The World Bank Economic Review, 31(1), 2017, 196–219 doi: 10.1093/wber/lhv031 Article A New Cross-National Measure of Corruption Laarni Escresa and Lucio Picci Abstract A new measure of cross-national corruption is constructed based on the geographic distribution of public offi- cials involved in cross-border corruption cases. A comparison is made between the Public Administration Corruption Index (PACI) and perception-based measures, considers the extent to which differences between them are driven by systematic factors, and concludes that they are not. As more data on cases of cross-border bribery incidents become available, the PACI will provide an increasingly valid cross-national measure of corruption. JEL classification: C18, C43, D73, F53, F55, H11, H50, K42 Key words: Corruption statistics, Causes of corruption, Corruption index This research presents a new measure of corruption, the Public Administration Corruption Index (PACI), formulated using data on cross-border corruption cases. The proposal is motivated by the need to find viable alternatives to currently available cross-national measures of corruption, which, despite their shortcomings, have been and are being extensively used in academia,1 within policy circles, and in public debates at large. There are two types of cross-national measures of corruption currently available. The most widely used type is perception-based, such as the Transparency International Corruption Perceptions Index (TI- CPI; Saisana and Saltelli 2012; Transparency International 2012) and the World Bank Control of Corruption Indicator (WB-CCI; Kaufmann, Kraay, and Mastruzzi 2009). The shortcomings of the cited type, however, are well known. For one, perceptions may have weak correlations with actual experien- ces of corruption (Seligson 2006; Olken 2009; Razafindrakoto and Roubaud 2010). Moreover, often the precise definition of corruption being assessed is not precisely defined. The second type of cross-national measure, meanwhile, is based on surveys assessing first-hand experiences of corruption.2 Surveys, Laarni Escresa is an assistant professor at the School of Economics, University of the Philippines; her email is laarni.escre- sa@eui.eu. Lucio Picci (corresponding author) is a professor at the Department of Economics, University of Bologna, Strada Maggiore 45, I-40125 Bologna, Italy; his email is lucio.picci@unibo.it. They would like to thank, for their comments on a previ- ous version of this paper, Rajeev Goel, Miriam Golden, Johann Graf Lambsdorff, Jerg Gutmann, Robert Klitgaard, Nicholas McLean, Angela Reitmaier, Susan Rose-Ackerman, participants to a workshop at the Inter-American Development Bank in Washington, DC, to the European Political Science Association 4th Annual Conference in Edinburgh, and to the 10th Annual Conference of the Italian Society of Law and Economics in Rome and three anonymous referees. Laarni Escresa acknowledges research funding from the Deutsche Forschungsgemeinschaft (German Research Foundation). Lucio Picci acknowledges the fi- nancial support of the Fondazione Cassa dei Risparmi di Forlı. All Website contents were accessed on June 8, 2015. 1 On the use of perception-based measures to inquire into the nature, causes and consequences of corruption, see Lambsdorff (2006 and 2007), Treisman (2007), and Rose-Ackerman (1999). 2 A well-known example is Transparency International’s “Global Corruption Barometer” (TI-GCB). See http://www. transparency.org/research/gcb/overview. C The Author 2015. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. V All rights reserved. For permissions, please email: journals.permissions@oup.com The World Bank Economic Review 197 however, are often costly endeavours, and their results are affected by respondents’ reticence in answer- ing questions related to their participation in corrupt activities.3 Proposing to use judicial statistics to develop a cross-national measure of corruption may appear ungainly. Even setting aside differences in legal definitions across jurisdictions, the fraction of observed corrupt transactions relative to their actual number is unknown and varies widely across countries. These differences could be so important that they could even imply a negative correlation between actual and observed corrupt transactions. After all, where corruption is endemic, the judiciary may also be cor- rupt or vulnerable to threats (Van Aaken et al. 2010).4 However, judicial statistics on cross-border corruption, which refers to corrupt transactions between firms headquartered in a particular country (hence, the “headquarters country”) and public officials else- where (the “foreign country”) permit the computation of a valid cross-national corruption index. A valid index in the lexicon of this article means one that increases along with the probability that a transaction is corrupt. The intuition behind such index is simple. To illustrate, the instances of corruption involving US firms caught bribing public officials abroad are not informative of the level of corruption in the United States; however, the distribution of these cases, with respect to the nationalities of the foreign public officials involved, reveals the relative level of corruption abroad. For example, if half of all the cases concerning US firms are seen to involve Chinese public officials, enough evidence would be pro- vided that public sector corruption in China is relatively high. Obviously, such a conclusion should take into account the extent of the subject countries’ bilateral transactions (the US interacts more often with China than, say, Denmark). As a suitable proxy for the number of bilateral transactions that could be vulnerable to corruption, export figures from the headquarters to the foreign country are used. In other words, the proposed measure uses information on the spatial distribution of cases enforced in a given country to evaluate levels of corruption in all other countries. Moreover, it considers cases arising not only from a single jurisdiction, but from all relevant ones.5 The necessary data are derived from reports of cross-border corruption ensuing from the criminaliza- tion of foreign bribery in several jurisdictions. With the passage of the Foreign Corrupt Practices Act (FCPA) of 1977, the United States became the first country to criminalize foreign bribery. On 15 February 1999, the Organisation for Economic Cooperation and Development (OECD) Anti-Bribery Convention came into force, requiring signatory countries to legislate measures intended to combat for- eign bribery in their own jurisdictions. To date, 41 countries have signed the Convention, and hundreds of cases of foreign bribery have been investigated worldwide.6 The number of observed cases, however, is not sufficient for the computation of a reliable annual measure. In light of this, this paper consolidated data culled from corruption cases reported over a 15- year period or from 1998 to 2012. However, as more and more countries sign the OECD Anti-Bribery Convention, and as more cases of cross-border corruption are exposed, there will arise the possibility of computing the PACI for shorter periods of time. Increased data availability would also enable the com- putation of more granular measures, which pertain to specific sectors of the economy or particular seg- ments of public administrations. 3 See for instance Treisman (2007), Clausen et al. (2011), and Kraay and Murell (2013). Other types of measures of corrup- tion have been proposed, but so far not at the cross-national level. See for example Golden and Picci (2005). 4 Judicial statistics have been used to measure corruption in studies at the sub-national level, where the assumption of spa- tially homogenous enforcement is plausible. See for instance, Glaeser and Saks (2006), Goel and Nelson (2011), Fisman and Gatti (2002), Alt and Lassen (2012), for the US; Chang et al. (2010), and Golden and Picci (2008). 5 This paper is arguably the first to propose an index of corruption based on cross-border incidents of corruption. McLean (2012) contemplates the possibility of computing such an index; see in particular his Figure 1. 