1-s2.0-S0305750X01001279-main_3

Choosing Rural Road Investments to Help

Reduce Poverty

DOMINIQUE VAN DE WALLE *The World Bank, Washington, DC, USA

Summary. — The paper first critically reviews past and current practices in how rural roadinvestments are selected. An attempt is then made to develop an operational approach that isgrounded in a public economics framework in which efficiency and equity concerns are inseparable,information is incomplete in important ways, and resources are limited. A key problem addressed isthat a potentially important, though unknown, share of the benefits to the poor from rural roadscannot be measured in monetary terms. The proposed method aims to identify places wherepoverty, inaccessibility and economic potential are high. The method is illustrated for Vietnam.� 2002 Elsevier Science Ltd. All rights reserved.

Key words — rural roads, poverty, evaluation, Southeast Asia, Vietnam

1. INTRODUCTION

It is widely agreed that the economic ap-praisal of development projects should helpselect projects that contribute most to socialwelfare. The chosen projects should yield largergains in social welfare than alternatives. Putin such general terms, the objective is clearenough. But its implementation, and particu-larly how to measure net benefits, are rarely soclear. This paper focuses exclusively on theappraisal and selection of investment projectsin the rural roads sector, where the specificobjective is taken to be poverty reduction. Thisis broadly defined to include relevant nonin-come dimensions of welfare. How one might goabout choosing between road investments isdiscussed in general terms with some specificillustrations from current work in Vietnam.A vocal group of rural road enthusiasts

has claimed that rural roads result in signifi-cant social benefits. Since these are difficult toquantify, they have typically been omitted fromconventional appraisal techniques. It is furtherargued that this has led to longstanding biasesagainst rural road projects and (since the poorare primarily rural) that there are biases againstpro-poor investments. As a remedy, specialtechniques have been devised for evaluatingand selecting rural road projects that simplytake the eventual flow of important socialbenefits as given. Unfortunately, there is as yet

little convincing empirical evidence that ruralroads affect social outcomes beyond what theywould have been without the road. Althoughthe argument that high social benefits will ensueis sometimes plausible, the evidence provided injustification is rarely so. Without better evi-dence, there can be no presumption that suchbenefits will be high or even positive.Given the poor quality of what we do know

in this area, there appear to be two solutions.The first is to simply halt all rural road con-struction that does not pass conventional cost–benefit rates of return criteria and wait untilunambiguous empirical evidence is available.Past experience does not suggest that thisoption will be favored. Decisions continue to bemade with imperfect knowledge and that is not

World Development Vol. 30, No. 4, pp. 575–589, 2002� 2002 Elsevier Science Ltd. All rights reserved

Printed in Great Britain0305-750X/02/$ - see front matter

PII: S0305-750X(01)00127-9www.elsevier.com/locate/worlddev

*This paper was in part written while visiting theUniversit�ee des Sciences Sociales, Toulouse. I would liketo thank PREMPO for funding support under its impactevaluation thematic group. I am grateful to DorothyjeanCratty, Ken Gwilliam, John Howe, Jerry Lebo, Chris-tina Malmberg-Calvo, Martin Ravallion, two anony-mous referees and seminar participants at the WorldBank‘s Economists’ Week for useful discussions, helpand comments on the paper and topic. The findings,interpretations, and conclusions expressed in this paperdo not necessarily represent the views of the WorldBank, its Executive Directors, or the countries theyrepresent. Final revision accepted: 19 November 2001.

575

likely to change here. The second option is toimprove on what is currently done and come upwith a coherent, internally consistent evaluationmethodology that explicitly says how muchlower a rate of return to measured benefits oneis willing to accept, and to test that with sub-sequent research. The point of this paper is notto decide which solution should be chosen butrather, to try to lay out the foundation for thesecond. The paper argues that there should beresearch on two fronts simultaneously. Specialefforts need to be directed at measuring the ex-istence and magnitude of the so-called socialbenefits from rural roads. At the same time,work needs to be done on improving themethods widely used to appraise and selectrural road projects in the absence of that evi-dence.To this end the paper proposes an alternative

approach. The proposal recognizes explicitlythat an important problem for some types ofpublic spending, including rural roads, is thatthere is a potentially sizable share of the benefitsthat cannot be measured in monetary terms soas to be aggregated consistently with monetarymeasures of other benefits and costs. But, re-search should at least be able to provide an as-sessment for a few selected cases, which canprovide a benchmark. There are participatorymethods for tapping local information to formjudgements of the relative importance of dif-ferent types of benefits in a specific setting. Theproposal tries to use the information availableto form a second best appraisal method, takingaccount of the informational constraints facedin practice.In the following sections, the paper argues

that a change in the transport sector’s currentapproach to rural road investment selection iswarranted along the lines described, buildingon some of the poverty-focused ‘‘hybrid’’methods found in recent rural road appraisalsat the World Bank and elsewhere. The paperfirst critically reviews the methods typicallyused for selecting roads, both conventionalcost–benefit analysis (Section 2) and the morerecent hybrid methods which combine cost–benefit methods for some projects with cost-effectiveness calculations for others (Section 3).Section 4 discusses efforts at quantifying typi-cally excluded benefits. This is followed, inSection 5, by an examination of the relevanceof the traditional approach in the context of apoor rural economy, using Vietnam to illustratethe points made. Survey data are used to testthe approach’s underlying assumptions. The

paper then proposes an alternative approach.Section 6 sets out the problem to be solved.Section 7 presents the proposed methodol-ogy and Section 8 looks at implementation.The paper ends with some concluding com-ments.

2. TRADITIONAL COST–BENEFITANALYSIS OF ROADS

There is some research on the importanceof infrastructure, and in particular road in-frastructure, to agricultural output, economicgrowth and poverty reduction (including Antle,1983; Binswanger, Khandker, & Rosenzweig,1993; Fan & Thorat, 1999; Jacoby, 2000; Jalan& Ravallion, 2002). For example, Jalan andRavallion (2002) found that road density wasone of the significant determinants of house-hold-level prospects of escaping poverty inrural China. It is far from clear, however, thatexisting methods of project appraisal for ruralroads will properly reflect the potential benefitsto the poor.Cost–benefit analysis methods for appraising

investments in the road infrastructure sec-tor were first developed for roads in more ur-banized, high-traffic density areas, drawing onmethods from a developed country literature onroad appraisal. Traditionally, road investmentsin World Bank financed projects have been se-lected based on benefit indicators derived fromconsumer surplus calculations of road usersavings, comprising both of vehicle operatingcost savings and journey time savings. Forecastsof traffic demand—reflecting both normalgrowth in traffic and that generated by theproject—are used to derive willingness to payestimates to proxy project benefits. Over time,the approach has been implemented at differentlevels of sophistication, anywhere from onlyconsidering benefits accruing to motorizedfour-wheel vehicles to also including gains tononmotorized traffic and pedestrians based onreduction of travel time savings. In some cases,estimates of the value of agricultural productionincreases induced by the road investment arealso included. 1 The appraisals have gener-ally not made distinctions between beneficiariesfrom different income or other socioeconomicgroups.A number of criticisms have been leveled at

this approach (Gannon & Liu, 1997; Hine,1982). One is that it tends to bias investmentstoward richer areas since the demand for traffic

