Background Paper to the

2002 Corruption Perceptions Index

Framework Document 2002

PD Dr. Johann Graf Lambsdorff∗

Transparency International (TI) andGöttingen UniversityJuly 2002

The CorruptionPerceptions Index is acomposite index. The dataused this year werecompiled between 2000 and2002.

Comparisons to last year’sindex should be based onscores. However, suchcomparisons can bemisleading because ofmethodological changesbetween years.

This document describesthe data that entered intothe index, discusses howcorruption is defined byour sources and relates theperceptions gathered toreality. A new methodologywas used for standardizingthe data and fordetermining confidenceintervals. Both these issuesare described here indetail.

∗ The author is utmost indebted to Walter Zucchini, Göttingen, for methodological innovations.Thanks also to Brad Efron, Susan Rose-Ackerman and Lynn Gale for helpful comments.

1. The methodologyTransparency International (TI) publishesits annual Corruption Perceptions Index(CPI) since 1995. This index has developedinto a leading indicator in social sciences.As in previous years, this frameworkdocument provides an in-depth explanationof the methodology and measurementprecision.

The goal of the CPI is to providedata on extensive perceptions of corruptionwithin countries. The CPI is a compositeindex, making use of surveys ofbusinesspeople and assessments by countryanalysts. It consists of credible sourcesusing diverse sampling frames and differentmethodologies. These perceptions enhanceour understanding of real levels ofcorruption from one country to another.

As pointed out in previousframework documents, unbiased, hard datacontinue to be difficult to obtain andusually raise problematic questions withrespect to validity. International surveys onperceptions therefore serve as the mostcredible means of compiling a ranking ofnations.

Overall, 15 sources could beincluded in the 2002 CPI, originating from9 independent institutions. The completelist of sources is presented in the appendix.All in all, the number of countries in theCPI increased from 91 last year to 102.

Sources in 2002

Prior to selecting sources guidelines havebeen set up which organize the underlyingdecision making process. These include theactual criteria that a source needs to meet inorder to qualify for inclusion as well asorganizational guidelines on how the finaldecision is reached with the help of theTransparency International SteeringCommittee. This process aimed at makingthe final decision as transparent and robustas possible. As a result of this it wasdecided that the 2002 CPI includes datafrom the following sources:

• The World Economic Forum (WEF)• The Institute for Management

Development, Lausanne (IMD)• PricewaterhouseCoopers (PwC)• the World Bank's World Business

Environment Survey (WBES)• The Economist Intelligence Unit (EIU)• Freedom House, Nations in Transit

(FH)• the Political and Eonomic Risk

Consultancy, Hong Kong (PERC)• Gallup International on behalf of

Transparency International (TI/GI)• the State Capacity Survey by Columbia

University (CU)

An essential condition for inclusion is that asource must provide a ranking of nations.This condition is not met if a sourceconducts surveys in a variety of countriesbut with varying methodologies.Comparison from one country to anotherare not feasible in this case and a rankingcannot be produced. Another condition isthat sources must measure the overall levelof corruption. This is violated if aspects ofcorruption are mixed with issues other thancorruption such as political instability ornationalism or if changes are measuredinstead of levels of corruption.

For example, the index “Corruptionin Government” from the InternationalCountry Risk Guide (ICRG), conducted bythe Political Risk Services (PRS), did notmeet these requirements, albeit beingwidely used in research as a measure oflevels of corruption. It does not determine acountry’s level of corruption but thepolitical risk involved in corruption. Aspointed out to us by the ICRG-editor, thesetwo issues can differ considerably,depending on whether there exists a high orlow tolerance towards corruption.Corruption only leads to political instabilityif it is not tolerated. Due to this, the data byPRS-ICRG did not qualify for inclusion inthe CPI. However, TI hopes to include amodified set of data by PRS in the future.

