See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/283360891

Corruption Perception Index 2003

Article · January 2004

CITATIONS READS

38 222

1 author:

Johann Graf Lambsdorff

Universität Passau
94 PUBLICATIONS   2,510 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Experiments in Macroeconomics, Ethics and Anticorruption View project

Determinants of Negative Reciprocity View project

All content following this page was uploaded by Johann Graf Lambsdorff on 11 September 2017.

The user has requested enhancement of the downloaded file.

Background Paper to the
2003 Corruption Perceptions Index
The Corruption Percep-
tions Index is a composite
index. The data used this
year were compiled be-
tween 2001 and 2003. 17
surveys of businesspeople
and assessments by country
analysts from 13 independ-
ent institutions have been
used.

All sources use a homoge-

Framework Document 2003 neous definition of “levels
of corruption”. These per-
ceptions enhance our un-
derstanding of real levels of
corruption from one coun-
try to another.

Comparisons with last

year’s index should be
based on scores. However,
such comparisons can be
misleading because of
methodological changes be-
tween years.

Non-parametric statistics
have been used for stan-
dardizing the data and for
determining the precision
of our results.

Prof. Dr. Johann Graf Lambsdorff

Transparency International (TI) and

University of Passau
September 2003
1. The methodology
Transparency International (TI) publishes • Freedom House Nations in Transit
its annual Corruption Perceptions Index (FH), 2003
(CPI) since 1995. This index has evolved • The Economist Intelligence Unit (EIU),
into a leading indicator in social sciences. 2003
As in previous years, this framework • PricewaterhouseCoopers (PwC), the
document provides an in-depth explanation 2001 Opacity Index.
of the methodology and measurement pre- • The Institute for Management Devel-
cision. opment, Lausanne (IMD). We will use
The goal of the CPI is to provide da- data for 2001-2003.
ta on extensive perceptions of corruption • The Political and Eonomic Risk Consul-
within countries. The CPI is a composite tancy, Hong Kong (PERC). We will for
index, making use of surveys of business- the last time be using the data from
people and assessments by country analysts. 2001.
It consists of credible sources using diverse • The World Bank (WBES), 2001.
sampling frames and different methodolo- • The World Economic Forum (WEF).
gies. These perceptions enhance our un- We will use data for 2001-2003.
derstanding of real levels of corruption • State Capacity Survey by Columbia
from one country to another. University (CU), 2003
As pointed out in previous frame- • Gallup International on behalf of Trans-
work documents, unbiased, hard data con- parency International (GI/TI), BPI
tinue to be difficult to obtain and usually 2002.
raise problematic questions with respect to • Information International (II), 2003.
validity. International surveys on percep- • A Multilateral Development Bank
tions therefore serve as the most credible (MDB), 2002.
means of compiling a ranking of nations. • The Business Environment and Enter-
Overall, 17 sources could be in- prise Performance Survey (BEEPS),
cluded in the 2003 CPI, originating from 13 2002.
independent institutions. The complete list • The World Markets Research Centre
of sources is presented in the appendix. All (WMRC), 2002.
in all, the number of countries in the CPI
increased from 102 last year to 133. An essential condition for inclusion is that a
source must provide a ranking of nations.
Sources in 2003 This condition is not met if a source con-
Prior to selecting sources guidelines have ducts surveys in a variety of countries but
been set up which organize the underlying with varying methodologies. Comparison
decision making process. These include the from one country to another is not feasible
actual criteria that a source needs to meet in in this case and a ranking cannot be pro-
order to qualify for inclusion as well as or- duced. Another condition is that sources
ganizational guidelines on how the final de- must measure the overall level of corrup-
cision is reached with the help of the Trans- tion. This is violated if aspects of corruption
parency International Steering Committee. are mixed with issues other than corruption
This process aimed at making the final de- such as political instability or nationalism
cision as transparent and robust as possible. or if changes are measured instead of levels
As a result of this it was decided that the of corruption.
2003 CPI includes data from the following For example, the index “Corruption
sources: in Government” from the International
Country Risk Guide (ICRG), conducted by
the Political Risk Services (PRS), did not Year-to-year comparisons
