39603 Poverty Reduction and Economic Management 273 April 2007 Findings reports on ongoing operational, economic, and sector work carried out by the World Bank and its member governments in the Africa Region. It is published periodically by the Operations Results and Learning Unit on behalf of the Region. The views expressed in Findings are those of the author/s and should not be attributed to the World Bank Group. Measuring Corruption: Myths and Realities Over the past decade measuring projects as well as more broadly in corruption has become an ever- developing countries. This in turn growing empirical field. Since the has also sparked debate on how best mid-nineties, we have under- to measure corruption and moni- taken various projects to measure tor progress in reducing it. In this corruption at the aggregate and context, some popular notions are disaggregated level. Among the espoused, which either lack clarity latter, we have carried out and or are not backed up by rigorous analyzed many surveys of the analysis or evidence. In this article enterprise sector. They have we highlight some of the main provided particular insights on the issues in these debates, in the private sector role in corruption. form of seven myths and their Such analysis of this empirical associated realities, and conclude work has led us to question by also pointing to some brief conventional approaches to implications for the private sector addressing private sector role in fighting corruption. corruption and investment climate constraints. This MYTH 1: Corruption cannot be empirical analysis question the measured traditional notion of viewing the REALITY: Corruption can, and is firm as an `investment climate being, measured in many forms. taker' and thus ignoring that Different approaches serve diffe- powerful conglomerates can also rent purposes, as seen in the shape the business climate and following three broad ways of thus become `investment climate measuring corruption: 2 makers'. This implies that it is warranted to move away from 1) By gathering the informed views simply blaming government of relevant stakeholders. These officials for prevailing corruption, include surveys of firms, public and to question the value of popular officials, and individuals, as well as initiatives such as voluntary--and views of outside observers in NGOs, often un-monitorable--codes of multilateral donors, the private conduct.1 sector and experts in investment rating agencies and think tanks. Progress in fighting corruption on These data sources can be used Findings all fronts requires measurement individually, or in aggregate of corruption itself, in order to measures which combine informa- diagnose problems and monitor tion from many such sources. Lite- results. This recognition has rally dozens of such sources are renewed interest in the World available, many of them covering Bank, and among aid donors, aid very large sets of countries, often recipients, investors, and civil so- over time for several years. These ciety, in developing measures of are the only available data sources corruption, both in aid-financed that currently permit large-scale cross-country comparisons and not appear to be substantial. It is uncertainty. No measure of monitoring of corruption over time. telling that perceptions of corruption `objective' or subjective, corruption from cross-country specific or aggregate, can be 100 2) By tracking countries' institutional surveys of domestic firms tend to percent reliable--in the sense of features. This provides information be very highly correlated with giving precise measures of that can be related to opportunities perceptions of corruption from ex- corruption. This imprecision or or incentives for corruption, such pert ratings in commercial risk measurement error stems from as procurement practices, budget rating agencies or multilateral two problems that are common to transparency. These do not development banks.3 all types of data, specific, subjective measure actual corruption, but or otherwise: can provide useful indications of Survey-based questions of corruption the possibility of corruption. In ad- have also become increasingly 1) There is measurement `noise' in dition of not constituting a direct specific, focused, and quantitative. specific corruption measures. A measure of corruption, these For example, we have survey question about corruption efforts as yet have limited country commissioned from the Global in the courts is subject to samp- coverage, and almost no time di- Competitiveness Survey ling error. An assessment of mension yet. coordinated by the World Economic corruption in procurement by a Forum the following specific commercial risk rating agency 3) By careful audits of specific question: "When firms like yours may not be accurate. Even a detai- projects. These can be purely do business with the government, led audit of a project cannot financial audits, or more detailed how much of the contract value conclusively distinguish between comparisons of spending with the must they offer in additional corruption, incompetence and physical output of projects. Such payments to secure the contract?". waste, and other sources of noise audits can provide information The results can be very specific-- in the data. about malfeasance in specific and also sobering--pointing in this projects within a very particular case to the extent and frequency 2) Specific measures of corruption context within a country, but not and extent to which firms-- are imperfectly related to overall about country-wide