Combating Corruption: On the Interplay between Institutional Quality and. Social Trust

Combating Corruption: On the Interplay between Institutional Quality and Social Trust Christian Bjørnskov * May 19, 2008 Abstract: The aim of this paper is to explore under which conditions institutional quality leads to lower corruption. A model of a simple economy where firms both chose between bribing or not bribing bureaucrats to avoid costs and between entering the official or unofficial economy shows that the effects of increasing institutional quality may be ambiguous due to perverse effects of institutions in the unofficial economy. Employing a recent index of corruption based on objective data, the paper shows that formal institutions are more effective in combating corruption in countries with high levels of social trust. The paper concludes by discussing the welfare and political implications of the findings. JEL Codes: D72, K42, Z13 Keywords: Corruption, Rent-seeking, Social trust * Aarhus School of Business, University of Aarhus, Department of Economics, Prismet, Silkeborgvej 2, DK-8000 Aarhus C, Denmark; Phone: +45 89 48 61 81; e-mail: ChBj@asb.dk. I am grateful to Axel Dreher for providing me with the corruption data, and to Partha Dasgupta and Maria Ziegler for valuable comments on an earlier version. Christina Bjerg provided excellent research assistance. The usual disclaimer naturally applies. 1

1. Introduction The economic and social consequences of corruption are difficult to neglect and, defined as the misuse of public office for private gains, one would think that it could fit into most conceptions of political economy or public choice without complications (Svensson, 2005). However, the economics of corruption only grew to be an important area with the advent of internationally comparable data since the mid-1990s. The voluminous amount of research published since then indicates that corruption, among other things, hampers economic growth (Mauro, 1995; Sekkat and Méon, 2005), distorts government spending (Mauro, 1998), affects productivity (Méon and Weill, 2004) and in the long run leads to very considerable losses of income and human welfare (Kaufmann et al., 2005). A number of studies have therefore explored why some countries seem to experience substantially more corrupt behaviour than others. Treisman (2000, 2007) summarizes the determinants of corruption as income, an established democracy, press freedom, rule of law, openness to trade, and not relying on resource exports such as oil or minerals. De Haan and Seldadyo (2006) instead employ confirmatory factor analysis to pool a large number of explanatory variables into a set of factors, of which only one regulatory capacity is robustly associated with corruption. This factor includes such variables as the rule of law, judicial independence, government effectiveness, political stability, regulatory quality, freedom from labour market regulations, international trade, and secondary school enrolment. A cluster of other variables including a number of Treisman s main findings also survives a somewhat weaker test of robustness. 2

However, cultural and social factors are also likely to be associated with the incidence of corruption, not least because corruption at the end of the day is evidence of unambiguously dishonest behaviour. Even if many studies do not discuss this factor, social trust often taken to be a measure of the strength of honesty norms in society has been identified as a statistically strong and quantitatively important determinant of corruption in recent empirical studies (Uslaner, 2002, 2004; Bjørnskov, 2003). It is nevertheless unclear whether social trust in itself is associated with corruption, or if other factors are stronger in its presence. As such, one of the many remaining questions in this strand of the corruption literature is whether the rule of law and social trust are substitutes or complements. It is easy to make the case for legal protection being more necessary in fighting corruption when the population is, in general, not trusting. Yet whether the legal system is also as effective in that situation remains an open question, and the argument could be made that social trust instead reinforces the effect of legal protection. The purpose of this paper is to explore whether social trust reinforces the effectiveness of the quality of formal institutions or more likely works as a substitute. The theoretical considerations that allow for an unofficial economy suggest that the legal system may only be effective when at least part of the population acts honestly. A set of empirical results based on unique panel data on corruption and legal quality supports that the quality of the legal system is substantially stronger associated with corruption when social trust scores are relatively high. The rest of the paper is organized as follows. Section two outlines a simple theory of how legal quality, moral behaviour and their interaction affect the likelihood of 3

corrupt transactions. Section three describes the data to be used in section four, which estimates a set of determinants of corruption. Section five concludes. 2. Theory Corruption is often defined as the misuse of public office for private gains (Treisman, 2000). Most models consequently assume that corruption occurs in the necessary interplay between government agencies or institutions and the business sector. As in many contributions to this literature, I therefore assume that firms need a permit in order to produce. In this setting, corruption can arise because bureaucrats at a government agency hold a monopoly of issuing these permits. The resulting game is depicted in Figure 1. Insert Figure 1 about here In all cases I use backwards induction to solve the game, starting with the choice of a bureaucrat working in the government agency responsible for issuing permits and monitoring their use. The choice the bureaucrat faces in both the official economy and his function vis-à-vis the unofficial economy is to accept a bribe or not. With a wage w, a bribe b, an outside option worth w out (which may include a fine or prison sentence), a moral cost M (including both external and internal costs such as shame, the risk of being socially ostracized etc.) of accepting a bribe, where M~iid(M A,σ M ), and a risk of 4