The 2002 CPI combines assessmentsfrom the past three years to reduce abrupt

variations in scoring that might arise due torandom effects. Some sources, such asTI/GI, WBES and PwC, provided only onerecent survey. Others such as PERC, WEFand IMD conducted various surveysbetween 1999 and 2001, which are allincluded. In addition to its GlobalCompetitiveness Report (GCR), the WEFalso published the Africa CompetitivenessReports (ACR) in 2000, which is alsoincluded.

While this averaging is valuable forthe inclusion of surveys, it is inappropriatefor application to the data compiled bycountry experts. Such assessments ascompiled by FH, CU and EIU areconducted by a small number of countryexperts who regularly analyze a country'sperformance, counterchecking theirconclusions with peer discussions.Following this systematic evaluation, theythen consider a potential upgrading ordowngrading. As a result, a country's scorechanges rather seldom and the data showslittle year-to-year variation. Changingscores in this case are the result of aconsidered judgement by the organizationin question. To then go back and averagethe assessments over a period of time wouldbe inappropriate. On the other hand, in thecase of elite surveys an averaging overvarious years produces a useful smoothingeffect. While some annual data may containrandom errors, these do not necessarilycarry over into the next year.

Year-to-year comparisons

Comparisons to the results from previousyears should be based on a country’s score,not its rank. A country’s rank can changesimple because new countries enter theindex and others drop out. A higher score isan indicator that respondents providedbetter ratings, while a lower score suggeststhat respondents revised their perceptiondownwards. However, year-to-yearcomparisons of a country's score do notonly result from a changing perception of acountry's performance but also from achanging sample and methodology. Withdiffering respondents and slightly differing

methodologies, a change in a country'sscore may also relate to the fact thatdifferent viewpoints have been collectedand different questions been asked. Theindex primarily provides an annualsnapshot of the views of businesspeople,with less of a focus on year-to-year trends.

However, to the extent that changescan be traced back to a change in the resultsfrom individual sources, trends cancautiously be identified. Noteworthyexamples of a downward trend between2001 and 2002 are Argentina, Ireland andMoldova. The considerable decline in theirscores does not result from technical factors– actual changes in perceptions aretherefore likely.

With the same caveats applied, onthe basis of data from sources that havebeen consistently used for the index,improvements can be observed for theDominican Republic, Hong Kong, Russia,Slovenia and South Korea.

2. Validity

All sources generally apply a definition ofcorruption such as the misuse of publicpower for private benefit, for examplebribing of public officials, kickbacks inpublic procurement, or embezzlement ofpublic funds. Each of the sources alsoassesses the “extent” of corruption amongpublic officials and politicians in thecountries in question:

• In 2002 the IMD asks respondents toassess whether “bribing and corruptionprevail or do not prevail in theeconomy.”

• The WEF asks in its 2002 GlobalCompetitiveness Report “in yourindustry, how commonly would youestimate that firms make undocumentedextra payments or bribes connected withimport and export permits, publicutilities and contracts, business licenses,tax payments, loan applications,influencing of laws and policies, andgetting favorable judicial decisions.” Aslightly different question had been

posed in 2000 and 2001 and in theAfrica Competitiveness Report

• The PERC asks in 2001 “How do yourate corruption in terms of its quality orcontribution to the overallliving/working environment”. A slightlydifferent question had been askedpreviously.

• The EIU defines corruption as themisuse of public office for personal (orparty political) financial gain and aims atmeasuring the pervasiveness ofcorruption. Corruption is one of over 60indicators used to measure “countryrisk” and “forecasting.”

• PwC asks for the frequency ofcorruption in various contexts (e.g.obtaining import/export permits orsubsidies, avoiding taxes).

• FH determines the "level of corruption"without providing further definingstatements.

• The WBES asks two questions withrespect to corruption, one determiningthe "Frequency of bribing" and anotherone relating to "corruption as aconstraint to business".

• Columbia University asks for the se-verity of corruption within the state.

