Corruption and Social Capital Formation

Transcription

1 Corruption and Social Capital Formation Tobias Cagala, Ulrich Glogowsky, Veronika Grimm, Johannes Rincke, Amanda Tuset Cueva May 20, 2016 Abstract We present controlled experimental evidence on how bad governance affects social capital. In our design, subjects make contributions to associations which provide real social capital to the population from which subjects are drawn. We compare contributions to social capital formation between a corruption condition, where an administrator can expropriate part of contributions, and a control condition without corruption. Compared to the control condition, subjects matched to an expropriating administrator contribute much less to social capital formation. This effect works through a specific channel: corruption diminishes the capacity of interpersonal trust to raise individuals contributions to social capital. Our results hold despite the fact that we keep the efficiency of social capital formation constant between conditions. We conclude that the anti-social aspect of bad governance makes individuals less inclined to behave pro-socially. JEL codes: D02; D03; H41 Keywords: Corruption; governance; social capital; trust 1 Introduction Economists tend to agree that bad governance is economically harmful. However, there is no comprehensive assessment of how shortcomings in governance such as Department of Economics, University of Erlangen-Nuremberg. Cagala: tobias.cagala@fau.de; Glogowsky: ulrich.glogowsky@fau.de; Grimm: veronika.grimm@fau.de; Rincke: johannes.rincke@fau.de; Tuset Cueva: amanda.tuset.cueva@fau.de. Financial support by the Emerging Field Initiative of the University of Erlangen-Nuremberg and the Hans-Frisch-Stiftung are gratefully acknowledged.

2 bureaucratic inefficiency, violations of the rule of law, and corruption affect society. Among the repercussions of bad governance that are well documented are disincentives to invest (Mauro, 1995; Campos et al., 1999; Wei, 2000), interference with public service delivery and its equity implications (Reinikka and Svensson, 2004, 2005; Olken, 2006, 2007), and negative consequences for economic performance (Acemoglu et al., 2001, 2014). Other effects of bad governance, including its impact on prosociality and cooperation, have repeatedly been discussed, but evidence on the causal impact of bad governance is scarce. This paper focuses on corruption as an important and very common form of bad governance. Specifically, we provide causal evidence on how corruption affects contributions to the social infrastructure individuals rely on when interacting with each other. Important elements of this infrastructure are clubs and associations that facilitate the exchange of information and the coordination of public goods provision. Borrowing the umbrella term social capital from Putnam (1993) as a convenient label, we interpret individuals monetary contributions to the social infrastructure as contributions to social capital. 1 The available evidence suggests that social capital produces tangible economic returns (Feigenberg et al., 2013) and is positively associated with long-run economic performance (Knack and Keefer, 1997; Zak and Knack, 2001; Guiso et al., 2008). Our work adds to a better understanding of how corruption affects social capital, and thus contributes to a more comprehensive assessment of the social cost of bad governance. Since Putnam (1993), an influential line of reasoning in political economy maintains that bad governance leads to low social capital. 2 This holds despite the fact that, at least for the case of corruption, there is no clear theoretical prediction on whether 1 An extensive literature in ethics, philosophy, and the social sciences elaborates on the role of social capital (Tocqueville and Bender, 1981; Bourdieu, 1977; Ostrom, 1990; Putnam, 1993, 1995; Dasgupta and Serageldin, 1999). Despite the use of different definitions, there is a broad consensus that social capital captures the conditions that allow individuals to reap the benefits from interacting cooperatively in a civic community. It is important to distinguish contributions to social capital in a society as a form of public goods provision from investments in individual social capital. In the terminology of Glaeser et al. (2002), individual social capital is the social component of human capital. Clubs and associations as platforms for social interaction can thus be understood as means by which a society can benefit from the positive cross-person externalities generated by the different types of individual social capital. 2 Putnam (1993) discusses the link between governance and social capital (and vice versa) for the south of Italy. More recent examples include Guiso et al. (2008) and Nannicini et al. (2013). 2