meet these requirements, albeit being Comparisons with the results from previous
widely used in research as a measure of years should be based on a country’s score,
levels of corruption. It does not determine a not its rank. A country’s rank can change
country’s level of corruption but the politi- simply because new countries enter the in-
cal risk involved in corruption. As pointed dex and others drop out. A higher score is
out to us by the ICRG-editor, these two is- an indicator that respondents provided bet-
sues can differ considerably, depending on ter ratings, while a lower score suggests that
whether there exists a high or low tolerance respondents revised their perception down-
towards corruption. Corruption only leads wards. However, year-to-year comparisons
to political instability if it is not tolerated. of a country's score result not only from a
Due to this, the data by PRS-ICRG did not changing perception of a country's perform-
qualify for inclusion in the CPI. However, ance but also from a changing sample and
TI hopes to include a modified set of data
methodology. With differing respondents
by PRS in the future. and slightly differing methodologies, a
The 2003 CPI combines assessments change in a country's score may also relate
from the past three years to reduce abrupt to the fact that different viewpoints have
variations in scoring that might arise due to been collected and different questions been
random effects. Some sources, such as II, asked. The index primarily provides an an-
TI/GI, BEEPS, WBES and PwC, provided nual snapshot of the views of businesspeo-
only one recent survey. Others such as ple, with less of a focus on year-to-year
WEF and IMD provided various surveys trends.
between 2001 and 2003, which are all in- However, to the extent that changes
cluded. can be traced back to a change in the results
While this averaging is valuable for from individual sources, trends can cau-
the inclusion of surveys, it is inappropriate tiously be identified. Noteworthy examples
for application to the data compiled by of a downward trend between 2002 and
country experts. Such assessments as com- 2003 are Argentina, Belarus, Chile, Canada,
piled by WMRC, FH, CU and EIU are con- Israel, Luxembourg, Poland, USA, and
ducted by a small number of country ex- Zimbabwe. The considerable decline in
perts who regularly analyze a country's per- their scores does not result from technical
formance, counterchecking their conclu- factors – actual changes in perceptions are
sions with peer discussions. Following this therefore likely.
systematic evaluation, they then consider a With the same caveats applied, on
potential upgrading or downgrading. As a the basis of data from sources that have
result, a country's score changes rather sel- been consistently used for the index, im-
dom and the data shows little year-to-year provements can be observed for Austria,
variation. Changing scores in this case are Belgium, Colombia, France, Germany, Ire-
the result of a considered judgment by the land, Malaysia, Norway, and Tunisia.
organization in question. To then go back
and average the assessments over a period
of time would be inappropriate. On the 2. Validity
other hand, in the case of elite surveys an All sources generally apply a definition of
averaging over various years produces a corruption such as the misuse of public
useful smoothing effect. While some annual power for private benefit, for example brib-
data may contain random errors, these do ing of public officials, kickbacks in public
not necessarily carry over into the next procurement, or embezzlement of public
year. funds. Each of the sources also assesses the
“extent” of corruption among public offi-
cials and politicians in the countries in problematic are these different factors
question: for the operation and growth of your
business: … corruption…”
• The IMD asks respondents to assess • MDB asks its staff to assess multiple
whether “bribing and corruption prevail countries with respect to the following
or do not prevail in the economy.” questions: “How widespread is the inci-
• The PERC asks “How do you rate cor- dence of corruption? (Widespread;
ruption in terms of its quality or contri- Somewhat widespread; Somewhat lim-
bution to the overall living/working en- ited; Limited; No judgement).
vironment?” A slightly different ques- • The WEF asks in its 2003 Global Com-
tion had been asked previously. petitiveness Report: 7. “In your industry,
• The EIU defines corruption as the mis- how commonly would you estimate that
use of public office for personal (or firms make undocumented extra pay-
party political) financial gain and aims at ments or bribes connected with:”
measuring the pervasiveness of corrup- 1 – exports and imports
tion. Corruption is one of over 60 indica- Common |1|2|3|4|5|6|7| Never occur