• Gallup International on behalf ofTransparency International asks “Howcommon are bribes to politicians, seniorcivil servants, and judges and howsignificant of an obstacle are the costsassociated with such payments for doingbusiness?”

The terms "prevalence","commonness", "frequency", "constraint ","contribution to working environment" and“severity” are closely related. They all referto some kind of “degree” of corruption,which is also aim of the CPI. This commonfeature of the various sources is particularlyimportant in view of the fact that corruptioncomes in different forms. It has beensuggested in numerous publications thatdistinctions should be made between theseforms of corruption, e.g. between nepotismand corruption in the form of monetarytransfers. Yet, none of the data included in

the CPI emphasize one form of corruptionat the expense of other forms. The sourcescan be said to aim at measuring the samephenomenon. As also emphasized in theframework documents of previous years,the sources do not distinguish betweenadministrative and political corruption.

The term "degree of corruption" mayimply different things. In particular, it mayrelate to the frequency of bribes or the sizeof bribes. But we know from the results ofour sources that frequency and the size ofbribes tend to correlate highly (asexpounded in the framework documents ofprevious years). In countries wherecorruption is frequent it also amounts to ahigh fraction of firms' revenues. In sum, theterm “degree of corruption” seems toequally reflect the two aspects, frequency ofcorruption and the total value of bribespaid.

3. Perceptions and reality

While the sources all aim at measuring thedegree of corruption, the sample designdiffers considerably. The data by IMD,WBES, PwC and WEF largely sampleresidents (sometimes also frommultinational companies). In contrast, thedata by PERC, FH, TI/GI, CU and EIUlargely relate to expatriates. Whether thisdifference between samples may lead todifferent outcomes still requires scientificstudy. For the purposes of the CPI it addedto the robustness of the resulting figures,since the data correlate well, irrespective ofwhether expatriates or residents had beenpolled. This correlation suggests that therebeing different samples makes no greatdifference to the results.

Interpreting perceptions

As the data collected relates to perceptionsrather than to real phenomena, it has to beconsidered whether such perceptionsimprove our understanding of what reallevels of corruption may be. Since actuallevels of corruption cannot be determineddirectly, perceptions may be all we have toguide us. However, this approach is

undermined to at least some extent, if theperceptions gathered are biased. Such apotential bias might originate from theparticular cultural background ofrespondents. Depending on whether thesample consist of locals or expatriates, thissuggests two potential biases to be relevant.

Imagine that being asked to assessthe level of corruption, a local estimates ahigh level of corruption in the country ofresidence. Such an assessment would be avalid contribution to the CPI only if therespondent makes the assessment as a resultof comparisons with the levels of corruptionperceived in other countries. But this is notnecessarily the viewpoint taken by therespondent. A respondent may also assignhigh levels by comparing corruption toother (potentially less pressing) problemsfacing the country, or by evaluating itaccording to a high ethical standard (e.g.which assumes any kind of gift-giving to apublic official to be corrupt and notculturally acceptable). In the case of suchan outlook, a high degree of observedcorruption may reflect a high standard ofethics rather than a high degree of realmisbehavior. Perceptions would be amisleading indicator for real levels ofcorruption. This bias can occur particularlyif only locals are surveyed, each assessingonly the level of perceived corruption intheir own countries. If respondents areasked to assess foreign countries or to makecomparisons between a variety of countries,this bias should not occur. Respondentswill, in this case, compare a foreign countrywith their home country or with an evenlarger set of countries. They will be forcedto apply the same definition of corruptionand make use of the same ethical standardfor all countries, which produces validcomparative assessments.

However, in this context a secondtype of bias might arise, originating fromthe potential dominance of a particularcultural heritage in the sample questionedor because expatriates lack a properunderstanding of a country's culture. Theresults would be meaningless to locals whohave a different understanding and

definition of corruption. While samples ofexpatriates are susceptible to this kind ofbias, surveys which question local residentsclearly avoid it.

