University of Groningen. Corruption and governance around the world Seldadyo, H.

University of Groningen Corruption and governance around the world Seldadyo, H. IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below. Document Version Publisher's PDF, also known as Version of record Publication date: 2008 Link to publication in University of Groningen/UMCG research database Citation for published version (APA): Seldadyo, H. (2008). Corruption and governance around the world: An empirical investigation Enschede: PrintPartners Ipskamp B.V., Enschede, The Netherlands Copyright Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons). Take-down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum. Download date: 19-03-2018

CORRUPTION AND GOVERNANCE AROUND THE WORLD An Empirical Investigation Harry Seldadyo Gunardi

Publisher: Printed by: PPI Publishers, Enschede, The Netherlands PrintPartners Ipskamp, Nederland ISBN: book: 978 90 367 3299 4; ebook: 978 90 367 3300 7 c 2008 Harry Seldadyo Gunardi Alle rechten voorbehouden. Niets uit deze uitgave mag worden verveelvoudigd, opgeslagen in een geautomatiseerd gegevensbestand, of openbaar gemaakt, in enige vorm of op enige wijze, hetzij elektronisch, mechanisch, door fotkopieën, opnemen of enige andere manier, zonder voorafgaande schriftelijke toestemming van de auteur. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system of any nature, or transmitted in any form or by any means, electronic, mechanical, now known or hereafter invented, including photocopying or recording, without prior written permission of the author.

RIJKSUNIVERSITEIT GRONINGEN CORRUPTION AND GOVERNANCE AROUND THE WORLD An Empirical Investigation Proefschrift ter verkrijging van het doctoraat in de Economie en Bedrijfskunde aan de Rijksuniversiteit Groningen op gezag van de Rector Magnificus, dr. F. Zwarts, in het openbaar te verdedigen op donderdag 20 maart 2008 om 16.15 uur door Harry Seldadyo Gunardi geboren op 8 januari 1966 te Jakarta, Indonesia

Promotor: Prof. dr. J. de Haan Beoordelingscommissie: Prof. dr. N. Hermes Prof. dr. M. Paldam Prof. dr. T. J. Wansbeek

Acknowledgements Ladies and Gentlemen... Now we have a winner. In the world corruption league, the grand prize goes to... Mr. Soeharto, the former president of Indonesia! Let s have a loud cheer for the most corrupt president in the globe! (see, Tranparency International: The Global Corruption Report 2004 ). That is my personal reason why I am now in the research area of corruption. But it would have never happened without Jakob de Haan. On the very first day I formally became a part of the University of Groningen, he asked me if I were interested in this issue. Since then till the day I completed this thesis, my days and nights were only about corruption. And, as will be shown in this dissertation, corruption is a serious problem. This thesis is the result of collaboration with many people. First of all, I have to mention again Jakob de Haan, a person with many roles: supervisor, discussion partner, and co-author; without him I could not have completed my PhD project. I also enjoyed working with Emmanuel Pandu and Paul Elhorst. Two chapters in this dissertation were written with them. Like Paul Elhorst, Maarten Allers inspired me to investigate spatial issues in governance. Tom Wansbeek s comments have very much improved the chapter on corruption persistence. Jan-Egbert Sturm and Richard Jong-A-Pin introduced the RATS program enabling me to run millions of regressions. Paramita Muljono of the UEA-UK kindly shared her data on the Global Competitiveness Report. Ward Romp gave me his beautiful L A TEXprogram. Also, I am indebted to the reading committee Niels Hermes (University of Groningen), Tom Wansbeek (University of Amsterdam), and Martin Paldam (University of Aarhus, Denmark) for their willingness to read the final manuscript and for their valuable comments. Certainly, I thank the

ii two paranymphs Poppy and Jeroen who have helped me a lot in arranging many things before and during the defence. Some other names have to be mentioned. Anneke Toxopeus and the other nice people at the ISD-RUG were always helpful whenever I had questions or came to their office with the blue envelope from the Belastingdienst. I thank the management of the Ubbo Emmius Scholarship program who made me possible to do my PhD project at this University. Astrid, Dity, Ellen, Rina, and Martin from SOM who have helped me a lot. Groningen, even the Netherlands, is geographically not a big place. Yet, there are many friends who have supported my study, directly and indirectly: those in PPI, PD, and GBI. Also, the relatives in the Netherlands have played their own roles. It is impossible to list all of them, but their names are written in my heart. I am thinking of my parents and parents in law in Indonesia. From a distance, I believe, they pray for me. Finally, a very special thanks must be given to Mia for her patience and support in days and nights, sadness and happiness, summer and winter. She is a blessing from God. I cannot imagine what I could do without her on my side. Now, as promised, I have more time to spend with her. Harry Seldadyo

