The 2017 TRACE Matrix Bribery Risk Matrix Methodology Report Corruption is notoriously difficult to measure. Even defining it can be a challenge, beyond the standard formula of using public position for private gain. This is due in part to the variety of situations in which corruption appears and the range of possible reasons for wanting to gauge its intensity. Different groups of interested parties activists, businesses, citizens, governments, lenders, policymakers, journalists are affected by corruption in different ways, and they all have different roles to play in the fight against it. Companies doing business transnationally have their own set of corruption-related concerns. Our publication of the TRACE Matrix in 2014 was designed to address one such concern: the risk of being asked for a bribe by a foreign public official. We concluded that this particular risk ought to be approached from a multidimensional perspective, taking into account not only the perceived expectation and tolerance of bribery within a given country, but the full range of factors that can affect a company s likelihood of encountering bribe demands. In this paper, we will briefly review the analytical model (originally developed by TRACE in cooperation with the RAND Corporation) that is at the core of the TRACE Bribery Risk Matrix. We will then discuss certain adjustments we have made to that model for the 2017 edition both substantive and methodological and our reasons for implementing them. Finally, we will address the effects of those changes on the Matrix s country scores and rankings, with some further observations on how the resulting numbers can be most meaningfully approached. The Model: Domains, Subdomains, and Data Sources The premise underlying the TRACE Bribery Risk Matrix is that bribery demands do not happen in isolation, but are made more or less likely by a variety of societal factors. We break those factors down into a set of four domains, each composed of two or three contributing subdomains. Domain 1 ( Opportunity ) Business Interactions with Government Contact with Government ( Interaction ) Expectation of Paying Bribes ( Expectation ) Regulatory Burden ( Leverage ) Domain 3 ( Transparency ) Government and Civil Service Transparency Transparency of Government Regulatory Functions ( Processes ) Transparency and Health of the Civil Service Sector ( Interests ) Domain 2 ( Deterrence ) Anti-Bribery Laws and Enforcement De Jure Anti-Bribery Laws ( Proscription ) De Facto Anti-Bribery Enforcement ( Enforcement ) Domain 4 ( Oversight ) Capacity for Civil Society Oversight Quality and Freedom of Media ( Free Press ) Human Capital and Social Development ( Civil Society ) 2017 TRACE International, Inc. All rights reserved. Page 1
The first domain ( Opportunity ) focuses on a company s (or its agents ) direct contact with foreign public officials, where the risk increases with the frequency of interaction (how many occasions there are for a bribe solicitation), along with the expectations surrounding bribery (how common it is and how thoroughly normalized) and the leverage an official might have at his disposal (how costly it could be made for the company to refuse a bribe demand). The second domain ( Deterrence ) addresses the intensity of the government s efforts to discourage bribery by enacting the necessary laws and by enforcing those laws effectively. With the third domain ( Transparency ), we examine how the government indirectly facilitates the detection of bribery by opening its books to inspection and maintaining them reliably. The fourth domain ( Oversight ) looks outside the government to consider whether the press and civil society are free enough and strong enough to provide a check on public corruption. With this model in place, the task is to identify available datasets containing relevant information for each domain and to select an appropriate subset of variables. Once the variables have been selected, the process of aggregating them to calculate domain and subdomain scores is relatively straightforward: normalize the data for each variable to a standard scale (with a mean of zero and a standard deviation of one); find for each country the arithmetic mean of all normalized variable scores 1 within each subdomain, then the arithmetic mean of all normalized subdomain scores within each domain; rescale the results for each domain and subdomain to yield a score between 1 and 100; and produce the total score as a weighted average of the four domain scores. 2 The 2014 and 2016 editions of the TRACE Matrix were based on publicly available data from nearly a dozen sources, including the World Bank, the United Nations, and the World Justice Project. 