon space, but a strong imperative todemonstrate that all relevant aspects havebeen considered and all assumptions madeclear (a defensive form of writing). There isalso need for a considerable amount ofmot i va ta ion (saying why material ispresented) and signposting (saying howmaterial is set out). Journal articles on theother hand are expected to be highlyorganised and concise; journals areexpensive to produce and have tight spacelimitations. Differences exist betweendifferent types of journals; science journalstend to follow the ‘aims – materials –method – results’ format, while journals inthe social sciences and humanities are alittle more expansive. Working papers andconference papers tend to be written morequickly with less finishing-off effort than

journal articles. When writing textbooks orother teaching materials, it is generallynecessary to provide extensive explanationto make sure that readers can understandthe concepts being presented, and toreinforce this with a large number ofexamples and perhaps case studies.Material which is intended for amultidisciplinary audience must containrelatively simple explanations. In the caseof extension material, this generally has tobe written in a more popular and catchystyle.

2. CRAFTING A RESEARCH PAPER

A research paper – e.g. journal article,conference paper, discussion paper – tendsto have a relatively standard structure. Thisvaries somewhat between journals, anddisciplines, so the following suggestionswon’t find precisely for every case, but arebased on experience for a variety ofpublication outlets.


Typical sections of a research paper are: anabstract (of about 150 words, sometimesfollowed by keywords and journalclassification codes1), acknowledgments(usually either a footnote to the first page ora section before the references),introduction, research method, a number ofsections forming the body of the paper,discussion, concluding comments,references, and perhaps one or moreappendix.

The Abstract should be a brief overview ofwhat the paper is about, a note on themethodololgy, and a concise summary ofthe main findings, such that a potentialreader can decide whether to read thepaper in detail. The Introduction shouldindicate what the paper is about in generalterms, perhaps review some relevantliterature, perhaps mention areas ofcontroversy of important information gaps,indicate in general terms what methodologyhas been adopted, and state how the paperis set out (the sequence of sections). Theamount of detail will vary between shortcommunications and longer papers. Thesetwo sections (Abstract and Introduction) areoften badly crafted in draft papers.

Literature review material may appear in aseparate section, but often this is containedboth in the Introduction and throughoutother sections of the paper. It is important tomake clear the Methodology which hasbeen adopted, often as a separate section.If the paper reports survey findings, thenthe study area, sampling frame, surveymethod, response rate, and method of dataanalysis would normally be indicated. In areview paper, there may be no need for aMethodology section.

It is difficult to lay down guidelines as towhat structure the body of the paper willtake. If results are being presented – e.g.from experiments, modelling or surveys –then a critical reviewer would normally lookto see if there was any statistical analysis(such as t-tests, regression analysis orANOVA). In qualitative research or reviewpapers, it may not be possible to providethis type of analysis. The main results of theresearch are normally presented in the

1 For example, in economics the JEL

Classification Codes are often included.

body of the paper, sometimes as a separatesection.

The Discussion section provides anopportunity to expand on research results,and any limitations of the methodology andfindings, and to highlight the mainimplications for management or policy.

The C o n c l u s i o n s section tends tosummarise and remind the reader of themain impressions they should take awayfrom reading the paper. This section is oftenbrief, and the heading Conclusion (singular)or Concluding comments is sometimesused, to signify the section concludes thepaper, and does not necessarily presentdetailed conclusions from the research. Aguiding principle for this section is ‘say whatyou found, not what you did’ – don’t repeatcomments on the methodology or paperlayout.

Appendices are useful when the authorwants to include information which is moretechnical and could be useful for aspecialist reader, or which is necessary tosupport the main arguments, but wouldinterrupt the flow of text if included in thebody of the paper. This keeps the paperconcise, but provides an opportunity toinclude the additional information.

3. HANDLING REFERENCES

The handling of references can presentmany difficulties, so is worth consideringseparately here. In general, references intext should indicate author’s name and yearonly; full references should be compiled atthe end of the paper (not provided asfootnotes).

There are many formats adopted byjournals – an almost unlimited combinationof sequences and syntax – and the writerhas little choice but to follow house styles. Acritical requirement is to obtain allnecessary information when initiallycapturing references; this may include bothvolume and issue for journals, and city (notcountry) of publication for books andreports. It is extremely painful and timewasting to have to chase up the originalsources because incomplete informationwas recorded initially. A recent personalexperience of this was when a journal

Report Writing and Publication Strategy 221

insisted on stating the number of pages inbooks and reports referenced.

In the case of personal communications, itis desirable to provide information about theinformants claim to expertise – designation,employer, location – in the referencessection. Anagrams are often preferable fortext references, especially for agencies withlong convoluted names – provided fulldetails are provided in the referencesection.2

The use of indirect references – of the form‘Brown (1988, cited in Bloggs, 2001)’ – is tobe avoided if at all possible, since itindicates a lack of effort to chase down theoriginal source.

While ‘Anon’ was commonly used in thepast for unknown authors, this tends not tobe acceptable now. Sometimes the name ofthe agency is cited, if there is no informationabout the particular persons who did thewriting.

Increasingly, material from the Web is citedin research papers. Reviewers of submittedpapers tend to be rather critical of the Webas an information source, since much of thematerial posted has undergone little or novalidation (peer review). It is desirable toindicate the full http address, the agency towhich it applies where possible, as well asthe date when the site was accessed.

For convenience, references may be kept ina reference database, such as Endnote.This allows them to be convenientlygrouped under topics, and extracted in avariety of formats, to suit the style of thepublication medium.3

4. FURTHER WRITING HINTS

2 In Australia, the Department of Employment,

Education, Training and Youth Affairs – whichwas responsible for the points system forgrading university publications – was muchmore simply referred to in text as DEETYA.Like many public sector agencies, the namehas been changed at least once since thismouthful.

3 This can also save time chasing up a refer-ence from another document on the computerhard drive, e.g. using Windows Explorer –Tools – Search – Containing text.

