Aggregating Governance With The Right Method

Download as pdf or txt
Download as pdf or txt
You are on page 1of 48

WP5 JIqS

Public Disclosure Authorized

POLICY RESEARCH WVORKING PAPER 2195


Public Disclosure Authorized

Aggregating Governance With the right method,


aggregate indicators can
Indicators provideusefulestimatesof
basic governance concepts as
well as measuresof the
Daniel Kafrfann imprecisionof these
Aart Kraay aggregate estimatesand their
Public Disclosure Authorized

Pablo Zoido-Lobat6n
components.
Public Disclosure Authorized

The World Bank


Development Research Group
Macroeconomics and Growth
and
World Bank Institute
Governance,Regulation,and Finance
October 1999 a
POLICYRESEARCHWORKINGPAPER2195

Summaryfindings
In recent yearsthe growinginterest of academicsand governancethan any singleindicator,the standarderrors
policymakersin governancehas been reflectedin the associatedwith estimatesof governanceare still large
proliferationof cross-countryindicesmeasuringvarious relativeto the units in whichgovernanceis measured.In
aspectsof governance. light of these marginsof error, it is misleadingto offer
Kaufmann,Kraay, and Zoido-Lobat6nexplain how a very preciserankingsof countriesaccordingto their level
simplevariant of an unobservedcomponentsmodel can of governance:smalldifferencesin country rankingsare
be used to combinethe informationfrom these different unlikelyto be statistically- let alone practically-
sourcesinto aggregategovernanceindicators.The main significant.Nevertheless,these aggregategovernance
advantageof this method is that it allowsquantification indicatorsare usefulbecausethey allow countriesto be
of the precisionof both individualsourcesof governance sorted into broad groupingsaccordingto levelsof
data and country-specificaggregategovernance governance,and they can be used to studythe causesand
indicators. consequencesof governancein a much larger sampleof
Kaufmann,Kraay,and Zoido-Lobat6nillustratethe countriesthan previouslyused (seefor examplethe
methodologyby constructingaggregateindicatorsof companionpaper by Kaufmann,Kraay,and Zoido-
bureaucraticquality,rule of law, and graft for a sample Lobat6n, "GovernanceMatters,"PolicyResearch
of 160 countries.Althoughthese aggregategovernance WorkingPaper 2196).
indicatorsare more informativeabout the levelof

This paper - a joint product of Macroeconomicsand Growth, Development Research Group; and Governance,
Regulation,andFinance,World BankInstitute-is part of a largereffort in the Bankto studythe causesandconsequences
of governancefor development.Copies of the paper are availablefree from the World Bank, 1818 H Street NW,
Washington, DC 20433. Please contact Diane Bouvet, room G2-136, telephone 202-473-5818,fax 202-334-8350,
Internet address [email protected] ResearchWorking Papers are also posted on the Web at http://
www.worldbank.org/html/dec/Publications/Workpapers/home.html. The authors may be contacted at
[email protected], [email protected],or [email protected]. October 1999. (39 pages)

The PolicyResearchWorkingPaperSeriesdisseminatesthe findingsof work in progressto encouragethe exchangeof ideasabot


development issues.An objective of the series is to get the findingsout quickly, even if the presentationtsare less than fully polisbed. The
paperscarry the names of tbe authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this
paper are entirely those of the authors. They do not necessarilyrepresent the viewvof the Worid P,ank, its Execu tive Directors, or the
countries
theyrepresent.

Produced by the PolicyResearchDisseminationCen1ter


AggregatingGovernanceIndicators

DanielKaufmann
Aart Kraay
FlabloZoido-Lobat6n

TheWorld Bank

Abstract:In recentyears,the growinginterestof academicsand policymakersin


govemancehas beenreflectedin the proliferationcross-countryindicesmeasuring
variousaspectsof governance.In this paperwe explainhow a simplevariantof an
unobservedcomponentsmodelcan be usedto combinethe informationfrom these
differentsourcesinto aggregategovemanceindicators.The mainadvantageof this
methodis that it allowsus to quantifythe precisionof the both individualsourcesof
governancedata as well as the aggregategovernanceindicators.We illustratethe
methodologyby constructingaggregateindicatorsof bureaucraticquality,rule of law,
andgraft, for a largesampleof 160 countries. Althoughthese aggregategovernance
indicatorsare more informativeatboutthe levelof governancethan any individual
indicator,the standarderrors associatedwith estimatesof govemanceare still large
relativeto the units in whichgovernanceis measured.

The World Bank,1818H StreetN.W.,Washington,D.C. 20433.


(dkaufmann(afworldbank. ora,akraavyworldbank.orq,ozoidolobatonAworldbank. ora).
The views expressedin this paperare the authors'alone,and in no way reflectthose of
the World Bank,its ExecutiveDirectors,or the countriesthey represent. We would like
to thankwithoutimplicationCraig Burnside,Brad Efron, EduardoEngel,Gil Mehrez,
JakobSvensson,ScottWallsten,Shang-jinWei, and seminarparticipantsat the World
Bank,UNDPand LatinAmericanEconometricSocietyMeetingsfor helpfulcomments.
1. Introduction

In recentyears,the growinginterestof academicsand policymakersin the


extent,causesand consequencesof governanceand misgovemancehas beenreflected
in the proliferationof cross-countryindicatorsof variousaspectsof governance.In an
accompanyingpaper(Kaufmann,Kraay,Zoido-Lobat6n(1999))we presenta large
databasecompilingseveralhundredcross-countryindicatorsof variousaspectsof
governance,producedby thirteenclifferentorganizations,and covering178 countries.
These indicatorsreportsubjectiveperceptionson a widerangeof issuesrelatingto
govemance,rangingfrom the extentto whichcorruptionin the politicalsystemaffects
foreigninvestment,to the efficiencyof publicservicesdelivery,to the likelihoodthat
citizensof a countryresortto extrajudicialmeansto settledisputes.

In this paper,we take the viewthat manyof these indicatorsserve as imperfect


proxiesfor one of a much smallernumberof fundamentalconceptsof governance.
Giventhis view, there are considerablebenefitsfrom combiningrelatedindicatorsinto a
small numberof aggregategovernanceindicators.First,the aggregateindicatorsspan
a muchlargerset of countriesthan any individualsource,permittingcomparisonsof
govemanceacrossa broadset of countries.Second,aggregateindicatorscan provide
more precisemeasuresof governancethan individualindicators. Third,it is possibleto
constructquantitativemeasuresof the precisionof boththe aggregategovemance
indicatorsand their components,allowingformaltestingof hypothesesregardingcross-
countrydifferencesin governance

We realizethese benefitsby constructingaggregategovernanceindicatorsusing


an unobservedcomponentsmodel. This modelexpressesthe observeddata as a linear
functionof unobservedgovernanceplus a disturbanceterm capturingperceptionerrors
and/orsamplingvariationin each indicator.The mainadvantageof this methodis that it
allowsus to obtainestimatesof the varianceof this disturbancetermfor each indicator.
These can be interpretedas a measureof howinformativeeach indicatoris aboutthe
broaderconceptof governanceit measures.We then computethe meanof the
conditionaldistributionof govemancegiventhe observeddata for each countryas a
naturalpointestimateof the levelof governancein that country. Similarly,the variance

1
of this conditionaldistributionprovidesa naturalestimateof the precisionof this
aggregategovernancemeasurefor eachcountry.

We illustrateour approachwith referenceto three fundamentalaspectsof


governance:rule of law, governmenteffectiveness,and graft. We group 31 indicators
constructedin 1997and/or 1998into three clusterscorrespondingto these three
conceptsof governance,and computeaggregateindicatorsspanning166, 156and 155
countriesrespectively.In our companionpaperdocumentingthe governancedatabase,
we constructsimilarindicesfor severalotheraspectsof governance.

Althoughthe unobservedcomponentsmethodologywe use is quitestandard,we


find its applicationto the constructionof compositegovemanceindicatorsinteresting.'
One of our majorfindingsis that the aggregategovernanceindicatorswe constructare
rather imprecise,despitethe highcorrelationsobservedbetweenvarioussourcesof
governancedata.In particular,a 90% confidenceintervalaroundthe pointestimateof
governancefor a typicalcountryspansalmostthe entire interquartilerange of the
distributionof estimatedgovemance.This impliesthat althoughit is possibleto robustly
identifytwentyor so countrieswith the best andworstgovemancein the world,it is
muchmore difficultto identifystatisticallysignificantdifferencesin govemanceamong
the majorityof countries.

Our resultsare basedon three key assumptions:(1) that the measurementerrors


in individualindicatorsof govemanceare uncorrelatedacrossindicators;(2) that the
relationshipbetweenunobservedgovemanceand observedindicatorsis linear,and (3)
that the distributionof unobservedgovernanceacrosscountriesis normal.Relaxingthe
first assumptionis difficultto do in practice,simplybecausewithoutthis assumptionwe
cannotdeterminewhetherthe correlationof observedscores acrossindicatorsis merely
due to correlatedperceptionerrorsor whetherit reflectsthe commonconceptof
governancebeingmeasured. However,underthe likelyaltemativethat perception
errors are correlatedacrosssources,the measuresof precisionwe report will be biased
downwards.As a result,the standarderrorswe report shouldbe interpretedas a lower
bound on the precisionof aggregategovemanceindicators. We considerthe

1Unobserved componentsmodelswere pioneeredin economicsby Goldberger(1972),and the closely-


relatedhierarchicaland empiricalBayesmodelsin statisticsby Efronand Morris(1971,1972).

2
consequencesof relaxingthe secondassumptionby proposinga methodwhichsimply
aggregatesthe ordinalinformationacrossindicators.Althoughthis has the advantageof
simplicityand does not requireassumptionsof linearity,it is also much less precisethan
the unobservedcomponentsmethodsince it discardsthe cardinalinformationin the
data. The third assumptionof a normaldistributionfor unobservedgovernanceimplies
that our estimatesof governancewill be clusteredaroundthe mean of this distribution.
This raisesthe possibilitythat the!difficultyin distinguishingbetweencountriesis in part
drivenby this assumption.We thereforeexplorethe robustnessof our results by
consideringaltemativesto this assumption,andfind that our conclusionsare materially
unaffectedby our assumptionson the shapeof the distributionof unobserved
govemance.