The strength of the CPI rests withthe idea that we include surveys which arenot susceptible to the first type of bias.Particularly these are EIU, TI/GI, CU, FHand PERC. Since the data provided by thesesources refers to assessments by expatriates,they are subject to a homogeneousdefinition of corruption and a consistentethical standard. In case of TI/GIrespondents have been asked to comparebetween different countries. This ensuresthat a consistent ethical standard is applied.

The CPI also incorporates the datafrom the IMD, WEF, PwC and WBES.Since these refer to assessments madelargely by local residents, they are lesslikely to represent the perception of acertain cultural heritage. The second type ofbias can clearly be rejected for thesesources.

Since the data from the EIU, TI/GI,CU, FH and PERC correlate well with theother data, there seems to be no support forthe suggestion that they might be influencedby the second type of bias. Similarly, sincethe data by the IMD, WEF, PwC andWBES correlate well with data from theother five institutions; the notion that thefirst type of bias might be present is clearlynot supported. The validity of the sources ismutually confirmed and no hint is found forthe existence of a bias in our data.

Another criticism of the CPI wasthat expatriates surveyed are often westernbusinesspeople. The viewpoint of lessdeveloped countries seemedunderrepresented. TI/GI now surveysrespondents from less developed countries,asking them to assess the performance ofindustrial countries. This balances thesample; yet, as shown in the correlations, itdoes not bring about noteworthy differentresults. Thus, the comparative assessmentsgathered in the CPI do notdisproportionately reflect the perceptions ofwestern businesspeople.

In sum, it seems that residents tendto have a consistent ethical standard withregard to assessments of corruption, whileexpatriates do not tend to impose aninappropriate ethical standard or to lackcultural insights. Our approach clearlysuggests that the perceptions gathered are ahelpful contribution to the understanding ofreal levels of corruption.

As was also explained in detail inthe 2001 framework document, theperceptions gathered well relate to actualexperience made and less to hearsay.1

4. The index

Standardizing

Since each of the sources uses its ownscaling system, aggregation requires astandardization of the data before eachcountry’s mean value can be determined.This standardization is carried out in twosteps.

Until 2001 a simple means andstandard deviation approach was adoptedfor step 1. The aim was to ensure thatinclusion of a source consisting of a certainsubset of countries should not change themean and standard deviation of this subsetof countries in the CPI. In 2001 the 2000CPI was the starting point forstandardization, except for older sourcesthat were already standardized previously(where the standardized values determinedin previous years were utilized).Standardization meant that the mean andstandard deviation of a new source musttake the same value as the respective subsetin the 2000 CPI. With S'(j,k) being theoriginal value provided by source k tocountry j, the standardized value, S(j,k),was determined by

1 See Lambsdorff, J. Graf (2001)“Framework Document.”, BackgroundPaper 2001 Corruption Perceptions Index:www.gwdg.de/~uwvw/2001_CPI_FD.pdf

-k)(j,S' k)S(j, [=

+(k))SD(S'CPI) SD(2000

(k)Mean(S' ]

CPI) Mean(2000

where the means and standard deviations(SD) for the source k and the 2000 CPIhave been determined for the joint subset ofcountries. After standardizing each source,the simple average was taken for eachcountry.

The step 2 is a final standardizationof the average values determinedpreviously. Taking the average implies thatthe resulting index has a standard deviationacross countries which is smaller than thatof the CPI of previous years. In order toavoid a year-to-year trend towards acontinuously smaller diversity ofassessments, the scores had to be stretched.This ensured that the standard deviation ofcountries in the index2 remained constantover time.