While choice of words is a personal thing,some guidelines can be laid down on whatpresentation is likely to be well received byeditors and reviewers. While some authorslike to use extravagant language, in generala guiding principle is to keep the text assimple, straightforward and uncluttered aspossible, both visually and in terms of themessage conveyed.

There are a number of words commonlyused in conversation which are to beavoided or minimized in written material.These include:

* vague terms, e.g. ‘very’ (‘highly’ may be abetter choice), ‘fairly’, ‘quite’, ‘different’ (asan adjective – ‘various’ is often a betterchoice).

* terms that have well accepted meaningswhich differ from the context in which theyare being used, and are sometimes usedexcessively in written material, e.g. ‘good’(means a product, or holy), ‘certain’ (canoften be replaced by ‘particular’ or ‘specific’)‘significant’ (a term which has a specificmeaning in statistics relating to probabilitylevels in hypothesis tests – ‘considerable’ or‘major’ may be better choices).

* terms which are misused in everydaylanguage and can be irritating to the reader,e.g. ‘cheap’ (not an economic term, andoften taken to mean ‘of low quality’),‘quality’ (used far too often and without anyqualification of ‘high’ or ‘low’ quality)

* uninformative terms, such as ‘results’ (it isusually possible to indicate what type ofresults are being provided)

* personal pronouns, such as ‘we’ and ‘you’.These can be used to effect for a reader-friendly ambience in a textbook, but in ajournal article are rather too informal.

* inappropriate use of adjectives, such as inthe expression ‘cold temperature’, ‘cheapprices’ and ‘early age’. Here the noun is anumber, not a characteristic, e.g. airtemperature is a number between –70 and+50 approximately, in celsius), but theadjective is designed to describe acharacteristic. In each case here, ‘low’would be a better adjective.


A construction to be avoided is tocommence a new paragraph with anexpression designed to link sentences inthe same paragraph, such as ‘Therefore’,‘Hence’, ‘As a result’ and ‘However’.

The wording of section headings should besufficient to convey to the reader what thesection covers, i.e. should be relatively‘stand-alone’ or convey context. It is betterto have long headings (even running on to asecond line) than overly brief (includingone-word) headings. It is undesirable tohave two headings running; normally, thereshould be some substantive or at leastsignposting text between headings.

References are normally cited in past tense,since the author may no longer have thesame view, or may not even be alive. Itwould not make much sense to say‘Faustmann (1844) argues that the optimalrotation is …’.4

Graphs and tables require particularattention. It is important to lable the axes orcolumns with the name of the variable (notjust the units, e.g. ‘(%)’. All columns of atable, including the first, should have acaption. A table must have a title, and besituated in document at a point after which ithas been mentioned in the text. (Somejournals require tables and figures to begrouped at the end of a paper when it issubmitted.) A table should have a minimumnumber of horizontal lines – usually at thetop and bottom of the row containing thecolumn captions and at the bottom of thelast row of the table. Vertical lines are to beavoided if at all possible. Following theseguidelines will result in an unclutteredappearance.

A number of style guides are available,which provide advice on both writing styleand syntax, e.g. the Cambridge guide andthe AusInfo guide. While these arerecommended reference works, it must beborne in mind that each journal has its ownin-house style. Also, the style of writingtends to vary between disciplines. Oneinteresting example is that in the sciences itis a guideline to include descriptive materialabout tables in the title, whereas in the

4 Here ‘argues’ should be changed to past

tense, but ‘is’ should remain in present tense.

social sciences the principle is to keep tabletitles as concise as possible and confine allexplanatory material to the text or a tablefootnote.

5. FORMS OF PUBLICATIONS

There are various forms of papers, and tosome these forms can be stages in theproduction of a journal article.

Conference papers

This is often the first form of publication ofresearch or review findings. A commitmentto a conference presentation provides adeadline which can increase a researcher’sfocus – one needs to prepare somethinguseful to say. A conference is a made-to-order feedback mechanism, though oftenthe most useful comments will be obtainedoutside the formal sessions. Audiencemembers may draw attention to otherrelated research. Often the proceedings willbe published, in which case these areclaimable publications. Sometimesconferences result in an invitation to submita paper to a journal.

In-house discussion papers andmonographs

These are sometimes used as a first run, oras a more durable form for an unpublishedconference paper, to establish claim in thefield, publicise research and obtainfeedback. These have higher status ifsubjected to peer review process.

Journal articles

In general, the greatest publication credit isobtained for having papers published inrefereed journals. This can be a timeconsuming and exacting task. Choice of anappropriate journal is critical. Ideally,articles will be submitted to ‘double blindreview’ by two or more reviewers, i.e. wherethe authors does not know who thereviewers are, and the reviewers do notknow who the authors are. Sometimessingle blind review, where the author’snames are disclosed to the reviewers, isappropriate; the reviewer’s then know‘where the authors are coming from’ withtheir views and can take this into account.Sometimes papers rejected in one journal


will be accepted by another.

Book chapters

These are another well respected form ofpublication, though often ranked not quiteas highly as journal articles. The reviewprocess for books typically is not asexacting as for journals.

Authored and edited books

Authored books tend to be regarded morehighly than collections of chaptersindividually authored by contributors. Also,books with a mainly research focus tend tobe regarded more highly than textbooks.Notably, the weight given to books dependsto some extent on the idiosyncrasies of theappraiser. For instance some people saythat nothing new ever appears in textbooks, and that the only worthwhilepublications are those presenting originalresearch in journal articles. Others regardbooks with publishers of high standing asimportant publications.

Consultancy reports

To some extent, taking on consultancies iscompetitive with undertaking researchprojects. In an academic environment,consultancy reports typically are notconsidered as ‘publications’ in terms of thenarrow criteria which bring academicrecognition, funding and promotion.However, taking part in consultancyprojects can be invaluable experience, canprovide an important service to industry orcommunity, can lead to development of newprofessional contacts, and can lead tosubsequent production of papers forpublication. Hence it appears desirable toundertake some consultancy work, but notat such a high level as to prevent othermore traditional university research.

Extension papers

These can perform a useful technologytransfer and service function. Unfortunately,in an academic environment, they are givenlittle if any credit.

