Corruption at the Grassroots-level – Between Temptation, Norms, and Culture: Themenheft Jahrbücher für Nationalökonomie und Statistik 2/2015 9783110511628, 9783828206144

A Very Short History of Corruption Research and a List of What We Should Aim For

177 35 14MB

English Pages 144 [148] Year 2015

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Inhalt / Contents
Guest Editorial: Special Issue on Corruption at the Grassroots-level What Can We Know About Corruption?
Abhandlungen/Original Papers
Corruption and Productivity: Firm-level Evidence
Bribing Behaviour and Sample Selection: Evidence from Post-Socialist Countries and Western Europe
Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China
Religious Loyalty and Acceptance of Corruption
Group Decision Making in a Corruption Experiment: China and Germany Compared
Mitigating Extortive Corruption? Experimental Evidence
Recommend Papers

Corruption at the Grassroots-level – Between Temptation, Norms, and Culture: Themenheft Jahrbücher für Nationalökonomie und Statistik 2/2015
 9783110511628, 9783828206144

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Corruption at the Grassroots-level Between Temptation, Norms, and Culture

Edited by Johann Graf Lambsdorff and Günther G. Schulze

With Contributions by Bühren, Christoph, University of Kassel De Rosa, Donato, The World Bank, Washington DC, USA Frank, Björn, University of Kassel Gooroochurn, Nishaal, London Metropolitan University, UK Görg, Holger, Kiel Institute for the World Economy and University of Kiel, Germany; IZA Gouda, Moamen, Graduate School of International and Area Studies, Hankuk University of Foreign Studies, Seoul Hinks, Timothy, University of the West of England, Bristol, UK Ivlevs, Artjoms, University of the West of England, Bristol, UK Khachatryan, Elina, University of Kassel

Lucius & Lucius · Stuttgart 2 0 1 5

Kube, Sebastian, University of Bonn Lambsdorff, Johann Graf, University of Passau Li, Sha, University of Kassel Liu, Qijun, School of Public Administration, Huazhong University of Science and Technology, China Park, Sang-Min, Technische Hochschule Mittelhessen - University of Applied Sciences Peng, Yaping, School of Public Administration, Huazhong University of Science and Technology, China Schulze, Günther G., University of Freiburg and Australian National University Qin, Haiying, Nankai University Vollan, Björn, University of Innsbruck

Guest Editors Prof. Dr. Johann Graf Lambsdorff Universität Passau Lehrstuhl für Volkswirtschaftslehre mit Schwerpunkt Wirtschaftstheorie Raum WIWI 108 Innstraße 2 7 94032 Passau [email protected] Prof. Dr. Günther G. Schulze Albert-Ludwigs-Universität Freiburg im Breisgau Institute for Economic Research Department of International Economic Policy Room 2512 Platz der Alten Synagoge 1 79085 Freiburg [email protected]

Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http://dnb.d-nb.de abrufbar ISBN 978-3-8282-0614-4

© Lucius &C Lucius Verlagsgesellschaft m b H · Stuttgart - 2 0 1 5 Gerokstraße 51, D-70184 Stuttgart Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlags unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Ubersetzungen und Mikroverfilmungen sowie die Einspeicherung und Verarbeitung in elektronischen Systemen.

Satz: inMEDIAlo, Digital- und Printmedien UG, Plankstadt Druck und Bindung: Neumann Druck, Heidelberg Printed in Germany

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2015) Bd. (Vol.) 235/2

Inhalt / Contents Guest Editorial: Lambsdorff, Johann Graf, Günther G. Schulze, What Can We Know About Corruption? A Very Short History of Corruption Research and a List of What We Should Aim For

100-114

Abhandlungen/Original Papers De Rosa, Donato, Nishaal Gooroochurn, Holger Görg, Corruption and Productivity: Firm-level Evidence Ivlevs, Artjoms, Timothy Hinks, Bribing Behaviour and Sample Selection: Evidence from Post-Socialist Countries and Western Europe Liu, Qijun, Yaping Peng, Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China Gouda, Moamen, Sang-Min Park, Religious Loyalty and Acceptance of Corruption Li, Sha, Christoph Bühren, Björn Frank, Haiying Qin, Group Decision Making in a Corruption Experiment: China and Germany Compared Khachatryan, Elina, Sebastian Kube, Björn Vollan, Mitigating Extortive Corruption? Experimental Evidence

115-138

139-167 168-183 184-206

207-227 228-241

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2015) Bd. (Vol.) 235/2

Guest Editorial: Special Issue on Corruption at the Grassroots-level

What Can We Know About Corruption? A Very Short History of Corruption Research and a List of What We Should Aim For Johann Graf Lambsdorff University of Passau

Günther G. Schulze University of Freiburg, Australian National University JEL D73; C9

1

Introduction

Corruption research has centered on three fundamental questions: What determines corruption? What effects does corruption have? And finally, what can we do to reduce corruption effectively and efficiently? Corruption has been, and continues to be researched so intensively that it is hard to keep abreast of all the latest developments in the various subfields of corruption research. Recent surveys of the vast literature include Aidt (2003), Svensson (2005), Lambsdorff (2007), Pande (2007), Olken and Pande (2011), and KisKatos and Schulze (2013). Instead of adding yet another survey, we focus on the main challenges of empirical corruption research and show by example how the literature has tried to address these challenges. 1 The first and foremost challenge of empirical corruption research is that corruption is not directly observable, at least not in any systematic fashion as it is illegal and therefore clandestine. Thus, proxies for actual corruption levels need to be found or controlled situations need to be created in which corruption is observable. Proxies frequently used are corruption perceptions, stated corruption experiences, audit results of public expenditure tracking analyses or quantitative service delivery surveys, corruption incidents registered with law enforcement agencies, and corruption convictions. Controlled situations include laboratory and field experiments. 2 Moreover, there are some natural experiments as well.

1

2

We do not discuss the usual challenges of empirical research such as the search for a convincing strategy to identify causal relationships or sample representativeness but focus on those that are particular for corruption research. Closely related is the literature on politically connected firms that looks into the value of these political connections for the firms and their differential behavior (Faccio 2006) and on excessive spending of governments on themselves rather than on people they govern (Sjahrir et al. 2014). Both may not be illegal in the strict sense but fit the common definition of corruption as 'misuse of public office for private gain'.

What Can We Know About Corruption? · 101

Second, corruption is highly context-specific as it depends on the institutional setup, stage of development, but also n o r m s and culture (inter alia Fisman/Miguel 2 0 0 7 ; KisKatos/Schulze 2014). This begs the question to w h a t extent results derived in one setting carry over to other settings as well. External validity is a m a j o r concern not only for experiments, both laboratory and real-world, but also for country-specific econometric studies. Third, the type of corruption for which sufficiently good proxies exist may not be the most harmful. For instance, corruption in the allocation of driver's licenses (Bertrand et al. 2007) or in issuing identity cards (Kaiser et al. 2006) may be h a r m f u l , but corruption between the political elite and the business community in f o r m of cronyism, preferential access to government contracts or freedom f r o m prosecution may be fundamentally more important and much harder to measure. The former may be proxied sufficiently well by surveys of users of public services, whereas the latter, often referred to as grand corruption, is extremely hard to quantify in all of its consequences. Corruption research in economics has a long history, seminal early articles include RoseAckerman (1975) and Shleifer and Vishny (1993). W e identify t w o generations of corruption research. The first generation has focused on cross-country analyses, mostly using corruption perception indices as a proxy for real levels of corruption. The second generation has taken a micro-perspective using microeconometrics and experiments to analyze corruption at the individual, household or firm level. We discuss the generations with the help of one or t w o examples. The first generation included also grand corruption. Data on perceived levels of corruption include assessments by senior business leaders, respondents w h o are capable of judging corruption at the elite level. The second generation has advanced our knowledge on the causal processes that underlie corrupt transactions. But it tends to focus more on petty corruption because this is easier to measure. This is just one of many trade-offs that we discuss subsequently. 2

Macro studies

M a c r o studies have used the variation of corruption levels across countries in cross-section or panel analyses. O n e strand of literature has focused on the consequences of corruption pioneered by M a u r o ' s (1995) article on the effect of corruption on growth and investment. This has initiated a wave of subsequent studies that largely corroborated M a u r o ' s findings; i.e. a negative impact of corruption on levels of investments (measured as a ratio to GDP), which is the most important causal link to the impact of corruption on the g r o w t h of G D P (Lambsdorff 2007: 7 3 - 7 5 ; 1 0 0 - 1 0 2 ) . A variety of further dismal effects of corruption have been found, for example relating to increased inequality of income or higher military spending. A second group of studies has focused on the causes of corruption. O n e representative example is Treisman (2000) w h o tests twelve hypotheses on the determinants of perceived corruption levels. H e runs on a sample of 34 to 64 countries four sets of cross-sectional regressions with Transparency international's Corruption Perception Indexes for the years 1996, 1997, 1998 and Business International's index for the early 1980s as endogenous variables. Explanatory variables comprise controls for the legal system, colonial status, percent age of Protestants, ethnolinguistic division, exports of fuel, minerals and metals, log G D P per capita, and federal state organization as well as democratic history, import penetration, state intervention, government salary levels and government turnover. Treisman finds, a m o n g other things, that countries with a higher share of Protestants,

102 • Johann Graf Lambsdorff and Günther G. Schulze

histories of British rule and with a democratic history and higher import penetration are less corrupt, federal states more corrupt. The challenge with cross-country studies, as it is often the case with macro data, is that causation often remains suggestive, even if statistics are carried out with utmost care. Two-way causation exists between corruption and many socioeconomic variables. For instance, corruption is likely to inhibit economic development and to increase poverty. Yet, corruption may also result from poverty partly because resources for anticorruption are scarce. More importantly, poverty and corruption may also be the simultaneous consequences of persistently bad institutions brought about by the colonial rulers ( Acemoglu et al. 2001) - they are both endogenous to the same cause. In order to identify causation, appropriate instruments have to be found for the endogenous regressors. This often proves difficult, especially if the endogenous regressors are time-variant. What might serve as an instrument for corruption that affects poverty only through corruption? Or, to identify the reverse causality, which variable might serve as an instrument for poverty without affecting corruption also through other channels? There have been two strategies to dealing with these problems, none of them reaching perfection. The first is a pragmatic one: Do not investigate relationships where two-way causation is a major problem. This would for example be true of the relationship between poverty and corruption. Instead of embarking on the mission impossible of finding instruments for one of these variables, one would rather look for relationships where two-way causation is less of a problem. Rather than testing for poverty, one might investigate the impact of corruption on economic growth if researchers consider reverse causation in this case to be less likely. Rather than seeking the "right" statistical model, such a pragmatic approach assembles supportive statistical evidence. Testing for a varying battery of control variables allows researchers to assess the robustness of their findings. Any such approach should end with a discussion of potential omitted variables and derive plausible arguments on their relevance. 3 This might be complemented by suggesting instruments and testing their properties. But one needs to remain pragmatic even here - the exogeneity of instruments is nowhere iron-clad. The strength of the findings thus hinges on theoretical arguments that such a bias is implausible. In some cases one might need to be modest and settle for interesting correlations rather than for causation. A second approach has been to exploit the dynamics that goes along with causation. A cause commonly precedes its effect, such that observations on the time-varying properties of certain variables may help identify causation. A simple example would be to use lagged data, stating the belief that these are exogenous and cannot be caused by more recent developments of the endogenous variable. This would allow for identifying within-country variation over time, for example by help of fixed effects panel regression. Yet, we still know rather little about the dynamics of corruption. It might well be that levels of corruption are rather persistent, remaining almost unaltered for decades. In this case, they are highly correlated over time, such that lagged data may not be exogenous. The same might be true of some fundamental causes of corruption, such as limits to trade, low levels of openness, abundance of natural resources, or some cultural variables. If these

3

An inherent problem of cross-country analyses such as Treisman (2000) are unobservable or immeasurable variables influencing corruption levels such as culture or a permissive attitude towards corruption leading to unobserved heterogeneity. If these variables are correlated with included control variables an omitted variable bias will result.

What Can We Know About Corruption? · 103

variables change little over time, focusing on within-country variations may not lead to statistically significant results. 4 The measure of corruption most often used in cross-country analyses is not beyond criticism. Corruption perceptions have three distinct advantages: They are easy to compile and readily available, different perception measures are in many instances highly correlated indicating that they measure "the same thing" and lastly perceived corruption (not necessarily actual corruption) is relevant for a number of economic decisions, in particular investment decisions. On the downside, perception-based measures may be prone to perception biases or interviewer biases, leading to a low correlation of perceptions and experiences of corruptions (Mocan 2 0 0 8 ; Donchev/Ujhelyi 2013). Corruption experiences are more accurately measurable than corruption perceptions as they do not require a common yardstick of what is a high, medium or low level of corruption. Also, they are not prone to perception biases. Still, measures based on experienced corruption are not beyond criticism. Experience is easily measured in street-level situations, where the general public is subject to harassment by low-level bureaucrats, such as for avoiding speeding tickets or getting access to school and hospitals. But these levels of experienced corruption may not well capture the cases of grand corruption. This brings us back to our third challenge. Perceptions data might be measured with less precision as compared to data on experienced corruption, but they might relate to areas where corruption is more harmful. 3

Microstudies

Cross-country (panel) analyses have provided valuable insights; their inherent drawbacks described above have led to a second generation of corruption studies that have complemented previous evidence. Their unit of observation is the firm, the household or the individual, mostly at the subnational or regional level. This reduces unobserved heterogeneity because the relevant institutional setup, culture, history and other factors that cannot be controlled for are rather similar. As a consequence the likelihood and the extent of a potential omitted variable bias is much smaller; at the same time external validity becomes a concern. To what extent are the results generalizable to other contexts? The microstudies differ with respect to the methodological approach taken to measure corruption. They include econometric analyses of household and firm surveys on corruption experiences, public expenditure tracking, and quantitative service delivery surveys that measure financial leakage and lacking supply. Experimental evidence measures the extent of corruption in a controlled environment in the lab or in the field. 3.1

Econometric evidence I: household and firm surveys, expenditure tracking

Household or firm surveys ask respondents for their experiences with corruption or for what they consider a "typical level of corruption" in their line of business (Reinikka/ Svensson 2006). This approach reduces perception biases as people refer to own experiences, yet it does not eliminate answering biases as corruption is illegal and in most settings

4

A good example for this is Van Rijckeghem and Weder ( 2 0 0 1 ) who show that corruption is significantly negatively related to bureaucrats' relative pay - a result that is highly significant in pooled OLS cross-country regressions, but not significant at the usual levels in fixed effects panel regressions, even though it is very plausible and has been corroborated in other contexts (e.g. Schulze et al. 2 0 1 3 ) .

104 · Johann Graf Lambsdorff and Günther G. Schulze

immoral and thus respondents are hesitant to implicate themselves. Consequently many respondents are reticent (e.g. Clausen et al. 2 0 1 0 ) . Examples for this approach are Rand and Tarp ( 2 0 1 2 ) , Henderson and Kuncoro ( 2 0 1 1 ) , and McCulloch et al. ( 2 0 1 0 ) . 5 These surveys mostly capture only frequent forms of corruption that many people experience, which may not be the most important ones. For instance corruption at the level of government, such as rigged public tenders in government procurement, may be experienced only by a few firms that may be very reluctant to make this public. Public expenditure tracking surveys seek to evaluate the leakage of funds that occurred from the source of the funds, i.e. the central government or a donor organization, to its final destination such as schools or hospitals. This information is either based on special surveys or audit reports. Examples include Reinikka and Svensson ( 2 0 0 4 ) for the former and Ferrez and Finan ( 2 0 0 8 ) for the latter. If audits are not random - as they are in the case of Ferrez and Finan - , inference will be made on a biased sample. Yet, from an auditing agency's perspective profiling rather than random sampling may be the best strategy. Quantitative public service delivery surveys are particularly suited to measure corruption that seeks non-monetary favor as in the case of teacher absenteeism (Kremer et al. 2 0 0 5 ) . 6 3.2

Econometric evidence II: All the rest

A relatively new and still small strand of literature has looked at law enforcement data as a measure for corruption (Glaeser/Saks 2 0 0 6 and Alt/Lassen 2 0 1 4 for the USA, and Schulze et al. 2 0 1 3 for Russia). This is arguably a more objective measure of corruption, if law enforcement and institutional setup are the same across all units of observation. Thus this measure is useable only in subnational studies and only if there is no regional variation in the intensity of law enforcement or if a possible difference can be appropriately controlled for. For instance Glaeser and Saks use only FBI data, as state and local police may have different zeal, resources, mode of operation and legal environment. Alt and Lassen ( 2 0 1 4 ) and Schulze et al. ( 2 0 1 3 ) control for effectiveness and resource endowment of law enforcement agencies. Another approach was employed by Escresa and Picci ( 2 0 1 4 ) by focusing on crossborder corruption. They process 7 3 4 court cases on cross-border corruption ( 2 9 8 from the USA, 8 4 from Germany, 5 5 from UK, 4 9 from France ...), embracing convictions and ongoing cases between 1 9 9 8 and 2 0 1 2 . They observe the frequencies of countries where the alleged transgression took place (74 in China, 4 0 in Nigeria and India, 2 2 in Kazakhstan, 2 1 in Brazil ...) and convert these figures into per-capita numbers. The resulting measure well represents the probability that a randomly chosen person in a country is investigated for cross-border corruption. The analysis builds on the assumption that a court in the USA or in Germany will apply identical standards of judgment and the respective law enforcement agencies investigate with homogeneous intensity, irrespective of whether corruption relates to business in Canada, China or Nigeria. The resulting measure may portray a country's overall level of corruption relatively well. Escresa and Picci's findings are interesting in so far as they largely replicate the T I Corruption Perceptions Index, indicating that perceptions data may in some contexts be a good proxy for real levels of corruption.

5

There are also cross-country studies using surveys on corruption experiences, see for instance M o c a n

6

Teacher absenteeism can also be analyzed by randomized control trials, cf. Duflo et al. ( 2 0 1 2 )

(2008).

What Can We Know About Corruption? · 105

3.3

Experimental evidence from the lab

There are three m a j o r , closely related advantages of laboratory experiments as compared to other methods for research. First, the experimentor is by definition in control of the experiment, which allows for identification of causation. If findings differ significantly across treatments, they must be related to the differences that the experimenter implemented f r o m one treatment to another. Second, incentives can be manipulated easily. Experiments can be designed in response to research questions. This is in contrast to macro studies and many field studies where the important data are unavailable, unretrievable, or t o o expensive to be obtained and where crucial questions must remain underresearched. A third advantage is that the laboratory allows for a closer look at the psychological determinants of corrupt behavior. W h a t do people in a corrupt environment expect f r o m each other, h o w guilty d o they feel, and h o w might such feelings be affected by the beliefs they hold vis-a-vis each other? This leeway in addressing important questions and the close scrutiny of h u m a n beliefs and incentives have led to a wave of experiments related to corruption. D o penalties impact behavior in ways that are predicted by theory? H o w d o subjects value monetary versus non-monetary goals, such as donating to a charity or delivering decent work? In h o w far does trust a m o n g criminal actors p r o m o t e corruption? Should the legal code provide leniency to those w h o confess their infractions? D o subjects feel responsible for their infractions and h o w might responsibility be affected by elections or monitoring? H o w might perpetrators justify their behavior or bias their self-image? Is transparency always helpful or might it backfire? These are just some of the many questions where substantial evidence has been collected recently (Lambsdorff 2012). T o illustrate this, m a n y approaches have been m a d e on whether w o m e n are less corruptible. Cross-country data revealed that corruption is less pronounced in countries with a high percentage of w o m e n in parliament and in the labor force. But causation is difficult to address. Evidence f r o m the lab has revealed that w o m e n tend to be more pro-social and more risk-averse (Chaudhuri 2012). But whether and if so h o w this finding may carry over to issues of corruption remained to be clarified. Early evidence was provided by Frank and Schulze (2000) and Schulze and Frank (2003), w h o show that w o m e n are no less corrupt in non-risky situations, but reduce corruption more than men if risk of detection and punishment is introduced. Lambsdorff and Frank (2011) and Frank and Lambsdorff (2010) find that men reciprocate a bribe more often than w o m e n , while w o m e n tend to cheat the briber, taking money without giving the advantage. This suggests that men have a higher sense for positive reciprocity. M e n are also seen to be more willing to engage in negative reciprocity. Bribers were given the option to exercise costly punishment. This option was exercised more often by cheated men. Similar results are reported by Rivas (2013) w h o runs a more complicated game across many periods. This finding has revealed h o w gender causes corruption, while at the same highlighting the deeper motivations that underlie this link. Another interesting finding relates to delegation. Bribery often involves intermediaries w h o have the criminal expertise and resources to carry out the dirty w o r k . Bribers may not only seek the expertise but alleviate the moral burden of their transgression. H a m m a n et al (2010) show for dictator games that contributions to recipients decrease almost to zero when dictators choose between competitive agents w h o announce u p f r o n t h o w much of the dictator's money they would transfer to the recipient. Acting through the intermediary allows dictators to distance themselves f r o m the n o r m of fairness. Consequently,

106 · Johann Graf Lambsdorff and Günther G. Schulze

they express little responsibility for the recipients' payoffs when having made use of intermediaries. This insight has been applied to corruption by Drugov et al. (2014), w h o find that officials expressed a higher willingness to take bribes from intermediaries and accepted lower bribes. Clients more frequently offered bribes when this was arranged by intermediaries. Intermediaries may thus enhance corruption by reducing the moral costs of bribery. Laboratory evidence has been instrumental in getting deeper insights into the motivations that underlie corrupt transactions and getting clearer guidance for reform. But the external validity of these findings continue to be the biggest challenge. This is already true for experiments on standard human behavior (Levitt/List 2007a,b) and holds even more for those related to corruption. Data are derived mostly from a pool of undergraduate students in an artificial environment where incentives are designed that are only loosely linked to the corrupt incentives one may find in reality. In particular, the moral implications of a lab situation may be significantly different from a real world situation as are the consequences of being detected and punished. Researchers thus often have a hard time stating that their findings from the laboratory may also hold in real-world circumstances. 3.4

Experimental evidence from the field

External validity appears to be stronger for field experiments as they can involve actors and environments that are the direct targets of reform. They can robustly reveal actual behavior- for instance they may show whether a reform method works in a specific environment. For the latter two important requirements have to be met: First, data must be obtained from an environment where the reform was implemented and compared to a control treatment with no such reform or a different type of reform. Second, treatment group and control group should not be endogenously determined. This would be the case if the reform was implemented in areas where it was needed more or believed to be more successful. One example for a field experiment relates to driver's licenses in India (Bertrand et al. 2007). The authors investigate the effect of bonus payments on the behavior of Indian applicants who wanted to obtain a driver's license. Individuals were randomly assigned to three different groups. One group was given free driving lessons, another a bonus payment for obtaining the license within 32 days and a control group where none of these measures were implemented. Data were obtained on whether the applicants were successful in obtaining the driving license, how many days this required, whether they participated in the official test, whether they engaged an agent to facilitate the application (which may involve bribery) and how they performed in an independent test of their driving qualifications. Those who were given a bonus were less qualified in driving, less often participated in the official test and more often engaged local agents to arrange things. The willingness to pay bribes is thus related to need (due to lack of driving capabilities) and advanced further by monetary incentives. The study highlights how important it is to randomly assign individuals to the groups to avoid that they might sort themselves into their preferred group. Differences between groups would no longer be caused by differences in treatments but by those of the underlying self-selecting sample. Another method for arriving at valid comparisons is when nature randomly selects individuals into treatments or when nature changed the environment such that differences can be observed over time. One example of such a natural experiment is Olken and Barron (2009), who study extortions payments paid by truck drivers along two main roads

What Can We Know About Corruption? · 107

in Aceh, an Indonesian province where separatist guerrillas had long been active. Over nine months in 2005 and 2006, data gatherers accompanied the truckers on 304 trips to and from Aceh, recording more than 6,000 illegal pay-offs at military roadblocks, police checkpoints, and weigh stations. This allowed some tests on the validity of economic theory. First, as the trucks neared their destination checkpoint officials demanded increasingly larger sums. This shows that drivers found themselves with a progressively stronger incentive to avoid hassle and safeguard their cargo, which gave the extortionists greater power over them. Midway through the study, after the Indonesian government had signed a cease-fire with the rebels, a phased withdrawal of 30,000 troops began, leading to a fall in the number of checkpoints. The amount lost to extortion decreased, but only by 36%. Fewer stops meant less frequent extortion, but this was offset by a rise in the amounts demanded at the remaining checkpoints, whose operators captured part of the newly liberated surplus. This reveals the economic logic that corrupt officials behave like monopolists, setting their prices so as to maximize their own revenue, without considering the response of the fellow at the next checkpoint, or whether their activities would deter truckers. Another example of a natural experiment is reported by Vicente (2010). He shows that the announcement of oil discoveries in Sao Tome and Principe in the late 1990s led to an increase in corruption as compared to Cape Verde, the control economy that had previously experienced a similar decline in corruption and that had a very similar colonial past and economic performance in the present. Households in Sào Tomé and Principe reported increased perceived levels of vote buying and corruption in customs.

3.5

The value of political connections

Closely linked to field experiments on bribery and corruption is the analysis of the value of political connections. Such studies use cross-section and panel econometrics as well as event studies. This literature investigates whether firms that have politicians or their relatives on the board or as owners perform differently. Since the seminal paper by Faccio (2006) a substantial literature has developed. The measurement of the effect of political connections may indicate corrupt behavior if, for instance, politically connected firms have preferential access to government contracts or finance by state-owned banks (Khwaja/Mian 2005). Yet, such correlations do not necessarily prove the existence of corrupt behavior; in particular, they do not prove causality: If politically connected firms were observed to be more productive, politicians could join the boards of more productive firms or firms could become more productive because politicians joined their boards. Event studies can explain changes in firm values by unexpected, exogenous political events. Such events lead to an exogenous change in the value of political connections and cause a change in the firm value (e.g. Fisman 2001 and Ferguson/Voth 2008). Such events are rare; moreover to establish the relevant political connections is far from straightforward. The advantage of this literature is that it captures indirectly a particular type of grand corruption for which in most cases sufficient data are unavailable. The abuse of political office for private gain can be brought about by direct payments such as kickbacks in public procurement (which is unobservable in large numbers) or through various channels that increase the firm value of the firms that are politically connected (and may have paid for the political connection one way or the other).

108 · Johann Graf Lambsdorff and Günther G. Schulze

4

Research desiderata: What is next?

Laboratory experiments provide experimenters with the highest level of control. Variations across treatments are in the hand of the experimenter. Data is collected not only on behavior, but also on attitudes, beliefs, and perceptions of the respective environment, for example the salience of social norms. Studies allow a deep understanding of human behavior and how it relates to beliefs about the (potentially corrupt) context. This depth of control and understanding will continue to make laboratory experiments a workhorse for research. But the contexts tend to be rather artificial and the subject pools are often not the ones that might be relevant for reform. External validity thus tends to be the biggest challenge. What is more, until now there has not been an idea of how to bring grand corruption to the lab. Field studies have the advantage of being more reliable with respect to external validity, but they are arduous to implement. Often they are implemented as part of a given government project (for example related to infrastructure in Indonesia, Olken 2 0 0 7 ) . Experimenters that are well connected to the authorities in charge of a project are in need of some control, for example allowing them to randomly assign recipients to treatment and control groups. But the respective government project may qualify primarily for reducing poverty, less for understanding anticorruption and advancing research. This may imply that field studies are the result of random opportunities rather than responding to identifiable gaps in research and the interests of the academic community. Likewise, natural experiments depend on a manipulation that was chosen by nature, not an experimenter. While this can advance our knowledge considerably, nature does not manipulate for the purpose of best helping research. Our first research desideratum is thus that laboratory and field experiments should not be seen as substitutes, but as complements. Each addresses questions that the other will find difficult to tackle. The ample evidence on the determinants of corruption has shown that corruption is context- and culture-specific and that results do not easily carry over to contexts other than those in which they have been derived. This holds true also for the effectiveness of anticorruption strategies. While information campaigns may work very well for government transfers in Uganda (Reinikka/Svensson 2 0 0 5 ) central auditing may be the measure of choice in community-based self-managed projects in Indonesia (Olken 2007). What makes a certain anti-corruption strategy successful in one context but not in another? The policy-relevance of this topic is obvious, yet the question has not been sufficiently addressed. With our second research desiderata we thus reemphasize the first one. Only joint efforts that involve a variety of methods can advance our understanding of corruption and reform and generate findings that are valid for a given context and culture. Field studies and laboratory studies must complement each other. The lab entails the advantage of allowing for replication. Conditions for running an experiment can be held (largely) comparable in different countries while operating only with different samples (see for example Banuri/Eckel 2 0 1 2 ; Armantier/Boly 2 0 1 3 ; Alatas et al. 2 0 0 9 ; Cameron et al. 2009). But laboratory experiments suffer from the uncertainty whether the behavior of laboratory subjects comes anywhere close to potential corrupters, officials or politicians. They should thus inspire field studies to heal this caveat and investigate behavior of those who are the target of reform. Field studies might not lend themselves easily to replication. Identical projects in a different country might not be available for experimentation. Results from the field should thus inspire experiments in the lab for detecting differences across contexts and cultures. Furthermore, possible differences between lab

What Can We Know About Corruption? • 109

and field data - if they existed - would be informative on the type of behavioral bias that the lab situation creates, which would inform future lab experiments. Ideal empirical evidence that identified context-specificity of corruption determinants and consequences would entail studies that were identical save for identifiable dimensions such as culture, regime type, religion etc. There are cases in which similar field and econometric data are available across countries, at least in principle: for instance standardized enterprise surveys, surveys of public services, or opinion surveys. These data could be used to identify country differences and their possible determinants, which might then feed into the design of cross-country lab experiments. But, even if empirical cross-country evidence is not ideal (as is often the case) new evidence on countries that have been largely neglected is desirable as it allows contrasting existing empirical evidence with evidence derived in different settings. Prime candidates are (semi) authoritarian regimes or regimes in transition such as China or Russia as reliable data are only becoming publicly available. Context specificity could also be analyzed in a national context, analyzing sector differences or subnational regional variation using the same data type (e.g. Glaeser/Saks 2006; Alt/Lasson 2014; Schulze et al. 2013 for law enforcement data). Such approaches have the distinct advantage that a number of determinants are almost identical across the units of observation such as legal environment, law enforcement, culture, tradition so that the focus can be placed on the remaining differences. 7 The third area which we consider especially fruitful for future research is methodology. Clandestine and illegal in nature corruption is typically not directly observable. As a consequence data are notoriously inaccurate and selection issues are abundant. For instance, how reliable are perception-based surveys in international or interregional comparison? Which respondents are reticent and under which circumstances (inter alia Azfar/Murrell 2009; Clausen et al. 2010; Friesenbichler et al. 2014), what do people mean when they state that firms 'like theirs' pay bribes (Clarke 2012) and w h o is entering into transactions in which bribes may be required? These and related issues are underresearched; yet they are very important in understanding the empirical results on corruption research. The fourth research desideratum is the analysis of grand and systemic corruption. It can neither sensibly be simulated in the lab nor analyzed via randomized control trials due to the sheer scale, nor does it lend itself to econometric analyses as the number of observations is small (often one). Yet, as systemic corruption affects the entire politicaleconomic system, its understanding is crucial for designing successful anti-corruption policies. This includes a comprehension of factors that have led to corruption being systemic (and not only frequent) and of the way the incentive structures were designed to keep the system corrupt. 8 Likewise the design of anti-corruption policies, both success stories and failures, need to be analyzed. Indonesia after Suharto's demise is a good example for a fairly successful development towards lower, non-systemic corruption that has suffered from a lot of setbacks (e.g. Butt 2011). One of the reasons why these highly relevant grand stories have not received the attention in the economics literature that they deserve may be that systemic political-economic corruption cannot be analyzed with the same methodological elegance as more small-scale, but more frequent corrup7

8

In p a r t i c u l a r sectors s h o u l d be s t u d i e d t h a t h a v e received relatively little a t t e n t i o n so far. T h e higher e d u c a t i o n sector is a case in p o i n t - even t h o u g h t h e r e is a m p l e a n e c d o t i c evidence t h a t g r a d e s a n d degrees a r e p u r c h a s e d in m a n y c o u n t r i e s , there a r e h a r d l y a n y empirical studies t h a t s h o w t h e d e t e r m i n a n t s of such a c o r r u p t b e h a v i o r . A n e x a m p l e f o r this is t h e d e s c r i p t i o n of systemic c o r r u p t i o n in S u h a r t o ' s I n d o n e s i a by M c L e o d (2008).

110 · Johann Graf Lambsdorff and Günther G. Schulze

tion. 9 As a consequence, corruption in the allocation of driver's licenses in India may be better analyzed (Bertrand et al. 2007) than the systemic corruption in Indonesia or China. W e suggest that if there is a trade-off between relevance and elegance (often times there is none), relevance should take precedence over elegance. M a c r o studies that are based on perceptions of grand corruption and also qualitative studies d o have their role. 5

This volume

The papers in this volume - t w o f r o m the lab and four f r o m the field - contribute to closing the gap for the first three research desiderata for empirical corruption research. They venture into unchartered waters as they study corrupt behavior in novel and relevant situations of strategic interactions (Li et al. and Khachatryan et al.) and analyze cross-cultural differences (Li et al.), evaluate corruption attitudes across religions (Gouda/Park), use cross-country firm-level data to study the interaction of micro- and macro-determinants on the effects of corruption on productivity (de Rosa et al.), analyze for the first time corruption in college admission in China (Liu/Peng) and lastly advance our methodological understanding by looking at selection issues in corruption research (Ivlevs/Hinks). These studies enhance our understanding of corruption in various ways by analyzing corrupt behavior at the grassroots level. Donato De Rosa, Nishaal Gooroocburn, and Holger Görg analyze the effect of bribery on firm level productivity using the EBRD/World Bank 2 0 0 9 Business Environment and Enterprise Performance Survey (BEEPS) of firms in 28 countries of Central and Eastern Europe and Central Asia. This cross-country firm-level data set allows investigating macro (country level) and micro (firm level) influences on individual firm productivity. Applying an augmented production function approach that controls for a wide variety of firm characteristics, firms' perceptions of institutional quality, and competition intensity, the authors measure the effect of corruption o n total factor productivity (TFP) in t w o dimensions: whether firm officials pay bribes frequently (monetary costs) and w h a t percentage of senior management's time is spent on dealings with public officials (time or opportunity costs). As corruption is endogenous they apply also an instrumental variable a p p r o a c h and find that productivity is reduced by monetary bribes, but not by the time that management spends on government and bureaucracy. The productivity reducing effect of corruption is stronger in high corruption environments. They find n o evidence for the "greasing the wheels" hypothesis which posits that bribing helps to 'get things done'. Artjoms lvlevs and Timothy Hinks address the issue of sample selection in corruption analyses, arguing that in order to face the decision whether to bribe people need to interact with public officials. Those that d o interact are a n o n - r a n d o m sample of the entire population and thus a selection bias may result. T o remedy this selection bias they apply a H e c k m a n procedure. lvlevs and H i n k s use the 2 0 1 0 'Life in Transition 2 ' survey conducted in 30 Central and Eastern European and Central Asian countries (and also five Western European countries), which asks for actual bribing experiences in dealings with eight different types of public officials, a n d find evidence for a significant selection bias. Their findings show some c o m m o n features across the countries studied, but also distinctive differences between country groups providing additional evidence for the context and culture-specific nature of corruption (Kis-Katos/Schulze 2014). ' An exception is the phenomenon of politically connected firms as stock market valuations can be easily measured over time and politically connected firms constitute a substantial share of firms in the national stock markets.

What Can We Know About Corruption? - 1 1 1

Qijun Liu and Yaptng Peng provide evidence of corruption in art college admissions in China. They have compiled a new data set on art students that have been admitted and find that a large share of them have bribed public officials in order to facilitate acceptance. They show that the likelihood of bribing depends on the perceived corruption level and the personal tolerance for corruption, but not on gender. Corruption is higher for middle income families than for poor or rich families and it is more frequent for the admission of lower tier art colleges. More capable (i.e. more experienced) candidates are less likely to bribe. Moamen Gouda and Sang-Min Park employ data from the World Values Survey to investigate the link between attitudes towards corruption and religion. They find that people who consider religion and God to be important and who attend religious services are less likly to state that paying a bribe is justified. Interestingly, this finding does not depend on the religious denomination. Relevant to levels of corruption is thus not the actual content of a religious belief but more its intensity. At the risk of overinterpreting these findings, the religious denomination does not provide a clear candidate for understanding behavior in a corrupt context. This would indeed be an interesting conjecture for experimental researchers, who may have to care about one context less. Sha Li, Christoph Bühren, Björn Frank, and Haiying Qin take a deep look at human behavior in a context that is highly relevant to reform. Anticorruption programs in the public sector often involve the rigorous implementation of the "four eyes principle". This mechanism is often employed as a safeguard against misuse. It has also been repeatedly involved in efforts to reduce bribe-taking and nowadays belongs to the standard arsenal in the public sector. Bribing two public servants, so goes the argument, is more demanding than bribing just one. If only one rejects the offer a corrupt relationship cannot be established. But in laboratory experiments in China and Germany the authors report the opposite. As compared to individual players, groups of three players were offered higher bribes and they more often reciprocated. The authors relate this finding to a diffusion of responsibility. Groups decrease their members' moral costs, making it more likely to engage in malpractice. Elina Khachatryan, Sebastian Kube, and Björn Vollan investigate citizens' reactions to extortive officials. Depending on the treatment, citizens can either reward officials for good behavior or punish them for bad behavior. Which of these methods is superior for reducing extortion? The authors find that this largely depends on the mode of interaction, in particular, whether officials and citizens are in a one-shot or a repeated exchange. Reporting is superior in reducing extortion in one-shot interactions, where laboratory subjects are randomly re-matched with a different partner. But in repeated exchange there is an advantage to rewarding: Citizens are less likely to pay the requested bribe. The findings might be explained by feelings of reciprocity. The chance to recommend an official may frame an environment of positive reciprocity. Being extorted in such an environment may stir higher levels of resistance. Certainly, whether these findings provide direction for reform, particularly in areas where also collusive bribery might arise, is something that will require further research. We believe this volume advances research on corruption in the four key area. The contributions not only look at the micro side but also the macro causes of corruption, the ones that tend to be understudied otherwise. They address robust differences across countries and identify where to look for context-specificity. They seek to join experiences in the lab with those in the field and, finally, take a critical look at methodological issues. The

112 · Johann Graf Lambsdorff and Günther G. Schulze

c o n t r i b u t i o n s bring a b o u t original findings t h a t should inspire r e f o r m . At the same time w e believe they can reveal h o w research o n c o r r u p t i o n should continue in years to come.