This approach was subject to intensedebate this year. It was decided that amodified approach should be used for step1: matching percentiles. Instead ofequalizing means and standard deviations,the ranks (and not the scores) of countries isthe only information processed from oursources. For this technique again thecommon subsamples of a new source andthe previous year’s CPI are determined.Then, the largest value in the CPI is takenas the standardized value for the countryranked best by the new source. The secondlargest value is given to the country rankedsecond best, etc.3 Imagine that a new

2 More precisely, we should talk about thestandard deviation of a common sub-sampleof countries in the index.3 In case two countries share the same rank,their standardized value is the simple meanof the two respective scores in the CPI. Thescores for countries where no CPI valuewas available are determined by referring toneighbor countries in the source’s ranking.Linear interpolation is applied to theirscores, suggesting that if a source assignssuch a country a score close to the upper

sources ranks UK best, Singapore second,Venezuela third and Argentina fourth. Inthe 2001 CPI these countries obtained thescores 9.2, 8.3, 3.5 and 2.8. Matchingpercentiles would now assign UK the bestscore of 9.2, Singapore 8.3, Venezuela 3.5and Argentina 2.8.

Matching percentiles is superior incombining indices that have differentdistributions. But, as it makes use of theranks, and not the scores of sources, thismethod looses some of the informationinherent in the sources. What tipped thebalance in favor of this techniques is itscapacity to keep all reported values withinthe bounds from 0 to 10: All countries inthe CPI obtain scores between 0 (verycorrupt) and 10 (highly clean). While wereport these absolute bounds, the previousstandardization provided no guarantee thatall values remained within these bounds. Ineffect, equalizing means and standarddeviations can bring about standardized val-ues above 10 or below 0. This has indeedhappened in the past with e.g. Finlandobtaining standardized scores above 10.Last year, standardized values forBangladesh of –1.7 have led to confusionamong observants. Matching percentiles, onthe other hand, guarantees that allstandardized values are within thesebounds. This results because anystandardized value is taken from theprevious year’s CPI, which by definition isrestricted to the aforementioned range.

In sum, matching percentiles has thedisadvantage of wasting some informationby processing only the ranks reported bysources. Yet, this disadvantage seems to beoffset because 1) the approach is free of

neighbor, also its standardized value iscloser to that of this neighbor. If such acountry is ranked best (or worst) by asource it would have only one neighbor, nottwo. The second neighbor is constructed byusing the highest (or lowest) attainablescore by the source and the CPI value 10(or 0). This approach guarantees that allvalues remain within the range between 10and 0.

assumptions regarding the distribution ofsources, and 2) all standardized valuesremain within the range from 0 to 10.

Step 2

Having obtained standardized values thatare all within the reported range, a simpleaverage from these standardized values canbe determined. As already argued before,the resulting index has a standard deviationwhich is smaller than that of the CPI ofprevious years. Without a second adjust-ment there would be a trend towards acontinuously smaller diversity of scores. If,e.g., Finland were to repeat its score fromthe previous year, it would have to scorebest in all sources listing this country. If itscores second to best in any source, thestandardized value it obtains after usingmatching percentiles and aggregation wouldbe lower than its current score. Thus, givensome heterogeneity among sources, itseems inevitable that Finland’s score woulddeteriorate. The opposite would be true ofBangladesh, which would obtain a betterscore if it is not consistently rated worst byall its sources. A second standardization isrequired in order to avoid a continuoustrend to less diversity among scores.

However, applying a simple meanand standard deviation technique mightagain bring about values that are beyondour range from 0 to 10. The proposal wouldtherefore be to apply a more complicatedstandardization for this second step: A beta-transformation. The idea behind thismonotonous transformation is to increasethe standard deviation to its desired value,but to keep all values within the range from0 to 10. Each value (X) is thereforetransformed according to the followingfunction:

∫ −− −∗1

0

11 )10/1()10/(10 dXXX βα

This beta-transformation is available instandard statistics programs. The crucialtask is to find the parameters α and β sothat the resulting mean and standarddeviation of the index have the desiredvalues. An algorithm has been determined

that carries out this task. Applying thisapproach to the 2002 CPI, the change in thescores is depicted by figure 1. The parame-ters were α=1.1756 and β=1.1812. Asshown in the figure, scores between 5 and10 are increased slightly, while thosebetween 0 and 5 are lowered.