6. GETTING ITEMS PUBLISHED

Even when a researcher writes highly

original and interesting papers, publicationis by no means assured, and there is agood deal of strategy involved in havingpapers accepted for publication. Choice ofjournal is important, since research qualityis assessed by quality of the journal inwhich it is published and it is thereforenecessary to aim for as journal of as highstanding as possible. Importance of high-standing, high impact, high citation ratejournals. This does involve to some extentmaintaining a clear disciplinary focus inone’s research program. While it is criticalfor junior faculty staff to build up the lengthof their publication record, the importance ofpublishing in high-standing, high impact,high citation rate journals cannot beoveremphasised. At the same time, thejournal should be clearly within theresearcher’s discipline area. In establishinga research reputation, it is necessary to beseen as a specialist in a particular area, andnot a generalist publishing over a widespectrum of outlets – ‘the jack of all tradesis a master of none’! Further comments onchoice of journal are provided in latersections of this module.

Sometimes it is difficult to know if a paperone has written is publishable. In thissituation, it is helpful if an opinion can beobtained from an experienced publisher,such as a departmental mentor. Often thewrite will not recognize the publicationpotential of their work.

Sometimes a publication outlet can befound for concept papers. This means forexample that a project proposal can beworked up into a publication. Obviously, alot of research often goes into a researchapplication, so this is not an unreasonableoutcome.

There can be advantages in jointauthorship. To the extent that length ofpublication list is important, it can bebeneficial to have a large number of jointlyauthored papers, rather than a smallnumber of sole-authored papers. Also, thecombined efforts of two or moreresearchers can result in stronger papersand hence acceptance in better journals.

When seeking to have books published, it isusually necessary to prepare a bookproposal for a publishing house. Most


publishers can provide a set of instructionson the kinds of information they require tomake a judgment about accepting a book.This information will include details on theauthors, competing books, special features,amount if international material, thetargeted market and potential sales,expected length, extent of graphics,potential reviewers, possibil i ty fortranslation into other languages, andvarious other things. It is usually necessaryto submit a few sample chapters, thoughsome publishers will want to see a draft ofthe overall manuscript before making acommitment. The information sought by thepublishers is usually designed to give theman indication of the market size andproduction cost. In some cases, thepublisher will have a preference for beingprovided with camera-ready copy; thisreduces their production costs, henceallowing smaller print runs, and greatlyspeeds up the publication process.

The manuscript will sometimes be sent forexpert review. Almost invariable, a copyeditor will make style corrections. Once thepage proofs of the text are finalised, theauthors will need to prepare an index. Thisusually involves preparing a list of items,sub-items and page numbers, most easilydone on a computer and using analphabetical sort. Modern wordprocessorpackages have the capability to form anindex, through selecting appropriate wordsin the document files, though it is not clearthat this actually saves time in preparationof an index.

Nowadays, Powerpoint slides are widelyused for conference, symposium andworkshop presentations. While these areoccasionally published, it should be kept inmind that Powerpoint slides have littleinformation content except in the presenceof the author providing a commentary onthem. If one has prepared a Powerpointdisplay, but not written a formal paper, itmay be possible to capture the essence ofthe presentation by use of a cassette (ormicrocassette) recorder.

It is interesting to observe habits of highlysuccessful researchers. One person with anotably impressive publication recorddevotes a block of time each day to writing(typically 5am to 10 am, before going to the

office), spends a good deal of timecontemplating and conceptualising, keepscollections of notes and ideas on variouspapers under development, does not use acomputer (but has research typing support),writes relatively short papers, produces alarge number of discussion papers whichare later developed into journal articles, andis able to bring papers to a publishablestate with remarkably few drafts.Unfortunately, some of these habits are notparticularly feasible for most of us, but thereare certainly lessons to be learned fromsuccessful researchers.

A useful principle to keep in mind is tobecome a habitual writer, e.g. type up fieldnotes and summaries on discussions, scriptlectures, write discussion papers, commit toconferences. Writing becomes easier, andconfidence is gained about writing ability,with practice.

7. CHOOSING THE RIGHT JOURNAL

Choosing the appropriate journal to publishan article in is one of the most importantdecisions that are made in the publicationprocess. Not all journals are equal in termsof quality and prestige. The next sectiondeals with two the key issues associatedwith selecting a journal, i.e. (1) matching thetype of article to the publication outlet, and(2) determining the quality of a journal.

Matching the article to the journal

Journals are selective in what they publish.They usually publish material on a particulardiscipline area such as forestry,environmental management, or nuclearphysics. It seems almost self-evident that itis important to select a journal thatpublishes material on the topic of yourpaper. The general discipline area of ajournal is usually evident from its title. Forinstance, it is obvious that the Journal ofForestry publishes papers in the generalarea of forestry. What may not be clear isthe specific focus of a journal. For instance,the discipline of forestry covers areas suchas silviculture, menstruation, foresteconomics, extension and ecology, andthere are many forestry journals in printwhich sever particular specializations in thediscipline. Some journals only publishpapers in quite specific specialty areas


within a discipline. Particular journals alsooften favour particular styles of articles.Some journals publish only highlyquantitative papers which involve large datasets obtained from rigorous experiments.Other journals publish material that isqualitative in nature or more policyorientated, while other journals publishmainly review material. It does not matterhow good a paper is, it is unlikely to beaccepted by a journal if it is not consistentwith the fields and style of material that thejournal publishes.

It is critical to spend some time determiningwhich journals are likely to be interested inthe article that you are preparing. It ispreferable to do this research at an earlystage of preparing the paper. An attemptshould be made to keep the style of thepaper as consistent as possible with thatused in the target journal. This reduces thelikelihood of the paper being rejected on thebasis of inconsistency with the style andnature of the journal.

The easiest way to identify whether ajournal is appropriate is to read past issuesand see if papers have been published inthe field of the your paper and which usesimilar methodologies. Usually you will referto a number of past studies in a manuscriptand a sound guide to suitable journals is tolook at what journals have been the sourcesof the key references of your paper. If stilluncertain, it is advisable to contact theeditor of a journal to see if they areinterested in publishing the type of paperyou are preparing.