Contents 1 Setting the Scene 1 1.1 Basic Story............................ 1 1.2 Main Findings.......................... 6 1.2.1 Determinants of Corruption............... 6 1.2.2 Corruption Persistence.................. 7 1.2.3 Governance and Growth................. 8 1.2.4 Spatial Dimension.................... 9 2 Determinants of Corruption: A Survey 11 2.1 Introduction............................ 11 2.2 Corruption: Definition and Measurement........... 12 2.2.1 Definition......................... 12 2.2.2 Measurement....................... 13 2.2.3 Criticisms......................... 17 2.3 Determinants of Corruption................... 20 2.3.1 Four Classes....................... 20 2.3.2 Empirical Issues..................... 29 2.4 Concluding Remarks....................... 31 3 On the Sensitivity of Corruption Determinants 33 3.1 Introduction............................ 33 3.2 The Setup............................. 34 3.3 Some First Results........................ 38 3.4 Effect of Observations...................... 45

iv Contents 3.5 Concluding Remarks....................... 54 4 Is Corruption Really Persistent? 57 4.1 Introduction............................ 57 4.2 Convergence: Preliminary Evidence............... 59 4.3 A Closer Look.......................... 70 4.4 Markov Chain Analysis..................... 74 4.5 Conclusion............................ 79 5 Governance and Growth Revisited 81 5.1 Introduction............................ 81 5.2 Governance: Concept and Construct.............. 83 5.3 Governance and Growth..................... 88 5.3.1 Parsimonious Model................... 88 5.3.2 Effect of Observations.................. 89 5.3.3 Sensitivity Analysis................... 90 5.4 Conclusion............................ 93 6 Geography and Governance: Does Space Matter? 95 6.1 Introduction............................ 95 6.2 Spatial Dependence........................ 97 6.2.1 Preliminary Evidence.................. 97 6.2.2 Spatial Regression Models................ 100 6.2.3 Data............................ 102 6.2.4 Results.......................... 104 6.3 Spatial Heterogeneity....................... 108 6.3.1 Local Statistics...................... 108 6.3.2 Geographically Weighted Regression.......... 112 6.4 Conclusion............................ 115 7 Putting the Pieces Together 117 7.1 Conclusions and Limitations of the Research......... 117 7.2 Policy Implications........................ 121

Contents v Bibliography 123 Appendix 139 Samenvatting (Summary in Dutch) 155

List of Tables 2.1 Indicators of Corruption, Correlation and Summary Statistics 18 3.1 Robustness Analysis (No F, Z up-to-3)............ 40 3.2 Robustness Analysis (F =0, Z up-to-4 and up-to-5)...... 43 3.3 Robustness Analysis (F =0, Z up-to-3, Various N)...... 48 3.4 Robustness Analysis (F =0, Z up-to-4, Various N)...... 51 3.5 Robustness Analysis (F =0, Z up-to-5, Various N)...... 53 4.1 Losers and Winners 1984-2003................. 61 4.2 Corruption Correlation Over Time............... 63 4.3 Ordered Logit Regression.................... 65 4.4 Ordered Probit Regression.................... 67 4.5 Tests of β-convergence...................... 69 4.6 Tests of σ-convergence...................... 70 4.7 Modality (Rescaled) ICRG data................. 74 4.8 Markov Chain Estimates, 1984-2003.............. 78 5.1 Correlations among the ICRG Indicators of Governance... 85 5.2 CFA Estimates.......................... 87 5.3 Parsimonious Growth-Governance Regressions......... 89 6.1 Moran s I and Geary s c for Governance............ 100 6.2 Spatial Governance using W 1.................. 105 6.3 Spatial Governance using W 2.................. 107 6.4 Normalization by Largest Eigenvalue (W 1 and W 2 ).... 110

viii List of Tables 6.5 Significance Tests......................... 115

List of Figures 1.1 10 Most Corrupt and Cleanest Countries 2005-2006...... 3 3.1 Corruption (y-axis) and Two Robust Determinants (x-axis). 50 4.1 Annual Kernel Density, 1984 (top-left) to 2003 (bottom-right) 73 5.1 Governance in Factor Analysis Model............. 86 5.2 Regressions with Various Samples................ 91 5.3 Sensitivity Analysis........................ 92 6.1 Governance and Distance to Worst and Best Practices, 2005. 99 6.2 Local Moran s I Statistics (Matrices W 1 and W 2 )...... 111 6.3 Local Geary s c Statistics (Matrices W 1 and W 2 )...... 111 6.4 Local Coefficients......................... 114

Chapter 1 Setting the Scene How do I tell the story of trillions of dollars of dirty money lodged around the world, derived from many types of illegal activities, shifting across some 200 countries, affecting both economic and political affairs? Raymond W. Baker (2005) 1.1 Basic Story Corruption is one of the unholy trinity of dirty money, together with criminal and illegal commercial activities (Baker, 2005). It is a dark side of society, claimed to cause people in corrupt countries to stay poor and illiterate, and to suffer from high infant and child mortality rates, low birth-weight babies, as well as high dropout rates in primary schools (Kaufmann et al., 1999; Gupta et al., 2001). Not only that, corruption is also said to deteriorate countries distribution of income (Li et al., 2000), quality of public infrastructure (Tanzi and Davoodi, 1997), and productivity (Lambsdorff, 2003a). Clearly, it is a serious problem and nothing good can be expected from corruption. As Tanzi (1997: 164-165) states: I must confess that I have little patience for those who try to find benefits in corruption. These individuals are often addressing artificial or unusual situations. Unfortunately, when corruption exists, it is often widespread. It affects not just some decisions or sectors but many decisions and most