corruption including many multinationals-- corruption--or to another manifesta- more generally. These tend to be continue to pay bribes to obtain tion of corruption. A survey question one-time confined to specific public procurement contracts. Sim- about corruption in the police need projects and countries, and so are ilar specific questions are also not be very informative about not suited for cross-country posed by other firm surveys like corruption in public procurement. comparisons or for monitoring over the World Bank's Business Even if an audit turns up evidence time. Environment and Enterprise of corruption in a project, this need Performance Survey (BEEPS). And not signal corruption in other MYTH 2: Subjective data reflect household surveys like the projects, or elsewhere in the pu- vague and generic perceptions of Gallup's Voice of the People and blic sector. corruption rather than specific Global Barometer Surveys and the objective realities Tracking particular forms of Latinobarometro ask respondents corruption, and especially overall REALITY: Since corruption usually to report actual percentages of corruption at the country level, leaves no paper trail, perceptions of corrupt officials or actual number inevitably runs into one or both types corruption based on individuals' of times they witnessed acts of of measurement problems. Efforts to actual experiences are sometimes the corruption. In fact, these surveys measure corruption should aim at best, and the only, information we rely on citizens and firms residing minimizing measurement error have. Perceptions also matter in the country--even multination- and be transparent about what directly: when citizens view the als, which are a small minority in inevitably will always remain as courts and police as corrupt, they the enterprise surveys, provide residual error. For example, the will not want to use their services, responses through questionnaires Kaufmann-Kraay-Mastruzzi regardless of what the `objective' administered in the subsidiary corruption indicator average many reality is. Similarly, firms will pay operating in the recipient country. different data sources for each less taxes if they believe that they MYTH 3: Subjective data is too country, in order to reduce will be wasted by corruption, and unreliable for use in measuring measurement error. Unusually, in they will invest less in their coun- corruption these aggregate indicators try. Further, while social norms (measuring six dimensions of might affect what people view as REALITY: All efforts to measure governance, one of which is corruption, in practice such corruption using any kind of data corruption), we also report explicit cultural bias in perceptions does involve an irreducible element of margins of error,4 summarizing the remaining unavoidable noise. estimates of differences in judiciary, etc), it is very useful in Unfortunately, the practice of being corruption across different types of identifying priority areas for ac- explicit and transparent about interventions (such as a likely tion. Specific objective indicators imprecision in estimates of project audit). of opportunities for corruption are corruption or other dimensions of no more "actionable"--in the governance is very uncommon, and One can also obtain objective data sense of guiding specific policy thus should be generalized. on institutional features such as interventions. For example, one procurement practices or budget can measure whether a country Users of governance data should not procedures that may create has an anticorruption commission confuse the absence of explicit opportunities for corruption, for or not, or whether competitive margins of error with accuracy: all example through the Public bidding is mandated by fiat (`in the approaches to measuring Expenditure and Financial books') for some areas of public corruption and governance and Accountability (PEFA) initiative for procurement. But this does not tell investment climate more broadly monitoring financial management us whether such reforms are (such as Doing Business procedures in the public sector. effectively implemented and indicators), involve elements of Such approaches can usefully enforced on the ground, or whether uncertainty and margin of error. document the "on the books" or of- implementing such reforms in Nor should they confuse specificity ficial description of specific rules these specific areas will of corruption measures with preci- and procedures. But these will only necessarily have a large impact on sion or reliability. Very specific be imperfect proxies for actual corruption. measures, such as proxying for the corruption, not least because the "on opportunity for corruption in the ground" application of these rules Moreover, tracking even quite gene- procurement based on a review of and procedures might be very diffe- ral perceptions about corruption can procurement practices (or through rent.6 Given the common gap also be a useful way--even if not specific survey questions) are between what procedures, policies, alone--of monitoring anticorruption affected by both types of budgets and regulations are on the programs. In fact, governments in measurement error, as illustrated books as compared with the democracies around the world rely above. implementation on the ground, on polling data to set policy there should be no presumption priorities and track their progress. MYTH 4: We need hard objective that objective data is necessarily measures of corruption in order more informative about the MYTH 6: There is