3 the effect on social capital should be positive or negative. If clubs and associations formed by citizens offer services that are substitutes for state-provided public goods, corruption could raise individuals contributions to social capital. This is because corruption tends to decrease the efficiency of public goods provision by the state, making private provision relatively more attractive. In contrast, corruption could also reduce individuals contributions to social capital. We point out three complementary mechanisms that could drive such an effect. The first mechanism relates to the notion of self-serving beliefs (Di Tella and Pérez-Truglia, 2015). It captures the idea that individuals form beliefs that legitimize their own selfish actions. In the context of social capital formation, individuals could respond to corruption by forming a belief that the society lacks pro-social attitudes. Such beliefs would help individuals to justify the decision not to contribute to social capital. Second, a related literature on investment decisions under risk (Bohnet and Zeckhauser, 2004; Bohnet et al., 2008; Cubitt et al., 2015) has shown that individual decisions are affected by betrayal aversion. In the context of corruption, contributions could be negatively affected if subjects interpret the actions of a corrupt bureaucracy as a betrayal. Third, if corrupt bureaucrats expropriate contributions designated for the social infrastructure, contributors could interpret corruption as unkind behavior. If they are motivated by reciprocity concerns (Rabin, 1993; Dufwenberg and Kirchsteiger, 2004), contributors might respond to this unkindness by lowering their contributions (and hence bureaucrats payoffs from expropriation). However, it is challenging to isolate how governance affects pro-social behavior from existing data. 3 We respond to this difficulty by implementing a controlled experiment. This relates our paper to an emerging experimental literature studying the effects of corruption on trust and investment behavior. For instance, using an artefactual field experiment, Beekman et al. (2014) examine how the presence of a corrupt leader affects private investment. Banerjee (2016a) uses a laboratory experiment to 3 The literature in political science and sociology has extensively discussed the link between governance and social capital (Fukuyama, 1995; Levi, 1998; Levi and Stoker, 2000), but attempts to identify causal effects have been limited. Using survey data, Brehm and Rahn (1997) and Uslaner (2003) provide suggestive evidence that perceptions of the quality of governance are in fact positively correlated with social capital. Also related to our study is OECD (2013), documenting a negative correlation between corruption and subjective tax morale in survey data. 3

4 study how bribery erodes trust. 4 Using a dynamic framework, Cagala et al. (2015) study how contributors in a public goods game interact with more or less trustworthy outside actors. Whereas the aforementioned studies rely on some representation of social capital in the laboratory (like behavior in the public goods game or the trust game), our design involves individual contributions to real social capital (i.e., outside the laboratory). We achieve this by providing subjects with an endowment which they can either keep for private consumption, or contribute to non-profit associations which form part of the local real-world social infrastructure. Using this setting, we identify the impact of corruption on social capital by comparing contributions in a corruption condition, where an administrator expropriates part of contributions, to a control condition without active expropriation. 5 Our main findings are as follows. First, we demonstrate that subjects matched to an expropriating administrator contribute much less to social capital than subjects in the control condition. The effect is substantial: the exposure to a corrupt administrator causes a 31 percent reduction in contributions. Our design makes sure that a one unit increase in contributions has the same impact on social capital in both treatments. This implies that differences in the efficiency of social capital formation cannot explain observed responses of contributors. We conclude that the anti-social aspect of bad governance is what affects social capital formation. Second, we use a simple surveybased measure of subjects baseline level of interpersonal trust to shed light on the heterogeneity in the response to corruption between trusting and non-trusting types. We find that non-trusting types make lower contributions to social capital than trusting types and do not react to the quality of governance. 6 In contrast, trusting types reduce 4 Banerjee (2016a) identifies the effect of bribery in a two-stage game. Subjects first play an ultimatum game, either with a bribery frame or with a neutral frame. The second stage consists of a trust game. 5 Our approach captures short-run effects of behavioral responses to corruption. There is a growing literature providing complementary evidence on long-run effects of governance. See, for instance, Greif (1994), Alesina and Fuchs-Schündeln (2007), and Becker et al. (2015). The focus on the consequences of corruption differentiates our work from studies dealing with agents motivation to engage in or punish corruption (Abbink et al., 2002; Barr and Serra, 2009; Cameron et al., 2009; Abbink and Serra, 2012; Banerjee, 2016b). 6 The positive association between the baseline level of trust and contributions to local associations links our approach to studies using measures of trust to proxy for social capital (Banerjee, 2016a). The fact that we use monetary contributions to measure social capital formation allows us to interpret baseline trust as a factor that shapes the impact of corruption on social capital. 4

