Forum

THE CAUSES OF CORRUPTION

also the definition that the World Bank employs. Examples of corruption thus defined include taking bribes in return for issuing licenses; accepting a kickback for granting a defense contract; not enforcing rules and regulations in exchange for a payment and so on. The Figure provides a snap-shot of the world distribution of corruption in 2010 based on the corruption perception index published yearly by Transparency International. Each bar represents one of the 178 countries surveyed in 2010. The index ranges from 0 to 10, with a score of 10 indicating absence of corruption and a score of 0 indicating widespread corruption. To the left with high index values, we find the Scandinavian countries and Singapore, in the middle of the distribution we find countries like Thailand and Greece, while the most corrupt countries located to the far right include Iraq, Afghanistan and Somalia. A major challenge for economists and other social scientists interested in the causes of corruption and in offering advice to policy makers is to explain not only why corruption varies over time, as suggested by the two historical examples given above, but also to explain the distribution of corruption across the world illustrated in the Figure.

While corruption is usually recognised when confronted with it, it has proved more difficult to find and agree on a precise and encompassing definition. Most economists, however, are satisfied with some version of the following definition: corruption Figure is an act in which the power of public office is used WORLD DISTRIBUTION OF (PERCEIVED) CORRUPTION 2010 for personal gain in a manCorruption perceptions index 10 ner that contravenes the rules of the game (Jain 2001) or corruption is the 8 sale by government officials of government property for 6 personal gain (Shleifer and Vishny 1993). The latter is 4
2

* Faculty of Economics and Jesus College, University of Cambridge, UK. E-mail: [email protected]. uk. This report was written while visiting the Institute for Quantitative Social Science at Harvard University. The hospitality of the Institute is gratefully appreciated. I am also grateful to Jakob de Haan for helpful advise.

0 Countries
Notes: The corruption perceptions index scores countries on an inverse scale from 10 (very clean) to 0 (highly corrupt). Each bar represents one of the 278 countries surveyed and ordered from low to high corruption. Source: Transparency International (http://www.transparency.org).

15

CESifo DICE Report 2/2011

Forum
A conceptual framework To structure thinking about a complex issue such as corruption, it is useful to have a conceptual framework in mind. At the risk of over-simplifying a vast theoretical literature on the causes of corruption, surveyed superbly in, e.g., Tanzi (1998), Rose-Ackerman (1999) or Jain (2001), at least three conditions are necessary for corruption in the public sector to arise and persist: 1. Discretionary power: the relevant public officials (bureaucrats, politicians, etc.) must possess the authority to design or administer regulations and policies in a discretionary manner. 2. Economic rents: the discretionary power must allow extraction of (existing) rents or creations of rents that can be extracted. 3. Weak institutions: the incentives embodied in political, administrative and legal institutions must be such that public officials are left with an incentive to exploit their discretionary power to extract or create rents. Much, if not all, theoretical work on the causes of corruption can be linked back to these three basic conditions. Two contrasting views about how these conditions interact, however, stand out. The first view is the institutional view. It emphasises the role of institutions (broadly defined to include political, bureaucratic, juridical and economic institutions) and presumes a causal order running from weak institutions to corruption to poor economic outcomes. To give an example, the absence of democratic accountability may foster political corruption which, in turn, encourages rent extraction and excessive regulation of economic activity. Corruption is, in other words, seen as the symptom that something else, more fundamental, is wrong. Of course, feedbacks from economic conditions to corruption are usually also acknowledged as being important because they affect the presence or absence of economic rents, and it is also possible over longer time spans to imagine that corruption could be the cause of weak institutions. Yet, the organising and in many cases very powerful analytic principle is to look for weaknesses in the institutional framework to find the root causes of corruption. The second view is the social interactions view. It emphasises that corruption is the outcome of selfreinforcing social processes and downplays the causal role of institutions. The basic idea can be illustrated with an example along the lines of Andvig and Moene (1990). Corrupt colleagues are less likely to report that you are corrupt than honest colleagues simply because corrupt colleagues do not want to risk triggering an external inquiry. This makes the probability of being discovered/reported a decreasing function of how many others in the relevant social group take bribes. The incentive of any one individual to accept a bribe is obviously stronger when the likelihood of being discovered is lower. Combining these plausible assumptions, two types of stable situations can emerge: one in which no one takes bribes because anyone who does would be reported and punished, and one in which everyone take bribes because no one will ever be reported and caught. In other words, societies with otherwise very similar institutions and economic conditions may end up with very different levels of corruption. Shifts in institutions and economic conditions may trigger a swing from one equilibrium to the other, but it requires big shifts. According to this view, the key to understanding corruption is to understand how underlying social interactions work.