being caught accepting the bribe given by the quality of the legal system that determines this risk, λ, the bureaucrat will chose to accept if the following inequality holds. 1 ( λ)( w + b) + λwout M > w 1 (1) Solving this inequality gives the minimum bribe, b min, acceptable to a risk-neutral bureaucrat, which will differ between bureaucrats since M is a stochastic term. λ M bmin = ( w wout ) + (2) 1 λ 1 λ Here, it is worth noting that the first term exhibits increasing returns to legal quality while the second term implies that social trust, which in the following is used as empirical proxy for M since it reflects individuals moral costs (Uslaner, 2002), is associated with a higher minimum bribe and is complemented by legal quality. Hence, the minimum bribe in any situation is determined by the interplay between the quality of the legal system and the idiosyncratic moral costs of the bureaucrats whom firms face. 2 Turning to firms profits, the economy consists of many even-sized firms that face the choice of either 1) bearing a cost F, which can be thought of as either a handling fee for receiving the permit or a cost of conforming to regulations, or 2) attempting to bribe 1 The relation between the theoretical moral costs and the measure of social trust employed in the empirical part of the paper is based on the assumption that the moral costs are increasing in social trust. Both the internal motivation to avoid dishonest behaviour, such as shame, and external motivation, such as the risk of being socially ostracized, are stronger when an individual believes that most other people are not likely to engage in such behaviour. 2 If it furthermore is assumed, as in Yavas (2007), that the moral cost decreases with the corruption of the surrounding society the possibility arises that a self-enforcing equilibrium without any corruption exists, even if the same distribution of moral costs in another society would result in some corrupt behaviour. For simplicity, I do not consider this option in the following. 5

the bureaucrat to avoid paying the fee. Firms profit function is given by π(φ) - c, where φ is a stochastic productivity parameter, and c is a fixed investment cost. 2.1. Firm behaviour in the official economy Focussing first on what might be considered the standard part of the model, illustrated in the lower half of the game tree in Figure 1, the question here is under which conditions bribes arise in the official economy. In other words, under which conditions are the expected profits from corrupt behaviour larger than those of honest behaviour as in the right-hand side of (3)? In (3), p is the probability that the average moral costs are sufficiently low to make it rational to attempt to bribe an average bureaucrat, i.e. a bureaucrat with the expected average moral cost. This long expression reduces to the short expression in (4), which simple states that firms will attempt to bribe an official as long as the necessary bribe is smaller than the expected fine. ( 1 λ) [ p( π ( ϕ) c bo ) + ( 1 p) ( π ( ϕ) c F )] + λ[ p( π ( ϕ) c F b ) + ( 1 p) ( π ( ϕ) c F )] > ( π ( ϕ) c F ) o (3) b o < (1-λ)F (4) This obviously means that all bureaucrats will require the maximum bribe since they collectively hold monopoly power and firms are either randomly assigned to a bureaucrat or are otherwise restricted from searching for a corruptible agent. Hence, the probability of a bureaucrat taking a bribe of the size in (4) is given by (5). { M < ( λ) F ( w )} 2 p = prob 1 λ (5) Focussing at first on an entirely official economy rather clearly indicates that the quality of the legal system, λ, is always effective, but may be more so in countries with a high average level of moral costs since the size of bribes in (2) increases w out 6

proportionally with λ but the probability of corrupt behaviour decreases disproportionally, depending on M. The latter conclusion thus also holds for the deadweight losses of corruption that in this simple economy can be approximated as the product of (2) and (5). 2.2. Allowing for an unofficial economy However, introducing the additional possibility that firms can choose to produce without a permit i.e. in an unofficial, grey, shadow or underground economy complicates the analysis. Allowing for this possibility is nevertheless closer to reality in most developing countries where the unofficial economy can affect up to three quarters of GDP (Schneider and Enste, 2000). I therefore effectively allow for an exit option from the lower half of the game tree as many firms in most countries exist in an unofficial economy, which in the context of this model will consist of firms producing without a permit. In most countries the agency employing the bureaucrat not only handles the approval rubber stamp or not but also controls whether firms conform to the regulations stipulated by the approval. Hence, I make the simple assumption that bureaucrats from the government agency issuing permits also make control visits to ensure that firms are in possession of a valid permit. Firms are nevertheless not controlled from the beginning of a period, but face an expected waiting time d, depending on the capacity of the government institutions, before receiving a control visit. First, for considering paying a bribe in the unofficial economy, the expected profits have to outweigh simply ending production after having produced for the fraction d of a period. As such, b u, the bribe paid in the unofficial economy, becomes 7

smaller over time, as the willingness of firms to pay a bribe to stay in production decreases. Investing in the capacity of the regulatory authorities can thus actually make bribery in this part of the economy more likely, given that firms choose to produce unofficially as in (6). π b ( ϕ) c bo > dπ ( ϕ) < ( 1 d ) π ( ϕ) o c (6) If the expected bribe is lower than this limit and given that q is the likelihood that the bureaucrat performing the control visit at the firm is corruptible in the sense that his minimum bribe is smaller than the benefit of paying it, going underground yields an expected profit π u of operating illegally. Making the simplifying assumption that the expected waiting time for a control visit is given by d = a - μλ, 3 i.e. that overall institutional quality is accompanied by capacity, bribes will occur in the underground economy when (7) holds. Consequently, bribes are more likely in the underground economy than in the official economy if (8) holds: λ M a out + (7) 1 λ 1 λ ( μλ) π ( ϕ) < ( w w ) π λ < 1 (8) a μλ ( ϕ) F As such, the structure of the model allows for a perverse effect of increasing the costs of getting a permit, as it is likely to drive some firms into the underground economy and at the same time is likely to increase the probability of corruption in this 3 This type of assumption allows for a variable association between institutional quality and institutional capacity, given by μ. While this association is likely to be positive, it may not be perfect; hence the likely parameter space is 0 < μ < 1. 8