This effect makes sure that the pre-vious standard deviation is preserved. Yet,once a score of 10 has been reached, thescore is not further increased. Equally, ascore of 0 is not further decreased. Thisguarantees that all values remain within therange.

The beta transformation is firstapplied to all values that were standardizedin step 1. Afterwards the average of theseare computed to determine a country’sscore. In our publication we also report thehigh-low range. This refers to allstandardized values after carrying out thebeta-transformation. This procedureensured that the high-low range isconsistently related to a country’s meanvalue. All these values remain within therange from 0 to 10.

While the methodologicaladjustments were considerable, their impacton the outcome was rather small. Had the

2002 CPI beendetermined with theprevious methodology,the result would correlate0.996 with our currentone. In spite of the

methodologicalmodifications, thereexists a high numericalcontinuity of the CPI.4

Reliability andPrecision

A ranking of countriesmay easily bemisunderstood asmeasuring theperformance of a countrywith absolute precision.This is certainly not true.Since its start in 1995 TIhas provided data on thestandard deviation and

the amount of sources contributing to theindex. This data already serves to illustratethe inherent imprecision. Also the high-lowrange is provided in the main table. Thisdepicts the highest and the lowest valuesprovided by our sources, so as to portraythe whole range of assessments. However,no quick conclusions should be derivedfrom this range to the underlying precisionwith which countries are measured.Countries which were assessed by 3 or 12sources can have the same minimum andmaximum values, but in the latter case wecan feel much more confident about thecountry’s score. In order to arrive at suchmeasures of precision, other statisticalmethods are required.

The strength of the CPI is based onthe concept that a combination of datasources combined into a single indexincreases the reliability of each individualfigure. As in previous years, the 2002 CPI

4 Matching percentiles as compared to alinear transformation provided aparticularly unfavorable score for Tunisia,bringing about a deterioration of about 0.5in its final score.

0 2 4 6 8 10

-1.0

-0.5

0.0

0.5

1.0

Figure 1: Beta Transformation

CPI-score prior to transformation

Cha

nge

due

to tr

ansf

orm

atio

n

includes all countries for which at leastthree sources had been available. The ideaof combining data is that thenonperformance of one source can bebalanced out by the inclusion of at least twoother sources. This way, the probability ofmisrepresenting a country is seriouslylowered. This is valid even in case thesources are not totally independent of eachother. Such partial dependency may arise ifsome respondents are aware of otherpeople's perception of the level ofcorruption, or of other sources contributingto the CPI.

An indicator for the overall reliabilityof the 2000 CPI can be drawn from the highcorrelation between the sources. Since thestandardization approach makes use only ofthe ranks (and not the scores) provided by

the sources, we report Kendall’s rankcorrelation. Referring only to countriesincluded in the CPI, this data is in table 1.5

5 Abbreviations relate to the sources used,Africa Competitiveness Report (ACR) ofthe World Economic Forum, EconomistIntelligence Unit (EIU), Freedom House(FH), Global Competitiveness Report(GCR) of the World Economic Forum,

As the correlations on average are 0.7, thesources do not differ considerably in theirassessment of levels of corruption. It shouldbe noted that the rank correlation is lowerthan the more commonly used Pearson-correlation, which is 0.84 on average for thevarious sources.

In addition to these correlations, thereliability of each individual country scorecan be determined. The larger the numberof sources and the lower the standarddeviation between the sources, the morereliable is the value for a country. Therelatively large standard deviation forPoland of 1.1 signifies that 95% of thesources ranged between a value of 1.8 and6.2. In contrast, the low standard deviationfor Canada of 0.2 means that 95% of thescores range between 8.6 and 9.4.