Thesefindingshaveseveralimplicationsfor policyand empiricalresearchon the


causesand consequencesof governancefor economicdevelopment.At a basiclevel,
the findingthat governanceis impreciselymeasuredshouldwarn againsttakingtoo
seriouslythe exactpoint estimatesof govemance,as well as countryrankingsbasedon
these estimates.At best,it is possibleto sort countriesinto broadcategoriesaccording
to their levelsof governance,and eventhen there is considerableuncertaintyregarding
the categoryto whichmany countriesshouldbe assigned.To emphasizethis point,we
avoiddiscussionsof specificcountriesin this paper. Second,since availableindicators
of governanceare noisy measuresof "true"govemance,empiricalwork whichuses
these indicatorsas explanatoryvariablesmaywell underestimatethe impactof
-governancedueto the usualattienuationbiascausedby badly-measuredright-handside
variables.Sinceour methodologyallowsusto quantifythe measurementerrors in these
2
variables,it is possibleto obtainroughmeasuresof the extentof this attenuationbias.
Finally,our resultssuggestthat if we want to more preciselydifferentiateamong
countriesaccordingto their leveilof governance,we needto improvethe qualityand
quantityof data gatheredon governance.

The remainderof this paperproceedsas follows. In Section2, we motivatethe


empiricalwork whichfollowsby describingthe indicatorsof governancewe use to

2 Inourcompanion
paperwe explorethisideain moredetail,usingcross-country
regressions
of percapita
incomeonvariousgovernance measuresinstrumenting forgovemanceusingmeasures
of thelinguistic'
composition
of thepopulation
- in thespiritof HallandJones(1999).

3
illustrateour ideas. In Section3, we lay out and implementthe unobservedcomponents
frameworkfor estimatinggovernance,and presentthe main resultsin Section4. In
Section5, we discussthe consequencesof relaxingseveralof the assumptions
underlyingthe model.We concludewith a discussionof the implicationsof our findings
for researchand policyadviceregardinggovernance.

4
2. Indicatorsof Governance

In this paper,we use datafrorn31 differentindicatorsof governanceconstructedin


1997and/or1998. These indicatorsare drawnfrom 13 differentsourcesand are
groupedinto three clusterscorrespondingto rule of law, governmenteffectiveness,and
graft. The key featuresof these indicatorsare summarizedin Table 1, and a detailed
descriptionof the sourcesandvariablescan be foundin Kaufmann,Kraayand Zoido-
Lobat6n(1999). In the first two columnsof Table I we identifyeach sourceof
governancedata by abbreviationand by name. In the nextthree columns,we reportthe
sourceof informationfor each measure(surveysof residentsor polls of experts),the
countrycoverage,and a measureof the extentto whichthe sampleof countriescovered
by each indicatoris representativeof the populationof countriesin the world. In the
remainingcolumnswe report the specificconceptsmeasuredby each sourcein each of
the three clusters.3

A quicklook at Table 1 showsthat these indicatorsdiffer alongseveraldimensions.


First,evenwithinclustersthere is considerablevariationin the particularconcept
measuredby eachindicator. Forexample,questionsaboutgraft rangefrom the
incidenceof "improperpractices' (WCY)to the likelihoodthat additionalpaymentsare
requiredto "getthingsdone"(WDR). Similarly,questionsregardingthe rule of law range
fromwhethercitizenscan successfullysue the stateto whethercitizensare likelyto
resolvedisputesextra-judicially.Despitethis heterogeneity,we takethe viewthat within
each cluster,eachof these conceptsis an imperfectindicatorof the corresponding
broaderconceptof governance.

The secondrespectin whichthese indicatorsdiffer is in the natureof the


respondentswho providethe information. Slightlyless than half of the indicatorsare
surveysof businesspeopleand/orresidentsof a country,while the remainingindicators
are pollsof expertswho rate a set of countriesaccordingto variouscriteria. As we
discussin more detaillater in the paper,this differencebetweenthese two typesof

3 For a numberof these sources,we usethe averageof severalquestionsrelatingto the corresponding


core conceptsof govemance.As we discusssubsequently, we are reluctantto includeindividualquestions
froma singlesourceseparatelyin our analysis,as the necessaryassumptionthat measurementerrorsare
uncorrelatedacrossindicatorsis muchmoredifficultto supportfor the caseof multiplequestionsfrom a
singlesource.

5
indicatorshas implicationsfor howwe interpretthe errorterms in the relationship
betweenobservedindicatorscoresand the underlyingconceptsof governance.

The third respectin whichthese indicatorsdifferis in the sampleof countriesthey


cover. A numberof indicatorscovera very largeand broadsampleof developedand
developingcountries(EIU, DRI, HFWSJ,PRSandWDR),while others cover very
narrowly-focusedsamplesof countries(PERCfor Asia, CEERand FHNTfor transition
economies). Someindicatorscover primarilydevelopedcountriesbut also includemajor
developingcountries(WCR,GALLUP,BERI). Thisdifferencebetweenindicatorsis
perhapsthe mostimportantfor the empiricalwork whichfollows. There is by now
considerableevidencethat govemanceon averagetendsto be betterin richercountries.
This impliesthat the distributionof governanceis likelyto be verydifferent in indicators
whichcover sets of countrieswith differentaverageincomelevels. Thesedifferences
needto be taken into accountwhen placingthe observeddata from variousindicators
into commonunitsand combiningtheminto aggregategovemanceindicators.

In orderto distinguishbetweenindicatorsin this dimension,we constructa simple


coverageindexwhichmeasuresdifferencesbetweenthe distributionof countriesacross
incomeand regionalclassificationsand the distributionof all countriesin the world
acrossthese categories.In particular,we dividethe world into a two-wayclassification
by regionand income,followingthe World Bank's1998World DevelopmentReport. For
eachof the sourcesof govemancedata,we reportone-halfof the sum of absolute
-deviationsbetweenthe shareof countriesin each of the 45 region/incomecategoriesin
that sourceand in the worldas a whole. By construction,this measurerangesfrom zero
to one, with low valuesindicatingmore representativeindicators.We reportthis number
in the fifth columnof Table 1. Thefive indicatorscoveringthe largestnumberof
countries(DRI,EIU, HF, PRSandWDR) are substantiallymore representative
accordingto this measurethan the others,with a valueof the coverageindexof less
than 0.25. In our subsequentempiricalwork we will referto these as representative
indicators,and the remainderas non-representative
indicators.

Finally,we notethat eachof these sourcesof governancedata uses different


units to measuregovernance.Mostpolls of expertsreportdiscretecategorical
responses(e.g.the prevalenceof corruptionon an integerscalefrom one to four),while

6
for mostsurveysof citizensor entrepreneurswe havethe mean responseacross
respondentsof discretecategoricalscores. We re-orientdatafrom each sourceso that
highervaluescorrespondto betteroutcomes(i.e. strongerrule of law, moreeffective
govemment,and less graft). In addition,we rescaleeach indicatorby subtractingthe
minimumpossiblescore and dividingbythe differencebetweenthe maximumand
minimumscores,sothat each indicatoris on a possiblescalefrom zero to one. Since
we rescaleeachindicatorusingthemaximumand minimumpossiblescores(ratherthan
the maximumand minimumactualscoresin the sampleof countriescoveredby each
indicator),this is nothingmorethan a convenientchoiceof units.

In Table2, we reportthe pairwisecorrelationsamongindicatorswithineach of the


three governanceclusters. Thegreat majorityof these are positiveand substantial,
frequentlygreaterthan 0.6. In the empiricalwork whichfollows,we will interpretthese
largecorrelationswithinclustersas reflectingthe commoncomponentof govemance in
these indicators. It is interestingto note that despitethe strongpairwisecorrelations
amongthese indicators,and despitethe favourableinterpretationthat these correlations
reflectthe commoncomponentof governanceratherthan correlatedperceptionerrors,
we neverthelessfind that governanceis not verypreciselymeasured.We providesome
intuitionsfor this in the followingsection.

7
3. EstimatingGovernance

In this sectionwe interpretthe data as beinggeneratedby an unobserved


componentsor multiple-indicatormodelin whichthe observeddata on governancecan
be expressedas a linearfunctionof unobservedgovernanceplus a randomerror term.
We reviewthe well-knownfeaturesof this model,and proposea simpleextensionwhich
deliversconsistentparameterestimatesfor representativeas well as non-representative
indicators.We then describehowthe parametersof this modelcan be estimatedand
can be usedto constructestimatesof each of the three aspectsof govemancein each
country.

The Model

Our data consistsof clustersof indicatorsof three aspectsof governance- rule


of law, governmenteffectiveness,andgraft. Letgo)denotean unobservedindexof one
of these three aspectsof governancein countryj, for example,graft. The observeddata
on graftconsistsof a clusterof k=1,...,Kindicators,each one providinga numerical
ratingof some aspectof graftin eachof the j=1,..,J(k)countriescoveredby that
indicator. We assumethat we can write the observedscoreof countryj on indicatork,
y(j,k), as a linearfunctionof unobservedgovernance,go), and a disturbanceterm, eo,k),
as follows:

(1) y(j,k) = a(k) + 1(k).(g(j) + F(j,k))

wherea(k) and ,B(k)are unknownparameterswhichmap unobservedgovernancego)


into the observeddata yo,k). We assumethat go)is a randomvariablewith meanzero
and varianceone. Our objectiveis to summarizeour knowledgeaboutgo)for each
countryj usingthe distributionof go) conditionalon the observeddata yo,k), k=1,...,Ko)
for countryj. The meanof this conditionaldistributionprovidesa naturalestimateof the
levelof govemancein countryj, andthe varianceof this conditionaldistributionis a
naturalmeasureof the precisionof this indicatorof governance.The assumptionof a
zero meanand unit variancefor governanceis an innocuouschoiceof units requiredto
identifythe parametersa(k) and ,B(k).Sincewe will allowthe varianceof the error term

8
to vary acrossindicatorsk, the fact that P(k)multipliesthe error term is an innocuous
rescalingwhichslightlysimplifiessomeof the expressionswhichfollow.