tors used to measure “country risk” and 2 - public utilities (e.g. telephone or
“forecasting.” electricity)
• PwC asks for the frequency of corrup- Common |1|2|3|4|5|6|7| Never occur
tion in various contexts (e.g. obtaining 3 - annual tax payments
import/export permits or subsidies, avoi- Common |1|2|3|4|5|6|7| Never occur
ding taxes). 4 – public contracts
• FH determines the "level of corruption" Common |1|2|3|4|5|6|7| Never occur
without providing further defining sta- 5 - loan applications
tements. Common |1|2|3|4|5|6|7| Never occur
• The WBES asks two questions with re- 6 - influencing laws and policies, regula-
spect to corruption, one determining the tions, or decrees to favor selected busi-
"Frequency of bribing" and another one ness interests?
relating to "corruption as a constraint to Common |1|2|3|4|5|6|7| Never occur
business". 7 – getting favorable judicial decisions
• Columbia University asks for the se- Common |1|2|3|4|5|6|7| Never occur.
verity of corruption within the state. From these questions the simple average
• WMRC assesses the amount of red tape has been determined.
likely to be encountered, as well as the • Gallup International on behalf of Trans-
likelihood of encountering corrupt offi- parency International (TI/GI) and simi-
cials and other such groups. The types of larly Information International ask:
corruption covered range from small- 13. Which are the countries, besides this
scale bribes right through to larger-scale one, with which you have had the most
kickbacks business experience in the last 3-5
• BEEPS asks “Thinking about officials years? Please name up to five countries.
… It is common for firms in my line of 13a. In [country 1], how common are
business to have to pay some irregular payments (e.g. bribes) to obtain or retain
‘additional payments’ to get things business or other improper advantages to
done”. (Always, Mostly, Frequently, senior public officials, like politicians,
Sometimes, Seldom, Never, Don’t senior civil servants, and judges?
know)” and “Using this scale (No Ob- In [country 1], how significant of an ob-
stacle=1 ; Minor Obstacle=2 ; Moderate stacle are the costs associated with such
Obstacle=3 ; Major Obstacle=4 ; Don’t payments for doing business? From
know/no answer=5) can you tell me how ‘Very significant [1] to ‘Insignificant’
[4]. Don’t know [88]. The questions
 continue for countries 2-5. sample residents (sometimes also from mul-
In [country 1], how frequently are pub- tinational companies). In contrast, the data
lic contracts awarded to business associ- by PERC, FH, TI/GI, II, MDB, CU,
ates, friends and relatives rather than on WMRC, and EIU largely relate to expatri-
a competitive bidding basis?” ates. Whether this difference between sam-
Scale for answers are from ‘Very Com- ples may lead to different outcomes still re-
mon [01] to ‘Very Uncommon / quires scientific study. For the purposes of
Never’[04]. Don’t know [88]. the CPI it added to the robustness of the re-
sulting figures, because the data correlate
The terms "prevalence", "common- well, irrespective of whether expatriates or
ness", "frequency", "constraint ", "contribu- residents had been polled. This correlation
tion to working environment" and “sever- suggests that there being different samples
ity” are closely related. They all refer to makes no large difference to the results.
some kind of “degree” of corruption, which
is also the aim of the CPI. This common Interpreting perceptions
feature of the various sources is particularly As the data collected relates to perceptions
important in view of the fact that corruption rather than to real phenomena, it has to be
comes in different forms. It has been sug- considered whether such perceptions im-
gested in numerous publications that dis- prove our understanding of what real levels
tinctions should be made between these of corruption may be. Since actual levels of
forms of corruption, e.g. between nepotism corruption cannot be determined directly,
and corruption in the form of monetary perceptions may be all we have to guide us.
transfers. Yet, none of the data included in However, this approach is undermined, to at
the CPI emphasize one form of corruption least some extent, if the perceptions gath-
at the expense of other forms. The sources ered are biased. Such a potential bias might
can be said to aim at measuring the same originate from the particular cultural back-
phenomenon. As also emphasized in the ground of respondents. Depending on
framework documents of previous years, whether the sample consist of locals or ex-
the sources do not distinguish between ad- patriates, this suggests two potential biases
ministrative and political corruption. to be relevant.
The term "degree of corruption" may Imagine that being asked to assess
imply different things. In particular, it may the level of corruption, a local estimates a