References Acemoglu, D., S. Johnson, J.A. Robinson (2001), The Colonial Origins of Comparative Development: An Empirical Investigation. The American Economic Review 91(5): 1369-1401. Aidt, T.S. (2003), Economic analysis of corruption: a survey. The Economic Journal 113: F632-F652. Alatas, V., L. Cameron, A. Chaudhuri, N. Erkal, L. Gangadharan (2009), Subject pool effects in a corruption experiment: A comparison of Indonesian public servants and Indonesian students. Experimental Economics 12(1): 113-132. Alt, J.E., D.D. Lassen (2014), Enforcement and Public Corruption: evidence from the American States. Journal of Law, Economics, and Organization, forthcoming, doi:10.1093/jleo/ews036. Armantier, O., A. Boly (2013), Comparing Corruption in the Laboratory and in the Field in Burkina cFaso and in Canada. Economic Journal 123: 1168-1187. Azfar, O, P. Murrell (2009), Identifying reticent respondents: assessing the quality of survey data on corruption and values. Economic development and cultural change 57(2): 387—411. Banuri S. and C. Eckel (2012) "Experiments in Culture and Corruption: A Review", in D. Serra and L. Wantchekon (eds.) N e w Advances in Experimental Research on Corruption, Research in Experimental Economics volume 15, Emerald Group Publishing. Bertrand, M., S. Djankov, R. Hanna, S. Mullainathan (2007), Obtaining a driver's license in India: an experimental approach to studying corruption. Quarterly Journal of Economics 122(4): 1639-1676. Butt, S. (2011), Anti-corruption reform in Indonesia: an obituary? Bulletin of Indonesian Economic Studies 47(3): 381-394. Cameron, L., A. Chaudhuri, N . Erkal, L. Gangadharan (2009), Propensities to engage in and punish corrupt behavior: Experimental evidence from Australia, India, Indonesia and Singapore. Journal of Public Economics 93: 843-851. Chaudhuri, A. (2012), Gender and corruption: A survey of the experimental evidence. Pp. 13—49 in: D. Serra, L. Wantchekon (eds.), N e w Advances in Experimental Research on Corruption. Research in Experimental Economics volume 15, Emerald Group Publishing. Bingley, UK. Clarke, G. (2012), What Do Managers Mean When They Say 'Firms Like Theirs' Pay Bribes? International Journal of Economics and Finance 4(10): 161-169. Clausen, B., A. Kraay, P. Murrell (2010), Does respondent reticence affect the results of corruption surveys? Evidence from the World Bank Enterprise Survey for Nigeria. Pp. 4 2 8 ^ 5 1 in: S. Rose-Ackerman, T. Soreide (eds.), International Handbook of Corruption, Vol. II, ch. 15. Cheltenham: Edward Elgar. De Rosa, D., N . Gooroochurn, H . Görg (2015), Corruption and Productivity: Firm-level Evidence, Jahrbücher für Nationalökonomie und Statistik 235(2): 115-138. Donchev, D., G. Ujhelyi (2013), What do corruption indices measure? Rev. working paper, University of Houston. Accessed April 7, 2014. http://www.class.uh.edu/faculty/gujhelyi/ corrmeasures.pdf. Drugov, M., J. Hamman, D. Serra (2014), Intermediaries in corruption: an experiment. Experimental Economics 17(1): 78-99. Duflo, E., R. Hanna, S. Ryan (2012), Incentives Work: Getting Teachers to Come to School. American Economic Review 102(4): 1241-1278. Escresa, L., L. Picci (2014), A New Cross-National Measure of Corruption. Unpublished manuscript. Online available at http://ssrn.com/abstract=2294477 (retrieved May 24, 2014). Faccio, M., (2006), Politically connected firms. American Economic Review 96(1): 369-386. Ferguson, T., H.-J. Voth (2008), Betting on Hitler - the value of political connections in Nazi Germany. Quarterly Journal of Economics 123(1): 101-137. Ferraz, C., F. Finan (2008), Exposing corrupt politicians: the effects of Brazil's publicly released audits on electoral outcomes. Quarterly Journal of Economics 123(2): 703-745.

What Can We Know About Corruption? - 1 1 3

Fisman, R. (2001), Estimating the value of political connections. American Economic Review 91(4): 1095-1102. Fisman, R., E. Miguel (2007, Corruption, norms, and legal enforcement: evidence from diplomatic parking tickets. Journal of Political Economy 115(6): 1020-1048. Frank, B., G.G. Schulze (2000), Does Economics Make Citizens Corrupt? Journal of Economic Behavior and Organization 43(1), 101-113. Frank, B., J. Graf Lambsdorff (2010), Bribing versus Gift-Giving - an Experiment. Journal of Economic Psychology 31(3): 347-357. Friesenbichler, K., G. Clarke, M. Wong (2014), Price competition and market transparency: evidence from a random response technique. Empirica 41: 5 - 2 1 . Glaeser, E., R. Saks (2006), Corruption in America. Journal of Public Economics 90(6-7): 1053-1072. Gouda, M., S.-M. Park (2015), Measuring the Effect of Religious Loyalty and Judicial Accountability on Corruption. Jahrbücher für Nationalökonomie und Statistik 235(2): 184—206. Hamman, J.R., G. Loewenstein, R.A. Weber (2010), Self-interest through Delegation: An Additional Rationale for the Principal-Agent Relationhip. American Economic Review: 1826-1846. Henderson, J.V., A. Kuncoro 2011, Corruption and local democratization in Indonesia: the role of Islamic parties. Journal of Development Economics 94(2): 164-180. Ivlevs, Α., T. Hinks (2015), Bribing Behaviour and Sample Selection. Evidence from PostSocialist Countries and Western Europe. Jahrbücher für Nationalökonomie und Statistik 235(2): 139-167. Kaiser, K., D. Pattinasarany, G.G. Schulze (2006), Decentralization, Governance and Public Services in Indonesia. Pp. 164-207 in: G. Peterson, P. Smoke, E.J. Gómez (eds.), Decentralization in Latin America and Asia: An Inter-Disciplinary Perspective. Cheltenham/UK: Edward Elgar Press, ch.7. Khachatryan, E., S. Kube, B. Vollan (2015), Mitigating Extortive Corruption? Experimental Evidence, Jahrbücher für Nationalökonomie und Statistik, 235(2): 228-241. Khwaja, A.I., A. Mian (2005), Do Lenders Favor Politically Connected Firms? Rent Provision in an Emerging Market. Quarterly Journal of Economics 120(4): 1371-1411. Kis-Katos, Κ., G.G. Schulze (2013), Corruption in Southeast Asia: a survey of recent research. Asian-Pacific Economic Literature 27: 79-109. Kis-Katos, K., G.G. Schulze (2014), Context-specificity of economic research: The example of corruption research in Southeast Asia. Pp. 187-210 in: M. Huotari, J. Rüland, J. Schlehe (eds.), Reflecting Methodology in Southeast Asian Studies. Palgrave. Knack, S., Ρ Keefer (1995), Institutions and economic performance: cross-country tests using alternative institutional measures. Economics and Politics 7(3): 207-227. Kremer, Μ., N. Chaudhury, F.H. Rogers, K. Muralidharan, J. Hammer (2005), Teacher Absence in India: A Snapshot. Journal of the European Economic Association 3(2-3): 658-667. La Porta, R., F. Lopez-de-Silanes, A. Shleifer, R. Vishny (1999), The quality of government. Journal of Law, Economics and Organization 15(1): 222-279. Lambsdorff, J. Graf (2007), The Institutional Economics of Corruption and Reform: Theory, Evidence, and Policy. Cambridge University Press. Lambsdorff, J. Graf (2012). Behavioral and Experimental Economics as a Guidance to Anticorruption. 2 7 9 - 2 9 9 in: D. Serra, L. Wantchekon (eds.), New Advances in Experimental Research on Corruption. Research in Experimental Economics volume 15. Emerald Group Publishing, Bingley, UK. Lambsdorff, J. Graf, B. Frank (2011), Corrupt reciprocity - experimental evidence on a men's game. International Review of Law and Economics 31(2): 116-125. Levitt, S., J.A. List (2007a), Viewpoint: On the Generalizability of Lab Behaviour to the Field. Canadian Journal of Economics 40(2): 347-370. Levitt, S., J.A. List (2007b), What do Laboratory Experiments Measuring Social Preferences Reveal about the Real World. Journal of Economic Perspectives 21(2): 153-174. Li, S., C. Bühren, Β. Frank, H . Qin (2015), Group Decision Making in a Corruption Experiment: China and Germany Compared. Jahrbücher für Nationalökonomie und Statistik 235(2): 207-227.

114 · Johann Graf Lambsdorff and Günther G. Schulze

Liu, Q., Y. Peng (2015), Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China. Jahrbücher für N a t i o n a l ö k o n o m i e und Statistik 235(2): 1 6 8 - 1 8 3 . M a u r o , P., (1995), Corruption and Growth. Quarterly Journal of Economics 110(3): 6 8 1 - 7 1 2 . McCulloch, N., G.G. Schulze, J. Voss (2010), W h a t determines firms' decisions to formalize? Evidence f r o m rural Indonesia. University of Freiburg IEP Discussion Paper N o . 13. McLeod, R. (2008) Inadequate budgets and salaries as instruments for institutionalizing public sector corruption in Indonesia. South East Asia Research 16(2): 1 9 9 - 2 2 3 . M o c a n , Ν . (2008), W h a t determines corruption? International evidence f r o m microdata'. Economic Inquiry 46(4): 4 9 3 - 5 1 0 . Olken, B. (2007), Monitoring corruption: evidence from a field experiment in Indonesia. Journal of Political Economy 115(2): 2 0 0 - 2 4 9 . Olken, B. (2009), Corruption perceptions vs. corruption reality. Journal of Public Economics 93(7-8): 9 5 0 - 9 6 4 . Olken, B.A., R. Pande (2011), Corruption in developing countries. NBER Working Paper N o . 17398. Cambridge/MA. Pande, R. (2007) Understanding political corruption in low income countries. Pp. 3 1 5 5 - 3 1 8 4 in: T.P. Schultz, J. Strauss (eds.), H a n d b o o k of Development Economics Vol. 4. Elsevier, Amsterdam. Rand, J., F. T a r p (2012), Firm-level corruption in Vietnam. Economic Development and Cultural Change 60(3): 5 7 1 - 5 9 5 . Reinikka, R., J. Svensson (2004), Local capture: evidence f r o m a central government transfer program in Uganda. Quarterly Journal of Economics 119(2): 6 7 9 - 7 0 5 . Reinikka, R., J. Svensson (2005 ), Fighting corruption to improve schooling: evidence f r o m a newspaper campaign in Uganda. Journal of the European Economic Association 3(2-3): 2 5 9 - 2 6 7 . Reinikka, R., J. Svensson (2006), Using micro-surveys to measure and explain corruption. W o r l d Development 34(2): 3 5 9 - 3 7 0 . Rivas, F. (2013), An experiment on corruption and gender. Bulletin of Economic Research 65 (1): 10—42. Rose-Ackerman, S. (1975), The economics of corruption. Journal of Public Economics 4: 187-203. Schulze, G.G., Β. Frank (2003), Deterrence versus Intrinsic Motivation: Experimental Evidence on the Determinants of Corruptibility. Economics of Governance 4(2): 1 4 3 - 1 6 0 . Schulze, G.G., B.S. Sjahrir, Ν . Z a k h a r o v (2013), Corruption in Russia. University of Freiburg IEP Discussion Paper N o . 22, M a r c h , Freiburg. Shleifer, Α., R.W. Vishny (1993), Corruption. Quarterly Journal of Economics 108: 5 9 9 - 6 1 8 . Sjahrir, B.S., K. Kis-Katos, G.G. Schulze (2014), Administrative overspending in Indonesian districts: The role of local politics. World Development 59: 1 6 6 - 1 8 3 . Svensson, J. (2005), Eight questions about corruption. Journal of Economic Perspectives 19(3): 19^12. Treisman, D. (2000), The causes of corruption: a cross-national study. Journal of Public Economics: 3 9 9 - 4 5 7 . Van Rijckeghem, C., B. Weder (2001), Bureaucratic corruption and the rate of temptation: do wages in the civil service affect corruption, and by h o w much? Journal of Development Economics 65(2): 3 0 7 - 3 3 1 . Vicente, P. (2010), Does oil corrupt? Evidence f r o m a natural experiment in West Africa. Journal of Development Economics 92: 2 8 - 3 8 . Prof. Dr. J o h a n n Graf Lambsdorff, Universität Passau, Lehrstuhl für Volkswirtschaftslehre mit Schwerpunkt Wirtschaftstheorie, R a u m WIWI 108, Innstraße 27, 94032 Passau, Germany. [email protected] Prof. Dr. Günther G. Schulze, Albert-Ludwigs-Universität Freiburg im Breisgau, Institute for Economic Research, Department of International Economic Policy, R o o m 2 5 1 2 , Platz der Alten Synagoge 1, 7 9 0 8 5 Freiburg, Germany. [email protected]

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2015) Bd. (Vol.) 235/2

Corruption and Productivity: Firm-level Evidence Donato De Rosa* The World Bank, Washington DC, USA Nishaal Gooroochurn London Metropolitan University, UK Holger Görg Kiel Institute for the World Economy and University of Kiel, Germany; IZA JEL 014; P37 Corruption; productivity; Central and Eastern Europe; CIS.

Summary Using enterprise data for the economies of Central and Eastern Europe and the CIS, this study examines the effects of corruption on productivity. Corruption is narrowly defined as the occurrence of informal payments to government officials to ease the day-to-day operation of firms. The effects of this "bribe tax" on productivity are compared to the consequences of red tape, which may be understood as imposing a "time tax" on firms. When testing effects in the full sample, only the bribe tax appears to have a negative impact on firm-level productivity, while the effect of the time tax is insignificant. We also find that the surrounding environment influences the way in which firm behaviour affects firm performance. In particular, in countries where corruption is more prevalent and the legal framework is weaker, bribery is more harmful for firm-level productivity.

1

Introduction

O n e of the most obvious facts a b o u t corruption is that p o o r countries tend to be the most corrupt (e.g., Bardhan 1997). Available data at the country level support this view. For instance, there is a 0.81 correlation between GDP per capita and Transparency International's Corruption Perception Index, one of the most commonly used measures of corruption. Beyond this simple observation, however, understanding corruption is not an easy task. For instance, it is debatable whether corruption is a cause of low incomes per capita, one of its consequences or, as it seems more likely, whether the relationship between corruption and income is an intricate one, made of a web of dynamic interactions, whereby some countries appear trapped in a condition of low incomes and high corruption. 1

* T h e a u t h o r s are grateful t o Irena Grosfeld, Erich G u n d l a c h , M a r i a n a Iootty, J o h a n n Graf Lambsdorff, Jorge Pena, Ekaterina Z h u r a v s k a y a , seminar participants at the Kiel Institute and the W o r l d Bank, and t w o a n o n y m o u s referees for many helpful c o m m e n t s . 1 T h e r e is an extensive literature o n the causes of c o r r u p t i o n , see, for example, Acemoglu et al. (2001) w h o focus on the role of inherited institutions, La Porta et al. (1999) w h o look at the importance of legal origin, or Brunetti and Weder (2001) w h o investigate the role of a free press.

116 • Donato De Rosa, Nishaal Gooroochurn, and Holger Görg

In common parlance and in academic research corruption is defined in various ways. Most often it is understood as bribery, whereby an official demands informal payments to perform an official task - e.g. issuing a license - or to circumvent laws and regulations. State capture may also qualify for the definition of corruption, when bureaucrats subject themselves to more or less legal forms of lobbying, involving monetary bribes or other forms of exchange of favours, to afford preferential treatment to certain private interests. Political patronage, nepotism and cronyism, whether or not they involve monetary kickbacks, may also be included in a broad definition of corruption. For our purposes, corruption is defined as a "bribe tax", a certain amount of money necessary to enforce a contract between an individual and the state. In this asymmetric relationship, the state - or its agents - define the property rights of individuals and enforce them with a monopoly on the legitimate use of force. The institutions that govern this type of "vertical" transactions between the state and its citizens are defined by Acemoglu and Johnson (2005) as property rights institutions and are distinguished from contracting institutions that regulate "horizontal" transactions among ordinary citizens. Property rights institutions are inefficient when they allow those who control the state to extract rents from producers (Acemoglu 2006) and the extortion of bribes from firms may be viewed as a form of rent extraction perpetrated by bureaucrats (Shleifer/Vishny 1998). The aim of this paper is to investigate the link between corruption and economic performance, more specifically, productivity. Our analysis exploits the advantages of firm level data by using the information contained in the 2009 Business Environment and Enterprise Performance Survey (BEEPS) of firms in 28 countries of Central and Eastern Europe and Central Asia. The sample of countries is very diverse. It covers all the formerly communist countries of Europe and the Former Soviet Union, which have undergone the profound institutional transformation connected with transition to a market economy. The group of formerly communist countries presents substantial variation, ranging from the low income economies of Central Asia, to high income Central European countries, which, as members of the EU, tend to have a fully developed market system. Unlike similar country level studies using broad definitions of corruption and institutional quality combined with aggregate measures of economic performance (e.g., Lambsdorff 2003; Meon/Sekkat 2005), investigation of the effects of bribery on firm level productivity will allow to be more precise regarding the incentives of economic agents to engage in corrupt behaviour and the consequences this has for productive efficiency. Variables for firm level bribery and productivity can be obtained from the BEEPS database, which, in addition to information on the occurrence of bribing and other aspects of firm operation and performance, allows estimating a measure of total factor productivity (TFP) at the enterprise level. The need to recur to bribery is often linked to the power of government officials to impose and enforce regulatory requirements on individuals and firms and to exact bribes in the process (see, for example, Djankov et al. 2002). In order to account for this possibility, it is necessary to identify some measure of the power that officials have over firms as enforcers of regulatory requirements. The BEEPS survey offers such a measure at the firm level. It refers to the time that enterprise managers are required to spend complying with government regulations, amounting to a time tax imposed on firms. This may be interpreted as an opportunity cost borne by firms, which, in isolation or in combination with the bribe tax, potentially constitutes a drag on enterprise performance. The availability

Corruption and Productivity: Firm-level Evidence - 1 1 7

of a firm level measure for the time spent dealing with bureaucracy offers the opportunity to perform a direct test of the so-called "efficient grease" hypothesis, which is explicitly defined in the literature in terms of bribery helping reduce the time required for some interaction between an economic agent and the state (e.g., Lui 1985). 2 The effects of corruption on productivity are modelled based on a firm level production function. 3 This paper offers a number of novelties with respect to the existing literature on the consequences of corruption. First, the analysis makes use of cross-country firm level data in order to investigate the effects of country-level characteristics in mitigating the negative productivity impact of bribe payments. Second, we deal with the potential endogeneity of corruption at the firm level using an instrumental variables approach. To do so, we make use of the rich firm level data we have available in the BEEPS data to come up with reasonable instruments. Third, the time spent dealing with bureaucracy, which is intimately linked to bribe payments, is directly examined. Results of the econometric analysis highlight some differences between the effects of bribery and the time tax. Across the entire sample, whereas the time devoted to complying with government regulations has no significant effect on firm level productivity, corruption has a statistically significant negative effect in instrumental variable specifications. Additionally, regression results show no evidence in favour of the "efficient grease" hypothesis, whereby bribing would be a second best option to achieve higher productivity levels by helping firms circumvent burdensome regulatory requirements. Namely, when bribing is made conditional on the time spent dealing with government regulations, the interaction term has no significant effect on productivity, implying that no trade-off emerges between the time and the bribe tax. We test the hypothesis that the effects of corrupt behaviour on firm performance vary depending on broader country characteristics. In order to do so, bribery experienced by individual firms is made conditional on broader country characteristics that may play a role in influencing individual choices to engage in corrupt behaviour. The first measure used is the Transparency International Corruption Perception Index (CPI), which provides an independent measurement of perceived corruption in 180 countries, based on 13 different expert and business surveys. Inclusion of country-wide corruption may be interpreted to reflect the extent to which peer effects may be conducive to corrupt behaviour, as in theoretical models that explain the persistence of corruption with social effects, such as Tiróle (1996). The second country-wide measure is the World Economic Forum (WEF) index of the effectiveness of the legal framework in resolving disputes,

2

3

Use of the time t a x as a specific outcome of the institutional inefficiencies linked with corruption distinguishes our paper from other studies, such as Méon and Sekkat ( 2 0 0 5 ) and M é o n and Weill ( 2 0 1 0 ) . These authors, for lack of more precise country level data explicitly reflecting the time involved in interactions with the state, use generic country level proxies of "governance" to test whether corruption does indeed act as efficient grease. Using the same BEEPS sample, Blagojevic and Damijan ( 2 0 1 3 ) differentiate the effect of corruption by firm ownership. Fisman and Svensson ( 2 0 0 7 ) look at the relationship between corruption and firm growth using firm level data for Uganda. Our paper is different in a number of aspects. First, we use firm level data for a number of countries in Central and Eastern Europe and Central Asia, which allows us to explore cross country heterogeneity. This is important, as we show, since we find strong differences in the relationship between corruption and productivity in different countries. Second, we look at productivity, not growth of sales as in Fisman and Svensson ( 2 0 0 7 ) .

118 • Donato De Rosa, Nishaal Gooroochurn, and Holger Görg

intended to capture the possibility that a higher likelihood of sanctioning by the legal system m a y act as a deterrent. 4 Regression analysis shows that firms that d o not pay bribes in environments with a high prevalence of corruption and inefficient legal f r a m e w o r k s experience higher productivity. Furthermore, when overall levels of corruption exceed a certain threshold, the total effect of corruption on productivity - i.e. the combination of individual and country effects is increasingly negative. This indicates that, whereas environmental circumstances are beyond the choice set of individual firms, managers still have some degree of a u t o n o m y in deciding whether to recur to bribery or not and this affects firm level productivity. The remainder of the study is structured as follows. The following section provides an overview of the relevant literature on the possible effects of corruption on economic performance. N e x t is a description of the BEEPS 2 0 0 9 data, as well as an exposition of the econometric methodology. The fourth section demonstrates the effects of corruption on productivity, both unconditional and conditional on the time tax experienced by individual firms and on country characteristics in terms of prevalence of corruption and efficiency of the legal f r a m e w o r k . The final section concludes. 2

Background literature

Corruption is sometimes seen a second-best option when it helps reduce the time involved in dealing with burdensome regulatory requirements. According to the proponents of this "efficient grease" hypothesis this would happen since, in spite of the transaction costs it entails, bribery would lead to lower effective red tape for the firm, hence increasing firm performance and productivity. A theoretical f r a m e w o r k for this efficiency enhancing role of corruption is provided by Lui's (1985) queuing model, where the size of bribes by different economic agents reflects their different opportunity cost, with more efficient agents more able or willing t o buy lower effective red tape, reflected in a lower "time t a x " . As a consequence, a license or contract awarded on the basis of bribe size could achieve Pareto-optimal allocation. However, K a u f m a n n and Wei (1999) identify a m a j o r shortcoming in Lui's (1985) assumptions, namely that the regulatory burden is treated as exogenous, independent of the incentives for officials to take bribes. This may not be the case since the incentives of bureaucrats can be modified by specific policy measures. Ultimately, because of this assumption, Lui's theory is partial equilibrium in nature, and may not hold in a general equilibrium. M o r e generally, Bardhan (1997) argues that red tape and corruption are not exogenous, as they are caused - or at least preserved or aggravated - by those w h o benefit f r o m an overregulated and corrupt system. Hence, as argued by Aidt and Dutta (2008), even if corruption helps overcome cumbersome regulation in the short term, it creates incentives to create more such regulation in the long term. Empirical evidence, especially at the micro level, is generally not supportive of the efficient grease hypothesis 5 , with corruption f o u n d

4

5

North (1990) distinguishes between "informal norms," "formal rules" and "enforcement" as pillars of the institutional framework of an economy. For our purposes, the CPI might also be viewed as a proxy of informal norms of behaviour dictated by social networks, while the WEF index of the efficiency of the legal framework might be interpreted as a proxy for the quality of formal rules or their enforcement. One of few exceptions is Egger and Winner (2005) who, based on country-level evidence, argue that corruption can help overcome regulatory obstacles and stimulate FDI.

Corruption and Productivity: Firm-level Evidence - 1 1 9

to increase the time spent by managers dealing with red tape (Kaufmann/Wei 1999) and to hamper firm growth (Fisman/Svensson 2007). 6 If corruption were a means to "greasing the wheels of commerce" it could possibly have positive effects on economic performance by reducing transaction costs in the vertical transactions between the state and its citizens. However, the theoretical and empirical evidence in favour of the opposite argument appears more convincing, highlighting the negative consequences of corruption for resource allocation, entrepreneurship, investment and innovation.7 The main argument is that the prevalence of corruption may distort resource allocation by increasing the returns to rent-seeking compared to those of productive activities (Baumol 1990). An extremely corrupt environment may induce individuals to minimize interaction with the state by expanding more slowly, operating in the informal sector or even forgoing entrepreneurial activity altogether. Corroborating this point, Djankov et al. (2002) find that entry of new firms is more difficult in the presence of greater corruption and larger unofficial economies. Corruption also affects the allocation of entrepreneurial talent, when, in highly corrupt environments, entrepreneurs may devote greater efforts to obtaining valuable licenses and preferential market access than to improving productivity (Murphy et al. 1991). When entrepreneurial talent is directed towards productive activity, the rate of innovation and investment is likely to increase with positive consequences for productivity and income growth. In contrast, when talent is directed towards rent extraction, returns to talent are maximized by appropriating wealth rather than wealth creation (Murphy et al. 1991, 1993; Acemoglu/Verdier 1998). The sources of productivity enhancements, technological progress and investment, may be directly affected in corrupt environments. For instance, entrepreneurs may have incentives to adopt inefficient "fly-by-night" technologies of production with an inefficiently high degree of reversibility, which allows them to react more flexibly to future demands from corrupt officials - and more credibly threaten to shut down operations (Svensson 2003). Additionally, vested interests may directly oppose the adoption of new technologies, which would threaten their position of influence by rendering obsolete the older technological vintages they control (Krusell/Rios-Rull 1996). Finally, corruption may erect de facto entry barriers into otherwise competitive markets with discouraging effects for investment decisions, in a mechanism similar to the one proposed by Alesina et al. (2005). Alternatively, the monetary cost involved in the payment of bribes may simply limit the amount of resources available to expand productive capacity via investment. In addition to distortionary allocation effects, the discretionary power of state officials will increase the risk of expropriation thus reducing the appropriability of returns to investment and innovation (Demsetz 1967 and Alchian/Demsetz 1973). This will further diminish rewards for entrepreneurial behaviour, while propping up inefficient firms engaged in corrupt practices. In this spirit, Johnson et al. (2002), using firm-level data from former communist countries in Europe, find a negative effect of paying bribes on investment and interpret this finding as the effect of insecure property rights.

6

7

Hellman et al. ( 2 0 0 0 ) also find a negative correlation between bribe payments and sales, although these authors do not instrument for bribe payments. For an overview of the consequences of corruption, see Lambsdorff ( 2 0 0 3 ) , Svensonn ( 2 0 0 5 ) , and Shleifer and Vishny ( 1 9 9 8 ) .

120 · Donato De Rosa, Nishaal Gooroochurn, and Holger Görg

Whereas corruption can impact economic performance through all these channels, its adverse effects may be non-linear and depend on the overall level of institutional quality (or governance) in the country. Two studies - both based on country level data - find such non-linearities in the relationship between corruption and growth, namely a more negative effect when institutional quality is poor. Méon and Sekkat (2005), based on a sample of 71 countries between 1970 and 1998 and using various proxies for both corruption and governance, 8 find that corruption is most harmful to growth where governance is weak. Méndez and Sepúlveda (2006) examine country-level evidence by using different proxies for corruption 9 , as well as the Freedom House index of political freedom as a proxy for overall institutional quality. They find that the relationship between corruption and growth is non-monotonic with corruption having negative effects only at high levels of incidence. 3

Data and methodology

To assess the effects of corruption on firm performance this paper uses the 2 0 0 9 EBRD/World Bank Business Environment and Enterprise Performance Survey (BEEPS) of firms in 28 transition and developed countries. 10 The BEEPS was specifically conceived to assess the extent to which government policies and practices facilitate or impede business activity. It therefore provides a vast array of information on the behaviour and performance of firms, which allows to explicitly model the possible influence of various firm characteristics on the occurrence and impact of corruption at the firm level. The main disadvantage of the data is that it is only available for a cross section in 2 0 0 9 , which means we cannot look at changes in firm performance over time. Also, it makes interpretation of the estimated effects as "causal" more difficult, hence we revert to an instrumental variables approach in the analysis below. Table 1 lists the countries included in the sample. It shows that there is substantial variation in terms of income group (based on the World Bank classification for 2 0 0 8 ) " and EU membership. Such high dispersion in income per capita provides a particularly rich sample, that allows controlling for specific country characteristics linked to the level of development and, in particular, to the quality of the institutional environment. The business environment is examined by asking firms to assess how various factors affect business operations, including infrastructure, financial services, government regulation, tax administration, judiciary functions. Corruption is also examined, allowing us to model

8

9

10

11

The authors use the Transparency International CPI, as well as a number of indicators from the World Bank Governance Indicators, as proxies for corruption and governance. The authors use the International Country Risk Guide (ICRG), the I M D index of corruption is published by the Institute for Management Development (IMD) and the corruption perceptions index (CPI) compiled by Transparency International. The data set and documentation is available at http://www.ebrd.com/pages/research/economics/ data/beeps.shtml. Previous rounds of the BEEPS surveys were carried out in 1 9 9 9 , 2 0 0 2 and 2 0 0 5 . Unfortunately, given the changes in survey design, a meaningful link between the 2 0 0 9 and earlier versions is not feasible. Also, the survey nature of the data leads to the loss of many observations in multivariate regressions, owing to non-response rates. Economies are divided according to 2 0 0 8 GNI per capita. The groups are: low income, $ 9 7 5 or less; lower middle income, $ 9 7 6 - $ 3 , 8 5 5 ; upper middle income, $ 3 , 8 5 6 - $ 1 1 , 9 0 5 ; and high income, $ 1 1 , 9 0 6 or more. See http://web.worldbank.org/WBSITE/EXTERNAL/ DATASTATISTICS/0„contentMDK:20420458~menuPK:64133156~pagePK:64133150~ p i P K : 6 4 1 3 3 1 7 5 ~ t h e S i t e P K : 2 3 9 4 1 9 , 0 0 . h t m l (accessed 15 October 2 0 0 9 ) .

Corruption and Productivity: Firm-level Evidence • 121

Table 1 List of countries in the sample Economy

Income group

Croatia Estonia Slovenia Czech Republic Hungary Slovak Republic Belarus Bosnia and Herzegovina Bulgaria Kazakhstan Latvia Lithuania Macedonia, FYR Montenegro Poland Romania Russian Federation Serbia Turkey Albania Armenia Azerbaijan Georgia Moldova Ukraine Kyrgyz Republic Tajikistan Uzbekistan

High income High income High income High income High income High income Upper middle Upper middle Upper middle Upper middle Upper middle Upper middle Upper middle Upper middle Upper middle Upper middle Upper middle Upper middle Upper middle Lower middle Lower middle Lower middle Lower middle Lower middle Lower middle Low income Low income Low income

EU mer / / / / / ncome ncome ncome ncome ncome ncome ncome ncome ncome ncome ncome ncome ncome ncome ncome ncome ncome ncome ncome

/ / /

/ /

Source: World Bank

its occurrence and impact on the operation and performance of firms. A list of variables used and their description is given in Table A l , and their descriptive statistics are in given in Table A2 in the appendix. A typical concern when using survey data is that of individual perception bias (Kaufman/Wei 1999). Some firms may, for instance, consistently provide positive or negative answers depending on their overall perception of the business climate. In principle, assuming that the bias is uncorrelated across groups of respondents, individual perception bias contributes only to the standard error of estimates obtained from the survey responses. In cross-country surveys, such as the BEEPS, the group within which the bias is likely to be correlated is the particular country in which respondents operate. Perception bias at the country level could originate from different cultural norms and degrees of political freedom across countries, which may influence the choice of specific ratings and the will-

122 · Donato De Rosa, Nishaal Gooroochurn, and Holger Görg

ingness of business people to criticise state institutions. Fries et al. (2003) check for such perception bias in the BEEPS 2 0 0 2 by statistically comparing measures obtained from the aggregation of survey responses to related objective measures and find no significant perception biases across the countries in the sample. Since the BEEPS 2 0 0 9 follows a similar methodology, we may be reasonably confident that perception bias will not affect the results of the analysis. However, as a further control, the analysis that follows will make use of sector and country level fixed effects. 3.1

The bribe tax and productivity

The aim of this study is to evaluate the extent to which corrupt practices by firms may be a drag on their productivity or lead to productivity improvements. In order to capture the complexity of the phenomenon of corruption and its potentially varied effects on the performance of individual firms, the empirical methodology will proceed in three steps. First, is an analysis of the effects of the bribe tax and of the time tax on individual firms, controlling for firm, sector and country characteristics that may influence both phenomena. Second, we proceed with an explicit test of the "efficient grease" hypothesis. Possible trade-offs between time consuming compliance with government regulation and the payment of bribes are modelled by including an interaction term between the time and the bribe tax and observing its effects on firm level productivity. Finally, the effect of individual corrupt conduct on firm level productivity is made conditional on the level of institutional quality in the country. That is, in addition to country fixed effects, the econometric specification includes an interaction term between the firm level bribe tax and independent assessments of the prevalence of corruption or the quality of the legal framework in the country. We model the effect of corruption on TFP using an augmented production function, including, in addition to factor inputs, the set of firm, industry and country characteristics that are assumed to have an effect on output. Hence, we include corruption explicitly in the determination of output, as in (1): vac =

+

+ ßcorruptionijc

+ « 3 Ujc + « 4 M , , ,

+ γΧ^

+ d¡ + dc + £¿/c

+

(1)

where y,jc is log output by firm i in industry j and country c and ^equipment^ ¿ and M are log of land, machinery, employment and materials, respectively. 12

12

The estimation of a one-step augmented production function to investigate the determinants of TFP is common in the literature, see, for example, Javorcik ( 2 0 0 4 ) . An alternative is a two step approach. A first step would be to estimate a simple production function only including Κ' , ^equipment ^ £ a n c j ^ a n ¿ s a v e the estimated residual as TFP. In a second step, TFP is then modelled as being determined by a number of firm and industry characteristics. The one step approach is more efficient than the two step approach. However, the latter has the advantage that it allows to take care of country level heterogeneity in productivity by estimating the first step production function separately for each country. We also employ the two step approach as a robustness check in the working paper version (De Rosa et al. 2 0 1 0 ) , results are similar to those reported below. Estimation of TFP at the firm level is, of course, a task fraught with methodological difficulties. Prime among these is the simultaneity problem in the estimation of factor inputs, which are likely to be endogenously determined with output (see, for example, Levinsohn/Petrin 2 0 0 3 ) . However, due to the cross section nature of our data we are not able to implement an approach a la Levinsohn and Petrin. This should be borne in mind in the interpretation of our empirical results.

Corruption and Productivity: Firm-level Evidence · 123

The main variable of interest is corruptionwhich is the measure of corruption at the firm level. It is defined as a "bribe tax", in the form of a dummy equal to one if a firm replies "frequently", "usually" or "always" to the question "is it common to have to pay some irregular additional payment or gifts to get things done with regard to customs, taxes, licenses, regulations, services, etc.'" 3 The same specification can be used to test the direct effects of the "time tax", defined as the percentage of senior management time devoted to dealing with bureaucratic requirements, by including it in the model as a substitute for the "bribe tax". Consideration of both variables allows verifying the extent to which the time and the bribe tax are different phenomena, with different implications for firm productivity. It should be noted that the overall effect of bribes on productivity might be underestimated due to selection bias, as firms that had to pay the largest bribes may have been driven out of business altogether and, therefore, they are not in the dataset. X¡jc is a vector of control variables that serve to detect observable aspects of firm heterogeneity in our data to allow identification of the effect of the bribe tax on productivity. It consists of sizeM, sizeL, age, exporter, innovator and foreign-owned. SizeM and sizeL are dummy variables representing medium and large firms, respectively. Larger firms are expected to have higher productivity due to the effects of scale economies. The variable age represents the age of firms in 2008, capturing effects such as the vintage of the firm or learning by doing externalities on TFP. Innovator, exporter, and foreign-owned are all dummy variables indicating whether the firm is engaged in innovation (in terms of having positive R & D expenditure), involved in exports and owned by foreign investors. These variables are expected to have a positive effect on productivity. In particular, innovation and R & D expenditures tend to positively affect firm productivity since they lead to the development of more efficient production technologies or to the more effective adoption of technologies developed outside the firm (Aw et al. 2008; Klette/Kortum 2004). At the same time, exporting activity has been found in several empirical studies to be positively associated with firm-level productivity (Wagner 2007). Foreign ownership is associated with various measures of firm performance, including investment, innovation and productivity, since foreign owners can be expected to transfer technology and know-how to domestic affiliates (see, for example, Girma/Görg 2007). In order to account for the possibility that increased competition may act as a form of control on corruption, while, at the same time, affecting firm level productivity, Xjjc also includes a variable for the perceived intensity of competition. The variable is defined " H o w much of an obstacle are competitors to your operations?". Specifically, firms are asked to rank whether competition is an obstacle on a scale from 0 (no obstacle) to 4 (very severe obstacle). We define our variable as the difference between the individual firm's response and the country average. As mentioned earlier, the rationale for including the competition variable is that, as firms' profits are driven down by competitive pressure, there are no excess profits from which to pay bribes (Ades/Di Telia 1999).

13

The BEEPS also includes questions that ask more specifically a b o u t the level of bribery, i.e., the a m o u n t that is typically paid by a firm to "get things done". Perhaps unsurprisingly, this very delicate question is only answered very rarely by firms. For example, for the regression in Table 3, column 1 (for which we have 1,666 observations when using the dummy) we would only have 275 observations if we defined bribery using the responses to the question about the amount of money spent. Hence, we prefer the dummy variable for our analysis.