2 Chapter 1 sectors. Thus, in reality, corruption is likely to distort markets and to impose major costs on the economy. As corruption is a world-wide phenomenon, not surprisingly it has attracted a lot of attention in both the policy and academic arenas. Some decades ago, there were only a few studies on corruption. Since then, numerous academic articles have been published. 1 This partly reflects an increased public concern for the problems caused by corruption, and partly explains the emergence of various perceived-corruption indexes that have made corruption measurable. Today, the world is still split in countries that are governed by clean governments and those that are not. 2 The group of clean countries mainly consists of Western nations, while the group of dirty countries predominantly contains African, Asian, and Latin American nations. The International Country Risk Guide (ICRG) that produces a perceived-corruption index reports that 79 per cent of the around 140 nations in the world (2005-2006) are run by corrupt bureaucrats. An aggregated corruption index of Kaufmann et al. (2007) of the World Bank (WB) gives a lower figure: 59 per cent of 207 countries and territories are corrupt. 3 Figure 1.1 shows the 10 most and least corrupt countries in 2005-2006 according to both indexes. Finland is always regarded as the cleanest country by these indexes. Meanwhile, Somalia is the most corrupt country according to the WB index, and Zimbabwe according to the ICRG data. But, what is corruption? Corruption can be viewed from different angles. A commonly accepted view is Waterbury s (1973: 533) definition explaining corruption as the abuse of public power and influence for private ends. Jain (2001) gives a socio-political context of corruption, namely discretionary power, economic rents, and a weak judicial system in defining 1 The Google Scholar search engine (12 November 2007) indicated some 157,000 articles on corruption in social sciences, arts, and humanities. 2 Later, we will show that there is a tendency of convergence via upward and downward dynamics. 3 The ICRG index is scaled between 0 and 6, while the WB index ranges between 2.5 and +2.5, where a higher score denotes less corruption. The division of the countries into corrupt and clean countries and territories is based on the middle score of each index.

Setting the Scene 3 2.5 1.5 0.5-0.5-1.5-2.5 Somalia Myanmar Eq.Guinea N.Korea Afghanistan Haiti Rep.Congo Iraq N.Caledonia Zimbabwe Austria Netherlands Norway Switzerland Sweden Singapore N.Zealand Denmark Iceland Finland (a) WB Index 6.0 5.0 4.0 3.0 2.0 1.0 0.0 Zimbabwe Kenya Rep.Congo Gabon Iraq Lebanon Myanmar Niger N.Korea PNG Austria Canada Luxembourg Netherlands Norway Sweden Denmark Iceland N.Zealand Finland (b) ICRG Index Figure 1.1: 10 Most Corrupt and Cleanest Countries 2005-2006

4 Chapter 1 corruption. He (p. 73) defines corruption as an act in which the power of public office is used for personal gain in a manner that contravenes the rules of the game. Corruption also reflects poor governance of government, i.e., public officials are not qualified, laws are not adhered to, and lack of public transparency and accountability. Government governance itself refers to the process of decision-making by the government and the process by which decisions are implemented (or not implemented). Poor governance is often regarded as one of the root causes of all evil within societies. Although definitions of governance differ, there seems to be broad consensus that good governance has various characteristics: It is accountable, transparent, responsive, effective and efficient, and follows the rule of law, thereby assuring that corruption is minimized. Clearly, corruption is a failure of good governance or a manifestation of poorly functioning state. indicated by Hassan (2004: 32). In his words, How corruption reflects poor governance is In the political realm, corruption undermines democracy and good governance by subverting formal processes and rules of conduct. It erodes the institutional capacity of government as established procedures are disregarded, resources are siphoned off, and officials are assigned or promoted without regard to performance. Corruption in elections usually elects the wrong people, those who are parasites and put personal greed over national interests. Corruption in legislative bodies undermines accountability and representation in policy making. In the administrative realm, corruption results in the unequal provision of services, which undermines the States legitimacy and, in extreme cases, may render a country ungovernable and lead to political instability and social conflict. Corruption in the judiciary circumvents the rule of law, and justice is often delivered late or even denied. This dissertation mainly deals with the issues of corruption and governance. It raises a general question about the causes and consequences of corruption and governance. As the causes of corruption may be deeply rooted in economic, political, and socio-cultural as well as geographical environment (Rose-Ackerman, 1999; Andvig et al., 2000; Jain, 2001), the first research question that we deal with is what determines the cross-country variation in corruption.

Setting the Scene 5 Apart from being widespread, many authors argue that corruption is persistent. There is some theoretical support for this view (Basu et al., 1992; Tirole, 1996; Mauro, 2004; Mishra, 2006) and some scant empirical evidence as well (Herzfeld and Weiss, 2003; Alesina and Weder, 2002; Damania et al., 2004). However, like other institutions, corruption is a not static phenomenon. Over a subtantial time span, we may observe a change in corruption. Hence, the second research question is whether corruption is really persistent. As corruption is closely linked to governance, the third issue to investigate is the relationship between economic growth and government governance. Authors like Knack and Keefer (1995), Barro (1997), Keefer and Knack (1997), and Chong and Calderon (2000) argue that the impact of governance on growth is positive, but others report the opposite (Quibria, 2006). A deeper look at country cases (Qian, 2003; Pritchett, 2003) tends to support the latter view. China and Vietnam, for example, are still far away from good governance, but their growth rates are remarkable. Under the kleptocratic regime run by Soeharto, Indonesia was one of Asian tigers. Contributing to this debate, we re-examine the relationship between growth and governance. The idea to combat corruption via the promotion of good governance has spread across the globe. At the same time, nations in the world are not isolated from international economic and political interactions. Hence, neighbors may matter in the formation of good governance to result in a spatial dependence among that are close nations. This spatial dependence may be due to spillovers and diffusion-adoption processes, as also found in democracy, war and peace, or economic liberty (O Loughin et al., 1998; Ward and Gleditsch, 2002; Simmons and Elkins, 2004). It also can stem from policy convergence (Mukand and Rodrik, 2005), interdependency of policy decisions (Brueckner, 2003), or transmission of government forms (Starr, 1991). This implies that countries may be geographically clustered according to their quality of governance. At the same time, the effect of the determinants of governance may not be homogenous over space, i.e., the