relate to the frequency of bribes or the size high level of corruption in the country of
of bribes. But we know from the results of residence. Such an assessment would be a
our sources that frequency and the size of valid contribution to the CPI only if the re-
bribes tend to correlate highly (as ex- spondent makes the assessment as a result
pounded in the framework documents of of comparisons with the levels of corruption
previous years). In countries where corrup- perceived in other countries. But this is not
tion is frequent it also amounts to a high necessarily the viewpoint taken by the re-
fraction of firms' revenues. In sum, the term spondent. A respondent may also assign
“degree of corruption” seems to equally re- high levels by comparing corruption to
flect the two aspects, frequency of corrup- other (potentially less pressing) problems
tion and the total value of bribes paid. facing the country, or by evaluating it ac-
cording to a high ethical standard (e.g.
3. Samples, perceptions and reality which assumes any kind of gift-giving to a
public official to be corrupt and not cultur-
While the sources all aim at measuring the ally acceptable). In the case of such an out-
degree of corruption, the sample design dif- look, a high degree of observed corruption
fers considerably. The data by IMD, may reflect a high standard of ethics rather
WBES, BEEPS, PwC and WEF largely than a high degree of real misbehavior. Per-
ceptions would be a misleading indicator Because the data from the EIU,
for real levels of corruption. This bias can MDB, CU, FH and PERC correlate well
occur particularly if only locals are sur- with the other data, there seems to be no
veyed, each assessing only the level of per- support for the suggestion that they might
ceived corruption in their own countries. If be influenced by the second type of bias.
respondents are asked to assess foreign Similarly, the data by the IMD, WEF,
countries or to make comparisons between BEEPS, PwC and WBES correlate well
a variety of countries, this bias should not with data from the first group; the notion
occur. Respondents will, in this case, com- that the first type of bias might be present is
pare a foreign country with their home clearly not supported. The validity of the
country or with an even larger set of coun- sources is mutually confirmed and no hint
tries. They will be forced to apply the same is found for the existence of a bias in our
definition of corruption and make use of the data.
same ethical standard for all countries, A third group of more recently in-
which produces valid comparative assess- cluded sources is less susceptible to both
ments. biases simultaneously: these are TI/GI, II
However, in this context a second and MDB. They survey either their staff
type of bias might arise, originating from members (in the case of MDB) or respon-
the potential dominance of a particular cul- dents from emerging economies and less
tural heritage in the sample questioned or developed countries (TI/GI and II). In the
because expatriates lack a proper under- latter case respondents are asked to assess
standing of a country's culture. The results the performance of industrial countries and
would be meaningless to locals if they have neighboring countries. Those polled are not
a different understanding and definition of asked to assess their home country or indi-
corruption. While samples of expatriates are vidual foreign countries, but to provide a
susceptible to this kind of bias, surveys that comparative assessment of various foreign
question local residents clearly avoid it. countries. This approach makes sure that a
The strength of the CPI rests with consistent ethical standard is applied to all
the idea that we include surveys that are not countries, that only those countries are as-
susceptible to the first type of bias, in sessed where sufficient experience and cul-
particular EIU, WMRC, CU, FH and tural insights are available and that the
PERC. Because the data provided by these viewpoint of less developed countries is
sources refer to assessments by expatriates, well represented. Yet, as shown in the cor-
they are subject to a homogeneous relations, this different approach does not
definition of corruption and a consistent bring about noteworthy different results.
ethical standard. EIU, WMRC, CU and FH Thus, the comparative assessments gathered
are assessments carried out by country ana- in the CPI do not disproportionately reflect
lysts (mostly from North America and the perceptions of western businesspeople.
Western Europe). These assessments are In sum, it seems that residents tend
discussed and reviewed by peers in order to to have a consistent ethical standard with
guarantee consistency across countries. regard to assessments of corruption, while