part of the economy. Based on this assumption, the probability of a bureaucrat accepting a bribe in the unofficial economy, i.e. on a control visit, is given by q in (9). { M < ( a μλ)( 1 λ) π ( ϕ) ( w )} q = prob λ If the expected bribe in the unofficial economy is larger, the expected profit is given by (10). Hence, firms will chose to operate in the unofficial economy if the total expected profit of an underground firm E{π u } exceeds that of a corrupt firm in the official economy E{π o } in (11). w out [ ] = q( π ( ϕ) b ) + ( q) π ( ϕ) c E π 1 (10) u u q ( π ( ϕ) bu ) + ( 1 p) q( 1 d ) π ( ϕ) + ( 1 q) π ( ϕ) ( π ( ϕ) b ) + ( 1 p) ( π ( ϕ) b F ) c p o o c > (11) This inequality holds whenever: 1 π ( ϕ) < pq b + 1 F (12) qd qd ( 1 p) Hence, the model generates the intuitive result in (12) that less productive firms are more likely to enter the underground economy. 4 However, the relation between the likelihood of going underground and institutional quality, λ, will be ambiguous as shown formally in the appendix. For relatively low λ s, the relation may be positive, i.e. increased monitoring due to better institutions drives firms underground and thus into a corrupt environment instead of staying in the official economy where they remain corrupt, but willing to pay higher bribes the faster they are monitored. In other words, moving from very bad to relatively better institutional quality, thereby coming from a likely situation of a relatively large underground economy, an 4 To obtain this result formally, it must be kept in mind that q also depends on π(φ). However, it can be shown by simple differentiation that the likelihood of staying in the official economy is increasing in φ. 9

improvement of institutional quality has ambiguous effects on the level of corruption. Theoretically, allowing for the existence of an underground economy affects the association between institutional quality and corruption due to the perverse incentives arising from firms ability to go underground. In summary, the theoretical model above thus comes to imply three testable propositions to be explored in section 4. 1. Institutional quality is more effective in combating corruption in conjunction with high moral costs of accepting corrupt deals. 2. Corruption is less likely when average moral costs are high regardless of institutional quality. 3. In countries with particularly poor institutional quality, i.e. relatively large unofficial economies, investing in overall institutional quality can in principle lead to more corruption. 3. Data To test these broad implications, I draw data from a number of different sources. First, the corruption index employed in this paper comes from Dreher et al. (2007) and is derived by a MIMIC estimator, a method that instead of providing a direct indicator triangulates the potential corruption of a country, based on a dataset of variables that the literature either associates with corruption as causes or consequences. This set consists of, among other things, legal quality, secondary education rates, cement consumption, GDP per capita, the severity of capital restrictions, financial development, latitude, the age of democracy and a dummy for German legal origins. Dreher et al. (2007) thus follow the literature on robust determinants of corruption summarized in the introduction to estimate countries levels of corruption from objective data. The 10

corruption data are, as all other data in the following, arranged in four periods: the period up to 1980, the early half of the 1980s, the period around 1990 and the late 1990s up to 1997. Second, social trust, my proxy indicator of the M in the model, is measured as is standard in the literature by the percentage of a population answering yes to the question In general, do you think most people can be trusted? Although the question is fairly broad, it has been used since Rosenberg (1956) introduced it and is usually considered a reasonable proxy for actual trust and trustworthiness. Uslaner (2002) shows that the answers measure the extent to which respondents think that strangers can be trusted to abide by a common set of norms of morality. Knack (2001), using data from a walletdrop experiment conducted by Reader s Digest in 16 cities around the world and 12 major American cities, furthermore show that the percentage of a population answering yes to the social trust question above is a strong predictor of the share of the dropped wallets that were returned to their owners with the contents intact. Most authors therefore treat the trust scores as proxies for the average trust and honesty of citizens of a country or region (cf. Knack, 2002; Uslaner, 2002; Bjørnskov, 2007a). 5 These data are from the four waves of the World Values Survey conducted between 1981 and 2001, the 1995 and 2003 LatinoBarometro surveys, the 2001-2004 Asian and East Asian Barometers, and the 2002-2004 Danish Social Capital Project. As most countries are only observed in one period, I use the average of all available observations and thus impose the restriction upon the data that social trust does not vary 5 However, a small literature is sceptical about the use of the aggregate scores of the trust question. See for example Moore (1999), who argues that it is unclear what is standard reputation effects and what is outcomes of social trust, or Beugelsdijk (2006) who claims that trust simple proxies for the standard of formal institutions. 11