6 See Carr and Outhwaite (2008) and http://www.oecd.org/daf/anti-bribery/oecdantibriberyconvention.htm. 198 Escresa and Picci Table 1. An Illustrative Example HQ FO (1) Cases, first (2) Cases, as percent of (3) Exports HQ! FO Ratio Col. (2)/Col. (3) pursued in HQ total no. cases first pursued in HQ as percent of total exports of HQ USA China 65 20.63% 9.29% 2.22 USA Austria 1 0.32% 0.31% 1.02 Germany China 4 7.14% 3.96% 1.96 Germany Austria 3 5.36% 5.62% 0.95 Notes: Cases reported are those first pursued in the headquarters country (HQ), 1998–2012. FO: Foreign country. The total number of cases first pursued in the United States is 315, while the total number of cases first pursued in Germany is 56. Data sources are reported in Appendix B. The PACI is described in the next section. Section II introduces the dataset and presents the computa- tion of the index. Section III shows that the PACI is highly correlated with existing perception-based measures of corruption and considers the determinants of the observed differences. Section IV discusses available evidence regarding the assumptions that ensure the validity of the index, which are described in detail in Appendix A. Section V concludes. I. The Public Administration Corruption Index To illustrate the mechanics of the PACI, it is necessary at the outset to distinguish cases of cross-border corruption (involving firms from the headquarters country i bribing public officials in the foreign coun- try j) based on where they were enforced first, that is, the country whose judiciary was the first to take action on a particular corruption case. The vast majority of cross-border corruption cases were first enforced in headquarters countries, comprised mostly of developed nations. This is expected, because the FCPA and the OECD Anti-Bribery Convention were envisioned precisely to address the lax policing of corruption in many foreign countries where multinational firms operate. It is also possible that cor- ruption cases are first enforced in a third country, mainly because of broad interpretations of the extent of the jurisdiction of the United States, which covers companies registered in its Securities and Exchange Commission (SEC) and, generally, entities carrying out their businesses in the United States.7 The intuition behind the PACI is best described by means of a simple example, illustrated in Table 1. The available data (to be presented in detail in Section III) indicate the presence of 315 cases of cross- border alleged corruption, first enforced in the United States and involving firms headquartered in the same country. If the public officials of all countries trading with the United States were equally corrupt, the 315 cases could be expected to be distributed according to the number of bilateral transactions, which can be proxied by bilateral trade flows. Exports to China represented 9.29 percent of total US exports, implying an expected number of 29.26 cases (Table 1). However, the number of actual observed occurrences of corruption is 65, or 20.63 percent of the total. Thus, the number of cross-border corrup- tion cases involving Chinese public officials was 2.22 times more than what could be expected if the total number of cases first enforced in the United States, and involving firms headquartered there, were dis- tributed geographically according to the countries’ shares in US exports (see the last column of Table 1). This provides evidence that the level of corruption exhibited by public officials in China is higher than the average level demonstrated by the public officials of all trade partners of the United States. In comparison, Austria was pegged as the destination of 0.31 percent of total US exports (Table 1). It registered one reported case of bilateral corruption, corresponding to 0.32 percent of the total. Based on 7 After a case is first enforced in the headquarters country or in a third-country jurisdiction, it may also fall under the radar of the foreign country’s judiciary. The World Bank Economic Review 199 these figures, the level of corruption demonstrated by Austrian public officials is shown to be close to the average level, with the ratio tallied at 1.02. Similarly, occurrences of corruption from the point of view of other countries may be considered. Germany, for instance, could be designated as the headquarters country. The available data indicate that there were four cases involving Chinese public officials, out of a total of 56 documented cases of corrup- tion involving firms headquartered in Germany (and first enforced in the same country). The tally would also lead to the conclusion that China’s public officials are more corrupt than the average trade partners of Germany. Meanwhile, Austria registered fewer cases of cross-border corruption despite having a larger share of German exports (see the last column of Table 1). The PACI is based not only on one or two “points of observation”, but it aggregates cases involving firms headquartered in any country. This reasoning does not need any assumption with respect to the probability that cases in a given country would be enforced. Notably, that there were 315 cases of cor- ruption involving firms headquartered in the United States and only 56 in Germany (with enforcement occurring first in the respective countries) has no implications with respect to levels of corruption in the United States or in Germany. It is the distribution of these cases (regardless of how many there are) abroad which conveys information that could help determine the level of corruption in the countries where the bribery of foreign officials takes place. The PACI The incidents of corruption involving firms in i that implicate public officials in j and first enforced in the headquarters’ country i is indicated by cases obs HQi;j . The PACIz compares the total number of those cases with the expected number of corrupt transactions that would be observed if their spatial dis- tribution reflected bilateral trade shares between the headquarters countries and z: P N cases obs HQi;z PACIz ¼ i¼1 Á 100; with i 6¼ z: (1) P N Eðcases obs HQi;z Þ i¼1 P The numerator, cases obs HQi;z , is the total number of observed corrupt exchanges between officials from country z and firms from all other headquarters countries, first enforced in those countries. The denominator is the total number of similar cases which could be observed if cases of corruption were dis- tributed according to the ratio of exports of country i to z (Xiz ) with respect to the total amount of coun- try i exports to the rest of the world: X N X N Xiz X N Eðcases obs HQi;z Þ ¼ Á cases obs HQi;j : (2) i¼1 i ¼1 P N Xij j¼1 j¼1 The denominator may be interpreted as the total number of cross-border corruption cases involving country z public officials, first enforced elsewhere, that could be observed if the level of corruption of public officials were the same in all countries. If the actual and expected values are equal, then the PACI equals 100. The lowest value that the index may take is zero, which obtains when no corruption case (first enforced in all headquarters countries) is registered in country z. The composite PACI ALL The PACIz or “composite PACI” follows the same logic of the simple PACIz . However, it accounts for cases that were first enforced not only in all headquarters countries but also in other countries 200 Escresa and Picci except z. Cases that are first enforced in a third country w, meanwhile, are denominated as cases obs OTHiw ;j (where i,j 6¼ w). The index is then shown as: P N D P P N cases obs HQi;z þ cases obs OTHiw ;z (3) ALL i¼1 w¼1 i¼1 PACIz ¼ Á 100 P N PD P N Eðcases obs HQi;z Þ þ Eðcases obs OTHiw ;z Þ i¼1 w¼1 i¼1 with i6¼j, w6¼i,j, D is the number of third countries that first enforced cases, and for the denominator the following holds: D X X N XD X N Xiz X N w Eðcases obs OTHz ;i Þ ¼ cases obs OTHiw ;j w¼1 i¼1 w¼ 1 i ¼ 1 P N j ¼ 1 Xij j¼1 ALL The interpretation of the PACIz is conceptually the same as that of the PACIz (Eq. 1), but it considers all available cases of observed cross-border corruption, first enforced either in the country where firms are headquartered or within the jurisdictions of third countries. This is the version of the index computed in this article. Conditional Probability of Observing Zero Cases It is also important to demonstrate the varying precision with which the PACI measures corruption, which is influenced by the total number of transactions between countries as represented by bilateral trade flows. To fix ideas, consider the case of a foreign country that has no reported incident of cross- border corruption. Such a situation could be interpreted as a signal that corruption in the said foreign country is relatively low, but not necessarily equal to zero. The strength of such a signal depends on the number of cross-border transactions, proxied by imports. To express this concept, the “probability of observing zero cases, conditional on the probability of corruption being equal in all foreign countries” is introduced: XN Przero casesPACI z ¼ Prð cases obs HQi;z ¼ 0 j Prcorr FOz ¼ cÞ i¼1 where Prcorr FOz is the underlying probability that a public official accepts a bribe (see Appendix A) and c is a constant. The measure expresses the probability that no case involving country z public offi- cials (first enforced in the headquarters country) is observed when the true country z probability that a cross-border transaction involving its public officials is corrupt is the same everywhere in the world. Przero casesPACI z may be computed as the product of individual probabilities: YN Przero casesPACI z ¼ i ¼1 ðPrcases obs HQi;z ¼ 0 j Prcorr FOz ¼ cÞ Each one of these probabilities, referring to a rare event, is described by a Poisson distribution and may be easily computed by making the condition Prcorr FOz ¼ c correspond to an ideal situation where occurrences of corruption are distributed according to trade shares. High values of Przero casesPACI z indi- cate that the country in question has relatively few cross-border transactions – that is, information is rel- atively scarce. II. Computing the PACI Information on cases of cross-border corruption from 1977 until the end of 2012 were collected using various sources (See Appendix B). However, only cases reported in 1998 onwards were used, The World Bank Economic Review 201 approximately coinciding with the effectivity of the OECD Anti-Bribery Convention. Out of a total of 979 cases detailed in this study, 796 cases were first enforced either in the headquarters country (569), in the United States acting as a third country jurisdiction (177), or in other third country jurisdictions (50). The remaining 183 