WORLD DEVELOPMENT576

and hence, willingness to pay measures, arehigher for the rich. Another is that it is ap-propriate for high, but not for low traffic areas;and relatedly, that it fails to capture some im-portant, but hard to quantify, benefits fromroad investments. For these reasons, some ob-servers argued that the method led to under-investment in rural roads and in particular,rural roads serving poorer populations. Thereare projects that, by conventional cost–benefitanalysis based on poorly measured benefitstreams, do not have an internal rate of returngreater than the critical level (typically set at12%), yet yield higher social welfare gains thanthe projects that do pass the test.In the late 1970s, a number of papers inside

the World Bank argued for replacing, or sup-plementing, consumer surplus measures withproducer surplus benefit measures for roadswhere traffic levels are low (see Beenhakker &Chammari, 1979; Carnemark, Biderman, &Bovet, 1976). The case for a change in focusrested on the induced agricultural develop-mental impacts of roads not captured by trafficcost savings when traffic is low. Producersurplus estimates aimed to capture gains inagricultural incomes resulting from transportimprovements and concomitantly higher farm-gate, and lower input, prices. The aim wasto prevent biases caused by sole emphasis onconsumer cost savings in predominantly agri-cultural areas. Complementary agricultural de-velopment programs were also emphasized inorder to maximize road investment returns(Beenhakker & Chammari, 1979).Cost–benefit (CB) analysis as currently

practiced in the transport sector continues tobe riddled with problems in how benefits aremeasured. Valuing benefits for nonmarketgoods for which prices are not known andthe consumption of which is subject to quan-tity constraints is difficult (Cornes, 1995). Oneproblem concerns a lack of agreement on thesocial welfare function on which these valuationjudgments are ultimately based. ConventionalCB analysis does not unambiguously answer thequestion of how much should be spent on ruralroads. A fundamental source of the ambiguityhas to do with the weights people attach to themultiple objectives of policy. Of course, theseproblems are not faced by the road and trans-port sectors alone. These issues are sharedthroughout public finance and public policy.The main problems in conventional methods

for assessing rural roads relate to alleged sys-tematic exclusion of certain benefits, faulty

measurement of the included benefits, and fail-ure to recognize that the assumptions needed tojustify ignoring distributional impacts—and sofocus solely on efficiency gains—do not hold inpractice.It is claimed that conventional appraisal

methods, even when combining consumer andproducer surplus, are still likely to result in theunderfunding of rural roads. Critics maintainthat the techniques omit some key benefits,such as those accruing to individuals and tosociety from increased attendance to schools,health and other facilities rendered accessibleby the road investments. Accompanying dis-tributional benefits are also ignored. Further-more, there may well be large but omitted riskinsurance benefits from linking isolated poorerpopulations to national transport and com-munication networks. Quantification of suchpurported benefits remains largely intractable.These omitted benefits would be of less concernif it could be argued that they are positivelycorrelated with the included benefits. That isnot, however, plausible. Rural roads may wellhave high omitted benefits but low includedbenefits. Ranking road investment options interms of observable benefits may be onlyweakly correlated with the ranking in terms oftotal benefit. If the alleged social benefits arereal, conventional methods are unlikely to be areliable guide to project selection.Current methods of estimating the included

benefits are also questionable. Both consumerand producer surplus are problematic as cur-rently measured. Typical consumer surpluscalculations for roads tend to exclude consumergains from changes occasioned by the road innontransport goods prices. Average daily trafficmeasures frequently used in forecasting benefitsare hard to predict. Similarly, producer surplusmeasures tend to be incomplete and arbitrary inwhat is included. Why focus solely on farmersand agricultural produced surplus? Impacts onnonfarm employment and other income-earn-ing opportunities are typically not factored in.Producer surplus measures also often rely onthe same supply response parameters acrossregions, on spotty production data and makeuse of averages across income groups not al-lowing for household and geographic specificfactors that influence marginal benefits (van deWalle & Gunewardena, 2001).The use of distributional weights to counter

biases against poor areas has tended to befrowned upon within the sector (Gannon& Liu, 1997). As Gannon and Liu state

RURAL ROAD INVESTMENTS 577

‘‘Economic efficiency is widely accepted as theprimary objective of transport sector opera-tions and is used, through CB analysis, to guideproject selection and design’’ (Gannon & Liu,1997, p. 23). They argue that distributionalconcerns should be handled at the macro-eco-nomic level such as directly through the taxsystem and, that income distribution decisionsare essentially a political responsibility. Theyalso feel that ‘‘use of distributional weights is,by and large not appropriate’’ since they ‘‘aresubjective, vulnerable to misinterpretation andopen to manipulation’’ (Gannon & Liu, 1997,p. 26).The argument that the transport sector

should be geared to maximize efficiency isbased on a first-best model of the economy,whereby one aims for efficiency in production,and redistributive instruments such as the taxsystem and lump-sum transfers are used toachieve the redistribution objective. There aretwo problems with this view. First, the objec-tive can be questioned, and second, its imple-mentation is problematic in practice. The keyassumption underlying the ‘‘maximize effi-ciency’’ view is that a complete set of marketsexists and that other instruments are availablefor meeting the equity objective. Then theproductive sectors, such as transport, can beleft to deal solely with efficiency. Given marketfailures (including incomplete markets) andlimitations on redistributive instruments, therealism of the argument that transport shouldcare solely about efficiency can be questioned.Given that equity is valued, if one cannot es-tablish that there already exist the instrumentsneeded for redistribution, and that marketswork well, then focusing solely on efficiencybecomes unsupportable.Second, even if we agree that efficiency is the

objective of road investments, benefits muststill be measured properly and thoroughly.Otherwise, it is entirely possible that the effi-ciency objective is not in actual fact being met.If important net social benefits exist, then ben-efits are typically not being thoroughly mea-sured. Consequently, as discussed above, themeasurement of benefits will tend to emphasizebenefits to the better off and omit those thatfavor the poor. It can be argued that the ben-efits that one cannot measure tend to be thosethat accrue to the poor, so that achieving thepartial efficiency objective may well bias in-vestments against the poor. Thus, in both cases,biases in evaluation practice may go againstprojects that favor the poor.