significantly in the medium run. While this assumption is clearly not borne out in the case of the United States in which the overall trust level has been declining from the early 1970s to the early 1990s, social trust seems remarkably stable in most other countries (cf. Uslaner, 2002; Bjørnskov, 2007a). 6 The use of MIMIC-based corruption data gauged from objective or quasiobjective primary indices is a point of considerable importance in the present context. 7 First of all, since these estimates on corruption across countries and time are primarily based on objective data, they are probably as free of pure perceptions as such data can be. The bulk of the data used in previous research on the association between social trust and corruption has been perceptional and derived from surveys, making the use of corruption data based on hard facts of great importance in this specific context: when exploring non-economic determinants of corruption, an additional problem arises if underlying, but possibly erroneous, perceptions of the status of a country confound both trust and corruption. In that case, the use of a perceptions-based corruption index, as in most of the empirical corruption literature, might give rise to a spurious association 6 That social trust has been declining in the US is clear in all studies, yet it is worth noting that the decline has not been consistent across the states. While some states, for example California, have seen large declines, trust levels in other states as Massachusetts have been stable (Bjørnskov, 2007b). Likewise, social trust has varied in only few countries besides the US, declining slightly in the United Kingdom and increasing in Denmark and Uruguay. Furthermore, the known trust changes do not seem to have affected changes in corruption as the data for Denmark, the US and the UK exhibit strong declines since the late 1970s while increasing in recent years in Uruguay. 7 In addition, it is worth mentioning that while the Dreher et al. (2007) data are based on more or less objective data, the coefficient of variation in period-to-period changes is the same (approximately 4) as in the perceptions-based ICRG measure. Compared to previous research, the findings in the following are there unlikely to be attributable to artificial or a priori improbable variability of the new measure. 12

between social trust and corruption. By using the data derived by Dreher et al. (2007), this worry is substantially reduced, if not entirely eliminated. Second, using as objective data as are available avoids the pitfalls of the many alternative cross-country indicators of corruption, all based on perceptions, which Knack (2007) outlines. One of Knack s main results is that most broad corruption indices such as Transparency International and the Kaufmann et al. control of corruption index seem to measure administrative corruption instead of state capture. Knack is also seconded by, among others, Abramo s (2007) Brazilian survey results and Olken s (2006) unique study from Indonesia, in assessing that surveys of the experience with corruption among ordinary citizens reveal that corruption perceptions may be rather weak proxies for actual corruption, at least as measured by respondents actual experience, and that perceptions-based indices may underestimate actual corruption. Instead, such perceptions may contain no more information than is obtained by asking about poverty, human rights etc. (Abramo, 2007, 35). Andvig (2004) also criticizes perceptions-based measures on the basis of their risk of being influenced by informational cascades, which for example would make them artificially stable over time. For example, many decision makers and experts contributing to perceptions-based measures are likely to both be aware of alternative measures and to employ the same information, creating potentially spurious correlations between the array of indicators available at any point in time (Knack, 2007). On the other hand, Kaufmann et al. (2006) defend perceptions-based measures (see also Svensson, 2005). First, an inherent problem of the critical studies advocating experience-based indicators is that the actual experience with corruption of most people can also be a misleading indicator. In countries relatively free of corruption, 13

most people may not experience bribery or similar actions. Instead, the less actual corruption there is in a country, the more is it likely that ordinary people stretch their concept of corruption to include nepotism, lobbyism, and unpopular but legal fees, and therefore infer corrupt behaviour from outcomes of professional networking or merely expensive public regulations. Moreover, even in countries with relatively prevalent corruption, the majority of the respondents of international surveys may not have experienced corruption themselves if types of illegal behaviour are focused around specific business sectors or bureaucratic agencies. As such, Kaufmann et al. (2006, 2) note that survey questions about corruption in, for example, the police with which ordinary people are likely to have first or second-hand experience with, need not be informative about corruption in public procurement. Such effects could explain the discrepancy between, for example, the non-negligible World Values Survey measure of experienced corruption and most other indicators in extremely clean countries such as Denmark or Switzerland. Knack (2007) also notes that most broad perceptions-based indicators tend to measure administrative corruption instead of state capture and more specific incidences of corruption. This point is of some importance in the following as the objective indicator employed here, which is highly correlated with the popular broad corruption measures, may suffer from some of the same problems while solving others. 8 As the MIMIC data of Dreher et al. (2007) solves some of the potential problems associated with alternative indicators, I use these data in the following to estimate the relation between trust, institutional quality and corruption using a baseline consisting of 8 The similarities are rather obvious as the correlations between the indicator employed in this paper and alternatives are -.69 with the ICRG measure and -.80 with the Kaufmann et al. control of corruption indicator. These correlations change only marginally when controlling for legal quality and remain significant when controlling for GDP per capita. 14

legal quality, social trust, the logarithm to GDP per capita, openness, government size, political competition (the Herfindahl index), and a set of dummies for postcommunist countries, Latin America, the MENA region, Asia, Sub-Saharan Africa, and the four time periods. Employing this baseline yields an unbalanced panel of 194 observations from 74 countries with full data. All sources for the additional data are given in the appendix. All estimates are obtained using a feasible least squares estimator with random effects as Hausmann tests in the majority of cases strongly support this choice. In one case, I also provide an instrumental variables estimate of the influence of social trust so as to address the possibility that corruption or factors entering the MIMIC procedure generating the corruption data may affect individuals perceptions of the trustworthiness of their fellow citizens. 4. Results At a first glance at the data, there is a quite strong association between social trust and corruption as the simple correlation of -.64 is clearly visible in Figure 2 although the assumption of stability of the trust scores might well serve to bias this correlation. In particular, if trust depends on corruption one would expect that the present trust scores are most strongly correlated with corruption in the latest period. However, the correlations with corruption in the four periods differ only between -.64 and -.68. The partial correlations when controlling for the quality of the legal system similarly exhibit no change over time. Such simple correlations can nonetheless obviously be misleading and Table 2 consequently applies the full baseline specification. The table first of all shows that 15