cases that were first enforced in foreign countries were not considered for the pur- pose of computing the PACI. Each case was coded according to the observed outcome. They were classified “positive” if the accused party was either found guilty or, while not admitting guilt, conceded to the payment of a fine (as in the case of a “consent to a cease-and-desist order” in the US);8 “not positive” if the case was eventu- ally dropped and no action was taken; or “ongoing” if no available evidence was found to determine whether the case is “positive” or “not positive.” The term public official is used in a broad sense, encom- passing both bureaucrats and politicians. Table 2 indicates that out of 796 cases that may be used to compute the PACI, 444 are classified as positive, 272 as ongoing, while the rest have either been dropped or have resulted in an acquittal. Firms are shown to be headquartered in 40 countries, comprised mostly of industrialized nations. Among them, the United States takes the lion’s share in the number of cases enforced, reflecting both its early adoption of the FCPA and the proactive stance adopted by its Department of Justice (DOJ) and SEC after its ratification of the OECD Anti-Bribery Convention. Germany, Britain, and France follow in the list. The list of countries where public officials are seen to be on the receiving end of alleged bribery is more extensive (Table 3). At least one case is recorded in 128 countries. China leads the list with 88 cases, followed by Nigeria. ALL To compute the index of public administration corruption, PACIz , all 796 cases, regardless of their outcomes, were considered (Table 4). The inclusion of all the cases is justified by the presence of a high burden of proof in order to lead to a conviction. Thus, false negatives are likely to be more numerous than false positives. However, in order to accommodate a more agnostic view on this issue, the study computes the index by excluding cases that have been dropped or have resulted in an acquittal. It also computes the PACI by excluding cases involving public agencies in charge of health and telecommunica- tions sectors, for reasons discussed in Section IV of this paper. The combination of these alternatives results in the computation of four versions of the PACI, whose reciprocal Spearman rank correlations range between 0.885 and 0.945 (Table 5). These computations will be discussed in their entirety in the succeeding section. The probability of observing zero cases, under the condition that probability of corruption is equal in all foreign countries, Przero casesPACI z , is used to rank those countries that have no observed cases of cor- ruption first enforced abroad, and also to exclude very small countries for which the available informa- tion is not deemed to be sufficient. Canada and Finland, for instance, reported zero cases of corruption (first enforced abroad) involving their public officials. For Finland, such probability equals 1.3 percent— being a middle-sized economy, it is rather unlikely for Finland not to observe any cases, if the probability that public officials accept bribes were the same in Finland as in the rest of the world. For Canada, on the other hand, Przero casesPACI z ¼0.0000. In this case, the signal provided by the PACI, indicating that the level of corruption in Canada is low, is very strong and illustrates that Canada, which is a bigger country, generates more information than Finland. Przero casesPACI z provides a useful indication of the precision of the PACI also for countries with at least one case of corruption on record. For smaller countries, the signal that the PACI provides is rather noisy because of the rarity of observed corruption events. For example, the probability of observing zero cases in Tunisia would be equal to 0.31 in a situation where the level of corruption is the same 8 It should be emphasized that coding the outcome of a case as positive does not imply conviction or guilt on the part of the accused. 202 Escresa and Picci Table 2. Total Number of Cases by Headquarters’ Country, 1998–2012 Country of firm’s headquarter Total Cases Positive Cases Ongoing Cases United States 331 227 82 Germany 86 50 36 United Kingdom 54 25 24 France 47 27 18 Switzerland 46 40 6 Italy 27 5 16 Spain 22 0 8 Australia 19 5 14 Canada 19 4 8 Japan 17 11 5 Netherlands 15 8 5 Sweden 15 1 12 Korea 14 14 0 Portugal 13 0 7 Norway 9 5 3 China 8 7 1 Argentina 6 1 2 Austria 6 2 4 Brazil 6 0 5 Finland 5 4 1 Bermuda 4 0 0 Chile 3 0 2 Denmark 3 0 3 Israel 3 2 1 Belgium 2 0 2 Hungary 2 1 1 Angola 1 0 0 Bangladesh 1 1 0 Czech Republic 1 0 0 Ghana 1 1 0 Ireland 1 0 1 India 1 1 0 Luxembourg 1 1 0 New Zealand 1 0 1 Poland 1 0 1 Russia 1 0 0 Slovak Republic 1 0 1 Turkey 1 0 1 British Virgin Islands 1 1 0 South Africa 1 0 1 Total 796 444 272 Notes: Cases are those first enforced in the headquarters country or in any third-country jurisdiction. The “headquarters country” is where the firm which allegedly bribed public officials abroad is headquartered. Positive Cases refer to cases that were concluded with a judgment in favor of the prosecution or a settlement. Ongoing cases are those that are still pending. everywhere in the world. Notably, two cases were actually observed, close to twice as many as the expected figure if cases were distributed according to trade shares, with a resulting PACI of about 170 and a ranking measure for Tunisia similar to that indicated by the WB-CCI and the TI-CPI (see the last two columns of Table 4). The World Bank Economic Review 203 Table 3. Total Number of Cases by Country where Alleged Corruption Takes Place, 1998–2012 Foreign country Total Cases Positive Cases Ongoing Cases China 88 49 34 Nigeria 42 32 6 India 29 12 14 Russia 28 15 10 Indonesia 24 18 5 Libya 24 6 16 Brazil 22 7 11 Kazakhstan 22 8 9 Angola 19 9 8 Egypt 17 14 3 Argentina 16 9 7 Philippines 15 9 5 Greece 14 9 4 Mexico 14 10 4 Thailand 14 11 3 Saudi Arabia 12 7 4 United States 12 12 0 Vietnam 12 11 1 Venezuela 11 7 3 Algeria 10 3 6 Poland 10 6 2 Turkey 10 6 1 United Arab Emirates 9 5 4 Iraq 9 3 6 Ghana 8 2 3 Iran 8 4 2 Korea 8 3 5 Malaysia 8 5 1 Romania 8 3 3 Bangladesh 7 5 2 Bulgaria 7 4 1 Czech Republic 7 2 4 Hungary 7 3 3 Liberia 7 4 1 Serbia 7 4 2 Congo, Republic of 6 4 2 Costa Rica 6 3 3 Croatia 6 3 2 Italy 6 2 3 Kenya 6 4 2 Panama 6 2 3 Uganda 6 3 2 Uzbekistan 6 4 2 South Africa 6 2 3 Austria 5 2 2 Peru 5 1 3 Pakistan 5 4 1 Syria 5 3 2 Bahrain 4 4 0 France 4 4 0 Equatorial Guinea 4 2 1 Haiti 4 2 1 204 Escresa and Picci Table 3. (continued) Foreign country Total Cases Positive Cases Ongoing Cases Cambodia 4 2 2 Mali 4 3 0 Qatar 4 2 2 Slovenia 4 2 2 Tanzania 4 4 0 Azerbaijan 3 3 0 Coˆ te d’Ivoire 3 2 1 Ecuador 3 3 0 Gabon 3 2 0 Georgia 3 1 1 Hong Kong 3 3 0 Honduras 3 3 0 Kuwait 3 1 2 Morocco 3 1 1 Mauritania 3 2 1 Malawi 3 1 1 Oman 3 1 2 Rwanda 3 2 1 Taiwan 3 2 1 Ukraine 3 1 1 Zimbabwe 3 0 2 Albania 2 1 1 Colombia 2 1 1 Germany 2 1 0 Spain 2 1 1 Guinea 2 1 1 Lithuania 2 1 1 Latvia 2 1 1 Mozambique 2 1 0 Nepal 2 2 0 Sudan 2 2 0 Singapore 2 1 1 Slovak Republic 2 0 2 Senegal 2 1 1 Somalia 2 0 0 Turkmenistan 2 2 0 Tunisia 2 0 2 Yemen 2 1 1 Zambia 2 0 1 Afghanistan 1 0 1 Bosnia 1 1 0 Belgium 1 1 0 Burkina Faso 1 1 0 Benin 1 1 0 Brunei 1 1 0 Bolivia 1 1 0 Bahamas, The 1 1 0 Belarus 1 0 0 Cameroon 1 0 1 Cuba 1 1 0 Djibouti 1 1 0 Dominican Republic 1 0 1 The World Bank Economic Review 205 Table 3. (continued) Foreign country Total Cases Positive Cases Ongoing Cases Eritrea 1 0 1 Jamaica 1 0 1 Jordan 1 1 0 Japan 1 0 1 North Korea 1 0 1 Luxembourg 1 1 0 Madagascar 1 1 0 Macedonia 1 1 0 Myanmar 1 1 0 Mongolia 1 1 0 Moldova 1 0 1 Niger 1 1 0 Netherlands 1 1 0 Norway 1 0 1 Nairu 1 0 1 French Polynesia 1 1 0 Portugal 1 0 1 ao Tome S~ ıncipe ´ and Pr 1 0 0 El Salvador 1 0 1 Turks and Caicos Islands 1 0 1 Chad 1 1 0 Trinidad and Tobago 1 0 0 United Kingdom 1 1 0 Uruguay 1 0 0 Total 796 444 272 Notes: Cases are those first enforced in the headquarters country or in any third-country jurisdiction. The “foreign country” is the country where the act of (alleged) corruption took place. III. The PACI and Perception-Based Measures: A Comparison The differences in the rankings yielded by the PACI and the WB-CCI and TI-CPI are also explored (see last columns of Table 4), where the perception-based measures used are for the year 2005, the middle of the time interval of reference of the PACI. A positive value indicates that a given country is considered to be relatively more corrupt based on the PACI. In most cases, differences are modest. However, results diverge considerably for a few countries. These information are summarized by showing the Spearman rank correlations between the different versions of the PACI and the perception-based alternatives (Table 5). The rank correlations of PACI1, the version which includes all cases (presented in Table 4), with perception-based measures are rather high, typically pegged at above 0.7.9 While the ranking based on the PACI is similar to the rankings based on the two most popular perception-based indices, the scales appear to be very different (Figure 1). Panel A shows a scatter dia- gram of the PACI together with the WB-CCI for 2005, while Panel B shows the same indices, with the PACI log-transformed. The Pearson correlation of log(PACI) and WB-CCI is quite high (equal to À0.841). The study also seeks to examine whether the observed differences between the PACI and perception- based