POLICY RESEARCH WORKING PAPER 1550

Governance and Returns There is a strong statistical

link between a country's civil
on Investment liberties and the performance
of its aid-financed govern-
An Empirical Investigation ment investment projects. But
type of politicalregime
(whether authoritarian or
Jonathan Isbam democratic) and the status of
Daniel Kaufmann more purely political liberties
Lant Pritchett do not appearto significantly
affectprojectperformance.

TheWorldBank
PolicyResearchDepartment
PovertyandHumanResourcesDivision
November1995
VKIICY
RESEARItH WORKINM PAPER 1550

Summary findings
lUsing data frornithe World Bank's Operations Evaluation civil liberties records projects have an economic rate of
D)epartnmcilt,Ishalan,Kaufmann, and Pritchett examine return between 8 and 22 percentage points higher than
the link between the performance of Bank-financed the rate of return in countries with the worst civil
projects and varioLusinLdicatorsof coountrygovernance. liberties. (The average rate of return in the sample is 16
They find that: percent.)
T'here is a strong statistical, and possibly causal, link * The type of political regime (whether authoritarian
berveen civil liberties and project performance. After or democratic) and the status of more purely political
conltrolling for a variety of determiiinalntsof project liberties do not appear to significantly affect project
performance, they find that in countries with the best performance.

T'his paper -- a prodUct of the Poverty and Human Resources Division, Policy Research Department - is part of a larger
effort in the department to understand the donor and country determinants of aid effectiveness. The study was funded by
the Bank's Research Support Budget under the research project "Bank Project Effectiveness and Country Policy
Envirolnmeit" (RPO 679-49). Copies of this paper are available free from the World Bank, 1818 H StreetNW, Washington,
DC 2043.3. Please contact Sheila Fallon, room N8-030, telephone 202-473-8009, fax 202-522-1123, Internet address
sfallon@(aworldbank.org.(44 pages)

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about
deuelopmrpent issues. At obje ctive of the series is to get the findings out quickly, even if the presentations are less than fully polished. The
papers carry thc names of the authors and should be used and cited accordingly. The findings, interpretations, and conclusions are the
authors' ownr atnd should no?tbe attributed to the World Bank, its Executive Board of Directors, or any of its member countries.

Produced by the Policy Research Dissemination Center

Governanceandthe Returns to Investment:

An Empirical Investigation

Jonathan Isham

Daniel Kaufmann

Lant H. Pritchett
 Governance and the Returns to Investment:An Empirical Investigation'

Tntroductinn

The quality of governance is widely thought to be an important determinant of a country's

economic development (Brautigam 1992)2. However, to find convincing empirical evidence of

the effect of governance on economicperformance is a challenge. Definitional and measurement

issues plague both the evaluation of inputs--what is the appropriate measurement of good

govemance?--and of outputs--what is the effectiveness of government?3 . We are able to make

empirical progress in exploring the connection between the governance and economic performance

only by severely limiting our focus. We first isolate one observable indicator of economic

performance: the returns on investment projects of governments that were financed by the World

Bank. We then relate this performance indicator to just two limited dimensions of governance:

a) the degree of civil liberties; and b) the political regime type and the degree of political liberties.

Despite the serious definitional and measurementproblems with each of these indicators (discussed

Isham:the IRISCenter at the Universityof Maryland. Kaufmannand Pritchett:theWorldBank. We

would like to thank Deon Filmer and Phil Keefer for helpfulcomments. One caveat,althoughstandard,
deserves special mention given potentialsensitivityof the topic: all views presented in this paper are
exclusivelythose of the authors and do not necessarilyreflectthose of the World Bank.

2 For instance,the World Bank's policystatementon governanceand developmentsummarizes: 'Good

governance is central to creating and sustainingan environmnentwhich fosters strong and equitable
development,and it is an essentialcomponentto soundeconomicpolicies."(WorldBank1992).

3 Creating an objectivemeasure of the efficacyof governmentaction is plaguedby both normative

disagreementsabout the appropriateaims of governmentpolicy and positive disagreementsabout the
instrumentsempiricallylikely to achieveany givenaim. For example,a governmentmaybe very effective
(and even cost effectivein the limitedsense of achievinga givenobjective)at banningimports;yet this is a
policywhichmanymay regard as an ineffectiveand inefficient(in the broad sense)actionfor a government
to undertake. See Putnam(1993)for ingeniousattemptsto avoidthese problemsand measuregovernment
efficacy.
 as they arise) we feel that the data are of sufficient quality and interest to merit examination.

The first section describes the data (particularly the data on project performance),

establishesthe basic specification of the determinants of project performance excluding governance

variables, and introduces our classification of governance variables. The second section

establishes the positive link between the degree of civil liberties and project performance and

argues that this relationship is causal. The third section shows the lack of such a relationship

between either political liberties or political regime type. The fourth section discusses the

interpretation and implications of the empirical results. The conclusion, as is customary,

concludes.

II Data description, hasic specificrtinn and results

A Project dat

The data on the performance of projects is assembled by the World Bank's Operations

Evaluation Department (OED)4 . After e wchWorld Bank loan is fully disbursed--typically 5 to 8

years after the opening of the loan--Bank and borrower country staff write a Project Completion

Report (PCR) to assess project performance5 . As one part of this assessment, two performance

4 We do not differentiateprojects financedby the IBRD (InternationalBank for Reconstructionand

Development)loans, whichare relatively 'hard" loans at near commercialterms, and IDA (International
DevelopmentAssociation)credits, "soft' loans restricted to the poorest countries. Private sector IFC
(InternationalFinanceCorporation)loans are not includedin this analysis.

5 The PCR-recently rechristenedthe Implementation CompletionReport--isusuallywrittenby a staff

member in the divisionthat supervisedthe loan, but notby anyonewith major project responsibilities.As
such,theincentivesof thestaffto dissembleaboutprojectperformance,whilepresent,are not overwhelmingly
strong.

-2 -
indicators are created. For all projects, OED staff assign an overall performance indicator on

whether the project was 'satisfactory' or 'unsatisfactory' in achieving its development objectives.

For those projects in eight economic sub-sectors where the stream of project benefits can be

readily quantified and valued--infrastructure, agriculture, industry, energy, water, urban,

transport, and tourism--Bank project staff, sometimes in collaboration with OED staff, calculate

an ex post economic rate of return (ERR)6 .

The ERR is the discounted stream of project costs and benefits over the life of the project,

evaluated at economic (as opposed to financial) prices. Typically, it is calculated by roughly

following the methodology of Squire and van der Tak (1975)'. The ex post ERRs are typically

calculated about two to three years after project completion, in contrast to ex ante ERRs which

are calculated when the project is first assessed8 . Thus, at the time the ex post ERRs are

calculated, project evaluators know the actual investment costs and are somewhat better informed

about actual operating costs and demand; but they still must estimate most of the future stream of

benefits'.

6 Assessingadjustmentlendingis a whole other kettle of fish, a ketde whichhas been fried on several
occasionsbothbyi.heWorldBankand its staff (World Bank 1991, 1992, 1993;Pritchettand Summers 1993)
as well as othersless sympathetic.We excludeadjustmentoperationsfrom our universeof projects. For a
discussionof projectperformanceand adjustmentlending,see Ishamand Kaufmann(1992).

' Little and Mirlees(1992)discussthe degree of importanceof economic(versusfinancial)pricing in

WorldBank appraisalsand the qualityof cost-benefitanalysisoverall.

'There is an enormousgapbetweenthe ERR calculatedex anteand ex post (between6 and 10 percentage

pointson average)anda hugevariabilitybetweenthesetwomeasures (theR2 of regressingex post on ex ante
only about 0.2). The determinantsof this gap have been studiedbut are difficultto determine(Pohl and
Mihaljek, 1992).

9 Follow-upstudiestendto find that even the ex post ERRstendto overstatethe "true"rate of return as
in many cases, thebenefitflowsare not sustainedas long as anticipatedin the ex post ERR calculations.

-3-
 Table 1 shows the basic information about the ERR and overall ratings used in this

analysis, decomposedby region, from 1974 to 199310. The average rate of return was 16.1

percent. The ERRsvariedsubstantiallyacross regions, from nearly 18 percentin both South and

East Asia to only 14 percent in Sub-SabaranAfrica. About 73 percentof projectswere rated as

satisfactory, ranging from 83 percent in East Asia to only 64 percent in Africa.