The CPI also incorporates the data expatriates do not tend to impose an inap-
from the IMD, WEF, BEEPS, PwC and propriate ethical standard or to lack cultural
WBES. These ratings are less likely to rep- insights. The approach suggests that the
resent the perception of a certain cultural perceptions gathered are a helpful contribu-
heritage because they refer to assessments tion to the understanding of real levels of
made by local residents. The second type of corruption.1
bias can clearly be rejected for these
sources. 1
As was also explained in detail in the
2001 framework document, the perceptions
4. The index scores 8.7, 9.3, 2.5 and 2.8. Matching
percentiles would now assign UK the best
Standardizing score of 9.3, Singapore 8.7, Venezuela 2.8
Each of the sources uses its own scaling and Argentina 2.5.
system, requiring that the data be standard- Matching percentiles is superior in
ized before each country’s mean value can combining indices that have different dis-
be determined. This standardization is car- tributions. But, as it makes use of the ranks,
ried out in two steps. and not the scores of sources, this method
For step 1 each source is standardized loses some of the information inherent in
using matching percentiles. The ranks (and the sources. What tips the balance in favor
not the scores) of countries is the only in- of this techniques is its capacity to keep all
formation processed from our sources. For reported values within the bounds from 0 to
this technique the common sub-samples of 10: All countries in the CPI obtain scores
a new source and the previous year’s CPI between 0 (very corrupt) and 10 (highly
are determined. Then, the largest value in clean). This characteristic is not obtained by
the CPI is taken as the standardized value an alternative technique that standardizes
for the country ranked best by the new the mean and standard deviation of the sub-
source. The second largest value is given to samples. Matching percentiles, on the other
the country ranked second best, etc.2 Imag- hand, guarantees that all standardized val-
ine that a new source ranks only four coun- ues are within these bounds. This results
tries: UK is best, followed by Singapore, because any standardized value is taken
Venezuela and Argentina respectively. In from the previous year’s CPI, which by
the 2002 CPI these countries obtained the definition is restricted to the aforemen-
tioned range.
In sum, matching percentiles has the
gathered relate well to actual experience disadvantage of wasting some information
made and less to hearsay. See Lambsdorff, by processing only the ranks reported by
J. Graf (2001) “Framework Document”, sources. Yet, this disadvantage is offset be-
Background Paper 2001 Corruption Percep- cause 1) the approach is free of assumptions
tions Index: regarding the distribution of sources, and 2)
http://wwwuser.gwdg.de/~uwvw/download all standardized values remain within the
s/2001_CPI_FD.pdf range from 0 to 10.
2
In case two countries share the same rank,
their standardized value is the simple mean Step 2
of the two respective scores in the CPI. The
scores for countries where no CPI value Having obtained standardized values that
was available are determined by referring to are all within the reported range, a simple
neighbor countries in the source’s ranking. average from these standardized values can
Linear interpolation is applied to their sco- be determined. As already argued before,
res, suggesting that if a source assigns such the resulting index has a standard deviation
a country a score close to the upper which is smaller than that of the CPI of
neighbor, also its standardized value is previous years. Without a second adjust-
closer to that of this neighbor. If such a ment there would be a trend towards a con-
country is ranked best (or worst) by a tinuously smaller diversity of scores. If,
source it would have only one neighbor, not e.g., Finland were to repeat its score from
two. The second neighbor is constructed by the previous year, it would have to score
using the highest (or lowest) attainable best in all sources listing this country. If it
score by the source and the CPI value 10 scores second to best in any source, the
(or 0). This approach guarantees that all standardized value it obtains after using
values remain within the range between 10 matching percentiles and aggregation would
and 0. be lower than its current score. Thus, given
some heterogeneity
among sources, it
seems inevitable that
Finland’s score would
deteriorate. The oppo-
site would be true of
Bangladesh, which
would obtain a better
score if it is not con-
sistently rated worst
by all its sources. A
second standardization
is required in order to
avoid a continuous
trend to less diversity
among scores.
However,
applying a simple
mean and standard