measures of corruption are systematic, and to the extent that they might be, what determines 9 Note that more corruption corresponds to lower values of the WB-CCI and the TI-CPI indices, but to higher values for the PACI and the TI-GCB. 206 Escresa and Picci Table 4. Public Administration Corruption Index (PACI), 1998–2012 ALL Country PACIz Przero casesPACI z Country Rank Rank difference, WB CCI Rank difference, TI CPI Canada 0 .0000 1 À10 À11 Switzerland 0 .0000 2 À2 À3 Australia 0 .0000 3 À4 À4 Sweden 0 .0001 4 À1 0 Ireland 0 .0002 5 À9 À12 Denmark 0 .0018 6 4 4 Finland 0 .0128 7 6 6 UK 2.58773 .0000 8 À2 À2 Japan 3.017315 .0000 9 À10 À9 Netherlands 3.520985 .0000 10 2 1 Germany 3.723345 .0000 11 À1 À2 Belgium 3.786931 .0000 12 À5 À4 Spain 8.576681 .0000 13 À5 À6 France 10.58484 .0000 14 À2 À1 Singapore 14.45795 .0000 15 12 12 Portugal 19.10472 .0053 16 À4 À4 Norway 19.3984 .0058 17 11 11 USA 21.54232 .0000 18 3 4 Italy 22.29935 .0000 19 À22 À11 Taiwan 25.40663 .0000 20 À4 À4 Mexico 29.16911 .0000 21 À35 À31 Korea 46.02413 .0000 22 À5 À7 Austria 47.04836 .0000 23 14 15 Dominican Republic 50.16856 .1362 24 À52 À42 Colombia 65.02713 .0462 25 À27 À17 Luxembourg 80.35295 .2881 26 13 15 Slovakia 90.9612 .1109 27 À2 À11 Bahamas 99.06574 .3644 28 À97 À94 El Salvador 99.22466 .3650 29 À38 À10 Malaysia 106.3388 .0005 30 À8 2 South Africa 107.8829 .0038 31 3 À4 Jamaica 119.4216 .4328 32 À33 À16 Turkey 119.5222 .0002 33 À12 À20 Czech Republic 125.0943 .0037 34 4 À2 Poland 126.9792 .0004 35 À5 À21 China 144.8671 .0000 36 À43 À26 Trinidad and Tobago 148.6733 .5104 37 À10 À7 Morocco 149.0878 .1337 38 À19 À22 Ukraine 154.7996 .1440 39 À43 À51 Hungary 168.4742 .0156 40 14 9 Tunisia 173.0795 .3148 41 À7 9 Saudi Arabia 190.4156 .0018 42 À8 À17 Belarus 192.3195 .5945 43 À54 À46 Jordan 198.2679 .6038 44 9 17 Kuwait 202.4064 .2271 45 20 11 Uruguay 204.3132 .6129 46 25 21 Brazil 213.2271 .0000 47 À6 0 Lithuania 217.4639 .3986 48 9 15 Cuba 236.8499 .6555 49 13 4 India 241.05 .0000 50 À14 À22 Thailand 255.6769 .0041 51 0 5 Qatar 264.63 .2205 52 29 29 The World Bank Economic Review 207 Table 4. (continued) ALL Country PACIz Przero casesPACI z Country Rank Rank difference, WB CCI Rank difference, TI CPI Slovenia 268.9919 .2260 53 31 31 Romania 276.0276 .0551 54 À1 À11 Venezuela 281.1082 .0199 55 À45 À47 Russia 310.2815 .0001 56 À35 À42 Ecuador 315.465 .3863 57 À29 À40 Peru 322.8425 .2125 58 0 9 Honduras 339.1537 .4128 59 À28 À28 Brunei 351.3141 .7522 60 À63 À64 Latvia 367.9257 .5806 61 27 20 Costa Rica 378.6169 .2050 62 30 22 Pakistan 380.5453 .2687 63 À44 À50 Greece 388.5903 .0272 64 31 27 Iran 396.8289 .1331 65 À5 À10 Oman 460.9047 .5215 66 29 45 Afghanistan 472.1338 .8091 67 À52 À27 Bosnia 472.9085 .8094 68 14 À6 Bolivia 475.1096 .8101 69 À20 À24 Benin 479.1134 .8116 70 À29 À1 Philippines 496.6861 .0488 71 À6 À21 Algeria 533.9379 .1536 72 4 À4 Croatia 539.2692 .3286 73 31 18 Yemen 545.6701 .6931 74 À18 À9 Cameroon 576.2632 .8407 75 À33 À33 Argentina 586.4659 .0653 76 7 À1 Macedonia 672.15 .8617 77 6 À5 Bulgaria 696.8326 .3662 78 34 35 Vietnam 714.1484 .1863 79 À11 À9 Sudan 775.2856 .7726 80 À38 À32 Indonesia 846.4598 .0586 81 À15 À26 Panama 884.8087 .5075 82 20 32 Bahrain 900.8724 .6414 83 52 57 Senegal 924.2173 .8054 84 38 23 Syria 955.4548 .5925 85 2 27 Coˆ te d’Ivoire 1061.689 .7538 86 À26 À30 Albania 1144.235 .8396 87 À1 À13 Burkina Faso 1171.676 .9181 88 39 34 Gabon 1176.169 .7748 89 15 21 Madagascar 1219.936 .9212 90 47 10 Egypt 1253.273 .2575 91 19 34 Iraq 1268.119 .4917 92 À28 À17 Azerbaijan 1280.879 .7911 93 À9 À17 Serbia 1356.355 .5968 94 31 30 Kenya 1481.514 .6669 95 À6 À16 Bangladesh 1549.458 .6365 96 À18 À24 Haiti 1651.818 .7849 97 À20 À21 Niger 1812.204 .9463 98 14 À1 Georgia 1819.622 .8480 99 38 À4 Mozambique 1900.602 .9001 100 27 21 Burma 1929.848 .9495 101 À20 À18 Turkmenistan 1947.478 .9024 102 À14 À15 Zambia 1961.419 .9030 103 10 17 Nigeria 2101.201 .1354 104 À5 À10 208 Escresa and Picci Table 4. (continued) ALL Country PACIz Przero casesPACI z Country Rank Rank difference, WB CCI Rank difference, TI CPI Ghana 2191.095 .6941 105 46 54 Chad 2207.392 .9557 106 À9 À15 Mongolia 2302.75 .9575 107 32 40 Tanzania 2307.337 .8408 108 28 35 Guinea 2410.733 .9203 109 5 À6 Djibouti 2425.423 .9596 110 29 41 Kazakhstan 2475.69 .4112 111 8 27 Congo 3088.969 .8234 113 8 12 Eq. Guinea 3102.567 .8790 114 À8 À9 Mauritania 3242.89 .91164 115 55 52 Liberia 3426.804 .8152 116 10 11 Libya 3550.355 .5086 117 19 26 Zimbabwe 4122.554 .9298 118 5 33 Uzbekistan 4290.974 .8695 119 9 13 Cambodia 4645.204 .9174 120 9 16 Nepal 5354.408 .9633 121 43 25 Sao Tome & P. 5905.44 .9832 122 27 41 Uganda 5911.488 .9034 123 29 28 Mali 6340.174 .9388 124 58 54 Malawi 8598.199 .9657 125 40 47 Notes: Index computed using all cases regardless of outcome and administration. Countries for which PACIz ¼ 0 have been ranked according to the negative of Przero casesPACI z . Countries for which PACIz ¼ 0 and Przero casesPACI z > 0.015 and countries for which PACIz > 10000 have been excluded from the list. WB-CCI: World Bank Corruption Control Index, 2005. TI-CPI: Transparency International Corruption Perceptions Index, 2005. them. The residual of a linear regression between the log of the PACI and the WB-CCI (as in Figure 1.B) or the TI-CPI is considered and denoted as “measurement residuals” – mesresWB-CCI and mesresTI-CPI, respectively. Negative (positive) values indicate that a given country appears to be less (more) corrupt according to the PACI as opposed to the perception-based index used. Whether these differences are correlated with Przero casesPACI z is determined first. If the precision of the PACI deteriorates significantly with respect to smaller countries, while the precision of the perception-based measures remains intact (or deteriorates to a lesser degree), then the absolute value of the measurement residuals could be expected to be positively correlated with Przero casesPACI z . This, however, is not the case. The correlation between the absolute values of mesresWB-CCI and mesresTI-CPI and Przero casesPACI z is seen to be insignificant, pegged at À0.0524 and À0.1001, respectively. This result indicates that any problem which may beset the precision of the PACI in measuring corruption in small countries, would also be shared by the two perception-based alternatives. Subsequently, the study selected a set of variables which may be linked to biases in favor of or against either measure.10 Variables of an economic and demographic nature, namely: per capita GDP based on purchasing-power-parity (gdp_cap), population (pop), and the ratio of public expenditure over GDP (r_g/gdp) are considered. Variables that express the ease with which publicly relevant information is gen- erated and debated upon, as well as the democratic attributes of a country, are also contemplated. Also included in such examination are emp_rights, an index of empowerment rights; free_press and free_- speech, which respectively measure the operationalization of the guarantees to free press and free speech; voice_acc, a measure of voice and accountability; democ, an index of democratization; checks, which measures systems of checks and balances; and stability, which measures political stability. To test 10 A detailed description of the variables used is found in Appendix B. The World Bank Economic Review 209 Table 5. Spearman Rank Correlations between Different Indexes of Corruption PACI1 PACI2 PACI3 PACI4 TI-CPI WB-CCI TI-GCB PACI1 1 PACI2 0.938 1 (123) PACI3 0.940 0.886 1 (123) (123) PACI4 0.885 0.945 0.941 1 (123) (123) (123) TI-CPI À0.779 À0.736 À0.751 À0.729 1 (123) (123) (123) (123) WB-CC À0.768 À0.711 À0.729 À0.692 0.954 1 (123) (123) (123) (123) (123) TI-GCB 0.755 0.731 0.769 0.747 À0.800 À0.767 1 (56) (56) (56) (56) (56) (56) Notes: PACI: Public Administration Corruption Index, 1998–2012. The subscript indicates: 1: All cases, all administrations (the same values shown in Table 4; our preferred index). 2: All cases, with the exclusion of health and telecom administration. 3: Only “positive” and “ongoing” cases, all administrations. 