Table 1: Summarystatisticsof project performance (1974-1993)

Region EconomicRate of return (ERR) Fraction of projects rated as

. _________________ "Satisfactory"

Region Average Number of Average Number of

l _____________ projects projects
All 16.1 1824 0.73 3435
South Asia 17.9 235 0.78 439
East Asia _ 17.7 340 0.83 588
EMENA 17.1 338 0.81 613
Latin America 15.5 364 0.70 701
Sub-Saharan 14.0 547 0.64 1094
A frica _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

Notes: Includesall projectsevaluatedby the World Bank's OperationsEvaluation

Departmentfrom 1974to 1993.
Source: Authors' calculationsfrom OED data.

R Rasic specificatinn and clacssification of independent variahles

In this paper, we are focusedonly on the relation between selectedaspects of

'1 An annualpublicationby OEDon evaluationresults (e.g. WorldBank 1993)uses this data to examine
project performanceby a numberof characteristics.

-4 -
 governanceand project performance. Nevertheless, we must still account for country

structuraland policy characteristicswhich, as previously shown (Isham and Kaufmann1995,

Kaufmannand Wang 1995)are determinantsof project success. Accordingly,for each

governancevariable tested, we estimatefour specificationswhich include variousdegreesof

control variables: a) country and structural characteristicsonly; b) these variablesplus

regional dummies; c) country and structural characteristicswith "policy" variables;and d)

1.
these variablesplus regional dummies"

By estimating and reportingeach of these specifications,we are able to explorepossible

endogeneityof policy variablesand governancevariables. The independentvariableshave

been classified as follows:

* Exogenousand/or structural variables (denoted Xs), including the country

capital/laborratio, termsof trade changes, and a dummy for project complexity12;

* Possibly endogenouspolicy and economic variables (denoted Zs) that could be

correlated with each other and/or with the governance variables, includingthe black

market premium, the fiscal deficit, and GDP growth;

* Regional dummiesfor South Asia, East Asia, Sub-SaharanAfrica, Latin America,

and Europe and the Middle East.

When only structuralvariablesare included (specificationsA and B) the estimateof the

" Wedid not includesectoraldummiesin the basecase althoughthereare differencesacrosssectors

(whetherthisisdueto realdifferences
or methodologies
is unclear). Theirinclusiondoesnotappearto affect
the resultsas countryportfoliocompositiondoesnot appearto be correlatedwith the otherdependent
variables.
12 FollowingIshamandKaufnann(1995),thisincludesall integratedrural development,
irrigationand
drainage,andlivestockprojects.
-5 -
partial impact of better governanceon project performancecould be overstatedbecauseof

omittedvariablesbias, conversely, when policy variables are included(specificationsC and

D), this estimate could understatethe true total impact of governanceif part of the impactof
3.
better governanceis through better policies"

The inclusionof the regional dummies, which are obviously exogenous,is simply a

robustnesstorture (for us) test. In order to be persuasivewe feel the results shouldsurvive the

introductionof regional fLxedeffects; otherwise, the results may simply be capturingsome

other unmeasuredhistorical, culturalor ideologicaleffect that covaries across regions and is

perhaps correlated with both project returns and the governancevariables.

C Basicresuits without governance variahles

Three econometric issuesrelated to the project evaluation data deservementionbefore

reporting the base case results. First, the ERRs are truncated from below as, accordingto

OED conventionthe lowest ERR reported is negative 5 percent. Thus, unless otherwisenoted,
14 . Second, we must match the time period of the
the reported results use a Tobit regression

3 If the set of equationsfor determiningthe rate of return on project with governance(G), exogenous

ERR, = 3*Gi+ 6*X] +cx*Z. + e,

Zj = y *Gi+Tl,

variables(X),and policyvariables(Z) is as above, thenthe pardal effect of governanceholdingall variables

constant (includingthose that normallywouldbe affected) is just P. The total effectof governance,which
includesthe effect of governancethroughits effect on policies(Z), is P+a *y.

" In somecircles,the Tobitregressionis out of favor becauseit is non-robust,as its parameterestimates

dependon theassumptionof normallydistributederror term. Since a relativelysmallfractionof this sample
is at the truncationpointof -5 percent, the Tobitestimatesare quite similarto simpleOLS(Greene1981).

-6 -
 time varyingvariables, such as black market premium or terms of trade, to the period relevant

to the dependent variables, the economic rate of return on projects. This is difficult, as the

projects are implementedover a long (on average seven years) and variable period and are

expected to yield benefits over an extendedperiod as well. While there are argumentsin favor

of various weights, we use a three year weighted average of the time varying variable, going

back from the year in which the project evaluationwas done. Third, although the projects

vary tremendouslyin total cost, from $1.7 million to $5.7 billion, we do not weight the

projects, nor adjust for heteroskedasticity,as the standard tests did not indicate any conditional

heteroskedasticityas a function of project size'5 .

The "base case" regressions--withoutany governancevariable--are presentedin table 2.

In table 2 and in the tables below, we report p-levels of the test whether the coefficientis zero

rather than the usual test (t or chi-square) statisticsthemselves. The p-level is the significance

level at which the null hypothesiscould be rejected. A p-level of less than 0.05 indicatesa

rejection of the null hypothesisat (at least) the 5 percent level. As noted above, specification

A is the regression of the project ERR on exogenousand structural variables (Xs) alone,

specificationB is the Xs with regional dummies, specificationC is the Xs -andother possibly

endogenouspolicy indicators (Zs), and specificationD is the full regression: Xs, Zs, and

regional dummies.

The results in table 2 are substantiallythe same as those of Isham and Kaufmann(1995)

and hence will not be discussedin any depth. Most of the results are intuitive: ERRsare

' Ina preliminary

versionof thispaperregression
resultswereweighted,butsubsequent
analysisrevealed
thattheresultswerenotaffectedby theweighting(andthat wehadincorrectlycomputedtheweights).
-7 -
lower with a larger capital-labor ratio, greater project complexity, a higher black market

premia, and a larger fiscal deficit. In addition, even after controlling for all these factors,

ERRs are substantially lower in Sub-Saharan Africa and modestly lower in Latin America and

EMENA16 .

16 The regionaldefinitionsare based on the (then) standardBank groupingsinto Latin America and
Caribbean,Sub-Saharan
Africa,SouthAsia, East Asia, and the somewhatmixedbag of Europe, MiddleEast
and NorthAfrica(EMENA).

-8 -
Table 2: Non-g vernance determinants of project returns (ERRs): 1974-1987
Specification:

A B C D
Exogenous Exogenous Exogenous Exogenous,
with region and policy policy, and
durnmies region
dummies
Exogenous
ln(capital/labor) -1.28 -1.73 -1.34 -1.39
(.029)a (.050) (.024) (.122)
Dummy for project -5.80 . -5.81 -4.99 -5.06
complexity (.0001) (.0001) (.0003) (.0002)
Terms of trade 0.086 0.096 0.064 0.077
shock (.278) (.218) (.417) (.316)
Policy
Black market -0.046 -0.040
premia (.0001) (.0001)
Fiscal deficit 0.178 0.229
(.194) (.116)
GDP growth 0.233 0.056
(.272) (.799)
Regional Dummiesb
East Asia -0.61 -2.69
_________________ (.791) (.246)
Latin America -3.85 -4.90
(.140) (.065)
EMENA -6.13 -5.51
(.036) (.071)
Sub-Saharan Africa -8.54 -9.12
(.0001) (.0001)
Notes: a) p-levels in parenthesis, b) based on World Bank regional classifications. Sample
size = 761. Source: Authors' calculations

-9-
 Since we focus on the impact of governance variables, we will (as a presentational matter)

only report the coefficients of the various governance indicators when added to these four base

specificationsrather than repeat all the results for each control variable for each regression. Little

of substance is lost in not repeating these results: none of the coefficients on any of the variables

above--includingthe black market premia and fiscal deficit--change dramatically or interestingly

with the inclusion of any of the governance indicators17 .

D. Measuring "governance"

Discussions of "governance" typically generate more rhetorical heat than empirical light

as politics, like religion, is a topic where beliefs are strong and reliable empirical measurement

is difficult. Even a consensus on definitions is elusive: what does one mean by "governance" or

18 . There is probably agreement on governance in the extremes. A

especially "goodgovernance"?

stable democracy where basic prescriptive human rights are honored and with a competent and

honest civil service would no doubt receive the label "good governance" from most observers; by

contrast, an unstable autocracywhere human rights are abused and with a demoralized and corrupt

civil service would receive the label "bad governance". There is probably also agreement that the

'' One finaleconometricnote:in order not to sacrificetoo many observationswhenthe non-governance

controlvariablesare missing,we imputedthe meanvaluefor missingobservations on the fiscaldeficitand then
includedin the estimationa dummyvariableinteractedwith thesevariablesfor thoseobservationsfor which
the data are imputed.At leastin the case where the independentvariablesare uncorrelated,this procedure
producesconsistent estimatesfor thevariableswith imputedmissings (bypackingthe impactof measurement
error dueto imputation ontotheinteractiveterm) and at leastpotentiallyimprovesthe efficiencyof estimation
of the governancevariablesby notthrowingout any observationswith dataon governancebecauseof missing
valuesof the controlvariables.