As a general rule, try to publish in the bestjournal that will accept your work. Sayingthis, it is also important to be realistic aboutthe quality of the work that you produce.Some pieces of research that you producewill be better than others. Some articles willbe good enough to get into top internationaljournals while other pieces of work might bemore suited to regional journals. Comparingyour work with articles published in thejournal under consideration is probably thebest way to whether it is of suitable quality.Important things to keep in mind include thesize of your data set compared with otherstudies, the uniqueness of the work,thoroughness of analysis, the types ofstatistical techniques used, the level of

theory development and extent ofreferencing to recent relevant literature. Topjournals seldom publish research or reviewarticles that have methodological flaws orlack a sound theoretical framework.

Usually the Chief Editor of a journal acts asthe ‘gatekeeper’ and will not allowunsuitable papers to proceed to the reviewprocess, although this is not always thecase. Publication can be delayedsubstantially by submitting a paper to thewrong type of journal, particularly if it goesinto the review process before beingdeemed to be unsuitable.

It is almost inevitable that at some stage apaper one submits will be rejected. Do notfeel discouraged – this happens toeveryone – even the best of researchers.Aim to submit the article to the next journalon the list that you have identified. When apaper is rejected, the editor is likely toprovide some feedback on what wasconsidered unsatisfactory in it. The rejectionmay have been on the basis of the articlenot being consistent with fields and style ofthe journal. In other cases the paper mayhave gone to review and was rejected onthe basis of reviewers’ comments. If thiswas the case then the editor is likely toforward the reviewers’ comments as well assome comments of their own. Thecomments of the editor and reviewers areoften useful feedback and should beaddressed as far a possible beforeresubmitting the paper to another journal. Itis likely that the next reviewers will havesome similar views to the previousreviewers. Before submission, it isimportant to rework the paper following theparticular style requirements of the newtarget journal.

Quality vs quantity – which is moreimportant?

The point has been made elsewhere thatsome avenues for publication are betterthan others. Generally, journal articles areconsidered to be better than book chapters,which are in turn considered to be muchbetter that published conference anddiscussion papers. While this is true, it isalso true that some journals are muchbetter than others. It is also true thatsomething is better than nothing.


The question of quantity versus quality is animportant one and bears some discussion.Developing a record of publication isprobably the most important aspect ofdeveloping a track record. Often whenpeople are assessing a person from theirresume or curriculum vitae (CV) – forexample, for a job or in appraising a grantapplication – they will skim the initialinformation such as degrees and then flipstraight to the list of publications of theapplicant. It is thus critical to actstrategically when submitting material forpublication.

The relative importance of quality andquantity of publications differ depending onthe stage of career. Assessors for grantsand those on selection committees for jobslook for different things depending on thecareer stage of the applicant. For instance,an early career researcher or academicwould not be expected to have 10 or 15papers in leading journals. In reality, earlycareer people often have no publications intop journals and assessors and selectionpanels understand this. What these panelsoften look for is potential. In thesesituations, the adage ‘something is betterthan nothing’ is highly appropriate. Panelslook for ‘indicators’ or ‘flags’ that anapplicant has potential, and actualpublications (no matter what they are) arepreferred to none. An applicant who haspublished several conference papers andmaybe has a journal article in a third tierjournal will be preferred to one who has nopublications.

Getting journal articles into top journals canbe a frustratingly long process. The reviewprocess for top journals is particularlyrigorous and the research on which anarticle is based must be of high quality. Thetime between commencing research andhaving it published in a top journal can besix or seven years and sometimes longer.Such delays are less critical for establishedresearchers because they invariably have anumber of projects running simultaneouslyand which are at different stages ofcompletion. For an early career researcher,presentation of preliminary research resultsat conferences is an effective way ofobtaining publications.

In the case of early career researchers the

balance between quality and quantity ofpublications is strongly in favour of quantity.As stated before – something is better thannothing. This does not mean that quality isnot important – it simply reflects the factthat quality publications take time toproduce and have published. It is importantfor early career researchers to producesome publications in the intervening period.A reasonable target is to produce at leasttwo published papers a year, or more if theyare multi-authored.5

Quality and type of publications becomeincreasingly important with more seniorpositions, particularly in faculty positions. Inthe most senior positions, such asAssociate Professor and Professor, qualityis critical. Quantity is also important butmore as a necessary minimum amount.

In Australia there is some degree ofconflicting messages being sent in terms ofquality versus quantity of publications.Promotions committees place a strongemphasis on quality of publications,particularly at the more senior levels. Thisconflicts with the funding formula foruniversities, which draws no distinctionbetween the quali ty of refereedpublications. The situation is different in theUK where much more emphasis is placedon the impact of research and this is howuniversities are assessed. Academics areassessed on the impact of their top articles,not on volume of publication.

8. ASSESSING THE QUALITY OFJOURNALS

Since journals vary in standing, it is worthspending some time exploring journalquality in more detail. Publishing inparticular journals carries a substantialamount of prestige. How then does onegauge how ‘good’ a journal really is? Somesuggestions are now made on how todetermine the quality of journals, withspecific comment on key forestry journals.

5 The average number of published papers per

person in schools of the University ofQueensland is between one and two (basedon a count of 1/n papers when there are nauthors).


‘Quick and dirty’ guides

The simplest guide to what journals arebest to publish in is to look at the journalscited in the references of one’s paper. Inany field there are often a number of keyarticles that are regularly cited – thejournals in which these are published areusually the leading ones in the field.

A reasonable guide to the quality andimpact of a journal is to look at whopublishes it. Journals published byacademic publishers such as SpringerVerlag, Cambridge University Press, CABI,Blackwells, Urban and Fisher, and Elsevierare generally of a high standard and have ahigh wide readership and impact becauseof the distribution and marketing support ofthe publishing houses. Journals publishedand dist r ibuted by professionalorganizations and University departmentsare much more variable in quality andimpact.