We use this unobservedcomponentsmodel,whichtreats unobserved


govemanceas a randomvariableratherthan as a fixed parameterto be estimated,for a
pragmaticreason. We will shortlyalso assumethat the varianceof the disturbanceterm
e(j,k)maydiffer acrossindicators.In this case,we cannottreat the ga)s as fixed
parametersto be estimatedfor eachcountry, sinceindividualeffectsare not identifiedin
4 Moreover,it should be clear
a fixed effectsmodelwith heteroskedasticdisturbances.
from Equation(1) that naiveaggregatessuch as a simpleaverageof rescaledindicators
for each countrywill not result in sensibleestimatesof govemance,as longas the
parametersa(k) and ,3(k)differ acrossindicatorsand differentcountriesappearin
differentsets of indicators.It is also not possibleto removethe dependenceof the
observeddata on these nuisance!parametersby standardizing(i.e. by removingthe
samplemeanfrom each indicator,and dividingby the samplestandarddeviation). This
is becauseif indicatork is non-representative,
the samplemeanwill reflectnotonly a(k),
but alsothe meanof go) in the sampleof countriescoveredby indicatork. Evenif an
indicatoris representativein the sensethat the standarddeviationof unobserved
governanceis equalto one in the sampleof countriesit covers, the samplestandard
deviationof observedscoreswill reflectnotonly ,8(k),but alsothe standarddeviationof
the disturbances.

The disturbanceterm s(j,k)capturestwo sourcesof uncertaintyin the relationship


betweentrue governanceandthe observedindicators. First,the particularaspectof
governancecoveredby indicatork is imperfectlymeasuredin each country,reflecting
either perceptionerrorson the partof experts(in the case of pollsof experts),or
samplingvariation(in the case of surveysof citizensor entrepreneurs).Second,the
relationshipbetweenthe particularconceptmeasuredby indicatork andthe
correspondingbroaderaspectolFgovernancemay be imperfect. For example,evenif
the particularaspectof graft coveredby someindicatork, (suchas the prevalenceof

4 Toseethis,consider
thespecialcase!
wherea(k)=Oand13(k)=1
forallindicators.Wecanmakethe
likelihood
functionoftheobserved dataarbitrarily
largesimplyby estimating
go)as theobserved scoreona
particularindicator,
forexample ga)=y(j.K),
foreverycountryj, andsettingcr(K)=0.Kiefer(1980)provides
a
detaileddiscussionof thispoint.

9
"improperpractices")is perfectlymeasured,it mayneverthelessbe a noisy indicatorof
graft if there are differencesacrosscountriesin what "improperpractices"are
consideredto be.

We assumethat the disturbanceterm has zero mean,E[Eo,k)1=O;


hasthe same
varianceacrosscountrieswithina given indicatorbut a differentvarianceacross
indicators,E[gj,k)2]=ao(k)
2; and is independentacrossindicatorsand countries,

E[E(,k)gj',k')]=Oif j#j' or k•k'. The varianceof the errorterm can be interpretedas a


measureof how informativeindicatork is aboutgo), and is likelyto vary across
indicators.The assumptionthat the errors are independentacrossindicatorsis a strong
one, but unfortunatelyone that is alsodifficultto relax. Intuitively,withoutthis
assumptionwe cannotidentifywhetherthe correlationof scoresbetweentwo indicators
is dueto their commoncomponentof governancego),or whetherit simplyreflectsthe
correlationof errors.In contrast,this identifyingassumptionmaintainsthat all of the
correlationof scoresacrossindicatorsis attributableto their commonestimateof
govemance.We will considerthe consequencesof relaxingthis assumptionin the
followingsection. For now,we simplynotethat this identifyingassumptioncorresponds
to a "bestcase" scenarioregardingthe precisionof govemanceaggregates,sinceit
assumesthat eachindicatorprovidesindependentinformationon a particularaspectof
governance.As a result,we are if anythinglikelyto overstatethe precisionwith which
governanceis measured.

The parametersc(k) and P(k)map unobservedgovemanceinto the observed


data. Althoughall of our indicators(afterrescaling)are nominallyin the same units and
are measuredon a scalefrom zero to one, there are neverthelessthree reasonswhy
these parametersmay differacrossindicators. First,notall indicatorsuse the entire
rangeof possiblescores. For example,althoughWDRmeasuresperceptionsof graft on
a scalefrom one to six, the lowestobservedscoreis only 2.36. This suggeststhat a(k)
on this indicatormaybe greaterthan that of an indicatorsuch as EIUwhich usesthe full
rangeof possiblescores. Second,a givenindicatormight be 'easy" ("tough")relativeto
other indicatorsin the sensethat it tends to overestimate(underestimate)a particular
aspectof governancein countrieswhereit is in fact low (high). This would be reflected
in a relativelyhigh (low)valueof a(k) on that indicator.Third,considera non-
representativeindicatorthat covers a set of countriesin whichthe averagelevelof a

10
particularaspectof governanceis betterthan in the worldas a whole (e.g. BERI,which
coversprimarilydevelopedcountries). Supposefurtherthat this sourcetends to score
countriesrelativeto each other,so that the worst(best)countryin the samplereceives
the lowest(highest)possiblescore?
of zero (one). Thiswould be reflectedin a relatively
highvalueof ,8(k),since relativelysmalldifferencesin true govemanceare magnified
into relativelylargedifferencesin observedscores.

The ConditionalDistributionof Governance

Our objectiveis to summarizeour knowledgeaboutgovernancein each countryj


by the distributionof governanceconditionalon the observeddata in countryj. Thistask
is greatlysimplifiedby assumingthat bothg() and the disturbancesso,k) are jointly
normallydistributed. In this case,go) andy(,k), k=1,...,Ko)are jointly normal,andthe
conditionaldistributionof g() giventhe data is also normal,with mean and variance
given by:.

(3) VE[(g)ly(j)]= 1y- -c

wherey() is a Kj)xl vectorwhichstacksthe K(j) data pointsfor countryj, a is the


correspondingK()xl vectorof a(k)s, B and , are K&)xK()diagonalmatriceswith the
corresponding,B(k)sand a,(k)2s on the diagonal,and t is a K()xl vectorof ones.We
referto the conditionalmean in (2) as the estimatedvalueof that aspectof govemance
in countryj. With a slight abuseof terminology,we referto an intervalfrom the (o/2)t
percentileto the (1-8/2)thpercentileof the conditionaldistributionof go)as an o-percent
"confidenceinterval"aroundthis estimate,andwe referto the squarerootof the
5
conditionalvariancein (2) as the "standarderror' of this estimate.

5 Thisframeworkhasa distinctlyBayesianinterpretation.The distributionof go) conditionalon the observed


datayj) can be vieweda posteriordistrlibution,and the meanof this distributionas an estimatorof g() would
bejustifiedas a point estimateof go) by a quadraticexpectedposteriorlossfunction. Similarly,the
"confidenceinterval'is analogousto a Bayesianhighestposteriordensityinterval.

11
These expressionshavea very naturalinterpretation.If the parametersa(k),
2 were known,a sensibleway to estimatego) would be to rescalethe
,(k) and acn(k)
observedscoresby subtractinga(k) and dividingby ,B(k),and then constructa weighted

averageof these re-scaledscores. In particular,let 9(j,k) 1 k) ( - =g(j)+ e(j,k)


P(k)
denotethe rescaledvalueof y(,k). Then the conditionalmeanin (2) is a weighted
averageof these standardizedscoresfor countryj on each of the KG)indicatorsin which
it appears,with weightscorrespondingto the inverseof the variance of the errorterm on
eachindicator,i.e.

E[(j)jy(j)] t(k) *(j,k) The conditionalvarianceis simply


k= 1 + E a (k) -2
k=1

( K(j)
V[g(j)ly(j)J = (1+ (k)- 2 ), whichis decreasingin the numberof indicators
k=1

availablefor that country,K(), and is increasingin the varianceof the errorterm in each
of these indicators,ue(k)2.

Estimatingthe UnknownParameters

In orderto implement(2) and (3), we needto first estimatethe unknown


parametersa(k), P(k)and o,(k)2 for everyindicatork. Forthe set of representative
indicators,we can use the assumptionof normalityof go) and so,k)to write downthe
likelihoodfunctionof the observeddata. Providedthat we haveat leastthree such
indicators,the modelis identifiedand it is straighfforwardto maximizethis functionwith
2Sto obtainestimatesof the unknownparameters
respectto the a(k)s, P(k)s, andcre(k)
6
for the representativeindicators.

6 Althoughmaximumlikelihoodestimationof theseparametersrequiresthe assumptionof normality,it is


also possibleto dispensewiththis assumptionand applya methodof momentsprocedure.In the just-
identifiedcase of three indicators,these methodsleadto identicalparameterestimates. In the overidentified
case of morethan threeindicators,thesemethodsdiffer onlyin the weightsappliedto the variousmoment
conditions,and in practicethis makeslittledifferencefor the parameterestimates.

12
indicators.To see why,
We cannotapplythis methodto non-representative
considerthe maximum-likelihood
estimateof a(k), which unsurprisinglyis the mean
score acrosscountrieson indicatork. It is straightforwardto see from Equation(1) that
the expectedvalue of the sampleimeanof scoreson indicatork is a(k) + ,B(k)* g(k),
where g(k) denotesthe averagelevelof governancein the sampleof countriescovered
by indicatork. Forrepresentativeindicators,our choiceof units for governance
normalizesg(k) = 0. However,for a non-representative
indicatorwherethe average
level of govemanceis differentfrom the worldas a whole, g(k) • 0 and the sample
meandoesnot providea consistentestimateof a(k).