3. POVERTY-FOCUSED HYBRIDMETHODS

Transport sector experts typically do notmake decisions about how much of the budgetshould be allocated to the sector, or even acrossbroad within-sector categories (such as ruralversus urban roads). They are presented with aset budget for investment in rural roads, say,and must then choose what road projects to do.This means that there are ways of choosingprojects that allow a more comprehensive as-sessment of the benefits, but do not claim tomeasure the social rate of return.A key difference between CB analysis and

cost-effectiveness (CE) calculations is that thelatter work only in situations where total ex-penditures for a program are fixed. In such acase, one only needs to decide how to allocatethe budget in the best possible way. There is noneed to use a consistent metric of benefits thatcould be the basis for comparisons with otherprograms or resource uses. Nor is there a needfor this benefits indicator to be expressed inmonetary units or for it to be comparable withindicators used for other programs. The onlyrequirement is to obtain an outcome indicatorper amount spent. It is an indicator specific tothe particular program and would not neces-sarily be of interest to any other program.Thus, although CB and CE both measure theratio of benefits to cost, the ‘‘benefit’’ units aredifferent. To put the CE indicator in a broadercontext would require a comparable measure ofthe social value of the project outcomes.A number of projects in the World Bank and

elsewhere have turned to CE calculations totake account of a broader set of benefits—suchas potential health and education benefits—yetget around the problem of putting a monetaryvalue on them. The method is sometimes re-ferred to as multicriteria analysis (Cook &Cook, 1990). It has typically been used whentraffic volume is too low (<50 vehicles per day)for conventional consumer surplus measures tomake sense, yet, it is strongly believed that therewill be important social benefits. In general, aleast-cost approach is adopted. A thresholdlevel of costs is arbitrarily designated and pro-ject investments costing less are exempt from aconventional CB analysis that aims to maximizeefficiency alone. 2 The eligibility of subprojectsis then subject to ‘‘social criteria’’ such as pov-erty indicators meeting some predeterminedlevel. 3 In practice, the ‘‘social criteria’’ areoften no more than the number of population in


the zone of influence per unit cost. In othercases, potential subprojects are ranked accord-ing to indices based on a series of variablesdeemed to identify needier locations (see, forexample, the Zambia project, World Bank, 1997).If one accepts that the project as a whole

must reach some minimum internal rate of re-turn (though recognizing that this is based on apartial measure of benefits) then it is unclearwhy one would only measure the rate of returnfor subprojects above some cost level. There isno reason to suppose that the cheaper projects(of which there may be many) would have thesame (conventionally measured) rate of return.So there could well be a selection bias in thismethod. It would be better to estimate the rateof return to a random sample of subprojects.A further concern about past ‘‘hybrid meth-

ods’’ is with the benefits measure, which tends tobe crude. For example, a priori, there can be noassurance that higher population served perunit cost will translate into higher benefits froma road investment. Given identical numbers ofpotential beneficiaries, it is conceivable that ahigher investment cost due to worse terraincould produce considerably higher benefits, as aresult of resolving a worse access problem. Fur-thermore, it is not always clear why some variableis included in the benefit index, and even why it isweighted positively. For example, lower literacyis often treated in this way. Yet, lower literacy inan area might instead be taken as a positive in-dicator of need (in effect, a distributional weight)or a negative indicator of benefit, assuming thatthose among the poor who are literate will havethe highest marginal gains from access to a road.A sharper conceptual distinction is needed be-tween the ‘‘benefits’’ and how they are weightedto reflect concerns about distribution.A final concern is that the process of deter-

mining the variables and their weights shouldmore fully exploit the knowledge of local ex-perts and of the poor themselves. Road expertscan help on technical matters, but are unlikelyto be the best people to make the decisionsabout what information should be included inmaking a comprehensive assessment of the so-cial gains, and how that information should beaggregated.

4. ASSESSING THE EXCLUDEDBENEFITS FROM RURAL ROADS

Recognizing the possibility that some po-tentially important benefits arising from rural

road provision and rehabilitation are not in-cluded by conventional methods of measuringbenefits, there have been efforts to quantifysocial gains and add them to transport costsavings. 4 For example, in attributing educa-tion gains it has been assumed that better roadaccess will increase enrollments by an amountderived from mean national rates; previouslynonattending children are assumed to completeschool, and their life-time earnings predictedbased on a comparison of earnings for educatedand noneducated individuals nationally. Totaladditional earnings, appropriately reduced totake account of the costs of education, are thenadded into the road benefits measure.Such methods require strong assumptions.

Implicitly, road access is treated as the soleconstraint to children attending school. Yet,there could be a host of contributing reasonsthat may in turn, partly explain why that par-ticular road has not previously been built. De-mand for schooling could be low as a result ofhigh local poverty and the opportunity cost ofchildren’s time. Alternatively, there may becultural reasons keeping girls away, the returnsto education may be perceived to be low, or thequality of the school and teaching may be af-fecting the schooling decision. Second, it is alsoa strong assumption that when these childrenjoin the labor market, economic conditions willbe identical and that current earning differen-tials will persist.In attempting to account for these benefits

that are difficult to quantify, it is not uncom-mon for road project appraisals and impactevaluations to draw on socioeconomic indica-tors across geographic entities (villages, re-gions), delineated by whether they are servicedby a road, for evidence of such benefits andtheir magnitude. This is part of the approachintended for the research effort mentionedabove within the CB framework (South AsiaRegion, 1999), but this technique is also used asevidence in CE calculations. As is well knownfrom the evaluation literature, however, draw-ing policy conclusions from such statistics canbe misleading. Table 1 illustrates the potentialbiases using a simple model in which roadplacement is endogenous, and based in part onthe outcome indicator used for assessing im-pact. By simply comparing outcomes in villageswith roads versus those without, the evaluatorcan easily conclude that there are large benefitswhen in fact there are none.The general point here is that unless road

placement is truly random—which seems most


unlikely—simple comparisons of outcome in-dicators in villages with roads versus withoutthem can be very deceptive. Using such data asevidence of benefits without accounting for theprocess by which the road came to be built in aspecific location may lead to very deceptivepolicy conclusions and decisions. (Indeed, thereis nothing preventing a health project fromcoming along and replacing the ‘‘with andwithout road’’ to a ‘‘with and without a healthintervention,’’ and attributing the same incomegains to the health policy.)