IS The WorldBank'spolicypaper defines governanceas "the mannerin whichpower is exercisedin the

managementof a country'seconomicand social resourcesfor development."(WorldBank1992).

- 10 -
quality of governance can be evaluated along at least the following three dimensions:

accountability (including legitimacy, institutional pluralism and participation), openness and

transparency,and predictabilityand the rule of law (Brautigam, 1992).

However,even thoughthe many dimensionsof "good governance"typicallycovary (e.g.,

political stability, transparentrule of law, competent administration,democracyand respect for

human rights), they are conceptually,logically, and empirically distinct. There are clearly

examples of effective but non-democratic governments (e.g. some East Asian countries),

democratic but corrupt governments (e.g. some South Asian countries), and democratic

governments which abuse human rights (e.g. some Latin Americancountries). As an extreme

example,the UnitedStateswas inarguablya democracyin the ante-bellumperiod, yet it tolerated

the most egregious of humanrights violations: slavery.

In spite--andbecause--ofall thesedefinitionaldifficulties, we isolatefor analysisonly two

of the many possible elementsof governance: the extension of civil liberties; and the extension

of politicalliberties and type of politicalregime. As noted, these are regrettablybut necessarily

narrowaspectsof governance.We are ignoring a number of other potentiallyimportantfactors:

we do not treat governmentalpolicy-making(Haggard and Webb 1994), nor directlymeasure the

degreeof governmentaccountability(Paul 1992, 1994), nor addresspoliticalinstability(Alesina

and Perroti 1993), nor use otherindicators of govemrnentperformance, such as those generated

by privaterating servicesfor foreigninvestors(Mauro 1995, Knack and Keefer 1994) . We have

chosenthe civil and politicaldimensionsfor three reasons: they are at least plausiblyquantifiable,

andempiricalmeasuresdo exist;they (albeitimperfectly)span the three dimensionslisted above--

accountability, openness, and the rule of law; and these same indicators of civil and political

- 11 -
liberties are increasingly being tested in cross-country growth regressions.

TI) Civil liberties and1projectperformance

A) Indicatorsof civil liberties

Developing a meaningful cross-country indicator of civil liberties is obviously difficult.

There is little international consensus on liberties that ought to be permitted. Moreover, when

those liberties about which there might exist some consensus are suppressed, it is almost always

done surreptitiously. That said, several large efforts have attempted to rank countries by degree

of civil liberties. In this study, we use data from three of these efforts:

* Freedom House19 (1994 and previous years) has constructed a ranking of civil liberties

for 165 countries from 1972 to 1994. This ranking--on a seven point scale--is based upon

a fourteen item checklist of civil liberties20 ;

* Humana (1986) constructed an index of human rights achievement in 89 countries for

the year 1985. This index, on a scale of zero to 100 (actual range is 13 to 98) was based

upon the definition of human rights adopted by the General Assembly of the United

Nations in 1966 under the International Covenant on Civil and Political Rights;

19 This indexis more commonlyknownas the Gastilindex, after its originator(Gastil1987).

20 The fourteenitemsare: mediafree of censorship;open publicdiscussion;freedomof assemblyand

demonstration; freedomof politicalorganization;nondiscriminatoryrule of law in politicallyrelevantcases;
free from unjustifiedpoliticalterror; free trade unions and peasant organizations;free businessesand
cooperatives;freeprofessionalandotherprivateorganizations;free religiousinstitutions;personalsocialrights
(e.g. property, internal and externaltravel); socioeconomicrights; freedom from gross socioeconomic
inequality;and freedomfrom grossgovernmentindifferenceor corruption.

- 12 -
 * Coppedge and Reinicke (1990) constructed five series on upolyarchy"22 in 170 countries

for the year 1985. We use "media pluralism" and 'freedom to organize' as indicators of

two fundamental civil liberties.

While each of these indices of civil liberties are subjective and debatable, their cross

correlationsare reasonably high, which creates some confidence that they measure the same thing

and do so reasonably well (althoughwe note that, since Coppedge and Reinicke used the Freedom

House and Humana studies in their own ranking procedure, part of the high correlation between

'
"The set of institutionalarrangementsthat permits public oppositionand establishesthe right to
participatein politics" (Coppedgeand Reinicke1990). The other series are 'fair elections","extensionof
suffrage",and "freedomof expression".We do notreport the resultson these series becausetheyeitherhave
lessvarianceand/orcapturepoliticafreedoms.Consistentwith reports reportedbelow on politicalindicators,
they are in no case statisticallysignificantacrossspecifications.

- 13 -
 the lattertwo and former two series is by construction)3. However, we wish to stress that by

usingtheserankingsneitherthe authors-nor especiallythe World Bank--placeany importanceon

the rankingfor any particularcountry. Moreover, by using these numbers strictly for our cross-

sectionaleconometricexercise, we do not imply acceptanceof the numbers nor any commenton

the politics, rights, or practices of any country.

R) Civil fiherties and prnject performance

Table 3 shows the results of including each of the measures of civil liberties in the project

performance regressions. There is a consistent, statisdcally significant and empirically large effect

23 Thecorrelationsamongstthe 'civil liberties"measuresare:

FH Civil(79-86) Humana CR Organize
Humana 0.83
CR Organize 0.78 0.68
CR Media 0.81 0.79 0.82

24 If differentobservationson 'civil liberties," say A (FreedomHouse) and B (Humana)differ onlyby

independent measurement error then the correlation between the two measures is
02
_ _ _ _ _ _ _ _ 2 2 2
p = X ,where a(. is the variance of the "true' variable and is the

measurement error variance for measurement A(B). If the measurement error variance is equal
a 5 thena correlationof .8 impliesthe ratio of measurementerror (noise)to the varianceof

(2
the true variable(signal), g is about .2.

- 14-
of civil liberties on the return to projects25. Taking the estimates from specificationD, if the

FreedomHousecivillibertiesindexwere to improvefrom the worst (1) to the best (7, as in Costa

Rica for all evaluated years), the ERR is predicted to increase by 7.5 percentage points.

Similarly, with the estimatesusing the Humana index, improvingfrom the worst civil liberties

(13)to one of the best (91, as in CostaRica) would improvethe ERR by 22.5 percentage points.

Since these civil liberties indices are on a different scale, a more standard method for

comparisonis to calculatehow much the ERR is predictedto increaseif each index were improved

by one standarddeviation(column5). An increaseof this magnitudein the Freedom House index

wouldraise the predictedERRby 1.6 points; a similar increasein the Humana index would raise

the ERR by 5.2 points; a standard deviation increase in "mediapluralism" would improve the

predicted ERR by 3.1 points.

5 For the FreedomHouseand the Coppedgeand Reinickeindicators,we havereversedthe scalesfor

comparability.Thus, for all indices,a highervalue representsmore liberties.

- 15 -
 Table 3: Impact of civil liberties on project rates of return
Specificationa: Effect of
one std.
dev.
A B C D increase on
________
_________ _______ ~~~ERR
d:

Freedom House 1.81 1.16 1.71 1.07 1.57

Civil (1978- (.0005)c (.079) (.002) (.114)
19 87 )b (N =649)
Humana (1982- .290 .299 .296 .289 5.19
1985) (.003) (.007) (.002) (.013)
(N=236)_
Media pluralism 4.61 4.45 3.66 3.43 3.12
(1983-1987) (.0001) (.002) (.001) (.026)
(N=448)
Freedom to 3.17 1.81 2.41 -0.26 2.70c
organize(1983- (.0001) (.184) (.006) (.854)
1987) (N =448) I ___
Notes: a) for descriptionof the specificationssee table 2; b) Annualvalues from 1978-
87. The other three indicesare single values extrapolatedto cover the listed time
period; c) p-levels in parenthesis;d) for the calculation in column V, the standard
deviations--forthe entire samplefor which each variable is available--are1.47, 17.97,
0.91 and 1.12 respectively(see appendixtable Al.1); e) uses estimatefrom column C.
Source: Authors' estimates.

Thisfindingof a positiverelationshipbetween civillibertiesandERRs is the central

positivefindingof our paper. We completethis sectionby showingthe statisticalrobustnessof

this finding;we arguein the followingsectionthat it is plausiblethat better civillibertiesactually

are a factor in producingbetter projects.