Cross-national studies on the causes of corruption At the macroeconomic level, much of what we know about the causes of corruption originates from crossnational comparisons of survey-based corruption indices. These indices can be divided, roughly speaking, into two categories. The first category contains indices based on corruption perceptions, i.e., subjective assessments by business consultants, by local and international businessmen or by ordinary citizens about how much corruption there is in a given country. The three most popular indices of this type are the Corruption Perception Index, published by Transparency International (and shown in the Figure), the Control of Corruption Index, published by the World Bank, and the Corruption Index published by the International Country Risk Guide. These measures of corruption have many weaknesses (see, e.g., the extensive discussion in Lambsdorff 2005); one of the major ones being that there may be a significant gap between perceptions and facts. The second category includes indices of experienced corruption. As the name suggests, these derive from selfreported experiences with corruption. The World Banks World Business Environment Survey, for example, asks managers in many countries to respond to the following statement: It is common for firms in my line of business to have to pay some irregular additional payment to get things done.

CESifo DICE Report 2/2011

16

Forum
Clearly, these indices are also problematic because of reporting biases, etc., but other than data for a few countries on criminal convictions or legal cases related to corruption (e.g., Italy and the US, see Goel and Nelson 2011), these are the macro data available, and they have, despite their weaknesses, given a significant boost to research into the causes of corruption during the past decade. In fact, 35 (and still counting) published, empirical studies have investigated the relationship between these indices and at least 75 potential causes of corruption. rion for robustness is that 95 percent of the cumulated density associated with the estimated coefficients on the variable of interest (e.g., GDP per capita) across all the models considered should be on one side of zero. Sturm and de Haan (2005), however, advocate going one step further and demand that a robust variable should be significant at the 5 percent level in at least 90 percent of all the regressions considered (test 1) and when there is more than one outcome variable (as is the case with corruption where one needs to study alternative indices) that this should be true for the majority of outcome variables considered (test 2). The result of such a Sensitivity Analysis applied to the determinants of five widely used corruption perception indices (including the three mentioned above) are reported in de Haan and Seldadyo (2005) and Seldadyo (2008). The Table summarises the results by listing the 12 most robust determinants of corruption and their correlation with corruption. All 12 variables pass Sala-i-Martin (1997)s criterion for robustness for at least one corruption index, but we see that only six, in addition, pass test 1 and that only two pass test 2. Based on

The potential causes can usefully be divided into four groups: (1) economic and demographic factors; (2) political institutions; (3) judicial and bureaucratic factors; and (4) geographical and cultural factors. The typical study limits attention to a small number of potential causes of particular interest and examines the reliability of the results by varying the set of control variables (see, e.g., Treisman 2000, Paldam 2002 or Lambsdorff (2005) for examples of this approach). Unfortunately, it is not uncommon to discover that variables such as press freedom, the size of the public sector, economic freedom or the level of democraTable cy found to be significantly corThe top-12 most robust determinants of cross-national corruption perceptions related with corruption in one Sign of study are insignificant in another Test 1 Test 2 Variable name Group correlation which uses an alternative model Government effectiveness (3) () Yes Yes specification. Moreover, Treisman Rule of law (3) () Yes Yes (2007) points out that standard GDP per capita (1) () Yes No determinants of perceived corrupRegulatory quality (3) () Yes No tion, such as GDP per capita and Political polarisation (2) () Yes No measures of political and bureauProtestant fraction of the cratic institutions, are largely un(4) () Yes No population correlated with measures of exPresidentialism (2) () No No perienced corruption. This is clearAbsolute latitude (4) () No No ly unsettling. One way out of the Voice and accountability (2) () No No forest, originally proposed by Wage bill in % of GDP (3) (+) No No Sala-i-Martin (1997) in a different Population size (1) (+) No No context, is to use so-called SensiEconomic freedom (1) () No No tivity Analysis to establish which Notes: For the full set of variables included in the Sensitivity Analysis and for of the many potential determiinformation on definitions and sources, see Appendix 2 in Seldadyo (2008). nants or causes are robustly corThe 12 variables are the only ones that pass Sala-i-Martins criterion for related with corruption. This inrobustness for at least one of the five corruption perception indices studied. Group refers to the four categories of causes of corruption listed in the volves a systematic evaluation of main text. Sign of correlation refers to the sign of the coefficient on the all possible linear (regression) morelevant variable in 95% of the regressions. The outcome variable of these regressions is one of the five corruption indices, ordered such that higher dels with corruption as the depenvalues mean higher levels of corruption. Each regression includes up-to-three dent variable and a fixed number potential determinants, lagged by a suitable number of years. Test 1 requires that the variable is significant at the 5% level in at least 90% of all of potential determinants (typithe regressions considered. Test 2 requires that test 1 is passed for three of cally 3 to 5) from the target list of the five corruption perception indices considered. up to 75 as the explanatory variSource: Seldadyo (2008: chapter 3). ables. Sala-i-Martin (1997)s crite-