5. HOW RELEVANT IS ALL THIS TOPOOR RURAL DEVELOPING

COUNTRIES?

Many of the aforementioned limitations ofconventional rural road investment appraisaland selection apply directly to poor, largelyrural developing economies. 5 For one, theassumptions underlying the ‘‘maximize effi-ciency’’ goal are generally not plausible in suchsettings. There tend to be few redistributiveinstruments available to policymakers. Indeed,we look to sectors such as rural infrastructureand roads to help achieve redistributive objec-tives. In addition, it cannot be assumed thatinvestments in rural roads will automatically bepro-poor. Failure to consider the equity ob-jective alongside the efficiency one will thus biassectoral investments against poorer areas andpoor people.

Consumer and producer surplus as conven-tionally measured appear to be inadequatemeasures of expected benefits in these settings.For example, in countries such as Vietnam andthe other Asian transition economies, it isparticularly difficult to predict how agriculturaloutput will alter, or how traffic levels will de-velop, given how many factors can begin tochange all at once. In many areas, labor andland markets will be newly developing along-side the road investment. Roads have been justone of many constraints to development. Theireconomic and social benefits will depend onmany factors including, for example, whetheraffordable transport services follow the roadinvestment. We know little about how the re-habilitation of a road link interacted with theother changes in the economy will eventuallyalter traffic flows and composition, the pro-vision of transport services, agricultural andother sectoral employment, input and outputmarkets. This may well also be true in morestatic economies—such as in sub-SaharanAfrica—where, due to a series of other con-straints, effects from the road may not bereflected in traffic levels or agricultural pro-ductivity for a long time. On the other hand,the argument that there may be substantial pro-poor gains from rural roads that are difficult tomeasure and to include in conventional CBanalysis could also apply to most of these ruralsettings. For these reasons, working within theCE framework and attempting to refine it,seems to be the most appropriate, as well aspromising, means of tackling rural road ap-

Table 1. Deceptive assessments of the gains from rural roadsa

Mean incomes in villages with and without a road ($/day/person)

Without road (n ¼ 56) With road (n ¼ 44) % increase (t-test)

Case 1: Road yields 20% income gain 1.287 2.413 87% (2.29)Case 2: Road yields no income gain 1.287 1.976 54% (2.00)

a The table shows mean incomes for a group of villages that do not have road access and a group that does. Meanincome is much higher for the villages with roads. From such statistics the conclusion is sometimes drawn that theroads generated these large gains—87% increase in mean income for one group of villages and 54% for the other inthis particular case.These numbers were created, however, by a model in which roads generated an income gain of only 20% for case 1,

and no gain for case 2. The model’s pre-intervention incomes were drawn randomly from log normal distributions.Road placement was determined endogenously, as a function of village income (with 25% weight) and a secondindependent log-normal random variable (75%). The latter could represent population size, ethnicity, likely votes,historical accident, or any other variable influencing road location by the government. Thus, roads are distributedacross villages in terms of a latent variable z ¼ 0:25y þ 0:75x where y is log income and x is the other determinant ofroad placement. The model gave a road only to villages with positive values of z.Of course, the evaluator does not know the true impact of the roads and is tempted to base an estimate on the

observed differences in mean incomes between villages with a road and those without. This yields a large overesti-mate.


praisal, once the decision to invest in ruralroads has been taken.Assuming that roads bring net benefits, as-

suring minimum access to all may be pro-poorin countries where the better-off are well-servedby past road investments. Further expansionwill tend to reach the poor. But, this may not bethe case in the poorest countries. Vietnam is acase in point. The country had negligible in-vestment in infrastructure for decades coupledwith destructive wars. The road stock remainssparse and in severe disrepair. There has clearlybeen a tendency to concentrate first on re-building higher level networks as opposed toinsuring basic access to isolated and poorcommunes. One would need a very large budgetto ensure a level of ‘‘minimum access’’ to alland yet, the benefits from any lesser goal willtend to be captured by better-off areas. Oneshould also consider whether providing roadaccess to isolated, poor communities is in allcases a cost effective use of scarce resources forpoverty alleviation. Thus, given an objective ofraising living standards in a cost-effective way,and given the fixed nature of most rural trans-port projects—where the total allotted budgetfor the project is almost certainly not sufficientfor ensuring some defined minimum access toall households—a method is still needed forranking road projects that takes into accountboth equity and efficiency.One response to the above arguments is that

we need not worry since inaccessibility is anadequate proxy for poverty in rural developingeconomies. It is also strongly implied that high-poverty areas have low economic potential.Such convictions underlie the rhetoric and jus-tification for current poverty-focused appraisalapproaches to rural road projects—whereby,typically, a budget is set aside for noneconomicor ‘‘social’’ objectives, not subjected to ordi-nary economic analysis, and projects are cho-sen so as to maximize the population providedwith ‘‘basic access’’ for a cost deemed accept-ably low. Under this perspective, the appro-priateness of a selection formula that aims toidentify places where poverty and economicpotential are high and access is low, is open toserious doubts.This paper argues that if one wants to use a

transport intervention to reduce poverty, it isimportant to worry about all three factors.Among places where benefits will be high, thereare both poor and nonpoor places; among poorplaces there are ones where access is bad andones where access is already good. Spending on

roads will not help the poor much if they al-ready have good access. Alternatively, in somepoor and low-access places, the costs may faroutweigh the potential benefits from improvedaccess. Other interventions—such as facilitatingoutmigration—are more cost-effective ways toreduce poverty.Only data can help resolve these tensions.

For example, data can throw light on the ar-gument that the poor are concentrated in areaswhere access is bad and vice versa. If the em-pirical evidence supports that view, one vari-able can be dropped from the formula used toidentify appropriate interventions.A commune-level data base covering 200 of

Vietnam’s communes in six provinces allows aninvestigation of these issues. 6 These data wereused to create measures of poverty, inaccessi-bility and economic potential by commune.Poverty is represented by an index that com-bines the rate of infant mortality, the rate ofmalnutrition for children under five, and theincidence of hungry households in the com-mune. 7 Inaccessibility takes into account theexistence of passenger and freight transportservices, kilometers of commune roads per area,access to different levels of road, railroads,navigable waterways and whether a paved all-weather, or paved sometimes impassable, com-mune-level road runs through the commune.Economic potential reflects population density,agricultural potential (here represented by irri-gated agricultural land per capita), the numberof social and economic facilities, human capital(% of children 15 and under who have com-pleted primary school) and number of otherdevelopment programs. Each of the index com-ponents was attributed points reflecting low, me-dium or high values—determined by the range ofthe data—for a maximum of 100 points for eachindex. One can certainly quibble both with thevariables included, as well as with how they areaggregated. Yet, the general conclusion was notin the least altered by sensitivity tests changingthe combination and aggregation of the variables(including using far fewer variables to constructthe inaccessibility and poverty indices in casedifferent components cancel each other out).The communes were ranked according to

each of the three measures. Figure 1 plots thecommune rankings by inaccessibility againstrankings by poverty; Figure 2 does the same forinaccessibility and economic potential; andFigure 3 does so for economic potential andpoverty. As is readily seen, there is very littlecorrelation between any of these rankings.