124

Donato De Rosa, Nishaal Gooroochurn, and Holger Görg

X¡j c also includes two measures of the firm's perception of the quality of the institutional environment. The first is a dummy variable equal to one if the firm responds that the quality of courts is a major or very severe obstacle to operating a business. The second is a dummy variable that is similarly defined if a firm sees political instability as a severe problem. Including these two measures allows us to capture some aspects of institutional quality that may be correlated with corruption and, if not controlled for, may therefore bias our results. Finally, d¡ and dc include a full set of industry and country dummies, respectively, and uijc is the idiosyncratic error term, which allows for clustering at the country-industry level. 3.2

Efficient grease: trade-offs between the bribe tax and the time tax

The model in equation (1) can be expanded to investigate the extent to which bribes may be a second best outcome in a context where inefficient bureaucracy leads to a time tax for producers. In other words, when regulation is overly restrictive, corruption may aid entrepreneurs in their interaction with the state, thus leading to a beneficial impact on productivity. A direct way to test this hypothesis would be to include the bribe tax and the time tax jointly in the empirical specification, together with their interaction. The latter would test the extent to which the effect of bribes on productivity is conditional on time consuming dealings with bureaucracy; in other words, it would allow a direct test of the efficient grease hypothesis, as in equation (2): y¡jc = aiKl?"d

+ aixe.^H'Pment

+ /¡3 (corruption¿/c

+

a^Lijc + a4M¿yc + ß\corruption¿;c

* timetax¡jc)

+ ß2timetaxijc

+ yX¡jc + d¡ + dc + e¡jC

+ (2)

A significant coefficient for will indicate that the effect of corruption on productivity depends on the degree to which the firm is engaged in time consuming relations with the state. In particular, a positive coefficient for β3 would indicate that a high time tax is accompanied with less negative - or even positive - effects of corruption on productivity, thus providing evidence in favour of "efficient grease", with corruption helping to mitigate the effects of burdensome regulation. The same result could also be consistent with a setting where bribe-revenue maximizing bureaucrats may use red tape (the time tax) as a screening device to give production licenses to high-productivity firms (Banerjee 1997). In such a model, every firm pays the same amount of bribes while high productivity firms spend more time with bureaucrats. 3.3

Institutional quality: interaction between firm-level and country-level effects

As a further step in our analysis, in order to check whether the effect of corruption on firm level productivity differs depending on country characteristics, we extend equation (1) and interact corruption with country level measures of institutional quality that may be hypothesized to be relevant in determining individual incentives for corrupt behaviour. For this we use two alternative indicators. First is the Transparency International Corruption Perception Index (CPI) for the year 2008, which allows investigating whether the prevalence of corruption at the country level has implications for firm behaviour and, consequently, productivity performance. The CPI captures the perceived levels of publicsector corruption in a given country and is a composite index, drawing on different expert

Corruption and Productivity: Firm-level Evidence · 125

and business surveys. It may be interpreted to reflect the possibility of social effects as described earlier, whereby in a more corrupt environment individual entrepreneurs would have stronger incentives to behave corruptly. The CPI ranges from zero (highly corrupt) to ten (highly clean). It varies across countries and is fixed across sectors for a given country. As can be seen in Table 2, the average levels of productivity, bribe tax and CPI are different between low CPI and high CPI countries 14 . These variations could imply that the impact of bribe tax on productivity could be different for high and low CPI countries. No major differences can be depicted in the level of time tax, however.15 Table 2 Summary statistics by group of countries Non EU

Total

EU

CPI < CPI

CPI > CPI

Mean

Std. Dev.

Mean

Std. Dev.

Mean

Std. Dev

Mean

Std. Dev.

Mean

TFP

0.148

0.916

0.150

0.941

0.144

0.874

0.239

0.983

0.090

0.867

Bribe T a x

0.099

0.298

0.139

0.346

0.030

0.171

0.195

0.396

0.030

0.169

Time T a x

14.528

18.223

14.313

19.819

14.884

15.220

14.669

21.172

14.430

15.851

3.393

1.150

2.863

0.904

4.288

0.949

2.287

0.362

4.191

0.814

CPI Note:

T F P is calculated as a residual from a simple production function, y,¡c = aìKl'"d

4-

Std. Dev.

+

a 3 í_ í ) C

+

ctiMiic + e¡jC

As a robustness check, we also use an alternative measure of institutional quality, namely the World Economic Forum's index of the effectiveness of the legal framework is solving legal disputes, which can be interpreted as a proxy for the ability of formal institutions to enforce contracts and prevent or sanction the occurrence of corrupt practices. This is also a country level index for the year 2 0 0 8 , where increases in the index imply better legal quality. Including either of the indices in the model gives the following equation (3) as yac = « i r f ? + « 2 ^ t P m e n t + +ßcorruptiotijjc + X(cormptionl¡c +YXljc

+ « 4 Mljc + * institutionc)

+ dj + dc + eljc

+ κ institutionc

+ (3)

In equation (3) a significant coefficient for λ will indicate that the effect of corruption on productivity depends on the country's level of institutional quality, as represented by the diffusion of corruption and the efficiency of the legal system. In particular, a positive (negative) coefficient of λ will indicate that high institutional quality will lessen (strengthen) the negative effect of corruption at the firm level on productivity.

14

L o w and high CPI countries are defined as countries with CPI below and above the mean level,

15

T h e use of perceived c o r r u p t i o n (the CPI index by T r a n s p a r e n c y International) m a y be problematic

respectively. if there are discrepancies between perceived and actual c o r r u p t i o n . Olken ( 2 0 0 9 ) s h o w s that this m a y be the case by examining a specific case of a r o a d building project in rural Indonesia. H e c o m p a r e s c o r r u p t i o n perceived by villagers with a m o r e objective measure of c o r r u p t i o n based o n missing expenditure. While a difference between actual and perceived c o r r u p t i o n m a y potentially be a problem, one should keep in mind that T r a n s p a r e n c y International reports broad c o u n t r y level indices which are based o n e x p e r t and business surveys.

126 · Donato De Rosa, Nishaal Gooroochurn, and Holger Görg

4

Econometric analysis

We now turn to the results of the econometric analysis in the three stages outlined above. Namely, we examine the effects of corruption on productivity; of the interaction between corruption and the time tax; and of the relevance of country characteristics for firm level outcomes. As discussed, the determinants of productivity are estimated using a one-step augmented production function. In order to address the potential endogeneity of firm level bribe tax and time tax, equation (1) is estimated with an instrumental variable (IV) approach. Implementation of the IV method requires adequate instruments that fulfil two conditions, namely being correlated with the endogenous variable and being uncorrelated with the error term in equation (1). While it is difficult to find fully excludable instruments, the BEEPS data set offers a number of potential candidates. For instance, firms are asked whether or not the owner is female. This is likely to be correlated with corruption, as there is evidence that women are less likely to engage in or condone corruption (e.g., Swamy et al. 2001). It is also established in the literature that women are more risk averse than men (Barsky et al. 1997), which may be one explanation for why they are less likely to engage in corrupt behaviour, which can be seen as a risky activity. It is not a priori obvious, however, that gender should have a direct influence on productivity that is not related to the indirect channel of risk aversion. Also, the data relate to the owner of the firm, not the manager. While the owner may have more influence on fundamental decision such as whether or not a firm should engage in illegal activities such as corruption, the owner may matter less for operational decisions that affect productivity. Hence, we would argue that this may be a relevant and valid instrument for corruption. An alternative instrumental variables candidate is a firm's reply to a question as to whether they submitted an application for an electricity connection over the last two years. This allows us to generate a dummy variable equal to 1 if the firm did not submit an application and 0 if it did. This variable is arguably likely to be correlated with corruption, since it would imply interaction with public officials who have to grant the firm its right to be connected with the electricity grid. This would, hence, be a good opportunity for corrupt officials to demand a payment, either in cash or in terms of time. On the other hand, a dummy whether or not a firm submitted an application for electricity is unlikely to be correlated with productivity. Since one may assume that a firm needs some access to electricity to start operating, it appears reasonable that the application that is mentioned in the survey must relate to an additional or new connection. There is therefore no a priori reason why, conditional on the covariates in equation (1), there should be a correlation between the incidence of the application and TFP. It is also important to point out that this variable relates to the incidence of the application, not the actual connection to electricity. While this may of course lead to a new connection in the future, which may then possibly (but not necessarily) lead to an increase in productivity, this is unlikely to be the case in the current period. Furthermore, we employ two additional instruments. These are the country-industry averages of bribe tax and time tax. These instruments are also employed by Fisman and Svensson (2007). Firms' experiences and perceptions of corrupt practices or of the burden in terms of time associated with red tape are likely to be influenced by the experiences of other competitors in the same industry. Hence, we would expect our additional instruments to be correlated with the firm-level bribe tax and time tax variables. The necessary

Corruption and Productivity: Firm-level Evidence • 127

assumption for the validity of the instruments is that there is no direct effect of the sectoral average on a firm's level of productivity conditional on the included covariates. This would not be the case if there were processes at the industry level that affect firm level productivity and bribes. An example may be governments favouring sectors that are particularly productive (or unproductive) in their attitude towards corruption. As in Fisman and Svensson (2007), we are not aware of any systematic evidence to support this claim. Hence, we are cautiously confident that our instrument does not just pick up any unobserved industry effects that are correlated with firm level productivity. We based our selection of instruments on initial tests for the validity of overidentification restrictions and relevance of the instruments using standard tests. These tests are in line with the assumption that the dummy whether the owner is female and the industry average are valid instruments in all cases. The dummy capturing whether a firm applied for an electricity connection only appears valid and relevant in the case of the bribe tax, hence, we only use it when looking at the effects of bribes on productivity. It is important to point out, however, that we cannot conclusively test for the validity of instruments. This should be kept in mind in the interpretation of results. The results reported in Table 3 are, hence, based on using a female dummy, electrical connection application dummy, and industry level bribes, as instruments for bribe tax. For time tax, we do not use the electrical connection dummy, only the female dummy and industry level time tax as instruments. We report tests for the relevance of the instruments in these specifications, using a joint F test to verify whether the instrument candidates are correlated with the endogenous variable in Table 3. The F-statistics are higher than 20 in both cases confirming that the instruments are jointly highly correlated with the respective firm level corruption variable. Furthermore, we provide a Hansen-Sargan J test of overidentification restrictions to check that the IV candidates are uncorrected with the error term in equation (1). The p-values of the Hansen-Sargan test confirm the validity of the chosen IV, as we cannot reject the null of instrument validity. The second stage results of the effects of time and bribe taxes on productivity according to the baseline estimation of equation (1) are presented in Table 3. Columns (1) and (2) present the results using an OLS estimator, while columns (3) and (4) show IV estimates. 16 Note firstly that the coefficient on the production factors capital, labour, land and materials are all positive as expected. Furthermore, exporters and foreign-owned firms are more productive, ceteris paribus, as expected. Strikingly, larger firms tend to be less productive, perhaps a sign of incomplete restructuring that prevents firms from exploiting the benefits of scale economies. Whereas innovation would be expected to be associated with higher productivity, the innovation dummy appears as insignificant in all specifications. This may indicate that the innovation activities carried our within firms may be insufficient to have an impact on productivity. This result could indicate a prevalence of defensive as opposed to strategic restructuring by the firms in the sample, where the former is related to short-term cost-cutting measures, while the latter is focused on increasing the longterm efficiency and viability of the firm, by investing in labour training, fixed assets and

16

In some applications using firm level data it is found that extreme observations, or outliers, tend to shape the results, see Verardi and Wagner ( 2 0 1 1 ) for a discussion. In order to see whether this is an issue for our data, we re-estimated columns (1) and (2) using a robust regression estimator which takes account of outliers. W e find that this does not alter much the coefficients in the model. The only exceptions are the innovation and competition variables, which are statistically significant when using robust regression. The results are not reported here to save space, but can be obtained from the authors upon request.

128 • Donato De Rosa, Nishaal Gooroochurn, and Holger Görg

Table 3 Productivity regression results: baseline specifications Dependent Variable: Log Output Bribe Tax

(1)

OLS

-0.002 (0.005)

Time Tax L M ^Equipment

((Land

Exporter Age SizeM SizeL Foreign Owned Innovator Competition Courts Political Stability Constant Country Dummies Sector Dummies

(2)

0.281 (0.039)*** 0.353 (0.027)*** 0.062 (0.015)*** 0.012 (0.011) 0.017 (0.004)***

(3)

IV

(4)

-0.051 (0.019)*** 0.000 (0.000) 0.273 (0.041)*** 0.350 (0.028)*** 0.068 (0.016)*** 0.020 (0.012)* 0.016 (0.004)***

0.285 (0.043)*** 0.354 (0.028)*** 0.063 (0.016)*** 0.008 (0.012) 0.019 (0.004)*** -0.000 -0.000 -0.000 (0.000) (0.000) (0.000) -0.036 -0.037 -0.033 (0.008)*** (0.008)*** (0.009)*** -0.019 -0.019 -0.017 (0.005)*** (0.005)*** (0.005)*** 0.012 0.012 0.015 (0.005)*** (0.005)** (0.005)** 0.006 0.006 0.006 (0.004) (0.004) (0.004) -0.002 -0.001 -0.000 (0.001) (0.001) (0.001) -0.003 -0.005 0.004 (0.004) (0.004) (0.005) 0.004 0.003 0.006 (0.004) (0.004) (0.004)* YES YES YES YES YES YES YES YES YES

-0.001 (0.001) 0.274 (0.042)*** 0.355 (0.029)*** 0.069 (0.015)*** 0.015 (0.012) 0.017 (0.004)*** -0.000 (0.000) -0.038 (0.009)*** -0.018 (0.005)*** 0.011 (0.005)** 0.004 (0.004) -0.001 (0.001) -0.003 (0.004) 0.003 (0.004) YES YES YES

R squared Wu-Hausman (p-value) Hansen J (p-value) F-Stat

0.86

0.87

0.77 0.01 0.59 43.49

0.78 0.11 0.45 27.51

Observations

1666

1519

1629

1490

Noie: Standard errors clustered by country-industry in brackets. * significant at 10%; ** significant at 5%; *** significant at 1 %. Instruments: bribe tax: country-industry bribe tax, dummy for female owner, dummy for electricity connection application. Time tax: country-industry time tax, dummy for female owner. * significant at 10%; ** significant at 5%; *** significant at 1%

Corruption and Productivity: Firm-level Evidence · 129

other innovation related activities such as R & D (Grosfeld/Roland 1997; Frydman et al. 1999). The other controls are statistically insignificant. Examination of the OLS results in columns 1 and 2 shows that the coefficients on both time and bribe tax are statistically insignificant. It is, however, unlikely that the corruption variables are exogenous in this productivity estimation. For example, highly productive firms may have a better ability to engage in bribing or may be preferred targets of bureaucrats aiming at exacting bribes. This would introduce reverse causality in the equation or, more formally, a correlation between the right-hand-side variable and the error term. Another potential source of endogeneity is the impact of unobserved institutional characteristics at the firm level. We argue that our measures of perception of the quality of courts and political instability go some way to address these concerns. The Wu-Hausman test is performed to check whether bribe is endogenous and the results are given at the bottom of Table 3. The significant p-value rejects the null hypothesis of exogeneity of bribe tax. This is not the case for time tax, however, where we cannot reject exogeneity. However, in both cases we implement an instrumental variables (IV) technique to estimate equation (1) to check the implications this has for the coefficient on corruption. The bribe tax has a negative and significant effect on productivity when adjusting for potential endogeneity bias (Table 3, column 3). The negative and significant coefficient of bribe tax indicates that firms that pay bribes to officials experience lower productivity than other firms. The size of the coefficient suggests that a firm that pays bribes is on average around 5 percent less productive than a non-corrupt firm.17 While not directly comparable, it is interesting to note that Fisman and Svensson (2005) find that bribes reduce firm growth by around 3 percent. The table also shows that, none of the controls for institutional quality are significantly correlated with firm-level productivity. We also still fail to find a statistically significant impact of time tax on firm level productivity (Column 3). 4.1

Trade-offs between the bribe tax and the time tax

The preceding analysis has shown that the payment of bribes is negatively associated with the productivity of the bribing firm, time spent dealing with bureaucratic requirements per se appears to be irrelevant. However, it has been argued that the occurrence of corruption may not be independent of the length of bureaucratic processes. These may, in fact, be deliberately established by state officials with the intent of exacting bribes. In this context, the payment of bribes might help "grease the wheels of commerce" by speeding up bureaucratic requirements, as captured by the time tax, and lead to a second best outcome for the bribing firm. The challenge is, therefore, to examine whether the (negative) effect of bribes on productivity is somehow dependent on the time that firms have to spend dealing with red tape. A direct way to investigate the trade-offs between bribes and red tape and to test whether such trade-offs are of the "efficient grease" type, is to include the bribe tax and the time tax jointly in the empirical specification, together with their interaction. Examination of the sign, significance and magnitude of the coefficient on the interaction term would allow drawing conclusions on the nature of the relationships between bribe payments

17

Calculated as e x p ( - 0 . 0 5 1 ) - l

130 · Donato De Rosa, Nishaal Gooroochurn, and Holger Görg

Table 4 Trade-off between the bribe tax and the time tax Dependent Variable: Log Output Bribe tax Time tax

(1) OLS 0.001 (0.006) 0.000 (0.000)

(2) IV -0.045 (0.020)** -0.001 (0.001)

0.272 (0.041)*** 0.350 (0.028)*** 0.068 (0.016)***

0.071 (0.015)***

0.001 (0.008) 0.000 (0.000) 0.000 (0.000) 0.272 (0.041)*** 0.350 (0.028)*** 0.068 (0.016)***

0.020 (0.012)* 0.016 (0.004)*** -0.000 (0.000) -0.038 (0.008)*** -0.019 (0.005)*** 0.012 (0.005)** 0.006 (0.004) -0.001 (0.001) -0.005 (0.004) 0.003 (0.004) YES YES YES

0.014 (0.013) 0.018 (0.004)*** -0.000 (0.000) -0.035 (0.009)*** -0.017 (0.005)*** 0.014 (0.006)** 0.005 (0.004) 0.000 (0.001) 0.003 (0.006) 0.005 (0.004) YES YES YES

0.020 (0.012)* 0.016 (0.004)*** -0.000 (0.000) -0.038 (0.008)*** -0.019 (0.005)*** 0.012 (0.005)** 0.006 (0.004) -0.001 (0.001) -0.005 (0.004) 0.003 (0.004) YES YES YES

Time tax * Bribe tax L M ^Equipment I^Land

Exporter Age SizeM SizeL Foreign Owned Innovator Competition Courts Political Stability Constant Country Dummies Sector Dummies

(3) OLS

0.278 (0.045)*** 0.352 (0.029)***

(4) IV -0.090 (0.071) -0.001 (0.001) 0.004 (0.005) 0.271 (0.044)*** 0.352 (0.029)*** 0.076 (0.018)*** 0.016 (0.014) 0.019 (0.005)*** -0.000 (0.000) -0.036 (0.010)*** -0.017 (0.006)*** 0.011 (0.007) 0.005 (0.004) 0.000 (0.001) 0.001 (0.007) 0.003 (0.005) YES YES YES

R squared Wu-Hausman [p-value] Hansen J [p-value] F-Stat [time tax] F-Stat [bribe tax] F-Stat [time tax * bribe tax]

0.87

0.78 0.03 0.69 11.26 29.48

0.87

0.77 0.05 0.89 11.26 29.48 6.90

Observations

1519

1490

1519

1490

Note: Standard errors clustered by country-industry in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%. Note: Instruments used in column (2) are the same as in Table 4. In column (4), we include the interaction of industry level bribe tax and industry level time tax as additional instrument in order to be able to test for overidentification restrictions using the Hansen J test.

Corruption and Productivity: Firm-level Evidence · 131

and red tape in our sample and to verify the extent to which the effect of corruption on productivity is conditional on time consuming dealings with bureaucracy. Table 4 shows that the interaction of time tax and bribe tax is insignificant, failing to provide evidence of a link between inefficient bureaucracy, corruption and productivity. The time tax remains statistically insignificant, whereas the effect of the bribe tax for productivity remains negative but becomes statistically insignificant when including the interaction term. This suggests that the specification with the interaction term does not fit the data well. The coefficient for factor inputs and other control variables remain largely unaltered compared to Table 3. 4.2

Does the institutional environment matter?

An interesting question that can be answered with our data is whether there are any systematic variations in the effects of corruption on productivity across groups of countries. In order to do so, we explicitly consider the potential influence of broader institutional characteristics of the country. For this purpose, two variables, obtained from sources other than the BEEPS, are used. As discussed above, the first is the Corruption Perception Index computed by Transparency International. The second is a measure of the quality of the legal framework taken from the Global Competitiveness Report. As shown in Table A3 in the Appendix, mirroring the large differences in income per capita, these two variables also present substantial variation across the countries in the sample. We, therefore, posit that the effect of bribe and time tax on productivity may depend on the overall prevalence of corruption in the country - in the spirit of theories highlighting the role of social effects on individual behaviour - and on the effectiveness of the legal framework in preventing and sanctioning corrupt behaviour. This hypothesis is tested by estimating model (3) and the results are given in Tables 5 and 6. Overall, our results indicate that the relationship between corruption and economic performance is conditional on the overall level of institutional quality. In particular, the coefficients on time tax and its interaction are statistically insignificant, the coefficient of bribe tax is still negative and significant in all specifications, whereas the interactive term, bribe χ institution, is positive and significant for both the CPI and the quality of the legal framework. 18 For Table 6, the coefficient estimates imply that for the average firm in a country with the average level of CPI, the effect of a bribe is an increase in productivity by roughly 5 percent. For the same firm in a country with a CPI at one standard deviation less than the mean, the effect of bribery is to reduce productivity by 8 percent." This implies that in highly corrupt environments (i.e. for lower values of the CPI) or in countries with low levels of legal quality, bribes have higher negative impact on productivity, in line with Meon and Sekkat (2005). At the same time, as the value of the CPI or the legal quality indictor increases, the total effect of bribe on productivity becomes less negative and, beyond a certain threshold, could even be positive. This could be because, in an environment that is generally free of corruption, paying a bribe might result in a 18

As a robustness check, we also estimated the model using labour productivity (output per worker) as dependent variable. In this case, the results for bribe tax and the interaction with CPI hold, however, the coefficients of bribe tax and the interaction with the legal framework indicator are statistically insignificant. Results are available from the authors upon request. " The mean level of CPI is 3.393, the standard deviation is 1.151, see Table 2. T h e first result is calculated as e x p ( - 0 . 3 4 5 + 0.116*3.393) - 1; the second as e x p ( - 0 . 3 4 5 + 0.116*2.232).

132 • Donato De Rosa, Nishaal Gooroochurn, and Holger Görg Table 5 Productivity regression results: Country-level corruption Dependent Variable: Log Output Bribe tax Bribe tax * CPI

(1)

OLS

0.009 (0.016) -0.004 (0.005)

M [(Equipment ((Land

Exporter Age SizeM SizeL Foreign Owned Innovator Competition Courts Political Stability Constant Country Dummies Sector Dummies

IV

(4)

-0.345 (0.186)* 0.116 (0.065)*

-0.010 (0.004)** 0.281 (0.039)*** 0.353 (0.027)***

-0.067 (0.029)** 0.301 (0.058)*** 0.367 (0.035)***

0.000 (0.005) -0.000 (0.001) -0.026 (0.025) 0.272 (0.041)*** 0.352 (0.030)***

0.061 (0.015)***

0.068 (0.016)***

0.088 (0.025)***

0.067 (0.015)***

0.013 (0.011) 0.017 (0.004)*** -0.000 (0.000) -0.037 (0.008)*** -0.019 (0.005)*** 0.012 (0.005)** 0.006 (0.004) -0.002 (0.001) -0.003 (0.004) 0.005 (0.004) YES YES YES

0.020 (0.012)* 0.016 (0.004)*** -0.000 (0.000) -0.037 (0.008)*** -0.019 (0.005)*** 0.012 (0.005)** 0.006 (0.004) -0.001 (0.001) -0.005 (0.004) 0.003 (0.004) YES YES YES

-0.015 (0.023) 0.023 (0.007)*** -0.000 (0.000) -0.019 (0.013) -0.008 (0.008) 0.020 (0.007)*** 0.004 (0.004) -0.001 (0.002) 0.008 (0.008) 0.005 (0.004) YES YES YES

0.020 (0.015) 0.016 (0.004)***

Time tax * CPI

L

(3)

0.000 (0.000) -0.000 (0.000) -0.009 (0.005)* 0.272 (0.041)*** 0.350 (0.028)***

Time tax

cpi_2008

(2)

-0.000

(0.000) -0.037 (0.008)*** -0.018 (0.005)*** 0.011 (0.006)** 0.005 (0.005) -0.001 (0.001) -0.004 (0.004) 0.003 (0.004) YES YES YES

R squared Wu-Hausman [p-value] Hansen J [p-value] F-Stat

0.86

0.87

0.72 0.05 0.54 1.99

0.80 0.99 0.93 3.49

Observations

1666

1519

1252

1490

Note: Standard errors clustered by country-industry in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%. Instruments: as in Table 3, additionally; country-industry time tax; country-industry bribe tax.

Corruption and Productivity: Firm-level Evidence • 133 Table 6 Productivity regression results: Quality of the legal framework Dependent Variable: Log Output Bribe tax Bribe tax * LF

(1)

OLS

-0.033 (0.035) 0.010 (0.012)

M «Equipment «Land

Exporter Age SizeM SizeL Foreign Owned Innovator Competition Courts Political Stability Constant Country Dummies Sector Dummies

IV

(4)

-1.506 (0.787)* 0.503 (0.271)* -0.005 (0.015) 0.002 (0.005) -0.092 (0.085) 0.235 (0.053)*** 0.383 (0.037)***

0.063 (0.016)*** 0.007 (0.013) 0.018 (0.004)*** -0.000 (0.000) -0.040 (0.008)*** -0.020 (0.005)*** 0.014 (0.005)*** 0.006 (0.004) -0.002 (0.001)* -0.001 (0.004) 0.005 (0.004) YES YES YES

0.016 0.009 (0.013) (0.016) 0.017 0.011 (0.004)*** (0.007)* -0.000 0.000 (0.000) (0.000) -0.042 -0.032 (0.008)*** (0.011)*** -0.021 -0.017 (0.005)*** (0.007)** 0.014 0.027 (0.005)*** (0.010)*** 0.006 0.007 (0.004) (0.005) -0.002 -0.000 (0.001) (0.002) -0.003 0.004 (0.004) (0.007) 0.003 0.008 (0.004) (0.005) YES YES YES YES YES YES

0.013 (0.015) 0.018 (0.006)*** -0.000 (0.000) -0.035 (0.012)*** -0.016 (0.007)** 0.022 (0.006)*** 0.006 (0.005) -0.002 (0.001)* 0.000 (0.005) 0.003 (0.005) YES YES YES

Time tax * LF

L

(3)

-0.000 (0.001) 0.000 (0.000) 0.026 -0.130 (0.009)*** (0.144) 0.247 0.285 (0.043)*** (0.054)*** 0.355 0.333 (0.032)*** (0.042)*** 0.071 0.080 (0.016)*** (0.025)***

Time tax

Legal framework

(2)

0.026 (0.008)*** 0.261 (0.042)*** 0.357 (0.031)***

0.070 (0.020)***

R squared Wu-Hausman [p-value] Hansen J [p-value] F-Stat

0.87

0.87

0.55 0.00 0.85 2.52

0.80 0.72 0.68 2.75

Observations

1504

1363

1470

981

Note: Standard errors clustered by country-industry in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%. Instruments: as in Table 3, additionally country-industry time tax; country-industry bribe tax.

134 · Donato De Rosa, Nishaal Gooroochurn, and Holger Görg

competitive advantage, perhaps reflected in a marginal gain in firm level productivity. On the other hand, in a highly corrupt environment, social effects of the type modelled by Tiróle (1996) may induce most market players to pay a bribe. Hence there would be no competitive edge or gain in productivity to be obtained by paying a bribe. Quite the opposite, paying more bribes allocates resources away from their most productive use, reducing productivity of the firm. Hence productivity gains are more likely to incur to the firms that do not bear the cost of bribes. 5

Concluding remarks

Corruption is often identified as one of the primary causes for the underdevelopment of many economies. Nevertheless, some see corruption as a necessary evil that, by "greasing the wheels of commerce", mitigates the negative effects of inefficient bureaucracy, which may be seen as imposing a "time tax" on individuals and firms. At the same time, both the incentives for - and the impact of - corruption may be different across countries, depending on the nature of the surrounding environment, namely on the diffusion of corruption and on the ability of the legal system to sanction corrupt behaviour. Based on these premises, this study investigates the effect of corruption - interpreted as a "bribe tax" - on firm-level productivity across a diverse sample of countries in Central and Eastern Europe and the Former Soviet Union. The findings of econometric analysis corroborate the hypothesis that corruption has, on balance, negative consequences for enterprise performance. However, the relationship between corruption and economic performance presents some nuances. First, a comparison of the effects of the bribe tax and the time tax indicates that only bribery negatively affects firm productivity, while lengthy bureaucratic requirements per se have no significant consequences. Second, an explicit test of the hypothesis that bribes help to mitigate the negative effects of time consuming dealings with bureaucracy does not find confirmation in our data. Contrary to previous studies addressing the same question at the country level, our conclusion is based on a precise definition of the institutional inefficiencies that corruption is supposed to "grease" -namely lengthy bureaucratic requirements - rather than generic measures of "governance". Third, broader environmental circumstances turn out to play a significant role in determining the impact of firm level corruption on productivity. Results indicate that in highly corrupt environments and where the legal framework is weaker, firms that do not pay bribes are more productive. Furthermore, as the level of institutional quality decreases, the total effect of corruption is increasingly negative. This suggests that, whereas environmental circumstances are beyond the choice set of individual firms, managers retain some degree of autonomy in deciding whether to recur to bribery or not and this affects enterprise performance.

Corruption and Productivity: Firm-level Evidence • 135

Appendix Table A1 Variable definitions Variable

Definition

TFP

Total factor productivity, calculated as residual from a production function

Y

This establishment's total annual sales

Κ

This establishment spend on purchases of machinery, land and building

L

Total annual cost of labour (including wages, salaries, bonuses, social security payments)

M

Total annual cost of raw materials and intermediate goods used in production

Time Tax

Percentage of time spent by senior management with public officials in order to obtain favourable interpretation of regulations

Bribe Tax

Dummy = 1 if firm replies frequently, usually or always to the question "it is common to have to pay some irregular additional payment or gifts to get things done".

Courts

Dummy = 1 if firm replies that courts are a major obstacle or very severe obstacle to the operations of the firm

Political stability

Dummy = 1 if firm replies that political instability is a major obstacle or very severe obstacle to the operations of the firm

CPI

Corruption Perception Index at the country level. It relates to perceptions of the degree of corruption as seen by business people and country analysts, and ranges between 10 (highly clean) and 0 (highly corrupt).

Legal Framework

Indicator from the Global Competitiveness Report at the country level. It provides a measure of the efficiency of the legal framework in settling disputes (1 = extremely inefficient; 7 = highly efficient).

Exporter

Dummy = 1 if firm has positive exports

Foreign_Owned

Dummy = 1 if firm has foreign ownership

Innovator

Dummy = 1 if firm has positive expenditure on R&D

Age

Age of firm (years)

SizeS

Dummy = 1 for small firm (less than 20 employees)

SizeM

Dummy = 1 for medium firm (between 20 and 99 employees)

SizeL

Dummy = 1 for large firm (larger than 99 employees)

Competition

Difference between firm's perception and country level average on question "competition is an obstacle for operations of the establishment" (ranked between 0 and 4)

136 · Donato De Rosa, Nishaal Gooroochurn, and Holger Görg

Table A2 Descriptive statistics of variables used Variable

Obs

Mean

TFP Time tax Bribe Tax Courts Political stability CPI Legal Framework Exporter Foreign_Owned Innovator Age SizeM SizeL Competition

1666 1519 1666 1666 1666 1666 1504 1666 1666 1666 1666 1666 1666 1666

0.22 12.81 0.11 0.19 0.32 3.57 2.98 0.49 0.11 0.21 20.34 0.29 0.37 0.04

Std. Dev. 0.90 14.79 0.31 0.39 0.46 1.29 0.51 0.50 0.30 0.41 20.33 0.45 0.48 1.37

Table A3 Descriptive statistics of variables used by country

Albania Belarus Georgia Tajikistan Ukraine Uzbekistan Russia Poland Romania Serbia Kazakhstan Moldova Bosnia Azerbaijan FYROM Armenia Kyrgyz Estonia Czech Rep Hungary Latvia Lithuania Slovakia Slovenia Bulgaria Croatia Montenegro

TFP Mean Std. Dev.

Bribe Tax Mean Std. Dev.

Time Tax Mean Std. Dev.

CPI Mean Std. Dev.

0.85 0.12

0.03 0.09 0.03 0.13 0.21 0.36 0.23 0.04 0.18 0.10 0.20 0.09 0.09 0.21 0.08 0.08 0.19 0.02 0.03 0.03 0.02 0.00 0.07 0.01 0.01 0.00 0.03

20.38 19.00 2.00 14.98 15.01 12.77 20.17 14.14 13.50 17.24 6.11 10.49 14.02 2.66 14.80 13.95 4.03 8.32 13.46 18.64 8.74 10.37 7.38 9.21 17.23 13.17 9.85

3.40 2.00 3.90 2.00 2.50 1.80 2.10 4.60 3.80 3.40 2.20 2.90 3.20 1.90 3.60 2.90 1.80 6.60 5.20 5.10 5.00 4.60 5.00 6.70 3.60 4.40 3.40

0.22

0.36 0.25 0.42 0.02 -0.01 0.19 -0.06 0.08 0.47 -0.21 0.03 0.20 0.38 0.86 0.02 0.85 0.28 -0.10 0.05 -0.04 -0.25 0.20 -0.19 0.28

0.95 0.83 0.89 0.87 1.11 1.04 0.83 0.84 0.92 0.92 0.79 0.96 0.69 0.64 0.65 0.97 1.46 0.87 1.77 0.91 0.70 0.65 0.74 0.74 0.61 0.73 1.12

0.17 0.29 0.18 0.33 0.41 0.48 0.42 0.21 0.38 0.30 0.40 0.29 0.29 0.41 0.27 0.27 0.40 0.15 0.18 0.18 0.15 0.00 0.26 0.10 0.07 0.06 0.16

22.56 20.20 6.84 19.19 21.93 13.84 25.66 13.77 17.95 18.62 11.12 17.24 17.63 4.95 15.36 16.90 8.03 10.74 14.67 18.71 12.88 12.16 12.44 9.26 15.20 15.67 11.33

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Legal Framework Mean Std. Dev. 3.60

0.00

3.20 3.40 2.30

0.00 0.00 0.00

2.90 2.80 2.90 2.60 3.40

0.00 0.00 0.00 0.00 0.00

1.80 3.80 3.20 2.80 2.60 4.20 3.40 3.10 3.10 3.50 3.00 4.00 2.80 2.60 3.90

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Corruption and Productivity: Firm-level Evidence

137

References Acemoglu, D. (2006), A Simple Model of Inefficient Institutions. Scandinavian Journal of Economics 108(4): 5 1 5 - 5 4 6 . Acemoglu, D., S. Johnson (2005), Unbundling Institutions. Journal of Political Economy 113(5): 949-995. Acemoglu, D., J.A. Robinson, S. Johnson (2001), T h e Colonial Origins of Comparative Development: An Empirical Investigation. American Economic Review. December, 9 1 : 1 3 6 9 - 1 4 0 1 . Acemoglu, D., Th. Verdier (1998), Property Rights, Corruption and the Allocation of Talent: A General Equilibrium Approach. The Economic Journal 108: 1 3 8 1 - 1 4 0 3 . Ades, Α., R. Di Telia (1999), Rents, Competition, and Corruption. American Economic Review 89(4): 9 8 2 - 9 9 3 . Aidt, T.S., J. Dutta (2008), Policy Compromises: Corruption and Regulation in a Democracy. Economics and Politics 20(3): 3 3 5 - 3 6 0 . Alesina, Α., S. Ardagna, G. Nicoletti, F. Schiantarelli (2005), Regulation and investment. Journal of the European Economic Association 3(4): 7 9 1 - 8 2 5 . Aw, B.Y., M.J. Roberts, D.Y. X u (2008), R & D Investments, Exporting and the Evolution of Firm Productivity. American Economic Review 98(2): 451—456. Banerjee, A.V. (1997), A Theory of Misgovernance. Quarterly Journal of Economics 112(4): 1289-1332. Bardhan, P. (1997), Corruption and Development: A Review of Issues. Journal of Economic Literature 35(3): 1 3 2 0 - 1 3 4 6 . Barsky, R.B., F.Th. Juster, M.S. Kimball, M . D . Shapiro (1997), Preference Parameters and Behavioral Heterogeneity: An Experimental Approach in the Health and Retirement Study. Quarterly Journal of Economics 112(2): 5 3 7 - 5 7 9 . Baumol, W.J. (1990), Entrepreneurship: Productive, Unproductive, and Destructive. The Journal of Political Economy 98(5): 8 9 3 - 9 2 1 . Blagojevic, S., J.P. Damijan (2013), The impact of corruption and ownership on the performance of firms in Central and Eastern Europe. Post-Communist Economies 25(2): 1 3 3 - 1 5 8 . Brunetti, Α., Β. Weder (2003), A Free Press is Bad N e w s for Corruption. Journal of Public Economics 87(7-8): 1 8 0 1 - 1 8 2 4 . De Rosa, D., N . G o o r o o c h u r n , H . Görg (2010), Corruption and Productivity: Firm-level Evidence f r o m the BEEPS Survey World Bank Policy Research W o r k i n g Paper 5348. Djankov, S., R. La Porta, F. Lopez-De-Silanes, A. Shleifer (2002), T h e Regulation of Entry. Quarterly Journal of Economics 117(1): 1 - 3 7 . Egger, P.H., H . Winner (2005), Evidence on Corruption as an Incentive for Foreign Direct Investment. European Journal of Political Economy 21: 9 3 2 - 5 2 . Fisman, R., J. Svensson (2007), Are corruption and taxation really harmful to growth? Firm level evidence. Journal of Development Economics 83(1): 6 3 - 7 5 . Fries, St., T. Lysenko, S. Polenac (2003), The 2 0 0 2 Business Environment and Enterprise Performance Survey: Results f r o m a survey of 6,100 firms. EBRD, London. Frydman, R., Ch. Gray, M . Hessel, A. Rapacyinski (1999), W h e n Does Privatisation work? T h e Impact of Private Ownership on Corporate Performance in the Transition Economies. Quarterly Journal of Economics 114(4): 1 1 5 3 - 1 1 9 1 . Girma, S., H . Görg (2007), Multinationals' Productivity Advantage: Scale or Technology? Economic Inquiry 45(2): 3 5 0 - 3 6 2 . Glaeser, E.L., R. La Porta, F. Lopez-de-Silanes, A. Shleifer (2004), Do Institutions Cause Growth? Journal of Economic G r o w t h 9(3): 2 7 1 - 3 0 3 . Grosfeld, I., G. Roland (1997), Defensive and Strategic Restructuring in Central European Enterprises. Journal of Transforming Economies and Societies 3: 8 3 - 1 1 6 . Hellman, J.S., G. Jones, D. K a u f m a n n (2003), Seize the State, Seize the Day: State Capture, Corruption and Influence in Transition. Journal of Comparative Economics 31(4): 7 5 1 - 7 7 3 . Johnson, S., J. McMillan, Ch. W o o d r u f f (2002), Property Rights and Finance. American Economic Review 92(5): 1 3 3 5 - 1 3 5 6 .