Another way of gauging what journals arehighly regarded is to look at what otherpeople in your field consider to beimportant. For instance, an excellent guideto what forestry journals are highly regardedin the USA can be obtained by looking atthe profiles of US forestry faculty containedin the various web pages of leading forestrydepartments. The practice in the USA is forstaff to list only their high quality or mostsignificant publications in the lists. A quickperusal of these pages provides a soundindication about what journals are highlyregarded. For instance, faculty commonlyhighlight Forest Science, Forest Ecologyand Management, Canadian Journal ofForest Research and the Journal ofForestry. Publications in journals such asSociety and Natural Resources, SoilScience Society of America Journal andForest Products Journal are also commonlylisted.

Journal citation and impact indices

One of the best – and certainly the mostobjective – measure of the quality andimpact of journals is provided by citationand impact statistics. The ISI JournalCitation Reports are the leading source ofinformation about citation rates and impactof journals. This is a web-based service

which can be found at http://jcrweb.com//.Two databases are available, one forscience journals and the other for socialscience journals. Within these databases,journals are further classified in subjectcategories. ‘Forestry’ is one of the disciplineareas in the biological sciences. Within thesocial sciences list is ‘Environmentalstudies’. Information on citation rates andimpact of 29 journals is available within theforestry group (Table 1).

The journals can be sorted in order of totalnumber of citations and impact factor.Figures 1 and 2 provide screenshots of theISI JCR for forestry journals demonstratingsorting in total cites (i.e. citations) andimpact factors. These rankings provide aneffective guide of the top journals in aparticular field and hence the best ones inwhich to publish. Importantly, reference canbe made to the ranking of the journal tojustify statements about its quality whenpreparing a CV or grant application.

Journals can also be sorted on the based ofImpact Factor, Immediacy Index, Cited HalfLife and Citing Half Life (Figure 3). Detailsof the basis of how these indices arederived is provided in Figure 4.

9. SEEING IT THROUGH

Bringing a paper to publication stage can bea long and frustrating process, and calls forconsiderable persistence. Some highlyoriginal thinkers and capable researchersmove on to the next topic too soon, anddon’t do the ‘hard yard’ in report writing.Even when a paper is written and is suitablefor a journal, this is not the end of the story.The 80%: 20% rule should be kept in mind:finishing off the last 20% of a paperproduction process takes 80% of the effort.

Notably, in publication success breedssuccess. When a researcher has a strongtrack record, invitations to contributechapters in books and papers for regularand special issues of journals are likely tofollow.


Table 1. Forestry journals listed in the ISI Journal Citation Reports

Abbreviated journal titleYear 2000total cites

Impactfactor

Immediacyindex

Year 2000articles

Citedhalf-life

AGR FOREST METEOROL 2891 1.588 0.950 141 7.8

AGROFOREST SYST 824 0.918 0.250 60 5.8

AI APPLICATIONS 87 0.500 99.999 0

ALLG FORST JAGDZTG 141 0.239 0.031 32 6.9

ANN FOR SCI 53 0.576 0.203 69

ANN SCI FOREST 757 1.897 99.999 0 6.4

CAN J FOREST RES 5597 0.955 0.117 206 9.1

EUR J FOREST PATHOL 494 0.696 99.999 0 9.5

FOREST CHRON 555 0.417 0.203 64 8.0

FOREST ECOL MANAG 3231 0.982 0.172 332 5.5

FOREST PATHOL 2 0.000 34

FOREST PROD J 953 0.329 0.041 121 >10.0

FOREST SCI 2073 0.966 0.200 15 >10.0

FORESTRY 477 0.698 0.068 44 >10.0

FORSTWISS CENTRALBL 132 0.263 0.036 28 10.0

HOLZFORSCHUNG 1500 0.981 0.093 108 8.2

IAWA J 426 0.738 0.065 31 9.1

INT J WILDLAND FIRE 188 0.400 0.000 21 6.0

J FOREST 1140 0.451 0.120 25 >10.0

J VEG SCI 1924 1.589 0.085 82 5.7

NAT AREA J 330 0.452 0.095 42 6.8

NEW FOREST 184 0.417 0.027 37 5.6

PLANT ECOL 500 0.822 0.054 112 3.2

SCAND J FOREST RES 660 0.519 0.155 71 7.7

SILVAE GENET 660 0.312 0.074 27 >10.0

TREE PHYSIOL 2292 2.052 0.436 140 5.2

TREES-STRUCT FUNCT 735 1.122 0.191 47 6.0

WOOD FIBER SCI 585 0.446 0.096 52 >10.0

WOOD SCI TECHNOL 673 0.291 0.059 34 >10.0


Figure 1. Sorting on basis of total cites

Figure 2. Sorting on the basis of impact factor


Figure 3. Some citation details for an individual journal (Forest Ecology and Management)

Figure 4. The basis of calculation of some key citation inidices


10. SOME JOURNALS TO TARGET FORFORESTRY-RELATED ARTICLES

The following journals are some of the oneswhich would seem appropriate for forestryresearch, particularly with a systems orsocio-economics orientation.

Agricultural SystemsAustralian ForestryAnnals of Tropical ResearchAustralian Journal of EnvironmentalManagementAustralian Journal of Agricultural andResource EconomicsEcological EconomicsEconomic Analysis and PolicyEconomic GeographyForest Ecology and ManagementForest Policy and EconomicsForest ScienceJournal of Environmental ManagementJournal of Forest ProductsJournal of ForestryJournal of Sustainable ForestryNatural Resources ForumSmall-scale Forest Economics,Management and PolicySociety and Natural Resources

11. SOME REFLECTIONS AS AJOURNAL EDITOR

A number of observations may be madefrom experience as a journal editor. Thefollowing comments arise from four yearsexperience as editor of Economic Analysisand Policy (for the Economic Society ofAustralia Inc. – Queensland Branch). Thisjournal, which has been published for morethan 30 years, is issued twice yearly, withoccasional special issues, individual issuetypically containing about seven papersplus several book reviews.

The paper processing system

When submitting papers to a journal forpublication, it is desirable to have a clearunderstanding of the paper processingsystem adopted by the journal. Thefollowing is a typical example of how thisprocess works.