6 Chapter 1 determinants may have a different impact in different countries. These are the final issues covered in this dissertation. To summarize, the four issues discussed in this dissertation are: (1) what determines corruption?; (2) is corruption persistent?; (3) what is the relationship between governance and growth?; and (4) does space matter in explaining cross-country variation in the quality of governance? The following section summarizes our approach and main findings. 1.2 Main Findings 1.2.1 Determinants of Corruption As shown in Chapter 2, the theoretical literature provides many view points in explaining the existence of corruption. In the absence of a theory-based consensus on what determines corruption, empirical researchers typically experiment with a set of variables that may be correlated with corruption. Others focus on a particular variable of interest, using a set of control variables. If one runs a regression using a particular combination of explanatory variables, it is possible to find that one variable of interest is significant but becomes insignificant when other combinations of explanatory variables are used. The same holds for the sign of the estimate: the impact of a variable may change if different variables are controlled for. Using the Sensitivity Analysis (SA) of Sala-i-Martin (1997), we deal with this issue in Chapter 3. The idea behind this approach is to generate a series of estimates (β s) of particular variables of interest using all possible combinations of control variables drawn from a pool of variables. The pool contains variables argued to be correlated with corruption. Our focus is to examine whether the distribution of β lies on one side of the cumulative distribution function. We employ 45 variables that previously have been claimed to be significant in explaining cross-country variation in corruption. We also experiment with different numbers of observations to examine the robustness of the finding. We find that only two variables having stable coefficients, namely government effectiveness and rule of law.

Setting the Scene 7 To come up with the list of variables, we first survey what researchers report on the determinants of corruption. Updating surveys by Andvig et al. (2000) and Jain (2001), Chapter 2 starts with a discussion of the definition and measurement of corruption. Although there is a widely accepted definition, corruption is difficult to measure. Corruption can be indirectly measured via the perception of respondents in surveys, but there are also some studies trying to directly measure it in money metric terms. Nowadays, perceived-corruption indexes have been widely used. At the same time, these indexes have been criticized. Some scholars criticize such indexes from a technical point of view (Galtung, 2006; Kurtz and Schrank, 2007), others from a more substantive perspective (Olken, 2006; Donchev and Ujhelyi, 2007). Some of these criticisms, however, have limited validity. In addition, this chapter discusses determinants of corruption as found in the literature. While other categorizations are possible, we may group these determinants into four broad classes. First, economic and demographic determinants including economic institutions. Second, variables categorized as political institutions. Third, those falling in the area of judicial and bureaucracy determinants, and finally, geography and culture. 1.2.2 Corruption Persistence Theoretical and empirical research on corruption generally concludes that corruption is persistent. However, using ICRG data for the period 1984-2003, we find strong evidence that corruption changes over time. Many corrupt countries saw their level of corruption decline, while many clean countries became less clean over the same period. We also find that this convergence process is not continuous: there has been an improvement in world corruption in the first part of our sample period, but a worsening in the second half. We start with a simple check on the correlation between the current and past levels of corruption, and find that the correlation shrinks when the time lag increases. We also find evidence on β-convergence when we

8 Chapter 1 regress the change in corruption on its initial value. Examining the trend as indicated by the standard deviation and coefficient of variation of crosscountry corruption over time, we discover σ-convergence. Also, the results from a set of ordered logit and probit regressions give additional support to the conclusion that corruption is not persistence. These finding are confirmed in our further analysis of the dynamics of the distribution of corruption data. There is a significant modality shift in the corruption distribution over time. Using a Gaussian kernel function we detect a transformation in the distribution of corruption from a bimodal to a unimodal distribution. Finally, on the basis of a Markov chain analysis, we also find interclass upward and downward shifts of countries. These issues are discussed in Chapter 4. 1.2.3 Governance and Growth The literature on the governance-growth relationship does not provide clear cut conclusions about the relevance of governance for growth. There are two issues in this debate: the robustness of the relationship and the way governance is measured. We contribute to this debate by introducing our governance index generated using Confirmatory Factor Analysis (CFA) on ICRG governance indicators, namely democratic accountability, government stability, bureaucracy quality, corruption, and law and order. CFA is a latent variable approach that can be used to analyze proxies for theoretical concept of governance. We discover that the five indicators of governance can be combined into one single index with an impressive goodness of fit. Using our CFA-based index of governance, we test the stability of the growth-governance nexus via a set of parsimonious models, recursive regressions, and the Sensitivity Analysis of Sala-i-Martin mentioned previously. We discover that our index is fairly robust in a series of experiments using different control variables as well as different numbers and compositions of observations. In the majority of cases, we find that good governance promotes economic growth significantly. Chapter 5 discusess this issue in more