138 • Donato De Rosa, Nishaal Cooroochurn, and Holger Görg

Kaufmann, D., S.-J. Wei (1999), Does grease money speed up the wheels of commerce? NBER Working Paper No. 7093. Klette, T.J., S. Kortum (2004), Innovating Firms and Aggregate Innovation. Journal of Political Economy 112(5): 986-1018. Krusell, P., J.-V. Rios-Rull (1996), Vested interests in a positive theory of stagnation and growth. Review of Economic Studies 63: 301-329. Lambsdorff, J. Graf (2003), How corruption affects productivity. Kyklos 56: 457—474. La Porta, R., F. Lopez-de-Silanes, A. Shleifer, R.W. Vishny (1998), Law and Finance. Journal of Political Economy 106: 1113-1155. La Porta, R., F. Lopez-de-Silanes, A. Shleifer, R.W. Vishny (1999), The Quality of Government. Journal of Law, Economics and Organization 15(1): 222-279. Levinsohn, J., A. Petrin (2003), .Estimating production functions using inputs to control for observables.. Review of Economic Studies 70: 317-341. Lui, F. (1985), An Equilibrium Queuing Model of Bribery. Journal of Political Economy 93(4): 760-781. Méndez, F., F. Sepúlveda (2006), Corruption, Growth and Political Regimes: Cross-country Evidence. European Journal of Political Economy 22(1): 82-98. Méon, P.-G., K. Sekkat (2005), Does Corruption Grease or Sand the Wheels of Growth? Public Choice 122: 69-97. Méon, P.-G., L. Weill (2010), Is Corruption an Efficient Grease? World Development 38(3): 244-259. Murphy, K., A. Shleifer, R.W. Vishny (1991), The Allocation of Talent: Implications for Growth. Quarterly Journal of Economics 106: 503-530. Murphy, K., A. Shleifer, R.W. Vishny (1993), Why is Rent-seeking so Costly to Growth? American Economic Review 83(2): 4 0 9 ^ 1 4 . North, D.C. (1990), Institutions, Institutional Change and Economic Performance. Cambridge University Press, Cambridge. Olken, B.A. (2009), Corruption perceptions vs. corruption reality. Journal of Public Economics 93(7-8): 950-964. Shleifer, Α., RW. Vishny (1998), The Grabbing Hand: Government Pathologies and Their Cures. Harvard University Press, Cambridge, MA. Svensson, J. (2003), Who Must Pay Bribes and How Much? Quarterly Journal of Economics 118(1): 207-230. Svensson, J. (2005), Eight Questions about Corruption. Journal of Economic Perspectives 19(3): 3-28. Swamy, Α., St. Knack, Y. Lee, O. Azfar (2001 ), Gender and corruption. Journal of Development Economics 64: 25-55. Tiróle, J. (1996), A Theory of Collective Reputations (with Applications to the Persistence of Corruption and to Firm Quality). The Review of Economic Studies 63(1): 1-22. Verardi, V., J. Wagner (2011), Robust Estimation of Linear Fixed Effects Panel Data Models with an Application to the Exporter Productivity Premium. Jahrbücher für Nationalökonomie und Statistik 231: 546-557. Wagner, J. (2007), Exports and Productivity: A Survey of the Evidence from Firm Level Data. The World Economy 30(1): 60-82. Donato De Rosa, Ph.D., Financial and Private Sector Development, Vice Presidency, The World Bank, Washington D.C., USA. [email protected] Nishaal Gooroochurn, Ph.D., London Metropolitan University, London, UK. [email protected] Prof. Holger Görg, Ph.D., Institut für Weltwirtschaft, Kiellinie 66, 24105 Kiel, Germany. [email protected]

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2015) Bd. (Vol.) 2 3 5 / 2

Bribing Behaviour and Sample Selection: Evidence from Post-Socialist Countries and Western Europe Artjoms Ivlevs and Timothy Hinks* University of the West of England, Bristol, UK JEL C13; D73; P2 Bribing; sample selection; transition economies.

Summary We study the individual-level determinants of bribing public officials. Particular attention is paid to the issue of respondents' n o n - r a n d o m selection into contact with public officials, which may result in biased estimates. Data come f r o m the 2 0 1 0 Life in Transition Survey, covering 30 post-socialist and five Western European countries. The results suggest that the elderly tend to be less likely to bribe public officials, while people with higher income and, especially, low trust in public institutions are more likely to bribe. Several determinants of bribery - ethnic minority status, the degree of urbanisation, social trust - are context specific, i.e. they change signs or are statistically significant according to the geographical region or the type of public official. The results show that not accounting for sample selection effects may produce a bias in estimated coefficients.

1

Introduction

T h e r e is a general a g r e e m e n t a m o n g a c a d e m i c s a n d p o l i c y m a k e r s t h a t c o r r u p t i o n leads t o i n f e r i o r s o c i o - e c o n o m i c o u t c o m e s . A large a n d well-established b o d y of l i t e r a t u r e suggests t h a t , a t c o u n t r y level, higher c o r r u p t i o n is a s s o c i a t e d w i t h l o w e r e c o n o m i c g r o w t h ( M a u r o 1 9 9 5 ; E v e r h a r t et al. 2 0 0 9 ; J o h n s o n et al. 2 0 1 1 ) , l o w e r p r o d u c t i v i t y ( L a m b s d o r f f 2 0 0 3 ) , h i g h e r inequality a n d p o v e r t y ( G u p t a et al. 2 0 0 2 ; J o n g - s u n / K h a g r a m 2 0 0 5 ) , a n d l o w e r i n t e r n a t i o n a l t r a d e a n d f o r e i g n i n v e s t m e n t (Wei 2 0 0 0 ; C u e r v o - C a z u r r a 2 0 0 6 ; E g g e r / W i n n e r 2 0 0 6 ; J o n g / B o g m a n s 2 0 1 1 ). In this c o n t e x t , f i g h t i n g c o r r u p t i o n h a s b e c o m e a m a j o r p r e o c c u p a t i o n f o r g o v e r n m e n t s in m a n y d e v e l o p i n g , t r a n s i t i o n a n d , in c e r t a i n cases, d e v e l o p e d c o u n t r i e s a c r o s s t h e w o r l d . ' T o deal w i t h c o r r u p t i o n , o n e m u s t k n o w its causes. T h e s e causes a r e likely t o be b o t h ' e x t e r n a l ' a n d ' i n t e r n a l ' t o individuals involved in c o r r u p t e x c h a n g e s . ' E x t e r n a l ' f a c t o r s

* The authors would like to thank two anonymous referees for their helpful comments and suggestions. The paper also benefited from comments made by participants of the Economics Internal Seminar Series at the University West of England. 1 For a survey of literature on the causes and consequences of corruption, see Lambsdorff (2006) and Kis-Katos and Schulze (2013).

140 · Artjoms Ivlevs and Timothy Hinks

associated with corruption 2 include a slower pace of structural reforms (Iwasaki/Suzuki 2012), less freedom of the press (Brunetti/Weder 2003), a lower ratio of government to manufacturing wages (Van Rijckeghem/Weder 2001), a lower share of women in the labour force (Swamy et al. 2001), a lower share of Protestants in the population (Treisman 2000) and higher levels of official development aid (Ali/Isse 2003). 'Internal' causes of corruption are also of interest, as they help the policymakers understand why, within a particular country, some people are more likely to engage in corrupt acts than others. The availability of large-scale survey data has recently led to a number of empirical papers on the individual-level determinants of corruption behaviour and corruption attitudes (Swamy et al. 2001; Gatti et al. 2003; Mocan 2008; Guerrero/RodriguezOreggia 2008; Hunt 2007; Hunt/Laszlo 2012; Tavits 2011; Truex 2011). This literature suggests, for example, that wealthier people are more prone to paying bribes (Guerrero/Rodriguez-Oreggia 2008; Hunt/Laszlo 2012), being asked for bribes by public officials (Mocan 2008) and find corrupt behaviour justifiable (Gatti et al. 2003). This paper continues the exploration of corruption using survey data and makes several contributions to the literature. First, it addresses a methodological gap in the empirical literature - an issue of potential sample selection bias which arises from a non-random selection, based on respondents' unobservable characteristics, into contact with public officials. Previous work by H u n t (2007), Guerrero and Rodriguez-Oreggia (2008), Tavits (2011) and Hunt and Laszlo (2012) does not explicitly consider this issue. Secondly, this paper uses the largely unexplored 'Life in Transition-2' survey, administered in 2010 in thirty economies of Central and Eastern Europe and Central Asia and five Western European countries. An important advantage of the data used is that respondents are asked about their actual corruption experience - paying a bribe to a public official. Much of the previous literature on the micro-determinants of corruption has concentrated on attitudes towards corruption (Soot/Rootalu 2012; Gatti et al. 2003; Truex 2011) and the probability of being asked for a bribe (Mocan 2008). Several studies (Tavits 2010; Hunt/Laszlo 2012; H u n t 2007; Guerrero/Rodriguez-Oreggia 2008) have addressed the actual bribing experience, but only in the context of one or two countries (Estonia, Peru, Uganda, Mexico). Our study analyses the determinants of actual corruption experience (completed corrupt exchanges) in a multi-country context. The paper primarily focuses on transition economies as corruption remains a widespread phenomenon there (Transparency International 2011; Iwasaki/Suzuki 2012). The high levels of corruption in the region can be attributed to several factors: high corruption could be a legacy of the Communist regime where it was considered "institutionalized, socially necessary evil for achieving goals and maintaining the national economy" (Iwasaki/Suzuki 2012: 54); it could have resulted from the plan-to-market transition process itself, which, in the absence of strong rule of law, created ample opportunities for corrupt exchange (Tavits 2011; Iwasaki/Suzuki 2012); and, finally, it could be linked to mentality, culture and considered a norm in a particular society. Our analysis initially pools all of the countries and reveals common determinants of bribery in the post-socialist world. However, we are also interested in cultural and context-specific determinants of bribery, which is why we estimate our model for different geo-political regions and different types of public officials. 2

Note that because of reverse causality and omitted variables it is often difficult to prove causal effects of particular variables on corruption. Most results in corruption literature report conditional correlations rather than causalities.

Bribing Behaviour and Sample Selection • 141

T h e remainder of the paper is structured as follows. Section 2 presents the data and discusses the corruption variables used. Section 3 presents the method o f estimation. Section 4 reports and analyses the results. A conclusion follows.

2 2.1

Data and variables Data

T o test for possible sample selection bias and find the determinants o f bribery, we use data from the "Life in Transition 2 " survey (LITS-2), conducted by the E B R D and the W o r l d Bank in autumn 2 0 1 0 . Thirty post-socialist economies of Central and Eastern Europe and Central Asia, as well as five Western European countries (France, Germany, Italy, Sweden and the UK), participated in the survey. T h e nationally representative samples consist of 1 , 0 0 0 respondents per country ( 1 , 5 0 0 respondents in the case of Russia, Ukraine, Uzbekistan, Serbia, Poland and the UK). In each country, the households were selected according to a two-stage clustered stratified sampling procedure. In the first stage, the frame o f primary sampling units was established using information on local electoral territorial units. In the second stage, a random walk fieldwork procedure was used to select households within primary sampling units. Steves ( 2 0 1 1 ) provides the survey summary, including detailed information on survey design and implementation methodology. O u r main analytical focus will be on the broad region (30 countries) of Eastern Europe and Central Asia. W e exclude the five Western European countries - the UK, Germany, France, Italy and Sweden - from the main sample, as they are likely to represent a qualitatively different group compared with the post-socialist region. Western countries have longer histories of functioning democracy and more transparent institutions, which would make petty corruption a less widespread phenomenon. T h e debate about whether corruption can be objectively measured requires consideration before proceeding further. Country-level measures o f corruption, notably Transparency International's Corruption Perception Index and the W o r l d Bank's World-wide Governance Indicators, have raised the profile o f corruption and been important in calls for greater transparency o f governments. However, since these indicators are largely based on perceptions o f experts from outside o f the countries involved, they are open to a number of criticisms raised by the authors themselves, e.g. Lambsdorff ( 2 0 0 5 ) , and by a number of other researchers. These include a lack o f conceptual underpinning (Arndt/Oman 2 0 0 6 ; Anderson/Heywood 2 0 0 9 ) and questionable primary data sources and methods, used to compile the indicators (Knack 2 0 0 7 ) . Olken ( 2 0 0 9 ) finds that actual corruption is not correlated with peoples' perceptions o f corruption. Razafindrakoto and R o u b a u d ( 2 0 1 0 ) find compelling evidence that experts over-estimate corruption in low-income countries compared to household level surveys, because they are biased towards a free market ideology. These findings raise questions about whether corruption and bribery can ever be measured accurately. Finding objective measures o f corruption is conceptually hard but there are numerous examples in the literature. Reinikka and Svensson ( 2 0 0 4 ) measure embezzlement as the gap between an independent estimate of government spending and what governments claim to spend. Di Telia and Schargrodsky ( 2 0 0 3 ) and Hsieh and M o r e t t i ( 2 0 0 5 ) compare government prices to the market prices to measure price-mark-ups or 'kickbacks'. Gorodnichenko and Peter ( 2 0 0 7 ) estimate the level and scale of bribery within the public sector by analysing consumption-income gaps o f public and private sector workers.

142 · Artjoms Ivlevs and Timothy Hinks

Reinikka and Svensson (2006) use public expenditure tracking surveys (PETS) and q u a n titative service delivery surveys (QSDS) to find that most public spending on education in Sub-Saharan Africa leaks f r o m the system. 3 However, this type of information is costly to gather and requires the help of those officials w h o m a y have a vested interest in hiding corruption in the first place (principal-agent problem). Because of the cost and principal-agent problems of objective measures of corruption, m a n y researchers have instead analysed individual experiences of corruption assuming these experiences are good proxies for actual (particularly petty) corruption. Proponents of this measure (e.g. Svensson 2 0 0 3 ; Reinikka/Svensson 2 0 0 6 ; Seligson 2 0 0 6 , and R a z a f i n d r a k o t o / R o u b a u d 2010) are aware of its weaknesses, notably that people are likely to under-report their participation and hence experiences in any corrupt or bribing activity which will bias any results. However this under-reporting is likely to vary in its degree depending on the cultural acceptance of such behaviour. H u n t (2007) suggests that the stigma associated with answering questions about bribery honestly may be lower in high-corruption countries t h a n low-corruption countries. Given this paper is concerned with transition economices only, we expect relatively low under-reporting compared to developed countries. 2.2

Variables

Dependent variable

Whether a bribe is made or not is subject to contact with public officials/institutions. The LITS-2 survey contains several questions relating to public official use a n d self-reported corruption behaviour. This paper uses t w o questions f r o m it. W i t h reference t o eight types of public services (interacting with road police; requesting official documents (visa, passport) from authorities; going to courts for a civil matter; receiving public education (primary or secondary); receiving public education (vocational); receiving medical treatment in the public health system; requesting unemployment benefits; requesting social security benefits), the first question asks, "During the past 12 m o n t h s have you or any member of your household used these {eight) services?" (italics added). If answered in the affirmative, then information is gathered on whether " . . . any member of your household (has made) an unofficial payment or gift when using these services over the past 12 m o n t h s ? " (italics added). W e will use the first question to capture the selection into contact with public officials/ institutions, and the second to capture bribery conditional on contacting public officials. 4 Table 1 reports, for each type of public official/institution, 1) the number of bribery episodes, 2) the proportion of bribery episodes relative to the full sample, 3) the p r o p o r t i o n of respondents w h o have used a public official relative to the full sample, and 4) the proportion of bribery episodes relative to the n u m b e r of respondents w h o have contacted 3

4

Monitoring institutional features of fiscal procedures in the public sector has been targeted by the Public Expenditure and Financial Accountability (PEFA) initiative established in 2001, www.pefa.org. In the 1980s and 1990s empirical work looking at how aid was spent and the impact it had on government spending was inconclusive, e.g. Pack and Pack (1990, 1993) with monitoring of aid an issue as was reliable country-level data. Depending on the type of public official, the non-response rate for the first question ranges from 1 to 2 % , and the non-response rate for the second question paying bribes ranges from 2 to 4%. These missing values have been excluded from the analysis. See also Kis-Katos and Schulze (2013) for a discussion of non-responses in corruption research.

Bribing Behaviour and Sample Selection · 143

Table 1 Distribution of bribery and official use across official types Bribery Bribery rate Public official Bribery rate use rate (relative to episodes (relative to (bribe was total (relative to public paid) total sample) official use) sample) Road police Requesting documents from authorities Courts Public education (primary and secondary) Public education (vocational) Public health system Unemployment benefits Other social security benefits

1901 1386 197 1302 1014 5839 176 320

0.057 0.042 0.006 0.039 0.030 0.175 0.005 0.010

0.188 0.230 0.044 0.254 0.164 0.687 0.054 0.095

0.303 0.181 0.134 0.154 0.185 0.255 0.098 0.101

Source: Life in Transition-2 Survey (2010)

a public official. In absolute terms, the highest number of bribery episodes is observed in the public health system (17.5% of all respondents were involved in bribery), followed by the road police (5.7%) and requesting official documents (4.2%). Together, these three types of public officials account for 7 5 % of all bribery episodes. However, if the comparison is made relative to those respondents who have used a public official, the road police emerges as the most corrupt public service (30.3% of those who have interacted with road police paid a bribe), followed by the public health system (25.5%) and the vocational public education (18.5%). Courts, unemployment and other social security services have the lowest rates both in terms of official use (4-10%) and giving a bribe conditional on official use (10-13%). 5 Explanatory variables Following the empirical literature on the micro-determinants of corruption behaviour and attitudes (Tavits 2010; Guerrero/Rodriguez-Oreggia 2008; Soot/Rootalu 2012; Hunt/Lazslo 2012; H u n t 2007; Truex 2011), our regressions will include the following respondent characteristics: dummy variables for gender, six age groups, 6 three education levels (primary, secondary, tertiary) and being employed; and the following householdlevel variables: linguistic minority status (if the main language spoken in the family is different from the official language(s) of the country), and dummy variables for three types of settlement (rural, urban, and metropolitan). Unfortunately, the survey does not contain information on the actual household (or individual) income. Therefore, we had to consider different proxies for household income. 5

It is i m p o r t a n t t o n o t e t h a t , f o r c e r t a i n types of p u b l i c institutions, it is possible t o o b t a i n t h e service w i t h o u t c o n t a c t i n g h a v i n g c o n t a c t t o a p u b l i c official. F o r e x a m p l e , electronic a n d p o s t a l a r r a n g e m e n t s m a y be in place f o r r e q u e s t i n g a n d o b t a i n i n g official d o c u m e n t s , as well as m a k i n g p a y m e n t s f o r p u b l i c services. If p e o p l e h a v e a c h o i c e t o c o n t a c t t h e p u b l i c official or n o t , w e m i g h t e x p e c t t h e m t o a v o i d c o n t a c t if t h e y a n t i c i p a t e t h a t they will be e x t o r t e d . For m a n y types of p u b l i c services, h o w e v e r , c o n t a c t w i t h officials will be inevitable (e.g. being s t o p p e d by a police o f f i c e r , d e a l i n g w i t h c o u r t s , h o s p i t a l s or e d u c a t i o n a l e s t a b l i s h m e n t s ) .

6

T h e survey also c o n t a i n s i n f o r m a t i o n o n t h e h o u s e h o l d h e a d s ' g e n d e r a n d age. W e h a v e c h e c k e d t h e r o b u s t n e s s of o u r results t o using these variables instead of t h e respective r e s p o n d e n t c h a r a c teristics - t h e results r e m a i n qualitatively u n c h a n g e d . T h i s is p a r t l y b e c a u s e in t h e 6 1 % of cases t h e r e s p o n d e n t s w e r e t h e h e a d s of h o u s e h o l d s (Steves 2 0 1 1 ) .

144 • Artjoms Ivlevs and Timothy Hinks

First, the survey contains information on household monthly expenditure on different goods (food, utilities, transport, education, health, clothing and durable goods), as well as information on household monthly savings. We used this information to create a total expenditure and savings adult equivalence variable. 7 Closer inspection of this variable revealed a 'don't know'/ non-response rate of 4 6 % and it was decided not to use this variable because of the huge loss in information this would cause. 8 Instead we used information on where respondents thought they were on a ten-step income ladder, where the first (tenth) step captures the poorest (richest) 1 0 % of the country. The response rate to this question was 9 8 . 3 % and is used as an as an income proxy, although it is likely to suffer from subjectivity bias, as there is no guarantee that everyone imagines the ten-step income ladder in the same way. To add a more tangible dimension of economic wealth, we also use information on household assets (car, secondary residence, bank account, debit card, credit card, mobile phone, computer and internet access at home) to create a wealth index using principal components. 9 Next, we want to include variables capturing respondents' trust in institutions and trust in people. The literature suggests that those with a high level of trust in public institutions are less likely to tolerate corruption and break the law (Soot/Rootalu 2 0 1 2 ; Marien/Hooghe 2 0 1 1 ) . Therefore, we expect they will also be less likely to be involved in bribery. Institutional trust will be captured by a composite variable, ranging from 4 (complete distrust) to 2 0 (complete trust), formed by adding together trust in four institutions: 1) the government/ cabinet of ministers, 2) local government, 3) courts and 4) the police (trust in each institution ranges from "complete distrust" (1) to "complete trust" (5)). We also include a variable capturing trust in people or what is termed 'social trust' in the literature. One could argue that participants involved in an illegal transaction need to trust each other in order for corruption to be sustained (Rose-Ackerman 2 0 0 1 ) . The variable is based on the question: "Generally speaking, do you think most people can be trusted with answers range from "complete distrust" (1) to "complete trust" (5). Tavits (2011) considered trust in people as a potential determinant of the probability of paying a bribe in Estonia, but obtained a statistically insignificant coefficient. La Porta et al (1997) found social trust to have a significant negative association on corruption at the country-level. We want to test the significance of this variable in a broader sample of East European and ex-USSR countries. The inclusion of trust measures raises the broader issue of endogeneity in our model since many of our right-hand-side variables can feasibly be caused by whether the individual pays a bribe or not. It is easy to think of paying bribes to public officials as having a direct impact on causing lower trust in institutions. 10 The quality of institutions is also important here, since lower quality means a poorer reputation and poorer experience

7

This variable used the O E C D income equivalence scale. W e subsequently split the variable into deciles within each country to avoid comparison of expenditures in different currencies.

8

The non-response rate is higher for particular countries: e.g. 6 6 % of respondents did not provide an answer for the expenditure on durable goods in Uzbekistan.

9

The correlation between the self-reported position on income ladder and the wealth index is 0 . 3 4 . The expenditure/ savings variable, which has 4 6 % missing values, has correlation coefficients of 26€/o and 3 6 % with the self-reported position on income ladder variable and the wealth index, respectively.

10

The reverse causality between corruption and trust though is not so clear-cut. Uslaner ( 2 0 0 2 , 2 0 0 8 ) and Fukayama ( 1 9 9 5 ) argue that trust is formed in the early years and it is unlikely that a child will have any direct dealings with corruption and paying bribes.

Bribing Behaviour and Sample Selection • 145

in using these institutions. Dealing with the endogeneity of individual regressors would require either panel data or suitable instrumental variables. Unfortunately, our data do not contain suitable instruments for trust; dealing with endogeneity is therefore left for future research. Finally, to control for all country-wide influences (historical, cultural etc.) on household corruption behaviour, all regressions will include country-fixed effects. 3

Estimation strategy

A typical regression, estimating the effects of socio-economic characteristics on the likelihood of bribery, could be run on a sample of public official users only. This results in a potential sample selection issue if the sample of public official users are not drawn from a random distribution (Wooldridge 2010). Formally, the bribing model with sample selection can be described as follows. We assume that an individual i's propensity to bribe is captured by an underlying latent variable, y*, which is determined by a set of the respondent's characteristics, X¡, and an independent and normally distributed error term, uy,: y*br,be = X.p+Ulh

Ui~N(0,l)

The actual bribing behaviour is captured by a binary variable the latent propensity to bribe y*:

ybribe __ \(Χφ

(1) which is related to

> 0)

(2)

The bribing behaviour, however, is observed only if contact with public official has taken tact place. The binary variable capturing contact with public officials, yÇ°" ^ depends on contact the latent propensity to contact public officials, y* :

^contact _ ^ ^contact > g^j Uj ~ N(0, 1), y*contact =Zly+u2l, where Z, is a vector of respondent characteristics determining selection into contact with public officials and «2 is an independent and normally distributed error term. Sample selection bias arises if the error terms in the outcome and selection equations are correlated: corr{u\, «2) = Ρ, Ρ Φ 0. To correct for any sample selection bias we use extensions to Heckman's (1979) original model that allows for both equations to have discrete dependent variables (see e.g. Van der Ven/Van Praag 1981; Baum 2006). To operationalise the model, we need an identification variable which would affect the probability of interaction with public officials but not necessarily the probability of making a bribe. We consider particular household/ respondent circumstances that are likely to increase contact with each type of public officials/institutions. We observe that households having a car are more likely to interact with road police; households with children are more likely to contact primary and secondary education establishments, as well as institutions in charge of social security benefits; students are more likely to contact vocational education establishments, as well as request official documents (e.g. passports); the less healthy are more likely to receive medical treatment; those who have recently

^ (4)

146 · Artjoms Ivlevs and Timothy Hinks

lost job are more likely to contact institutions in charge of unemployment benefits; and those who rent or have inherited their house/flat are more likely to go to courts for a civil matter. We use this information to construct a binary variable (for each type of public official), which takes the value of 1 if the characteristic is observed and 0 otherwise. Given that each respondent was asked about the actual corrupt behaviour eight times in consideration of eight different types of public officials - we follow Hunt and Laszlo (2012) to generate a stacked dataset which contains eight observations per household corresponding to interactions with eight institution/public official types. The unit of observation thus becomes the household-public official pair. To control for possible interdependence of responses provided by the same household and account for different types of institutions/officials, we cluster standard errors at the household level and include dummy variables for each type of public official in both the selection and outcome equations. The socio-demographic characteristics (age, gender, education, income, minority status, employment status and the area of residence), institutional trust, social trust, as well as country fixed effects, are included in both the selection into contact and outcome equations. 11 Besides comparing the results of the outcome equation in the Heckman correction model with a naïve probit regression, which does not account for sample selection, we will report the determinants of selection into contact with public officials. They are of interest, as selection into contact could be viewed as an integral part of the bribing process: e.g., Hunt (2007) and Hunt and Laszlo (2012) posit that factors increasing the need for public services increase bribery indirectly. To find out whether the determinants of bribery are context-specific, we will also run regressions for different country groups and different types of public officials.

4

Results

Base results At the outset, we want to note that our results represent conditional correlations rather than causal effects, as some of our regressors are potentially endogenous. This means applying extra caution when interpreting estimated coefficients. Table 2 reports the results of the two stages - bribery once contact with public officials has taken place and selection into contact with public officials - of the Heckman probit model, as well as those of two naïve probit models of bribery not correcting for sample selectivity: one estimated on a sample of public official users (censored) and another estimated for the total sample of respondents (uncensored). All models are estimated for the broad

11

The outcome (bribery) equation would also benefit from the characteristics of public officials (which, unfortunately, are not available in the survey), since the probability of bribery is arguably affected by the interests of both the consumers of public services and public officials. However, public officials will have little influence over who contacts them in the first place - we assume that contacting public schools, hospitals, registry offices etc. is largely driven by the demands of public service consumers. Note that in some cases, e.g. when the consumers of public services are private businesses, public officials may be able to create entry requirements to target customers who are more likely to pay bribes if asked to do so.

Bribing Behaviour and Sample Selection · 147

region of Central Eastern E u r o p e a n d Central A s i a . T o facilitate the interpretation of the results, only (average) marginal effects are reported a n d discussed. 1 2 W e first notice that the rho term is statistically significant (p = 0 . 0 4 1 ) , meaning that sample selection bias is present a n d that naïve probit results w o u l d be inconsistent, and the H e c k m a n selection procedure is a p p r o p r i a t e . It is also negative (—0.073), implying an omitted variable(s) which h a s an o p p o s i t e effect on contact with public officials a n d bribery. O n e variable w e can think of is the individual perception of h o w corrupt a particular public service is: if a p e r s o n thinks a service is c o r r u p t , she will minimise contact with it (probably preferring to deal with a private-sector provider if it is available), but if contact has taken place, the person will be m o r e likely to bribe, driven by the perceived behavioural n o r m . Another variable could be the accessibility of public services, defined in a b r o a d sense. People w h o , for different r e a s o n s (geographical distance, inconvenient w o r k i n g hours), find it difficult to access public services will have a lower likelihood of contacting public officials. H o w e v e r , such people m a y also be m o r e likely to engage in bribery a s they are keen to get the public service on the rare o c c a s i o n s they contact public officials. While w e d o include regressors potentially capturing the perceived corruptability of public officials (institutional trust) and limited access to public services (e.g. people living in rural a r e a s , employed), m a n y aspects of these variables are likely to remain unobserved a n d , therefore, uncontrolled for. C o m p a r i n g the findings of the H e c k m a n correction model with the results of a naïve probit w e observe larger (in a b s o l u t e terms) marginal effects in the latter. N o t controlling for selection effects thus p r o d u c e s an inflating bias in coefficients (and marginal effects). For instance, the naïve probit suggests that people are 3 . 8 percentage points less likely to bribe public health officials than the r o a d police; this declines to 2 . 5 percentage points when the s a m p l e selection bias is corrected for. Concerning specific determinants of bribery, c o l u m n (1) of T a b l e 2 reports the marginal effects of the H e c k m a n correction m o d e l ' s o u t c o m e equation. Considering first the sociod e m o g r a p h i c characteristics, respondents aged 4 5 - 5 4 , 5 5 - 6 4 a n d 6 5 + are 1, 2 . 5 a n d 1.4 percentage points, respectively, less likely to bribe public officials, c o m p a r e d with individuals a g e d 3 4 - 4 5 (the reference g r o u p ) . This finding is consistent with Guerrero a n d R o d r i g u e z - O r e g g i a ( 2 0 0 8 ) , M o c a n ( 2 0 0 8 ) a n d S o o t and R o o t a l u ( 2 0 1 2 ) , w h o find that the elderly are less likely to pay bribes, be a s k e d for bribes by public officials a n d tolerate corruption. T h e H e c k m a n correction model suggests that here is no significant difference in bribing behaviour between males a n d females. This is inconsistent with similar w o r k using the W o r l d Value Surveys by S w a m y et al ( 2 0 0 1 ) a n d Gatti et al ( 2 0 0 3 ) w h o find males to be significantly m o r e prone to corrupt behaviour. Experimental evidence f r o m Alatas et al. ( 2 0 0 9 ) also find that females f r o m Australia are less tolerant and less likely to engage in corruption. G u e r r e r o and R o d r i g u e z - O r e g g i a ( 2 0 0 8 ) find that in M e x i c o males a n d females have the s a m e disposition to bribe but that females d o not pay bribes whereas men d o . Level of education plays no significant role in corruption behaviour in the p o o l e d regression. This is consistent with the findings of Gatti et al ( 2 0 0 3 ) but inconsistent with those of S w a m y et al ( 2 0 0 1 ) , M o c a n ( 2 0 0 8 ) and Guerrero a n d R o d r i g u e z - O r e g g i a ( 2 0 0 8 ) w h o find the m o r e educated are m o r e prone to corrupt behaviour. This is likely to be explained by cultural characteristics a n d geo-political differences between countries a r o u n d the globe.

12

The full econometric output is available on request.

148 · Artjoms Ivlevs and Timothy Hinks

This argument is substantiated by the findings of Truex (2011) that the more educated in Nepal were less accepting of corrupt behaviour. The linguistic minorities are 1.5 percentage points more likely to pay bribes. This finding is corroborated by the results of two studies on corruption in Estonia: Soot and Rootalu (2012) find that the ethnic minorities are more likely to tolerate corruption and have lower corruption awareness, and Tavits (2010) finds that minority public officials are more likely to engage in hypothetical corrupt deals. Several explanations of why the minorities are more likely to engage in corruption can be provided. First, the minorities are concentrated in certain sectors of the economy (e.g., the informal sector) which are more conducive to corrupt behaviour. Second, minorities, who are often more vulnerable and less able to seek/obtain protection, could be an easy target for extortion by public officials. Third, historical factors might play a role: in many countries of our sample, today's linguistic minorities originate from the former political elites (e.g. the Russian-speakers in the successor states of the USSR). These political elites might have been particularly prone to corruption under the previous regime (Iwasaki/Suzuki 2 0 1 2 ) , and transferred their corruption attitudes and norms to the present day. The perceived income is positive and statistically significant: moving one step on the imaginary income ladder is associated with a 0.4 percentage point higher likelihood of bribery. This finding conforms to existing literature on micro-determinants of corruption: wealthier people have been found to be more likely to pay bribes conditional on contact with public officials (Hunt/Lazslo 2 0 1 1 ; Guerrero/Rodriguez-Oreggia 2 0 0 8 ) , be asked for bribes by public officials (Mocan 2 0 0 8 ) and find corrupt behaviour justifiable (Gatti et al. 2003). Typical explanations for richer individuals' higher propensity to bribe include their greater ability to pay, their higher opportunity costs of time, as well as the conjecture that rich individuals are the prime target for corruption-prone public officials. 13 Institutional trust is a strong determinant of the probability of paying a bribe. A one unit increase on the institutional trust scale (4 - complete distrust, 2 0 - complete trust) is associated with a 0.8 percentage point lower likelihood of paying a bribe. The finding is consistent with Soot and Rootalu (2012), who find that people with more trust in institutions tolerate corruption less and have a stronger awareness of corruption. As discussed previously though, the causality could run in both directions. In particular, people who have experienced corruption-free encounters with public officials may develop a deeper trust in public institutions. If this is the case, the obtained coefficient on institutional trust is upward biased and should be interpreted with caution. 14 The road police (the reference group) appears to be the most corrupt public service in our sample. These are followed by the public health services (people are 2.5 percentage

13

Note, however, that a reverse causality between income and bribery may exist if people engage in bribery in order to increase their income; we would then observe an upward bias in the income variable estimates.

14

Given a potential interdependence between institutional trust and the socio-demographic controls (see Table A 2 of the appendix for a correlation matrix of regressors), we have estimated our model without the two trust variables (see Table A 3 of the appendix). Some of our results remain qualitatilely unchanged (in particular, the older and the ethnic majorities are less likely to bribe). However, the wealth index rather than the perceived income decile now becomes a positive and significant predictor of bribery; in addition, females and those aged 1 8 - 2 4 are 0 . 7 and 1.1 percentage points, respectively, less likely to pay bribes if trust variables are excluded. W e have also estimated the model on the institutional trust variable and country fixed effects alone and obtained virtually the same negative and highly significant coefficient on institutional trust.

Bribing Behaviour and Sample Selection · 149

points less likely to pay bribes to public health officials relative to road police), registry, courts and vocational education establishments (10-11 percentage points less likely), and primary and secondary schools (15 percentage points less likely). Among the eight public institutions/ services, those dealing with unemployment and other social security benefits appear to be the least corrupt (17.5-19.5 percentage points lower likelihood to pay bribes compared to the road police). Controlling for individual characteristics, we observe an important heterogeneity of selfreported bribery rates at the country level. Compared to Poland (the reference country), people in Kosovo, Georgia, Slovenia and Croatia are 5 - 1 0 percentage points less likely to pay bribes to public officials. In several countries - FYR of Macedonia, the Czech Republic and Slovakia - the likelihood of being involved in bribery is the same as in Poland. However, people are 4 - 5 percentage points more likely, relative to Poland, to pay bribes in Bosnia and Herzegovina, Bulgaria, Serbia, Latvia, Turkey and Estonia; 10-14 percentage points more likely to pay bribes in Armenia, Montenegro, Hungary Mongolia, Belarus and Russia; 1 6 - 1 7 percentage points more likely to pay bribes in Kazakhstan, Romania, Moldova and Lithuania; and, finally, 2 7 - 3 9 percentage points more likely to pay bribes in Ukraine, Uzbekistan, Albania, Tajikistan, Kyrgyzstan and Azerbaijan. Note that the country dummies capture the aggregate effect of all possible country-level influences on conditional bribery rates; by construction, they do not allow us to isolate the effects of separate country-level factors. However, one notices some association between the GDP per capita and conditional bribery rates (bribery tends to be more widespread in poorer countries). Also, the ex-USSR Central Asian states tend to have higher conditional bribery rates than other countries in the sample. Before we turn to the determinants of bribing behaviour and the presence of sample selection bias in different country groups, it is useful to consider the factors which affect the probability of being selected into contact with public officials (the selection equation of the Heckman correction model; column 2 of Table 2). As mentioned earlier, selection into contact can be viewed as having an indirect effect on bribery. The results suggest that, compared to people aged 3 5 - 4 4 (the reference group), those aged 18-24 and 2 5 - 3 4 are 1 and 2 percentage points, respectively, less likely to contact public officials. Beyond the age group 35—44, the demand for public services declines with age: compared with the reference group 35-44, those aged 4 5 - 5 4 , 5 5 - 6 4 and 6 5 + are 0.8, 3.6 and 5.5 percentage points, respectively, less likely to contact public officials. Thus, people aged 3 4 - 4 5 are the most active users of public services, while the elderly are the least active. This finding could be explained by lower engagement in the labour market and lower family commitments of the elderly. The two proxies for household income - perceived income decile and wealth index are statistically significant predictors of contact with public officials. Their coefficients, however, have different signs. The wealth index is positively associated with the probability of contacting public officials and is consistent with the findings of H u n t and Laszlo (2011). However, those who perceive themselves as having higher income are less likely to contact public officials. A possible explanation for this finding could be the willingness of people positioning themselves higher on an imaginary income ladder to substitute public-sector services with private-sector alternatives. 15 15

If only the wealth index is included in the analysis, its coefficient is positive and significant in both selection and outcome stages. If only perceived income is included, its coefficient is positive and significant in the stage, but positive and insignificant in the selection stage.