no need to mo- to progress in the fight against concrete reality than reports from nitor corruption closely since corruption experts, citizens or firms on the many countries with high corruption have also had fast REALITY: Since corruption is ground--irrespective of the extent growth clandestine, it is virtually impossible of `subjectivity' of the latter. to come up with precise objective Indeed, we have estimated the Skeptics of the anti-corruption measures of it. An innovative effort margins of error in the so called agenda are quick to point out to monitor corruption in road buil- `objective' indicators to be at least countries such as Bangladesh that ding projects in Indonesia as substantial as those in other score poorly on most cross-country illustrates the difficulties involved indicators. assessments of corruption, yet in constructing direct objective MYTH 5: Subjective measures of have managed to turn in measures of corruption.5 The audit corruption are not "actionable" impressive growth performance compared reported expenditures on and so cannot guide policymak- over the past decade. One should building materials with estimates ers in the fight against corruption not confuse these exceptions to the of materials actually used, based more general strong empirical fin- on digging holes in the roads and REALITY: Several different surveys ding that corruption adversely assessing the quantity and quality of firms and individuals ask detai- affects growth in the medium- to of materials present. But led and disaggregated questions long-run. Studies have shown that separating sand from gravel and about corruption in different areas of a one standard-deviation increase both from the soil present before government. As mentioned, there in corruption lowers investment the road was built, is difficult and are many kind of specific detail on rates by three percentage points inevitably involves substantial procurement bribery for instance and lowers average annual growth measurement error. As a result the that can nowadays be gathered by about one percentage point.7 study could not provide reliable through surveys. While such detail estimates of the level of does not always point to what These results are at some level corruption, although it was still specific detailed reforms needed difficult to interpret when we useful as it could provide good (say, within procurement, or recognize that corruption is likely to be a symptom of wider aggregate governance indicators (as detailed in the papers, in the institutional failures. A large body provide specific information on the web, and in this article), may of recent empirical work has private sector propensity to pay continue to serve a useful purpose documented that broader bribes to public officials from the for governance and anti-corruption measures of institutional quality perspective of the private-sector monitoring and analysis. They will explain a significant portion of "briber". Indeed, there is a very also challenge conventional income differences across high correlation between the notions--often devoid of empirical countries. One widely-cited study extent of corporate ethics by firms evidence--on what actually is tran- found that an improvement in in a country and overall country spiring on the public and private institutional quality from levels corruption as measured by our side of the corruption equation, observed in Nigeria to those in aggregate country indicator on and what can be done about it. Chile would translate into a seven- corruption. Further, the bribing fold difference in per capita practices by firms (which we do This article was authored by Daniel incomes in the long run.8 This type measure, including bribes for Kaufmann, Aart Kraay and Massimo of evidence suggests that policy- licenses, connection to utilities, Mastruzzi. It has been excerpted makers ignore corruption, and the tax evasion, procurement, judicial from the World Bank Institute's institutional failures that permit rulings, capture of policies, laws Development Outreach, September it, at their peril. and regulations, and the like), are 2006. explicit inputs into the corruption MYTH 7: Country level index in the aggregate governance Daniel Kaufmann is Director of Glo- aggregated indicators of indicators. Unethical behavior by bal Programs, The World Bank Insti- corruption refer specifically to the private sector in its conduct tute. Aart Kraay is Lead Economist, the extent of corruption within with the public sector is thus also Development Research Group the public sector, as assessed by encompassed within the overall (DECRG), The World Bank. a few experts. cross-country corruption indi- Massimo Mastruzzi is a Consultant, cators. Global Programs, The World Bank The interpretation of country Institute. indicators of corruption tends to In conclusion, for monitoring ignore the role of the private sector purposes, corruption can and is References in corruption. Specifically, often being measured through a wide the measure for corruption in our variety of innovative approaches. Acemoglu, Daron, Simon Johnson, aggregate governance indicators Given the imperfections of any and James Robinson. "The Colonial tends to be associated exclusively individual approach, it is Origins of Comparative with the extent of corruption within appropriate to rely on a wide va- Development." American Economic the public sector, often because riety of different indicators, both Review. 