17

CESifo DICE Report 2/2011

Forum
this, the two most robust determinants of corruption are government effectiveness and rule of law, from the World Banks Governance Matter Database (Kaufmann et al. 2006). Both of these variables correlate negatively with perceived corruption, however measured. Government effectiveness is a composite index related to the quality of public services, the quality of the bureaucracy, the competency of civil servants, and to the independence of the civil service from political pressures, while rule of law is an index related to the extent to which agents have confidence in and abide the rules of society, the effectiveness and predictability of the judiciary and the enforceability of contracts. A single variable from each of the groups also passes test 1 (but not test 2). They are GDP per capita, the degree of political polarisation, regulatory quality and the fraction of Protestants in the population, which are all negatively correlated with corruption. While the Sensitivity Analysis methodology is helpful in finding robust correlations, it does not resolve the much more thorny issue of causality. In fact, for many of the variables that make it to top-12, we suspect that the causality might run the other way. For example, perhaps the government is effective because corruption is low; or agents have trust in the rule of law because corruption is low; and GDP per capita is high because of low corruption, rather than the other way around.1 greater willingness to combat corruption or, more subtlety, high economic growth may provide incentives for public officials to eliminate corruption-induced economic inefficiencies in the present in order to hang on to their office or job for the future where even bigger rents can be extracted (Aidt and Dutta 2008). It is, of course, also possible that GDP per capita and corruption are jointly determined and caused by the same underlying institutional or cultural factors. Recent research has made some headway in sorting out this puzzle. When applied at the macro level, the theories of corruption based on social interaction effects suggest that corruption and economic development may feed on each other and that we should, therefore, not expect to observe simple linear and unidirectional relationships between proxies for the two. Corruption-development traps may lock some countries into a bad equilibrium with high-corruption-low-growth, while others may converge to a good equilibrium with low-corruption-high-growth (Blackburn et al. 2006). Aidt et al. (2008) and Aidt (2009) model such interdependencies empirically and find robust evidence that countries endogenously sort themselves into different corruption-growth regimes depending on the initial quality of their political institutions, and that, in the short to medium run, corruption and growth feed on each other: high growth tends to lower corruption which, in turn, enhances the growth potential. Moreover, using instrumental variables to isolate the link from corruption to growth, it appears that conditional on converging to the bad equilibrium, corruption has little effect on growth, while the impact is large and negative for countries converging to the good equilibrium. Gundlach and Paldam (2009), on the other hand, inquire into the causal order over the very long run. Their starting point is the widely accepted assumption that all countries had more or less the same (low) level of income per capita about 200 years ago and that the actual cross-national income distribution today, therefore, effectively represents differences in long-run growth rates. The work by Diamond (1997), moreover, suggests that deep pre-historical factors, such as the number of domesticable big mammals in pre-history, the number of domesticable wild grasses in pre-history, climatic conditions favourable for agriculture, relative EastWest orientation, etc., sowed the seeds that explain cross-national income patterns today. Since one can reasonably assume that such factors are exogenous to cur-

The corruption-development nexus The question of causality is particularly pressing in the case of the corruption-development nexus. Does the strong and robust correlation between GDP per capita and the various corruption indices tell us that development reduces corruption or does it tell us that corruption is an obstacle to development? Both directions of causation are plausible. On the one hand, corruption feeds on rents, and corrupt public officials have an incentive to create and maintain rent-creating but inefficient economic policies. The likely consequence of such inefficiencies is a reduction in national income. Likewise, corruption often works as a tax on factor accumulation and on investments in manufactured and human capital. This can retard economic growth. On the other hand, high levels of national income may be associated with a

Seldadyo (2008) uses lagged values of the potential determinants in his Sensitivity Analysis to partially address this issue.