We can neverthelessobtainconsistentestimatesof the unknownparametersby


usingthe followingsimpleargument. If go) were observable,we could estimatea(k),
P3(k)and aj(k) for any indicatorby regressingthe observedscoresy(,k) on go).
Althoughgo)is itselfnot observable,we do have an estimateof go)basedon the
representativeindicators. In partilcular,let g*() denotethe meanof go)conditioningonly
on the data from the representativeindicators.We can decomposethis conditional
mean into observedgovernancego) plus its deviationfrom the mean u0), i.e.
g*()=go)+uo) Sinceu0) is independentof go), we can view g*o) as measuringgo)with
error, i.e. as a classicerrors-in-variables
problem. It is well-knownthat OLS estimatesof
,8(k)from a regressionof y(,k) on g*() will producedownward-biased
estimatesdueto
the usualattenuationbias impartedby measurementerror in g*(). In particular,the

probabilitylimit of the OLSslopecoefficientis 1,(k) - VA *WI . Sincethe variance

of uO)is simplythe varianceof the conditionalmeanof go)givenin Equation(3), and


since V[g*0)]is observable,we can correctthe OLS coefficientsfor this attenuationbias
to arriveat consistentestimatesof the parametersof the non-representative 7
indicators.

indicatorswould be to imputedatafor the


7 An altemativeapproachto the problemof non-representative
missingobservations(in the spiritof Rubin(1987)).We do not pursuethis approachhere simplybecauseit
is difficultto specifythe key ingredientof the imputationprocess- the conditionaldistributionof the
unobserveddatagiventhe observeddata- in our application.

13
It is worth notingthat this estimationmethodurewardsconformity",in the sensethat
indicatorsthat are highlycorrelatedwill havelow estimatedvariancesand hencewill be
perceivedas more precise. Given our assumptionthat the disturbancetermsare
independentacrossindicators,it makessenseto treat highlycorrelatedindicatorsin this
way. If on the other handindicatorsare correlatedsimplybecausetheir disturbances
are correlated,this interpretationwouldbe inappropriate.We take this issueup in more
detailin Section4, and arguethat it will result in even less preciseestimatesof
govemancethan thosewe obtainhere.

14
4. Results

In this sectionwe implementthe unobservedcomponentsmodellaid out in the


previoussectionfor three conceptsof governance:governmenteffectiveness,rule of
law, and graft. We first presentestimatesof governanceand associatedstandarderrors
for eachcountry,andthen considerthe consequencesof these standarderrorsfor
identifyingcross-countrydifferencesin govemance.We concludewith a simpleexample
whichrelatesthe pairwisecorrelationsobservedamongindividualgovernanceindicators
directlyto the measuresof precisionof the aggregateindicators.

Estimates of Governance

Our mainfinding is that thoaavailabledata do not permitvery preciseestimatesof


governance.We illustratethis pointin Figure1. In each of the three panelsof Figure1,
we order countrieson the horizontalaxis by their estimateof governance,and on the
verticalaxis we plot the correspondingpointestimateof governance,i.e. the conditional
expectationof go)giventhe observeddata for countryj, and a 90-percentconfidence
intervalaroundthis pointestimate,i.e. the 5t and 95t percentilesof the conditional
distributionof governancefor eachcountryj. The size of these confidenceintervals
variesby country,reflectingthe fact that differentcountriesappearin differentnumbers
of sources,and that differentcountriesappearin differentsets sourcesof differing
precision.To providea senseof the dispersionin the pointestimatesof governance,the
three horizontallinesin eachgraphdelineatethe quartilesof the distributionof the point
estimatesof governancefor each cluster.

The moststrikingfeature of Figure1 is that these confidenceintervalsare large


relativeto the units in whichgovernanceis measured.For example,for a typicalcountry
the standarddeviationof the conditionalmeanof rule of law or graft is around0.3, so
that a typical90% confidenceintervalextends0.5 aboveand belowthe pointestimateof
graft. In the case of governmenteffectiveness,the standarddeviationof the conditional
meanis on averageslightlylargerand equal to 0.33,so that a 90% confidenceinterval
extends0.55 aboveand belowthe pointestimateof rule of law. These confidence
intervalsare largein the sensethat they are comparablein size to the entireinterquartile
rangeof the distributionof estimatesof governance.Moreover,it should be notedthat

15
these confidenceintervalsdo not reflectthe samplingvariationin the point estimatesof
the unknownparametersa(k), ,8(k)anda.(k). If this uncertaintywere also taken into
account,the standarderrorswould be evenlarger.

The parameterestimatesreportedin Table 3 revealsome interestingdifferences


acrossindicators.To interpretthe estimatesof the a(k)s and f3(k)s,note that our
assumptionof a standardnormaldistributionfor governanceimpliesthat the vast
majorityof countrieswili havegovemancerangingfrom -2 to 2. Sincethe observed
data rangefrom zero to one, one mightexpectthat a representativeindicatorwould
havea(k)=0.5and ,(k)=0.25. Interestingly,there are significantdeparturesfrom this
benchmark.Severalindicators(e.g.WDR)haveestimatedvaluesof o(k) substantially
lowerthan this benchmark,and highervaluesof a(k), indicatingthat they do not use the
entire rangeof possiblescores. There is alsoa great dealof variationin the estimates
of the standarddeviationof the errorson eachindividualindicator,c6(k), suggestingthat
the precisionwith whichindividualsourcesmeasuregovemancevarieswidely.

AssessingCross-CountryDifferencesin Governance

An advantageof this methodologyis that it permitsstraighfforwardtests of


hypothesesregardingcross-countrydifferencesin govemance. However,the largesize
of the confidenceintervalsdocumentedin Figure1 suggeststhat it will be difficultto find
statisticallysignificantdifferencesin govemancebetweenmany pairsof countries.We
illustratethis pointwith two simpleexercises.Supposefirst that for eachof the three
aspectsof governance,we want to groupcountriesinto quartilesaccordingto their level
of govemance. A naturalway to do this is to groupcountriesaccordingto their point
estimatesof govemance,i.e. accordingto the meanof the conditionaldistributionof
govemancein each country. Moreover,a naturalway to assessthe confidencewith
whichcountriesare assignedto quartilesis to considerthe corresponding90%
confidenceintervalsshownin Figure 1. In particular,if the 90% confidenceintervalfor a
countryfalls entirelywithina givenquartile,the probabilitythat this countryin fact
belongsin anotherquartileis less then 10%. For a smallgroupof countriesat each end
of the distributionof governance,we can concludewith a great deal of confidencethat
these countriesare in fact in the top and bottomquartiles. However,for the middle

16
quartilesthe situationis muchless clear,as veryfew countries'90% confidenceintervals
lie entirelywithin a givenquartile,for each of the three aspectsof govemance.

Clearly,the numberof countrieswe can assignto a particularquartileusingthis rule


dependson the size of the confidenceinterval. If we insteadconsidershorter
confidenceintervals,such as 75%or 50%intervals,we can betterdiscriminateamong
countries,albeitwith lowerconfidence.We explorethis possibilityin Table 4, wherewe
reportthe proportionof all countriesfor whichan x% confidenceintervalfalls entirely
withinthe indicatedquartile,for the threegovemanceaggregatesin tum. We consider
three possibilities,x=90%,x=75%and x=50%. At all significancelevels,a substantial
fractionof countriesin the top and bottomquartilescan be clearlyidentifiedas belonging
in these groups. As the size of the confidenceintervaldeclines,more and more
countriescan be significantlyassignedto quartiles. Nevertheless,evenat verylow
significancelevels,only one-quarterto one-halfof the countriesin the middletwo
quartileshaveconfidence intervalslyingentirelywithintheir respectivequartiles.

A relatedissue is the significanceof pairwisedifferencesin governance.In


particular,for every pair of countriesin whichour pointestimateof governancein
countryj is greaterthan in country j', we can investigatethe hypothesisthat countryj in
fact has bettergovemanceby computingthe probabilitythat g(j)>g(').8 For countries
with similarpoint estimatesoFgovemance,this probabilitywill be closeto 0.5,while for
countriesfar apart in the distributionof govemance,this probabilitywill approachone.
-Toillustratethis pointsystematically,for eachcountryj in the sample,we computethe
probabilitythat, conditionalon the observeddata for countriesj andj', g(j)>g(j')for every
comparatorcountryj'. We then computethe proportionof comparatorcountriesfor
whichthis probabilityis betwesen
5% and 95%. This is analogousto countingthe
numberof comparatorcountiriesfor whicha conventionaltest at the 10%significance
levelof the null hypothesisthat govemanceis the same in these two countriescannotbe
rejected.We summarizethe resultsof these pairwisecomparisonsin Figure2. We
again ordercountriesin ascendingorderaccordingto their pointestimatesof
governanceon the horizontailaxis,and we plot this proportionof comparatorcountries

8 Sincethe ga)ly() and g(')ly(') are jointly normaland independentby assumption,this calculationinvolves
a straighfforwardintegrationof the areaunder a bivariatenormalprobabilitydensityfunction.

17
as dark pointson the verticalaxis. We also repeatthe exercise,but insteadreportthe
largerproportionof countriesfor whichthis probabilityis between25% and 75%,which
correspondsto a test at the 50%significancelevel. This proportionis shownas a light
dot in Figure2.

Not surprisingly,at the two endsof the distributionthere are significantdifferences


betweenthe levelof govemancein these countriesand most other countries,especially
at the 50% significancelevel.However,there is also a stronginvertedU-shapedpattem
in this graph,reflectingthe fact that a largefractionof countriesare clusterednearthe
middleof the distributionof estimatedgovemance,and it is relativelydifficultto
distinguishamongsuch countries. In particular,for the "typical"countryaroundthe
middleof the distributionof govemance,govemanceis not significantlydifferentfrom
nearlyhalf of all other countriesin the world, at conventionalsignificancelevels.

Intuitions

Our findingthat governanceis impreciselymeasuredis somewhatsurprising.


After all, in Section2 we documentedthat fact thatthe pairwisecorrelationsamong
variousgovernanceindicatorsare substantial,andthe identifyingassumptionof
independenterrorsacross indicatorsimpliesthat the only sourceof this observedhigh
correlationamongindicatorsis the unobservedcommoncomponentof governance.
One mightthereforeeasilyconcludethatgovemanceis quite well measuredand that it
is straightforwardto distinguishamongcountries'govemanceusingthis data. We now
illustratewhy this intuitionis misleading,unlessthe correlationsin the observeddata are
very highindeed.