Setting the Scene 9 details. 1.2.4 Spatial Dimension Researchers widely recognize the role of geography in shaping countries quality of governance. Several proxies have been used for geography including latitude, climate, temperature, country size, climate-related diseases, or dummies indicating that countries are landlocked, islands, or belong to a particular region (Acemoglu et al., 2001, 2002; Easterly and Levine, 2003; Rodrik et al., 2004; Olsson and Hibbs Jr., 2005). These proxies, however, do not capture the spatial dimension of governance, whereas cross-country data are generally characterized by spatial dimension. We consider two spatial dimensions of governance, namely spatial dependence and spatial hetereogeneity. While the former refers to the degree of dependence of a country s governance on that of its neighbors, the latter deals with varying effects of the determinants of governance. Our preliminary analysis indicates that the closer the country s distance to the best (worst) practice of governance, the higher (lower) its level of governance. We also discover a positive global spatial dependence, where poorly (well) governed countries are geographically clustered with poorly (well) governed countries. As a further check we apply a spatial econometric method introduced by Anselin (1988). This approach is superior to other techniques in capturing the behavior of neighbors through a weight matrix. The weight matrix does not only consist of physical-geographical distance, but also political distance where countries regime types are taken into account. We find that governance in one country exhibits a positive relationship with governance in neighboring countries. Finally, we also run Geographically Weighted Regression (GWR) of Brunsdon et al. (1999) that allows us to analyze the issue of spatial nonstationarity defined as the variation in relationships and processes over space (Brunsdon et al., 1999: 497). We discover that the coefficients of

10 Chapter 1 the determinants of governance are not constant, but vary across observations. Chapter 6 discusses the spatial dimension of governance in more detail. Chapter 7 offers the conclusions of this dissertation.

Chapter 2 Determinants of Corruption: A Survey If government were a product, selling it would be illegal. P. J. O Rourke (1993) 2.1 Introduction Corruption is a world-wide phenomenon that is multi-faceted. While it also exists in the private sector, corruption primarily involves government officials. Hence, it is not surprising that corruption is labeled as endemic in all governments (Nye, 1967: 417), where... no region, and hardly any country, has been immune (Glynn et al., 1997: 7). Corruption is probably as old as government itself. Corruption affects almost all parts of society. Like a cancer, as argued by Amundsen (1999: 1), corruption eats into the cultural, political and economic fabric of society, and destroys the functioning of vital organs. The World Bank (WB) has identified corruption as the single greatest obstacle to economic and social development. It undermines development by distorting the rule of law and weakening the institutional foundation Earlier versions of this chapter joint work with Jakob de Haan were presented at the European Public Choice Society meetings in 2005 and 2006

12 Chapter 2 on which economic growth depends. 1 This explains why anti-corruption measures rank high on the policy agenda of the World Bank and the United Nations (2007) as exemplified by its recent Stolen Asset Recovery (StAR) Initiative. Corruption has also attracted the attention of researchers in the academic arena; not only in economics, but also in sociology, political science, law, etc. Research in this area includes detailed descriptions of corruption scandals, case studies, and cross-country studies. It also ranges from theoretical models to empirical investigations. This chapter updates the surveys by Andvig et al. (2000) and Jain (2001) and reviews empirical studies 2 on the causes of corruption. Since corruption has a negative impact on economic outcomes (Mauro, 1995; Tanzi and Davoodi, 1997; Gupta et al., 1998; Lambsdorff, 2001), it is important to know which factors determine corruption. The rest of this chapter is constructed as follows. In section 2.2 we discuss the concept of corruption and its measurement. Section 2.3 explores some empirical issues concerning the determinants of corruption. Section 2.4 concludes. 2.2 Corruption: Definition and Measurement 2.2.1 Definition The Oxford Advanced Learner s Dictionary (2000, p. 281) describes corruption as: (1) dishonest or illegal behaviour, especially of people in authority; (2) the act or effect of making somebody change from moral to immoral standards of behaviour. Here, corruption is linked to two important elements: authority and morality. Authors like Gould (1991: 468) explicitly define corruption as a moral problem, i.e., it is an immoral and unethical phenomenon that contains a set of moral aberrations from moral standards of society, causing loss of respect for and confidence in duly constituted au- 1 www1.worldbank.org/publicsector/anticorrupt/index.cfm. 2 An excellent survey on theoretical studies on corruption is Aidt (2003).

Determinants of Corruption: A Survey 13 thority. This is in line with Dobel (1978: 960) who labels corruption as the moral incapacity to make disinterested moral commitments to actions, symbols, and institutions which benefit the substantive common welfare. These normative definitions, however, are not without problems. Moral norms differ from place to place and change from time to time. For example, whose moral standard should be used, or what is the appropriate moral benchmark if there is more than one standard? In African traditions gift giving is a common practice (de Sardan, 1999), but in Western cultures it is often regarded as corruption (Qizilbash, 2001). Also, what was not regarded in the past as corrupt acts, now may be labeled as corruption, and the other way around. Moreover, viewing corruption merely as a moral problem tends to individualize this social phenomenon and ignores the wider socio-political context of corruption. For corruption to exist, according to Jain (2001), three conditions should be fulfilled: discretionary power, economic rents, and a weak judicial system. 3 Discretionary power relates to authority to design and administer regulations, which, in turn, is accompanied by the presence of extracted rents associated with power. A weak judicial system implies a low probability of detection and lack of sanctions. Jain (p. 73) thus defines corruption as an act in which the power of public office is used for personal gain in a manner that contravenes the rules of the game. It brings us back to Waterbury s (1973: 533) definition which is now widely accepted, i.e., corruption is the abuse of public power and influence for private ends. Under this definition, corruption takes many forms varying from the minor abuse of influence to institutionalized bribery and systematic kleptocracy. 2.2.2 Measurement The literature provides three ways of measuring corruption. First, direct estimates in money metric terms that mainly capture corruption by politicians. For example, the Transparency International Global Corruption Re- 3 In the same vein, Klitgaard (2000) constructs an equation saying that corruption equals monopoly power plus discretion minus accountability.