These data clearly show that, in Vietnam atleast, one cannot simply reduce the choice toplaces with either poor access, or high poverty,or economic potential. It will be important tofigure out how to combine and weight thesefactors so as to select the places where roadswill have the greatest impact on poverty giventhe cost. Clearly, there are places where inac-cessibility, poverty and economic potential areall high, identified by the northeast quadrantin each figure. Project selection needs to be ableto identify the intersection of the three. This iswhere returns to road investments will behighest. Of course, even among these places

further targeting choices will exist but they willmatter much less. 8

6. THE APPRAISAL PROBLEMREVISITED

Let us assume initially that a fixed budget isavailable for raising living standards throughthe construction or rehabilitation of rural roadlinks. How should the budget be allocated? Inanswering this question, one must consider theallocation between regional entities, such asprovince, district and commune. One must also

Figure 1. Communes ranked by inaccessibility and poverty.

Figure 2. Communes ranked by inaccessibility and economic potential.


consider geographical coverage within each ofthese levels. In making these choices one wantsto assure a cost-effective use of resources, giventhe overall objective of reducing poverty.A number of issues arise in addressing this

appraisal problem. First, how can we measureexpected benefits—accounting for factors thatcan be expected to influence the benefit stream,but also attaching values to each factor to allowa ranking of all potential investment projects?Existing road attributes (e.g., road density inthe area), as well as commune and populationcharacteristics (e.g., human resource develop-ment; presence of development projects andcomplementary infrastructure) will clearly in-fluence the social returns from an investment.But they are likely to do so in differing degrees. 9

This suggests that some kind of weightingscheme must be designed to reflect each factor’ssignificance. How are such weights to be deter-mined?How are distributional goals to be incor-

porated? All else equal, preference shouldclearly be given to poorer beneficiaries of theroad investments. But how should ‘‘poverty’’be measured? In practice, data availability andcomparability across the potential roads’ zonesof influence, are likely to be the decisive factor.A further concern relates to what tradeoffs willbe accepted between reaching the poor andother objectives, such as traffic volume.Another question concerns the ability to ap-

praise centrally all potential road links—thatcould run to tens of thousands—individually. Isit acceptable to rely on a limited number of

‘‘representative’’ road link appraisals and toextrapolate to other areas? Do alternatives tothis common solution exist? Finally, there is noguarantee that the initial budget allocation wasoptimal. Can something be said about whethertoo much or too little is being allocated to thesector as a whole?The following section proposes one approach

to resolving these issues, within realistic infor-mation constraints.

7. THE PROPOSED APPROACH

A total budget C is available for rural roadinvestments. It is assumed that there are manyroad links that are potential candidates for theproject and that C is not sufficient to fund themall. The task of the appraisal is to provide aranking among these potential links by defininga selection formula that identifies places wherepoverty, inaccessibility and economic potentialare high. Ultimately, we want a ranking for-mula that can reflect tradeoffs between thesevariables and still be implementable.It is assumed that each road link has a set

of encompassing communities ðECÞ and thatbenefits are confined to those communities.Although this is unlikely to be the ideal way todefine a road’s zone of influence, it reflects apragmatic attempt to resolve the data collectionproblem. The approach takes advantage of thefact that data are often, and/or more easily,collected at the community level. For datareasons, too, inequality within the EC is

Figure 3. Communes ranked by economic potential and poverty.


ignored—all those within a given EC are trea-ted the same way. Road links here refer to anytype of road—whether provincial (state), dis-trict or commune.The benefit to a typical user of a proposed

link is estimated from data on existing physicalinfrastructure, human development, economicpotential of the region and other factors thatmay influence the marginal gains from a roadinvestment. One could then calculate totalbenefit (multiplying by the number of people inthe EC) and hence the benefit to cost ratio forthe link. But, this treats different users in dif-ferent ECs the same way, and so does not re-flect equity concerns. Instead, we want to givehigher weight to poor users. This is done byattaching a social weight to each EC, reflectinghow poor residents are on average. Thus a so-cially weighted benefit–cost ratio is created.Social welfare ðSW Þ is defined as

PSiBiNi,

where Si is the social (equity) value attached toa typical user of the ith link, taken to be theaverage person living in its EC; Bi denotes theefficiency gain per person for the ith link; andNi is the number of people in the ECs of the ithlink. (I will return to the problem of measuringSi and Bi.) Let Ci denote the total cost of re-habilitating the ith link (cost per unit lengthtimes length.)The problem is then to maximize SW subject

toP

Ci ¼ C. To find the best allocation, allpotential road links should first be ranked bythe benefit-to-cost ratio: SiBiNi=Ci. (If Bi ¼ Bfor all i, then the ranking is simply done bySiNi=Ci.) Arbitrary thresholds for differenttypes of roads, or for amounts set aside for thepoor, are not required. The same criteria areused for all road links.If a minimum pecuniary rate of return is also

stipulated then this is a further constraint thatmust be satisfied. If the minimum rate of returncondition is not satisfied, then one finds thefeasible project composition which achievesthat rate of return at maximum SW, or onespends less than the initial budget. If the con-straint is satisfied then that suggests a case forexpanding the initial budget. It is, however,important that in setting the minimum pecu-niary rate of return one takes account of non-pecuniary benefits (see below).