As a specificexample,supposethat there are onlythree representative


indicatorsassociatedwith a particulargovernanceconcept,i.e. K=3,and that the
pairwisecorrelationsamongthe observedscoresare all equalto p. It is straightforward

to showthat in this case,the estimatedvarianceof the residualwill be a (k)2 =P

18
for each of the surveysk=1,2,:3.9Insertingthis into Equation(3),the varianceof the
aggregategovernanceindicatorbasedon this hypotheticaldatawill be the samefor

eachcountryand is equalto V[g(j)ly ( = 1+2 . To give an ideaof the magnitude

of the corresponding90% confidenceintervals,we superimposethem on the


hypotheticaldistributionof governancein the upperpanelof Figure3, for variousvalues
of p. As p increases,the confidenceintervalsbecomeshorter. However,for the
correlationsof around0.75typicallyobservedin our governancedata,this confidence
intervalremainslargerelativeto the unitsin whichgovemanceis measured.

In the lowerpanelof Figure3, we relatethis to the significanceof cross-country


comparisons.A simplesummarystatisticis the proportionof countrieswhosetrue level
of governancelies withinthe 90%confidenceintervalof a particularcountry. This
proportionwill dependon the locationof the referencecountry,and on the correlationsin
the observeddata. We plot tliis proportionfor the mediancountryandthe countryat the
first quartileof the distributionof govemance,for variousvaluesof p. For the observed
correlationof indicatorsof around0.75,the 90%confidenceintervalaroundthe point
estimateof govemancefor the mediancountryencompassesthe true levelof
governancein abouthalf of all other countriesin the world, and somewhatlessfor a
countryat the first quartile. Onlyif the observedcorrelationsare verylarge is it possible
to distinguishthe mediancountryfrom most othercountrieswith a highdegreeof
confidence.

9 Toseethis,it is onlynecessary relatingthethreesamplemeansand


to solvethe systemof nineequations
thesixuniqueelements
of thesamplecovariance
matrixof theindicators
to theirpopulation
counterparts
and solve for the unknownparameters.

19
4. Extensions

In this sectionwe considerhow our resultsdependon three assumptions


underlyingthe unobservedcomponentsmodelof the previoussections:that the
disturbancesare independentacrossindicators,that the mappingfrom unobserved
governanceinto observeddata is linear,and that the distributionof unobserved
governanceis normal. We find that the first two assumptionsif anythingoverstatethe
precisionwith whichgovemanceis measured.Relaxingthe third assumptiondoes not
materiallyaffect our results.

CorrelatedDisturbances

In the previoussectionwe assumedthat the disturbancesEo,k)were


independentacrossindicators. Intuitively,this assumptionallowedus to attributeall of
the observedcorrelationof scoresacrossindicatorsto the commoncomponentof
governancego), and hencepermittedusto identifythe portionof the variationin scores
acrosscountrieswithineachindicatordue to measurementerror. A consequenceof this
assumptionis that any indicatorwhichis not very correlatedwith the otherswas
interpretedas havinga largeresidualvariance.

Althoughuseful,the assumptionthat the errorsare independentacross


indicatorsmay not be valid,for at leastthree reasons. First,in the case of pollsof
experts,it is possiblethat the perceptionsof expertswho rank countrieson a particular
indicatorare influencedby their knowledgeof countries'rankingson other indicators.
Second,the errors in surveysof residentsmight be correlatedacross countriesif
residentsof a particularcountryhavea tendencyto systematicallyoverstateregulatory
and govemanceobstacles,due to a broad-basedpredispositionto reporta worse
0 Finally,it is possiblethat perceptionsof
situationthan is objectivelywarranted.1
governancefrom variousindicatorsare undulyinfluencedby a single event,such as a
high-profilescandalwhichis not representativeof the levelof graft in that country.

10SeeKaufmanand Zoido-Lobat6n(1999).

20
Althoughit is not possibleto statisticallyidentifythe correlationof the
disturbancesacrosssources,it is straighfforwardto see the consequencesof positively-
correlatederrorsfor our results.If the errorsare correlatedacross indicators,each
additionalindicatorcontributesless informationto our estimateof governance.Thiswill
be reflectedin the varianceof the conditionaldistributionof governancein eachcountry.
In particular,it is straightforwaird
to showthat holdingconstantthe varianceof the
residualson each indicator,the varianceof the conditional distributionof govemanceis
11
increasingin the correlationbetweenthe errorson any two indicators.

We illustratethe practicalconsequencesof this observationin Table 5. For each


of the three aggregateindicators,we re-estimatethe varianceof the conditional
distributionof govemance,imposinga rangeof assumptionson the correlationof the
disturbances.For the purposesof this example,we restrictourselvesto a set of three
representativeindicators(EIU, DRIand PRS),and alsoto the set of about100 countries
As the assumedcorrelationamongthe errorsrises
which appearin all three indicators.12
from 0 to 0.5,the aggregategovemanceindicatorsbecomeless precise,althoughthe
magnitudeof the effectdependson the indicator(sincethe estimatedvariancesof the
disturbanceschangeas well). In the case of governmenteffectiveness,the standard
error of the aggregateincreasesonly slightly,from 0.32 to 0.35. In contrast,for rule of
law the standarderror doublesfrom 0.33 to 0.66.

It is difficultto adequatelyaddressthe problemof correlateddisturbancessimply


becauseit is not possibleto separatelyidentifythe correlationsbetweenthe errors.
Nevertheless,it is usefulto realizethat the estimatedstandarderrors associatedwith
pointestimatesof governanceare likelyto be substantiallyunderstatedunderthe
assumptionof independenterrors. This reinforcesour argumentof the previoussection
that cross-countrycomparisonsof the levelof governanceshouldbe madewith caution.

11If the positivecorrelationsbetweenthe residualsdifferacrosspairsof indicators,the relativemagnitudes


of the estimatedvarianceswill also be affected.In particular,supposethat two indicatorsof bureaucratic
qualityare highlycorrelatedwith each other,but notvery correlatedwiththe third. In the previoussection
we assumedthat the errorswere iindependent, and so the highcorrelationbetweenthe first two indicators
impliedthat the varianceof the errorswassmallon these indicatorsrelativeto the third. However,if we
knewthat the highcorrelationbetweenthefirst two indicatorswas due to correlatederrors,then the
estimatedvarianceson theseindicatorswouldbe largerelativeto that of the third indicator. As a result,the
relativerankingsof countriesmightalso be affected.
12
We do this only for simplicity,sinceit allowsus to reportjust one standarderrorper aggregate,whichis
the samefor all countries.

21
Non-Linearities

In the previoussectionwe assumedthat the relationshipbetweenunobserved


governanceand observedindicatorscoreswas linear. This assumptionplacesstrong
restrictionson the units in whichgovernanceis measuredin the variousindicatorsin our
sample. For example,consideran indicatorsuch as Gallupwhichasks respondents
how many casesof corruptionthere are amongpublicofficials,and offersthe choiceof
four broadcategories:'none", "a few","many"and "a lot". Our observeddata consistsof
numericalscoreson a scale of one to four correspondingto these categories.The
assumptionof a linearmappingfrom unobservedgraft into observedscoresimpliesthat
the differencein graft betweena countrywith a score of 4 and one with a score of 3 (i.e.
the differencebetween"a lot" and umany"casesof corruption)rsthe same as the
differencebetweentwo countrieswith scoresof 3 and 2 (i.e.the differencebetween
"many"and 'a few"). Giventhe somewhatvagueresponsecategories,it is not at all
clearthat the assumptionof linearityis warranted.Moreover,even if these categories
were equally-spacedaccordingto some appropriatemetric,the fact that the observed
data are discretewhile our unobservedgovemanceindicatoris continuousviolatesthe
assumptionof a linearrelationshipbetweenthe two.1 3

Finally,the merefact that indicatorsare non-representative


may also contribute
to a non-linearrelationshipbetweengovernanceand observedscores. A numberof the
indicatorsin our samplecoverprimarilydevelopedcountriestogetherwith a few
developingcountries. It is possiblethat the developingcountriesin these indicators
sufferfrom a "curseof inclusion"in the sensethat they receiveworse scoresthan they
might otherwisehave receivedsimplybecausethey are implicitlybeing comparedwith
4 For
countriesin whichvariousaspectsof govemanceare likelyto be much better.1
example,representativesurveyssuch as DRI or PRSassignmoderatescores of 55/100

13
The straighfforward solutionto this problemwouldbe to rely on an orderedmultinomialchoicemodelwith
individualeffectsin the latentvariable. HoWever,for sucha modelto be identified,it wouldagainbe
necessaryto assumethat the variancesof the errorsare identicalacrossindicators.
14This problemmaybe particularlyacutefor polls of expertswho considera largeset of countriesat once,
since in contrastto surveysof residents,expertsare muchmorelikelyto be awareof relativecomparisons
of countries.

22
and 3/6 to Mexicofor graft,while on a much less representativesurveysuch as BERI
whichcoversprimarilydevelopedcountries,Mexicoreceivesa rather poor ratingof 1/7.