14 Chapter 2 port 2004 estimates the size of stolen asset by the top 10 world kleptocrats, ranging from 0.07-0.08 billion $US (Josep Estrada, the President of the Philippine, 1998-2001) to 15-35 billion US$ (Mohamed Soeharto, the President of Indonesia, 1967-1998). 4 Second, estimates based on micro level data. Some authors employ a direct approach (e.g., Henderson and Kuncoro, 2004 and Kuncoro, 2004), where the reported bribe payments to public officials is used to calculate the magnitude of corruption at the micro level. Others employ an indirect approach. A good example is the study of Olken (2006) who compares Indonesian villagers beliefs about the likelihood of corruption in a roadbuilding project with a measure of missing expenditures in the project. Third, estimates based on the perception of the likelihood, frequency, or level of corruption of respondents in surveys. There are more than 25 organizations around the world that collect and publish this type of corruption data, either as poll-based data (primary source) or as poll-of-polls-based data (secondary source). A good example of the poll-based data is the International Country Risk Guide (ICRG) data covering almost 150 countries since the beginning of the 1980s making it the largest panel dataset available. 5 The ICRG data are expert-based assessments of political, economic, and financial risks. Corruption is one of the 12 political risk components, with scores ranging between 0-6, where a higher score means less corruption. Corruption is captured via actual or potential corruption in the form of excessive patronage, nepotism, job reservations, favor-for-favors, secret party funding, and suspiciously close ties between politics and business. 6 Another example is the World Economic Forum (WEF) dataset reported in The Global Competitiveness Report. Released since 1979, it initially cov- 4 The list appeared also in the UN-WB report StAR Initiative 2007, but some of the estimates have been actually published in Time Asia (24 May 1999) and The Guardian (26 March 2004). 5 The complete list of primary sources can be found in the reports of Kaufmann et al. (Governance Matters) and Lambsdorff (Framework Document). 6 www.prsgroup.com/icrg Methodology.aspx

Determinants of Corruption: A Survey 15 ered only 16 countries, but over time has been expanded to almost 120 countries. It is based on an annual opinion survey that records the perspectives of business leaders around the world who compare their own operating environment with global standards on a wide range of dimensions. Some of these provide information on the perception of corruption. For instance, one issue refers to favoritism in decisions of government officials when deciding upon policies and contracts. Other dimensions include irregular payment in export and import, government procurement, tax collection, public contracts, and judicial decision as well as business cost of corruption. Each of these variables is scaled between 1-7, where a higher score means less corruption. The third example of primary data is the one reported in the World Competitiveness Yearbook of the Institute for Management Development (WCY-IMD). The WCY has been published since 1987 and was initially a joint initiative with the WEF. The IMD reports corruption on the basis of a survey among thousands domestic and foreign firms operating in about 50 countries. There are various variables that are related to corruption, but the most explicit one is an indicator labeled bribery and corruption exist in public sphere. This indicator has a scale ranging from 0 (highly existing) to 10 (not existing). A well-known example of the poll-of-polls-based data is the Transparency International (TI) index, called Corruption Perception Index (CPI). 7 This data covers about 130 countries. At least three primary surveys or sources should be available for a country to be included in the calculation of CPI. Computed by Lambsdorff on behalf of the TI since 1995, the CPI aggregates various perception-based indicators of corruption to a new index on 0-10 scale index where a higher score means less corruption. The index is constructed via two steps. First, standardization of the primary data is done by a two-sequential procedure: matching percentile and β-transformation. 8 7 www.transparency.org/policy research/surveys indices/cpi/ 8 Before 2002 the standardization was done by a simple mean and standard deviation approach.