150 • Artjoms Ivlevs and Timothy Hinks

Considering the country fixed effects, the least intensive contact with public officials is observed in Estonia and H u n g a r y (not significantly different compared with the reference country of Poland), and the most intensive in Albania, Azerbaijan, Kosovo, FYR of M a c e d o n i a and Uzbekistan ( 1 3 - 1 9 percentage points more likely to contact officials compared with Poland). This heterogeneity could be explained by the way in which public services are provided in different countries (e.g., they could be provided electronically, w i t h o u t having to physically meet public officials), as well as the availability of privately provided alternatives to public services (e.g., private education and healthcare can be more present/accessible in some countries). As well as controlling for omitted variable bias, the H e c k m a n a p p r o a c h provides a more nuanced picture on w h a t is likely to drive corruption at the individual and country levels, compared with studies which concentrate only on the outcome stage or ignore selection into contact altogether. For example, the elderly are both less likely t o contact public officials and pay bribes, conditional on contact with pubic officials, pointing at a rather clear-cut negative relationship between age and the likelihood of corruption. M o r e interesting are the cases when a characteristic has opposite signs in the selection and outcome equations. For example, the coefficients of Georgia, Croatia, Kosovo and Slovenia in the outcome equation are negative, suggesting that, conditional o n contact with public officials, people in these countries are 5 - 1 0 percentage points less likely to pay bribes compared with Poland (the reference group). However, these countries are 7 - 1 9 percentage points more likely to contact public officials, which would attenuate the conditional bribery effect in evaluating average bribery in a country. Estimating the probability of bribery on the whole sample (i.e. including into analysis people w h o did not contact public officials), the marginal effects for which are reported in column [4] of Table 4, indeed yields a statistically insignificant marginal effect for Kosovo, and the marginal effects ranging between —1.4 and —2.1 percentage points for Georgia, Croatia and Slovenia. 16 Results for different country groups T h e post-socialist world is far f r o m being a g r o u p of homogenous countries. I m p o r t a n t country level variations in the degree of democracy, rule of law and, indeed, control of corruption exist. In this subsection, we look at the determinants of bribery and the presence of sample selection bias in different parts of the Central Eastern European and Central Asian region, and contrast them with the five Western European countries. W e create, along geo-political lines, the following country groups: the Balkans (Albania, Bosnia and Herzegovina, Bulgaria, Croatia, Kosovo, FYR of Macedonia, M o n t e n e g r o , R o m a n i a , Serbia), the Baltics (Estonia, Latvia, Lithuania), the Caucasus (Armenia, Azerbaijan, Georgia), Central Asia (Kazakhstan, Kyrgyzstan, Tajikistan, Uzbekistan), Central Europe (Czech Republic, H u n g a r y , Poland, Slovakia, Slovenia), Slav ex-USSR (Belarus, Russia, Ukraine), and Western Europe (Germany, France, Italy, Sweden, the UK). 17 Table 3 reports the marginal effects of the H e c k m a n correction model outcome equation (probability of bribery once the contact with public officials has taken place), 18 as well as the estimated coefficient of correlation between the error terms in the selection and 16

17

18

This simple comparison also suggests that the total bribery effect cannot be obtained by summing the coefficients in the outcome and selection equations. Moldova, Mongolia and Turkey are excluded from the analysis, as it is difficult to assign them to a particular country group. The results of the selection stage are available in the appendix to this paper.

Bribing Behaviour and Sample Selection • 151

Table 2 Determinants of bribery and contact with public officials, Heckman probit and naïve probit average marginal effects Heckman probit

Naive probit (censored)

Naive probit (uncensored)

Paid a bribe (Outcome)

Contacted official (Selection)

Paid a bribe

Paid a bribe

[1]

[2]

[3]

[4]

Female

-0.005

-0.001

-0.006

-0.003*"

Age group 18-24 25-34 35-44 45-54 55-64 65+

-0.010 0.002 Ref. -0.010» -0.028*" -0.016"

-0.011 0.003 Ref. -0.012* -0.032*** -0.018" 0.017"*

-0.003* -0.003** Ref. -0.006*** -0.015*** -0.014***

0.005*"

0.001 ***

0.003

0.006*"

Linguistic minority Income ladder

0.015*"

-0.010*" -0.020*** Ref -0.008*** -0.036*** -0.055"* -0.003

0.004"*

-0.002"*

Wealth index

0.002

Education Primary Secondary Tertiary

-0.000 Ref. -0.000

-0.003 Ref 0.001

-0.000 Ref. -0.000

-0.001 Ref. 0.001

Employed

-0.004

-0.000

-0.005

-0.001

0.001 Ref. -0.003 -0.008***

0.002 Ref 0.004

0.002 Ref. -0.004

0.002* Ref. -0.000

-0.009*"

-0.002***

Type of settlement Rural Urban Metropolitan Institutional trust Social trust

0.003

0.016***

-0.000 0.001

0.003

0.002

0.001

Type of public official Road police Registry Courts Education (primary and secondary) Education (vocational) Public health Requesting unemployment benefits Requesting other social security benefits

Ref. -0.100"* -0.110"* -0.149"* -0.100*" -0.025*** -0.175*** -0.195***

Ref. 0.105*** -0.163*** 0.063*" 0.046*** 0.376"* -0.079*** -0.079***

Ref. -0.122*" -0.132"* -0.172"* -0.120*" -0.038"* -0.209"* -0.227*"

Ref. -0.014"* -0.079"* -0.017*** -0.026*** 0.056*** -0.083*** -0.065***

Country Albania Armenia Azerbaijan Belarus Bosnia and Herzegovina Bulgaria

0.300*** 0.089*" 0.393"* 0.132"* 0.036** 0.040***

0.126"* 0.019"* 0.151*** 0.088"* 0.046*** 0.055*"

0.343*** 0.101 *** 0.451"* 0.152*" 0.040" 0.046***

0.098"* 0.031"* 0.128*** 0.055"* 0.021 *** 0.021 * "

152 • Artjoms Ivlevs and Timothy Hinks

Table 2 Continued Heckman probit

Croatia Czech Republic Estonia Georgia Hungary Kazakhstan Kosovo Kyrgyzstan Latvia Lithuania Macedonia, FYR Moldova Mongolia Montenegro Poland Romania Russia Serbia Slovakia Slovenia Tajikistan Turkey Ukraine Uzbekistan Selection into contact Number of household-official pairs Number of households Censored observations Prob > Chi 2 Pseudo R 2 Rho Prob > Chi 2 (Rho=0)

Naïve probit (censored)

Naïve probit (uncensored)

Paid a bribe (Outcome)

Contacted official (Selection)

Paid a bribe

Paid a bribe

[1]

[2]

[3]

[4]

-0.098*** 0.016 0.050** -0.073*** 0.107*** 0.159*** -0.046*** 0.385*** 0.040** 0.174*** -0.007 0.170*** 0.120*** 0.100*** Ref. 0.169*** 0.139*** 0.042** 0.021 -0.086*** 0.311*** 0.053*** 0.270*** 0.283***

0.069*** 0.073*** 0.001 0.075*** 0.006 0.107*** 0.185*** 0.126*** 0.062*** 0.051*** 0.129*** 0.102*** 0.066*** 0.048*** Ref. 0.027*** 0.081 *** 0.057*** 0.068*** 0.073*** 0.100*** 0.011 0.090*** 0.132*** 0.160***

-

255624 32556 201469 0.000 -

-0.073 0.041

-0.114*** 0.017 0.058*** -0.084*** 0.123*** 0.182*** -0.055*** 0.442*** 0.045*** 0.200*** -0.009 0.195*** 0.137*** 0.115*** Ref. 0.195*** 0.160*** 0.048*** 0.024 -0.100*** 0.356*** 0.060*** 0.309*** 0.325*** -

54155 25852 -

0.000 0.204

-0.021*** 0.016*** 0.012** -0.014** 0.027*** 0.069*** 0.008 0.120*** 0.020*** 0.057*** 0.019*** 0.069*** 0.044*** 0.038*** Ref. 0.049*** 0.056*** 0.024*** 0.017*** -0.019** 0.103*** 0.016*** 0.086*** 0.106*** -

255624 32556 -

0.000 0.250

-

-

-

-

Notes: * denotes significance at 10% level, ** - 5 % , *** - 1%. The unit of observation is household-official pair. Standard errors (not reported to save space) clustered at household level.

outcome equations (rho). The sample selection bias is present in the Caucasus, Slav exUSSR and the Central Europe subsamples: w h e n statistically significant the estimated rho term is always negative (as for the pooled results in Table 2), meaning that unobservable variables in these country groups are systematically increasing the likelihood of engaging with officials but are then reducing the likelihood of paying a bribe (or vice versa).

Bribing Behaviour and Sample Selection · 153

The results reveal considerable variation in the determinants of the bribery decision at country group level. Gender is statistically insignificant in all country groups but Central Asia, where women are 3.1 percentage points less likely to pay bribes than men. The negative association between bribery and age is observed in Central Asia, and Slav ex-USSR; the relationship in other country groups is less clear-cut or the coefficients are insignificant. Linguistic minorities are more likely to pay bribes in the Baltics and, especially, the Western European group, while the coefficient for other country groups is statistically insignificant. Perceived income is positive and significant only in the Balkans and Western Europe. In Western Europe, moving from the lowest to the highest step on the ten-step income ladder is associated with a 4 percentage point higher likelihood of bribing public officials. In the Balkans, the association is stronger both in size and significance: moving from the bottom to top on the relative income ladder is associated with an 8 percentage point increase in the likelihood of bribing public officials, which is significant at the 99 per cent level. The wealth index is also positively correlated with paying a bribe in the Slav ex-USSR, but negatively correlated in Western European countries. Higher levels of education tend to be associated with a lower probability of paying bribes to public officials. In Central Asia and Western Europe, people with primary education are 3.8 and 1.2 percentage points, respectively, more likely to pay bribes compared with those with secondary education (the reference group). At the same time, in the Balkans and Western Europe, people with tertiary education are 2.5 and 1.5 percentage points, respectively, less likely to pay bribes relative to the reference group. The exception is Central Europe, where the tertiary educated are 1.5 percentage points more likely to bribe public officials. For several regressors, the sign of the estimated coefficient depends on the country group. For example, the employed are 1.8 percentage points less likely to pay bribes in Central Europe, while the regressor is insignificant in other country groups. We also observe an important variation in the relationship between bribery and type of settlement. For several country groups, bribery is more likely to be committed in urban areas: compared with people living in urban-non-metropolitan areas (the reference group), those from rural areas in the Caucasus and Central Europe are less likely to report bribing public officials. Similarly, those living in metropolitan areas in the Slav ex-USSR region are 8.3 percentage points more likely to pay bribes compared with the reference group. An opposite association is observed in Central Asia, where bribery is more widespread in rural areas: compared with the reference group, the village dwellers there are 6.1 percentage points more likely, and the metropolitan dwellers are 14.2 percentage points less likely, to bribe public officials. A less clear-cut relationship between the degree of urbanisation and the probability of bribery is observed in the Baltics, where both rural and metropolitan dwellers are less likely to bribe than people living in urban-non-metropolitan areas. Institutional trust is negatively associated with the probability of bribing public officials in all country groups, except the Baltics. The marginal effect is particularly high in Central Asia, where a one unit increase on the institutional trust scale (4 - complete distrust, 20 - complete trust) is associated with a 2.4 percentage point lower probability of bribery. Contrary to the consistent nature of the institutional trust variable, the relationship between bribery and social trust differs across country groups. In particular, the conjecture that in order to bribe one needs to trust people is supported in the Caucasus, Central Asian and Western European sub-samples. In these three country groups, an extra step on the 1 (no trust) —5 (complete trust) social trust scale is associated with 1.1, 3

1 5 4 • Artjoms Ivlevs and Timothy Hinks

and 1 percentage point increase in the probability of bribing public officials, respectively. However, in the Balkans, people with more social trust are less likely to engage in bribery, an extra step on the 1 - 5 interpersonal trust scale being associated with a decrease of 0.6 percentage points in the probability of bribing. Considering conditional bribery rates for different types of public officials, the road police are the most corrupt public service in the Balkans, Central Asia and Slav ex-USSR, while public health is the most corrupt in the Caucasus and Central Europe. In the Baltics, the road police and public health are equally corrupt and have the highest bribery rates among the eight types of public officials. In Western Europe, there is no statistically significant difference in the conditional bribery rates for the road police, registry, public health and services dealing with unemployment benefits, and there is 2.2-3.2 percentage point decrease in the likelihood of observing a corrupt exchange in vocational education and services dealing with social security benefits other than unemployment. Overall, in most country groups, services dealing with the unemployment and other social security benefits, as well as education, tend to be the least corrupt. Comparing the individual-level determinants of bribery in Western Europe (West) with those in the broad Central Eastern European/ Central Asian region (East), one notable similarity is that people with lower institutional trust are more likely to bribe public officials in both parts of the world. However, important differences also exist. Higher values of the wealth index are negatively associated with bribery in the West and positively in the East (Slav ex-USSR). Higher levels of education and lower levels of interpersonal trust are associated with lower bribery rates in the West, while the relationship in the East is more opaque. Older respondents are less likely to bribe in the East, while the estimated coefficients of age groups in the West sub-sample are statistically insignificant. Finally, linguistic minority status is a significant predictor of bribery in the West and only in one part of the East (the Baltics), and insignificant elsewhere. 19

19

We have also compared the average bribery rates, conditional on the observed socio-demographic characteristics, of all countries participating in the survey. Our expectation was that the bribery rates in the West would be lower than those in the East. Running a regression which includes both Western and Eastern European/ Central Asian countries, we found that the public services users in Sweden, Italy, France and the UK are 18.4, 14.3, 9.6 and 7.2 percentage points, respectively, less likely to bribe public officials, compared with the public official users in Poland (the reference group). However, unexpectedly, we found that the probability of bribing, condition on public official use, was 5.6 percentage points higher in Germany than in Poland. This is a puzzling result. One explanation could be that people in Germany tend to be more honest about their involvement in corrupt exchanges, while people in other countries underreport it. Another reconciling factor is that, according to our model, the probability of contacting public officials is lower in Germany than in any other country in the sample (the average marginal effect is -6 percentage points). This has an indirect negative effect on bribery, especially if one wants to calculate the bribery rate for whole population of the country (the users and the non-users of public services). The full set of results (the outcome and selection equations) for the whole sample is available on request.

Bribing Behaviour and Sample Selection • 155

£ υ υ α. « S ω 3 gu,

^

tv . τ- ko in r O f O O O o o ^ o o o o o o o o

at 1 / 1

rO o o

min o o Ò o

c « 41
X

Inemployment enefits

LU ~ 1/1

ducation vocational)

κ o -o

ΙΑ •e 3 O u

ducation irimaty and econdary)

o> tu "5 Q-

equesting fficial ocuments

Table A5 Determinants of contacting different types of public officials; Heckman probit selection equation average marginal effects

l/l J2

0.013*** 0.003 0.017*** -0.005 -0.028"* -0.045"* 0.009* -0.014*** -0.005*" 0.010"

0.003 -0.007

-0.005 -0.010" -0.029**'' -0.030***

-0.002

-0.001

-0.007 0.003"* 0.007*** 0.115*"

0.011 " '' -0.006 -0.001 " '' -0.000 -0.001 0.000 0.071 " '' 0.053***

0.004

References Alatas, V., L. Cameron, A. Chaudhuri, N. Erkal, L. Gangadharan (2009), Gender, Culture, and Corruption: Insights from an Experimental Analysis. Southern Economic Journal 75(3): 663-680. Ali, A.M., H.S. Isse (2003), Determinants of economic corruption: a cross-country comparison. Cato Journal 22: 4 4 9 ^ 6 6 . Andersson, S., P. Heywood (2009), The Politics of Perception: Use and Abuse of Transparency International's Approach to Measuring Corruption. Political Studies 57: 746-767. Arndt, C., C. Oman (2006), Uses and abuses of goverance indicators. Paris: OECD Development Centre. Baum, C. (2006), An Introduction to Modern Econometrics Using Stata. Stata Press, College Station, TX. Brunetti, Α., Β. Weder (2003), A free press is bad news for corruption. Journal of Public Economics 87: 1801-1824. Cuervo-Cazurra, A. (2006), Who cares about corruption? Journal of International Business Studies 37: 803-822. Di Telia, R., E. Schargrodsky (2003), The role of wages and auditing during a crackdown on corruption in the city of Buenos Aires. Journal of Law and Economics 46(1): 269-292. Egger, P., H. Winner (2006), H o w Corruption Influences Foreign Direct Investment: A Panel Data Study. Economic Development and Cultural Change 54(2): 459-486. Everhart, S., J. Martinez-Vazquez, R. McNab (2009), Corruption, governance, investment and growth in emerging markets. Applied Economics 41(13): 1579-1594.

166 · Artjoms Ivlevs and Timothy Hinks

Fukuyama, F., (1995), Trust: The Social Virtues and the Creation of Prosperity. N e w York: Free Press. Gatti, R., S. Paternostro, J. Rigolini (2003), Individual attitudes t o w a r d corruption: do social effect matters? World Bank Policy Research Working Paper 3122. W o r l d Bank, Washington. Glaeser, E.L., D.I. Laibson, J.A. Schienkman, C.L. Soutter (2000), Measuring Trust. Quarterly Journal of Economics 115(3): 8 1 1 - 8 4 6 . Guerrero, M.A., E. Rodriguez-Oreggia (2008), O n the individual decisions to commit corruption: A methodological complement. Journal of Economic Behavior and Organization 65: 357-372. Gupta, S., H . Davoodi, R. Alonso-Terme (2002), Does Corruption Affect Income Inequality and Poverty? Economics of Governance 3: 2 3 ^ 5 . H e c k m a n , J. (1979), Sample selection bias as a specification error. Econometrica 47: 1 5 3 - 1 6 1 . Hsieh, C-T., E. Moretti (2006), Did Iraq cheat the United Nations? Underpricing, Bribes, and the Oil and Food Program. Quarterly Journal of Economics 121(4): 1 2 1 1 - 1 2 4 8 . H u n t , J. (2007), H o w corruption hits people when they are d o w n . Journal of Development Economics 84(2): 5 7 4 - 5 8 9 . H u n t , J., S. Laszlo (2012), Is bribery really regressive? Bribery's costs, benefits, and mechanisms. World Development 40(2): 3 5 5 - 3 7 2 . Iwasaki, I., T. Suzuki (2012), The determinants of corruption in transition economies. Economics Letters 114: 5 4 - 6 0 . Johnson, N . , C. La Fountain, S. Yamarik (2011), Corruption is bad for growth (even in the United States). Public Choice 147: 3 7 7 - 3 9 3 . Jong, De.E., C. Bogmans (2011), Does corruption discourage international trade? European Journal of Political Economy 27(2): 3 8 5 - 3 9 8 . Jong-sun, Y., S. Khagram (2005), A comparative study of inequality and corruption. American Sociological Review 70 (1): 1 3 6 - 1 5 7 . Kis-Katos, K., G. Schulze (2013), Corruption in Southeast Asia: a survey of recent research. Asian-Pacific Economic Literature 27(1): 7 9 - 1 0 9 . Knack, S. (2007), Measuring Corruption: A critique of indicators in Eastern Europe and Central Asia. Journal of Public Policy 27(3): 2 4 5 - 2 9 1 . Lambsdorff, J.G. (2003), H o w Corruption Affects Productivity. Kyklos 56(4): 4 5 7 - 4 7 4 . Lambsdorff, J.G. (2005), The methodology of the 2 0 0 5 corruption perceptions index, September. Transparency International and University of Passau. Lambsdorff, J.G. (2006), Causes and consequences of corruption: w h a t do we k n o w f r o m a cross-section of countries? Pp. 3 - 5 1 in: S. Rose-Ackerman (ed.), International H a n d b o o k on the Economics of Corruption. Edward Elgar, Cheltenham, UK. La Porta, R., F. Lopez-De Silanes, A. Shleifer, R.W. Vishny (1997), Trust in Large Organisations. American Economic Review, Papers and Proceedings CXXXVII(2): 3 3 3 - 3 3 8 . Marien, S., M . H o o g h e (2011), Does political trust matter? An empirical investigation into the relation between political trust and support for law compliance. European Journal of Political Research 50: 2 6 7 - 2 9 1 . M a u r o , P. (1995), Corruption and growth. Quarterly Journal of Economics 110(3): 6 8 1 - 7 1 2 . M o c a n , Ν . (2008), W h a t determines corruption? International evidence f r o m micro data. Economic Inquiry 46(4): 4 9 3 - 5 1 0 . Olken, B.A. (2009), Corruption Perceptions vs. Corruption Reality. Journal of Public Economics 93(7-8): 9 5 0 - 9 6 4 . Pack, H., J.R. Pack (1990), Is foreign aid fungible? The case of Indonesia. Economic Journal 100(3): 1 8 8 - 1 9 4 . Pack, H., J.R. Pack (1993), Foreign aid and the Question of fungibility. Review of Economics and Statistics 75(2): 2 5 8 - 2 6 5 . Razafindrakoto, M . , F. R o u b a u d (2010), Are International Databases on Corruption Reliable? A Comparison of Expert Opinion Surveys and Household Survey in Sub-Saharan Africa. World Development 38(8): 1 0 5 7 - 1 0 6 9 . Reinikka, R., J. Svensson (2004), Local Capture: Evidence from a Central Government Transfer Program in Uganda. Quarterly Journal of Economics 119(2): 6 7 9 - 7 0 6 .

Bribing Behaviour and Sample Selection · 167

Reinikka, R., J. Svensson (2006), Using Micro-Surveys to Measure and Explain Corruption. World Development 34(2): 3 5 9 - 3 7 0 . Rose-Ackerman, S. (2001), Trust, Honesty, and Corruption: Reflection on the State-Building Process. European Journal of Sociology (42): 2 7 - 7 1 . Seligson, M.A. (2006), The Measurement and Impact of Corruption Victimization: Survey Evidence from Latin America. World Development 34(2): 3 8 1 - 4 0 4 . Sööt, M.-L., K. Rootalu (2012), Institutional Trust and Opinions of Corruption. Public Administration and Development 32(1): 8 2 - 9 5 . Steves, F. (ed.) (2011), Life in transition: After the crisis. London: European Bank for Reconstruction and Development and World Bank. Svensson, J. (2003), W h o must pay bribes and h o w much? Evidence from a cross-section of firms. Quarterly Journal of Economics 118(1): 2 0 7 - 2 3 0 . Swamy, Α., S. Knack, Y. Lee, O . Azfar (2001 ), Gender and Corruption. Journal of Development Economics 64(1): 2 5 - 5 5 . Tavits, M . (2010), W h y do People Engage in Corruption? The Case of Estonia. Social Forces 88(3): 1 2 5 7 - 1 2 8 0 . Transparency International (2011), Corruption Perceptions Index 2011. Transparency International. Treisman, D. (2000), The Causes of Corruption: A Cross-National Study. Journal of Public Economics 76(3): 3 9 9 - 4 5 7 . Truex, T. (2011), Corruption, Attitudes, and Education: Survey Evidence f r o m Nepal. World Development 39(7): 1 1 3 3 - 1 1 4 2 . Uslaner, E.M. (2002), The M o r a l Foundations of Trust. Cambridge University Press, N e w York. Uslaner, E.M. (2008), The foundations of trust: macro and micro. Cambridge Journal of Economics 32: 2 8 9 - 2 9 4 . Van de Ven, W . P . M . M . , Β.M.S. Van Praag (1981), The demand for deductibles in private health insurance: A probit model with sample selection. Journal of Econometrics 17: 2 2 9 - 2 5 2 . Van Rijckeghem, C., B. Weder (2001), Bureaucratic corruption and the rate of temptation: d o wages in the civil service affect corruption, and by h o w much? Journal of Development Economics 65: 3 0 7 - 3 3 2 . Wei, S.-J. (2000), H o w taxing is corruption on international investors? Review of Economics and Statistics 82: 1 - 1 1 . Wooldridge, J . M . (2010), Econometric Analysis of Cross Section and Panel Data. Cambridge, M A : M I T Press. Corresponding author: Artjoms Ivlevs, Department of Accounting, Economics and Finance, University of the West of England, Bristol BS16 1QY, UK. [email protected] Timothy Hinks, Department of Accounting, Economics and Finance, University of the West of England, Bristol BS 16 1QY, UK. [email protected]

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2015) Bd. (Vol.) 235/2

Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China Qijun Liu and Yaping Peng* School of Public Administration, Huazhong University of Science and Technology, China JEL D73; D03; P26 Corruption; bribery; China; university admissions; university examinations; expressive behavior.

Summary We use a unique dataset based on reported direct personal bribes paid by arts students in China at examinations and in the college admissions process to study willingness to bribe. We find that individual willingness to bribe depends on personal characteristics rather than on the attributes of the admissions process at different colleges and universities. The perceived level of corruption, personal attitudes towards corruption, academic attainment, and the rank of a college are significant predictors of bribery. Based on self-reporting, students from middle-income families have a higher likelihood of engaging in bribery than students from poor or rich families. There are no significant gender differences in bribing behavior. We acknowledge and seek to account for the possibility of identity-confirming expressive behavior in the survey responses.

1

Introduction

In China, as in low and middle income countries in general (Haque/Kneller 2009; Swaleheen 2011), there has been pervasive corruption, although, in contrast to other countries, corruption in and associated rent seeking and rent extraction in China have occurred in the presence of high rates of economic growth in transition years (Chen/Liu 2015). The scope of corruption in China is indicated by official statistics showing more than one million convictions for corruption over the past three decades, with an average of over 30,000 convictions per year. 1 In addition, many corrupt activities may not have been subject to official detection. Bribery or "gift payments" are so pervasive that Articles 383 and 385 of the Chinese Criminal Law stipulate that a government official is not to be prosecuted unless the sum of monetary values of bribes exceeds 5,000 RMB (equivalent to US $800 at the exchange rate at the time of this research). 2 The high levels of corruption make China a primary source of data for research on corruption.

* Helpful comments from the two anonymous reviewers and suggestions from Professor Arye L. Hillman and Professor Günther G. Schulze are gratefully appreciated. This research was supported by the National Science Foundation of China (No. 7 1 0 7 3 0 6 0 ) . 1 See work reports by the Supreme People's Procuratorate of China to the Congress, available at http://www.spp.gov.cn/site2006/region/00018.html. 2 A Chief Judge of the Supreme Court suggested that the legal minimum bribe be increased. See http://www.chinanews.com/gn/news/2009/ll-04/1946195.shtml (accessed August 14, 2014)

Determinants of Willingness to Bribe: M i c r o Evidence from the Educational Sector in C h i n a · 169

In general, corruption involves persons in government (Abed/Gupta 2002). The source of corruption can also be within-government relations: Kahana and Liu (2010) describe corruption of a type that exists in China that is internal to the government bureaucracy, whereby the need to pay bribes for promotion in the bureaucracy results in bribes extracted from the public to finance bureaucrats' internal promotion. When pervasive corruption exists, we expect a culture of corruption that extends beyond government. In China, the culture of corruption affects the educational system. In this paper we study corruption in China in college entrance examinations and the admission process for arts major students. Using data that we collected from students admitted to 139 colleges across China, we find that willingness to bribe depends on the perceived corruption level, the applicant's tolerance for corruption, academic attainment, family income, and the rank of a college the students attend. 3 Our data show that bribery for college admission is common among arts major students in China. The students and/or their parents pay to establish personal connections with admissions officials and workers and bribe the examiners for higher examination scores. A typical case detected in a musical college in central China's Hubei province reveals that students from 5 7 families paid bribes that summed up to 1,828,300 RMB in a college in 2008 and that a single bribe reached 170,000 RMB (Liu 2011). To obtain detailed information on the bribing process, we visited the colleges and universities at which corruption cases in college admission occurred, interviewed the personnel in charge of college admission and enrollment as well as the students with bribery experiences, and went to jails in order to interview the former examiners and admissions officials convicted for corruption. Appointed examiners were busiest the night before the examinations, during which large groups of students queued outside the examiners' houses for interviews, some accompanied by their tutors who were acquaintances of the examiners. A student entered to announce his or her name and registration number for the examinations, paid the bribe, and left. The interview lasted two or three minutes, only for the student's face to be remembered. Other students bribed the director of a department, a relative of the examiner, or another examiner with w h o m they had established personal connections. Students without direct personal connections bribed all potential examiners and the least-informed contacted professional intermediaries. The most capable students also engaged in bribery. Following the widely accepted definition of corruption, which is the misuse of public authority for private gains (Jain 2001), the majority of both theoretical and empirical research has focused on the incentives of public officials, which is the demand side of corruption (Tiróle 1996; Aidt 2003). Our study focuses on the behavior of bribers, that is, the supply side of corruption. Shifting the focus from the officials to the behavior of individual bribers contributes to our understanding of corruption by directing attention away from the recipients of bribes to the largely neglected bribe-giving part of the corruption transaction. An advantage of our study is the analysis of bribing behavior based on actual individual experiences. Economists have reported as many as 71 variables as significant predictors of corruption (Liu 2010), including economic, political, judicial, bureaucratic, geographical,

3

Corruption related to education exists not only in China. See Levin and Satarov (2015) on the market for academic degrees in Russia.

170 • Qijun Liu and Yaping Peng

and cultural influences, and socio-demographical quality and structure (Treisman 2000; Paldam 2002; Seldadyo/de Haan 2006; Olken/Pande 2012; Kis-Katos/Schulze 2013). A data problem usually arises because corruption and associated rent seeking and rent extraction are usually hidden from view and certainly are not reported in official statistics (Hillman 2013). Empirical studies of corruption therefore often use subjective measures in cross-country research at the macro level. Popularly used indices of corruption research include the Corruption Perception Index (CPI) compiled by Transparency International, the World Bank's Control of Corruption Index (CCI), and the index issued by the International Country Risk Guide (ICRG). The data used in our study are from reports of individual bribing experiences. In other studies at the individual level, Hillman and Schnytzer (1986) used reports of criminal prosecutions to study cases of individual bribery in the Soviet Union. Swamy et al. (2001) used micro data to study the relationship between corruption and gender. Doh et al. (2003) and Svensson (2003) used firm-level data to analyze the incidence of corruption. Olken (2009) collected micro data in fieldwork conducted in an Indonesian road building project to gauge the gap between corruption perceptions and corruption reality. Seligson (2006) and Mocan (2008) used data at the individual level from the International Crime Victim Survey (ICVS) compiled by the United Nations Interregional Crime and Justice Research Institute (UNICRI); the data was based on interviews or telephone conversations with individuals who were exposed to the risk of being required to give bribes. 4 The paper is organized as follows: Section 2 describes the institutional background of admissions policies to arts colleges in China; Section 3 describes the individual bribing decision; Section 4 presents the empirical model; Section 5 presents the survey, data; Section 6 reports and discusses the empirical results; and Section 7 concludes.

2

Entrance examinations and admissions to arts colleges in China

To enter a college, a high school student has to attend the National College Entrance Examinations, standard written examinations held once a year simultaneously throughout China in June. Universities select students by the scores in these examinations. Applicants for arts and related programs have to undergo examinations of arts skills and related knowledge in addition to the standard tests. These examinations are conducted before the National Examinations, usually from February to May, by the colleges that offer arts and related programs and sometimes by college unions in each province. Applicants for arts programs are allowed to take as many exams offered by the arts colleges as they wish. To enhance opportunities, most students attend several examinations offered by individual colleges, some even over twenty. For example, over 20,000 applicants in central China's Hubei province attended such an examination over 200,000 times in 2005, according to the Examination Administration of Hubei Province. 5 Admissions policies are different across the arts programs. Some colleges select applicants by the total scores of the standard tests and arts examinations combined; some 4

5

As is revealed in the ICVS questionnaire, some of the respondents paid the money upon request while others did not pay when demanded. The data we use in this study are different from the ICVS data in that ours arebased on bribing experiences of the survey subjects who actually paid the bribes. For a detailed description of ICVS and the data, see Seligson (2006) and the Appendix in Mocan (2008). The data are available at: http://www.unicri.it/icvs. Available at: http://www.hb.xinhuanet.com/zhuanti/2006-01/18/content_6078861.htm.

Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China • 171

set a minimum score for the standard tests and select applicants by the scores of the arts examinations; and others choose applicants by the total scores of the standard tests with a minimum score for the arts examinations. Despite those differences, the threshold scores of the standard tests required for arts programs are consistently much lower than those for other majors. This attracts many students who are weak in natural sciences and humanities even though they might have no interest in arts. Entering a college is the sole target for such students. As the arts examinations are independently conducted by the art colleges, there are opportunities to raise test scores and thus the probability of admission by bribing the examiners and the admissions staff. Most arts examinations are difficult to standardize because assessing a student's art potential is based on subjective evaluation. Moreover, a college is assigned the total number of enrollments in all programs by the State Education Commission each year but may change the number of enrollments between individual programs. An applicant who ranks lower among the candidates for one program can be admitted into another program and changed later back into the sought program. Thus both the examiners and admissions staff play a key role in college admission for applicants of arts programs and thus become targets of bribery.

3

The decision whether to pay a bribe

We view bribery as rational economic behavior. Students are engaged in a contest for admission to a college or university and evaluate expected utility when they decide whether to pay a bribe. In principle, with complete symmetry and common knowledge, the students confront a prisoners' dilemma as set out in Table 1. In the symmetric Nash equilibrium at (2,2) in Table 1, both students pay a bribe and neutralize each other's advantages. Both would be better off at (3,3) where no bribes are paid. The best personal outcome 4 is attained when the other student does not pay a bribe (and has benefit 1). Table 1 shows a 2 person game. In effect, there is a multi-person prisoners' dilemma/ Table 1 Whether to bribe as a prisoners' dilemma

Person 1 does not bribe Person 1 bribes

Person 2 does not bribe

Person 2 bribes

3, 3 4, 1

1, 4 2, 2

In practice, students also take into account non-monetary constraints, such as law and law enforcement, social conventions, and the specific institutional context (Guerrero/Rodriguez-Oreggia 2008). In our study, students have the same political, legal, cultural, social, and institutional background as they live in the same country and face the same laws and national standards for college entrance examinations. They are similar in many characteristics, such as age (around 20 years old), marital status (unmarried), education (college students), and occupation (unemployed). We therefore are not required to consider legal, political, or socio-demographical factors that may influence the individual's willingness to bribe for college admission. Asymmetries that affect the decision whether to bribe and therefore go beyond the symmetric game in Table 1 involve stu6

The same type of prisoners' dilemma arises with regard to the decision whether to participate in a rent-seeking contest {Hillman 2 0 0 9 , chapter 2) and in contests for status (Congleton 1989).

172 · Qijun Liu and Yaping Peng

dents' personal characteristics and specific conditions related to entrance examinations and admissions procedures. The prisoners' dilemma assumes common knowledge but in practice there is asymmetric information. Our investigations reveal that students or their parents search for information pertaining to the need to bribe. It is natural in a society in which corruption is endemic, as in China, for people to enquire about the necessity to bribe when they need a public service (Liu/Peng 2012). To collect information, the applicants and their parents often go to the targeted colleges and universities to consult the students, or consult intermediaries, tutors of arts training programs, or high school teachers. On the basis of this information, parents and the students form their perception of the level and scope of corruption that they confront. When they find that it is customary to bribe examiners to obtain higher exam scores, their rational response is to pay a bribe. For our empirical analysis, the prediction is that probability of giving a bribe increases when individual perceptions are that the existing level of corruption is greater. More bribes are offered when it is felt that others are also bribing. On the other hand, the prisoners' dilemma suggests that it is also rational for the parents or students to bribe examiners when others are not paying bribes, although, if others are not paying bribes, because of asymmetries in objective merit, more competent or optimistic students may not regard a bribe as necessary. The integrity of the examiners and admissions officials affects bribes. When corruption is endemic, officials have a reputation of bribe taking (Tiróle 1 9 9 6 ) . Once this reputation is established, corruption persists (Andrianova 2001). Corruption then becomes part of social norms and individuals pay bribes for any government service or to be able to conduct business. The probability of bribery thus increases, as the public level of trust in honesty of public officials decreases. In our context, we expect a negative relation between the level of the students' trust in the integrity of the examiners and admissions officials and the probability of the students giving a bribe. Personal tolerance for corruption also may influence individual willingness to pay a bribe. Based on experimental evidence, Cameron et al. (2009) examined whether higher levels of exposure to corruption promote tolerance for corruption that is reflected in norms of behavior. They found a wide variation in tolerance towards corruption across cultures but levels of tolerance and propensities to engage in actual corrupt behavior were not related. A question that we study is whether personal attitudes regarding corruption affect willingness to bribe in the same social environment. Competition on the supply side (many corrupt officials seeking to extract rents) has been suggested to deter corruption (Ades/Di Telia 1999). What might be the result of competition on the demand side (many beneficiaries of officials' decisions)? With the shortage of higher- education opportunities in China, greater competition on the demand side should increase the likelihood of corruption. When more applicants contest the limited positions available at a college, it is more likely that more students will choose to bribe, if bribery can increase the chance to be admitted to a college. We therefore expect the intensity of competition for college admission to be positively correlated with the incidence of corruption. Information transparency has been found to be negatively related to corruption (Renikka/Svensson 2 0 0 5 , 2 0 1 1 ; Olken 2007) and has been proposed as an effective means to curb corruption (Brunetti/Weder 2003; Björkman/Svensson 2009; Lindstedt/Naurin 2010; Goel et al. 2012). This literature looks however at rent extraction by government

Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China · 173

officials akin to theft f r o m public funds. In the college admission process, benefits are sought by the bribe-givers. The role of information is to guide the bribe-giver regarding examination types, implementation rules, and admissions procedures. W h e n the information is readily available, the applicants and their parents d o not need to pay bribes to obtain information. We hypothesize that the more open is access to information regarding examinations and admission and the more clearly the rules are announced to the public before examinations, the more likely that bribes in examinations and admission will be detected, and the lower the occurrence of bribery will be. It has been found that, when corruption exists, money becomes a substitute for capability (Liu/Peng 2012). An asymmetry affecting the prisoners' dilemma in Table 1 is that students weaker in academic merit have a greater incentive to give bribes. Another asymmetry is that the propensity to engage in bribery increases for students f r o m a rich family that can more afford to pay higher bribes than a p o o r family. A student with lower capability, ceteris paribus, is therefore expected to bribe, as is a student f r o m a rich family. Gender has been f o u n d to be a determinant of corruption. Dollar et al. (1999) have provided evidence that a higher representation of w o m e n in parliament is associated with lower corruption levels. Swamy et al. (2001) propose that w o m e n are usually less likely to be involved in corruption because they wish to set a good example to their children (which also implies that childless w o m e n are more corrupt than w o m e n with children). Frank and Schulze (2000) and Schulze and Frank (2003) find that w o m e n are less likely to be corrupt if they face possible detection and punishment because they are more risk averse than men. In our context, we expect a lower probability to bribe a m o n g female than male students.