economic development, political competition and legal quality are negatively associated with corruption while countries with larger government sectors tend to have more corruption. The results also exhibit strong differences across regions of the world with postcommunist countries having most corruption, all other things being equal, followed by the well-known Latin American problems and a weakly significant tendency to more corruption in the Middle East and North Africa region. While the association with legal quality is warranted by most other studies but is here entirely an effect of the MIMIC method used to construct the corruption data, it is worth noting that the other findings do not appear by construction but are still broadly in line with previous studies using data on perceptions of corruption (Aidt, 2003; de Haan and Seldadyo, 2006; Treisman, 2000, 2007). Insert Figure 2 about here Turning to the association between social trust and corruption, the association is strongly significant in both statistical and economic terms. The coefficient indicates that a one standard deviation shock to social trust (14 percentage points) is associated with a corruption decline of 26 percent of a standard deviation (.06 points), all other things being equal. The IV estimate reported in a lower panel in Table 2 provides support for the notion of a causal effect from trust to corruption even though the estimate is somewhat larger. The results in column 2, in which observations with residuals larger than ±2 standard deviations have been removed, in general gives much clearer results 16

consistent with the existing literature, as well as support for a strongly significant estimate of social trust, which is only slightly smaller than that in column 1. 9 In the following columns of Table 2, I explore the question whether social trust complements or substitutes that of legal quality, as captured by simply adding an interaction term. The results rather clearly support the former, as the interaction term is strongly and significantly negative when evaluated at the sample mean. The interaction has a cut-off point at a score of about three on the legal quality index, indicating that neither legal quality nor social trust are associated with corruption at very low levels of legal quality. Hence, this is broadly consistent with the implications of the simple theoretical model in section 2. Another way to evaluate the shape of this relation is to split the sample in two equal halves, one with relatively poor legal quality, and the other with relatively good quality (not shown in table). The result of this exercise is that in the half-sample with poor legal quality, the coefficient on trust is -.174 (standard error.058) while in the top half, it is -.673 (standard error.183). Yet, it is equally worth noting that legal quality is not significant in the bottom half. While this may of course be an artefact of the legal quality index, it is noteworthy that legal quality fails significance even though it forms one of the constituting parts of the MIMIC method used to generate the corruption 9 Only two countries exhibit consistently large residuals across periods in the regressions. The notable outliers are Japan, which appears substantially more corrupt in the raw data than predicted by the model, and Australia that is substantially less corrupt than predicted. The Japanese case is particularly interesting as certain studies questioning the standard trust measure are based on comparisons between the US and Japan (e.g. Miller and Mitamura, 2003). This and other studies therefore may suggest that the problem is not with the trust measure per se, but its implementation in Japan where it may fail to accurately estimate actual trust. 17

scores. Given that legal quality is also associated with the size of the unofficial economy (cf. Schneider and Enste, 2000), this particular finding is consistent with the implications of the theoretical model in which institutional quality has ambiguous effects in countries with relatively large unofficial economies. On a broad level, the results indicate that the effectiveness of the legal system in combating corruption is increasing in the level of social trust. Employing the instrumental variables technique to the two subsamples provides further support, as the IV estimate of social trust is -.943 (standard error.238) in the upper half with relatively good legal systems, whereas it is -.235 (standard error.108) in the comparatively bad half. Even while both estimates are somewhat larger than the random effects coefficients, the ratio between the effects in the two halves is 4.0 and thus not substantially different from the ratio between the corresponding random effects estimates at 3.7. Insert table 2 about here These results could of course be spurious for multiple reasons. 10 A number of different effects could in principle drive the heterogeneous trust effect. First of all, both corruption and social trust are arguably more precisely measured in relatively richer countries, which could give rise to a spurious interaction between trust and legal quality 10 Reviewing the more recent literature on social trust published since 2000, Nannestad (2008) in particular stresses that many studies exhibit two problems. First of all, the literature suffers from a lack of robustness analysis, and second, many well-known and often-quoted studies do only little in terms of causality analysis. Both points are emphasized in Bjørnskov (2007a) that show how few results from the previous literature on the determinants of social trust survive addressing these problems. 18

as the latter is correlated with GDP per capita (r=.47 in the present sample). Second, legal quality could in principle also proxy for democracy (r=.44), regulatory freedom, (r=.28) political competition (r=-.34), government size (r=-.19) or income inequality (r=-.36), and the results could also capture a non-linear effect of social trust due to its correlation with legal quality (r=.39). These possibilities are tested for in Table 3 that includes the additional variables and their interaction with social trust. Insert Table 3 about here The table shows that these worries are likely to be unfounded. While the size of the coefficient on social trust varies by 50 percent from the minimum to the maximum estimate in Table 3, the interaction term only changes about seven percent, indicating that the overall findings are very robust to this exercise. The cut-off point of legal quality at which the estimate of social trust becomes positive is also stable, shifting between index values of 3 to 3.5. As for the further results of the exercise (not shown), the interaction with legal quality are unchanged but two results may deserve mention. First, trust interacted with Gini coefficients (income inequality) is significant (at five percent), suggesting that trust may be more effective in unequal countries. 11 Second, the Hausmann test becomes strongly significant in column 10, indicating that the random effects estimator may not be consistent. However, the fixed effects results pertaining to the interaction term remain significant at conventional levels (not shown). 11 Another possible explanation is that social trust is measured with systematic error in relatively unequal societies. This would nevertheless represent a potentially severe problem for the trust literature, as one of its most robust results is that high levels of inequality are detrimental to social trust. 19