16 Chapter 2 The former is to tackle the differences in scaling system among the primary sources as well as in the distribution of data, and also to ensure that the resulting index lies between 0-10. The latter is to handle the tendency of the standard deviation to attenuate over time. In the second step, the final index is computed as the unweighted average of the standardized and transformed data (Lambsdorff, 2002, 2003b, 2004). Due to the differences in the standardization and transformation and the use of different sources over time, the CPI is not a consistent time series. In Lambsdorff s (2000: 4) words, year-to-year changes may not only result from a changing performance of a country... changes can result from the different methodologies... not necessarily from actual changes. 9 The other main source of poll-of-polls corruption data is the World Bank (Kaufmann and Kraay, 2002; Kaufmann, et al., 1999, 2006, and 2007). 10 Corruption is one of the six components of the Kaufmann governance index. 11 Reporting corruption data since 1996, this index covers almost 200 countries and territories and draws upon about 40 data sources produced by more than 30 different organizations. To aggregate the various corruption indicators, Kaufmann et al. (1999, 2002) use a latent variable approach in which the observed perception indicators of corruption are expressed as a linear function of the latent concept of corruption plus a disturbance capturing perception errors and sampling variation in each indicator. The resulting index ranges between 2.5 (most corrupt) and +2.5 (least corrupt). Table 2.1 reports the correlation among the indicators discussed above. Both the Pearson and polychoric correlation coefficients demonstrate the closeness of the corruption indicators. Apart from the TI, the World Bank, and the ICRG indicators, the table also shows seven indicators of corrup- 9 In line with this, since it relies heavily on independently conducted surveys and expert polls, the CPI is not available for a significant number of countries. Sometimes countries are no longer included if the required minimum number of sources is missing. Galtung (2006) therefore concludes that the CPI does not measure trends as it is a defective and misleading benchmark of trends. 10 www.worldbank.org/wbi/governance/wp-governance.html 11 The other components are voice and accountability, political instability and violence, government effectiveness, regulatory quality, and rule of law.

Determinants of Corruption: A Survey 17 tion drawn from the WEF, namely irregular payments in public utilities, public contract, judicial decision (d1-d3), export-import, tax collection (d5- d6), and business cost of corruption (d4), as well as favoritism in policy making (d7). Using these WEF indicators, we also construct an aggregate index of corruption using Principal Component Analysis (PCA). 12 It is clear that the correlation among the WEF indicators is very high, and so is the correlation among the resulting PCA-based WEF index, the Lambsdorff, Kaufmann, and ICRG indicators of corruption. 13 In other words, although the indicators are different, they proxy the same phenomenon: corruption. 2.2.3 Criticisms Some of the critique on these indicators of corruption depart from the definition underlying these indicators. Galtung (2006), for example, argues that the CPI does not explicitly distinguish between corruption in the civil service and political corruption. Likewise, Kurtz and Schrank (2007) criticize that the measures of corruption that combine questions about the presence of nepotism, cronyism, and bribe taking place in government with the intrusiveness of the bureaucracy or the amount of red tape may be problematic. In their words (p. 543), intrusiveness and red tape can be a sign of either effective or ineffective governance, depending on the content of the policies being enforced. In line with this is critique referring to the use of different sources to come up with an indicator of corruption. Different sources may give different definitions of corruption and different forms of corrupt act. Andvig (2005) questions whether the different sources underlying the CPI cover the same phenomenon. Also, Søreide (2003: 7) writes, 12 Other latent variable estimates such as Principal Factor (PF) and Maximum Likelihood (ML) based Factor Analysis used to produce aggregate indexes give very similar results. Each of the three techniques produces only one factor with a high eigenvalue and very high loadings of the underlying indicators. 13 The correlation between the two aggregated indexes (WB and TI) and the three individual indexes are certainly high, because the former indexes contain the later indexes.

18 Chapter 2 Table 2.1: Indicators of Corruption, Correlation and Summary Statistics Corruption WB TI ICRG WEF WEF Indicators IMD Indicators (PCA) WEF1 WEF2 WEF3 WEF4 WEF5 WEF6 WEF7 a. WB 0.99 0.95 0.93 0.86 0.89 0.90 0.94 0.87 0.85 0.84 0.97 b. TI 0.97 0.94 0.95 0.88 0.92 0.91 0.95 0.89 0.86 0.85 0.98 c. ICRG 0.89 0.88 0.87 0.80 0.84 0.83 0.88 0.80 0.77 0.80 0.93 d. WEF 0.93 0.94 0.78 0.91 0.98 0.97 0.96 0.96 0.92 0.88 0.93 d1. WEF1 0.85 0.85 0.70 0.92 0.86 0.85 0.83 0.95 0.89 0.69 0.84 d2. WEF2 0.87 0.90 0.74 0.97 0.88 0.95 0.92 0.92 0.87 0.87 0.91 d3. WEF3 0.88 0.89 0.72 0.95 0.82 0.91 0.92 0.92 0.89 0.82 0.89 d4. WEF4 0.92 0.93 0.80 0.95 0.82 0.89 0.92 0.87 0.82 0.90 0.94 d5. WEF5 0.86 0.88 0.72 0.96 0.91 0.92 0.88 0.87 0.92 0.74 0.85 d6. WEF6 0.85 0.86 0.67 0.94 0.92 0.89 0.87 0.85 0.93 0.68 0.82 d7. WEF7 0.81 0.82 0.71 0.85 0.67 0.83 0.79 0.85 0.74 0.68 0.87 e. IMD 0.97 0.98 0.93 0.93 0.84 0.91 0.89 0.94 0.85 0.82 0.87 Observations 207 168 140 117 117 117 117 117 117 117 117 53 Mean 0.00 4.05 2.50 0.00 5.04 4.06 4.57 4.36 4.66 4.88 3.21 4.63 Std. Dev. 1.00 2.14 1.18 2.47 1.09 1.06 1.25 1.13 1.13 1.19 0.87 2.62 Minimum -1.68 1.80 0.00-4.69 2.10 2.10 2.40 2.20 2.20 2.20 1.60 0.51 Maximum 2.49 9.65 6.00 4.96 6.80 6.40 6.80 6.80 6.70 6.80 5.40 9.38 Correlation is based on the 2005-06 data. The lower (upper) diagonal is Pearson (polychoric) correlation. PCA is based on indicators d1-d7. WEF1-WEF3: irregular payments in public utilities, public contract, and judicial decision; WEF4: business cost of corruption; WEF5-WEF6: irregular payments in export-import and tax collection; WEF7: favoritism in policy making.