Whatcan be doneaboutthese non-linearities?A verygeneralsolutionto this


problemmightbe to combineonly the ordinalinformationin each indicator,i.e. the
relativerankingsof countrieswithineach indicator. In particular,one can think of a given
indicatoras providinga rankingof pairsof countriesaccordingto their levelof
governance.The informationin the relativerankingsof countriesfrom variousindicators
can then be combinedby notingthat if severalindicatorsconsistentlyrank countryA as
havingbettergovemancethan countryB, this providesevidencethat governanceis in
fact betterin countryA than in countryB. Clearly,this approachhas severaladvantages.
First,it is computationallyvery simple. Second,we do not haveto knowthe choiceof
units in whichgovernanceis measured, or whetherindicatorsare representativeor not.
Third,it does not requireany assumptionof linearityin the relationshipbetween
governanceand observedscores. However,this methodwill result.inlargerstandard
errorsfor pairwisecomparisonssince it discardsinformationin differencesin the levelof
scoresacrosscountriesand indicators.15

To illustratethe relativeimprecisionof such an ordinalaggregate,for every pair of


countriesj andj' on every indicatork, we constructan indicatorvariablex0j,j',k)which
takesthe value 1 if countryj is rankedhigherthan countryj' with respectto the aspectof
governancecoveredby indicatork, and zero otherwise.16A naturalnull hypothesisto
test is that governancein countriesj andj' is the same,i.e. that the probabilityx(,j',k) is
equalto one is 0.5. Underthe assumptionthat the errors are independentacross

15A furtherdrawbackof this methodis that it is difficultto constructan aggregaterankingof countries


accordingto governance.One possibilitywouldbe to averagethe indicatorvariablesx0,j',k) over all surveys
k and partnercountriesj', and rankoDuntriesaccordingto this index.However,it is difficultto put standard
errorson sucha ranking,since even if the errorsare independentacrosssurveys,the x(,j',k) will not be
independentacrosspartnercountriesj'. A deeperproblemwiththis methodis that thereis a fundamental
resultfromsocialchoicetheorywhich placesstrongrestrictionson thepropertiessuch an aggregateranking
mayhave.Accordingto Arrow'sImpossibility Theorem,it is impossiblefor any aggregationof each
indicators''preferences" to simultaneiously
satisfythreeintuitiveand desirableproperties:(i) the aggregation
respectsunanimity-- if everyindicatorsaysthat A is morecorruptthan B, then so shouldour aggregate,and
(ii) the aggregationdisplaysthe independenceof irrelevantalternativesproperty-- the rankingof A and B
does not dependon any indicators'rankingbf A or B relativeto anyothercountryC; and (iii) the aggregation
is non-dictatorial- the aggregaterankingof A and B is not uniquelydeterminedby a singleindicators
rankingof A and B. In particular,Arrow'stheoremtellsus that if (i) and (ii) hold, then (iii) does not hold.
16Forthosesurveyswhich reportdiscretecategoricalscores,we discard'ties" as uninformativeaboutthe
relativelevel of govemance.

23
indicators,this hypothesiscan be testedusingthe data on the proportionof indicatorsin
which countryj is rankedhigherthan countryj' in a simplebinomialproportionstest.

We reportthe resultsof this exercisein Figure4, whichis analogousto Figure2.


We again ordercountriesin ascendingorderaccordingto their point estimatesof
govemanceon the horizontalaxis, and we plot the proportionof all comparatorcountries
for whichthe null hypothesisthat governancein these two countriesis equal cannotbe
rejectedat the 10%significancelevelas dark dots on the verticalaxis. The light dots
reportthe same information,but at the 50%level. ComparingFigures2 and 4, it is clear
that the ordinalaggregateallowsuse to identifyfar fewer statisticallysignificant
differencesin govemanceacrosscountries.In fact,for many countriesit is impossibleto
rejectthe null at the 10%levelthat govemancein this countryis the sameas for every
othercountryin the worldusingthis method! Althoughthis ordinalmethodis a useful
vehiclefor makingroughcomparisonsacrosscountriesand requireslittle in the way of
assumptionson the underlyingdata, it is muchmore difficultto obtainstatistically
significantdifferencesamongcountries.

Alternative Distributions for Governance

In Section2 we assumedthat unobservedgovemanceand the disturbances


werejointly normallydistributed.As we noted,this assumptionhas a significantpayoff
in terms of analyticaltractibility,as it ensuredthat the distributionof govemance
conditionalon the observeddata was normal,with simpleexpressionsfor its meanand
variance. However,giventhe bell-shapeof the normaldistribution,this approach
embodiesthe implicitassumptionthat a relativelylargefractionof countriesin the world
havesimilarmoderatelevelsof governance,and relativelyfew haveeither verygoodor
verybad governance.There are two reasonsto questionthis assumption. First,it is not
at all cleara priori that this providesan accuratedepictionof the true cross-country
distributionof governance.Second,it is possiblethat our findingthat it is difficultto
statisticallydistinguishdifferencesin governancebetweena largeproportionof countries
in the worldis accentuatedby the assumptionof normality,whichforcesa
disproportionatefractionof countriesto be clusterednearthe meanof the distributionof
governance. If insteadwe assumedthat governancewas moredispersed,then it is

24
possiblethat it is easierto identifystatisticallysignificantdifferencesin governance
acrosscountries.

It is not clear howone might identifythe shapeof the true distributionof


governanceacrosscountries,sinceit is difficultto disentanglethe shapeof this
distributionfromthe shapeof the distributionof the error terms. However,it is possible
to explorethe robustnessof the resultsto differentchoicesfor the distributionof
governanceitself. We do this by insteadassumingthat unobservedgovernancefollows
a Beta[a,b]distribution.We considerthree choicesof parameterscorrespondingto
three very differentshapesof the possibledistributionof govemance. Thesethree
possibilitiesare illustratedin the left-handcolumnof Figure5. We first considera=b=5,
whichgeneratesa symmetricbell-shapeddistribution.This case servesas a benchmark
in that it is verysimilarto the normaldistributionwe havebeenusingso far. We also
considerthe possibilitythat the distributionof govemanceis skewedto the right (a=2,
b=5),with relativelyfew countrieswith very goodgovernancein the right tail of the
distribution. Finally,we consicder
the possibilitythat govemanceis uniformlydistributed
(a=b=1),with a similarproportionof countriesat eachpossiblelevelof govemance.We
continueto assumethat the diistributionof the disturbancesis normal.

On the right-handside of Figure5, we explorethe consequencesof these


alternativeassumptionsfor our conclusionsaboutthe significanceof cross-country
differencesin govemance. As a specificexample,we focuson an aggregateof the
three largestrepresentativeindicatorsof graft (EIU, DRIand PRS),and againrestrict
ourselvesto the sampleof about100 countriesappearingin all three indicators. For
each country,we reportthe pointestimatesand standarderrorscorrespondingto each
1 7 As the assumedshape of the
assumptionon the distributionof governance.
distributionof true governancechanges,not surprisinglyso does the distributionof point
estimates. The more importantobservationis that our resultson the difficultyof
distinguishingbetweencountriesdo not change. It is clearfrom Figure5 that the

17
We computethese as follows. First,usinga methodof momentsargumentwe constructestimatesof the
parametersof the modelcorresponding to the assumeddistributionof govemance(notethat the meanand
varianceof this beta distributionchangeas we varythe parameters).We then constructthejoint distribution
of go)andy() as theappropriate
mixtureof a normalanda betadistribution,
andthenobtainthemarginal
distributionof yj) and conditionaldistributionof go) givenyG)by appropriatenumericalintegrationsof this
joint distribution.We numericallyevaluatethe meanof this distribution,and the 5e and 95 percentiles,and
reportthesefor eachcountryin Figure5.

25
number of countries with 90% confidence intervals falling entirely within particular
quartiles is essentially unchanged for each quartile, as we vary our assumptions on the
shape of the distribution of govemance.

26
5. Conclusions

In this paper,we havetalkentheviewthat the many differentavailableindicators


of governanceprovideimperfectsignalsabouta relativelysmallnumberof fundamental
aspectsof governance,such as rule of law, govemmenteffectiveness,and graft. We
groupedthe many availableindicatorsinto three clusterscorrespondingto these
conceptsof governance,and useda linearunobservedcomponentsmodelto obtain
aggregateestimatesof thesethree aspectsof govemance. Despiteseveraloptimistic
assumptions,we find that govemanceis not very preciselymeasuredusingthese
aggregateindicators. In particular,althoughit is possibleto identifystatistically
significantdifferencesbetweencountriesat oppositeendsof the distributionof
govemance,it is much moredifFicultto discriminateamongthe majorityof countrieswith
any degreeof confidence.

Nevertheless,we find the aggregategovernanceindicatorsto be usefulfor


severalreasons. First,they are'basedon a methodologywhichprovidesa consistent
frameworkfor placingdata fromvarioussourcesinto commonunits,taking into account
that the samplesof countriesincludedin differentsourcesmay not be representativeof
the world as a whole. Second,the aggregateindicatorsspan a much largersampleof
155 or morecountries,permitting(admittedly-imprecise)
comparisonsacrossa much
largerset of countriesthan is possibleusingany singleindicator. Third,althoughthe
aggregateindicatorsare not as preciseas one mighthave hoped,they are nevertheless
muchmore reliablethan any individualindicator. Finally,we believethat it is usefulto
havequantitativemeasuresof the precisionof aggregateindicatorsin orderto caution
usersof both individualand aggregateindicatorsof the substantialmarginsof error
associatedwith cross-countrycomparisonsof govemance.

Empiricalresearchon governanceissuescan also benefitfrom the aggregate


indicatorspresentedhere. Manyempiricalstudieswhichuse govemanceindicatorsas
eitherleft-handor right-handside variablesare limitedto smallsamplesby the poor
countrycoverageof manyindicators.This can potentiallyintroducea varietyof sample
selectionbiases. In addition,the measuresof precisionwe reportcan be usedto correct
for the attenuationbiasdue to measurementerror in govemanceindicatorsusedas
dependentvariables.

27
In the longterm, however,our resultsalsopoint to the inadequacyof existing
governancemeasures.jt is very unsatisfyingthat existingdata,evenwith favourable
assumptions,allows us to identifyrelativelyfew statisticallysignificantdifferencesin
governanceacrosscountries. Moreover,existingdata providesat besttenuouslinks
betweenperceptionsof govemanceand objectivepolicyinterventionsthat govemments
interestedin improvingthe qualityof governancecan undertake.There is thereforea
needto improvethe qualityand quantityof govemancedata,both by improvingand
extendingcross-countrysurveywork of govemanceperceptions,as well as employing
1 8 Manyof the polls and surveyswe
country-specificin-depthgovemancediagnostics.
use sufferfrom deficiencies,such as poorly-wordedquestionsabout ill-definedand
excessivelybroadconcepts. There is roomto improvethese instrumentsby asking
respondentsabouttheir directexperienceswith well-definedeventsand using
transparentunitsto measuregovemance. However,these are time- and resource-
intensiveexercises,and intemationally-comparable
high-qualitydata of this sort is years
away.

18Detailedcountrydiagnosticexercisessuch as thosecurrentlybeingpilotedby the World Bankhave the


potentialto providemuchmoredetailedinformationon the specific
institutionalfailureswhich contributeto
perceptionsand the realityof poorgovemance.Kaufmann,Pradhanand Ryterman(1998)providea
descriptionof theseexercises.