4: Only “positive” and “ongoing” cases, with the exclusion of health and telecom administrations. TI-CPI: Transparency International Corruption Perceptions Index, 2005. WB-CCI: World Bank Corruption Control Index, 2005. TI-GCB: Percentage of persons who answered “yes” to the question: “In the past 12 months, have you or anyone living in your household paid a bribe in any form?” Transparency International Global Corruption Barometer, 2005. Number of observations are between parentheses. All the estimated coefficients are significant at less than 1 percent. Countries for which PACIz ¼ 0 and Przero casesPACI z > 0.015 and countries for which PACIz > 10000 have been excluded. whether the observed residuals follow some recognizable geographic pattern, a set of geographic dum- mies is also considered. Pairwise correlations between mesresWB-CCI, mesresTI-CPI, and these variables are thus shown (Table 6). Results excluding the ten biggest (in absolute value) measurement residuals, roughly corresponding to ten percent of the available observations, are also presented. Two of the geographic dummies are significantly correlated with both types of measurement resid- uals. Countries in the American continents, on average, appear to be less corrupt according to the PACI, compared to when either of the perception-based measures are used. The opposite, however, can be seen for countries in Africa and the Middle East. Populous countries also appear to fare better according to the PACI. Countries that are more democratic and have stronger checks and balances appear to be less corrupt on average when the PACI is used. The significance of some of the other variables, meanwhile, depends on which measurement residual is considered and on whether outliers are included. In interpret- ing the results, the squared estimated correlation coefficient should be noted to represent the fraction of the variance of the measurement residuals that is explained by a given variable (the R2 of the bivariate ordinary least-squares regression). Such fraction is always rather small, even when the estimated correla- tion coefficient is statistically significant. To go beyond simple bivariate correlations, a multivariate regression is also considered (Table 7). Here, the dependent variable is either mesresWB-CCI or mesresTI-CPI and the regressors consist of all the explanatory variables described earlier.11 As in the bivariate analysis, results obtained by excluding the ten observations of the dependent variable having the greatest absolute value were also reported. Only few of the explanatory variables considered are statistically significant and all the regressors only explain 30 to 35 percent of the total variability of the measurement residuals. A significant negative effect for the dummy variable pertaining to the American continents, in three out of four cases, can still be found. The effect of the dummy variable on Africa and the Middle East, detected in the bivariate 11 The dummy for Europe and Central Asia was excluded to avoid the dummy variable trap. The estimated coefficients of the other dummies should then be interpreted as the estimated effect relative to that reference group of countries. 210 Escresa and Picci Figure 1. Comparison between PACI and WB Corruption Control Index Notes: In Panel B, when the PACI equals zero (countries: FI, DK, IE, SE, AU, CH, CA), it has been set equal to arbitrary small numbers PACI for the purpose of computing the log. Countries for which PACIz ¼ 0 and Przero casesz > 0.015 and countries for which PACIz > 10000 have been excluded. analysis, is now insignificant. Countries where the index of democratization (democ) is higher appear to be less corrupt according to the PACI. The same applies to countries having a high share of public expen- diture over GDP (r_g/gdp). Population is also seen to have a significant effect in two out of four cases. Overall, differences between the PACI and the two leading perception-based cross-national measures of corruption appear to be rather idiosyncratic, at least with respect to the set of factors considered. The World Bank Economic Review 211 Table 6. Correlation Coefficients between Measurement Residuals and Selected Variables All observations Dummy EU Asia Dummy Asia Pacific Dummy America Dummy Africa ME Gdp_cap Pop R_g/gdp mesresWB-CCI À0.0604 À0.1405 À0.2507*** 0.3621*** À0.0090 À0.2122** À0.1182 (0.5071) (0.1211) (0.0052) (0.0000) (0.9242) (0.0185) (0.1946) mesresTI-CPI À0.1297 À0.0810 À0.2854*** 0.4109*** 0.0510 À0.2234** À0.1010 (0.1527) (0.3732) (0.0014) (0.0000) (0.5916) (0.0130) (0.2683) Free_press Free_speech Voice_acc Emp_right Democ Checks Stability mesresWB-CCI 0.0368 À0.0739 À0.0604 À0.0677 À0.2687*** À0.2180** 0.1518* (0.6876) (0.4163) (0.5067) (0.4572) (0.0033) (0.0177) (0.0938) mesresTI-CPI 0.1579* À0.1221 À0.1620 À0.1415 À0.3310*** À0.2807*** 0.1113 (0.0823) (0.1786) (0.0735) * (0.1184) (0.0003) (0.0021) (0.2205) Excluding the ten biggest outliers Dummy EU Asia Dummy Asia Pacific Dummy America Dummy Africa ME Gdp_cap Pop R_g/gdp mesresWB-CCI À0.0400 À0.1297 À0.1928** 0.2860*** 0.0478 À0.2234** À0.0933 (0.6737) (0.1708) (0.0408) (0.0021) (0.6315) (0.0174) (0.3278) mesresTI-CPI À0.1200 À0.0758 À0.2216** 0.3476*** 0.1124 À0.2319** À0.0772 (0.2056) (0.4247) (0.0184) (0.0002) (0.2585) (0.0134) (0.4186) Free_press Free_speech Voice_acc Emp_right Democ Checks Stability mesresWB-CCI À0.0221 À0.0673 0.0019 À0.0391 À0.1882* À0.2234** 0.2073** (0.8171) (0.4787) (0.9839) (0.6904) (0.0511) (0.0195) (0.0276) mesresTI-CPI 0.1159 À0.1104 À0.1161 À0.1153 À0.2682*** À0.2818*** 0.1619* (0.2238) (0.2443) (0.2208) (0.2241) (0.050) (0.0030) (0.0866) Notes: Number of observations. Panel A: between 124 and 117; Panel B: between 104 and 113. P-values between parentheses. *p-value < .1; **p-value < .05; ***p-value < .001. For a description of the data, see Appendix B. Countries for which PACIz ¼ 0 and Przero casesPACI z > 0.015 and countries for which PACIz > 10000 have been excluded. IV. The Validity of the PACI This section discusses the assumptions necessary for the PACI to be considered a valid measure of cor- ruption. While Appendix A presents these assumptions formally, to show how they imply index validity, the focus here is on their overall meaning and, most importantly, on the extent to which they may hold in practice. The first assumption is that the probability of observing a corrupt transaction involving firms from country i (and enforced first in the same country, or in third country jurisdictions) and public officials in country j does not depend on the identity of country j. It implies that the judiciary, when deciding which cases to pursue, does not “discriminate” based on the foreign officials’ nationality. The possibility of assessing the levels of corruption in the foreign country by looking at the geographical distribution of cases first enforced elsewhere hinges on this key assumption. Whether this assumption can be relied upon is an empirical question. McLean (2012) evaluated the relevance of foreign policy considerations and opportunities for enforcement cooperation in determining the geographic distribution of FCPA cases, and found that bilateral frameworks for securities regulation and enforcement cooperation appear to be associated with higher levels of FCPA implementation. However, the magnitude of such an effect was found to be rather modest. The same study did not find other candidate explanatory variables to have any significant effect. Choi and Davis (2012) also found little evidence contradicting the first assumption. The assumption could also be violated if the probability of detecting corruption cases depended on the conditions surrounding freedom of expression and information in the foreign country. However, the available data indicate that most cases are first enforced based on evidence gathered in the headquarters 212 Escresa and Picci Table 7. Multivariate Analysis of the Determinants of the Measurement Residuals Dep. var: mesresWB-CCI Dep. var. mesresTI-CPI Regressors All observations Excluding10 outliers All observations Excluding10 outliers Asia & Pacific À0.4102 À0.2714 À0.0992 0.1070 (0.3213) (0.2886) (0.3243) (0.2972) America & Caribbean À0.9691** À0.6412* À0.8116** À0.4874 (0.4018) (0.3569) (0.3778) (0.3353) Africa & Middle East 0.3450 0.4022 0.4826 0.4655 (0.3502) (0.3281) (0.3624) (0.3453) Per capita GDP. PPP 11.7404 8.8195 21.7674* 11.8142 (10.0708) (9.5360) (11.189) (8.4577) Pop À0.0063 0.0052 À0.0008* À0.0010** (0.0006) (0.0006) (0.0004) (0.0004) R_g/gdp À0.0102*** À0.0097*** À0.0075* À0.0010** (0.0035) (0.0027) (0.0041) (0.0004) Free_press À0.0036 À0.0008 0.0127 0.0068 (0.0120) (0.0116) (0.0113) (0.0109) Free_speech À0.1566 À0.0559 À0.0909 À0.0357 (0.2883) (0.280) (0.266) (0.24089) Voice_acc 0.4953 0.467 0.1817 0.2087 (0.3785) (0.361) (0.3578) (0.3176) Emp_right 0.0734 0.0114 0.1048 0.0539 (0.0777) (0.0663) (0.0790) (0.0682) Democ À0.0440** À0.0234 À0.03245* À0.0247 (0.0194) (0.0176) (0.0178) (0.0164) Checks À0.0436 À0.0642 À0.0332 À0.0228 (0.0457) (0.0420) (0.0404) (0.0346) Stability 0.1302 0.1582 0.1605 0.1560 (0.1705) (0.1491) (0.1592) (0.1467) Observations 105 95 105 96 R-squared 0.328 0.295 0.354 0.337 Notes: OLS estimates. Robust standard errors are between parentheses. **p-value < 0.1; **p-value < 0.05; ***p-value < 0.001. For a description of the variables, see the note at the bottom of Table 6 and Appendix B. Countries for which PACIz ¼ 0 and Przero casesPACI z > 0.015 and countries for which PACIz > 10000 have been excluded. countries. Furthermore, supposing that the relevant perception-based measure is unbiased, the variables that capture a liberal environment for expression and the circulation of information in the foreign coun- try would be expected to positively affect the measurement residuals introduced in the earlier sections (implying that the PACI in those cases would reveal a higher level of corruption compared to its perception-based counterpart). Three variables expressing the various dimensions of the ease of expres- sion and circulation of information (free_press, free_speech and voice_acc; see Appendix B for a descrip- tion) were considered and the results of both bivariate and multivariate analysis (Table 6 and 7) indicate a statistical significance only in very few cases, while pointing to the presence of marginal effects at most.12 12 The probability of detection and enforcement may also be sector-specific. For example, corruption in the arms trade is likely to be more difficult to detect as national security concerns may limit the actions available to the judiciary (see also Rose-Ackerman 1999). To address such issues, the PACI could be computed separately for different sectors of the economy. The World Bank Economic Review 213 