5 their contribution to social capital to the level of non-trusting types in the presence of expropriating administrators. This finding suggests that the overall effect of corruption on social capital works through a specific channel: corruption prevents interpersonal trust to translate into higher contributions to social capital. Can we generalize the insights from our experiment to real-world settings? Using data from the World Values Survey, we provide suggestive evidence that contributions to social capital formation are in fact negatively associated with individuals exposure to corruption. The survey evidence is in line with a positive correlation between social capital and indicators of good governance in cross-country data (Knack and Keefer, 1997). While more research is needed to test the external validity of our experimental findings, a tentative conclusion is that the negative link between social capital and corruption observed in real-world data likely reflects a causal relationship. Bad governance thus hinders individuals to reap the benefits from the cooperative social interactions that social capital facilitates. 7 The remainder of the paper is organized as follows. Section 2 explains our research design, Section 3 describes our main findings, Section 4 discusses robustness and external validity, and Section 5 concludes. 2 Experimental Design Measuring the Formation of Real Social Capital We investigate the effect of corruption on social capital formation in a laboratory experiment. Importantly, the experiment involves the formation of social capital outside the laboratory. Specifically, we measure subjects contributions to social capital as the monetary support for associations which use the funds to provide real social capital to the population from which subjects are drawn. We achieve this as follows: we conduct a laboratory experiment with students enrolled at the department of economics and business administration at the University Erlangen-Nuremberg. In the experiment, we provide subjects with 7 Available estimates of the social cost of corruption consider a number of consequences of administrative inefficiency, but do not take into account the effect on social capital. For examples, see Kaufmann (2005) and Olken and Pande (2012). 5

6 an endowment which they can either keep for themselves or contribute to one of the various nonprofit associations operating at the department. The associations facilitate social interaction at the department and are a central part of what sociologists would call the local civic community (Putnam, 1993, 1995). They offer a broad range of services, including placement into international exchange programs, students workshops, tutoring services, and counseling for students. The associations also participate in the organization of various social events at the department throughout the academic year. 8 By fostering social interaction between students, teachers, and other staff, the associations increase the productivity of both individual student effort and inputs supplied by the university (e.g., teaching and tutoring services). In short, the associations benefit the entire department community and play an important role in the process of social capital formation at the department. Treatment Conditions The experiment consists of a simple one-shot game. To identify the effect of corruption on contributions to social capital, we implement two treatment conditions: a corruption condition and a control condition. In both conditions, we randomly determine subjects player type, either contributor or administrator. In addition, we randomly allocate subjects into groups each comprising three contributors and one administrator. Both types of players receive a fixed endowment of 100 experimental currency units (ECU), which equals 10 Euro. Furthermore, both conditions consist of two stages. We first describe the corruption condition. In the first stage of this condition, each contributor selects one out of five associations to which she can contribute an amount between 0 and 100 ECU in the second stage. The instructions list the associations by name and briefly inform subjects about the associations main activities. For instance, the international student association AIESEC is introduced as AIESEC: Placement for international exchanges and internships and corresponding counseling services; intercultural tutoring. 9 While contributors select an association, each administrator makes 8 The following paragraph and the instructions in the appendix provide more detailed information on the associations subjects could contribute to. The associations are very active in advertising their services to the department community through announcements in courses, posters in the department building, and social media. 9 The other four associations on the list include two different students unions (activities: tutoring 6