When papers are received, the editor willusually have a quick look at them, andperhaps seek an opinion of a colleague or

member of the editorial committee. Aneditor can often judge quite quickly thesuitability of a paper and the probability offavourable reviews. If they judge the paperis not suitable for publication, even withsome further work, then the editor is likelyto make an executive decision to reject thepaper and advise the author accordingly.This could occur for example if the researchreported lacks substance, the analysisappears weak, or the literature review isabsent of weak.

In other cases, the editor may correspondwith the author and say the paper haspotential but is not suitable to send toreview in its current form, e.g. it may be toolong, too unsatisfactory in format (noAbstract, inadequate Introduction), orcontaining too much algebra for the journalaudience. If the editor is generous with theirtime, they may provide considerablefeedback at this stage, even to the extent ofproviding various suggested text changes.

If these hurdles are passed, the editor ortheir committee will identify suitablereviewers, and ask them if they areprepared to review the paper for the journal,typically sending the full paper or theabstract to assist them to make a decision.

In due course, the reviewer’s comments arereceived. These may be to reject the paper,or that the paper is suitable for publicationprovided specified changes are made. Thereview comments are usually forwardedunattributed to the authors. If the reviewsare mixed, the editor must make a decisionabout whether to proceed with the paper. Ifthe decision is to proceed, then the editorwill provide an indication to the authors ofwhat changes they think are needed for thepaper to be satisfactory.

Once revisions are made to the paper, theeditor may decide themself whether thesechanges are adequate, or may send therevised paper to the reviewers and askthem if they are satisfied with the changes.In the latter case, if a reviewer is stillunhappy with the paper, the process maybe prolonged.

Hopefully, the paper is eventually judged tobe acceptable, at which stage the editormay require wording changes, more


complete details on references. They mayalso request camera ready copy of complexgraphics or computer screen images, andmay discourage coloured diagramsbecause they can lose definition in blackand white. The paper eventually goes to thetypesetter, after which page proofs areproduced and forwarded to the authors.Once all authors have corrected their pageproofs, the typesetter makes finalcorrections and the issue is printed.

As indicated by these notes, paperprocessing can be a long and tortuousprocess.

Submitting papers

Journals typically provide instructions toauthors of how papers are to be formatted.Often about four copies are to be provided,which must follow quite detailed formattinginstructions. Presumably, the copies aredate stamped on receipt, one copy isretained by the editor, one is filed away,and two are posted to reviewers. This is arather archaic system, and nowadays it isgenerally more convenient if a single copyis submitted as an email attachment, notnecessarily is a precise format, and copiesare emailed to reviewers. The formattingaccording to in-house style can be then bedone if and when the article is accepted. Itis worth asking the editor if this method ofsubmission is acceptable.

Contacting the editor

Some authors are loath to contact the editorafter a paper is submitted. (Others phonethem repeatedly.) In general, if there hasbeen no response from the editor (otherthan an acknowledgment of receipt of apaper) after a few months, it is a good ideato contact the editor and inquire aboutprogress of the paper. This of course willdepend on the journal. Some have longturnaround times, but the authors are ratherprecious and will not respond well tocontact. But generally, a follow-up is a goodidea, the check that the editor has not lostthe paper, or (more likely) failed to chase upslack reviewers.

Responding to reviewer’s comments

When the editor indicates a paper will be

considered further subject to makingspecified revisions, it is critical for theauthors not only to make these revisions orprovide good reasons for not doing so (thegeneral rule is that ‘the reviewers arealways correct’), but also to provide a list ofthe changes they have made.

Handling page proofs

Page proofs may be provided as hard copyor in electronic form. At this stage onlyminimal and essential changes areacceptable. Generally, the author shouldforward had copy with clearly markedchanges, although it the changes are minorit may be acceptable to advise by email.

What not to do

There are a number of things authors cando which will upset editors. In that it is wizeto develop a friendly relationship with aneditor, with a view to submitting furtherpapers, these need to be kept in mind.

* Simultaneous submission of papers tomore than one journal is unacceptablebehaviour. There is a lot of work for editorsand reviewers in processing a paper, and ifthe paper is withdrawn from one journal tobe published elsewhere or they becomeaware that it is being considered by anotherjournal, then your name is likely to beremembered should you submit to thatjournal again.

* Submitting a paper that is substantially thesame as has been published elsewhere (bythe contributor or by someone else) is alsounacceptable. A ‘repeat’ paper may bedetected by the reviewers, in which casethe editor will probably say cautiously thatthe paper cannot be accepted because it istoo similar to something that is publishedelsewhere. An even worse outcome (for thecontributor and journal) would be for thepaper to be published and then forsomeone to submit a comment paperpointing out the lack of originality.

* Submitting a paper which contains manyviews or subjective statements but lacksreferences can come across as presentingthe author’s personal philosophy of theworld. Unless the author is quite brilliant, itis probable that the ideas have already


been thought of by others, and that aliterature exists on them.

* Other things which will upset editorsinclude: failure to provide a rejoinderexplaining what changes have been madein response to reviewers’ comments; failureto shorten a paper when the direction is thatit must be cut down in length; failure toprovide promptly camera-ready graphicswhen this is requested; long turn-aroundtime with page proofs.

Length of papers

How long should the paper be? This willvary between disciplines. In the physicalsciences, papers as short as about 2000words are often published. Longer papersare acceptable in the social sciences,though manuscripts longer than about 7000words approximately are generally notwelcomed by editors. They take up toomuch space in the journal, such that onelong paper may displace two shorter ones.They test the endurance of readers. Theymay invoke the response that the editorwould be prepared to consider a shortenedversion. A length of 5000-6000 words isprobably more optimal. Often, journalinstructions will indicate, after a statementabout journal articles, that ‘shortercontributions are welcome’. Sometimesthere is a separate section in journals forshorter comments.

Some further observations

A few further observations may be madeconcerning dealing with journals:

* Bear in mind that the editor is alwayslooking for suitable papers to publish. Withincreased advertising of journals, includingadvertising on web sites, the number ofpapers submitted is increasing (increasingdemand for publication), but there havebeen various new journals launched in thelast few years, so the supply of journaloutlets is increasing.