(a) Benefits and equity weights

The measure of benefit for the ith road linkðBiÞ is derived from the values of a series of

variables Xi which help to determine the aver-age benefit that can be derived from the roadinvestment within the link’s ECs. These includeattributes of the road and attributes of thepeople served. Some factors may lessen bene-fits, some may increase the stream of benefits.Careful thought needs to go into ensuring thatrelevant variables are accounted for as much aspossible. But, what is considered will ultimatelydepend on what can be measured at the en-compassing community level. Certain factorswill be of more consequence to the road bene-fits, as well as to overall project objectives, thanothers. For example, we may want to put ahigher weight on connectivity to the existingnetwork than to the state of the existing road.Hence a system of weights ðwjÞ needs to beestablished which reflects the relative impor-tance of each observed variable in X in thedetermination of eventual benefits. So Bi ¼P

j wjXij, whereP

wj ¼ 1. For each link i, theweighted values of the Xs are then added up toget a measure of the total expected benefit fromthe road link. This should then be expressed ona per capita basis.In a similar fashion, we can postulate that the

social weights Si are a weighted sum of thevalues taken by a vector of measurable vari-ables Zi describing the socio-economic condi-tions in the EC of the ith road link. The poorerthe average person in an EC, the higher thevalue of S. Thus, Si ¼

Pk vkZik , where vk is the

weight attached to the kth factor deemed rele-vant to the overall social weight, whereP

vk ¼ 1.An important issue is the scaling of B and S,

since this determines the overall importanceattached to equity versus efficiency (as mea-sured by the Bs). This is a value judgment. Oneway to decide the issue is to fix the ratio of themaximum B to the minimum B and similarlyfor S. Bi can be normalized to vary between 0and 100, say. Similarly the minimum S (for theleast poor EC) can be set to zero. The decisionon the maximum S (for the poorest EC) thendetermines the relative weight attributed toequity versus efficiency by the formula.Finally, the resulting measure of benefits is

divided by the estimated costs. The costs willvary according to the road type and the plan-ned work. The ratios are then used to rank allroad link investment proposals. The first dis-bursement from the budget goes to the linkwith the highest benefit–cost ratio. The nextgoes to the next highest ratio, and so on till thebudget is exhausted.


(b) Nonpecuniary benefits and minimum rates ofreturn

Given that valuation problems are likely tobe worse for certain projects, it cannot be op-timal to insist that all projects achieve the samerate of return as required for a public invest-ment with known (measured) benefits. Theapproach common in current practices in thesector is that a project must either achieve acertain return (constant across all types ofprojects) or it is taken completely outside thenormal evaluation system. This can hardly bethe best solution. In reality, for all types ofprojects we are able to measure some benefitsreasonably straightforwardly but not others,and the extent of the valuation problem is nodoubt greater for some types of projects thanothers.Instead of putting certain projects outside the

evaluation process, it may be better to explicitlyacknowledge the problem of nonpecuniary re-turns by setting lower target monetary ratesof return for certain types of projects, reflect-ing our best guess of the value of unmeasuredbenefits. By this approach, decision makerswould have to set benchmark estimates of themagnitude of average nonpecuniary benefits foreach type of project; this can be done in theform of the appropriate ‘‘discounts’’ on thepecuniary rate of return. For example, if it isbelieved that only two-thirds of the benefits arebeing captured in the existing CB calculationsfor rural roads and the overall minimum rate ofreturn is 12% (when all benefits are monetaryand observed) then the minimum monetary rateof return on rural roads should be 8%.There will of course be ample scope for de-

bate on how to set these discounts to allow forunmeasured benefits. While some guidancemight be found in existing research, there willbe considerable uncertainty around any esti-mate. But it is arguably better to directly con-front this problem, and set explicit ‘‘best guess’’estimates rather than putting important classesof development projects outside the evaluationsystem, such that we have little or no idea if weare investing too much or too little in thesetypes of projects. Looking ahead, the need tomake explicit allowances for nonpecuniarybenefits will no doubt stimulate research toprovide better information on these benefits inthe future. (The current practice provides littleencouragement to improve valuation data andmethods for the types of projects that aredropped from the set to which a minimum rate

of return is applied.) Proper, careful evalua-tion based on the latest evaluation best prac-tices, that allows for endogenous placement or,where possible, uses experimental methods,could greatly improve our knowledge aboutthese benefits. More explicitly recognizing ourlack of knowledge in this area will add impetustoward resolving the issue in a believable waytaking proper account of biases such as due tothe nonrandomness of program placement (seeTable 1). This means setting up focused andrigorous research projects that aim to coverenough project types to provide an idea onvarious nonpecuniary benefits.A number of judgments will need to be made

to implement the above approach, notablyin setting the various weights (including theoverall weight on equity versus efficiency). Thenext section suggests how well-informed judg-ments, consistent with social values in eachsetting, might be formed in practice.

8. PUTTING THE APPROACH INTOPRACTICE 10

The following gives a step by step exampleof possible implementation in Vietnam. Obvi-ously, it is important to be flexible and allowfor institutional and other local constraintsin implementation. The approach needs to bepiloted, revised after a first cycle and alteredin the light of experience. All players must bewilling to accept set-up costs including the timenecessary for data collection and analysis, aswell as for all project proposals to be made.A fixed budget is available for the rehabili-

tation of rural road links. All provinces (cov-ered by the project) compete for this budget.This will create incentives for the provinces tocome up with efficient cost proposals, withinthe bounds of the construction standards setby the project. The selection formula’s spe-cific variables and their weights are devisedby the project team in collaboration with thegovernment, following an extensive consulta-tive process. The idea is to then decentralize theformula to the provinces that will be responsi-ble for making proposals and bid for themoney. The steps are as follows:Step 1. Availability of data at the com-

mune or district level, and consultations witha wide array of data, poverty, rural develop-ment and infrastructure specialists in govern-ment and academia, allows the Bank and thetransport ministry teams to delineate a set of


Xs—encompassing commune and road char-acteristics—that must be taken into accountin estimating expected efficiency benefits. 11 Apotential list of the variables that determineefficiency gains might include the following:road density in area; local human resourcedevelopment (as measured, say, by percent-age of children completing primary school);other (complementary) development projects inarea; accessibility to social service facilities;accessibility to other forms of transport (train,waterways); agricultural development potential(as measured, say, by unused land with agri-cultural potential); current road condition;linkages with the existing road network.Step 2. Next the scale and key variables de-

termining the imputed social value of the ben-efits from a link must be determined. Givenproject objectives, the poverty level is an im-portant characteristic of ECs. Ideally, compa-rable commune-level poverty data would beavailable centrally. Data often exist at com-mune or district level but there is currently nosystem for compiling it nationally. Recent ad-vances in combining census and household-level survey data to estimate poverty levels atvery disaggregated regional level are promisingin this respect (Minot & Baulch, 2001). In themeantime, one possibility is to rely on theprovinces to come up with an internal povertyranking of all their communes using a reason-able and widely accepted indicator of welfare.This should be done outside the provincial de-partments of transport, for example, by theMinistry of Labor, Invalids and Social Affairs’provincial office (DOLISA), or the authoritiesresponsible for the provincial implementationof the national ‘‘Hunger Elimination and Pov-erty Reduction’’ (HEPR) program.If a single poverty measure does not exist, the

ranking can be based on a composite index ofavailable variables ðZÞ that influence Si. Thecontent and scale of Si must be identical acrossall provinces and command wide acceptance.For example, it might include one or more ofthe following: infant mortality rates (IMR),average incomes, literacy, share of school-agechildren attending secondary school, undernu-trition, etc. Since such indicators are typicallyexpressed in different units, a different scalemust be determined so that the numbers can beadded up (note that this applies also to the Xs).Most communes report such data to the dis-trict. The province authorities should then beable to collect the information from each oftheir districts.