Onepossibleconcernwiththe econometricresultsaboveis that they are drivenby a few

outlyingobservations,as someprojectshave very high estimatedrates of return. We have dealt

with that problemin two ways. First, in additionto a Tobit specificationaccountingfor the lower

- 16 -
truncation,we truncatedthe ERRsabove at the more or less arbitrarylevelof a 50 percentrate of

return. Thistruncationdidnot affectthe results. Second, we in additionto OLS,we estimated

specificationD using quantile(median)regression,a techniquethat is morerobust to extreme

observations. Again,allthe civillibertiesvariablesthat were significantin specificationD in table

3 were statisticallysignificantusingmedian regressionestimates.

The results are qualitativelysimilar with the binary 'satisfactory/unsatisfactory"rating.

Using only this rating as the measureof project performance allows a larger sampleof

projects, as social sectorprojectsthat normally do not receive an ERR are rated by OED'2.

Table 4 reports the estimatesof a Probit regression for specificationsC and D (results for A

and B were similar). Naturally,since the binary indicator discards a great deal of statistical

information, these results are less precise: the p-levels are lower, and the estimatesfor the

Humana ranking are insignificant. For the other variables, the estimatesshow large increases

in the likelihood of a good project when implementedunder higher civil liberties. For

instance, at the mean of the FreedomHouse variable, a one standarddeviationincrease in civil

liberties would lower the probabilityof an unsatisfactoryproject by 3.2 percentagepoints,

which reducesthe predictedfailurerate by 16 percent (from the mean of 20 percent).

Similarly, a one standarddeviationimprovement in media pluralismwould reducethe failure

rate by almost 5 percentagepoints, or 25 percent from the mean7 .

26 See Kaufmannand Wang(1995)for a discussionof the performanceof socialsectorprojects as a

functionof macroeconomicpolicies.

2 The comparisonbetweenthe linear regressions and the Probit effects is complicated,but these
magnitudesare roughlysimilar. That is, if we created a dummyvariablefor 'unsatisfactory"based on the
ERR faiingbelowsomecriticallevel,thenthemarginalchangein probabilityof failurefromthe linear model
at a particularpoint (if the error term were normal) wouldbe PIo*o(.), where P is the slopecoefficient,o

- 17 -
 Table 4: Impact of civil liberties on the probability of a project being rated as
satisfactory,using Probit regression.
Specification:
C D
Freedom House Civil (1 9 7 8 _9 0 )b .018c .022
N= 1155 (.056)d (.060)
Humana (1982-86) -.00067 .0012
N=604 (.589) (.388)
Media pluralism (1983-90) .022 .054
N=740 - (.296) (.045)
Freedom to organize (1983-90) .042 .040
N =740 (.009) (.085)
Notes: a) for descriptionof the specificationssee table 2, b) Annual values from 1978-87
while for the other three indicesare single values for the listed time period; c) the value
reported is not the Probit coefficient,but the marginal change in the probabilityof a
successfulproject as the variablechanges, evaluated at the means of all independent
variables; d) p-levels of the test that the Probit coefficient is zero in parenthesis.
Source: Authors' calculations

C) Civil liberties andprojectperformance Disentangling cauIsality

This empirical relationshipbetweenperformance of projects and civil libertiesis

striking. Yet the interpretationof this partial correlation is problematic: it may well be that

some countryconditions causeboth greater civil liberties and better projects. We argue in two

ways that the results suggesta causativeeffect from better civil liberties to better project

performance. First, prior work suggeststhat beneficiary involvementand accountabilityof

is the error standarderror, and4() is thestandardnormalpdf evaluatedat a particularpoint. Takingthe

valuesfrom theregressionsin table3, themeansof ERR from table 1, and an estimateof a givesresults from
.0266=1.07/14.7*@((16.1-10)/14.7)(specificationD) and .042= 1.71/14.7*¢((16.1-10)/14.7)which are
slightlyhigherthan the estimatesof .018and .022 from Probit.

- 18 -
 implementingofficials are both key elementsof project success: both of these facets of

governancecould suggest a mechanismfor a causal link between civil libertiesand

performance. Second, we show that the data between manifestationsof civil strife and project

successshow a positive effect that is best understoodas a mediating mechanism.

1) Other literature civil liberties and performnance

Experience in developingcountiesilluminatestwo aspects of governancerelated to civil

libertiesthat affect project performance.

Beneficiaryinvolvementand participation. There is increasing consensusamongaid

agenciesthat input from and involvementof potential beneficiaries in governmentinvestment

projects is key to their success(e.g. World Bank 1995). Myriad case studies (e.g., Korten and

Siy 1988) illuminatethe mechanismsbetween beneficiaryparticipation and performance.

Isham, Narayan, and Pritchett (1995)use data from 121 water projects to show that greater

participationby potential beneficiariesdirectly causedbetter project performance. It is likely

that beneficiaryinvolvementin projects is greater in countries which score higher on a ranking

of civil liberties. At least six of the 14 elementsof the Freedom House civil rightsindex, for

example, are directly compatiblewith beneficiaryparticipationin projects: mediafree of

censorship,open public discussion, freedom of assemblyand demonstration,free trade unions,

peasant organizations,businessesor cooperatives,free professional or other private

organizations,and freedom from gross governmentindifferenceor corruption.

Public sector accountability. A related dimensionof the performanceof government

projects is the degree to which the public sector officials are held accountablefor their

- 19 -
performance. For an extreme instance, Dreze and Sen (1991), in their study of famines

elaborateon the fact that no major famine has ever happened in a country with a free press.

They postulatethat free flow of informationforces even non-democraticgovernmentsinto

actions to prevent economic catastrophessuch as famines. Empirical studies of accountability

are rare, but suggest that the degree to which public sector employees are responsibleis an

importantdimensionto performance(Wade 1994, Paul 1996).

Both greater beneficiary involvementand greater accountabilityof public sector

officials are facilitatedby an environmentin which basic civil liberties--suchas the freedom to

speak out and the ability of groups to organize to protect and advance their interests--are

recognized.

2) Civil unrest, civil liherties and project performance

Another intuitive argumentfor a causal mechanismbetween civil liberties and project

performanceis a chain that runs from civil liberties through indicators of civil unrest to project

perforrnance. The data suggestthat, controllingfor population, higher indicatorsof civil

strife, such as an increased numbersof riots, protest demonstrationsand strikes, is strongly

positively correlated with project performance. This finding at first seems paradoxical, but we

show that greater civil libertiesare associatedwith higher values of these civil strife indicators

and that, controllingfor the degree of civil liberties, there is little additional impacton project

performance.

Table 5 shows that countrieswith high average project performancealso have on

average much higher levels of civil unrest. When we sort countries into groups based in their

- 20 -
average ERR, we find countries in the 'high ERR' category had average rates of return twice

as high as those countries in the "low ERR" category. Interestingly, these high ERR countries

had many more riots, demonstrations and strikes per capita (adjusted for population effects)

than countries with poor project performance2 8 .

Table 5: Indicators of political and civil unrest by average ERR by country.

Indicators of political
ERR Average Number Regional Number unrest, averages per year
categorya ERR of distribution of by country (deviations
countries projects from population-adjusted
means):
Riots Demon- Strikes
strations

High 22.2 6 S. Asia: 3; 181 2.48 0.30 3.19

E Asia: 3
Medium 17 11 LAC: 5; SSA: 2; 253 0.00 0.16 -0.02
EMENA: 3;
S. Asia: i
Low 11.2 12 SSA: 9; LAC: 2; 209 -0.19 -0.04 -0.23
S. Asia: 1I___

Notes: a) ERR categories are determined by average rates of return classified by country for all
countries with at least 10 projects over the period from 1974-1987.
Source: Autifors' calculations

2' Thecivilunrestvariables(riots, protestdemonstrations,and strikes,)came as numberof incidentsper

country per year (Banks1979,updates). This meant that countrieswith larger populationshad a greater
absolutenumberof incidents. However, it did not seemright to simnply normalizeto per capita,as there is
plausiblysome increasingreturns to scale in civil unrest. Consequently,for each of the three variableswe
regressedtheabsolutenumberof incidentson population*ln(population) (which is equivalentto adjustingthe
per capitalevelfor thetotalpopulationin semi-logform) and reportthe residualof this regressionas 'excess'
civilunrestovertheamountexpectedfor a givenlevelof population.Thepopulationadjustmentwasalso very
significantand the R-squaredvaried from .02 (strikes) to .18 (riots). The results reportedbelow were
unchangedby usingother concavefunctionalforms in place of this semi-logform.

- 21 -
 That greater civil unrest is associated with better projects appearsat first to be

puzzling. Typically, suchmanifestationsare thoughtto be associatedwith worse performance.