* In general, papers should not contain alarge amount of graphics. These make apaper look too much like a slide show.Tables provide more information thangraphs, because with tables numbers are

known exactly.

* Don’t wait for the editor to get back to you,if the time delay is too great. They mayhave lost the paper, or not got around tochasing up a slack reviewer.

* Choice of reviewers can make or break apaper. The author is unlikely to have anysay over choice of reviewers, but in somecircumstances can make suggestions. Inthe case of invited papers, or convertingworkshop papers into journal articles,sometimes a deliberate choice if reviewersis made such that they will not be‘unfriendly’.

* The focus or editorial policy of the journalmust be kept in mind. Some papers nomatter how good simply are not suitable forthe journal. ‘Swiggle papers’ (highlymathematical, with lots of Greek characters)should be sent to ‘swiggle’ journals.Sometimes an editor will say it is necessaryto ‘Move the maths to an appendix, andexpand the policy section’.

* It can be difficult to obtain acceptances for‘review’ articles, as distinct from results oforiginal research.

* In general, when a paper presents resultsof a survey, some hypotheses andstatistical analysis would be expected,together with some test statistics anddiagnostics to justify assumptions.

* Journals are generally short of bookreviews. Unsolicited book reviews areusually welcome.


The publication record is invariablyregarded as a key criterion of ability ofresearchers, including those working withinuniversities. Different publication strategieswill work best different people, and it isnecessary to think strategically abouthaving research findings published. Tosome extent, success is a matter ofpersistence, and skills can only bedeveloped by experience. The observationsand suggestions presented here aredesigned to stimulate further thought aboutpublication strategies.

235

20. Experiences Gained and Lessons Learnt fromthe Training Workshop

Steve Harrison, Ed Mangaoang and John Herbohn

This review module attempts to draw together some of the experiences gained and lessonslearned from conducting the training workshop on socio-economic research methods inforestry research. A workshop of this nature had been an ambition of The University ofQueensland forestry research group for more than two years, so it was gratifying to see thisambition fulfilled. As a first run by the research group at such a training workshop, this wassomething of an experiment, and impressions gained here may be of assistance should the‘training package’ be employed in other situations. This paper reflects on the workshopexperience and on the feedback obtained from delegates and presenters. No formal surveywas undertaken to gain impressions of participants, but considerable informal feedback wasobtained.

1. OBJECTIVES OF THE WORKSHOP

In designing the workshop, a number ofobjectives were in mind:

• The workshop would fulful part of thecapacity-building objective of ACIARp r o j e c t ASEM/2000/088 –Redevelopment of a Timber IndustryFollowing Extensive Land Clearing. Inpar t icu lar , there would bedevelopment of new skills byparticipants in socio-economicanalysis and in modelling forestrysystems.

• This would be an opportunity tocommunicate the forestry researchmindset of the Australian team to theFilipino team, and obtain feedback,and hence to explore interactively theparadigms, methods of analysis andassumptions upon which each groupis working.

• The planning of the ACIAR projectwould be advanced, particularlythrough the research ‘case studies’which were tightly tied to objectives ofthe project. This would includedevelopment of a prototype model forsimulation of community forestrysystems.

• The workshop would establish thebasis for improved communicationand closer understanding between

the researchers throughout theremainder of the ACIAR project.

• A further objective in planning theworkshop was to provide a stimulatingand enjoyable experience forparticipants.

• A planned tangible output was thistraining manual, to be suppliedsubsequently to all participants.

2. PLANNED WORKSHOPARRANGEMENTS

Formal presentations were designed on anumber of specific topics, but with a view togroup participation and discussion. Morespecifically, the workshop meetings weredesigned to include the followingcomponents:

1. Presentations by authors of the notes2. Presentations by discussants3. Small-group development and

presentation of case studies4. A fieldwork component5. ‘Hands-on’ computing sessions.

Ideally, the instruction papers would havebeen prepared and distributed to delegatesin advance of the workshop. This wouldhave allowed formal sessions toconcentrate on specific issues, andprovided a basis for contributions bydiscussants. Nominated discussants couldhave been invited to prepare brief


presentations on the basis of materialsprovided to them in advance. In particular,they could have lead discussion on therelevance of the particular concepts andmethods in forestry research, from theirorganizational perspective. However, ingeneral this was not possible, and a foldercontaining some of the printed modules wasprovided to participants at the time ofregistration, with other modules providedduring the workshop.

An objective of the workshop was to haveeach participant prepare a small simulationmodel using the Simile package, and mostwere able to achieve this.

The first three days of the workshopcovered the main presentation sessions,with the emphasis for the next three dayson case studies. The field trip wasscheduled to provide a change from theformal sessions, and to collect data forsubsequent used in development of theprototype community forestry model.

3. LOCAL ARRANGEMENTS

The organisation on the part of staff ofLeyte State University was excellent, interms of meeting facilities, copying ofmaterials, transport and meals. This kind ofmeeting always imposes a heavy burden onthe local organisers, and their efforts weregreatly appreciated. A number of socialevents were organised, which made theoccasion memorable and built morale andcohesion in the group.

4. REVIEW OF TOPICS COVERED

Five major topic areas were covered,namely:

1 . Basic computing skil ls, usingspreadsheets, discounted cash flowanalysis and statistical analysissoftware.

2. Specific research techniques (samplesurveys, cost-benefit analysis, non-m a r k e t v a l u a t i o n , l i n e a rprogramming).

3 . Modell ing and simulat ion ofcommunity forestry systems withSimile.

4 . G r a n t s e e k i n g , r e s e a r c hadministration and publicationstrategy.

5. Practical exercise in developing aresearch project.

5. SUITABILITY OF THE CONTENT ANDENGAGEMENT OF PARTICIPANTS INDISCUSSIONS

As a first run, it was to be expected thatthere would be some glitches. An ambitiousagenda was planned for the workshop, andthough much material for presentations hadbeen drafted before the meeting, continuedmodule development took place betweenworkshop sessions.