Institute for Management Development(IMD), Political and Economic RiskConsultancy (PERC),PricewaterhouseCoopers (PwC), WorldBusiness Environment Survey of the WorldBank (WBES), Gallup International onbehalf of Transparency International(TI/GI) and Columbia University (CU).

Confidence range

We have been providing readers with theadditional information on the confidencerange for some years now. These werebased on the determination of the standarderror for a country’s average score and aresulting parametric assessment of a 95confidence range. This approach requiredthe assumption that there is no imprecisionassociated with the source’s values and thatthese values are independent of each other.Another strong assumption required is thaterrors are normally distributed. While it isstatistically difficult to relax the first twoassumptions, one can relax the assumptionof a normal distribution and apply testswhich are valid throughout any type ofdistribution. Another drawback of the olderconfidence ranges was, again, that theysometimes violated the given range from 0to 10. For example, while in 2001Bangladesh had a score of 0.4, its 95%confidence range was between –3.6 and 4.4.For Finland, on the other hand, the rangewent as high as 10.4. This type of a range isconfusing even for an expert. Since it is incontradiction to the official range reported,the public is equally disoriented.

In order to restrict the confidencerange to our pre-specified limits, this yearwe introduce a different approach: a non-parametric approach applying the bootstrapmethodology. The principal idea of such abootstrap confidence range is to resamplethe sources of a country with replacement.If five source values (3, 5, 4, 4.5, 4.2) hadbeen given, an example of such a samplewould be (5, 5, 4.2, 3, 3). A sufficientlylarge number of such samples (in our case10,000) are drawn from the available vectorof sources and the sample mean isdetermined in each case. Based on thedistribution of the resulting means,inferences on the underlying precision canbeen drawn. The lower (upper) bound of a90% confidence range is then determined asthe value where 5% of the sample’s meansare below (above) this critical value. Inaddition to the “percentile” method justdescribed, more complicated approachesexist. First, the confidence levels can be

adjusted if (on average) the mean of abootstrap sample is smaller than theobserved mean. The relevant parameter iscalled z0. Another adjustment is to assumethe standard deviation also to be dependenton the mean of the bootstrap sample. Therelevant parameter is a. If both theseadjustments are considered, the resultingapproach is called a bootstrap-BCa-method(bias-corrected-accelerated). A precisedescription of this approach can be obtainedfrom Efron and Tibshirani (1993, chap.14.3, 22.4 and 22.5).6 One concern with theBCa approach is that it is throwing a lot ofmachinery at very few observations. Due tostatistical considerations, a simple methodmight prove superior. Brad Efron hadtherefore suggested the use of a BC-approach for our purpose. In this case, z0 isdetermined endogenously from thebootstrap sample but a is set equal to zero.There are two interesting characteristics ofthe resulting confidence range.

1) When requiring a 90% confidence range(which allows with 5% probability thatthe true value is below and with 5%probability that the value is above thedetermined confidence range) the upper(lower) bound will not be higher (lower)than the highest (lowest) value providedby a source. This implies that our rangefrom 0 to 10 will never be violated.

2) The confidence range remains valid evenif the data (i.e. the standardized valuesfor a given country) are not normallydistributed. The range is even free ofassumptions with regard to thedistribution of these data.

It should not be ignored that confidenceranges cannot be very solid when only veryfew sources are available. This is true forany methodology applied. Regardless ofwhether a normal distribution is assumed ora bootstrap approach is taken, theconfidence range must not be overrated

6 See Efron, B. and R. Tibshirani (1993), AnIntroduction to the Bootstrap, Chapman &Hall: New York and London: 202-219.

when only 3 sources exist. It serves only asa rough guide in this case. Above that, therecan arise boundary effects when only 3 or 4sources exist. Since only 10 differentcombinations are possible in the case of 3sources, a 5% confidence point can “hit”one resulting boundary. If this is the case,the BC-approach could produce at randomtwo different values for the upper (or thelower) confidence point. These boundaryeffects have been identified and, if existent,the more conservative range is reported inthe table.