4

The empirical model

The empirical model focuses on sources of asymmetry that affect willingness to bribe in a departure f r o m the symmetric prisoners' dilemma. The sources include perception of the existing corruption level, personal trust in the accountability of the examiners and admissions staff, the personal attitude towards corruption, the level of competition in the admission process, the availability of information about the examinations and admission, the student's family income, academic merit, and gender. W e postulate the following basic empirical model: Bribe, =f(X„Zt)

(1)

where Bribe, is the individual bribing decision; denotes an individual student; X, represents a g r o u p of personal characteristics including the perceived level of corruption in the specific environment, the individual trust level in the honesty of examiners and admissions workers, the individual tolerance level of corrupt activities by others, the student's family income, academic merit, and the gender of the student; and Z, refers to a g r o u p of environmental factors that have potential effects on individual willingness to bribe, including information transparency in the examination and admissions policies and the competitiveness in application for admission. Apart f r o m these influences, the individual bribing decision may also be affected by different admissions policies for arts programs in different colleges. Admissions policies change f r o m year to year in China, and each college has the right to establish its own admissions

174 • Qijun Liu and Yaping Peng

procedures,7 which may affect the willingness to bribe of applicants in different years. To capture the possible effects therein, we control for the year of enrollment of the students. The type of college can be another source of differences in willingness to bribe. In China, top colleges and universities are more attractive to most students and enjoy the privilege to be the first to select applicants. A higher intensity of competition is expected among the applicants for arts major programs as a consequence of higher quality education at these institutions and thus we expect a higher propensity of willingness to bribe for admission. To control for the possible effect of the college type on individual willingness to bribe, we use the top colleges as a reference to determine whether applicants to better educational institutions are more likely to bribe. To check the robustness of our estimations, we control for the potential effects of time variant and time invariant factors. 5

Survey and Data

5.1

Survey design and administration

The data used in this study are collected from a survey administered in early 2 0 1 0 . We designed a survey similar to the International Crime Victim Survey by the UN Center for International Crime Prevention (Seligson 2006), consisting of 27 questions. As corruption is a highly sensitive topic, we chose to use scaled answers for questions of high privacy. We also tried to reduce the sensitivity by avoiding offensive or morally charged wording. For example, to find out whether an applicant bribed for admission, we asked the question: "How much did you or your parents spend on personal connections with the examiners and admissions workers?" instead of asking "How much did you bribe for admission?" In fact, we needed only a yes or no answer to this question, but five choices were offered so that the respondent would focus on the selection of the options. 5.2

The misreporting problem of surveys

Survey data can be compromised by misreporting (Bertrand/Mullainathan 2 0 0 1 ; KisKatos/Schulze 2013). Respondents can answer expressively in line with an identity that they wish to portray rather than answer truthfully (Hillman 2010). We took two measures to address misreporting. First, all respondents were guaranteed anonymity and confidentiality at the beginning of the questionnaire. Second, we employed a validity detection technique in the construction of the questionnaire. A few seemingly unrelated questions were inserted in different places in the questionnaire so that the students might not notice their relation once they started to answer the questions. For example, one question asked about the total costs of attending the examinations including registration fees, accommodation costs, and extra money spent on personal connections. Other questions asked for the same information, either separately or simply repeated a previous question, which allowed us to find inconsistencies pointing towards misreporting. We selected these questionnaires out of the sample and excluded them. Still, the problem of misreporting could persist in our survey responses, given the low cost of not reporting the truth and the subjective benefit from describing oneself in line with a pleasing identity. On the other hand, we do not believe that, in our sample, students might be led to deny that they paid bribes when in fact bribes were paid, because paying of bribes is the norm of behavior.

7

For a description of the admissions policies and procedures for arts major students, see Section 3.

Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China · 175

5.3

Selection bias

Another concern in connection with survey data is selection bias. To begin, our sample includes only successful applicants to colleges and universities. There can be selection bias among admitted students. To reduce the latter bias, we took two steps in subject selection. First, we chose the sample colleges by dividing the country into seven geographical regions. For each region, we selected the colleges from each of the four types that exist in China 8 by the ratio of colleges and universities that offer arts programs and identified 139 colleges and universities, or 11.8% of the relevant higher educational institutions across China. 9 In the second step, we chose the respondents from these sampled colleges, with five students randomly selected from each grade. The students were called into a classroom. The questionnaire was distributed and collected 25 minutes later when the students had finished answering the questions. 5.4

Survey results and data

From our survey, we received 2,780 returned questionnaires by arts program students in 139 colleges across China. To check for validity, we examined the answers and dropped those questionnaires with pattern choices - the same choice for a cluster of consecutive questions - and those that failed to pass the misreport detection mentioned above. After this procedure, we had 1,541 observations, 55.4% of the original sample. Misreporting may still exist in these questionnaires, but the degree of misreporting should have been substantially reduced. The definitions of the variables and the statistics of the data are reported in Table 2. In addition to the results of our survey, we also used official statistics in our study. In order to measure the competitiveness in the college admission process, the number of applicants and that of enrollments for each of the four years in the colleges concerned were required. We obtained those data from the Ministry of Education of China. Each year, the Ministry of Education of China issues enrollment quotas to the colleges and universities, specifying the number of students to be admitted in a program. After the admission and enrollment, the colleges and universities are expected to report the final enrollments to the Ministry. However, some colleges, especially in the Top Class, often change the enrollments between programs as long as the total enrollments are consistent with the quotas. The number of applicants to each program should also be sent to the Ministry of Education, but is not always reported. In the end, we have only 1,117 observations for the variable of competitiveness (see Table 2).

8

9

T h e T o p Class of 4-year colleges and universities, the Second Best Class of 4-year colleges and universities, the T h i r d Class of 4-year colleges and universities, and 3-year colleges. T h e higher educational institutions surveyed include: 15 colleges and universities in Hubei Province and 11 in H e n a n Province in Central China, 20 in Beijing in N o r t h e r n China, 7 in Jilin Province and 11 in Liaoning Province in N o r t h e a s t e r n China, 12 in Shaanxi Province and 2 in Xinjiang A u t o n o m o u s Region in N o r t h w e s t e r n China, 9 in Sichuan Province and 4 in C h o n g q i n g City in Southwestern China, 12 in G u a n g d o n g Province in Southern China, and 10 in Shanghai, 15 in Jiangsu Province, and 11 in S h a n d o n g Province in Eastern C h i n a .

176 · Qijun Liu and Yaping Peng

Table 2 Variable definitions and descriptive statistics Variable name

Definition

Dependent variable Bribes Dummy variable =0 if the answer to the question: "How much did you or your parents spend on personal connections with the examiners and admissions workers?" is "zero", and =1 otherwise. Explanatory variables Corruption perception Level 1 Dummy variable =1 if the answer to the question: "Is it common for your classmates to pay extra money for admission by your knowledge?" is "no", and =0 otherwise.

Obs.

Mean

Std. Dev.

1541 0.5308241

0.499211

1541 0.0902012

0.2865627

Level 2

Dummy variable =1 if the answer to the question is "rare", and =0 otherwise.

1541 0.2316677

0.4220348

Level 3

Dummy variable =1 if the answer to the question is "common", and =0 otherwise.

1541 0.3848151

0.4867095

Level 4

Dummy variable =1 if the answer to the question is "very common", and =0 otherwise.

1541 0.2783907

0.4483523

Level 5

Dummy variable =1 if the answer to the question is "no exception", and =0 otherwise.

1541 0.0149254

0.1212937

Dummy variable =1 if the answer to the question: "How much do you trust in the honesty of the examiners and admissions workers?" is "absolutely", and =0 otherwise.

1541

0.2283644

Level 2

Dummy variable =1 if the answer to the question is "quite trust", and =0 otherwise.

1541 0.1505516

0.357727

Level 3

Dummy variable =1 if the answer to the question is "not so trust", and =0 otherwise.

1541 0.3841661

0.4865554

Level 4

Dummy variable =1 if the answer to the question is "little", and =0 otherwise.

1541 0.312135

0.463515

Level 5

Dummy variable =1 if the answer to the question is " n o " , and =0 otherwise.

1541 0.073329

0.2607604

Dummy variable =1 if the answer to the question: "Do you consider it tolerable if you find others bribed for admission?" is " n o " , and =0 otherwise.

1541 0.3011032

0.4588864

Level 2

Dummy variable =1 if the answer to the question is "a little intolerable", and =0 otherwise.

1541 0.3861129

0.487015

Level 3

Dummy variable =1 if the answer to the question is "understandable", and =0 otherwise.

1541 0.2167424

0.4121594

Level 4

Dummy variable =1 if the answer to the question is "very understandable", and =0 otherwise.

1541 0.0778715

0.2680563

Level 5

Dummy variable =1 if the answer to the question is "absolutely understandable", and =0 otherwise.

1541 0.01817

0.1336093

Trust Level 1

Tolerance Level 1

0.055159

Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China · 177

Table 2 Continued Variable name

Definition

Family income Below Dummy variable =1 if the student's annual family 10000RMB income is below 10 000 RMB, and =0 otherwise.

Obs.

Mean

Std. Dev.

1541 0.2907203

0.4542421

1541

0.449708

0.4976258

60001Dummy variable =1 if the student's annual family 110000RMB income falls between 60001 and 110000 RMB, and =0 otherwise.

1541 0.1791045

0.3835643

110001Dummy variable =1 if the student's annual family 160000RMB income falls between 110001 and 160000 RMB, and =0 otherwise.

1541 0.0519143

0.2219261

Over 160000 RM Β

Dummy variable =1 if the student's annual family income is over 160 000 RMB, and =0 otherwise.

1541 0.0285529

0.1666002

Female

Dummy variable =1 if the respondent is a female student, and =0 otherwise.

1541 0.5723556 0.4948976

Capability

Years of training in arts skills.

1541 3.084491

2.407111

1541 0.1518494

0.3589914

1000160000RMB

Dummy variable =1 if the student's annual family income falls between 10001 and 60000 RMB, and =0 otherwise.

Information access Level 1 Dummy variable =1 if the answer to the question: "Is it easy for you and your parents to obtain information of admissions requirements and process?" is "easy", and =0 otherwise. Level 2

Dummy variable =1 if the answer to the question is "quite easy", and =0 otherwise.

1541 0.4120701

0.4923674

Level 3

Dummy variable =1 if the answer to the question is "not so easy", and =0 otherwise.

1541 0.3848151

0.4867095

Level 4

Dummy variable =1 if the answer to the question is "quite difficult", and =0 otherwise.

1541 0.0421804

0.2010658

Level 5

Dummy variable =1 if the answer to the question is "very difficult", and =0 otherwise.

1541 0.009085

0.0949122

Competition

Ratio of applicants to enrollments.

1117 17.75389

18.37534

Dummy variable =1 if the student is from a top class college/university, and =0 otherwise.

1541 0.3043478

0.46028

2nd best

Dummy variable =1 if the student is from a second class college/university, and =0 otherwise

1541 0.3497729

0.477053

3rd class

Dummy variable =1 if the student is from a third class college/university, and =0 otherwise

1541 0.277093

0.447708

3-year

Dummy variable =1 if the student is from a 3-year college/university, and =0 otherwise

1541 0.0687865

0.253173

College type Top class

178 · Qijun Liu and Yaping Peng

6

Empirical results and discussion

To identify the influences on reported individual willingness to bribe, we use probit estimation. We first run a regression on the independent variables indicating individual characteristics. Then, we include in the regression the environmental factors that have potential effects on individual bribing decision. In the third regression, we control for the college type to examine if corruption incidence varies across different college categories. To capture the potential effects of time invariant factors and time variant factors such as variation in the admissions policies on individual willingness to bribe, we introduce the province dummies and dummies of time of the year in the fourth regression. The results are represented in Table 3. In the full model (Column 4 in Table 3), we find that the first significant determinant of corruption incidence is the individual perceived level of corruption in society. Compared with the base group of the students who do not believe that there is corruption in the arts examinations and admissions process, the students who have perceived corruption from the slight level to the highest level of "no exception" have positive probability to paying a bribe for admission. Detailed observation indicates that this probability does not rise consistently with the perceived levels of corruption: those who hold the view that there is only "a little" corruption actually have a higher probability of paying a bribe than those who believe that corruption is common in the examinations and admissions process (30.6% vs. 21.6%), although the other two groups that perceived corruption to be "very common" and "no exception" have higher bribing probabilities (37.8% and 56.0%, respectively) than the rest of the students (see Column 4 in Table 3). These results show that the existing level of corruption in a society has significant positive effects on the individual bribing decision and hence corruption incidence. We therefore add to evidence on a self-reinforcing mechanism in corruption (Aidt 2003; Mishra 2006) as well as for the theory of corruption persistence suggested by Tiróle (1996). 10 Does individual trust affect bribery at the micro level? A high level of trust is believed to reduce corruption (Bjernskov 2003; Liu/Peng 2012). Our regression results imply that this is not always the case. Of the four estimates of the trust variable, only the last one has a significant value of 0.149, which means that only those students with the least trust in the integrity of the examiners and admissions staff have a probability of 14.9% to bribe when compared with those with full trust. There is no significant evidence for the other groups of students of a higher probability to bribe for admission as their distrust increases, as none of the other estimates is statistically significant. Does personal tolerance for corruption play a role in bribery? Our regression results suggest that it does. Three estimates have positive values statistically significant at the level of 1%, with the exception of the last one, which also has a positive value but is insignificant at the conventional level.11 And the estimated values increase as the level of tolerance increases: when compared with the base group with no tolerance with corruption, the other three groups of students with little median and much tolerance with corruption have higher bribing probabilities of 8.6%, 10.9% and 21.8%, respectively.

10

11

However, in distinction to previous studies on corruption persistence in low-income countries with pervasive corruption (Haque/Kneller 2009; Olken/Pande 2011), our study offers evidence from the supply side rather than the demand side. The last group with the absolute tolerance with corruption consists of 28 students, about 1.8% of the sample. All four estimates of the variable tolerance have positive values at the significance level of 1% when we regress bribing decision on tolerance only.

Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China · 179

Table 3 Determinants of willingness to bribe at the individual level: probit models, average marginal effects Independent variables

1

Individual characteristics Corruption perception Level 1 Level 2 Level 3 Level 4 Level 5

0.278 0.185 0.358 0.431

2

(0.049)*" (0.047)*" (0.048)*" (0.111)*"

3

0.294 0.206 0.370 0.558

Trust Level 1 Level 2 Level 3 Level 4 Level 5

0.014 -0.055 -0.088 -0.162

Tolerance Level 1 Level 2 Level 3 Level 4 Level 5

0.085 0.129 0.202 0.132

(0.029)* (0.034)* (0.049)* (0.095)

Family income Below 10000RMB 10 001 -60 000RMB 60001-110 000RMB 110001-160000RMB Over 160000RMB

0.060 0.132 0.181 0.034

(0.029)* 0.039 (0.036)*" 0.143 (0.059)*** 0.210 (0.077) -0.032

Gender Female Capability Environmental factors Information access Level 1 Level 2 Level 3 Level 4 Level 5 Competition College type Top class 2nd best 3rd class 3-year Year dummy Province dummy Log likelihood Observations

(0.057)*** (0.054)*** (0.056)*** (0.144)***

0.298 0.208 0.374 0.572

4

(0.057)*" (0.054)*" (0.056)*" (0.145)*"

0.306 0.216 0.378 0.560

(0.058)"* (0.055)*" (0.057)*** (0.145)"*

(0.054) 0.003 (0.062) 0.001 (0.062) 0.009 (0.062) (0.048) -0.049 (0.056) -0.051 (0.056) -0.050 (0.055) (0.049)* -0.066 (0.058) -0.061 (0.058) -0.053 (0.055) (0.064)" -0.156 (0.079)" - 0 . 1 5 6 (0.079)** -0.149 (0.080)*

0.078 0.103 0.210 0.046

(0.035)* (0.041)* (0.059)* (0.104)

0.080 0.105 0.211 0.050

(0.034) 0.042 (0.044)*** 0.145 (0.068)*** 0.216 (0.090) -0.018

(0.035)* (0.041)* (0.059)* (0.103)

0.086 0.109 0.218 0.094

(0.035)* (0.041)* (0.059)* (0.104)

(0.034) 0.040 (0.034) (0.044)*" 0.132 (0.044)* (0.069)*" 0.217 (0.069)* (0.090) -0.003 (0.09)

0.013 (0.008) 0.006 (0.009) 0.007 (0.009) 0.007 (0.010) -0.014 (0.005)***-0.014 (0.006)" -0.014 (0.006)** -0.013 (0.006)*

0.031 -0.003 0.130 0.073 0.000

(0.044) (0.047) (0.081) (0.181) (0.001)

0.028 -0.010 0.127 0.069 0.001

(0.044) (0.048) (0.081) (0.181) (0.001)

0.064 (0.040) 0.074 (0.050) 0.170 (0.075)** -983.75364 1541

-710.38061 1117

-707.59713 1117

0.016 -0.006 0.127 0.041 0.002

(0.044) (0.047) (0.081) (0.181) (0.001)

0.106 (0.048) 0.139 (0.059) 0.158 (0.079) YES YES -696.69192 1117

Standard errors are in parentheses. ***, ** and * represent levels of significance at 1%, 5 % and 10%, respectively.

180 · Qijun Liu and Yaping Peng

These results provide the missing link in the study by Cameron et al. (2009): the personal attitude towards corruption, or rather corruption tolerance, is a significant factor that has a positive relation to corruption incidence. We predicted that richer families are more likely to bribe than the poor. The regression results do not, however, support this prediction. Rather, we find that individuals with middle family incomes have a higher probability to bribe. When compared with the group whose annual family income is below 10,000 RMB, the students from families with annual income between 60,001 to 110,000 RMB and between 110,001 to 160,000 RMB have bribing probabilities of 13.2% and 21.7%, respectively. The estimate for the students from the modest income (between 10,001 to 60,000 RMB) families is positive but not statistically significant. The estimate for the students from rich families with annual income over 160 000 RMB is not statistically significant and the sign of the value changes across regressions. A possible explanation is expressive, with poor and rich wishing to associate their selection with merit rather than bribery. Investigation of the reason for this result awaits further research. Previous studies indicate that females are less likely to engage in corruption (Dollar et al. 1999; Swamy et al. 2001; Mocan 2008) 12 . We do not find significant support for this conclusion. The coefficients across all four regressions are positive but are not statistically significant. What is the relationship between capability and bribing probability? Our estimated results show a negative relation. The variable capability, proxied by the years of training received in arts skills, has negative values in all four regressions at the significance level of 5 % . By the estimate in the full model, every additional year of training that a student receives reduces the bribing probability by 1.3 percentage points. Considering that the average training spans 3.08 years for the sample students, capability reduces an average student's bribing probability by 4 . 0 % . Regarding the effects of information transmission and transparency on corruption (Björkman/Svensson 2009; Lindstedt/Naurin 2010; Goel et al. 2012), the prediction is that students who had the most difficulty of access to information have the greatest probability of paying bribes for admission. However, none of the estimates for the variable in our regressions is statistically significant. Nor do we find evidence for the relationship between competition and corruption. These results imply that the environmental factors that we consider do not have significant effects on individual willingness to bribe. H o w ever, the level of competition is found to be positively correlated with bribing probability: an increase in the ratio of enrollment quota to the number of applicants by 1 % increases individual bribing willingness by 0.2 percentage points. We find difference in corruption incidence across the college categories. The regression results show that corruption increases as the college rank decreases, which means that corruption is more prevalent in colleges of lower ranks. Compared with the base group, that is, the students in the top colleges, the students applying for admission into the second class of colleges have a bribing probability that is 10.6 percentage points higher; those applying for admission into the third class colleges have a bribing probability that is 13.9 percentage points higher than the base probability; and the students seeking to enter the 3-year colleges have the highest differential probability of 15.8 percentage points to paying a bribe for admission. 12

O u r results are thus in line with Frank/Schulze 2 0 0 0 and Schulze/Frank 2 0 0 3 w h o find that women are equally likely to bribe if they do not face the risk of detection and punishment.

Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China · 181

The regression results are net of the potential effects of both time-variant and timeinvariant factors, as we have controlled for the year dummies and province dummies.

7

Conclusions

W e have studied the determinants of the willingness to bribe at the micro level f r o m the supply side of corruption. O u r sample consists of students w h o were admitted to a college or university. W e d o not k n o w the bribing behavior of those w h o were not admitted. T h e prisoners' dilemma as a symmetric game with c o m m o n knowledge predicts that all students seeking admission to a college or university will engage in bribery. O u r empirical study has investigated asymmetries in willingness to bribe. Using a unique dataset for direct personal bribing experiences, we find that reported individual willingness to bribe differs according to personal characteristics and does not depend on the specific environment. The perceived level of corruption, the personal attitude towards corruption, capability in academic merit, and the rank of a college are the significant predictors of bribery that occurs in the examinations and admissions process for arts students in China. W e find n o significant evidence for a negative relation between information transparency and corruption. N o r d o we find evidence that females engage less in bribery. Individual willingness to bribe is independent of the intensity of competition a m o n g the potential bribers. Only those individuals with absolute trust in the accountability of public servants are less likely to bribe than others. The finding of this study that is perhaps the most interesting is that the individuals with middle family incomes are more likely to bribe t h a n either the p o o r or the rich - or perhaps are more likely to report their bribery. If there has not been misrepresentation in reporting, perhaps the p o o r could not afford to bribe and the rich may have felt that, because of their connections through privileged tutoring arrangements, bribes were not required. W e noted the possibility that the responses of the p o o r a n d rich are expressive. This implies that neither the p o o r nor the rich were willing to admit that bribery, and not exclusively merit, was the basis for their examination grades and their successful admission to college. The students f r o m middle-income families are then more pragmatic in acknowledging the conditions of endemic corruption in which the society functions.

References Abed, G.T., S. Gupta (eds.) (2002), Governance, Corruption, and Economic Performance. Washington DC: International Monetary Fund. Ades, Α., R. Di Telia (1999), Rents, competition, and corruption. American Economic Review 89(4): 982-993. Aidt, T.S. (2003), Economic analysis of corruption: A survey. Economic Journal 113: F632-F652. Andrianova, S. (2001), Corruption and reputation. Scottish Journal of Political Economy 48(3): 245-259. Bertrand, M., S. Mullainathan (2001), Do people mean what they say? Implications for subjective survey data. American Economic Review, Papers and Proceedings, 91(2): 67-72. Björkman, M., J. Svensson (2009), Power to the people: Evidence from a randomized field experiment on community-based monitoring in Uganda. Quarterly Journal of Economics 124 (2): 735-769. Bjornskov, Ch. (2003), Corruption and social capital. Working Papers 03-13, University of Aarhus, Aarhus School of Business, Department of Economics.

182 · Qijun Liu and Yaping Peng

Brunetti, Α., Β. Weder (2003), A free press is bad news for corruption. Journal of Public Economics 87(7-8): 1 8 0 1 - 1 8 2 4 . C a m e r o n , L. A. Chaudhuri, N . Erkal, L. G a n g a d h a r a n (2009), Propensities to engage in and punish corrupt behavior: Experimental evidence from Australia, India, Indonesia and Singapore. Journal of Public Economics 93(7-8): 8 4 3 - 8 5 1 . Chen, K., Q . Liu (2015), Economic development and corruption in China in the shadow of rent seeking. In: Roger D. Congleton and Arye L. Hillman (eds.), C o m p a n i o n to Political Economy of Rent Seeking, Cheltenham UK and N o r t h a m p t o n MA: Edward Elgar. Congleton, R.D. (1989), Efficient status seeking: Externalities, and the evolution of status games. Journal of Economic Behavior and Organization 11(2): 1 7 5 - 1 9 0 ; reprinted in pp. 6 9 7 - 7 1 2 : R.D. Congleton, A.L. Hillman, K.A. Konrad (eds.) (2008), Forty Years of Research on Rent Seeking 2 - Applications: Rent Seeking in Practice, Heidelberg: Springer. D o h , J.P., P. Rodriguez, K. Uhlenbruck, J. Collins, L. Eden (2003), Coping with corruption in foreign markets. Academy of M a n a g e m e n t Executive 17(3): 1 1 4 - 1 2 9 . Dollar, D., R. Fisman, R. Gatti (1999), Are w o m e n really the fairer sex? Corruption and w o m e n in government. Journal of Economic Behavior and Organization 46(4): 4 2 3 - 4 2 9 . Frank, B., G.G. Schulze (2000), Does economics make citizens corrupt? Journal of Economic Behavior and Organization 43(1): 1 0 1 - 1 1 3 . Glaeser, E.L., J. Scheinkman, A. Shleifer (2003), The injustice of inequality. Journal of M o n e t a r y Economics 50(1): 1 9 9 - 2 2 2 . Goel, R.K., M.A. Nelson, M.A. Naretta (2012), The internet as an indicator of corruption awareness. European Journal of Political Economy 28(1): 6 4 - 7 5 . Guerrero, M.A., E. Rodriguez-Oreggia (2008), O n the individual decisions to commit corruption: A methodological complement. Journal of Economic Behavior and Organization 65(2): 357-372. H a q u e , M.E., R. Kneller (2009), Corruption clubs: Endogenous thresholds in corruption and development. Economics of Governance 10(4): 3 4 5 - 3 7 3 . Hillman, A.L. (2009), Public Finance and Public Policy: Responsibilities and Limitations of Government (2nd Ed.), N e w York NY: Cambridge University Press. Hillman, A.L. (2010), Expressive behavior in economics and politics. European Journal of Political Economy 26: 4 0 3 - 4 1 8 . Hillman, A.L. (2013), Rent seeking. Pp. 3 0 7 - 3 3 0 in: M . Reksulak, L. Razzolini, W.F. Shughart II (eds.), The Elgar C o m p a n i o n to Public Choice (2nd Edition), Cheltenham UK and N o r t h a m p t o n MA: Edward Elgar. Hillman, A.L., A. Schnytzer (1986), Illegal economic activities and purges in a Soviet-style economy: A rent-seeking perspective. International Review of Law and Economics 6(1): 8 7 - 9 9 ; reprinted in pp. 5 4 5 - 5 5 7 : R.D. Congleton, A.L. Hillman, Κ. Konrad (eds.) (2008), Forty Years of Research on Rent Seeking 2 - Applications: Rent Seeking in Practice, Heidelberg: Springer. Jain, A.K. (2001), Corruption: A review. Journal of Economic Surveys 15(1): 7 1 - 1 2 1 . Kahana, N., Q . Liu (2010), Endemic corruption. European Journal of Political Economy 26(1): 82-88. Kis-Katos, K., G.G. Schulze (2013), Corruption in Southeast Asia: A survey of recent research. Asian-Pacific Economic Literature 27(1): 7 9 - 1 0 9 . Levin, M.I., G.A. Satarov (2015), Rent seeking through control of the state in Russia. In: R.D. Congleton A.L. Hillman (eds.), C o m p a n i o n to the Political Economy of Rent Seeking, Cheltenham UK and N o r t h a m p t o n M A : Edward Elgar. Lindstedt, C., D. Naurin (2010), Transparency is not enough: M a k i n g transparency effective in reducing corruption. International Political Science Review 31(3): 3 0 1 - 3 2 2 . Liu, Q. (2010), H o w economists study corruption. Journal of Anticorruption Research 2: 1 1 0 - 1 1 8 . Reprinted in pp. 3 2 8 - 3 4 4 : Y. Z h o u , G. Chen (eds.) (2011), Research on Anticorruption (Series 10). H a n g z h o u : Zhejiang University Press, H a n g z h o u . Liu, Q . (2011), Corruption in college entrance examinations for arts m a j o r students and admissions process. Project Report to the Ministry of Education of China, Beijing.

Determinants of Willingness to Bribe: Micro Evidence from the Educational Sector in China · 183

Liu, Q., Y. Peng (2012), Behavioral environment, social capital and the evolutionary mechanism of corruption equilibrium. Comparative Economic and Social Systems 163(5): 1 4 5 - 1 5 8 . Mishra, A. (2006), Persistence of corruption: Some theoretical perspectives. World Development 34(2): 3 4 9 - 3 5 8 . M o c a n , Ν . (2008), W h a t determines corruption? International evidence from microdata. Economic Inquiry 46(4): 4 9 3 - 5 1 0 . Olken, B.A. (2007), Monitoring corruption: Evidence from a field experiment in Indonesia. Journal of Political Economy 115(2): 2 0 0 - 2 4 9 . Olken, B.A. (2009), Corruption perceptions vs. corruption reality. Journal of Public Economics 93: 9 5 0 - 9 6 4 . Olken, B.A., R. Pande (2012), Corruption in developing countries. Annual Review of Economics 4: 4 7 9 - 5 0 9 . Paldam, M . (2002), The cross-country pattern of corruption: Economics, culture and the seesaw dynamics. European Journal of Political Economy 18: 2 1 5 - 2 4 0 . Reinikka, R., J. Svensson (2005), Fighting corruption to improve schooling: Evidence from a newspaper campaign in Uganda. Journal of the European Economic Association 3(2-3): 259-267. Reinikka, R., J. Svensson (2011), The power of information in public services: Evidence from education in Uganda. Journal of Public Economics, 95(7-8): 9 5 6 - 9 6 6 . Schulze, G.G., Β. Frank (2003), Deterrence versus intrinsic motivation: Experimental evidence on the determinants of corruptibility. Economics of Governance 4(2): 1 4 3 - 1 6 0 . Seldadyo, H., J. de H a a n (2006), The determinants of corruption: A literature survey and new evidence. Paper prepared for the 2 0 0 6 EPCS Conference, T u r k u , Finland, 2 0 - 2 3 April 2006. Available at http://conferences.wcfia.harvard.edu/sites/projects.iq.harvard.edu/ files/go v2126/files/seldadyo_determinants_corruption.pdf Seligson, M.A. (2006), The measurement and impact of corruption victimization: Survey evidence from Latin America. World Development 34 (2): 381—404. Svensson, J. (2003), W h o must pay bribes and h o w much? Evidence from a cross section of firms. Quarterly Journal of Economics 118(1): 2 0 7 - 2 3 0 . Swaleheen, M . (2011), Economic growth with endogenous corruption: An empirical study. Public Choice 146(1-2): 2 3 - 4 1 . Swamy, Α., St. Knack, Y. Lee, O . Azfar (2001 ), Gender and corruption. Journal of Development Economics 64: 2 5 - 5 5 . Tiróle, J. (1996), A theory of collective reputation (with applications to the persistence of corruption and to firm quality). Review of Economic Studies 63(1): 1 - 2 2 . Treisman, D. (2000), The causes of corruption: A cross-national study. Journal of Public Economics 76(3): 3 9 9 - ^ 5 7 . Corresponding author: Qijun Liu, School of Public Administration, H u a z h o n g University of Science and Technology. Postal address: 1037, Luoyu Rd., H o n g s h a n District, W u h a n , China Postal code: 4 3 0 0 7 4 . qij un_liu@ 163 .com Yaping Peng, School of Public Administration, H u a z h o n g University of Science and Technology.

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2015) Bd. (Vol.) 2 3 5 / 2

Religious Loyalty and Acceptance of Corruption Moamen Gouda* Graduate School of International and Area Studies, Hankuk University of Foreign Studies, Seoul Sang-Min Park Technische Hochschule Mittelhessen - University of Applied Sciences JEL A1; DO; D1; D7; K4; Z1 Religion; corruption; institutions; preferences.

Summary This study investigates the relationship between religiously-induced internalized values of individuals and their specific attitudes regarding the acceptance of corruption. The dataset on which our study is based was collected by the World Values Survey f r o m 141,326 individuals in 78 countries surveyed during a period of 13 years. We propose that individual attitudes towards corruption and religion are associated given certain societal and institutional contexts. O u r results show that although there is a negative and statistically significant effect of religiosity on the acceptance of corruption on the individual level, this effect is small. We find that there is a threshold value of religiosity below which corruption is m o r e easily accepted by individuals. O u r interpretation for this result is simple: individuals with minimal religiosity are generally less constrained by religious norms; specifically, religious n o r m s that are opposed to corruption are less binding on these individuals, resulting in them having a greater propensity to accept corruption. Religiosity, therefore, does lower the acceptance of corruption only when it exceeds a certain threshold for a specific individual.

1

Introduction

T h e c o n c e p t of c o r r u p t i o n h a s n o universally a c c e p t e d d e f i n i t i o n ( B h a t t a r a i 2 0 0 9 ) . N e v ertheless, i n t e r n a t i o n a l o r g a n i z a t i o n s h a v e r e a c h e d a de f a c t o c o n s e n s u s a b o u t t h e grave n e g a t i v e effects of c o r r u p t i o n o n b o t h g l o b a l a n d local levels. T h e W o r l d B a n k classifies c o r r u p t i o n as " t h e single g r e a t e s t o b s t a c l e t o e c o n o m i c a n d social d e v e l o p m e n t " ( D u a s a 2 0 0 8 : 2), w h i l e T r a n s p a r e n c y i n t e r n a t i o n a l identifies c o r r u p t i o n in its m i s s i o n s t a t e m e n t

* The authors thank Gouda Abd Al-Khalek, Jerg Gutmann, Shima'a Hanafy, Bernd Hayo, Michael Kirk, Johann Graf Lambsdorff, Helmut Leipold, Marcus Marktanner, Mustapha Κ. Nabli, S. Ramachandran, Günther G. Schulze, Stefan Voigt as well as two anonymous referees for their helpful comments and suggestions. Special thanks goes to Jeffrey Nugent for his enlightening comments. We also thank the participants at the 12th International Meeting of the Middle East Economic Association (MEEA) in Speyer, Germany, The Fourteenth Mediterranean Research Meeting (2013) in Mersin, Turkey, the 2013 Association for the Study of Religion, Economics, and Culture (ASREC) meeting in Washington, D.C., as well as Economic Research Forum (ERF) 20th Annual Conference in Cairo, Egypt for their helpful comments. This work was supported by Hankuk University of Foreign Studies Research Fund. The usual disclaimer applies.

Religious Loyalty and Acceptance of Corruption · 185

as " o n e of the greatest challenges of the contemporary world. It undermines good government, fundamentally distorts public policy, leads to the misallocation of resources, harms the private sector and private sector development and particularly hurts the p o o r " (Transparency International 2011). In a world economy that was w o r t h USD 30 trillion in 2 0 0 1 - 2 0 0 2 , The World Bank estimates that a b o u t USD 1 trillion are paid out in bribes each year, globally (World Bank 2004). Transparency International, in its annual report on global corruption, calculated in 2 0 0 4 that, worldwide, public procurement lost at least USD 4 0 0 billion per year due to bribery (Transparency International 2006). This study investigates the relationship between religiously-induced internalized values of individuals and their specific attitudes regarding the acceptance of corruption. Based on the principles of the N e w Institutional Economics (NIE), we propose that individual attitudes t o w a r d corruption and religion are associated with certain given societal and institutional contexts. We use data collected by the World Values Survey (WVS) f r o m 141,326 individuals in 78 countries surveyed in 13 different years. O u r results show that, although there is a negative and statistically significant effect of religiosity on the acceptance of corruption by individuals, the effect is small. Interestingly, we find that those people w h o have a value of religiosity below a certain threshold have a greater acceptance of corruption. O u r interpretation for this result is simple: individuals with minimal religiosity are generally less restricted by religious norms, including those norms that are opposed to corruption, resulting in them having a greater propensity to accept corruption. Religiosity does lower the acceptance of corruption only when it exceeds the threshold for a specific individual. W e find that religiosity's effect on the acceptance of corruption does not systematically differ a m o n g individuals of different religious denominations. In addition, our results show that the more accepted corruption is at the societal level, the less of a mitigating effect religiosity has on the individual's acceptance of corruption. This study is divided into six sections. The next section provides a multi-disciplinary literature review on religiosity and corruption while section three presents the theoretical basis for this study as well as our hypotheses. This is followed by a description of our methodology in section four, the empirical results in section five and conclusions in section six.

2

Literature Review

There are t w o previous studies that analyze individual attitudes t o w a r d corruption and religiosity. Guiso et al. (2003) (hereafter GSZ) investigate the effect of religion on people's economic attitudes, while controlling for country-fixed effects. GSZ use data collected by the World Values Survey (WVS) in three surveys (1981-1984, 1990-1993 and 19951997) which covered 66 countries. G S Z employ three distinct measures of religiosity simultaneously in each estimation: • "Raised religiously"; the answer takes on a value of 1 when the respondent answered positively to the question "Were you brought up religiously at h o m e ? " •

"Currently religious"; the answer takes on a value of 1 when the respondent affirms having attended religious services (apart f r o m weddings, funerals and christenings) at least once a year.