28
References

Efron,Bradleyand Carl Morris(1971). "Limitingthe Riskof Bayesand EmpiricalBayes


Estimators- Part 1: The BayesCase'. Joumalof the AmericanStatistical
Association.66:807-815.

Efron,Bradleyand Carl Morris (1972)."Limitingthe Risk of Bayesand EmpiricalBayes


Estimators- Part 1: The EmpiricalBayesCase". Journalof the American
StatisticalAssociation.637:130-39.

Goldberger,A. (1972). "MaximumLikelihoodEstimationof RegressionsContaining


UnobsevableIndependentVariables". IntemationalEconomicReview. 13:1-15.

Hall, Robertand CharlesJones(1999). "WhyDo SomeCountriesProduceSo Much


MoreOutputPerWorkerThan Others?". QuarterlyJoumalof Economics.
114(1):83-116.

Kaufmann,Daniel, Aart Kraay,and PabloZoido-Lobat6n(1999). "Governance


Matters". Manuscript,The WorldBank.

Kaufmann,Danieland PabloZoido-Lobat6n(1999). "Corruption,Unpredictabilityand


Performance".Manuscript,The World Bank.
Kaufmann,Daniel, SanjayPradhanand RandiRyterman(1998). "New Frontiersin
Diagnosingand CombatingCorruption",PREMNote Number7, The World Bank.

Kiefer,NicholasM. (1980). "Estimationof Fixed EffectModelsfor TimeSeriesof Cross


Sectionswith ArbitraryIntertemporalCovariance".Joumalof Econometrics.
14:195-202.

Rubin,Donald(1987). MultipleImputationfor Non-Responsein Surveys. NewYork:


Wiley.

29
Table 1: GovernanceIndicators

Code 6e Coe
(Sura C eptls
cir (P)
Pnli 12M-t

BEPI Businoss P 50 mostly 0.44 Buresauratic delays Erforceability of contracts Mewntality"regarding


Environment Risk developed corruption
Intelligenco countries

CEER Central Ewopean P 26 transition 0.84 Rule of law Effect of conotion on


Economic Review economies 'attractiveness of com"tiy
as a piace to do business'

DRN Standard and P 106 developed 0.23 Governmentineffectiveness, Enforceability contracts, Comuptionamong pubic
Poor's DRI and developing instiutional hfaihe costs of crime officials, offectivenessof
countries antkorruption iitiativs

EIU Economist P 115developed 0.19 Insitutional efficacy, red tape Cme, corruption in Corruptionemong public
Intelligence Unit and developing bakng sector of4ciats
counties
FHNT Freedom House P 28 transiion 0.82 Quality of goverment and Rule of law Perep,tions of corrlp,tiOn
economies public administration in civil service, business
interests of polkymakers

GALLUP Galup S 44 mostly 0.50 Freqwencyof "cases of


Intemational developed corruption"ramongpublic
countries offcials

GCS Gobal S 59 develped 0.42 Competence of pubic Cdizens can file awsuts Frequency of "kregular
Competitiveness and developing sector, polrtial pressures on against government payments" to officials and
Survey countries civi servants, time spent citizns acoept legal judiiary
with tbrreauciats e4udicln,
independence of judiciary,
costs of crime

GCSA Global S 23 African 0.73 Competence Ofpblic Citiwenscan file lawsuits Frequency of "iregular
Competitiveness countries servants, commitment to against goversnent, paymentsWto offiials and
Stivey, Afria polices of previous cdizens acceot legal judiciary
govefmments edudication,
independence of judiciary,
costs of mime

HF Heritage P 160developed 0.06 Law and ordertradition,


Foundation anddeveloping prevalence of black
countries market activiies

PERC Poliical and S 12 Asian 0.83 Effect of corrupton on


Economic Risk economies busins environment for
Consultancy foreign companies

PRS Poitical Risk P 131 developed 0.10 Bureatic qulaity, policy Rule of law Corruptionin the political
Services and developing stability system sa "threat to
countries foreign investment"
WCY World S 46 primarily 0.59 Efflcien implementation of Tax evasion, confidence "Improperpratices in
Competitiveness devetoped goennrert decisions in atblity o ato ties to the publi sptere
Yearbook countries political pressures on civil protect property,
servants confidence in
admrinistrationof justice
WDR World S 74 developed 0.25 Efficiency of govemment in Unpredictabilityof the Corruption as "obstacle to
Deveopment and develoing delivenrig services, judiciary, ttheft end crime, busine, frequency of
Report countries predictabiliy of rules, time ability of state to protect "additiol payments" to
spent wih bureaucrats private propeity "get things done"

Notes: Detailson these sourcesof governancedata, and definitionsof the concepts


measured,may be found in AppendixA of Kaufmann,Kraayand Zoido-Lobat6n
(1999).

30
Table 2: CorrelationsAmongGovernanceIndicators

GovernmentEffectiveness
gee,u gedn goprs gewdr geberi geefint gegcs gogcs gswcy
gaeiu 1.00
114
gadri 0.77' 1.00
96 106
geprs o.60- 0.61 1.00
111 100 140
gewdr 0.78 * 0.68 0.36 1.00
58 57 65 74
gsberi 0.74 * 0.71 * 0.54 * 0.73 1.00
49 so 50 30 so
gent 066 * 0.71 024 0.62 - 0.67 1.00
19 24 21 20 6 28
gegcs 0.76 0.74 0.52 * 0.85 0.70 0.75 * 1.00
64 62 72 45 46 6 75
gagcsa 0.64 0.69 0.55 0.26 1.00 0.56 1.00
1 15 20 - 14 2 0 19 23
gewcy 0.55 0.53 0.48' 0.81 0.65 0.06 0.92 1.00
43 44 46 27 41 4 46 0 46

Ruleof Law
ddi diu rdMf rIprs rdwdr dberi dceer Mint dges dgsa ljz dwqcy iscore
rtdi 100
106
dwiu 0.73* 1.O
96 114
Mhf 0.73 0.686 1.00
105 112 16O
rtpus 0.75 0.75* 0.62 1.00
1oo' 111 137 140
dwdr 0.58 0.75 0.76 0.59s 1.00
57 58 72 es 74
rberi 0.73 0.70* 0.82* 0.62' 0.54 1.00
so 49 50 50 30 50
de 0.69 * 0.79 * 0.90 * 0.52 * 0.51 - .0.86 1.00
24 19 25 20 20 6 27
dfhrt 0.76' 0.66' 0.86 0.33 0.394 -0.75' 0.91 * 1.00
24 19 26 21 20 6 27 28
dg3 0.70* 0.768 0.78* 0.70 0.74* 0.77* 0.96 0.78 1.00
53 se 59 59 33 46 6 6 59
dugsa 0.43 031 0.03 0.27 0.61 * -1.00 0.87 1.00
15 18 23 20 14 2 0 0 3 23
rjlkz 0.63 - 0.45 0.37 0.57 0.13 0.64 0.19 0.00 0.50 0.16 1.00
6s 73 76 74 48 44 19 20' 49 8 77
dwCy 0.63 0.73 0.77' 0.68' 0.786 0.75' 0.98' 0.94' 0.94 0.47' 1.00
44 43 46 46 27 41 4 4 46 0 40 46
drscor 1.00 0.99 0.95 0.91 0.72 0.63 0.99 0.986 0.986 . 0.37 0.96' 1.00
5 5 S 5 5 5 5 5 5 0 5 4 5

Graft
gWrd grei grpn gWwdr guSeti grow grit gupilup grgrs grgca gWpercgnucy grscore
grdri 1.00
106
greiu 0.80 100
97 115
guprs 0.65 0.64 1.00
100 112 140
grwdr 0.69e 0.80* 0.46 100
57 5s 65 74
gubiei 0.57 0.58 0.48 0.78 1.00
50 50 50 30 50
gnroer 0.91 0.76 0.68 0.56' 0.38 100
24 19 20 20 6 26
grntd 0.79 o0.60 0.72n 0.58 0.40 0.92' 100
23 19 21 20 6 25 28
grgalkp 0.63 0.78- 0.62 0.81 0.46 - 0.71 0.69 1.00
42 42 44 26 29 9 9 44
grgcs 0.77' 0.88' 0.57 0.87 0.73' 0.98 0.80 * 0.60 1.0o
53 57 59 33 46 6 7 35 59
grgcsa 0.59 0.53 * 0.45 0.61' 1.00 . . 0.9 0.79 1.00
15 18 20 14 2 0 0 3 3 23
grperc 0.58 0.95' 0.33 0.96' 0.76' . . 0.84' 0.89 * 1.00
12 12 12 6 11 0 0 5 12 0 12
gr*cy 0.69 0.-84 0.65' 0.93 0.62 -0.32 -0.51 0.67' 0.856 . 0.93 100
44 44 46 27 41 4 4 31 46 0 11 46
grscere 0.91 0.92' 0.82' 0.869 0.74' 0.89' 0.67' 0.73 0.82 0.64 0.82- 0.82' 1.0o
51 51 51 51 30 13 13 25 33 11 6 27 51

Thistable reportspairwisecorrelationsbetweengovemanceindicatorswithineach
govemancecluster. The numbersbelowthe correlationcoefficientsindicatethe
numberof countriescommonto eachpair of indicators.* indicatessignificanceat the
90% level.