However, all the projectsin this analysis are financed by governmentsand, unlike in the

private sector, civil unrest does not create the same kind of risks for performance. By

contrast, there may be a channel whereby civil tension leads to better project choice and

implementation. While markets for private goods rely on informationfrom consumers

(expressedin the form of the aggregation of individualpurchase decisionsmade in the

market), governmentsmust rely on other channels for expressionsof citizen's preferencesand

for the monitoringof the performanceof government agents in carrying out their functions It

may be that with more open channels, all forms of expressionof popular will--includingcivil

unrest--are greater.

What is the relationshipbetween the indices of civil libertiesand riots, strikes, and

demonstrations? Table 6 shows that for all indicators except for Humana, greater civil

liberties are stronglyassociatedwith greater degrees of civil unrest.

- 22 -
 Table 6: Correlationsbetweenindices of civil liberties and civil unrest
Civil unrest
(all variablesadjusted for populationeffects)'
Civil liberties
indicator Riots Demonstrations Strikes

Freedom House .27 .17 .34

(Civil)b (.00001) (.0001) (.0001)l
(1978-87) l l
Humana .06 -.01 .22
(1982-85) (.34) (.87) (.0002)
Media Pluralism .14 .24 .29
(1983-87) (.0011) (.0001) (.0001)
Freedom to .30 .29 .36
organize (.0001) (.0001) (.0001)
(1983-87) ___

Notes: a) indicatorsof civil unrestper capita adjusted for total population

size as describedin footnote28; b) Annual values from 1978-87. The
other three indices are singlevalues extrapolated to cover the listed time
period; c) p-levels in parenthesis.
Source: Authors' calculations

Table 7 uses the same regressionbase specification as above and showsthat there is, if

anything, a modest positiveeffect of various indicators of civil unrest on projectreturns. In

both of the specificationswithoutregional dummies, the number of riots, protest

demonstrations,and politicalstrikes are positively and significantlyrelatedto the rate of

return. Projects apparentlydo better in environments with greater civil strife when civil

liberties are not includedas a determinant(e.g. a coefficient of 0.56 in specificationA for

- 23 -
riots). However, with the additionof any of the indicators of the degree of civil liberties29 ,

the impact of political manifestationsis reduced in magnitude. For instancethe coefficient on

riots falls from 0.56 to 0.32 in specificationA. That is, for any given level of civil liberties,

neither riots nor strikesare associatedwith better performance, but protest demonstrations

seem to still have some independentaffect.

The results support a chain of causation that runs from greater civil libertiesto higher

levels of the citizen's involvement--includingcivil manifestations--andto better projects. This

is not to suggest that civil unrest is itself the mechanism: it is more likely that environmentsin

which civil unrest is possibleare also those in which other mechanismsfor expressionof

popular (dis)contentwith governmentperformances are available and effective.

9 Only the FreedomHousecivil indicatoris shown in table 8, but the resultsfor the other three are
similar.

- 24 -
Table 7: Indicators of civil unrest and project returns, without and with
controls for civil liberties.
Specificationa:
A C D
| with
without with without with Freedom
Freedom Freedom House
House House
Adding just riots
Riots 0.56 0.32 0.42 0.21 -0.34
(0.062) (0.148) (0.040) (0.34) (0.245)
Freedom House 1.48 - 1.51 1.19
Civil (0.090) (0.093) l (.083)
Adding just protest demonstrations
Protest 1.04 0.88 0.81 0.68 0.17
demonstrations (0.0001) (0.014) (0.003) (0.013) (0.607)
Freedom House - 1.46 - 1.48 1.08
Civil (0.053) (0.006) (.112)
Adding just political strikes
Political strikes 1.58 0.201 1.67 0.45 -0.81
(0.127) (0.857) (0.097) (0.683) (0.520)
Freedom House 1.77 1.61 1.09
Civil (0.002) (0.006 I 09)
Adding all three civil strife variables
F-testcfor all three 4.39 3.66 2.69 2.30 1.09
indicatorswithout (0.004) (0.012) (0.045) (0.076) (0.352)
and with civil l
liberties _ l l l __ _
Notes: a) for a description of the specificationssee table 2; b) p-levels in
parenthesis;c) F-tests calculatedwith and without all three indicators. Samplesize
= 649.
Source: Authors' calculations.

- 25 -
 TV) Political regime tye political liberties and project performance

Civil and political liberties are undoubtedly associated with each other and with

democratically-elected governments. Yet there is a clear analytical distinction among these

two types of liberties and the type of political regime: for example, the degree of civil and

political liberties varies widely among non-democracies. Therefore, finding that more civil

liberties are associated with better ERRs does not imply that different types of political regimes

are associated with better performance. To try to disentangle these relationships, we test for

the possible association between ERRs and these two related aspects of governance: political

liberties and political regime type.

A) Political liberties

The most widely used measure of political liberties is an index also published by

Freedom House, based on 11 indicators of political rights3 0 . Like the ranking of civil

liberties, it is a subjective ranking from 1 to 731 Another possible measure of political

liberties is the index of human rights violations (IHRV) constructed by Pourgerami (1988),

based on reports of Amnesty International on the extent of human rights abuses for the years

30 The elevenare: chief authorityrecentlyelectedby a meaningfulprocess; legislaturerecentlyelected

by a meaningfulprocess; fair electionlaws; fair reflectionof voter preference in distributionof power;
multiplepoliticalparties; recent shiftsin power throughelections;significantoppositionvote; freedomfrom
dominationby the military, foreignpowers, and other powerfulgroups; no major group or groupsdenied
reasonable self-determination;decentralizedpoliticalpower; and informal consensus(de facto opposition
power).

31In manystudies(e.g., Helliwell1992),thesetwo closelycorrelatedindicesare summedto createone

overallindexof liberties.

- 26 -
1984-86. It takes a value of 4 for the least and 1 for the most human rights abuses31.

Table 8: Political libertiesand project performance

______
______Specifiation
2:_ _ _ _ _ _ _

_________ A B C D
Freedom House (Political)b 1.16 -.016 1.05 -. 115
(N=649) (.0087)c (.977) (.026) (.840)
Index of Human Rights Violations(IHRV) (N=425)
Dummy variablefor each level of violations (default is most violations, IHR' 1)
iIHRV=2 5.46 3.33 4.23 2.92
(.018) (.193) (.078) (.292)
IHRV=3 0.08 1.66 1.30 3.04
(.974) (.534) (.586) (.310)
IHRV=4 (least violations) 0.90 3.96 4.87 7.83
l _____________________ (.788) (.297) (.154) (.045)
Notes: a) for descriptionof the specifications see table 2; b) Annual values from 1978-
1987. IHRV is a single value extrapolated to cover 1980-1986;c) p-levels in parenthesis.
Source: Authors' calculations

The results reported in table 8 do not show any strikingpositive impact of purely

political rights on ERR. The Freedom House index shows a significantpositive effect in the

least demandingspecification(A), but whereas civil libertiesis robust, the political result is

not. It is driven out in significanceby the policy controlsand, in both magnitudeand

significance,by the inclusionof regional variables.

The index of human rights violations (IHRV) similarlydoes not show any clear pattern

of impact on projectperformance. Although in specificationD there is some significant

effect, the result is highly non-robust and appears to be driven by a few observationswith very

`2 Actually,thehighestpossiblevalue on the scale is 5, whichonly one country, Jamaica, achievesso for

our purposesJamaicais re-classified
to 4, the levelwhichincludesCostaRica, Venezuela,Mexico,Singapore
and Nigeriafrom 1984-86.

- 27 -
high ERRs33. Besidesthe possibledistinction between the top and the bottom, the results show

no clear pattern amongstlevelsof the index. In most of the specifications,whilecountries

with the second to worst amountof violations (IHRV = 2) do better than countrieswith the

most violations, they also do much better than countries with an even better record (e.g. 5.46

for RV=2 versus 0.08 for IHRV=3 in specification A) and about as well as those with the

best record (e.g., 3.33 for IHRV=2 versus 3.96 for IHRV=4 in specificationB). Therefore,

while there is some mild evidencethat it is better not to be in the worst group, there are no

discernibledistinctionsamongstthe other levels of IHRV 3 4 .

These weak results on the importance of civil liberties are the result of introducingthe

political variable into the base specificationwithout any indicatorof civil liberties. If we ask

what the effect of politicallibertiesare, conditional on the level of civil libertieswe find that

the civil liberties indicatorsretain all of their iInportancewhile the FreedomHouse political

liberties variable producesweak,or even results which suggesta greater level of political

liberties, holding civil libertiesconstant, worsens project performance. Table 9 shows that,

controllingfor civil libertiesraising political liberties actually reduces the ERR. While not too

much shouldbe read into these results (as the multicollinearityproblemsinvolvedwith the

estimationof the variablestogether are severe) these do strengthenthe interpretationthat civil

liberties are of primary importancein explaining project returns.