186 · Moamen Gouda and Sang-Min Park

• "Actively religious"; the answer takes on a value of 1 when the respondent affirms having attended religious services (apart from weddings, funerals and christenings) at least once a week. As for the dependent variables, G S Z categorize economic attitudes into six categories; attitudes toward cooperation, women, government, thriftiness, the market economy and its fairness, and legal rules. G S Z base their measurements of the latter category on the answer to a specific question asking if the respondents think that certain illegal acts are justifiable using a scale between 1 (never justifiable) and 10 (always justifiable). The illegal acts in question are; claiming government benefits to which the respondent is not entitled, avoiding paying the fare on public transport, tax frauds, buying stolen goods and accepting bribes. O L S regression results measuring the effect of religiosity on acceptance of bribery, mostly find negative effects, regardless of whether they include all religious denominations or estimate separately for each denomination. However, as some of these estimated effects are not statistically significant, G S Z conclude that no inference is possible concerning which religion might be better in terms of the economic outcomes. The second study that we build upon is that of Gatti et al. ( 2 0 0 3 ) (GPR hereafter). While the study of G S Z is broader than ours with respect to the variables, G P R ' s study investigates a wider range of social effects of attitudes toward corruption. G P R also use data from the W V S , although with a much smaller sample size than that used by G S Z and us, because their analysis mainly focuses on the W V S ' third survey. Effects of religiosity are captured by including denomination dummies and a dummy for regular church attendance. G P R find that regular church attendance is negatively associated with acceptance of corruption. They also find that Catholic/Jewish respondents are characterized by a higher/lower acceptance of corruption. Our study differs from those of G S Z and G P R on several levels. (1) W e focus on the relationship between attitudes toward corruption and religiosity from an N I E perspective; (2) we use a larger sample, as we are able to include the latest wave of W V S survey responses; (3) we use a synthetic religiosity index instead of three dummy indicators or just a dummy for church attendance; (4) we treat attitudes toward corruption as a binary indicator, due to a very skewed distribution. Attempts to build a theoretical model to analyze corruption's causes and consequences are numerous (Nas et al. 1 9 8 6 ; Caiden 1 9 8 8 ; S h l e i f e r / V i s h n y 1 9 9 3 ; Mishra 2 0 0 6 ; Khan 2 0 0 6 ; Guerrero/Rodriguez-Oreggia 2 0 0 8 ; Matei/Matei 2 0 0 9 ) . These studies are not only based on economic perspectives but also on other specialist fields, including finance, public administration, sociology and political science. However, economists were generally interested in specifically modeling the relation between corruption and economic development (Macrae 1 9 8 2 ; Ehrlich/Lui 1 9 9 9 ; Barreto 2 0 0 0 ; M a u r o 2 0 0 2 ; Barreto/Alm 2 0 0 3 ; Basu 2 0 0 6 ) . As for the empirical research, a growing number of studies investigate the causes and effects of corruption across countries (Mauro 1 9 9 5 ; Ades 1 9 9 7 ; Van Rijckeghem/Weder 1 9 9 7 ; Wei 1 9 9 7 ; M a u r o 1 9 9 8 ; Lambsdorff 1 9 9 9 ; Rose-Ackerman 1 9 9 9 ; Jain 2 0 0 1 ; Herzfeld/Weiss, 2 0 0 3 ) . These studies have searched for empirical correlations between corruption and a variety of economic and non-economic determinants. H o w ever, there is still no commonly accepted theory on which to base an empirical model of the causes of corruption (Alt/Lassen 2 0 0 3 ) . Corruption was only recently incorporated in the studies of sociologists through their examination of social deviance (Naumova 2 0 0 9 ) . Marquette ( 2 0 1 0 ) asserts that the influence of religion on attitudes towards corruption is not clear, as many other factors come

Religious Loyalty and Acceptance of Corruption • 187

into the formation of these attitudes such as gender, age, education level and the nature of religion and the religious community involved. This claim is supported by the study of Hirschi and Stark (1969) who investigate the relation between church attendance and delinquent attitudes and behavior. They find that attendance at church influences neither actual delinquent acts nor attitudes towards delinquency, even amongst respondents who believe in a literal hell and the devil. The results of the latter study contradict those of Tittle and Welch (1983) wherein the demographics and the religious affiliations of residents of several US states are surveyed. The authors find that there is little or no difference between religious and non-religious respondents regarding behavior that is condemned by society as a whole, such as major theft, assault and tax evasion. However, significant differences are found when it comes to behavior that is not widely condemned by society, such as pot smoking and not standing for the national anthem. This implies that the deterrent impact of religion on attitudes concerning corrupt behavior positively correlates with the intensity of social condemnation of such behavior. Beets (2007) gives two main arguments as to why religiosity might encourage people to resist corruption: (1) the ill-treatment of others, theft and dishonesty are not compatible with adherence to a religion; (2) religion provides moral guidance to its adherents. These two arguments are supported by Treisman (2000), Brunetti and Weder (2003), Herzfeld and Weiss (2003), Braun and Di Telia (2004), Kunicova and Rose Ackerman (2005), and Lederman et al. (2005). North et al. (2013) argue that a religious society is expected to demonstrate a higher degree of morality than a non-religious one. Therefore, it is assumed that in countries where religion plays an essential role in the lives of most people, civic employees, as well as others, are likely to obtain their ethical framework at least partly from their religion: this, in turn, will directly influence their tendency to commit corrupt acts. Religion is said to provide its followers with a code of ethics, some of which are of significant importance in the battle against corruption. However, as Marquette (2010) argues, there is a logical error in this argument because it presupposes that all religions emphasize the same moral codes. In actuality, a considerable body of literature proposes that followers of different religions - or even sects of a religion - hold divergent opinions on what constitutes morality (Guiso et al. 2003; Al-Marhubi 2004; Durkheim 1912/1915; Weber 2010; Jagodzinski 2009). Luxmoore (1999) attempts to rebut this claim by assuming that because certain values such as fairness and honesty are basic teachings of most - if not all - religions, these same religions can therefore be used as an antidote for corruption. In Table 4, we document how some of the main sources of the major world religions stress the immorality of theft and bribery. Contradicting the assumption by Luxmoore (1999), Marquette (2010) "many of the most corrupt countries in the world (according to Transparency International's Corruption Perception Index) also rank high in terms of religiosity (using indicators such as those used by the Pew Global Attitudes Project)". This apparent contradiction has two main explanations. First, in countries where a high level of both religiosity and corruption exists, other endogenous factors may be affecting them both: for example, the presence of a corrupt theocratic leadership in a certain country. Investigating such endogenous factors is beyond the scope of most literature focused on the religion-corruption nexus. However, some control variables relating to the political and social environment in sampled countries are taken into consideration. Second, the level of religiosity might not be the only important explanation when investigating its relation to the perceived corruption levels and the type of religion. For that reason, various studies use the type of religion as an

188 · Moamen Gouda and Sang-Min Park

explanatory variable and show that it has a significant effect on the level of corruption in the sampled countries (La Porta et al. 1999; Treisman 2 0 0 0 ; Paldam 2 0 0 1 ; Beets 2 0 0 7 ; M u t a s c u 2010). However, other studies find an insignificant relation between the public level of adherence to a certain religion in a country and the country's perceived level of corruption (North et al. 2 0 1 3 ; Flavin/Ledet 2008). Several economic literature studies investigate the relation between religion and corruption (La Porta et al. 1999; Treisman 2 0 0 0 ; Paldam 2 0 0 1 ; Chang/Golden 2007). Treisman (2000) shows that religion reduces corruption since it helps civil society to be more organized and ensures that citizens are more likely to monitor elite groups. Paldam (2001) argues that religion may limit the effects of corruption and notes, for example, that the percentage of Protestants in a country is reciprocally related to corruption level. According to Flavin and Ledet (2008), scholars debate the appropriate measurement of religiosity, and explain that this is partly due to ". . . disagreements. . . about h o w best to quantify religion and an individual's underlying "level" of religious belief and devotion when referring to the different ways in which religiosity can be assessed". Religion was assessed through examining the " d o m i n a n t religion" or "the religion of m a j o r i t y " in a certain country. However, most of these studies are flawed because they assume that, if an individual ascribes to a certain religion, their behavior is b o u n d by the rules of that religion. The level of adherence to a religion is not investigated in these studies and that might significantly impact the overall conclusion. W e can deduce f r o m the foregoing, that, although theoretical arguments supporting the negative relation between religion and corruption may seem valid and logical, the results of the considerable body of empirical literature addressing this issue remain controversial and inconclusive. M a r q u e t t e (2010) states " t h a t the evidence for a causal relationship between religion (or types of religion) and either higher or lower levels of corruption is in n o way convincing". The author argues that the data - on religion - used in the majority of these studies are aggregated at the country level. Therefore, such studies are ill-equipped to examine aspects such as: (1) the influence of religion on h o w attitudes are formed, (2) h o w individual attitudes t o w a r d s corruption are formed a n d (3) w h a t are the possible strategies that the religion(s) permits or encourages its adherents to follow in order to change corrupt behavior. In order to avoid these shortcomings, this study will be based on a dataset collected by the W o r l d Values Survey. Regarding aspect (1), we hypothesize that religion forms an essential c o m p o n e n t of the individual's morality in social groups with high rates of religiosity. As for aspect (2), we hypothesize that religions, in general, endorse honesty a n d suppress corruption. Therefore, as the degree of religiosity increases on the individual level, their general attitudes t o w a r d s corruption c o n f o r m more and more with the religion's f u n d a m e n t a l morality teachings. It is difficult to fully assess the specific a p p r o a c h of every religion t o w a r d s corruption. Therefore, we test the effect of religiosity on corruption by either first, considering the type of religion or second, disregarding the religious type under investigation. This allows us to assess whether religiosity generically affects corruption or, adherence to a certain religion is the main influence on the level of corruption in any given country. Concerning aspect (3), it is essential to point out that actions endorsed by a certain religion against corruption are not practiced in a vacuum. In other words, social institutions (at both g r o u p and country level), can play a m a j o r role in influencing individual attitudes t o w a r d s corruption by offering various positive attitudes and appropriate actions against corruption (e.g. whistle-blowing) and limiting or suppressing other u n w a n t e d attitudes and actions (e.g., citizens' vigilante behavior against corrupt officials).

Religious Loyalty and Acceptance of Corruption · 189

Following the tradition of the N e w Institutional Economics (North 1990), we argue that individual attitudes about corruption are affected not only by the legal system, i.e., formal institutions, but also by the prevailing morals and values in a society, i.e., informal institutions. Consequently, individuals w h o are constantly exposed to a certain religion will to a significant extent - adopt its prescribed system of beliefs and values (i.e., informal rules) which f r a m e their own constraints w h e n tempted by corruption. 3

Theoretical background

Stark and Bainbridge (1985: 5) and Iannaccone ( 1 9 9 8 : 1 4 6 6 ) define religion as any shared set of beliefs, activities, and institutions based u p o n faith in supernatural forces. All religious sects investigated through this study emphasize the immorality of theft and bribery in their theological teachings. Table 4 provides a modest survey of religious texts that deal with stealing and bribery: these texts represent the foundations of these religions. A more comprehensive survey is beyond the scope of this study. T h e effect of religiosity on attitudes towards corruption remains an issue for debate in the empirical literature of sociology and economics. There is a schism between theology and social sciences on the stance of different religions regarding stealing and bribery. Several theories are proposed to explain the source of this confusion 1 . We n o w summarize the main points of these theories. Middleton and Putney (1962) conclude that some uncertainty is created by confusion of the scope of empirical research related to religion and morality. They emphasize that there is commonly a failure to distinguish between t w o different kinds of ethical standards: the ascetic (i.e., sexual inclinations, gambling) and the social (i.e., cheating, theft). Violations of social standards are harmful to every member of society, both religious and nonreligious people. However, since violations of ascetic standards are usually not directly harmful to society as a whole, the nonreligious are expected to be more prescribed by these standards t h a n the religious. Accordingly, differences in behavior between the religious and the nonreligious are apparent in specific areas only, and are a product of divergence in standards rather than to a differential upholding of standards. Tittle and Welch (1983) argue that individual religiosity has a significant impact on suppressing deviant behavior in highly secularized and run-down communities. In contrast, it is less of a deterrent in highly integrated and organized communities where religious morality is redundant given the other sources of moral authority and social control. Van Vleet et al. (1999: 12) state that most of the research investigating the relation between religion and delinquency hinges on a theory of "religious ecology". This theory proposes that religion is negatively associated with deviant behavior only when it is a part of widely accepted social values and norms that prohibit such behavior (Chadwick/Top 1993). Stark et al. (1982: 4) observe that "...conflicting findings stem f r o m variations in the religious ecology of the communities studied. In communities where religious commitment is the n o r m , the more religious an individual, the less likely he or she will be delinquent. However, in highly secularized communities, even the most devout teenagers are n o less delinquent than the most irreligious." Kohlberg (1981) presents a different perspective, claiming that religiosity and moral reasoning are essentially separate areas of h u m a n concern. Although moral decision-making is mainly influenced by the level of cognitive development (based on, e.g., education) 1

Marquette (2010) provides an impressive overview of these theories.

190 · Moamen Gouda and Sang-Min Park

and exposure to the socio-moral climate, religious reasoning is based u p o n teachings by religious authorities that emphasize morality. In other words, moral reasoning provides moral prescriptions and religious reasoning affirms these moral judgments as meaningful. This conclusion is also apparent in Kohlberg's (1984) theory on stages of moral development, where the author argues that moral reasoning has six identifiable developmental stages. Each of these stages is more suitable for responding to moral dilemmas t h a n its predecessor. Kohlberg (1984) states that the process of moral development is chiefly concerned with justice, and that it continues t h r o u g h o u t the individual's lifetime, a notion that spawned dialogue on the philosophical implications of such research. As for religion and morality, Power and Kohlberg (1981) suggest that a seventh stage should be integrated into the theory under the title "transcendental morality" or "morality of cosmic orientation" which links religion with moral reasoning. However, Kohlberg's difficulties in obtaining empirical evidence for even the sixth stage, lead him to underline the speculative nature of this proposed seventh stage (Power/Kohlberg 1981). The present study focuses on investigating the specific link between the religiously-induced internalized values and beliefs of individuals on the one hand and their specific attitudes regarding corruption on the other, which gives us our main hypothesis: Hypothesis corruption.

1 : Higher individual-level

religiosity

is associated

with lower acceptance

of

W e focus our analysis specifically on the micro-level rather than the macro-level. An individual's degree of exposure to religion is thought to be reflected by their internalized values and beliefs and is measured through variables that demonstrate their level of religious adherence. The attitudes and beliefs concerning corruption are reflected through the survey respondents' acceptance of corrupt actions. Because our f r a m e w o r k explicitly accounts for the micro-macro interaction in values and beliefs, i.e., the social context, we also propose the following hypothesis: Hypothesis 2: Stronger acceptance of corruption at the societal level is associated higher individual-level acceptance of corruption.

with

This hypothesis basically states that the average individual in society will - in his attitudes regarding corruption - adhere to p r e d o m i n a n t societal attitudes regarding corruption, rather than swimming against the current. The latter, if accompanied by according behavior, w o u l d very likely be associated by social sanctions, such as exclusion f r o m social groups. Finally, regarding second-tier aggregate influencing factors, let us mention democracy and income. In the ideal sense, democracy is associated with checks and balances which could lead to lower individual acceptance of corruption. In a more differentiated view, one could argue that - in young immature democracies - individual acceptance of corruption can even increase up to a certain point, due to an increase in opportunities for corrupt behavior 2 . W e also expect increasing incomes to be associated with lower individual acceptance of corruption, due to (1) a decrease in the need for illegal sources of incomes and (2) via a general increase in institutional quality.

2

For an analogous line of argument with regard to aggregate corruption and democracy, see Mohtadi and Roe (2003).

Religious Loyalty and Acceptance of Corruption · 191

4

Methodology and model specification

Because our hypotheses relate to individual level attitudes, it is appropriate that we test them with data from individuals. Naturally, experimental data about such attitudes is difficult to generate or find', therefore, we use survey data taken from the World Values Survey (WVS), which measures values and attitudes in representative samples from more than 80 countries around the world. Index i denotes individuals surveyed, j denotes country of residence and t denotes year of survey. Our estimation sample is composed of 141,326 individuals in 78 countries surveyed in 13 different years. Following a considerable body of literature 4 , the degree of acceptance of corruption is measured by responses to the WVS question "Please tell me for each of the following statements whether you think it can always be justified, never be justified, or it's somewhere in between, that someone accepts a bribe in the course of their duties." The responses range from 1 (never justifiable) to 10 (always justifiable). Because the responses to this question are very skewed towards 1 (almost 7 5 % ) , we recode this information into a binary format. Our dependent variable Corruption¡ takes on a value of 0 if respondents answered that bribes are never justified, and a value of 1 otherwise. Intuitively, this procedure is in line with the observation that religious norms usually do not permit any intermediate level of corruption. The main variable of interest, religiosity, is measured through responses to four WVS questions: (1) "Indicate how important it is in your life. Would you say it is: Religion?" (2) " A p a r t from weddings, funerals and christenings, about how often do you attend religious services these days", (3) "Independently of whether you go to church or not, would you say you are a religious person?", and (4) " H o w important is God in your life?" We construct a weighted index Religiosity¡ from these questions, weighted according to the results from a factor analysis. In order to control for denominational differences, we also include dummies for the 7 main denominations with which the respondent might be affiliated. In order to capture the micro-macro interaction of the theoretical framework (hypothesis 2), we include as independent variables the aggregated country-level mean for corruptibility (Acceptancejt). We estimate ^(Corruptiont

= 1 | X ) = G(ß0Religiosityi

+ Ζ)βχ + Zftß2 + u,)

(1)

with maximum likelihood, where G is the standard normal cumulative distribution function, Zj i is the vector containing our individual level controls and Z? is the vector containing our country-level controls. Several individual-level control variables are included in Zj. We control for sex (Male¡), age (Age¡), education (Education¿), marital status (Marriedj), employment status {Unemployed,) and trust towards others (Trustj) of the respondent. T o control for income-related differences, we include an ordinal variable (Income¡) which is a subjective, self-reported

3

4

Armantier and Boly ( 2 0 1 1 ) provide evidence from a controlled field experiment that religiosity, measured through a post-experimental question of how often the subject goes to church, is associated with a lower probability of subjects accepting bribes. See, for example, Swamy et al. ( 2 0 0 1 ) , Gatti et al. ( 2 0 0 3 ) , Y o u and Khagram ( 2 0 0 5 ) and Esarey and Chirillo ( 2 0 1 2 ) .

192 · Moamen Gouda and Sang-Min Park

assessment of the respondent's income level. We also control for the respondent's financial satisfaction (Financial^satisfaction,), because low financial satisfaction might be associated with higher acceptance of corruption. Unobserved heterogeneity refers to variables which cannot be accounted for, such as, the respondents' acceptance of corruption and religiosity may be affected by the year that the survey was carried out in or the country or region that they live in. There are, accordingly, two distinct approaches to arrange the vector Z \ . The first approach is to

ft

include a full set of country and year dummies that accounts for any level differences there might be between countries in different years. The second approach is to include a wide set of time-varying country-level variables which might be relevant in influencing acceptance of corruption on a macro-level. A country's colonial history might influence its formal and informal institutions, which is why we control for it with a set of appropriate dummies (Colonial_Xj)5. More institutional variables include an indicator for quality of democracy (Democracyy) and age of democracy (Age_democracyj t ). We also control for per capita income (GDP ; f ) 6 . The two approaches for arranging the vector are mutually exclusive as combining country, region and year dummies with multiple country level variables would result in near perfect multicollinearity. Details on all variables can be found in Table 1 and 2.

5

Estimation results

In Probit estimations, the marginal effect of any explanatory variable is (1) inherently non-linear and (2) conditional on values of all other covariates. The main effect of interest, i.e., the marginal effect of religiosity on the probability that corruption will be accepted is

aPr(

SSr" -

+ +

* 4« • Λ

where g(z) = 3 d G d z { z ) . It is immediately apparent that this marginal effect needs to be evaluated at specific values of (Religiosity ¿, ZK ι Z?) μ in order to be interpreted in a meaningful way. We can also see that interpretations of interaction effects are possible without explicitly including interaction terms as explanatory variables. In Table 3, we present the average marginal effects (AMEs) from our estimations. In order to compute these, equation (2) is calculated for each observation using the estimated coefficients and then averaged over all observations. This gives us a first impression of the effects we are interested in. Columns (1) to (4) represent estimations with different sets of fixed effects, while column (4) represents an estimation with country-level controls, but only year fixed effects. We interpret the results as being quite robust across specifications.

5

6

La Porta et. al. (1999), Treisman (2000), and Herzfeld and Weiss (2003) show that former British colonies have lower levels of corruption. See, for example, van Rijckeghem-Weder 1997; Ades/Di Telia 1999; Treisman 2 0 0 0 ; Rauch/Evan 2000; Paldam 2 0 0 2 ; Sandholtz/Gray 2 0 0 3 ; Tavares 2003; Dreher et al. 2 0 0 4 ; Chang/Golden 2 0 0 4 ; Kunicova/Ackerman 2 0 0 5 .

R e l i g i o u s

L o y a l t y

a n d

A c c e p t a n c e

o f

C o r r u p t i o n



1 9 3

m o in οι I in

y?

m in in

σι σ> σι t

y?

σ\

IN

(Ν co σ\ ιη m co

\ο

νο 00 m σ\ «t

m m S VO σ\ ΓΝ

iri

CO in VD

ID m 00 co o m

in vo σ\ Pv

IN



ò

t

a.

m τCM r^ 00 ΓΝ

00

m

VD m m 00

ΓΜ σ> ο

ΓΝ ο

o

00 t 00 co o

o

o

_ o Ο

ο νο fN m οο o o o 6

VO Tili» σ» Μ- (χ νο οο m ο

o

o

o

on rx σ» τin

00 00 m

σ\ τ* rx

ο ο ο νο VD νο νο (Ν (Ν ΓΝ (Ν m m no m τΐ- tj- -s- Tf

>>T3 y υ S C

rx o o

fx ο

c I OE O (J "O Ν- Ό Ο Sì

σ\

s O

73 co =3 X = °3 = Κ = S

ττ o σ» in o

o o

_ =5 g

> Q

o Χ E o Q. «I 3

.60

-S o o .36

ε' Ì«j ο C c ν Q

6o CJ > ce αϊ 3

O U

>η>

υ ΐ; c ι/ι co •S Q. .tw Β" ce


0

6o

3

O O (J U

c 0Q if

ο CD 00 ¡0 oí > CO Jp C 3 O υ

6o

O U

tSí i =

î» DO s*=

aj

£

a 5 > 1

>

c

ε

I



° -o Ο a s s sI- ς ™ 8 4) 0) S S s CL O £ AS C O Q. Ä £ « o o JS

u

11

ti -c

re

Ο

Ρ

S

™ ο

S ¿ - s ±3 3 0)

α ^ ·σ

(0

υ

5

Ζ "S · Ο α> >

^

i

E -C

?> ζ

u

E «

1

•E . 2

c

Π3 _

rtJ 4)

3

5 5

.

C

υ

Μ ΙΛ

I

ts

I l § 1

ä M < X c : 2¡ < U Ε c 2 .

-à £ fe 0) (5 * 5 Ό

43 ο ι

^

α

-O ¡fc M-" °

V

t : 0J O

Ε

73

ι

V Q.

E g

β

QO £ O» — -C I S E

¥



·

l

:

i a; « a. ι>

o

: l •

-

I

g J2 d ε ,«>| 5 ?

c

a> >

° =

CL > H c ο 'W)

Corruption

ο E Ίλ Μ) cfd O Û< ü

"o 3 rr j3 C (-< n"ï C ^ £ ζ; ο-Ό ¡5

-

ν

·-

£

Mí' C ** '> Ό -S O U υ S ra δ1 o c s; -o ï ^o . 'I oO Ι3 Λ (çd l-

δ =ra4)TD < ÌC 5m= .O CL _0) •. -ο 0)H ~ < o£ U O U< < oJ ifc -C rt 1> S"° δ ut-JS δ

db cû«J1dj _< C "F = Q -S · , ra )ra p ^ S, s ä υ o !! ü «I oοo ΈΩ ^ S. Ζ ra-α r Sif ~Ό ±ί 0 : δ C "Ό2 E «J i τ9 δra ra >Q-raE -c£ m I ; s ix in * 1 € ï S· i: δ l its ^ ra jι \ .. ^ £ ¿ ìJDl • > -π ^ ira> I u 3 : .>· I J! -ρ .5P a) < E O C 0ΛM_O O S "a ui aj o S ί C Dû ~ ®Ο Ό ^

, o ^ JDra u ¿'J ra í! -i i_ > ¿ Ln û_ u o^ σ•

Religious Loyalty and Acceptance of Corruption • 203

References Ades, A.D. (1997), National Champions and Corruption: Some Un-pleasant Interventionist Arithmetic. The Economic Journal 107(443): 1023-1042. Ades, Α., R. di Telia (1999), Rents, Competition, and Corruption. The American Economic Review 89(4): 982-993. Al-Marhubi, F. (2004), The Determinants of Governance: A Cross-Country Analysis. Contemporary Economic Policy 2(3): 394-406. Alt, J.E., D.D. Lassen (2003), The political economy of institutions and corruption in American states. Journal of Theoretical Politics 15(3): 341-365. Armantier, O., A. Boly (2011 ), A controlled field experiment on corruption. European Economic Review 55 (8): 1072-1082. Barreto, R.A. (2000), Endogenous corruption in a neoclassical growth model. European Economic Review 44(1): 35-60. Barreto, R.A., J. Aim (2003), Corruption, Optimal Taxation, and Growth. Public Finance Review 31(10): 1-34. Basu, P.K. (2006), Corruption: A Theoretical Perspective and Relevance for Economic Growth. International Review of Business Research Papers 2(2): 59-68. Beets, S.D. (2007), Global corruption and religion: an empirical examination. Journal of Global Ethics 3(1): 69-85. Berggren, N., C. Bjornskov (2011), Is the Importance of Religion in Daily Life Related to Social Trust? Cross-Country and Cross-State Comparison. Journal of Economic Behavior and Organization 80: 459-480. Bhattarai, P.C. (2009), Countering Corruption: Globally or Locally? Nepalese Journal of Public Policy and Governance XXXIV(l): 89-98. Bible-New International Version (2011), search words: "theft", "steal", "bribe". Retrieved November 11, 2011, from Bible Gateway: http://www.biblegateway.comy Braun, M., R. di Telia (2004), Inflation, Inflation Variability, and Corruption. Economics and Politics 16(1): 77-100. Brunetti, Α., Β. Weder (2003), A free press is bad news for corruption. Journal of Public Economics 87(7-8): 1801-1824. Caiden, G.E. (1988), Toward a General Theory of Official Corruption. Asian Journal of Public Administration 10(1): 3-26. Chadwick, B.A., B.L. Top (1993), Religiosity and Delinquency among LDS Adolescents. Journal for the Scientific Study of Religion 32(1): 51-67. Chang, E.C., M.A. Golden (2007), Electoral Systems, District Magnitude and Corruption. British Journal of Political Science 37(1): 115-137. Dreher, Α., C. Kotsogiannis, S. McCorriston (2004), Corruption around the world: Evidence from a structural model. Journal of Comparative Economics 35(3): 443-466. Duasa, J. (2010), Tendency of corruption and its determinants among public servants: A case study of Malaysia. Indian Development Review 8(1): 279-292. Durkheim, E. (1912/1915), The Elementary Forms of the Religious Life. Chicago: Free Press. Ehrlich, I., F.T. Lui (1999), Bureaucratic corruption and endogenous economic growth. Journal of Political Economy 107(2): 270-293. Esarey, J., G. Chirillo (2012), 'Fairer Sex' or Purity Myth? Corruption, Gender, and Institutional Context. Politics and Gender 9(4): 390-413. Flavin, P., R. Ledet (2008), Religiosity and Government Corruption in the American States. Proceedings of the Annual meeting of the MPSA Annual National Conference. Chicago: MPSA Annual National Conference: 1-22. Gatti, R., S. Paternostro, J. Rigolini (2003), Individual attitudes toward corruption: do social effects matter? World Bank Policy Research Working Paper Series 3122: 1-24. Guerrero, M.A., E. Rodriguez-Oreggia (2008), On the individual decisions to commit corruption: A methodological complement. Journal of Economic Behavior and Organization 65(2): 357-372.

204 · Moamen Gouda and Sang-Min Park

Guiso, L., P. Sapienza, L. Zingales (2003), People's opium? Religion and economic attitudes. Journal of Monetary Economics 50: 225-282. Herzfeld, T., C. Weiss (2003), Corruption and Legal (In)Effectiveness: An Empirical Investigation. European Journal of Political Economy 19(2003): 621-632. Hirschi, T., R. Stark (1969), Hellfire and delinquency. Social Problems 17(2): 202-13. Iannaccone, L.R. (1998), Introduction to the Economics of Religion. Journal of Economic Literature XXXVI (September 1998): 1465-1496. Jagodzinski, W. (2009), The Impact of Religion on Values and Behavior. Kwansei Gakuin University School of Sociology Journal 107: 19-34. Jain, A.K. (2001), Corruption: A Review. Journal of Economic Surveys 15(1): 71-121. Khan, M. (2006), Corruption and Governance. Pp. 200-221 in: K.S. Jomo, Β. Fine (eds.), the New Development Economics. London and New Delhi: Zed Press and Tulika. Kohlberg, L. (1981), Essays on Moral Development, Vol. 1: The Philosophy of Moral Development. San Francisco: Harper and Row. Kohlberg, L. (1984), Essays on Moral Development: The Psychology of Moral DevelopmentThe Nature of Moral Stages (Vol. II). San Francisco: Harper and Row. Kunicova, J., S. Rose-Ackerman (2005), Electoral rules and constitutional structures as constraints on corruption. British Journal of Political Science 35(4): 573-606. La Porta, R., F.L. de-Silanes, A. Shleifer, R. Vishny (1999), The Quality of Government. Journal of Law, Economics and Organization 15(1): 222-279. Lambsdorff, J.G. (1999), Corruption in empirical research- a review. Transparency International Working Paper: 1-17. Lederman, D., N. Loayza, R.R. Soares (2005), Accountability and corruption: Political institutions matter. Economics and Politics 17(1): 1-35. Luxmoore, J. (1999), Churches urged to help fight global corruption. Catholic New Times 23(2): 12-13. Macrae, J. (1982), Underdevelopment and the Economics of Corruption: A Game Theory Approach. World Development 10(8): 677-87. Marquette, H. (2010), Corruption, religion and moral development. Working Paper. University of Birmingham: 1-30. Marshall, M.G., K. Jaggers (2002), Polity IV project: Political regime characteristics and transitions, 1800-2002. University of Maryland, College Park: Center for International Development and Conflict Management (CIDCM). Matei, L., A. Matei (2009), Corruption in the Public Organization towards a Model of Cost Benefit Analysis for the Anti-corruption Strategies. Transylvanian Review of Administrative Sciences 27: 145-71. Mauro, P. (1995), Corruption and Growth. Quarterly Journal of Economics 110(3): 681-712. Mauro, P. (1998), Corruption and the Composition of Government Expenditure. Journal of Public Economics 69: 263-279. Mauro, P. (2002), The Effects of Corruption on Growth and Public Expenditure. Pp. 339-352 in: A.J. Heidenheimer, M. Johnston (eds.), Political Corruption: Concepts and Contexts. New Brunswick, NJ: Transaction Publishers. Middleton, R., S. Putney (1962), Religion, normative standards, and behavior. Sociometry 25(2): 141-152. Mishra, A. (2006), Persistence of corruption: some theoretical perspectives. World Development 34(2): 349-358. Mohtadi, H., T.L. Roe (2003), Democracy, rent seeking, public spending and growth. Journal of Public Economics 8 7 ( 3 ^ ) : 445-466. Mutascu, M. (2010), Corruption, Social Welfare, Culture and Religion in European Union 27. Transition Studies Review 16(4): 908-917. Nas, T., A.C. Price, C.T. Weber (1986), A policy-oriented theory of corruption. The American Political Science Review 80(1): 107-119. Naumova, S. (2009), Legal-ociological Parameters of the Fight Against Crime and Social Deviance. Pp. 11-22 in: M. Serafimova, S. Hunt, M. Marinov, V. Vladov (eds.), Sociology

Religious Loyalty and Acceptance of Corruption • 205

and Law: The 150th Anniversary of Emile Durkheim 1858-1917. Newcastle upon Tyne: Cambridge Scholars Publishing. North, C.M., W.H. Orman, C.R. Gwin (2013), Religion, Corruption, and the Rule of Law. Journal of Money, Credit & Banking 45(5): 757-779. North, D.C. (1990), Institutions, Institutional Change, and Economic Performance. New York: Cambridge University Press. North, D. (1991), Institutions. Journal of Economic Perspectives 5(1): 97-112. Paldam, M. (2001), Corruption and religion: adding to the economic model. Kyklos 54(2-3): 383-414. Paldam, M. (2002), The big pattern of corruption. Economics, culture and the seesaw dynamics. European Journal of Political Economy 18(2): 2 1 5 - 4 0 . Paldam, M., E. Gundlach (2013), The Religious Transition. A Long-run Perspective. Public Choice 156(1-2): 105-123. Pickthall, M.M. (1995), The meaning of the glorious Qur'an. Retrieved November 11, 2011, from http://www.khayma.com/librarians/call2islaam/quran/pickthall/index.html Power, C., L. Kohlberg (1981), Moral Development, Religious Thinking, and the Question of a Seventh Stage. Zygon 16: 203-259. Przeworski, Α., H. Teune (1970), The Logic of Comparative Social Inquiry. New York: WileyInterscience. Rauch, J.E., P.B. Evans (2000), Bureaucratic structure and bureaucratic performance in less developed countries. Journal of Public Economics 75(1): 4 9 - 7 1 . Rock, M.T. (2009), Corruption and Democracy. The Journal of Development Studies 45(1): 55-75. Rose-Ackerman, S. (1999), Corruption and Government- Causes, Consequences and Reform. Cambridge: Cambridge University Press. Sandholtz, W., M . M . Gray (2003), International Integration and National Corruption. International Organization 57(4): 761-800. Seligson, M.A. (2002), The Renaissance of Political Culture or the Renaissance of the Ecological Fallacy? Comparative Politics 34(3): 273-292. Shleifer, Α., R.W. Vishny (1993), Corruption. Quarterly Journal of Economics 108: 599-617. Stark, R., W.S. Bainbridge (1985), The Future of Religion: Secularization, Revival and Cult Formation. Berkeley: University of California Press. Stark, R., L. Kent, D.P. Doyle (1982), Religion and Delinquency: the Ecology of a "Lost" Relationship. Journal of Research in Crime and Delinquency 19(1): 4 - 2 4 . Student, G. (2000), Theft from Gentiles. Retrieved November 11, 2011, from The Real Truth about the Talmud: http://www.angelfire.com/mt/talmud/theft.html Swamy, Α., S. Knack, Y. Lee, O. Azfar (2001), Gender and Corruption. Journal of Development Economics 64: 25-55. Tavares, J. (2003), Does foreign aid corrupt? Economic Letters 79: 99-106. Teorell, J., A. Hadenius (2007), Determinants of Democratization: Taking Stock of the Large-N Evidence. Pp. 69-96 in: D. Berg-Schlosser (ed.), Democratization: The State of the Art (The World of Political Science). Leverkusen Opladen: Barbara Budrich Publishers. Tittle, C.R., M.R. Welch (1983), Religiosity and deviances: toward a contingency theory of constraining effects. Social Forces 61(3): 653-82. Transparency International. (2006), Transparency International. Retrieved March 1, 2011, from Handbook: Curbing Corruption in Public Procurement: http://www.transparency.org./ global_priorities/public_contracting Transparency International. (2011), Transparency International. Retrieved October 9, 2011, from Mission Statement: http://www.transparency.org.au/mission.php Treisman, D. (2000), The Causes of Corruption: A Cross-national Study. Journal of Public Economics 76(2000): 399-457. Van Rijckeghem, C., B. Weder (1997), Corruption and Rate of Temptation: Do Low Wages in the Civil Service Cause Corruption? IMF Working Paper 97(73): 1 - 5 6 . Van Vleet, R.K., J. Cockayne, T.R. Fowles (1999), Examining Religion as a Preventative Factor to Delinquency. Utah Criminal Justice Center: 1 - 1 8 .

206 • Moamen Gouda and Sang-Min Park

Weber, M. (2010), The Protestant Ethic and the Spirit of Capitalism. North Charleston, South Carolina: Createspace. Wei, S.J. (1997), Why is Corruption so Much More Taxing than Tax? Arbitrariness Kills. NBER Working Papers 6255: 1 - 1 7 . World Bank (2004, April 8), World Bank. Retrieved October 23, 2011, from The Costs of Corruption: http://go.worldbank.org/LJA29GHA80 You, J.S., S. Khagram (2005), A Comparative Study of Inequality and Corruption. American Sociological Review 70(1): 136-157. Prof. Dr. Moamen Gouda, Graduate School of International and Area Studies, Hankuk University of Foreign Studies, 107, Imun-ro, Dongdaemun-gu, Seoul, 130-791, South Korea. [email protected] Dr. Sang-Min Park, Technische Hochschule Mittelhessen - University of Applied Sciences, Fachbereich Sozial- und Kulturwissenschaften, Wiesenstr. 14, 35390 Glessen, Germany. [email protected]

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2015) Bd. (Vol.) 235/2

Group Decision Making in a Corruption Experiment: China and Germany Compared Sha Li, Christoph Bühren, Björn Frank* University of Kassel Haiying Qin Nankai University JEL C91 ; C92; D73 China; corruption; experiment; group decisions.

Summary Much hope is put into the "four eyes principle" as an anti corruption device in many countries. However, as recent cases have shown, entire groups of decision makers can be corrupt as well. This paper reports on an experimental investigation of individual versus group decision making in a corruption experiment. We find that the group decisions, as compared to individual decisions, lead to a higher level of corruption, for bribers and for bribees, and in China as well as in Germany. Only German women are less corrupt in a group decision context than when deciding individually. Further differences between Germany and China with respect to the effect of the teams' gender composition were found. In Germany, groups that consist only of females are the most honest and the male groups are the most corrupt, whereas in China the groups with mixed gender combination have shown a higher inclination to make corrupt decisions than the groups that are homogenous with respect to gender.

1

Introduction

The four eyes principle is popular at least in terms of the likelihood to find it demanded, be it as a standard rule on procurement for large dam projects (Wiehen 1999), as a requirement for internal control of banks according to the Basle Committee on Banking (1997), or as part of a compliance management system in China (Stucken/Liu s.a.). Hoping for positive - i.e., corruption preventing - effects of mutual audits, the "four eyes principle" is also required in contemporary laws. For example, it must be used when public contracts are awarded in the German Länder Nordrhein-Westfalen and SchleswigHolstein. 1

* W e are indebted to T h e o d o r e Eisenberg, Elina K h a c h a t r y a n , J o h a n n Graf Lambsdorff and t w o a n o n y m o u s referees for helpful comments, and to the China Scholarship Council (CSC) a n d the D A A D for providing financial support. 1 See Innenministerium des Landes N o r d r h e i n - W e s t f a l e n (2005) a n d Korruptionsrichtlinie SchleswigHolstein (2008).

2 0 8 · Sha Li, Christoph Bühren, Björn Frank, and Haiying Q i n

On the other hand, there are corruption cases in which evidently whole groups of decision makers have been bribed. In 2006, Compass subsidiary Eurest Support Services won contracts to provide food to United Nations peacekeepers in Liberia, valued at 7 million $. Alexander Yakovlev, the UN procurement officer, and Vladimir Kuznetsov, head of the UN Committee for Administrative and Budgetary Issues, were arrested and indicted after taking nearly a one million $ bribe from Compass. After its own three-month internal investigation, Compass declared it had discovered "serious irregularities" in its UN business (Russell/Rosett 2006). Another impressive example is team sports: Before the final round of the second soccer league in Goa (India), two clubs were at the top with an equal number of points, promotion to the first division depending on the goal difference. The results of the final round were 55:1 for one of the clubs, 61:1 for the other (Buse 2004; Perry 2004), which is not possible unless more or less the entire opposing team is helping. (For a similar incident, a 54:1 in Germany, though in a lower league, see dpa, 2008.) And in China, three leading officials of the Football Association have been arrested for abusing their power on various occasions (see Ifeng 2010). These cases call the actual usefulness of the four/more eyes principle into question. Does it really lead to the expected reduction in corruption? And if groups' decisions differ from those made by individuals: does this effect depend on culture or on gender composition? As yet, theoretical investigations of corrupt behavior (e.g., Lambsdorff 2007) typically refer to individual behavior. They remain silent about possible differences between individual and group behavior. One exception is the idea that an agent who considers costs C and benefits Β of staying honest and finds that both C and Β depend on the share of other agents who are corrupt (Andvig/Moene 1990; Andvig 1991). 2 At least from a certain degree of corruption on, these costs will fall with the share of corrupt agents, as not many officials have an interest in investigating complaints about their corrupt colleagues. Two stable equilibria might result, one in which taking bribes does not pay because this is practiced by too few people, and a bad equilibrium in which costs of bribe taking are low for a single agent because his or her environment is highly corrupt. While this is still a theory of decisions made individually, it makes use of the idea that agents monitor each other, which is the intention behind the four eyes principle. However, agents monitoring each other is a concept that is clearly different from agents coming to a joint decision. Leaving corruption aside for a moment, what do we already know about group decision making? Laboratory evidence on differences between individual and group decision making as surveyed by Charness and Sutter (2012) indicates that single deciders are less "homo oeconomicus" than groups in a majority of the experiments conducted so far, including those featuring reciprocity and cooperation (also see Kocher/Sutter 2005, and Ambrus et al. 2009). However, this stylized fact is based on experiments in which individual rationality is in conflict with social norms. This is different in corruption experiments, where bribing violates the social norm and somewhat irrationally exhibits trust in the bribee, who could maximize his profit by not reciprocating. Hence it is unclear what kind of group behavior to expect in our corruption experiment: Groups might be more rational (like in previous experiments), thus shying away from corrupt acts, or they might be more anti social (like in previous experiments), thus more corrupt.