Assumptions 2 and 3 in Appendix A are rather technical, and describe how the probability of offer- ing a bribe, or accepting a bribe when offered, may depend on the level of corruption in the other country. Of more interest is assumption 4, which establishes that the number of cross-border transac- tions is proportional to bilateral trade flows. An alternative proxy for cross-border transactions would be foreign direct investments (FDI) (as in Choi and Davis 2012 and Mclean 2012). However, many transactions are not reflected in FDI flows, nor stocks, and FDI eventually enable trade flows between the countries involved. For these reasons, the choice of proxy made in this study seems to be more appropriate. Differences in the scope of the public sector in different countries would also be a cause for concern. For instance, a pharmaceutical firm headquartered in country A successfully bribes hospital employees in country B and C in order to sell its products. If country B runs a public health system, the bribery inci- dent would qualify as a case of cross-border corruption of a foreign public official. On the other hand, if country C’s hospitals are managed by the private sector, the bribery incident would qualify as a case of private sector corruption and, as such, would not be included in this study’s dataset. At first glance, neglecting this difference may be seen to lead to an underestimation of the level of corruption in country B with respect to C since, ceteris paribus, more corruption cases are registered in the former than in the latter, due to the wider scope of the public sector in country B. Apparently, in this case, trade flows would not serve as a good proxy for cross-national exchanges involving public officials in a given coun- try, since for each dollar of imports there would be more interactions with public officials in country B than in country C. To account for this issue, this study computed the PACI without regard for cases involving procurement in the health and the telecommunication sectors (as opposed to transactions involving regulatory bodies in those sectors since they are invariably public). Arguably, these are the main sectors that are prone to stark variations in terms of government activism.13 Different choices, however, seem to deliver similar results (see Table 5).14 V. Discussion The PACI reflects a narrow definition of corruption: the propensity of public officials to accept bribes from foreign firms.15 In denominating the index as a general “public administration” measure of corrup- tion, the study implicitly assumes that what is observed by means of cross-country corruption statistics is also informative of the level of corruption in the public sector as a whole. Obviously, this assumption may be put to question, especially since levels of corruption may vary sensibly across public institutions within a given country. This article, however, argues that adopting narrow definitions of a phenomenon is desirable because it facilitates the testability of the assumptions on which hinges any extensive inter- pretation of the resulting measures. For this reason, the PACI represents a welcome departure from most indices of governance currently available, where “sometimes it is not clear what precisely is being measured, rendering questionable the validity of at least some of the proxies” (Klitgaard and Light 1998). Such a state of affairs may reflect a 13 Alternatively, the number of cases for the relative size of the public sector in different countries could be corrected. However, the concept of public sector is blurry, particularly in some countries. Generally, the decision on how to ac- count for the variable scope of public sectors worldwide depends on the position which the researcher takes on concep- tual issues that are the subjects of debate. 14 If the interest lies in measuring the magnitude of public sector corruption, instead of its frequency, then cross-country variations in the scope of activities of the public sector should not be a cause for concern. The scope of government is also arguably endogenous over the long-run since corrupt elites have an interest in maintaining and possibly expanding their reach. 15 A Bribe Payer’s Corruption Index (BPCI) may also be computed along the same lines as the PACI, but using cases first enforced in the foreign country. However, the scarcity of data proves to be an obstacle towards its computation. 214 Escresa and Picci situation “experienced in many ‘new areas’ of the social sciences: an explosion of measures, with little progress toward theoretical clarity or practical utility” (Klitgaard et al. 2005, 414). Quite likely, as “open data” become more widely available, more measures of governance, based on hard data, and nar- rowly defined like the PACI, would be ushered (see the discussion in Picci 2011, 117–119). Such a devel- opment, in turn, will facilitate greater theoretical clarity. Data availability, as seen, poses certain constraints to the usefulness of the PACI. This limitation is the reason why this paper opted to demonstrate its application using a 15-year period. However, when comparing this attribute of the PACI with the TI-CPI, it should be noted that Transparency International has warned against making comparisons with their index across time, in light of annual changes in the methodology and country coverage. In other words, to some extent, the advantage of hav- ing a yearly measure is more apparent than real.16 As more countries sign the OECD Anti-Bribery Convention, more cases are likely to be reported, lending more precision to the PACI and allowing its computation for shorter intervals of time. Also, more data would permit the separate computation of the PACI for different sectors, and the resulting measures could eventually be aggregated into a general index. This paper shows that, under a set of assumptions, the PACI is valid in the sense that it registers a higher value in countries where the probability that a transaction is corrupt is higher. The validity of the index is sufficient in delivering a correct ranking of countries. In a previous version of this article, it was shown that under a further assumption, the PACI represents the probability that a transaction is corrupt, relative to a world average. To illustrate, if the PACI equals 200 for a given country, it would imply that the probability of corruption in that country is twice a world average. However, it is left for future stud- ies to explore the plausibility of such an interpretation, which may be appealing due to its practical applications but may be unrealistic. A better understanding of this issue would be welcome, considering that current cross-national measures of corruption are wanting in this respect. For example, in the 2014 TI-CPI, Germany and Turkey registered scores of 79 and 45 and rank 12th and 64th in the list of coun- tries, respectively. The differences imply that the (perceived) level of corruption in Turkey is considerably higher than in Germany, but the index used does not allow any interpretation regarding how much more corruption there is in the latter compared to the former. Under this light it should also be interpreted the finding that the PACI is highly correlated with the main perception-based measures of corruption, but it has a different scale. It was shown earlier that the scales become comparable once the log of the PACI is taken. If the PACI indeed approximates the proba- bility that a transaction is corrupt with respect to a world average, then the exponential of the WB-CCI and the TI-CPI would be approximately proportional to levels of corruption. Thus, further work in this direction might help clarify the scales of existing perception-based indicators. Other venues for future research could be foreseen. First, the availability of the PACI spurs a reassess- ment of the literature on the causes and consequences of corruption. As mentioned earlier, more data would allow the computation of the PACI for separate sectors of the economy. It would also be interest- ing to study the differences among several versions of the same indices, computed by focusing on differ- ent jurisdictions (along the lines of the presentation in Table 1). Since they all measure the same concept of corruption, they are expected to provide similar results. Thus, any variance among them could be explained by sampling error but also, possibly, by the violation of one or more of the maintained assumptions—the bigger the violation, the bigger the divergence. Tests for the validity of the maintained assumptions could be developed by leveraging on the magnitude of such observed differences. 