3 For instancein a linearregression(for simplicity),whilethecoefficient

is 1.96withTobitestimation
(inspecification
D), it fallstoonly-.027in a medianregression.Similarly,if theERRsaretruncatedabove
at 50 percentagepoints,theestimatefallsto 1.1 (witha p valueof .207).
3' Moreover, beingintheworstgroupby thisindicatormaysimplyindicatethatthecountryis involved
in a seriouscivilconflict,andhenceonewouldexpectdisruptionof projectsindependent
of anyexogenous
impactonliberties.
- 28 -
 Table 9: Estimatedimpact of political liberties on project returns (ERR)while
introducingcivil liberties variables.
Civil liberties variable included in specificationD
Nonea Freedom Humana Media Freedom to
House (Civil) (1982-85) Pluralism Organize
(1978-87) (1982-87) (1982-87)
Freedom House -.115 -2.35 -1.44 -1.17 .040
(Political) ( 8 4 0)b (.013) (.241) (.175) (.967)
Civil liberties 3.33 .365 4.53 .216
variable (.0033) (.006) (.009) (.906)
N 649 649 236 448 448
Notes: a) same regressionas table 8, b) p-levels in parenthesis.

R) Political regime type

The first indicatorof regime type, developed by the Center on InstitutionalRefonn

and the Informal Sector (IRIS)and labeled here as the 'IRIS Indicator of RegimeType'

(IIRT), places countries annuallyin one of five classes: democracy(IIRT=5), partial

democracy; transition regime, partial autocracy; and autocracy (IIRT= 1). Matchedwith our

project/policy data set, this includesannual observations from 1974-1987for 51 countries.

Since this indicator.of differentregime types is not (necessarily) a cardinalvariable, we

include a dummy variablefor each type. The second indicator, constructedby Alberto

Alesina et. al. (Alesinaet al. 1992) and labeled here the "AlesinaDemocracyIndex"(ADI), is

an annual ranking of countriesby democratic status on a scale of one (mostdemocratic)to

- 29 -
three. We use the annualobservationsfrom 1974-1982for 48 countries.

The results from includingthese either of these two measures of politicalregime type in

the base specificationsare emphaticallyambivalent(table 10). The lowest rankinggroup of

countriesby IIRT, the "autocracy",tends to have a lower ERR than other categories(as

evidencedby the positive signs for most of the others). In specificationsA and C (without

regional dunmmies),IRT categories3 and 5 (most democratic) tend to have higher returns,

while in specificationD none of the differencesare empirically or statisticallysignificant.

Using the ADI, we also find in specificationsA and C some weak evidencethat more

democraticregimes tend to have higher returns, but this is reversed by the inclusionof

regional controls.

Moreover, except for a comparisonbetween the "best" and "worst" regimetype, there

is no clear pattern to the results. For instance, the increment to returns over the autocraciesis

nearly as high when IIRT=3 as when IIRT=5, and countries with IIRT=4 (partial

democracy)are predicted to have an ERR lower by 3.4 percentage points than even

autocracies. The most democratic(ADI= 1) countries are also predicted in specificationD (at

a very low level of significance)to have lower returns than autocracies.

- 30 -
 Table 10: Returns and politicalregime type.
Specificationa:
A B IC D
IRIS Indicator of Regime Type (IIRT) (1 9 7 4 -1 98 7 b; N=725)
Dummy variable by regime t ipe (default is least democratic, IIRT= 1)
IIRT=2 -0.48 -0.56 0.24 0.49
(.767) (.733) (.885) (.766)
IIRT=3 5.46 1.61 3.52 1.04
(.107) (.636) (.296) (.757)
IIRT=4 -0.076 -3.68 -0.82 -3.35
(.967) (. 058) (.657) (.091)
IIRT=5 (most democratic) 3.94 0.322 4.17 0.458
(.055) (.892) (.043) (.847)
Alesina democracy index (ADI) ( 19 7 4 - 19 8 2 b; N=369)
Dummy variable by democ level, default is least democratic (ADI=3)
ADI=2 3.10 2.86 2.38 2.41
(.225) (.257) (0.349) (.345)
ADI= 1 (most democratic) 2.93 -1.27 1.93 -1.52
1 (.117) (.542) (.311) (.465)
Notes: a) for description of the variables includedin each specification see table 2; b)
Annual values for time period; c) p-levels in parenthesis.
Source: Authors' calculations.

In summary, there is no clear pattern that suggests that countries with more democratic

regimes (as classifiedby the availableindices)have better projects, as measuredby ERRs.

The very least democratic countriesperhaps do more poorly than others, but movingtowards a

democracyfrom other levels not have any empiricallydiscernible impact on project returns.

- 31 -
 C) Other liter2ntre political freedomand performance

These ambiguous findings roughly agree with other assessments of the degree of

political freedom on aggregate economic performance. There is a sizeable literature which

examines the impact of "democracy" on aggregate growth (e.g. Weede 1983, Scully 1988,

Helliwell 1992, Barro 1994, Bhalla 1994, see Alesina and Perotti, 1994 for a review), much

of which uses the Freedom House index of political liberties3 5 as the measure of democracy.

While it is quite difficult to reconcile the strikingly different findings of these papers, in spite

of the fact they use almost identical dependent variables (economic growth) and measures of

political freedom, we think a fair summary of the current state of the macro level literature on

economic growth and political freedom would be involve six findings.

First, higher levels of income are associated with higher levels of the Freedom House

index of political liberties. Second, when some covariates are added, the level of the Freedom

House index of political liberties does not have independent explanatory power for growth of

per capita income. Third, controlling for reverse causation--from economic growth (hence

higher levels of income) to political liberties--reduces the estimated effect of political liberties

on growth. Fourth, the effect of political liberties on growth seems to be non-linear, the

middle levels of the index (e.g East Asia) have higher levels of growth than either very low

(e.g. OECD) or very high (e.g. Africa) levels. Fifth, as with nearly all growth regressions, a

great deal depends on how the newly industrializing economies of East Asia are treated. This

3 Or the sumof the two FreedomHouse indicators. Two other studies-(Kormendiand Meguire1985,
Greir and Tullock 1989)-use dummiesbased on the FreedomHouse civil index in cross-countrygrowth
regressions.

- 32 -
is especiallyimnportant
for politicalquestions, as the East Asian countries were by and large

modestlyauthoritarianbut had very rapid growth.

V What do the ratesof return indicate?

We have establishedtwo results: higher levels of civil liberties are associatedwith

better performanceon World Bank-financedprojects, and purely political libertiesand the type

of politicalregime do not affect project performance. How should these findingsbe

interpreted?

So far, we have used project ERRs as an indicatorwithoutbeing specific about exactly

what they are an indicatorof. Since the World Bank financesonly a fraction of aid-financed

projects--andan even smaller fraction any particular government's investmentportfolio--it

cannot be assumedthat the ERRs of Bank projects are representativeof overallgovernment

returns. Do these resultsapply to all aid-financedprojects? Is performanceof government-

implementedprojects a proxy for overall governmentefficacy? More broadly, does the return

on these aid-fnanced governmentprojects reflect the return to all investments? We address

these questionsin the next three sections.

projectcune
A) Are IRank among aid project?9

There is little reason to believe that World Bank-financedprojects are chosen very

differentlythan projects of other donor agencies. Moreover, it is particularlyunlikelythat the

selectionprocess for projects at the World Bank is biased in such a way as to produce results

for a relationshipbetweenprojects and governance that would not apply to projects of other

- 33 -
donors.

R) Cinvernmentperfnrmrnice?

The more difficultand important questions is whether the ERRs of World Bank-

fmanced projects indicatea degree of the efficacy of governmentacross countries. There are

arguments for and againstthis supposition.

On the one hand, sinceall countries are treated alike by the World Bank in terms of

project selectionand fund availability,it is likely that systematicdifferencesin project returns

do reflect countryspecific,rather than Bank specific factors. The Bank is quite centralized

and the internal standardsfor project selection, appraisal, review, implementationand

3 6. In addition, in the determinationof lending
supervisionare uniform acrosscountries

allocation--especiallyof 'soft' IDA loans--there are pressures for an allocationacross countries

based on considerationsotherthan expected performance of projects. Finally, the World

Bank's Articles of Agreementhave always been interpreted in such a way as to prevent an

explicit considerationof politicalfactors so that project selectionat least in theory shouldhave

been exogenouswith respectto the variables we are considering. Together, these factors

suggestthat differencesin ERRs across countries probably do not reflect large differencesin

Bank treatmentnor in projectselection rather it is likely that thesedifferencesreflect real

differencesin governmentinvestmentreturns across countries.