3 By aggregating this data at the domain and subdomain levels, we were able not only to present a set of overall risk scores specifically tailored to the anti-corruption concerns of companies doing business abroad, but also to provide a more focused level of detail regarding the nature and depth of the component risks. Substantive and Methodological Improvements for 2017 In preparing the 2016 edition, we began to recognize some ways in which the model could be improved in particular with regard to the breadth of the source data. We wanted to ensure that each subdomain score reflects a range of separate inputs, without overreliance on any one source. For the 2017 edition, we decided upon the following set of variables: 1 Because the available data is not always current, this year we have started discounting the normalized variable scores by age the weight dropping minimally at first (to 96% after one year), more dramatically in each successive year (to 84%, 64%, and 36%), until disappearing altogether after five years. 2 As in prior editions of the Matrix, Domain 1 is assigned the greatest weight, making up 45% of each country s total score. The remaining 55% is divided among the other three domains with one-seventh of that value coming from Domain 2 and three-sevenths each from Domains 3 and 4. (In other words: 45% for Domain 1; 7.86% for Domain 2; and 23.57% for each of Domains 3 and 4.) See RAND Corporation, Business Bribery Risk Assessment (2014), at 17 18 (available at https://www.rand.org/pubs/research_reports/rr839.readonline.html). 3 For complete details, see Business Bribery Risk Assessment at 34 37. 2017 TRACE International, Inc. All rights reserved. Page 2
Domain 1: Opportunity Subdomain 1.1: Interaction Public ownership in large firms CEPII Institutional Profiles Database Procedures to start a business Doing Business Procedures to obtain a construction permit Doing Business Procedures to get electricity Doing Business Procedures to register property Doing Business Documents to export goods Doing Business Documents to import goods Doing Business Senior management time spent dealing with the requirements of government regulation (%) If there were visits, average number of visits or required meetings with tax officials Share of government employment International Labor Organization Subdomain 1.2: Expectation ILOSTAT Government integrity Heritage Foundation Index of Economic Freedom Corruption between administrations and foreign businesses Bribery incidence (percent of firms experiencing at least one bribe payment request) Bribery depth (percent of public transactions where a gift or informal payment was requested) CEPII Institutional Profiles Database Irregular payments and bribes World Economic Forum Global Competitiveness Index Favoritism in decisions of government officials World Economic Forum Global Competitiveness Index Ethical behavior of firms World Economic Forum Global Competitiveness Index Government officials in the executive branch do not use public office for private gain World Justice Project Rule of Law Index Civil justice is free of corruption World Justice Project Rule of Law Index Subdomain 1.3: Leverage Difficulty of obtaining import licenses CEPII Institutional Profiles Database Days to start a business Doing Business Days to obtain a construction permit Doing Business Days to get electricity Doing Business Days to register property Doing Business Burden of government regulation World Economic Forum Global Competitiveness Index Burden of customs procedures World Economic Forum Global Competitiveness Index Government regulations are applied and enforced without improper influence World Justice Project Rule of Law Index 2017 TRACE International, Inc. All rights reserved. Page 3
Domain 2: Deterrence Subdomain 2.1: Proscription Anti-corruption policy Bertelsmann Stiftung Transformation Index Laws: Preventive anti-corruption policies and practices United Nations UNCAC Legal Library Laws: Preventive anti-corruption body or bodies United Nations UNCAC Legal Library Laws: Public sector United Nations UNCAC Legal Library Laws: Codes of conduct for public officials United Nations UNCAC Legal Library Laws: Bribery of national public officials United Nations UNCAC Legal Library Laws: Bribery of foreign public officials and officials of public international organizations Laws: Embezzlement, misappropriation or other diversion of property by a public official United Nations United Nations UNCAC Legal Library UNCAC Legal Library Laws: Trading in influence United Nations UNCAC Legal Library Laws: Abuse of functions United Nations UNCAC Legal Library Laws: Illicit enrichment United Nations UNCAC Legal Library Subdomain 2.2: Enforcement Prosecution of office abuse Bertelsmann Stiftung Transformation Index Regional cooperation Bertelsmann Stiftung Transformation Index Judicial independence CEPII Institutional Profiles Database Diversion of public funds World Economic Forum Global