16 Recent changes in the way the TI-CPI is computed should assure that “[it] will better capture changes in perception of corruption in the public sector of [a given] country over time. However, due to the update in the methodology, 2011 CPI scores are not comparable with CPI 2012 scores” (Transparency International 2012). The World Bank Economic Review 215 Lastly, the same intuition behind the use of judicial statistics in this article, could be applied to other domains. The essential ingredients needed for this are data culled from different jurisdictions that convey information on crimes committed in a given country by actors residing in another country. Some types of financial crimes may possibly lend themselves to a treatment based on the methodology introduced in this paper. Appendix A. Assumptions and Validity of the PACI. For a corrupt transaction to occur, both parties must be willing to engage in it. Firms headquartered in country i may decide to offer bribes to public officials in the foreign country j, with a probability that depends on the attributes of the foreign country. In particular, the probability that firms would offer bribes may be higher if the perceived level of corruption in the foreign country is high. This is because high levels of corruption would imply a lower risk of being caught and a higher social tolerance for bribery. The probability that a public official in the foreign country accepts a bribe when offered one may also depend on the attributes of the headquarters country. For example, if the latter is known to be proactive in curbing cross-border corruption, public officials may be deterred because the discovery of a corrupt act in the headquarters country may be followed by prosecution in the public officials’ respective countries. The probability that a public official in country j accepts a bribe, when he perceives that there is no risk of getting caught following a case first enforced in the headquarters country, is defined by prc orr FOj . The advantage of this concept is that, logically, it does not depend on the attributes of the headquarters country, as it purely reflects the propensity of country j’s public officials to accept bribes. This probability is identified as the level of corruption dem- onstrated by public officials in the foreign country. Meanwhile, a measure of corruption is deemed valid if it is monotonically increasing in the level of corruption. Definition. The PACI is valid if: @ PACIz =@ prc orr FOz > 0. The expected number of corruption cases observed and enforced first in the headquarters country i, involving pub- lic officials in the foreign country j, is determined as follows: cases obs HQi;j ¼ pro bs HQi;j Á corr exchi;j (A1) where corr exchi;j is the number of occurrences of corruption involving firms headquartered in country i and public officials in country j (which will later be equate to its expected value) and pro bs HQi;j is the probability that the cor- rupt exchange is observed, and enforced first in the headquarters country. The expected number of corrupt exchanges is: corr exchi;j ¼ prb ribe HQi;j Á prb ribe FOi;j Á transactionsi;j (A2) where prb ribe HQi;j is the probability that a firm headquartered in country i proposes a bribe to public officials in foreign country j, while prb ribe FOi;j is the probability that the public official in j accepts the bribe offered by the firm headquartered in i. Meanwhile, transactionsi,j is the total number of business transactions involving firms in country i and public officials in foreign country j. This formulation simply states that in order for a corrupt transaction to occur, both parties have to agree upon its execution. The possibility of extortion is ruled out, so that the probability that a transaction is corrupt is equal to a product of probabilities. Statistical independence between these events is not assumed, and the possibility that the probabilities of offering and accepting bribes are interdependent will be explicitly discussed subsequently. Assumption 1 The probability that a corrupt transaction involving firms from country i and public officials in country j is observed and first enforced in country i does not depend on the identity of the foreign country j: pro bs HQi;j ¼ pro bs HQi This assumption is of key importance because it is the basis for using the geographic distribution of the cases involv- ing country i firms to infer levels of corruption in all other countries. Assumption 2 The probability that for a given cross-border transaction, a firm headquartered in country i offers a bribe to public officials in country j is as follows: 216 Escresa and Picci prb ribe HQi;j ¼ prc orr HQi Á hðprc orr FOj Þ; where hð:Þ is a continuous and differentiable monotonically increasing function, and hð0Þ ¼ 1. This assumption describes how the probability that a firm will propose a bribe increases with the level of corrup- tion in the foreign country. The probability that a firm will offer a bribe abroad if it perceives that there is no risk of being caught following a case initiated in the foreign country is expressed by prc orr HQi . If such a risk is present, it would serve as a disincentive for firms in country i to propose a bribe. The second multiplicative factor, hð:Þ, is meant to capture such a deterrence effect. It equals one when there is no perceived risk that a corrupt transaction will be caught following a case first enforced in the foreign country, and it increases with the level of corruption of the for- eign country.17 Assumption 3 The probability that a public official in country j accepts a bribe when offered one by a firm headquartered in country i is as follows: prb ribe FOi;j ¼ prc orr FOj Á dð prc orr HQi Þ where dð:Þ is a continuous and differentiable function, and dð0Þ ¼ 1. Symmetrically, with respect to Assumption 2, this assumption expresses the probability that a public official accepts a bribe, and its dependence on the level of corruption in the headquarters’ country. The probability that a public official accepts a bribe if he perceives that there is no risk of getting caught following a case first enforced in the foreign country and then “spilling over” into the domestic jurisdiction, is expressed by prc orr FOj . This may be seen as the “underlying” probability of corruption of public officials in the foreign country j. Assumption 3 admits the possibility that, all else being equal, public officials in a given country may be more wary of accepting a bribe from a firm headquartered in a country that is very proactive in curbing cross-border corruption. It affirms that such a deterrence effect is functionally the same for all foreign countries – the function dð:Þdoes not depend on j. Assumption 4 Bilateral transactions are proportional to the value of exports from country i to country j, xij , according to a con- stant factor k: transactionsij ¼ k Á xij This assumption makes the number of cross-border transactions dependent on an observable variable – bilateral trade. Given Assumptions 1, 2, 3 and 4, equation (Eq. A1) becomes: cases obs HQi;j ¼ pro bs HQi Á prc orr HQi Á hðprc orr FOj Þ (Eq.A2) prc orr FOj Á dðprc orr HQi Þ Á k Á xij : By substituting this expression into the definition of the PACIz (Eq. 1), proving the following becomes straightforward: Proposition 1. The PACIz is valid: @ PACIz =@ prc orr FOz ! 0. Assumptions 1–4 guarantee that the PACI is valid in the sense that higher levels of probability of corruption of for- eign public officials results in a higher value for the index. Appendix B. Data Sources Corruption cases. The main sources of data are Trace International Compendium’s, a database on international anti- bribery enforcement (http://www.traceinternational.org/compendium), US DOJ and SEC documents, and OECD reports (various years). Other databases and publications, such as Shearman and Sterling 2013, Transparency International 2009 and 2013, and Cheung et al. 2012 were consulted. Various news sources, such as the Wall Street Journal Risk and Compliance Journal (http://www.wsj.com/news/risk-compliance-journal), and anti-corruption 17 In fact, the argument of the function hð:Þ should more appropriately be taken as expectations of levels of corruption in the foreign country. This, in particular, would lead to the possibility of expectations being partly self-fulfilling: in a country with a (initially, possibly undeserved) reputation for corruption, foreign firms would be prone to offer bribes more often, leading to more corrupt transactions. Considering prc orr HQi instead of expectations is justifiable if the lat- ter is assumed to depend monotonically from that underlying probability. The World Bank Economic Review 217 blogs like the FCPA Blog (http://www.fcpablog.com) were considered. Cases reported in multiple sources were labo- riously consolidated to avoid double counting. The reference period for each case is the year when the bribe was allegedly paid. However, in some instances, this date had to be presumed from the available data. A small number of cases could not be considered for the computation of the PACI because of lack of information on the identity of one of the two countries involved. All cases linked to the United Nations Oil-for-Food Programme, which allowed Iraq to sell oil in exchange for food, medicine and other humanitarian goods, were excluded because of their peculiar ori- gin. All the coding reflects information available on April 1, 2015. Raw data are available at http://www2.dse.unibo. it/picci/measure_corruption.html Exports: Barbieri and Keshk 2012. Gdp_cap: Gross domestic product based, per capita. Expressed in PPP dollar per person. Source: IMF-World Economic Outlook, October 2014 (variable name: PPPPC) (http://www.imf.org/external/Pubs/ft/weo/2015/01/). The secondary source of the following variables is the “Standard data” of the Quality of Government Institute dataset (Teorell et al. 2013) which assembles various sources. The May 2014 release of the dataset was used. A brief description of each variable is provided, taken verbatim from that dataset’s codebook. The original source, which the codebook identifies, is indicated. “Variable name” indicates