The resulting confidence range isreported in our publications. It is alsographically illustrated in figure 2. On theweb-sites www.uni-goettingen.de/~uwvwand www.transparency.org we provide thecomplete data for each country: the score,amount of sources contributing, standarddeviation, high-low range, the confidencerange and the amount of independentinstitutions that contributed to an averagevalue.

Appendix: Survey sources for the TICorruption Perceptions Index (CPI) 2002Number 1 2 3

Source Columbia University (CU) Political & Economic Risk Consultancy

Name State Capacity Survey Asian Intelligence IssueYear 2001 2000 2001Internet address http://www.asiarisk.com/Who wassurveyed?

US-resident country experts (policyanalysts, academics and journalists) Expatriate business executives

Subject asked Severity of corruption within thestate

Extent of corruption in away that detracts from the

business environmentfor foreign companies

How do you ratecorruption in terms of itsquality or contribution tothe overall living/working

environment?Number ofreplies 251 1,027 ca. 1,000

Coverage 121 countries 14 countries

Number 4 5 6

Source Institute for Management Development, IMD, Switzerland

Name World Competitiveness YearbookYear 2000 2001 2002Internet address www.imd.ch/wcyWho wassurveyed? Executives in top and middle management; domestic and international companies

Subject asked Bribing and corruption exist in the public sphere Bribing and corruption exist inthe economy

Number ofreplies 4,160 3,678 3,532

Coverage 47 countries 49 countries

Number 7 8

Source World Bank PricewaterhouseCoopers

Name World Business Environment Survey Opacity IndexYear 2001 2001

Internet address www1.worldbank.org/beext/resources/assess-wbessurvey-alt.htm www.opacityindex.com/

Who wassurveyed? Senior managers CFOs, equity analysts, bankers and PwC staff

Subject asked "Frequency of bribing" and "corruption as aconstraint to business"

Frequency of corruption in various contexts(e.g. obtaining import/export permits or

subsidies, avoiding taxes)Number ofreplies 10,090 1,357

Coverage 79 countries7 34 countries

7 The survey was carried out in 81 countries, but data for two countries was insufficient.

Number 9 10 11

Source Economist Intelligence Unit Freedom House World Economic Forum

Name Country Risk Service andCountry Forecast

Nations in Transit Africa CompetitivenessReport

Year 2002 2002 2000Internet address www.eiu.com www.freedomhouse.org www.weforum.org

Who wassurveyed?

Expert staffassessment (expatriate)

Assessment by US academicexperts and FH staff

Senior business leaders;domestic and international

companies

Subject asked

Assessment of thepervasiveness of corruption

(the misuse of public office forprivate or political party gain)

among public officials(politicians and civil servants)

Levels of corruptionHow problematic is

corruption? Are irregular,additional payments

required? In large amounts?

Number ofreplies

Not applicable Not applicable 1,800

Coverage 115 countries 27 transition economies 26 countries

Number 12 13 14

Source World Economic Forum

Name Global Competitiveness ReportYear 2000 2001 2002Internet address www.weforum.orgWho wassurveyed? Senior business leaders; domestic and international companies

Subject asked

Undocumented extra payments connected with import andexport permits, public utilities and contracts, business

licenses, tax payments or loan applications are common/notcommon.

Questions (in addition tothose mentioned left)

refer to paymentsconnected to favorableregulations and judicial

decisionsNumber ofreplies 4,022 ca. 4,600 ca. 4700

Coverage 59 countries 76 countries 80 countries

Number 15

Source Gallup International on behalf ofTransparency International

Name Corruption SurveyYear 2002

Internet address http://www.transparency.org/surveys/index.html#bpi

Who wassurveyed?

Senior businesspeople from 15emerging market economies

Subject asked

How common are bribes topoliticians, senior civil servants, and

judges and how significant of anobstacle are the costs associated withsuch payments for doing business?

Number ofreplies 835

Coverage 21 countries