Competitiveness Index Judicial independence World Economic Forum Global Competitiveness Index Organized crime World Economic Forum Global Competitiveness Index Government officials are sanctioned for misconduct World Justice Project Rule of Law Index Criminal Justice World Justice Project Rule of Law Index Domain 3: Transparency Subdomain 3.1: Processes Transparency in public procurement CEPII Institutional Profiles Database Transparency regarding revenues from the exploitation of natural resources CEPII Institutional Profiles Database Transparency regarding privatization programme CEPII Institutional Profiles Database Transparency of government policymaking World Economic Forum Global Competitiveness Index Publicized laws and government data World Justice Project Rule of Law Index Right to information World Justice Project Rule of Law Index Open Budget Index International Budget Partnership Open Budget Index 2017 TRACE International, Inc. All rights reserved. Page 4
Subdomain 3.2: Interests Reliability of state-owned firms' accounts CEPII Institutional Profiles Database Public information on the capital structure of firms CEPII Institutional Profiles Database Transparency of information on listed companies CEPII Institutional Profiles Database Government powers are effectively limited by independent auditing and review World Justice Project Rule of Law Index Domain 4: Oversight Subdomain 4.1: Free Press Freedom of expression Bertelsmann Stiftung Transformation Index Freedom of political reporting Freedom House Freedom of the Press Freedom of the press CEPII Institutional Profiles Database Freedom of opinion and expression is effectively guaranteed World Justice Project Rule of Law Index Press Freedom Index Reporters Without Borders Subdomain 4.2: Civil Society Press Freedom Index Civil society participation Bertelsmann Stiftung Transformation Index Mobilisation of society to take up challenges CEPII Institutional Profiles Database Government powers are subject to non-governmental checks World Justice Project Rule of Law Index Civic participation World Justice Project Rule of Law Index Human Development Index United Nations Development Programme Human Development Index While many of these variables were used in the earlier editions of the Matrix, there have been some notable changes. We have added information from the Institutional Profiles Database produced by the international economics research center CEPII; 4 the Index of Economic Freedom published by the Heritage Foundation; 5 and the Bertelsmann Stiftung s Transformation Index. 6 The Global Integrity Report, which was the source for a substantial number of inputs in previous years, is no longer being regularly updated, and so is no longer being used in the Matrix calculations. For some of the subdomains, the new variable selection yields a certain shift in emphasis. For example, Subdomain 3.2 ( Transparency: Interests ) used to focus primarily on expert assessment regarding the existence and effectiveness of regulations governing conflicts of interest and asset disclosure. The current model focuses instead on reliability in the state s bookkeeping and transparency in the private sector, along 4 http://www.cepii.fr/cepii/en/cepii/cepii.asp 5 http://www.heritage.org/index/ 6 https://www.bti-project.org/en/home/ 2017 TRACE International, Inc. All rights reserved. Page 5
with whether government powers are effectively limited by independent auditing and review. Similarly, the scores for Subdomain 4.4 ( Oversight: Civil Society ) used to be derived exclusively from the U.N. Development Programme s Human Development Index; now, it also includes measures of civic participation and mobilization, as well as the strength of non-governmental checks on government power. With the inclusion of a broader range of variables, we were able to increase the overall level of data coverage for each country that is, the extent of available information covering all data-points of interest. This presented us with a chance to reexamine our methodology with respect to missing information. Previous editions of the Matrix relied on a statistical technique called multiple imputation of missing values to produce estimated data-points where source information was lacking. 