Relying on these simple tests, the findings above therefore seem to be robustly indicating that social trust is significantly associated with corruption as well as making the legal system more efficient in combating corruption. As an alternative way to exemplify these findings, Table 4 provides some very simple dynamic statistics by splitting the set of data with more than one observation on corruption in two, according to their changes in corruption between 1980 and 1997. The result of this simple alternative test is that the good half in terms of success in combating corruption has seen a decrease in corruption of roughly -.075 points (34 percent of a standard deviation) compared to the global average while the bad half has seen a similar increase in the corruption index. Among the top performers are Denmark, Norway, Japan and the Netherlands, all high-trust countries in which only the latter has seen a marked improvement of the legal system in the same period. Among the top are also Belgium, Austria, the United Kingdom and the United States that have higher trust scores than the global average. In the bottom are such countries as Kenya and Nigeria that both score well below the global average. Insert Table 4 about here The table demonstrates that the general improvement of legal quality of 1.3 points does not differ between the two halves. Instead, while most countries in the world exhibited judicial improvements in this period, the simple average dynamics suggest that these improvements have, on average, been accompanied by a real decrease in the instance of corruption in countries with above-average levels of social trust, but generally not in countries with below-average trust. As such, even though such averages 20

should be interpreted with extreme care, they indicate that the static estimates in Tables 2 and 3 are sufficiently strong to be clearly visible in corruption dynamics over a longer period of time. Combined with the panel data estimates above, the findings thus provide clear support for the notion that social trust and institutional quality are complementary when combating corruption, a finding that is discussed and connected to the theory in the final section. 5. Conclusions This paper has revisited the question of whether moral norms of honesty, measured here by social trust, are negatively associated with the corruption problems many countries face. A simple model of potentially corrupt bureaucratic behaviour first illustrated that while moral costs relating to corrupt behaviour are unambiguously associated with less corruption, the association between corruption and the quality of formal institutions is ambiguous, at least at relatively low levels of quality and fragile moral norms. The reason for the theoretical ambiguity is the potential decision of entering the unofficial economy in which stronger institutional quality implies a smaller likelihood of bureaucrats being offered and accepting bribes but also implies more willingness to pay large bribes whenever they are offered. The source of this perverse effect is that as unofficial firms are on average audited earlier within a given period, the value of their potentially remaining time in the market increases, thus raising the opportunity costs of turning down a corrupt offer. While the specific theoretical intricacies cannot be explored with any data available today, the broader implications are tested in a panel dataset consisting of 74 countries observed in up to four periods, which yields 194 observations with full data. 21

The empirical analysis in the paper differs from previous research in using corruption data from a MIMIC exercise, i.e. the estimates on corruption across countries and time are triangulated from objective data. The data used in previous research on the association between social trust and corruption are perceptional indicators deriving from surveys, a difference that makes the use of corruption data based on hard facts of great importance in this specific context. If underlying, erroneous perceptions of the status of a country confound both social trust and perceptions of corruption, the use of a perceptions-based corruption index, as in most of the empirical corruption literature, can easily give rise to a spurious association between social trust and corruption. By using the data derived by Dreher et al. (2007), the paper avoids this pitfall of the existing literature. The results in the paper nonetheless closely follow the overall findings of the corruption literature that social trust is negatively associated with the incidence of corruption, and are consistent with recent work on links between social trust, corruption, and the shadow economy (cf. Ahmed et al., 2005; d Hernoncourt and Méon, 2008). Given that an instrumental variables approach can inform about causality, the estimates suggest that the association reflects an influence of social trust on corruption, although some feedback from the reverse direction cannot be rejected. The results are also of economic significance, as a one standard deviation increase of social trust (28 percentage points) is on average associated with a reduction of corruption of roughly a quarter of a standard deviation the difference between estimated corruption in Greece and Australia. Yet, the results rather clearly also point towards an increasing role of social trust in the course of the development of judicial institutions. At low levels of legal quality, 22

the estimates in Table 2 suggest that the association between social trust and corruption is weak and potentially zero at very low levels, whereas the association is strong at higher levels of legal quality. Splitting the sample nevertheless provides support for a small but significant effect of trust in countries with weak or deficient institutions. Following the implications of the theoretical model, this exercise also provides support for an ambiguous effect on corruption of improving formal institutions in such countries. While it must be stressed that the results are derived under the assumption that trust scores are sufficiently stable over time an assumption borne out in recent empirical studies (Uslaner, 2002; Bjørnskov, 2007a) the use of panel data in this paper therefore suggests that high levels of social trust tend to complement the development of formal institutions. The results consequently do not support the view that social trust can, for example, supplement the lack of formal protection of property rights as hinted at in some studies (e.g. Fukuyama, 1995), but instead follow the implications of the theoretical model in supporting complementarity. The effects of legal quality and trust thus come to differ across levels of economic and institutional development. On a broader level, the findings in this paper may therefore also hold implications for the literature on social trust. For example, this type of evidence seems to refute Beugelsdijk s (2006) claim that the answers to the social trust question simply measure the quality of formal institutions. It also refutes the notion found in some studies that social trust is a panacea for solving a number of very different social and economic problems. Taking the findings in this paper seriously leaves the question open whether some of the beneficial effects of having a trusting and trustworthy population only emerge under specific conditions. 23