2

See Murphy et al. (1993) for a similar model of bribers' or rent seekers' behavior.

Group Decision Making in a Corruption Experiment · 209

While it is difficult to hypothesize whether groups will be more or less corrupt than individuals, due to the unique moral connotations of corruption, it is important to find out, due to the practical importance of the four eyes principle. We contribute insights from a cross-country experiment. The remainder of the paper is organized as follows. Section 2 gives a literature overview, followed by a description of our experimental design in section 3. We present our results in section 4 and discuss the findings in section 5.

2

Related literature

We can build on previous corruption experiments (with single deciders), a branch of experimental economics that is still young, but flourishing for a good reason.3 As everyone involved in a corrupt activity typically has strong incentives to remain silent, it is extremely difficult to observe and to measure corrupt activities directly. Frank and Schulze (2000) started to generate individual data on corrupt behavior with a very simple experimental design, simulating the bribers and thus restricting their analysis on the inclination to accept bribes. An important feature of corruption is the negative welfare effect which has been demonstrated in many statistical analyses of international cross-section data.4 Abbink et al. (2002), who were the first to experimentally reconstruct not only the bribees', but also the bribers' decision, modeled this as a certain amount subtracted from the payoff of all other subjects in the experiment who play at the same time the same game. They find that the participants did not take the negative externality into account. However, this might have resulted from the symmetry in their design: every subject makes the same (corrupt) decision because she or he expects that the others pick the corrupt action as well. A method for avoiding this effect is imposing the externality on a third party not directly involved in the experiment, such as a students' film club in Frank and Schulze (2000) or Médecins Sans Frontières in Lambsdorff and Frank (2010). Not all experimenters who investigated corruption controlled for gender, but those who did inevitably found a gender effect (see Frank et al. 2011; or Chaudhuri 2012, for an overview), though more often one that is not suggested by folk psychology. It is probably in line with many people's intuition that men offer bribes significantly more often than women in Rivas' (2013) experiment; furthermore, those bribes that are paid are significantly larger if coming from men. However, Rivas (2013) did not find clear differences between women and men with respect to their inclination to accept bribes in the role of public officials to whom firm (i.e., subjects in the role of firms) pay bribes, hoping for being awarded a contract. Yet women are significantly more likely to behave opportunistically by keeping the bribe, but not awarding the contract to the briber. Other experiments that failed to find that female agents are less corrupt than men in a situation without monitoring are Frank and Schulze (2000) as well as Armantier and Boly (2013). The latter as well as Schulze and Frank (2003) do find, however, that women react more strongly to the risk of being detected. Finally, Lambsdorff and Frank (2011) also find remarkable differences between women and men in the role of agents. Women are markedly more likely to behave opportunistically and less likely to reciprocate, though no significant difference occurred with respect to whistle-blowing.

3 4

For a review see Abbink ( 2 0 0 6 ) , Andivig ( 2 0 0 5 ) or Serra and W a n t c h e k o n ( 2 0 1 2 ) . For a survey see Lambsdorff ( 2 0 0 6 ) .

210 • Sha Li, Christoph Bühren, Björn Frank, and Haiying Qin

T h e only one experimental study relating corruption and group-decision is Schikora (2011). Close t o the experimental set-ups used in Abbink et al. (2002, 2004) and Lambsdorff a n d Frank (2010), he modeled the four eyes principle as replacing a single official with a g r o u p of t w o officials deciding jointly according to a decision making process which secures veto p o w e r for the non-corrupt official. H e finds a negative effect of introducing the four eyes principle, which results f r o m the dominance of profit maximizing motives in the group decision making process despite the social dilemma of corruption. The results cast doubt on the usefulness of the four eyes principle. Furthermore, Schikora (2011) did not find any significant differences between groups with different gender constellation. However, this study does not take into account effects of different cultures. Cultural differences might be a very important determinant of the levels of bribery and bribetaking (Husted 1999), since corruption is perceived to be a moral and ethical issue by m a n y people. The informal and formal behavioral rules, such as social norms, social preferences, hierarchy or religions, vary f r o m one culture to another. Nevertheless, Engel et al. (2012) ran a corruption experiment in China and in Germany and did not find significant differences between subject pools with respect to the inclination to offer or to accept corrupt deals. However, some previous intercultural empirical studies showed a correlation between cultural variables and corruption. 5 Barr a n d Serra (2010) investigated a bribery game with 195 students taking the role of either private citizens or public servants. The participants were students at the University of O x f o r d , coming f r o m 43 countries. For "private citizens", Barr and Serra (2010) f o u n d a positive correlation between corruption in the h o m e country and corruption in the experiment, but only for undergraduates. The fact that the effect disappeared after a longer stay in the UK supports the presumption that corruption is, in part, a cultural p h e n o m e n o n . The first study that simultaneously investigates gender and culture effects in a corruption experiment is by Alatas et al. (2009). In Australia, they find the percentage of officials accepting a bribe to be significantly lower for w o m e n (80.0%) t h a n for men (92.1%). In India and Indonesia, this difference a m o u n t s to less t h a n one percentage point and is insignificant, while in Singapore, w o m e n appear to be more inclined to take bribes than men, yet the difference is not statistically significant. For bribe giving, the pattern is similar: Again, Australia was the only country where w o m e n turned out to be significantly less corrupt. In our corruption experiment, we analyze g r o u p vs. individual decision making and, moreover, control for culture as well as gender constellation effects.

3

Experimental design

Corruption is commonly defined as the misuse of public or private office for personal gain. In our experiment, we have modeled a sequential game between private business and administration. Figure 1 depicts the extensive f o r m of the game. A firm, which is called X, and a public official Β receive 30 or 2 0 experimental dollars, respectively. T o obtain a public contract, the firm X can decide on whether or not to offer a bribe to the 5

La Porta et al. (1997) and Knack and Keefer (1997) show a positive correlation between interpersonal trust and corruption at the country-level. La Porta et al. (1997, 1999) Treisman (2000) and Serra (2006) found a negative correlation between corruption and the dominance of nonhierarchical religions. For a review see Rose-Ackerman (2004); laboratory evidence on the interrelationship of corruption and culture is surveyed by Banuri and Eckel (2012).

Group Decision M a k i n g in a Corruption Experiment • 211

responsible official. If a bribe is paid, its size t c a n be 2 , 4 , 6 , o r 8 . If n o bribe is offered, then the public c o n t r a c t will be a w a r d e d to the fictitious c o m p e t i t o r firm Y . It is assumed here t h a t firm Y c a n d o the j o b m o r e efficiently t h a n firm X . If a bribe t is offered to the official B, then she c a n c h o o s e one o f the f o u r alternatives: 1. accepting the bribe and a w a r d i n g the public c o n t r a c t t o the briber 2 . keeping the bribe but a w a r d i n g the public c o n t r a c t t o firm Y 3 . rejecting the bribe, awarding the public c o n t r a c t to firm Y , and returning the bribe t o the briber 4 . rejecting the bribe, a w a r d i n g the public c o n t r a c t t o firm Y , a n d reporting the briber If the c o m p a n y X is a w a r d e d the c o n t r a c t , then she earns an additional 3 2 ; taking the initial e n d o w m e n t o f 3 0 a n d bribe into a c c o u n t , her net benefit is 6 2 - t . As is usual in c o r r u p t i o n e x p e r i m e n t s (e.g. A b b i n k 2 0 0 2 ) , w e assume the marginal utilities o f i n c o m e to differ between the private a n d the public sector. H e n c e by accepting the bribe the official c a n increase her p a y o f f n o t only by t, but by 2 t . H o w e v e r , t o c o v e r up her a c t , she m u s t a c c e p t a deduction o f 3 f r o m her p a y o f f , hence her net benefit is 1 7 + 2 t . In this case, a reciprocal relationship between the t w o parties is established a n d b o t h o f t h e m are c o r r u p t . If the official has reported the bribe, the c o m p a n y X will be punished a n d her p a y o f f is reduced by 8. H e n c e f r o m the initial e n d o w m e n t , 3 0 minus 8 minus bribe t are remaining. A t the s a m e time, the official's p a y o f f is reduced by 2 due t o (opportunity) costs o f reporting 6 , thus she earns 1 8 instead o f 2 0 (Figure 1).

Ξ Ss 30

20

62-t

17+2t

0

30-t

20+2t

25

30

20

25

22-t

18

30

25

no bribery

bribery t = 2,4,6,8 reporting

Figure 1 Extensive form of the bribery game Unlike m a n y c o r r u p t i o n e x p e r i m e n t s , w e presented the situation in a " l o a d e d " f r a m i n g , as w e think t h a t a neutral presentation could result in o v e r l o o k i n g o r a distortion o f

6

For example, she has to give testimony or appear in court etc.

212 · Sha Li, Christoph Bühren, Björn Frank, and Haiying Qin

possible cultural, social, or psychological factors of interest. 7 The actors in our experiment were called "firm X " and "public official", and the "bribe" was just called "bribe", not "transfer". Furthermore, in the instructions we mentioned a story behind the bribery game: Firm X bid for public tender of a highway construction. The World Wide Fund For Nature (WWF) served as the receiver of the negative externalities of corruption. If the inefficient firm X is building the highway, the damage to the environment is relatively high. Accordingly, in this case there will be no donation to the W W F which is known for its efforts in protecting the environment. If the public contract is awarded to the more efficient firm Y, the W W F receives a donation of 25 experimental dollars. In the case of rejecting and reporting, W W F receives 30 because the community is better informed on corruption and anti-corruption, and possible future bribers are deterred. Regardless of the choice of the parameters, a rational official will behave opportunistically in a one-shot setting, i.e. she accepts the bribe, but does not award the contract to the briber. Knowing this, a rational firm X does not offer any bribe at all. With the number of repetitions being fixed ex ante, as in our case, the subgame perfect equilibrium does not change compared to a one-shot game, yet one expects subjects to show a greater inclination to deviate from the Nash equilibrium (also in corruption experiments, see Abbink 2004). We conducted the experiment in two treatments: The individual decision (ID) serves as the control treatment and the group decision (GD) as the experimental treatment. In the ID treatment, one person plays in each role. In the GD treatment, both roles are taken by one three-person-group each. 8 The gender composition differs across groups, so than we can explore how this influences the decisions.' In both treatments, the game is repeated 10 times with partner matching, i.e., every firm always played with the same pubic official. 10 After each period, every subject or group, respectively, gets to know her opponent's choice and her own and opponent's payoffs. In the GD treatment, the reported payoffs are those that every member of the group receives. We ran the experiments from December 2009 to May 2010 at the University of Kassel in Germany and the University of Nankai in China. The two countries are representing different level of corruption" and different styles of group interaction 12 . A total of 504 subjects participated in the experiments, 64.1% of them female. 13 The number of inde-

7

8

9

10

11

12 13

For a more extensive argument in favor of loaded framings see Cooper and Kagel (2003). Unlike Abbink and Hennig-Schmidt (2006), Barr and Serra (2009) found the results of their corruption experiment to depend on whether instructions were loaded or not. We chose three-person teams because it may have important implications for the decision-making procedure, for example avoiding of unanimous decisions, see Luhan et al. (2007). Since we have groups of three we can also say something about the difference between unigender groups and groups which are not unigender but still have a strict majority of persons of a given gender. On the impact of gender composition on group decisions see for example Dufwenberg and Muren (2005). Li (2011) also reports on a staff rotation treatment, but this was played only with individual decisions. In Transparency International's Corruption Perceptions Index, Germany's score for 2010 is 7.9, while China's is 3.5 on a scale from 0 (highly corrupt) to 10 (very clean). For a detailed discussion see Li (2011), pp. 86-99 and 141-143. In each country 35 women and 31 men participated in the ID treatment. In the GD treatment, 96 women and 96 men in China, and 93 women and 87 men in Germany took part.

Group Decision Making in a Corruption Experiment • 213

pendent observations is 33 in ID-China, 32 in GD-China, 33 in ID-Germany and 30 in GD-Germany. 1 4 T o obtain further insights into the process and structure of g r o u p discussions, the communication within each group was only possible via paper and pencil, and the decision protocols were stored for analysis (see section 4.3). At the beginning of every round, each g r o u p member was asked to m a k e an individual proposal to his or her group. If all the proposals were identical, then this decision was entered directly into the PC. Otherwise, they had 3.5 minutes to discuss and agree on a joint decision and enter it into the PC. The experiments were computerized with z-Tree (Fischbacher 2007). A c o m m o n concern in intercultural experiments is that "between-country differences might arise because of uncontrolled procedural differences or because of uncontrolled personal differences a m o n g the experimenters" (Roth et al. 1991: 1071). In order to avoid this kind of possible experimenter effects, all sessions were presented by the same person, one of the authors, in Germany and China. Each session began with a brief verbal introduction. The written instructions (in German and in Chinese, respectively) including payoff tables which listed the payoffs for all possible strategy combinations were handed out to each subject. Participants had some time to read them carefully and afterwards possible remaining questions were answered. In the GD-treatment, every subject was asked in each round to complete a record for the g r o u p discussion, containing his or her individual proposal to the group, and rules applied for finding a joint decision. Furthermore, in all treatments the subjects were asked to fill out a questionnaire after the experiment, which included demographic information and individual experiences with corruption in their home country. T h e experiment lasted about 1 hour in the ID treatment or a b o u t 1.5 hours in the G D treatment. The average payoff was 12.95 € in Germany and in China 35.80 ¥ (i.e. 3.64 €). 1 5 4

Results

Section 4.1 illustrates our main findings, comparing individual and g r o u p decision making of firms and public officials in China a n d Germany. W h e n comparing the levels of corruption, we also consider different gender constellations within the groups and dynamics of the bribes offered. In section 4.2, we examine the effects of the more eyes principle and gender constellations in a multivariate analysis, and in section 4.3 we take a closer look at the decision making process of the groups. 4.1

Level of corruption

We measure the level of corruption by the following variables: the average bribe offered by firms and the frequency of reciprocal decisions of the public official.

14

15

We controlled the gender composition: In China: FFF (three female participants); FFM (two female and one male); FMM (one female and two male) and M M M (three male) 8 groups respectively. In Germany: FFF: 8 groups, FFM: 7 groups, FMM: 8 groups, and M M M : 7 groups. The conversion rate in the German sessions is 40 experimental Dollar = 1 Euro and in China 10 experimental Dollar = 1 RMB. This was based on the standard hourly wage paid for a student research assistant in Germany and China.

214 • Sha Li, Christoph Bühren, Björn Frank, and Haiying Qin

Average bribes Figure 2 depicts the average bribe offers of the firms in the ID and G D treatment. In the G D treatment, the Chinese average bribe offers are significantly higher than the Germ a n according to a t w o sided Mann-Whitney-U test 16 (p = 0.048). In the G D treatment, the Chinese offer on average 4 . 0 3 experimental dollars. C o m p a r e d to 3.29 experimental dollars in the Chinese ID treatment, this is an increase of 2 2 . 5 % (p = 0.129) showing a rather detrimental effect of the more eyes principle. In Germany, the introduction of the more eyes principle leads only to a small rise of 6 . 1 % (from 2.93 to 3.1, ρ = 0.751). 5

• China Germany

Figure 2 Average bribe offered

Bribes offered by gender In Figure 3, the average bribe offered is decomposed with regard to the different gender constellations. The G e r m a n all female groups (FFF) offered the lowest bribe to the public officials - on average 1.38 experimental dollars (compared to 3.73 of their Chinese counterparts). This is much less t h a n the average bribe of 4.43 given by the G e r m a n all men groups ( M M M ) , and the offers are also significantly smaller t h a n those by the German mixed gender groups. However, G e r m a n firms with t w o females and one m a n (FFM) offered significantly larger bribes than G e r m a n firms with one female and t w o men (FMM) (4.23 vs. 2.70). In China, we find n o remarkable different bribing decisions between purely male and purely female groups. 1 7 The only significant difference arises for FFM, which show the highest level of corruption. 1 8

lé 17 18

All p-values in section 4.1 are based on two sided Mann-Whitney-U tests. The average bribes are 3.72 and 3.73, respectively. 41.6% higher than women groups and 42.0% higher than men groups (p = 0.057 and 0.140); 55.3% higher than FMM groups (p = 0.114).

Group Decision Making in a Corruption Experiment · 215

2

I

F

t ππ mi FFF

FFM

FMM

• China • Germany

MMM

M

Figure 3 Average bribe offered, by gender constellation19

Public official's reaction Figure 4 shows the frequency of the four different actions the official is able to choose (see section 3). It serves as a first indication that the more eyes principle has a strong positive impact on reciprocal (i.e. most corrupt) decisions and a negative effect on the least corrupt decision to reject and report the bribe. Note, however, that we are not controlling for the size of the bribe in this subsection. Compared to individual decisions, the share of reciprocal decisions in the more eyes treatment is 7 3 . 1 % percent higher in Germany and 4 1 . 2 % in China (p = 0 . 0 1 2 and 0 . 0 5 1 , respectively). Correspondingly, rejecting and reporting decreases by 7 9 . 7 % in Germany and by 6 8 . 5 % and China (p = 0 . 0 1 9 and 0.037).20

• China • Germany

reciprocity

opportunistic behavior

rejecting and returning

rejecting and reporting

Figure 4 Public officials' decisions

19

F a n d M refer t o I D t r e a t m e n t a n d m e a n s the s u b j e c t is a female o r a m a l e . F F F , M M M ,

FFM

a n d F M M refer t o G D t r e a t m e n t . 20

N o significant differences c a n be observed with respect t o the public official's p r o p e n s i t y on o p p o r tunistic b e h a v i o r (bribe t a k i n g w i t h o u t reciprocity) a n d t o reject a n d return bribes.

216 · Sha Li, Christoph Bühren, Björn Frank, and Haiying Qin

Reciprocal decision by gender In a one shot two player game, Lambsdorff and Frank (2011) find that women are less likely to reciprocate if they were in the role of agents. Our experimental results confirm their findings: Figure 5 shows the average frequency of reciprocal decision of public officials if a bribe was offered to them. German pure women groups have the lowest percentage for reciprocal (i.e., extremely corrupt) decisions (15.63%) and German pure male groups have the highest tendency for reciprocity: 71.43% of their decisions are reciprocal when a bribe is offered to them. In China, reciprocity is only noticeable different between the mixed gender groups: In the FFM groups, the percentage of reciprocity is 37.15%. If two men participate in the group, it is significantly higher (67.33%, ρ = 0.035).

I China Germany

κ F

FFF

FFM

FMM

MMM

M

Figure 5 Public officials' decision to reciprocate after being offered a bribe Altogether, 3 2 % of the bribes of German individuals and 5 1 % of German groups were successful. In China, 4 0 % of the individual bribes and 5 6 % of the bribes conducted by groups were successful. In both countries, the differences in the success rate between individual and group decisions are highly significant according to two sided Fisher exact tests. Dynamics of bribe offers So far, we just looked at the independent observations, i.e. the averages of the ten rounds in the different treatments. Thereby, we might have overlooked important dynamic differences. As can be seen in Figure 6, the dynamics in our 10 round experiment are similar within the ID and GD treatments. 21 Starting from a relatively high level of corruption, there are little ups and downs during the bribery game (as a rule starting with a decrease, except for GD-Germany), ending with a strong last round effect, similar to public good games, where the average bribe is close to zero. To summarize, we cannot find evidence that the more eyes principle reduces the level of corruption. We find rather the opposite, particularly in China. Furthermore, we observe

21

The same is true for the different gender constellations and the dynamics of the choices of the official.

Group Decision Making in a Corruption Experiment • 217

6

5

4

3

2

1 0 Figure 6 Dynamics of bribes offered in the ID and GD treatments (10 rounds) a significant difference between w o m e n and men groups in German treatments, while in China this is not the case. In the following, we check these findings with probit regressions. 4.2

Regression results

We ran four ordered probit regressions. The first t w o (Table 1 ) try to isolate the effects of the more eyes principle, i.e. the comparison between g r o u p and individual decision making, and the other t w o (Table 2) analyze the effects of different gender constellations within the groups. 2 2 The dependent variables are the size of the bribe offered by the firm (0, 2, 4, 6, 8) and the a m o u n t of corruption chosen by the official (1: reporting, i.e. not corrupt; 2: rejection; 3: opportunism; 4: reciprocity, i.e. highly corrupt). W e pooled the German and Chinese observations and included a China d u m m y in the regressions as well as control variables taken f r o m the post-experimental questionnaire: available income, belief about h o w successful bribery is in general, and assessments a b o u t anti-corruption policy and trend of corruption in the home country. In order to catch the dynamics of our experimental data, we used rounds as single observations (rather than averaging over rounds as in section 4.1). Therefore, we adjusted the standard errors in our regressions for clusters at the individual level. As we can see in all of our regressions, the likelihood of being corrupt increases highly significantly if the negotiation partner is corrupt as well - it is more likely that a public official decides corruptly if a firm offers a high bribe, and it is more likely that a firm offers a high bribe after a corrupt decision of its public official. The control variables of our post-experimental questionnaire seem to play a minor role. The significant negative coefficients for the variable " r o u n d " in the regressions for the firms' decisions indicate a decreasing level of bribes over time that is mainly driven by the last round effect (Figure 6). The first regression of Table 1 confirms that Chinese participants in the role of firms offer significantly higher bribes. In combination with our descriptive analysis in Figure 2, we can conclude that the failure of the more eyes principle in firms is especially caused by high

22

O u r main results of Tables 1 and 2 remain intact if we exclude the control variables.

218 · Sha Li, Christoph Bühren, Björn Frank, and Haiying Qin

a m o u n t s of bribery of Chinese groups. The second regression of Table 2 draws an even worse picture of the more eyes principle in administrations: The likelihood that officials act more corruptly increases if they decide in a group. Having a look at the marginal effects in the appendix, we can see that for example the probability to choose the most corrupt action (reciprocity) increases by more than 15 percentage points w h e n officials decide in a committee. The group d u m m y is not significant in the first regression a n d significant with a negative sign in the second. Hence bribers seem not to anticipate that public officials will be more corrupt w h e n deciding in committees. 2 3 Table 1 Effects of the more eyes principle in ordered probit regressions Dependent variable: Amount of bribery from the firm (0-8)

Coefficient China (dummy) Group (dummy) Round (1-10) Choice of the official last round (1: reporting (least corrupt), 4: reciprocity (most corrupt)) Amount of bribery from the firm (0-8) Available income (1: less than 400 €, 4: more than 1200 €)

Dependent variable: Choice of the official if she or he gets a bribe (1: reporting (least corrupt), 4: reciprocity (most corrupt))

Standard error

Coefficient

Standard error

0.103

0.023

0.121

-0.051

0.105

-0.057***

0.121

0.392*** 0.012

0.100 0.014

0.511***

0.058

0.196***

0.024

0.320***

-0.097

0.069

0.141*

0.075

0.037

0.081

0.005

0.078

Assessment about anticorruption policy in home country (1: very ineffective, 5: very effective)

-0.075

0.083

0.022

0.078

Assessment about trend of corruption in home country (1: increasing, 3: decreasing) η

—0.119

0.098

0.129

0.117

728

756

Pseudo R2

0.057

0.098

Belief about success of bribes in general (1: extremely uncertain, 5: extremely certain)

Notes: *: significant at the 10% level; * * : at the 5 % level; * * * : at the 1 % level; standard errors adjusted for clusters at the individual level; see appendix for marginal effects. Pseudo R 2 not t o be interpreted as the R 2 in OLS regressions.

23

We thank an anonymous referee for suggesting this interpretation.

Group Decision Making in a Corruption Experiment • 219

The ordered probit regressions of Table 2 allow us to analyze the effects of different gender constellations. The first regression of Table 2 shows that firms that are exclusively led by women are less corrupt than firms that are solely led by men. The marginal effects reported in the appendix indicate e.g. that the probability to offer the highest possible bribe decreases by more than 19 percentage points when the decision is made by three women rather than three men. However, the second regression of Table 2 fails to confirm this finding for administrations. Table 2 Gender constellation (MMM is the reference category) Dependent variable: Amount of bribery from the firm (0-8)

China (dummy) FFF (dummy: group with 3 women) FFM (dummy: group with 2 women and 1 man) FMM (dummy: group with 1 woman and 2 men)" Round (1-10) Choice of the official last round (1: reporting (least corrupt), 4: reciprocity (most corrupt)) Amount of bribery from the firm

Dependent variable: Choice of the official if she or he gets a bribe (1: reporting (least corrupt), 4: reciprocity (most corrupt))

Coefficient

Standard error

Coefficient

Standard error

0.234 -0.556***

0.219 0.226

-0.056 -0.209

0.212 0.191

0.192

0.234

-0.189

0.155

-0.288

0.215

0.201

0.154

-0.072*** 0.776***

0.020 0.104

-0.012

0.022

0.252***

0.032

(0-8)

Available income (1: less than 400 €, 4: more than 1200 €) Belief about success of bribes in general (1: extremely uncertain, 5: extremely certain) Assessment about anticorruption policy in home country (1: very ineffective, 5: very effective) Assessment about trend of corruption in home country (1: increasing, 3: decreasing) η PseudoR 2

0.047

0.167

0.159

0.150

-0.061

0.191

0.045

0.113

-0.061

0.175

-0.001

0.094

0.035

0.303

-0.077

0.197

359 0.116

366 0.124

Notes: *: significant at the 10% level; **: at the 5 % level; ***: at the 1% level; standard errors adjusted for clusters at the individual level; see appendix for marginal effects. a: The coefficients for the FFM and FMM dummies differ significantly according to Chi 2 -tests (p = 0.016 for the firms and ρ = 0.046 for the public officials)

220 · Sha Li, Christoph Bühren, Björn Frank, and Haiying Qin

4.3

A look into the groups

In addition to the experimental data, we also studied the process of group discussion by asking our participants to fill out a protocol during the discussion and a questionnaire after the experiment. Before the group discussions, every participant was required to make an individual proposal for her or his group in every round. Figures 7 and 8 show the dynamics of the average group decisions of firms and the average individual proposal of firm members in China as well as in Germany. In the first two rounds in China (Figure 7), the average bribe within the groups is higher than the average of the individual proposals. Comparing the median proposals in these groups to the respective compromise levels, this difference is significant according to a Wilcoxon rank sum test (p = 0.06) which is in line with the comparison of individual and group decisions in Figure 2. Yet the differences of decisions and proposals diminish throughout the experiment. After round 2, the amount of the bribes suggested by Chinese individuals to their groups increase steadily. Their proposals and the group decisions level off at a very high level up to round 8, followed by the strong last round effect for both individuals and groups (cf. Figure 6). Having a look on rounds 1 to 10, the averages of the median proposals and group compromises do not differ significantly (4.08 vs. 4.14). In Germany, the dynamics of the individual bribe proposals and the group compromises differ from China. In the first 4 rounds, the average individual proposals are higher than the averages of the group decisions (Figure 8). From round 5 to 8, this relation reversed, again followed by the same last rounds effect as in China. That means that in the first half of our experiment, German groups find a moderate compromise that was mostly influenced by low bribe proposals. After that, the German groups agreed on higher bribes compared to the individual proposals, but still on a modest level. In Germany, the averages of the median proposals and group decisions are exactly the same (3.08). Having a look on the standard deviations of the individual proposals, we found the lowest standard deviation in German pure female groups: 2.38 compared to 3.36 on average in the other gender constellations of China and Germany. In combination with Figure 3 that means that most of the time German women could agree unanimously to offer low or no bribes. We also asked our participants ex post about their activity during the group discussion. According to their self assessments, there are significantly more "most active participants" in Germany than in China: 33.5% vs. 21.9%. 2 4 There is one group in Germany in which all participants consider themselves as being "most active", while there is no such group in China. Furthermore, in Germany there are 14 groups in which two participants are "most active" according to their self-assessment, but only 4 in China. Correspondingly, more subjects reported to be "least active participants" in China than in Germany. 25 These results could be due to the more individualistic culture of Germany, emphasizing self-realization and independence. Germans are more used to being confronted with different opinions, debate or dispute. On the other hand side, the (Chinese) participants that assessed themselves as least active might be characterized as modest, shy, or not willing to take responsibility for the groups' actions. When comparing the average bribe offered, we observe that the most active subjects in Germany are significantly more corrupt than the other group members. In contrast, the 24 25

ρ = 0.018 according to a Fisher exact test (two sided). 18.9% vs. 8.1%, p = 0.003 according to a Fisher exact test (two sided).

Group Decision Making in a Corruption Experiment · 221

— —

average of individual proposal

— — — average of group decision

0 1

2

3

4

5

6

7

8

9

10

Figure 8 Individual proposals and group decisions of bribes offered in Germany

Chinese most active participants were less corrupt than their g r o u p members. 2 6 A possible explanation might be that those w h o act as the most active g r o u p member are more liklely to decide independently f r o m the respective society's norms. This is speculation, however. W h e n it comes to the decision whether to accept bribes or not, we d o not observe any significant difference of the acceptance rates between the most active participants and other group members, neither in China nor in Germany. 26

In Germany, the average bribe is 3.69 (standard deviation: 3.43) from the most active subjects and 2.72 (standard deviation: 3.12) from the other subjects (p = 0.064, two sided Mann-Whitney-U test); in China 3.34 (standard deviation: 2.83) and 4 . 0 7 (standard deviation: 3.17), respectively (p = 0.376).

222 · Sha Li, Christoph Bühren, Björn Frank, and Haiying Qin

5

Conclusion

The four eyes principle failed to reduce corruption in our laboratory experiment; it rather increased the bribes paid and the likelihood of accepting them. Attempts to explain this surprising finding are necessarily speculative, but they provide hints with respect to fruitful approaches for future research. It is a classic result in social psychology that often groups take more risky decisions than their individual members w o u l d (Stoner 1968). Because bribers run the risk of whistleblowing by public officials, they face a kind of risk. However, this would be an insufficient explanation. O u r subjects' choices d o not only involve their own risk, but also others' payoffs, hence moral norms play a role. M a z a r and Aggarwal (2011) ran a psychological experiment in which a collectivist priming increased the inclination to bribe in a hypothetical choice situation (compared to an individualist prime); this effect was driven by subjects' lower perceived responsibility for their own actions. A beautifully simple demonstration h o w groups can decrease its members' moral costs when real payoffs are used has been provided by D a n a et al. (2007) in a modified dictator game. The " d i c t a t o r " is a team of t w o players. W h e n one of them chooses an inequitable outcome, this does not actually lead to unequal payoffs unless the other also opts against an equal distribution. Thus, subjects share the responsibility if they show egoistic, non-altruistic behavior, and the share of subjects doing so more than doubled (compared to a n o r m a l dictator game). Whatever channels are involved, we submit that the effect is a fragile one, as its size very much depends on national culture and gender composition. In our experiment, the introduction of the f o u r eyes principle led to a strong increase of corruption in China. In Germany, this effect was in the same direction but far smaller. This could be related to cultural differences of the t w o countries. According to Hofstede et al. (2010: 95-97), the G e r m a n culture is much more individualistic t h a n the Chinese. Chinese individuals are supposed to accept collective n o r m s more strongly and to behave according to them. The rapid process of rapprochement between individual proposals and the g r o u p decisions in China may well present this g r o u p orientation. In our German subsample, we found a strong gender effect: Purely female groups were considerably less corrupt than purely male groups. In China, we could not replicate this effect. However, we do not w a n t to oversell our results on culture and gender effects. Specifically, we refrain f r o m suggesting policy recommendations with respect to gender composition of groups. One reason is that our subjects did not self-select into our experiment, at least not in the way real managers and public officials choose careers in their respective fields. Groups composed of three female managers might be much more corrupt than the purely female groups in our experiment. As for national culture, we do not think that individual Chinese persons are inherently more corrupt than individual Germ a n persons. However, they are all influenced by the norms and traditions of the societies they live in. Even in an artificial environment such as a lab, expectations about others, even expectations about others' expectations, and hence cultural norms, play a role. Against this background, w h a t could m a k e the four eyes-principle work? One variant would be to introduce some kind of asymmetry between g r o u p members; they could be members of different hierarchy levels. Another new feature could be the opportunity of blowing the whistle after the g r o u p decision (by each member, or some members, of the group). Whatever the outcome of future w o r k in this direction m a y be, the existing evidence suffices to conclude that the implementation of the four eyes principle must be guided by facts, and its design must be thoughtful, otherwise it could lead to the disappointing results that we have demonstrated in our experiment.

Group Decision Making in a Corruption Experiment • 223

_

J3 I« οι "o-S-g.. V 3 C •«O i - O "O ti oβ>«t; «ι g -c ¡S ° '»•¡E » I »i· 1.2 Ι ό ».Ht g S-Ç e o^-e.«c =..s> .oc U

%

4» •8 fi « E >Ρ a, >> •a S SÎ ** û o ID XnIj Io ,Ο

> «

01 α. Ε «

σι coν 00 o •a- in m o ο Τ-Ο ο ο Ο Ο *

*

ιη Ο Ο ο Ο ο Ο

00 σ\ Γν ο Ο ο ο ο

tj- m m ιη 0 ο ο ο 8m Ο Ο ο ο ο ο ο Ο ο

I -O τ- tm flIN(N ÍN ÍX N • τq- OTl-Tloo oo oo oo do oo oo oo

(N ÍN o o o o o o I~

γμ σ\ o o o o o PO τ- o o o ο o o o o o o o I

VD ΙΏ ÍN O O o o o Iw

T- O m O O Τ- O Τo o Ο o Ο o o o o o o o o o Iw

lo m τ- O O O

Ϊ-Φ OID T- o o o o o o o d ow d o I I

(Ν vû o ο ο τ- ο ο ο ο ο d ow d os II

ι- Ν 00 Ν o 'SN O τ- m τ- mÍ τ- o o o o o d wo d o I —

ιη ιη o σι oo ο m in m (Ν in (Ν (Ν ro Tf m o o o o o o o o d o d o o so o o I I - I~

ΓΜ M- ΓΜ τ- o o O o (N o O O o o o O O O O o o o o d o d o I

S mo

ΓΜ ΓΜ O τ- o o m- m ο O O O O O O O ° § SS d o d o I " ~

τ- τ- o o o 88 88 88 o o d o d o o o o Io

Ν in IN ^ (Ν ι- O ·ί O O (NO O O o o od od do do I

8 Oo

,^ Γν m ιη 0 -m m Τ0 Τ-Ο Ο Ο ΟI Ο Ο ο

τ mΐ·t IN ο ο ο ο

Γ 8 88 *

*

ο ÍN ο ο ο ο ο

^ * ο 0 0( IΝ N τ-ο οI ο

mm τ- o oo oo oo oo

m τ- m O O o oo o o

^

^

m σ I> N δ ο οι ο

r^ m ro «í íN o -3· o oo oo dd

mm o o o o o o

in TJo ο o o o o

vo σ\ IN IN o o o o

ISI

«

VI Ο 0J Χ !t ΐ s S «-O JS ö .y i S > -c ¡tp ^ < -o < U O

•aOj ¿o (β ΙΛ»Λ % s 1 ÛQ = vi

O Q.

224

·

Sha

Li, C h r i s t o p h

(Ν (Ν Ο Ο ι

-O

Bühren,

Björn

Frank,

Tt co o ο

m co o ο ι

ΚΩ Γ*·* ο ο

ιη t-v ο ο ι

ΓΜ KD Ο Ο

ι η 00 ιη ο ο ο ο

rt· in o ο

Γν f ο ο

ο ιη ο ο

,s τ— τίο ο

KD τ ο ΓΜ o o d ö

m (Ν o o

in (Ν o ò

Ο

oo

o o

o d

ο ΚΩ Ο Ο

ο ο

and

Haiying

σ\ Ο ο Ο ο Ο ο ι

Qin

O O o

m o oo i n *£> KD io o o o ö o

(Ν to o

η

c S: β>

™ —

Q

O α> S -C

S O S

m o o o

·» E Ρ

φ

>

•ο

δ

S--S Q 'S 3 O E


m o o d

KD o o o o o o

r— 00 00 o o d o

"S

σ\ Tt

* _ 00 τ ΙΌ ο Ο ο ο I

m o

o

^*

•ao o

ο

(Ν ο ο ο ο ο ο I

•ίο ο ο

•ίο ο ο

ο ο

ιη ο ο ο I

00 ο ο ο

•ίο ο ο

(Ν ο ο ο

ΓΝ ο ο ο ο ο ο

ο ο ο

•q· ο ο ο

* * ΓΜ ο ο ο ο ο ο

I

ο

ιη ο ο ο

ID ο ο ο ι

KD Ο Ο Ο

ΓΜ ο ο ο

* * * Ο δ ο

ιη ο ο ο

ο ο ο

ο ο ο

I

r-. ο ο ο

t o o

o oo o o

m ΓΜ O o

Φ ΓΝ O O 1

ic o o o

# VO o o o 1

m o o o

*

ΓΜ o o o o o 1

m o ο

I

,, (Ν •ίo o

(Ν o o

1

,,

o o o o

m ΓΝ o o

m ΓΜ i n o o o o

o o o o

o o o

co o o o

o (N o o

m o o o

o o o o

ΓΝ o o o

ΓΜ o o o

vn o o o

τ-

m ΓΜ ο ο ι

m o o

,^

* ο ΓΜ (Ν ο ο ο ο

m o o o

m t— o O o o O I

o o o

1

,, ΓΜ ΙΌ Ο

ο

Ο

ο

ο ο

,^

r

ιη νο ο ο

,

m

m

ο ο

τι

,

Ο

r— ο ο ο

ιη ο ο ο

ΓΜ ο ο ο ο ο ο I

ΓΜ ο ο ο I

•ίο ο ο

ΤΟ ο ο I

ιη ο ο ο

ο ο ο

ο ο ο

ο ο ο

ο ο ο

ΚΩ (Ν ο ο ο ο ο I

m ο ο ο I

ιη ο ο ο

ο ο ο

ιη ο ο ο

ΓΜ ο ο ο

ιη ο ο

ο ο

ο

ο

00 ο ο ο

O o o

O ΓΜ rv

o o

00 ο ο

ο ο ο

I

ο

* * *

* * ιη ο ο ο I

ΓΝ Γ-. 00 f»O o o o w I

S

,

* * *

o o o

¿

. , ΚΩ ο ο ο

g o

g.,9-

o jz U

m ο ο ο

ao

Έ s .E S .y t Q - t

^ KD ιη ο Ο ο ο 1

fg £

- C ¡fc υ o

io

E

© ο ο

ο ο

-2í JD

Ε