Determinants of Corruption: A Survey 19 Most of the polls and surveys ask for a general impression of the magnitude of the problem, which actually means people s subjective intuitions of the extent of a hidden activity. For the TI index, only one source asks for people s personal experiences with corruption. The quantification of the problem is highly ambiguous. It is not clear to what extent the level of corruption reflects the frequency of corrupt acts, the severity to society, the size of the bribes or the benefits obtained. Most of the surveys do not specify what they mean by the word corruption. It can thus be quite difficult for the respondents to answer when asked about a quantification of the misuse of public office for private or political party gain or when encouraged to rate the severity of corruption within the state. These criticisms, however, have limited validity. As indicated in Table 2.1, various measures of corruption are highly correlated, suggesting that they measure the same phenomenon. Another criticism points to the design of surveys, e.g., surveys designed to capture perception may have a bias that affects the respondents interpretation of the questions in the surveys. As Bertrand and Mullainathan (2001) argue, the sequence of questions may substantially affect the respondents as they tend to answer questions in line with their answers to previous questions. Also, the prior questions may bring out certain memories which influence the answers. Other technical problems raised by Bertrand and Mullainathan include the time spent by the respondents to scrutinize each question, scoring effects, and other cognitive issues. A more substantive criticism is indicated by Olken (2006) and Donchev and Ujhelyi (2007). They argue that what is believed by the respondents may not reflect what actually happens. Fisman and Miguel (2006), however, provide some evidence that beliefs about corruption and actual corruption are closely related. They study the parking behaviour of UN diplomats who have immunity from parking enforcement actions and compare the data with perceived corruption. They find that diplomats from corrupt countries (based on existing perception-based indexes) accumulate significantly more unpaid parking violations demonstrating the abuse of power. The choice of respondents is also criticized as it may potentially influence the responses, i.e., different participants may give different opinions about a country under review. Kurtz and Schrank (2007) argue that surveys on

20 Chapter 2 corruption may reflect the narrow interests of respondents, say, business communities. A potential bias in sample selection may occur due to the exclusion of respondents who did not succeed in the marketplace, or... who were deterred from entering local markets by pervasive malgovernance or corruption itself. As a consequence, the sample is dominated by unrepresentative respondents, while those who show up in the surveys... are the beneficiaries of corruption and cronyism... therefore unlikely to report it accurately. Moreover, malgovernance is effectively reported... because it is not pervasive enough to create sufficiently strong distortions in firm-level survival or investor behaviour to induce selection bias. And thus in such contexts those who do not win from malfeasance can survive to report it! (p. 543). This criticism may be true for individual surveys, but less so for surveys aggregating such individual surveys. Kaufmann, et al. (2007), for example, use a mixture of sources of data coming from various agents: four crosscountry surveys of firms, seven commercial risk-rating agencies, three crosscountry surveys of individuals, six sets of ratings produced by government and multilateral organizations, and 11 data sources produced by a wide range of non-governmental organizations. Despite all these criticisms, some of which appear to be more valid than others, perception-based corruption indicators have created the possibility to study corruption. As a result, numerous studies have employed such indexes. In the following, we review studies on the determinants of corruption. 2.3 Determinants of Corruption 2.3.1 Four Classes We survey cross-country studies on the determinants of corruption published over the last 10 years. All studies that we are aware of are summarized in Appendix 1. While other categorizations are possible, we distinguish four broad classes of the underlying causes of corruption, namely (1) economic

Determinants of Corruption: A Survey 21 and demographic determinants, (2) political institutions, (3) judicial and bureaucracy environment, and (4) geography and culture. Economic and Demographic Determinants There is a wide range of economic variables that have been suggested to cause corruption. Income is a commonly-used variable to explain corruption (Damania et al., 2004; Persson et al., 2003; and van Rijckeghem and Weder, 1997; among others). Corruption can be seen as an inferior good, where the demand falls as income rises. Also, along with an increase in income, more resources are available to combat corruption. Mostly proxied by GDP per capita, income is also used to control for structural differences across countries. It is generally found that income has a negative and significant effect on corruption, even though Kaufmann et al. (1999) and Hall and Jones (1999) question the causal relationship between corruption and income. Two studies using panel data (Braun and Di Tella, 2004; Fréchette, 2006) deviate from this commonly-found result, as they report that higher income increases corruption, especially when country fixed effects are considered. Braun and Di Tella (p. 93) argue that this is due to the pro-cyclical nature of corruption, where moral standards are lowered during booms, as greed becomes the dominant force for economic decisions. Income distribution is also argued to affect corruption. As Paldam (2002: 224) puts it, A skew income distribution may increase the temptation to make illicit gains. Proxied by the Gini coefficient, he claims that income disparity significantly increases corruption. However, using the income share of the top 20% population, Park (2003) does not find a statistically significant relationship. Similarly, Brown et al. (2006) find no evidence that a greater income inequality increases corruption. The size of government is also put forward as a determinant of corruption, but the causality may run in two directions. If countries exploit economies of scale in the provision of public services thus having a low ratio of public services per capita those who demand the services might