deviation technique
might again bring score of 0 is not further decreased. This
about
are beyond
values
ourthat
range
arefrom 0 to 10. A more guarantees that all values remain within the
complicated standardization is required for range.
the second step: A beta-transformation. The The beta transformation is first ap-
idea behind this monotonous transformation plied to all values that were standardized in
is to increase the standard deviation to its step 1. Afterwards the average of these are
desired value, but to keep all values within computed to determine a country’s score. In
the range from 0 to 10. Each value (X) is our publication we also report the high-low
therefore transformed according to the fol- range. This refers to all standardized values
lowing function: after carrying out the beta-transformation.
1
This procedure ensured that the high-low
10 ∗ ( X / 10) α −1 (1 − X / 10) β −1 dX range is consistently related to a country’s
0
mean value. All these values remain within
This beta-transformation is available in the range from 0 to 10.
standard statistics programs. The crucial
task is to find the parameters α and β so Reliability and Precision
that the resulting mean and standard devia-
tion of the index have the desired values. A ranking of countries may easily be mis-
An algorithm has been determined that car- understood as measuring the performance
ries out this task. Applying this approach to of a country with absolute precision. This is
the 2003 CPI, the change in the scores is certainly not true. Since its start in 1995 TI
has provided data on the standard deviation
depicted by figure 1. The parameters are α=
and the amount of sources contributing to
1.24 and β=1.23. As shown in the figure,
the index. This data already serves to illus-
scores between 5 and 10 are increased
trate the inherent imprecision. Also the
slightly, while those between 0 and 5 are
high-low range is provided in the main ta-
lowered.
ble. This depicts the highest and the lowest
This effect makes sure that the pre-
values provided by our sources, so as to
vious standard deviation is preserved. Yet,
portray the whole range of assessments.
once a score of 10 has been reached, the
However, no quick conclusions should be
score is not further increased. Equally, a
derived from this range to the underlying
precision with which countries are meas- methods are required.
ured. Countries which were assessed by 3 The strength of the CPI is based on
or 12 sources can have the same minimum the concept that a combination of data
and maximum values, but in the latter case sources combined into a single index in-
we can feel much more confident about the creases the reliability of each individual
country’s score. In order to arrive at such figure. As in previous years, the 2003 CPI
measures of precision, other statistical includes all countries for which at least
three sources had been available. The idea tween a value of 1.7 and 7.7. In contrast,
of combining data is that the nonperform- the low standard deviation for Luxembourg
ance of one source can be balanced out by of 0.4 means that 95% of the scores range
the inclusion of at least two other sources. between 7.9 and 9.5.
This way, the probability of misrepresent-
ing a country is seriously lowered. This is Confidence range
valid even in case the sources are not totally We have been providing readers with the
independent of each other. Such partial de- information on the confidence range for
pendency may arise if some respondents are some years now. Up to 2001 this was based
aware of other people's perception of the on the determination of the standard error
level of corruption, or of other sources con- for a country’s average score and a result-
tributing to the CPI. ing parametric assessment of a 95% confi-
An indicator for the overall reliability dence range. This approach required the as-
of the 2003 CPI can be drawn from the high sumption that there is no imprecision asso-
correlation between the sources. This can ciated with the source’s values and that
be depicted from the standard Pearson cor- these values are independent of each other.
relation and Kendall’s rank correlation, Another strong assumption required is that
provided in tables 1 and 2. These data refer errors are normally distributed. While it is
to all countries, even those not included in statistically difficult to relax the first two
the CPI.3 The correlations on average are assumptions, one can relax the assumption
0.80 for the Pearson correlation and 0.65 of a normal distribution and apply tests
for Kendall’s rank correlation. The sources which are valid throughout any type of dis-
do not differ considerably in their assess- tribution. Another drawback of the older
ment of levels of corruption. confidence ranges was, again, that they