31
Table 3: ParameterEstimates

Govemment Effectiveness Rule of Law Graft

cx(k) Om oe) a( kfk S a(k) a (k) _s(k)

RepresentativeIndicators
DRI 0.539 0.239 0.588 0.668 0.179 0.583 0.539 0.221 0.618
EIU 0.432 0.226 0.396 0.442 0.285 0.445 0.312 0.309 0.322
HF 0.466 0.247 0.751
PRS 0.789 0.109 1.222 0.606 0.220 0.656 0.506 0.142 1.129
WDR 0.473 0.097 0.685 0.354 0.135 0.681 0.465 0.150 0.636

Non-RepresentativeIndicators
BERI 0.406 0.133 0.707 0.404 0.136 0.687 0.483 0.173 1.226
CEER 0.604 0.380 0.303 0.615 0.359 0.356
FHNT 0.624 0.396 0.440 0.566 0.397 0.454 0.561 0.513 0.511
GALLUP 0.470 0.149 0.709
GCS 0.398 0.151 0.600 0.526 0.148 0.594 0.493 0.247 0.457
GCSA 0.470 0.112 0.450 0.499 0.041 1.695 0.498 0.297 0.406
PERC 0.302 0.299 0.270
WCY 0.307 0.120 0.906 0.382 0.156 0.580 0.212 0.284 0.498

32
Table?
4: AssigningCountriesto Quartiles

Proportionof Countriesfor Which an x% Confidence


IntervalLies Entirelyin the IndicatedQuartile

x=90% x=75% x=50%

GovernmentEffectiveness
First Quairtile 0.31 0.54 0.72
SecondQuartile 0.00 0.00 0.26
Third Quartile 0.00 0.13 0.31
FourthQuartile 0.59 0.69 0.79

Ruleof Law
First Quartile 0.31 0.43 0.65
SecondQuartile 0.00 0.05 0.39
ThirdQuartile 0.12 0.24 0.55
FourthQuartile 0.55 0.63 0.84

Graft
First Quartile 0.13 0.23 0.49
SecondQuartile 0.00 0.03 0.26
Third Quartile 0.08 0.21 0.36
Fourth Quartile 0.65 0.72 0.80

This table reports the fraction of all countries whose point estimate of govemance
falls in the indicated quartile for which the corresponding x%/o confidence interval
also falls entirely within that quartile, for each of the three govemance aggregates
and for a range of values of x.
Table 5: Consequences of Correlated Disturbances

AverageStandardErrorof GovemanceAggregate
Basedon RepresentativeIndicators

AssumedError Correlation:
p=0 p=0. 2 5 p=0.5

Government Effectiveness 0.32 0.31 0.35

Rule of Law 0.33 0.47 0.66

Graft 0.31 0.42 0.44

Thistable reportsthe standarderror of an aggregateindicator(basedon a


balaricedpanel of three sources),underaltemativeassumptionsregardingthe
correlationof the disturbances.

34
Figure1: Estimatesof Governance

Go.,m.t EfIhws.

zT I TTT I
T4Si L*. R

.< i:.1U._
~~~~~~~~t. ._

2Z5

2o5

Countries areorderedon the horizontalaxisin ascendingorderaccordingto their


pointestimatesof governance, andtheirpointestimatesand90%confidence
intervalsareindicatedonthe verticalaxis. Thehorizontallinesdelineatethe
quartilesof thedistribution
of governanceestimates.Forreasonsof space,
countrynamesareindicatedonilyfor everyfifthcountry.

35
Figure2: Significanceof PairwiseGovernanceComparisons

I GovenmentEffectvenes

0.9
Z0.

a - 10%SignificanceLevel

071

. 10% SignMcanc Le* l


. Ol

X0.3
0.03**8

0.9
c 0.4

0.

_! 02 80%S10%ne LevelficSn ev
0. 6 - *
&0.29-
w~~~~~~~~~~~~
'2E0 1 * GaL
. f Lo

0.8

point estimates
of governance. For e1achScountr c,nweLpvloto h etclaih
.O0.7 - B
e t. Sis gnificanceL e< >*E<,, .+
0-
O

c 0.e areindc tevel

E0.3

between
5%and 95% (ark dots) 1ri0eten2%
Sglandc70eL (vligtdt) o
reasons of space,0 conr nm sae niatd0.7freer itcuty

gO S~~~3
Figure 3: Precision of GovernanceAggregates
as a Function of PatirwiseCorrelations of Observed Data

90% Confidenceintervalfor
mediancountryif
correlationof observeddata is:

p=0 .7 5
90 \
p=O.
Distributionof Actual
Govemance

-3 -2 -1 0 1 2 3
Govemance Index

0.7 -

5 0.6

0.5
0.5
O S = \ S , CountryatMedian

ouS0Country at F rst Quartile


° 0.3 -

o Z%0.2-
~E
to
-n

0.

0.5 0.6 0.7 0.8 0.9 1


Pairwise Correlation Among Observed Data

37
Figure 4: Significance of Pairwise Governance Comparisons
Based on Ordinal Method

GovemmentEffectivenea

01
Oa * . . s. S *,-a -

0.7 *
}07~~~~~ *-a *-,'g @ "
0.7

0. SigrificanceLevel
1~50%

0.3,
IL L

Rule of Law

0.9 - -t* -

:*: S}: *. * *'- 't :':'a

m 0.6 ,

j0.4 - #N 5

0.4 -aU ~ a55

03 so%SJgnmcae50%SignificanceLevel
0.2

Grt
-V.
i 0.9
6]107~~ * . U **a. *

10.8 U** .

§ &; 0.7 r >; . 10% Significance Level.

i 0 ° 5g * g , a.¢

0C3
16
li~~~~~~ ~~~~~~50%
° SignificanceLevel
a

0.2-
0.1
*- 0

Countriesareorderedon thehorizontalaxisin ascendingorderaccordingto their


pointestimatesof governance.Fpr each country,we plot on the verticalaxis the
proportionof all comparatorcountriesfor whichthe probabilitythat govemancein
the referencecountryis greaterthan that in the comparatorcountry is either
between5% and 95% (darkdots), or is between25% and 75%. For reasonsof
space,countrynamesare indicatedonly for everyfifth country.

38
Figure 5: Alternative Distributions for Governance

AssumedDistributionof Governance Distributionof GovernanceEstimates


e025 1

0.a
0.02 .

0.015

0.01 05-nMTE 9

0000~~~~~~~~~~~~~~WI

0.1

0 0.2 0.4 0.6 0.8 1 s

0.020

0.02 A >.

0.0210.

0.015 0000~~~~~~~~~~~~~~~~~~~~~~~~~.

02

0.02~~~~~~~~~~~~~~~~~~~~~~~~'
0.0Th 1
0.018~~~~~~~~~~~~~~~~~~~~~~~~~~~~.

0.015

o 0.2 04 0.a 08 1 0

0.2
coum we2 s oAthe shap of thsdsrbto.I0 h ih oun eso h
0.1

0005

0 0.2 0.4 0.6 06 1 ~~~~~~~~~~~~~~~02

The rowsofthis figurecorrespond


to the assumptions
thatunobserved
governancefollowsa Beta(5,5),Beta(2,5)anda Beta(1,1)distribution.Inthe left
columnweshowtheshapeofthis distribution.In therightcolumn,we showthe
correspondinganalogto Figure1,for a governance
aggregate constructed using
a balancedpanelof threeindicators
of graft.

39
Policy Research Working Paper Series

Contact
Title Author Date for paper

WPS2174 ProspectiveDeficitsand the Asian CraigBurnside September1999 E. Khine


CurrencyCrisis MartinEichenbaum 37471
SergioRebelo

WPS2175SectorGrowthand the Dual Economy Niels-HughBlunch September1999 H. Vargas


Model:Evidencefrom Cote d'lvoire, DorteVerner 37871
Ghana,and Zimbabwe

WPS2176 FiscalRisks and the Qualityof HanaPolackovaBrixi September1999 A. Panton


FiscalAdjustmentin Hungary Anita Papp 85433
Allen Schick

WPS2177FiscalAdjustmentand Contingent HanaPolackovaBrixi September1999 A. Panton


GovernmentLiabilities:Case Studies HafezGhanem 85433
Of the CzechRepublicand MacedoniaRoumeenIslam

WPS2178 NonfarmIncome,Inequality,and Land RichardH. AdamsJr. September1999 M. Coleridge-Taylor


In RuralEgypt 33704

WPS2179 HowChildLaborand ChildSchooling RanjanRay September1999 M. Mason


Interactwith Adult Labor 30809

WPS2180 RegulatingPrivatizedInfrastrictures Ofelia Betancor September1999 G. Chenet-Smith


and Airport Services RobertRendeiro 36370

WPS2181 Privatizationand Regulationcf the LourdesTrujillo September1999 G. Chenet-Smith


SeaportIndustry GustavoNombela 36370

WPS2182The Integrationof Transition ConstantineMichalopoulosSeptember1999 L. Tabada


Economiesintothe WorldTrading 36896
System

WPS2183MarketDisciplineand Financial Asli Demirgug-Kunt September1999 K. Labrie


Safety Net Design HarryHuizinga 31001

WPS2184 FinancialServicesand the World Aaditya Mattoo September1999 L. Tabada


Trade Organization:Liberalization 36896
Commitmentsof the Developiig
and TransitionEconomies

WPS2185 FinancialSector Inefficienciesand Pierre-Richard


Agenor September1999 T. Shiel
CoordinationFailures:Implications JoshuaAizenman 36317
for CrisisManagement

WPS2186Contagion,BankLendingSpreads, Pierre-Richard
Ag6nor September1999 T. Shiel
and OutputFluctuations JoshuaAizenman 36317
AlexanderHoffmaister
Policy Research Working Paper Series

Contact
Title Author Date for paper

WPS2187 Who DeterminesMexicanTrade Jean-MarieGrether September1999 L. Tabada


Policy? Jaimede Melo 36896

WPS2188 FinancialLiberalizationand the PedroAlba September1999 R. Vo


CapitaiAccount:Thailand,1988-97 LeonardoHernandez 33722
DanielaKlingebiel

WPS2189AlternativeFrameworksfor Stijn Claessens September1999 R. Vo


ProvidingFinancialServices DanielaKlingebiel 33722

WPS2190The Credit Channelat Work:Lessons GiovanniFerri September1999 K. Labrie


from the Republicof Korea'sFinancial Tae SooKang 31001
Crisis

WPS2191 Can NoAntitrust PolicyBe Better Aaditya Mattoo September1999 L. Tabada


Than SomeAntitrustPolicy? 36896

WPS2192 Districts,Spillovers,and Government RezaBaqir September1999 S. Devadas


Overspending 87891

WPS2193Children'sGrowthand Povertyin MicheleGragnolati September1999 M. Gragnolati


RuralGuatemala 85287

WPS2194 DoesDemocracyFacilitatethe Jean-JacquesDethier October1999 H. Ghanem


EconomicTransition?An Empirical HafezGhanem 85557
Study of Centraland EasternEurope EddaZoli
Andthe FormerSovietUnion

You might also like