On the other hand, this cannot be established. Countrieschoose whichof their possible

36Thiswasevenmoretrue in the periodprecedingthe reorganizationof 1987whichdecentralizedto the

regions manyformerlycentralfunctions.

- 34 -
set of projects to financethroughthe World Bank. This choice may involve'cream skimming"

--governmentsseekingBankfinancingfor projects with very high expectedERRs--or'laggard

offeringthe Bank the most problematicprojects. To the extent that

dumping"--governmnents

the mechanismwherebythe governmentand the Bank agree on whichprojectswill be fimanced

differs across countries, the average return across countries cannotbe interpretedas reflecting

governmentperformance.

The most worrisomefact against an interpretation of ERRs as an indicatorof

government efficacyis their lack of correlation with other measuresof governmentefficacy.

The BusinessEnvironmentalRisk Intelligence(BERI) and InternationalCountryRisk Guide

(ICRG) data rank countriesby various characteristics that indicatetheir attractivenessfor

foreign investment--e.g.,red tape encountered, corruption, and bureaucraticdelay. These

various measures are not significantlycorrelated with the ERRs in our data set: they are not

good measures of the samething. It is possible that these private sector ratings are flawed

indicators of governmenteffectiveness;they are designed for foreign investors,and

governmentsthat are not attractiveto foreign investors may be reasonablyeffectivein

implementingtheir own projects. Nevertheless, the lack of correlationof projectperformance

with other indicatorsof governmentefficacy does raise questions.

- 35 -
 Q Are projects retimq a
2proxy for overall returns tn invertment?

In earlier work on the effect of policy distortion variables--e.g., the blackmarket

premia--onproject performance(Isham and Kaufmann 1995), it has been arguedthat the ERRs

of World Bank-financedprojectsare reasonably representativeof overall investment

performance. How closelyrelated are overall returns to capital and these ERRs? This is

difficult to answer as the economy-widereturn to capital is almost never measured.

A simple growth accountingexercise can circumvent this measurementproblem.

Suppose that growth of outputper worker can be decomposedinto the growthof capitalper

worker and a residual as in equation 1:

y = ak*k

where lower case letters representper worker and the "dot" representsthe percentagetime rate

of growth. In a neoclassical(Solow)framework, the coefficient on capitalis just the share of

capital total output,

=r*K
ky

We typically do not observe r, the return on capital. However, we hypothesizethat r for each

country I varies systematicallyacross countries with the observed rate of return on projects,

ERR:

ri = r + P *ERRi

We can combine these three equationsand estimate the following equation:

- 36 -
 = r* * + *ERR*( f) *k

so that the first term identifiesthe average rate of return to capital and the secondparameter

the impact of an increase in ERR on the overall return to capital. The results of estimating

this equationusing cross nationaldata are presented in table 11. The estimatedaveragereturn

varies between 10.2 and 17.8 percent, a reasonable range of values. The impact of ERR on

the overall rate of return is between0.44 and 1.08 (with just two countries excluded). The

estimates suggest that the return on Bank projects, the ERR is related one-for-oneto the

economy-widerate of return which suggestsit is a very good proxy for overall investment

performance. This small piece of econometricevidence fortifies the argumentsin earlier work

for the use of the ERR as an indicator of the economy-widerate of return.

However, the multicollinearityproblems in identifyingthe interactiveterm are

extremelysevere. With just two variables, over 50 percent of the overall growthrate variance

is explainedand hence the F-tests of joint exclusion are overwhelming. However,the

estimates of each term individuallyare very imprecise and cannot reject zero (or any other

value for that matter) for the the individualterms. We are obviously not happywith this, but

see no solution.

- 37 -
 Table 11: Relationshipof overall returns to capital and the country averageof project
rates of return (ERR).
Unweighted Weightedby numberof
projectsa

Full Withoutoutliersb Full Withoutoutliersb

Return to capital .178 .102 .163 .106
(.107)c (.116) (.119) (.878)
Impact of ERR on .439 1.08 .626 1.06
return to capital (.750) (.821) (.705) (.722)
R-Squared .450 .507 .345d .506d

N 39 37 39 37
Notes: a) observationsare weightedby the square root of the number of projects, b) the two
outliers excludedare Syria and Pakistan, c) standard errors in parentheses,d) R-squaredof
the unweighteddependent variable.
Source: Authors' calculations.

Cnnclusimn

No one would pretendthat the degree of civil or political liberties or the choice of

politicalregime is--or ought to be--based on an assessmentof strictly economiccosts and

benefits. What is often meant by the definition of 'human rights" are exactlythose elements

of the interaction of human beingsthat go beyond any social welfare calculus.

That said, we have presentedempiricalevidence that, beyond their intrinsicmerit, civil

libertieshave direct instrumentalbenefits in improving the performance of a leasta subset of

governmentinvestmentprojects, those financed by the World Bank. We believe that this is an

additionalpiece of evidence for the view that increasingpublic voice and accountability--

- 38 -
through both participationand better governance--canlead to greater efficacyin government

action, includingdevelopmentassistance (Picciotto 1994; OECD 1995).

On the other hand, the empirical evidence does not provide evidenceof a relationship

between "democracy'and the particular performanceindicatorthat we examine(althoughwe

hasten to emphasizethe statisticalfact that lack of evidencefor is not necessarilyevidence

against). This merely suggestswith microeconomicdata that which is known from aggregate

data: while some authoritariangovermnentshave not providedeconomic benefits, other

countriesunder authoritarianregimes (particularlyin East Asia) have experiencedefficacious

governments,widespreadeconomic growth, and enormousreductionson poverty.

- 39 -
 References

Alesina, Alberto, Sule Ozler, Nouriel Roubini and Philip Swagel. 1992. "PoliticalInstability
and EconomicGrowth." NBER Working Paper No. 4173 (September). Cambridge,
Mass.: NationalBureau of Economic Research.

Alesina, Alberto and RobertoPerotti. 1993. "Income Distribution,Political Instability, and

Investment."NBERWorking Paper No. 4486 (October). Cambridge,Mass.: National
Bureau of EconomicResearch.

Alesina, Albertoand RobertoPerotti. 1994. 'The Political Economyof Growth: A Critical

Survey of the Recent Literature." Wnrld BRank
Fr2onomicReview 8(3): 351-373.

Banks, Arthur S. 1979(and subsequentupdates). 'Cross-NationalTime Series Data

Archive." Centerfor Social Analysis, State Universityof New York at Binghamton.

Barro, Robert. 1994. "Democracyand Growth." NBER WorkingPaper No. 4909 (October).
Cambridge,Mass.: National Bureau of EconomicResearch.

Bhalla, Surjit. 1994. "Freedomand Economic Growth: A Virtuous Cycle." Mimeo.

Brautigam, Deborah. 1992. 'Governance, Economy, and Foreign Aid." Studies in

ComparativeInternationalDevelopment (Fall) 27, 3: 3-25.

Coppedge, Michael and WolfgangH. Reinicke. 1990. "MeasuringPolyarchy." Studies in

ComparativeInternational Development (Spring) 25, 1: 51-72.

Dreze, Jean and AmartyaSen. 1989. Hunger and Public Action. New York: Oxford
UniversityPress

Gastil, Raymond. 1987. Freedomin the World: PoliticalRights and CivilLiberties. New
York: GreenwoodPress.

Grier, KevinB. and GordonTullock. 1989. "An EmpiricalAnalysis of Cross-National

EconomicGrowth, 1951-1980." Journal of MonetaryEconomics24, 259-276.

Greene, WilliamH. 1981, "On the Asymptotic Bias of the OrdinaryLeast Squares Estimator
of the Tobit Model."Econometrica, 49, 3: 505-513.

Haggard, Stephanand StevenB. Webb. 1994. Votingfor Reform: Democracy,

Liberalization,and EconomicAdjustment. New York: Oxford UniversityPress.

Helliwell,John. 1992. "EmpiricalLinkages Between Democracyand Growth," NBER

- 40 -
 Working Paper No. 4066 (May). Cambridge, Mass.: NationalBureauof Economic
Research.

Hunmana,Charles. 1986. WorldHuman Rights Guide. London: Hodder and Stoughton.

Isham, Jonathan and DanielKaufmann. 1992. "Does Policy BasedLendingAffect the

Productivityof InvestmentProjects?" Background paper for AdjustmentLending and
Mobilizationof Privateand Public Resourcesfor Growth. Washington:The World
Bank.