In addition to these correlations, the sometimes violated the given range from 0
reliability of each individual country score to 10. For example, while in 2001 Bangla-
can be determined. The larger the number desh had a score of 0.4, its 95% confidence
of sources and the lower the standard devia- range was between -3.6 and 4.4. For
tion between the sources, the more reliable Finland, on the other hand, the range went
is the value for a country. The relatively as high as 10.4. This type of a range is con-
large standard deviation for Lithuania of 1.6 fusing even to an expert. Since it is in con-
signifies that 95% of the sources range be- tradiction to the official range reported, the
public is equally disoriented.
3
Abbreviations are: BEEPS: Business En- In order to restrict the confidence
vironment and Enterprise Performance Sur- range to our pre-specified limits, since 2002
vey; CU: Columbia University; EIU: we apply a different methodology: a non-
Economist Intelligence Unit; FH: Freedom parametric approach applying the bootstrap
House, Nations in Transit; II: Information methodology. The principal idea of such a
International; IMD: World Competitiveness bootstrap confidence range is to resample
Report of the Institute for Management De- the sources of a country with replacement.
velopment; MDB: A Multinational Devel- If five source values (3, 5, 4, 4.5, 4.2) had
opment Bank; PERC: Political and Eco- been given, an example of such a sample
nomic Risk Consultancy, Hong Kong; would be (5, 5, 4.2, 3, 3). A sufficiently lar-
PwC: PricewaterhouseCoopers, Opacity ge number of such samples (in our case
Index; TI/GI: Gallup International on behalf 10,000) are drawn from the available vector
of Transparency International; WBES: of sources and the sample mean is deter-
World Business Environment Survey of the mined in each case. Based on the distribu-
World Bank; WEF: Global Competitive- tion of the resulting means, inferences on
ness Report of the World Economic Forum; the underlying precision can be drawn. The
WMRC: World Markets Research Centre lower (upper) bound of a 90% confidence
range is then determined as the value where It should not be ignored that confidence
5% of the sample’s means are below ranges cannot be very solid when only very
(above) this critical value. In addition to the few sources are available. This is true for
“percentile” method just described, more any methodology applied. Regardless of
complicated approaches exist. First, the whether a normal distribution is assumed or
confidence levels can be adjusted if (on av- a bootstrap approach is taken, the confi-
erage) the mean of a bootstrap sample is dence range must not be overrated when
smaller than the observed mean. The rele- few sources exist. It serves only as a rough
vant parameter is called z0. Another adjust- guide in this case. Above that, there can
ment is to assume the standard deviation arise boundary effects when only 3 or 4
also to be dependent on the mean of the sources exist. Since only 10 different com-
bootstrap sample. The relevant parameter is binations are possible in the case of 3
a. If both these adjustments are considered, sources, a 5% confidence point can “hit”
the resulting approach is called a bootstrap- one resulting boundary. If this is the case,
BCa-method (bias-corrected-accelerated). A the BC-approach could produce at random
precise description of this approach can be two different values for the upper (or the
obtained from Efron and Tibshirani (1993, lower) confidence point. These boundary
chap. 14.3, 22.4 and 22.5).4 One concern effects have been identified and, if existent,
with the BCa approach is that it is throwing the more conservative range is reported in
a lot of machinery at very few observations. the table.
Due to statistical considerations, a simple The resulting confidence range is
method might prove superior. Brad Efron reported in our publications. It is also
had therefore suggested the use of a BC- graphically illustrated in figure 2. On the
approach for our purpose. In this case, z0 is web-sites www.gwdg.de/~uwvw and
determined endogenously from the boot- www.transparency.org we provide the
strap sample but a is set equal to zero. complete data for each country: the score,
There are two interesting characteristics of amount of sources contributing, standard
the resulting confidence range. deviation, high-low range, the confidence
range and the amount of independent insti-
1) When requiring a 90% confidence range tutions that contributed to an average value.
(which allows with 5% probability that
the true value is below and with 5%
probability that the value is above the
determined confidence range) the upper
(lower) bound will not be higher (lower)
than the highest (lowest) value provided
by a source. This implies that our range
from 0 to 10 will never be violated.