91(5):1369-1401. there is a mistaken perception subjective and objective, individual that these aggregate governance as well as aggregate, cross-coun- Olken, Ben (2005) "Monitoring indicators are solely based on ass- try as well as country-specific. This Corruption: Evidence from a Field essments by a few experts (residing is important to monitor results on Experiment in Indonesia." NBER in selected capitals of the rich the ground, assess the concrete Working Paper No. 11753. world). This is incorrect. First, the reality of corruption, and develop Hall, Robert E., and Charles Jones dozens of institutions from which anticorruption programs. We (1999). "Why Do Some Countries we gather data for our governance continue to further refine the Produce So Much More Output per indicators are based in many aggregate indicators of governance Worker than Others?" Quarterly countries and regions around the and corruption. With the latest re- Journal of Economics, 114(1):83- globe. Second, the data does not lease of these indicators in Sep- 116. only draw from expert assess- tember 2005 we are also making ments, but also prominently from public a vast dataset of virtually all Hsieh, Chiang-Tai and Enrico the responses from surveys of of the underlying individual data Moretti (2006). "Did Iraq Cheat the citizens and of firms worldwide, sources that go into our aggregate United Nations? Underpricing, who report on their experience with governance indicators. This data, Bribes, and the Oil for Food Pro- corrupt practices. and accompanying papers, and web- gram." Quarterly Journal of Econo- based analytic tools, are available mics, forthcoming. Firm surveys report on firms' at www.worldbank.org/wbi/ behavior regarding payment of governance/govdata. Kaufmann, Daniel and Aart Kraay bribes, as mentioned above. Such (2002) "Growth without individual measures of corruption Such data, with the accompanying Governance." Economia, Volume 3, that serve as an input to our dose of caution for interpretation No.1, Fall. Kaufmann, Daniel, Aart Kraay and Endnotes 4 How much measurement error Massimo Mastruzzi (2004) is reduced by aggregation depends "Governance Matters III: 1 For details on the rationale for on the extent to which individual Governance Indicators for 1996, rethinking conventional data sources provide independent 1998, 2000 and 2002." World Bank approaches to address corruption, estimates of corruption. In Economic Review. and the specific recommendations Kaufmann, Kraay, and Mastruzzi as an alternative to conventional 2006) we develop tests of this Kaufmann, Daniel, Aart Kraay and initiatives to mitigate the private independence assumption and Massimo Mastruzzi (2005) sector side of corruption (which show that it is not unreasonable. "Measuring Governance Using overall have not been seen as Perceptions Data," forthcoming in bearing results) , see "`Click 5 Olken (2005). Susan Rose-Ackerman, ed. Hand- Refresh Button: Investment book of Economic Corruption, Edward Climate Reconsidered," by D. Elgar. 6 See for example Kaufmann, Kaufmann, in Development Kraay, and Mastruzzi (2005) who Kaufmann, Daniel, Aart Kraay and Outreach, March 2005 issue. show that much of the difference Massimo Mastruzzi (2006) between objective measures of bu- "Governance Matters V: Aggregate 2 Kaufmann, Kraay and Mastruzzi siness entry based on statutory and Individual Governance (2005) provide an exhaustive list requirements and firms' Indicators for 1996-2005." World of 22 different data sources that perceptions of the ease of business Bank Policy Research Department provide perceptions data on entry, can be explained by the Working Paper, corruption. Examples of measuring extent of corruption. w w w . w o r l d b a n k . o r g / w b i / institutional features that create governance/govdata. opportunities for corruption include 7 Mauro (1995). See also Knack the Public Expenditure and Finan- and Keefer (2005). Kaufmann, Daniel (2005). "`Click cial Accountability (PEFA) frame- Refresh Button: Investment work, and the Public Integrity In- 8 Acemoglu, Johnson and Robin- Climate Reconsidered," dex of Global Integrity. Examples of son (2001). Other studies include Development Outreach, World Bank audits include Olken (2005), Hsieh Knack and Keefer (1995), Rigobon Institute, March. and Moretti (2006). and Rodrik (2005), Rodrik, Knack, Steven and Philip Keefer Subramanian and Trebbi (2004), (1995). "Institutions and Economic 3 The correlation between Hall and Jones (1999), and Performance: Cross-Country Tests corruption ratings from the Global Kaufmann and Kraay (2002). Using Alternative Measures." Eco- Competitiveness Surveys and ex- nomics and Politics, 7, 207-227. pert polls such as Economist Intelligence Unit, and Global Mauro, Paolo (1995). "Corruption Insight, or Multilateral Institution and Growth." Quarterly Journal of ratings such as the World Bank's Economics. 110(3): 681-712. Country Policy and Institutional Assessments (CPIA) are very high. Rigobon, Roberto and Dani Rodrik A related critique is that assess- (2004). "Rule of Law, Democracy, ments of corruption produced by Openness, and Income: Estimating think-tanks and commercial risk- the Interrelationships." rating agencies display ideological Manuscript. MIT and Kennedy biases, generally pro-market and School. pro-right-wing. In Kaufmann, Kraay, and Mastruzzi (2004) we Rodrik, Dani, Arvind develop a test for such ideological Subramanian, and Francesco biases and find that they are Trebbi (2004). "Institutions Rule: quantitatively unimportant. The Primacy of Institutions over Geography and Integration in Economic Development." Journal of Economic Growth 9(2):131-165.