1995. "The Forgotten Rationale for Policy Reform" Background paper for
WorldDevelopmentReport 1991. Washington: The World Bank.

Isham, Jonathan, Deepa Narayanand Lant Pritchett. 1995. "Does ParticipationImprove

Performance:EstablishingCausality with SubjectiveData." 7he WorldBank Economic
Review 9(2):175-200

Kaufmann, Daniel and Yan Wang, 1995. "MacroeconomicPolicies and Project Performance
in the Social Sectors." WorldDevelopment 23(5):751-765.

Knack, Stephenand Philip Keefer. 1994. "Institutionsand EconomicPerformance:Cross-

Country Tests using AlternativeInstitutional Measures." Forthcomingin Economics
and Politics.

Kormendi and Meguire. 1985. "MacroeconomicDeterminantsof Growth." Journal of

Monetary Economics16:141-163.

Korten, Frances F. And Robert Siy, Jr., eds. 1988. TransformingBureaucracy:The

Experienceof the PhilippineNational Irrigation Administration. WestHartford,
Conn.: KumarianPress.

Little, I.M.D. and J.A. Mirrlees. 1991. "Project Appraisal and Planning TwentyYears On."
Proceedingsof the WorldBank Annual Conferenceon DevelopmentEconomics1990.
351-82.

Mauro, Paulo. 1995. "Corruption,Country Risk and Growth." QuartertyJournalof

Economics, forthcoming.

Originizationfor EconomicCooperationand Development. 1995. 'ParticipatoryDevelopment

and Good Governance.' DevelopmentCo-operation GuidelineSeries. Paris: OECD.

Paul, Samuel, "Accountabilityin Public Services: Exit, Voice, and Control,"World

Development,July 1992.

- 41 -
 ---------------. 1994. "Does VoiceMatter?" Policy Research Working Paper No. 1388.
Finance and Private Sector Division, the World Bank.

---------------. 1996. "A Citizen Report Card on Public Services: Mixing Barks and Bites?"
Mimeo.

Picciotto, Robert. 1994. "PuttingInstitutionalEconomicsto Work: From Participationto

Governance." Paper presentedat the conference: Economicand PoliticalInstitutions
for Development:Implicationsfor Assistance. Washington, DC.

Pohl, Gerhard, and DubravkoMihaljek. 1992, "Project Evaluation and Uncertaintyin

Practice: A StatisticalAnalysisof Rate-of-ReturnDivergencesin 1,015 World Bank
Projects," WorldBank EconomicReview 6(2):255-277.

Pourgerami, Abbas. 1988. "The Political Economyof Development:a Cross-National

Causality Test of Development-Democracy-GrowthHypothesis." PublicChoice
58:123-141.

Pritchett, Lant H. and LawrenceH. Summers, 1993, "The Structural AdjustmentDebate"

American EconomicReview 83:383-89.

Putnam, Robert with Robert Leonardiand RaffaellaNanetti. 1993. MakingDemocracy

Work. Princeton: PrincetonUniversityPress.

Scully, Gerald W. 1988. "The InstitutionalFramework and EconomicDevelopment."

Journal of PoliticalEconomy96(3):652-62.

Shea, John. 1993. "InstrumentalRelevancein Linear Models: A Simple Measure." Working

Paper No. 9312, SocialSystemsResearch Institute. University of Wisconsin-Madison.

Squire, Lyn and Herman van der Tak. 1975. EconomicAnalysis of Projects. Baltimore:
Johns Hopkins Press.

Wade, Robert. 1994. "PublicBureacracyand the Incentive Problem." Backgroundpaper for

WorldDevelopmentReport 1994.

Weede, Erich. 1983. "The Impact of Democracy on Economic Growth: SomeEvidencefrom

Cross-NationalAnalysis." Kyklos 36(1):21-39.

World Bank. 1992. Governanceand Development. Washington, D.C.: WorldBank.

1993. Annual Evaluation Results for 1991. Washington, D.C.: World Bank.

- 42 -
 . 1994. World Development Report 1994. New York: Oxford University
Press.

_ 1995. "The World Bank and Participation." Operations Policy Department,

World Bank.

- 43 -
AppendixTable A. 1: Summary statistics of independentvariablesa
Variable
nameb
1Mean
|
| Standard
Deviation
Range
(Possible)
Number of
countries
1Years
A) Civil liberties
Freedom 4.68 1.47 (1 to 7) 56 1974-1990
House
(Civil) _

Humana 55.13 17.97 (13 to 91) 38 1986

Media 2.50 0.91 (1 to 4) 56 1985
Pluralism
Freedom to 2.45 1.12 (i to 4) 56 1985
organize l
B) Political Rights
Freedom 4.73 1.85 (1 to 7) 55 1974-1990
House
(Political) _ _ l

Pourgerami 2.35 1.08 (1 to 4) 49 1984-86

Alesina 2.52 0.79 (I to 3) 55 1974-1982
Democracy
In d ex_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

C) Indicators of Civil Unrest _

Riots 0.14 1.61 (-3.83 to 56 1974-1989

17.50) l
Protest 0.29 1.63 (-0.79 to 56 1974-1989
Demonstra- 14.54)
tio n s _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

Strikes 0.07 0.50 (-0.43 to 56 1974-1989

3.50) l
Notes: a) Summarystatisticsfrom 1974 to 1990 for the 56 countrieswith projects in Tobit
and IV specifications;b) Data sources in the text and in the list of references.
Source: Authors' calculations.

-44-
 Policy Research Working Paper Series

Contact
Title Author Date for paper

WPS1531 Some New Evidence on Determinants Harinder Singh November 1995 S. King-Watson
of Foreign Direct Investment in Kwang W. Jun 31047
Developing Countries

WPS1532 Regulation and Bank Stability: Michael Bordo November 1995 D. Evans
Canada and the United States, 38526
1870-1980

WPS1533 Universal Banking and the Charles W. Calomiris November 1995 D. Evans
Financing of Industrial Development 38526

WPS1534 The Evolution of General Banking Forest Capie November 1995 D. Evans
38526

WPS1535 Financial History: Lessons of the Gerard Caprio, Jr. November 1995 D. Evans
Past for Reformers of the Present Dimitri Vittas 38526

WPS1536 Free Banking: The Scottish Randall iroszner November 1995 D. Evans
Experience as a Model for Emerging 38526
Economies

WPS1537 Before Main Banks: A Selective Frank Packer November 1995 D. Evans
Historical Overview of Japan's 38526
Prewar Financial System

WPS1538 Contingent Liability in Banking: Anthony Saunders November 1995 D. Evans

Useful Policy for Developing Berry Wilson 38526
Countries?

WPS1539 The Rise of Securities Markets: Richard Sylla November 1995 D. Evans
What Can Government Do? 38526

WPS1540 Thrift Deposit Institutions in Europe Dimitri Vittas November 1995 P. Infante
and the United States 37642

WPS1541 Deposit Insurance Eugene White November 1995 D. Evans

38526

WPS1542 The Development of Industrial Samuel H. Williamson November 1995 D. Evans

Pensions in the United States in the 38526
Twentieth Century

WPS1543 The Combined Incidence of Taxes Shantayanan Devarajan November 1995 C. Bernardo
and Public Expenditures in the Shaikh I. Hossain 37699
Philippines

WPS1544 Economic Performance in Small F. Desmond McCarthy November 1995 M. Divino

Open Economies: The Caribbean Giovanni Zanalda 33739
Experience, 1980-92
 Policy Research Working Paper Series

Contact
Title Author Date for paper

WPS1545 International Commodity Control: Christopher L. Gilbert November 1995 G. Ilogon

Retrospect and Prospect 33732

WPS1546 Concessions of Busways to the Jorge M. Rebelo November 1995 A. Turner

Private Sector: The Sao Paulo Pedro P. Benvenuto 30933
Metropolitan Region Experience

WPS1547 Testing the Induced innovation Colin Thirtle November 1995 M. Williams
Hypothesis in South African Robert Townsend 87297
Agriculture (An Error Correction Johan van Zyl
Approach

WPS1548 The Relationship Between Farm Size Johan van Zyl November 1995 M. Williams
and Efficiency in South African Hans Binswanger 87297
Agriculture Colin Thirtle

WPS1549 The Forgotten Rationale for Policy Jonathan Isham November 1995 S. Torres
Reform. The Productivity cf Daniel Kautman 39012
Investment Projects

WPS1550 Governance and Returns on Jonathan Isham November 1995 S. Fallon

Investment: Ar: Empirical Daniel Kaufman 38009
Investigation Lant Pritchett