2) The confidence range remains valid even

if the data (i.e. the standardized values
for a given country) are not normally
distributed. The range is even free of as-
sumptions with regard to the distribution
of these data.

4
See Efron, B. and R. Tibshirani (1993), An
Introduction to the Bootstrap, Chapman &
Hall: New York and London: 202-219.
Survey sources for the TI Corruption Perceptions Index (CPI) 2003

Number 1 2 3
Source World Economic Forum
Name Global Competitiveness Report
Year 2001 2002 2003
Internet address www.weforum.org
Who was sur-
Senior business leaders; domestic and international companies
veyed?
Undocumented extra payments
connected with import and ex-
In addition to questions mentioned on the left:
port permits, public utilities and
Subject asked payments connected to favorable regulations and ju-
contracts, business licenses, tax
dicial decisions
payments or loan applications
are common/not common.
Number of re-
4,022 ca. 4,600 7,741
plies
Coverage 59 countries 76 countries 102 countries

Number 4 5 6
Source Institute for Management Development, IMD, Switzerland
Name World Competitiveness Yearbook
Year 2001 2002 2003
Internet address www.imd.ch
Who was sur-
Executives in top and middle management; domestic and international companies
veyed?
Bribing and corruption exist
Subject asked Bribing and corruption exist in the economy
in the public sphere
Number of re-
3,678 3,532 > 4,000
plies
Coverage 49 countries 51 countries

Number 7 8
Source Information International World Bank
Name Survey of Middle Eastern Businesspeople World Business Environment Survey
Year 2003 2001
info.worldbank.org/governance/wbes/index
Internet address www.information-international.com
1.html
Who was sur- Senior businesspeople from Bahrain, Lebanon
Senior managers
veyed? and UAE
How common are bribes, how costly are they
for doing business and how frequently are pub- "Frequency of bribing" and "corruption as a
Subject asked
lic contracts awarded to friends and relatives in constraint to business"
neighboring countries
Number of re-
382 assessments from 165 respondents 10,090
plies
5
Coverage 31 countries 79 countries

5
The survey was carried out in 81 countries, but data for two countries was insufficient.
Number 9 10 11
Economist Intelligence Unit Freedom House World Markets Research
Source
Centre
Country Risk Service and
Name Nations in Transit Risk Ratings
Country Forecast
Year 2003 2003 2002
Internet address www.eiu.com www.freedomhouse.org www.wmrc.com
Who was sur- Expert staff Assessment by US academic
Assessment by staff
veyed? assessment (expatriate) experts and FH staff
Assessment of the pervasive- Perception of corruption in Red tape and the likelihood
ness of corruption (the misuse the civil service, the business of encountering corrupt
of public office for private or interests of top policy mak- officials. This includes
Subject asked
political party gain) among ers, laws on financial disclo- small-scale bribes, larger-
public officials (politicians and sure and conflict of interest, scale kickbacks and corpo-
civil servants) and anticorruption initiatives. rate fraud.
Number of re-
Not applicable Not applicable Not applicable
plies
Coverage 139 countries 27 transition economies 186 countries

Number 12 13 14
Political & Economic Risk
Source Columbia University (CU) PricewaterhouseCoopers
Consultancy
Name State Capacity Survey Asian Intelligence Issue Opacity Index
Year 2002 2001 2001
Internet address http://www.asiarisk.com/ www.opacityindex.com/
US-resident country experts
Who was sur- Expatriate business execu- CFOs, equity analysts, bank-
(policy analysts, academics
veyed? tives ers and PwC staff
and journalists)
Frequency of corruption in
How do you rate corruption
various contexts (e.g. ob-
Severity of corruption within in terms of its quality or con-
Subject asked taining import/export per-
the state tribution to the overall liv-
mits or subsidies, avoiding
ing/working environment?
taxes)
Number of re-
224 ca. 1,000 1,357
plies
Coverage 95 countries 14 countries 34 countries
 Number 15 16
Gallup International on behalf of
Source A Multilateral Development Bank
Transparency International
Name Survey Corruption Survey
Year 2002 2002
www.transparency.org/surveys/inde
Internet address
x.html#bpi
Experts within the bank were identi-
fied and multiple questionnaires
Who was sur- Senior businesspeople from 15 e-
(each relating to a different country)
veyed? merging market economies
were sent out to them. Roughly 40%
of the questionnaires were returned.
“How common are bribes to politi-
How widespread is the incidence of
cians, senior civil servants, and
corruption? (Widespread; Somewhat
Subject asked judges” and “how significant of an
widespread; Somewhat limited;
obstacle are the costs associated with
Limited; No judgment)
such payments for doing business?”
Number of re-
398 835
plies
Coverage 47 countries 21 countries

Number 17
Source World Bank and the EBRD
Business Environment and Enter-
Name
prise Performance Survey
Year 2002
info.worldbank.org/governance/beep
Internet address
s2002/
Who was sur-
Senior businesspeople
veyed?
Frequency of irregular “additional
payments”; how is corruption for the
Subject asked
operation and growth of your busi-
ness?
Number of re-
6500
plies
Coverage 25 transition countries

View publication stats