how the variable is tagged in the codebook. Pop: population. Source: Heston, Summers and Haten 2012. Variable name: pwt_pop). R_g/gdp: the share of public expenditure over GDP (Source: Heston, Summers and Haten 2012. Variable name: pwt_gsg) Emp_right: empowerment rights index (Source: Cingranelli and Richard 2010. Variable name: ciri_empinx_new). It is an “additive index constructed from the Foreign Movement, Domestic Movement, Freedom of Speech, Freedom of Assembly & Association, Workers’ Rights, Electoral Self-Determination, and Freedom of Religion indicators. It ranges from 0 (no government respect for these seven rights) to 14 (full government respect for these seven rights).” Free_press: freedom of the press index (Source: Freedom House. Variable name: fh_fotpc3). “The press freedom index is computed by adding four component ratings: Laws and regulations, Political pressures and controls, Economic Influences and Repressive actions. The scale ranges from 0 (most free) to 100 (least free).” Free_speech: freedom of speech (Source: Cingranelli and Richard 2010. See Teorell et al. for details. Variable name: ciri_speech). “This variable indicates the extent to which freedoms of speech and press are affected by govern- ment censorship, including ownership of media outlets”. Voice_acc: Voice and Accountability (The World Bank. Variable name: wbgi_vae). “‘Voice and Accountability’ includes a number of indicators measuring various aspects of the political process, civil liberties and political rights. These indicators measure the extent to which citizens of a country are able to participate in the selection of govern- ments. This category also includes indicators measuring the independence of the media [. . .].” Checks: a measure of “checks and balances” (Source: Database of Political Intitutions Variable name: dpi_checks). “Equals 1 if the Legislative Index of Political Competitiveness (dpi_lipc) or the Executive Index of Political Competitiveness (dpi_eipc) is less than six. In countries where dpi_lipc and dpi_eipc are greater than or equal to six, dpi_checks is incremented by one if there is a chief executive, by a further one if the chief executive is competitively elected (dpi_eipc greater than six), and by a further one if the opposition controls the legislature [. . .].” Democ: index of democratization (Source: Vanhanen 2011 Variable name: van_index). “This index combines two basic dimensions of democracy – competition and participation – measured as the percentage of votes not cast for the largest party (Competition) times the percentage of the population who actually voted in the election (Participation). This product is divided by 100 to form an index that in principle could vary from 0 (no democracy) to 100 (full democracy). (Empirically, however, the largest value is 49).” Stability: political stability (Source: The World Bank. Variable name: wbgi_pse). “‘Political Stability’ combines several indicators which measure perceptions of the likelihood that the government in power will be destabilized or overthrown by possibly unconstitutional and/or violent means, including domestic violence and terrorism.” References Alt, J. E., and D. D. Lassen. 2012. “Prosecutement and Public Corruption: Evidence from the American States.” Journal of Law, Economics, and Organization. Barbieri, K., and O. Keshk. 2012. Correlates of War Project Trade Data Set Codebook, Version 3.0. Online: http:// correlatesofwar.org. 218 Escresa and Picci Carr, I. M., and O. Outhwaite. 2008. “The OECD Anti-Bribery Convention Ten Years Later.” Manchester Journal of International Law 5 (1): 3–55. Chang, E., M. Golden, and S. Hill. 2010. “Legislative Malfeasance and Political Accountability.” World Politics 62 (2): 177–220. Cheung, Y. L., P. R. Rau, and A. Stouraitis. 2012. “How Much Do Firms Pay as Bribes and What Benefits Do They Get? Evidence From Corruption Cases Worldwide.” National Bureau of Economic Research Working Paper No. 17981: 1–38. Choi, S., and K. E. Davis. 2012. “Foreign Affairs Prosecutement of the Foreign Corrupt Pratices Act.” Unpublished paper. Clausen, B., A. Kraay, and P. Murrell. 2011. “Does Respondent Reticence Affect the Results of Corruption Surveys? Evidence from the World Bank Enterprise Survey for Nigeria.” In S. Rose-Ackerman, ed., International Handbook on the Economics of Corruption. Cheltenam UK: Edward Elgar Publishing House. Vol. 2. 428–50. Fisman, R., and R. Gatti. 2002. “Decentralization and Corruption: Evidence from US Rederal Transfer Programs.” Public Choice 113 (1): 25–35. Glaeser, E. L., and R. E. Saks. 2006. “Corruption in America.” Journal of Public Economics 90 (6): 1053–1072. Goel, R., and M. Nelson. 2011. “Measures of Corruption and Determinants of US Corruption.” Economics of Governance 12: 155–76. Golden, M., and L. Picci. 2005. “Proposal for a New Measure of Corruption, Illustrated with Italian Data.” Economics & Politics 17 (1): 37–75. ———. 2008. “Pork Barrel Politics in Postwar Italy, 1953–1994.” American Journal of Political Science 52 (2): 268–89. Kaufmann, D., A. Kraay, and M. Mastruzzi. 2009. “Governance Matters VIII: Aggregate and Individual Governance Indicators, 1996–2008.” World Bank Policy Research Working Paper 4978. World Bank, Policy Research Department, Washington, DC. Klitgaard, R., and P. C. Light. 1998. “Economic Growth and Social Indicators: An Exploratory Analysis.” Economic Development and Cultural Change, 46 (3): 455–89 Klitgaard, R., J. Fedderke, and K. Abramamov. 2005. “Choosing and Using Performance Criteria.” In R. Klitgaard, and P. C. Light, eds., High Performance Government. Structure, Leadership, Incentives. Santa Monica, CA: Rand Corporation Kraay, A., and P. Murrell. 2013. “Misunderestimating Corruption.” World Bank Policy Research Working Paper 6488. World Bank, Policy Research Department, Washington, DC. Lambsdorff, J. G. 2006. “Consequences and Causes of Corruption: What Do We Know from a Cross-Section of Countries?” In S. Rose-Ackerman, ed., International Handbook on the Economics of Corruption. Cheltenam UK: Edward Elgar Publishing House. 3–52. ———. 2007. The Institutional Economics of Corruption and Reform: Theory and Policy. Cambridge, UK: Cambridge University Press. McLean, N. M. 2012. “Cross-National Patterns in FCPA Prosecutement.” The Yale Law Journal 121 (7): 1970–2012. OECD, Various years. “Country Reports on the Implementation of the OECD Anti-Bribery Convention.” http:// www.oecd.org/daf/antibribery/countryreportsontheimplementationoftheoecdanti-briberyconvention.htm. Olken, B. A. 2009. “Corruption Perceptions vs. Corruption Reality.” Journal of Public Economics 93 (7): 950–64. Picci, L. 2011. Reputation-based Governance. Stanford, CA: Stanford University Press. Putnam, T. L. 2009. “Courts without Borders: The Domestic Sources of US Extraterritorial Regulation.” International Organization 63 (3): 459–90. Razafindrakoto, M., and F. Roubaud. 2010. “Are International Databases on Corruption Reliable? A Comparison of Expert Opinion Surveys and Household Surveys in Sub-Saharan Africa.” World development 38 (8): 1057–1069. Reinikka, R., and J. Svensson. 2006. “Using Micro-surveys to Measure and Explain Corruption.” World Development 34 (2): 359–70. Rose-Ackerman, S. 1999. Corruption and Government: Causes, Consequences and Reform. Cambridge, UK: Cambridge University Press. Saisana, M., and A. Saltelli. 2012. “Corruption Perceptions Index 2012 Statistical Assessment.” JRC Scientific and Policy Reports, European Commission. The World Bank Economic Review 219 Seligson, M. 2006. “The Measurement and Impact of Corruption Victimization: Survey Evidence from Latin America.” World Development 34 (2): 381–404. Sherif, M., and H. Cantril. 1945. “The Psychology of ‘Attitudes’: Part I.” Psychological Review, 52 (6): 295–319. Shearman & Sterling Llp. 2013. “FCPA Digest of Cases and Review Releases Relating to Bribes to Foreign Officials under the Foreign Corrupt Practices Act of 1977.” Shearman and Sterling LLP, New York. http://www.shearman. com/files/Publication/287c1af0-f9cb-4c11-805d-91c409975b41/Presentation/PublicationAttachment/83d9dc0b- b80c-4ca4-877b-9efbba0952e7/FCPA-Digest-Jan2013_010213.pdf. Teorell, J., N. Charron, S. Dahlberg, S. Holmberg, B. Rothstein, P. Sundin, and R. Svensson. 2013. The Quality of Government Dataset, version 20. Dec13. University of Gothenburg: The Quality of Government Institute, http:// www.qog.pol.gu.se. Transparency International. 2009. “OECD Anti-Bribery Convention Progress Report 2009: Prosecutement of the OECD Convention on Combating Bribery of Foreign Public Officials in International Business Transactions.” http://www.transparency-usa.org/news/documents/FinalOECDProgressReport2009.pdf. ———. 2012. “Corruption Perceptions Index 2012: An updated methodology.” http://cpi.transparency.org/cpi2012/ in_detail/. ———. 2013. “Exporting Corruption: Progress Report 2013: Assessing prosecutement of the OECD Convention on Combating Bribery. http://www.transparency.org/whatwedo/pub/exporting_corruption_progress_report_2013_ assessing_prosecutement_of_the_oecd. Treisman, D. 2007. “What Have We Learned About the Causes of Corruption From Ten Years of Cross-National Empirical Research?” Annual Review of Political Science 10: 211–44. Van Aaken, A., L. Feld, and S. Voigt. 2010. “Do Independent Prosecutors Deter Political Corruption? An Empirical Evaluation across Seventy-Eight Countries.” American Law and Economics Review 12 (1): 204–44.