The choice of topics of necessity dependedon the interests and skills of the group ofpresenters. In this regard, it was fortunatethat the visiting group covered a wide rangeof interests, and that senior staff from LeyteState University were also able to makepresentations.

The program proved to be highly ambitious,in scope, complexity and amount of materialpresented. However, flexibility was providedin the schedule, between presentations ofnew materials, discussion of case studiesand computing sessions. In general, daysran from 8 am to 6.30 pm. Despite thispace, by all impressions, delegates foundthe workshop absorbing, sometimescontinuing computing through coffeebreaks.

It was hoped prior to the meeting that thepresentations would simply be lead-ins totopics, and that there would beconsiderable interaction and debate atworkshop sessions. To some extent, thisambition was achieved. However, it is to benoted that the material presented wasrather foreign to most of the participants,and this at times constrained thediscussion. Although the participants had abackground in natural resourcemanagement, and were engaged to somedegree in policy and extension work, mosthad little background in economics or socialresearch.

No doubt, at times some participants feltrather lost with the topics, and it will require

Experiences Gained and Lessons Learnt from the Training Workshop 237

further reflection to gain a workingunderstanding of the techniques presented.Were a similar training workshop run in thefuture, with the materials which are nowavailable, it would be possible to providenotes for reading in advance and hencemake greater use of meeting time.

6. LEVEL OF COMPLEXITY ANDBACKGROUND KNOWLEDGE ANDSKILLS ASSUMED

One difficulty in holding a workshop for agroup of people from varying academic,industry, local government, NGO and otherbackgrounds is that the level of familiaritywith the presentation topics and the level ofexpertise with computers can be expectedto vary widely. In that the training workshopwas designed to broaden the skills baseand interests of the participants, and wasoutside the area of professional experiencefor most, some of the concepts ineconomics and sociology were found to benew and strange.

The modelling package Simile takesconsiderable effort to comprehend.Becoming adept at a complex computermodelling package such as this by naturetakes a good deal of time, for assimilation,experimentation and consolidation, to gainconfidence as a user. The workshopprovided an ideal introduction to Simile, butsubstantial further time will be requiredbefore participants can start to develop andrun realistic models. An added complicationhere is the need to become familiar with theconcepts and procedures of modelling andsimulation, which can be a hurdle for peoplewithout a background in this area.

It was found during the hands-on sessionsthat most of the delegates did not have thelevel of computing experience expected ofthe presenters, particularly in regard to theuse of the Microsoft Office package and theExcel spreadsheet. While a desktopcomputer and this software is regarded as astandard facility at The University ofQueensland, the high cost of computersrelative to university budgets can makeaccess more limited in lower-incomecountries. In the College of Forestry atLeyte State University, the stage has notbeen reached where each faculty membercan have exclusive access to a desktop or

notebook computer, and so there is lessfamiliarity with Excel, which was appliedwidely in the modules for financial modelling(including discounted cash flow analysis,cost-benef i t analysis and l inearprogramming). Hence it was necessary toprovide some background coaching in useof Excel.

7. WORKSHOP OUTCOMES

Areas found to be of high interest

There was clear enthusiasm for themodeling activities and hands-on computingin relation to the simulation modelingpackage Simile, development of financialanalysis spreadsheets with MicroSoft Exceland linear programming using the Solverfacility. The modules on researchadministration and publication strategy alsoappeared to be of high interest.

Benefits derived by delegates

In general, the workshop objectives aslisted above were achieved to a highdegree. As well as the more formal trainingaspects of the workshop, some practicalassistance in computing was provided tothe delegates, e.g.

• providing practical tips in computing,particularly with regard to use of theExcel spreadsheet package

• providing access to the add-ins in theExcel spreadsheet package, such asstatistical techniques and the linearprogramming package (Tools –Solver).

• providing copies of the Similemodelling and simulation packagedeveloped by the University ofEdinburgh, which was downloadedfrom the developer’s Web site.

Development of the prototypecommunity forestry model

An important objective of the workshop wasto make some steps towards developmentof a model of community forestry, whichcould be used to simulate forest industrydevelopment and carry out simulationexperiments with alternative reforestation


policies. It was recognized that only limitedprogress could be made during the shortperiod of the workshop and with thecompeting objectives planned. In particular,a field visit – unless organised as aparticipatory rural appraisal meeting – couldnot be expected to provide a great deal ofdetailed information for modelling purposes,although it did provide an excellent chanceto gain first hand observations of acommunity forestry site.

Suitability of OrganisationalArrangements

Some difficulty was found in providingsufficient access to computers during thehands-on computing sessions. Threenotebook computers were brought to theworkshop, and another notebook computerand one desktop computer were availablelocally, which meant that groups of aboutfive people were sharing computers.Groups of three would have probably beenpreferable.

It was noted that local delegates wereunable to attend all sessions. This was to alarge extent due to their involvement in thepreparation of materials and otherarrangements for the workshop, and partlybecause of other commitments. Thissuggests that there could be advantages inholding workshops of this kind away from

the normal place of employment ofdelegates, such that they become a ‘captiveaudience’, though this would of courserequire an out-of-semester time window,and would tend to increase costs in terms oftravel and accommodation.


In general, the workshop appears to havebeen highly successful. While sometentative recommendations might be madefor future training programs of this type –e.g. make available more computers toallow smaller user groups, gain a betteridea of the computer access and computingexperience of participants – it would seemdesirable to retain flexibility to meet theneeds of the audience and adaptpresentations as suits the progress of thegroup.

It was to be expected that some problemswould arise, but overall the workshopappears to have been highly successful inachieving its objectives. Lasting benefits areexpected for the ACIAR research project interms of a shared research methodologyand vision, and progress in planning ofspecific activit ies. This engendersconfidence for conducting such a trainingprogram as a capacity-building exercise inother locations.

SOCIO-ECONOMIC RESEARCH METHODS IN FORESTRY: A...

Documents

Transcript of SOCIO-ECONOMIC RESEARCH METHODS IN FORESTRY: A...