Step 3. To determine the weights on thevariables in B and S, and the value of the scalefor the poorest EC relative to the least poor,and hence the scaling of equity versus efficiencyconcerns, a multidisciplinary group of govern-ment and nongovernment Vietnamese expertscan be set up. The Bank team can design andparticipate in focus group meetings whence theweights can be defined by consultative process.Separate meetings may be needed for S and B.In the Vietnam context, it would be highlydesirable to bring together knowledgeablerepresentatives from the Women’s Union, theFarmer’s Union, the Commission for EthnicMinorities and Mountainous Areas, Ministriescovering transport, health, education, agricul-ture and rural development; and academicsspecializing in ethnic minority and genderissues, health, education, poverty, rural devel-opment and transport. By relying on a com-mission of local experts, it is expected that themeasurement of benefits will adequately reflectsocietal values.Step 4. A technical assistance team should be

provided to each province for a certain amountof time to explain the rules of the game, helpmake project plans and conduct consultationswith communities, and comment on the shelf ofpossible projects. It will also explain that vali-dation checks will be made.Step 5. All provinces produce proposals. The

methodology is applied to all types of roads.The provinces must carefully weigh the costs ofspot repairs, versus rehabilitation, versus fullupgrading in making their proposals. Theprovinces are also required to carry out con-sultations in the communities of potentialsubprojects prior to making their bids. This isto ensure responsiveness to local needs and thatthe views of prospective beneficiaries are takeninto account (concerning choices in levels ofroad upgrade and in setting priorities betweensubprojects within an intervention area). Thisshould be done with the help of impartialgroups, such as the Women’s Union (which isactive throughout the country) or local bran-ches of national NGOs, that are independentfrom the provincial department of transportofficials and can be trusted to represent theviews of local communities. Each provincedraws up a list of benefits and costs for all roadlinks put forward as potential subprojects. Theprocess should allow for proposals that includemore than one road link, and possibly com-binations of different levels of road links.For example, a benefit-to-cost ratio calculation


could be based on a network of two or morecontiguous links where it is persuasively arguedthat the benefits from one link are considerablyhigher if the other link is also rehabilitated. Thetechnical team would be responsible for ex-plaining this, ensuring that consultations withcommunities and all prospective beneficiariesare satisfactorily undertaken and results re-flected in the proposals, and for generally ex-tending assistance to the provincial teams. 12

Step 6. Since the formula is fully decentral-ized it may be desirable to introduce additionalincentives to play according to the rules. Vali-dation of the province assessments of numberscan be made on a random basis. A provincethat is found to have cheated is punished.Punishment can consist of being thrown out ofthe game, or face some appropriate penaltysuch as a tax on its costs.Step 7. The money is allocated to provinces

according to the lists. The first unit of moneygoes to the highest benefit for cost ‘‘subpro-ject,’’ the second to the next, and so on. Onepotential issue is that of the cross-provincefunding distribution. It is conceivable thatthe best projects will be concentrated in afew provinces. If this is a concern, there are anumber of ways to prevent this eventuality. Itmight be specified that the second pot of moneymust go to a different province from the first,and so forth, to avoid all the money ending upin only a few places. Or it may be decided thateach province originally selected to participatemust get a minimum of the total, (say 1/60th inthe case where 30 provinces are participating).Alternatively, a formula could be determinedby which one-half of the entire budget is allo-cated in proportion to province populationsize, or population and a provincial index ofinaccessibility and poverty, leaving the rest tobe allocated according to where the most cost-effective links are proposed. Either way, themoney is still allocated according to the lists ofrankings. If the minimum allocation has beenreached for each province, we stop. If not, thenwe will need to go back to the list and gothrough a process whereby the last chosen linkis dropped and (unless it is located in the un-derfunded province) replaced by the link withthe highest CB ratio from the underfundedprovince, and so on.Step 8. For a representative project within

each of the main road types, a conventionalinternal rate of return calculation is made basedon producer and consumer surpluses. This isused to estimate the overall rate of return to the

set of subprojects selected up to Step 7. Aminimum return that is adjusted for the current‘‘best guess’’ of the expected nonpecuniarybenefit levels is determined. If the average rateof return is at or above the minimum then onestops. If, however, it is below the minimum,then one has to substitute projects that had notpreviously been included for some that had.Thus substitution should be made so as to as-sure the least cost in terms of the more com-prehensive measure of benefits used in selectingprojects. The project with the lowest benefit–cost ratio in the road type category with thelowest rate of return should be dropped andreplaced by the project with the highest ratioamong those previously rejected. This con-tinues until the minimum rate of return isreached.

9. CONCLUDING COMMENTS

Measuring the benefits of rural roads isfraught with difficulty. Special selection andappraisal criteria for rural roads have evolvedthat simply assume important social benefits,despite a general lack of rigorous empiricalevidence. These are used as justification forabandoning economic analysis when, as is thecase in many rural areas of developing coun-tries, traffic levels are too low for conventionalconsumer surplus measures to make sense. Thispaper has argued, however, that efforts con-currently need to be made to improve on themethods that are currently being used to ap-portion budgets on rural road projects. Thepaper proposes such an improvement. Thisneeds to be backed up by serious research onthe impacts of rural roads on living standards,broadly defined.The approach proposed here builds on a

number of past approaches, observations andproject experience. The aim has been to focusthe discussion back squarely on the objective ofpoverty reduction, but doing so within a publiceconomics framework in which efficiency andequity concerns are inseparable, information isincomplete in important ways, and resourcesare limited. The approach tries to avoid thetendency to treat budgets for so-called ‘‘socialobjectives’’ outside the realm of hard-nosedeconomic analysis, but also recognizing theconstraints faced in practice in implementingrigorous appraisal with limited information.The advantages of proceeding as outlined in

this proposal include that it holds the hope of


building capacity, and is participatory; it ex-tracts local information that may not be readilyavailable to the center, and it appears to befeasible through its reliance on the participa-tion of local authorities and residents in the

appraisal of subprojects. The method promisesto assure that the most effective investments areselected from the point of view of povertyreduction, given both the information andresource constraints.