CESifo DICE Report 2/2011

18

Forum
rent corruption patterns, they can, in principle, be used as instruments for national income in an estimation of the income effect on corruption. Based on this approach, Gundlach and Paldam (2009) conclude unambiguously that the long-run causality runs from (high) income to (low) corruption or as they put it, societies grow honest. Adding the two pieces of evidence, it appears that in the short to medium run, corruption and economic development feed on each other in a self-reinforcing social process,2 while in the very long run, the process of development dominates and is what causes corruption levels to fall. The precise mechanism through which this is supposed to happen remains, however, unclear, but we may conjecture that it has to do with investments in stronger and more robust institutions.
Gundlach, E. and M. Paldam (2009), The Transition of Corruption: From Poverty to Honesty, Economics Letters 103(3), 14648. Jain A. K. (2001), Corruption: A Review, Journal of Economic Surveys 15(1), 71121. Kaufmann, D., A. Kraay and M. Mastruzzi (2006), Measuring Governance Using Cross-country Perceptions Data, in S. RoseAckerman, ed., International Handbook on the Economics of Corruption, Edward Elgar, Cheltenham, UK. Lambsdorff, J. G. (2005), Consequences and Causes of Corruption: What do We Know from a Cross-Section of Countries?, University of Passau, Passau. Paldam, M. (2002), The Cross-country Pattern of Corruption: Economics, Culture and the Seesaw Dynamics, European Journal of Political Economy 18(2), 21540. Rose-Ackerman, S. (1999), Corruption and Government, Causes, Consequences and Reform, Cambridge University Press, Cambridge, UK. Sala-i-Martin, X. X. (1997), I Just Ran Two Million Regressions, American Economic Review, Papers and Proceedings 87(2), 17883. Seldadyo, H. (2008), Corruption and Governance Around the World: An Empirical Investigation, Ph.D. dissertation, Rijksuniversiteit Groningen, PPI Publishers, Enschede, The Netherlands. http://dissertations.ub.rug.nl/FILES/faculties/feb/2008/h.s.gunardi/11_thesis.pdf Shleifer, A. and R. W. Vishny (1993), Corruption, Quarterly Journal of Economics 108, 599618. Sturm, J.-E. and J. de Haan (2005), Determinants of Long Term Growth: New Results Applying Robust Estimation and Extreme Bounds Analysis, Empirical Economics 30(3), 597617. Tanzi, V. (1998), Corruption Around the World: Causes, Consequences, Scope and Cures, IMF Staff Papers 45 (4), 55994. Treisman, D. (2000), The Causes of Corruption: A Cross-national Study, Journal of Public Economics 76, 39957. 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, 21144.

Concluding remarks So what are the causes of corruption? The short answer is that many factors play a role and in different ways at different time horizons. Yet, for practical purposes, applying the two general principles institutional weaknesses and social interaction effects alluded to above will in most cases not miss the target too badly and, therefore, offer a workable guide to those interested in understanding the causes of corruption and in doing something about them.

References
Aidt, T. S. and J. Dutta (2008), Policy Compromises: Corruption and Regulation in a Democracy, Economics and Politics 20(3), 33560. Aidt, T. S., J. Dutta and V. Sena (2008), Governance Regimes, Corruption and Growth: Theory and Evidence, Journal of Comparative Economics 36, 195220. Aidt, T. S. (2009), Corruption, Institutions and Economic Development, Oxford Review of Economic Policy 25, 27191. Andvig, J. C. and K. O. Moene (1990), How Corruption may Corrupt, Journal of Economic Behaviour and Organization 13(1), 6376. Blackburn, K., N. Bose and E. M. Haque (2006), The Incidence and Persistence of Corruption in Economic Development, Journal of Economic Dynamics and Control 30, 244767. de Haan, J. and H. Seldadyo (2005), The Determinants of Corruption: A Reinvestigation, unpublished paper prepared for the EPCS-2005 Conference Durham, England, 31 March3 April 2005. Diamond, J. (1997), Guns, Germs, and Steel: The Fates of Human Societies, Norton, New York. Goel, R. K. and M. A. Nelson (2011), Measures of Corruption and Determinants of US Corruption, Economics of Governance, in press.

Paldam (2002) refers to this as seesaw dynamics.

19

CESifo DICE Report 2/2011