References Abramo, Claudio Weber. 2007. How much do perceptions of corruption really tell us? Working paper 2007-19, Kiel Institute for the World Economy. Ahmed, Ehsan, J. Barkley Rosser, Jr. and Marina V. Rosser. 2005. A global perspective on the non-observed economy, inequality, corruption, and social capital. Mimeo, James Madison University. Aidt, Toke. 2003. Economic analysis of corruption: A survey. The Economic Journal, vol. 113, F632-652. Andvig, Jens Christian. 2004. The challenge of poor governance and corruption. Working paper 665-2004, Norwegian Institute of Foreign Affairs. Alt, James E. and David Dreyer Lassen. 2003. The Political Economy of Corruption in the American States. Journal of Theoretical Politics, vol. 15, 341-465. Beugelsdijk, Sjoerd. 2006. A note to the theory and measurement of trust in explaining differences in economic growth. Cambridge Journal of Economics, vol. 30. 371-387. Bjørnskov, Christian. 2003. Corruption and social capital. Working paper 03-13, Aarhus School of Business. Bjørnskov, Christian. 2007a. Determinants of Generalized Trust. A Cross-Country Comparison. Public Choice, vol. 130, 1-21. Bjørnskov, Christian. 2007b. On the Determinants of Social Trust in the United States. Paper presented at the Midwest Political Science Association conference, Chicago, April. Cembureau. 1998.World Statistical Review No. 18:World Cement Market in Figures, 1913/1995. The European Cement Association, Brussels, Belgium. 24

Cembureau. 1999. World Statistical Review Nos. 19 and 20. The European Cement Association, Brussels, Belgium. De Haan, Jacob and Harry Seldadyo. 2006. The Determinants of Corruption: A Literature Survey and New Evidence. Paper presented at the 2006 Meeting of the EPCS, Turku. D Hernoncourt, Johanna and Pierre-Guillaume Méon. 2008. The not so Dark Side of Trust: Does Trust increase Participation in the Shadow Economy? Paper presented at the 2008 Meeting of the EPCS, March 27-30, Jena. Dreher, Axel, Christos Kotsogiannis and Steve McCorriston. 2007. Corruption around the world: Evidence from a structural model. Journal of Comparative Economics, vol. 35, 443-466. Gwartney, J., Lawson, R. (2006). Economic Freedom of the World 2006. Downloadable from http://www.freetheworld.com/download.html Henisz, W. J. 2002. The institutional environment for infrastructure investment. Industrial and Corporate Change, vol. 11, 355-389. Kaufmann, Daniel, Aart Kray and Massimo Mastruzzi. 2005. Governance Matters IV: Governance Indicators fo 1996-2004. World Bank Policy Research Working Paper no. 3630. Kaufmann, Daniel, Aart Kray and Massimo Mastruzzi. 2006. Measuring Corruption: Myths and Reality. Mimeo, the World Bank. Kingston, Christopher. 2007. Parochial corruption. Journal of Economic Behavior & Organization, vol. 63, 73-87. 25

Knack, Stephen. 2001. Trust, associational life and economic performance. In John F. Helliwell (Ed.), The contribution of human and social capital to sustained economic growth and well-being. Quebec: Human Resources Development Canada. Knack, Stephen. 2002. Social capital and the quality of government: evidence from the US states. American Journal of Political Science, vol. 46, 772-785. Knack, Stephen. 2007. Measuring Corruption: A Critique of Indicators in Eastern Europe and Central Asia. Journal of Public Policy, vol. 27, 255-191. Marshall, Monty and Keith Jaggers. 2004. Polity IV dataset. University of Maryland, Center for International Development and Conflict Management. Mauro, Paolo. 1995. Corruption and Growth. Quarterly Journal of Economics, vol. 110 (3), 167-195. Mauro, Paolo. 1998. Corruption and the composition of government expenditure. Journal of Public Economics, vol. 69, 263-279. Méon, Pierre-Guillaume and Khalid Sekkat. 2005. Does corruption grease or sand the wheels of growth? Public Choice, vol. 122, 69-97. Méon, Pierre-Guillaume and Laurent Weill. 2004. Does better governance foster efficiency? An aggregate frontier analysis. Economics of Governance, vol. 6, 75-90. Miller, Alan S. and Tomoko Mitamura. 2003. Are Surveys on Trust Trustworthy? Social Psychology Quarterly, vol. 66, 62-70. Moore, Mick. 1999. Truth, Trust, and Market Transactions. Journal of Development Studies, vol. 36, 74-88. Nannestad, Peter. 2008. New work on trust: What have we learnt, if anything? Annual Review of Political Science, vol. 11, 413-436. 26