NOTES

1. Hine (1982) provides a good discussion of the most

commonly used methods of estimating benefits.

2. Recent examples include road projects in Peru,

Zambia, Andhra Pradesh, and China (World Bank,

1995, 1996, 1997, 1998a, respectively).

3. For example, eligibility under the social criteria for a

project in Peru requires that infant mortality rates be

over 80, the index of unmet basic needs be above 70%

(the index is an area specific composite of sanitation

facilities, housing quality, educational attainments,

school enrolments, employment and dependency based

on census data) and that there be more than 100

beneficiaries per kilometer (World Bank, 1995).

4. A recent example is South Asia Region (1999).

5. For a critique of current methods and a discussion

of the broader issues in the China context see Huene-

mann (2001).

6. See van de Walle (1999) for a description of the

data-base—the Survey on Impact of Rural Roads in

Vietnam (SIRRV).

7. Hungry households are defined nationally as those

with the income per person equivalent of less than 13 kg

of rice per month. This is a popular and widely collected

statistic in Vietnam.

8. The lack of correlation between inaccessibility and

poverty does not, per se, throw any doubt on the case for

geographic targeting for poverty reduction (Bigman &

Fofack, 2000). It does suggest that the aim should be to

target high poverty areas, not areas with low accessi-

bility when aiming to reduce poverty. But, those high

poverty areas must also have bad access if the pol-

icy instrument is to be road or other transport interven-

tions.

9. Numerous studies have remarked on the key role of

complementary inputs and mediating variables in ex-

plaining the gains from a rural road investment. For

example, see Hine (1982) and Cook and Cook (1990).

10. This section elaborates on an actual example from

a recent World Bank project (World Bank, 1998b).

Some points have been developed further than in the

project for expository reasons.

11. A number of variables that help determine the

efficiency gains might also enter the equity weights,

possibly with the opposite sign! For example, there is

evidence of significant complementarities between phys-

ical and human infrastructure investments (for example,

van de Walle, forthcoming). Thus, it is likely that the

marginal benefits from a road project will be higher in

areas where education and health status are higher. On

the other hand, one might want to favor ECs with lower

human capital, and hence welfare, with a higher

distributional weight.

12. It should be noted that although not a topic of this

paper, further local participation of beneficiaries may be

crucial to project sustainability, ownership and future

maintenance. The latter is often judged to be key to a

rural road investment’s success.

REFERENCES

Antle, J. M. (1983). Infrastructure and aggregateagricultural productivity: International evidence.Economic Development and Cultural Change, 31(3),609–619.

Beenhakker, H., & Chammari, A. (1979). Identificationand appraisal of rural roads projects. World BankStaff Working Paper No. 362, Washington, DC:World Bank.

Bigman, D., & Fofack, H. (2000). Geographical targetingfor poverty alleviation: Methodology and applications.Washington, DC: World Bank Regional and SectorStudies.

Binswanger, H., Khandker, S., & Rosenzweig, M.(1993). How infrastructure and financial institutionsaffect agricultural output and investment in India.Journal of Development Economics, 41(2), 337–366.


Carnemark, C., Biderman, J., & Bovet, D. (1976). Theeconomic analysis of rural road projects. WorldBank Staff Working Paper No. 241. Washington,DC: World Bank.

Cook, P., & Cook, C. (1990). Methodological review ofanalyses of rural transportation impacts in develop-ing countries. Transportation Research Record, 1274,167–178.

Cornes, R. (1995). Measuring the distributional impactof public goods. In D. van de Walle, & K. Nead(Eds.), Public spending and the poor: Theory andevidence (pp. 69–90). Baltimore, MD: Johns HopkinsUniversity Press.

Fan, S., Hazel P., & Thorat, S. (1999). Linkages betweengovernment spending, growth and poverty in ruralIndia. Research Report 110. Washington, DC:International Food Policy Research Institute.

Gannon, C., & Liu, Z. (1997). Poverty and transport.TWU discussion papers, TWU-30. Washington, DC:World Bank.

Hine, J. L. (1982). Road planning for rural developmentin developing countries: A review of current practice.TRRL Laboratory Report 1046, Overseas Unit.Crowthorne, Berkshire, UK: Transport and RoadResearch Laboratory.

Huenemann, R. W. (2001). Cost–benefit analysis ofrural roads in poverty areas of Western China.unpublished, Faculty of business, University ofVictoria, Victoria, BC, Canada.

Jacoby, H. G. (2000). Access to markets and thebenefits of rural roads. The Economic Journal, 110,713–737.

Jalan, J., & Ravallion, M. (2002). Geographic povertytraps? A micro econometric model of consumptiongrowth in rural China. Journal of Applied Economet-rics.

Minot, N., & Baulch, B. (2001). The spatial distributionof poverty in Vietnam and the potential for targeting.Mimeo, Washington, DC: IFPRI.

South Asia Region. (1999). Regional work programagreement: socio-economic impact of rural accessimprovements, with emphasis on social impact andpoverty alleviation: a regional study with Bank-wideapplications. Unpublished, May 14.

van de Walle, D. (1999). Assessing the poverty impact ofrural road projects. Unpublished. Washington, DC:World Bank Development Research Group.

van de Walle, D. (forthcoming). Are returns to invest-ments lower for the poor? Human and physicalcapital interactions in rural Vietnam. Review ofDevelopment Economics.

van de Walle, D., & Gunewardena, D. (2001). Doesignoring heterogeneity in impacts distort projectappraisals? An experiment for irrigation in Vietnam.World Bank Economic Review, 15(1), 141–164.

World Bank. (1995). Staff appraisal report: Peru ruralroad rehabilitation and maintenance Project, ReportNo. 14939-PE. Washington, DC: World Bank.

World Bank. (1996). Staff appraisal report: China,Second Henan provincial highway project, ReportNo. 15358-CHA. Washington, DC: World Bank.

World Bank. (1997). Staff appraisal report: Republic ofZambia, Project to support a road sector investmentprogram, Report No. 16539-ZA. Washington, DC:World Bank.

World Bank. (1998a). Economic analysis for the ruralroad component: Andhra Pradesh economic reha-bilitation project, India. Washington, DC: Infra-structure South Asia Region, World Bank.

World Bank. (1998b). Vietnam rural transport projectII: Project concept document, September 4. Wash-ington, DC: World Bank.


Documents

1-s2.0-S0305750X01001279-main_3