7 This year, we have chosen instead to base our initial subdomain-score calculations solely on data that are available for each country. Where variables do not have corresponding data, those variables are simply omitted from the subdomain average. Where a subdomain has no available variable data, no score is calculated for that subdomain. The domain scores are then based only on those subdomains for which a score has been calculated. For a fraction of countries in the Matrix (about one-eighth of them), it was impossible to generate a complete set of four domain scores using this method, because one or more of the domains had no underlying variable data at all. Although we have left blank the uncalculated domain scores for these countries, we have nevertheless assigned each of them a top-level score as an estimate of overall risk level. To do this, we measured the distance of the country s extant domain scores from the corresponding scores for each complete country, 8 then took the average of each missing domain score over those countries weighted by the inverse of the calculated distances 9 to yield phantom domain scores from which a final top-level score could be derived. Although there is necessarily something arbitrary about this approach, we note that it yields a wider spread of final scores for low-data countries than the multiple imputation approach closer to what would be expected, given the variety of such countries (ranging from Monaco and the Marshall Islands to Equatorial Guinea and North Korea). This difference in spread suggests another methodological change. In the past, the Matrix has supplemented its final country scores by adding a variance penalty based on the divergence among the four domain scores: the greater the variation, the greater the penalty. 10 We surmise, however, that the imputational gap-filling method may have had an across-the-board flattening effect, as it was involved to some degree in calculating almost all of the countries domain scores. This compression toward the center would have dampened the scores natural variability, marginally reducing the impact of any compressed outliers on the final scores. The variance penalty was an indirect way to counter this dampening effect reasonable under the circumstances but, we conclude, no longer necessary. 7 See Business Bribery Risk Assessment, at 39. 8 Euclidean distance in n-dimensional space, n being the number of relevant domains. 9 Greater weight was also given to the scores of countries with greater data coverage. 10 See Business Bribery Risk Assessment, at 18 & n.64. 2017 TRACE International, Inc. All rights reserved. Page 6
What Do the New Scores Mean? The changes we have described concern both content and method the sources from which information is drawn, and the manner of proceeding where information is unavailable. These changes do not fundamentally alter the conceptual framework of the TRACE Bribery Risk Matrix, but they do have an effect on particular outcomes. It will therefore not generally be a straightforwardly meaningful exercise to compare a country s 2017 scores (top-level or domain-specific) to those published in previous years. That doesn t mean comparison is impossible, but it should be undertaken with care and with caution. Most notably, top-level risk scores are markedly lower this year. This isn t due to a worldwide drop in corruption levels, but simply follows our discontinuance of the variance penalty. Because of this drop, the scale for assigning a general risk level to a country ( low, high, very high, etc.) needs to be revisited. Our practice will be to look to the number of standard deviations by which a score differs from the overall mean. With a mean score this year of 46.5 and a standard deviation ( SD ) of 17.32, the best-ranking country Sweden, with a score of 5 stands 2.40 SDs ahead of the curve, while the worst Somalia, at 88 is 2.39 SDs behind. As statistically expected, most countries (139 out of 200) fall within a single SD of the mean score. For the sake of meaningful classification, we will consider any country within half an SD of the mean (87 out of 200) to be medium risk. The low-risk and high-risk groupings will each extend one SD from that medium range (32 and 50 countries respectively), while very low-risk and very high-risk classification will be reserved for those countries with total risk scores more than one and a half SDs away from the mean (19 and 12 countries respectively). Of course, the TRACE Bribery Risk Matrix was not developed to function as a single best-to-worst index. With its independent evaluation of multiple contributory factors, it aims to provide businesses with a more nuanced understanding of country-specific bribery risk than can be conveyed by a single number. Trying to measure any aspect of something as complex as corruption will necessarily entail a certain lack of precision, and minor differences among country scores should not be overinterpreted. We also recognize that our model remains imperfect, and we welcome thoughtful feedback on ways in which it might be improved further. But as a whole, the TRACE Bribery Risk Matrix provides not only a more accurate assessment of the risks specific to transnational public-sector bribery, but also the possibility of greater insight into the conditions and trends that drive those risks. 2017 TRACE International, Inc. All rights reserved. Page 7