Rosenberg, Morris. 1956. Misanthropy and Political Ideology. American Sociological Review, vol. 21, 690-695. Schneider, Friedrich and Dominik Enste. 2000. Shadow economies: Sizes, causes, and consequences. Journal of Economic Literature, vol. 38, 77-114. Seldadyo, Harry, Emmanuel Pandu Nugroho and Jacob, De Haan. 2007. Governance and Growth Revisited. Kyklos, vol. 60, 279-290. Svensson, Jacob. 2005. Eight Questions about Corruption. Journal of Economic Perspectives, vol. 19, 19-42. Treisman, Daniel. 2000. The causes of corruption: A cross-national study. Journal of Public Economics, vol. 76, 399-457. Treisman, Daniel. 2007. What have we learned about the causes of corruption from ten years of cross-national empirical research? Annual Review of Political Science, vol. 10, 211-244. UNU. 2007. World Income Inequality Database v2.0. Helsinki: United Nations University-World Institute for Development Economics Research. Uslaner, Eric M. 2002. The Moral Foundations of Trust. Cambridge: Cambridge University Press. Uslaner, Eric M. 2004. Trust and Corruption. Pp. 76-62 in Johann Graf Lambsdorff, Markus Taube and Matthias Schramm (eds.), The New Institutional Economics of Corruption. London: Routledge. Uslaner, Eric M. and Bo Rothstein. 2005. All for One: Equality, Corruption, and Social Trust. World Politics, vol. 58, 41-72. World Bank. 2004. World development indicators. CD-ROM and on-line database. Washington DC: The World Bank. 27

Yavas, Cemile. 2007. The Ghost of Corruption. The B.E. Journal of Economic Analysis & Policy, vol. 7 (1), article 38. Appendix To show that the relation between the likelihood of going underground and instutional quality, λ, is ambiguous, take as a starting point that firms go underground when (11) holds, that is when: 1 pq 1 qπ ( ϕ) b F < 0 (A1) d d ( 1 p) Differentiating the left hand side with respect to λ yields equation A2 where f( ) is the density function associated with q: ( ϕ) f ( )[ μ( 1 λ) + a μλ] π ( ϕ) 1 pq f () [ μ( 1 λ) + a μλ] π ( ϕ) ( 1 p) d π 1 p w wout + M 1 λ ( 1 p) d ( ) 2 (A2) Here, the last term is unambiguously negative, but the sign of the first two terms is ambiguous, and can be easily shown to depend on the sign of μ + a 2μλ. For low values of λ relative to the other parameters, this implies that the sign becomes positive and therefore that the two first terms outweigh the influence of the last term. Insert Table A1 about here 28

Table 1. Descriptive statistics Average Standard deviation Observations Corruption.047.220 334 Log GDP per capita 8.635 1.062 333 Openness.718.486 334 Political competition.248.274 321 Government size.206.085 334 Legal quality 5.310 1.916 289 Postcommunist.033.179 334 Latin American.254.436 334 Asian.132.339 334 MENA.108.311 334 SSA.213.409 334 Social trust.276.139 304 Democracy 5.280 4.184 326 Checks (Keefer) 2.871 1.683 332 Political constraints III.277.207 329 Monarchy.186.389 334 Regulatory freedom 5.290 1.012 300 Log cement consumption 7.437 1.867 334 29

Tabel 2. Results RE RE, no outliers RE RE, no outliers 1 2 3 4 Log GDP per -.163*** -.135*** -.153*** -.129*** capita (.025) (.015) (.023) (.012) Openness.015 -.020.001 -.036** (.029) (.018) (003) (.015) Political -.149*** -.082** -.120** -.104*** competition (.052) (.035) (.047) (.027) Government.285*.248***.263*.112 size (.152) (.096) (.141) (.075) Legal quality -.009 -.011*** -.009* -.014*** (.005) (.004) (.005) (.003) Postcommunist.231***.199***.212***.209*** (.059) (.034) (.056) (.027) Latin.119** 109***.120**.110*** American (.055) (.031) (.052) (.025) Asian.038.062*.016.054** (.057) (.033) (.054) (.027) MENA.141*.127***.120*.121*** (.072) (.041) (.068) (.032) SSA.086.097**.079.087** (.077) (.044) (.072) (.037) Social trust -.416*** -.352*** -.346*** -.339*** (.123) (.068) (.117) (.051) Legal * trust -.168*** -.174*** (.028) (.017) RE, no large corr -.162*** (.027).026 (.029) -.140** (.057).070 (.165) -.005 (.006).199*** (.066).108* (.057) -.016 (.058).109 (.074).061 (.084) -.389*** (.125) -.164*** (.031) RE, no small corr -.115*** (.014) -.033* (.017) -.082*** (.032).194** (.088) -.009*** (.003).197*** (.033).126*** (.030).055* (.032).133*** (.039).106** (.042) -.232*** (.069) -.113*** (.024) Observations 194 174 194 154 171 172 Between R.821.921.841.958.823.911 square Within R.223.349.374.597.421.446 square Wald Chi 326.45 810.12 406.94 1463.78 329.46 706.3 square Breusch-Pagan 36.37*** 9.31*** 48.47*** 2.90* 49.55*** 13.38*** Hausmann test 4.12 5.28 3.84 6.63 9.03 7.43 IV estimate: Social trust -.598*** (.164) Hansen J- 1.958 statistic First stage F- statistic 32.05 Note: all regressions include period dummies and a constant; *** (**) [*] denotes significance at p<.01 (p<.05) [p<.10]; standard errors in parentheses. Instruments in the IV regression are monarchy, checks, political constraints III, and an interaction term between the latter variables. Outliers in columns 2 and 4 are defined as observations with residuals larger than 2± standard deviations. Column 5 excludes observations with corruption above.253, column 6 excludes observations with corruption below -.363. 30