7 a binary choice of whether or not to expropriate a fixed share of 10 percent of secondstage contributions in her group. Administrators who decide for expropriation know that this increases their payoff by 0 to 30 ECU, depending on second-stage contributions. We do not inform administrators about which associations contributors in their group select. Hence, administrators cannot condition their expropriation decision on which associations benefit from contributions. In the second stage, contributors learn about their administrator s decision (expropriation, yes or no) before they decide how much to contribute to the previously chosen association. Note that the game does not involve any interaction between contributors. We now turn to the control condition. The only difference to the corruption condition is that the first stage does not involve any active decision-making by administrators. Instead of deciding for or against expropriation, administrators in the control condition are informed about the outcome of a random draw that determines whether or not they receive 10 percent of second-stage contributions in their group. In the experiment, we use the label additional compensation for this part of administrators payoff. We choose the probability for the random draw to be equal to the empirical propensity of administrators choosing expropriation in the corruption condition. Importantly, we neither communicate the exact probability nor mention that the probability depends on the behavior of administrators in another condition. 10 Figure A1 in the appendix summarizes the timing of the experiment across both conditions. Further Details of the Design We need to further discuss three features of our design. First, the fact that we group together three contributors with one administrator does not change monetary incentives to contribute to social capital. We nevertheless assign individuals into groups of four players to ensure that, despite the moderate expropriation rate, our design implements substantial monetary incentives for administrators to opt for expropriation. Second, administrators and contributors make their services, organization of social events at the department, and representation of students in various department committees). The remaining associations offer a broad range of services to students, including counseling services and workshops of various types. See the instructions in the appendix for details. 10 The random draw in the control condition ensures that prospective outcomes (in terms of the efficiency of social capital formation) are equal between conditions. 7

8 decisions knowing that the experimenter doubles transfers to associations. 11 This ensures that contributions to the same associations would be less efficient in terms of social capital formation if contributors made them outside the laboratory. Third, administrators perform a simple administrative task after contributors make their contribution decisions. We thereby ensure that the administrators role in the experiment is consistent with the function implied by the administrator label. Specifically, administrators have to assign contributions to associations according to contributors choices. 12 Identification of the Effect of Corruption Our design generates observations from groups where the administrator benefits from contributions (through the expropriation decision or the random draw), and from groups where this is not the case. If not stated otherwise, we derive our results exclusively from studying contributors who interact with administrators who do benefit from contributions. Contributors from the remaining groups experience a situation where administrators do not interfere with social capital formation, and are thus of no particular use for our study. Hence, we identify the effect of corruption on social capital formation from comparing contributors in the corruption condition who were matched to an expropriating administrator to contributors in the control condition who were matched to an administrator receiving the additional compensation. Importantly, because the share of contributions accruing to administrators is equal in both conditions, this comparison involves contributors who face the same efficiency of social capital formation when deciding about their contribution. The only difference between both conditions is that administrators in the corruption condition make an active decision for expropriation, whereas administrators in the control condition benefit passively from contributions. Consequently, our design identifies how the anti-social aspect of the administrator s decision affects social capital formation and separates 11 In groups where administrators benefit from contributions (either actively through expropriation or passively through the additional compensation), the doubling applies to contributions net of the 10 percent share accruing to administrators. 12 The least popular association was selected by 9.4 percent of contributors while the most popular one was selected by 37.2 percent. The administrative task was implemented such that misallocations by administrators were ruled out. Contributors and administrators were informed about this. 8

9 this effect from the mere inefficiency associated with corruption. 13 Technical Details We conducted the experiment in the Laboratory for Experimental Research Nuremberg (LERN). In total, 384 subjects participated in the experiment, 96 as administrators and 288 as contributors. Figure A2 in the appendix shows the number of contributors conditional on administrators decisions and the random draw. As mentioned before, we do not consider the sample of 60 contributors from groups where administrators did not benefit from contributions in our main analysis. The experiment took one hour, with an average payoff of ECU. The average earning, including the show-up fee of 4 Euro, was Euro. We programmed the experiment with z-tree (Fischbacher, 2007) and recruited subjects with ORSEE (Greiner, 2015). Subjects were recruited from the subject pool at LERN without imposing any restrictions. Subjects solved control questions before the experiment and answered survey questions on individual characteristics and game-related issues after the experiment. We also informed subjects that we would send an reporting the total amounts transferred to the associations. 3 Corruption Impairs Social Capital Formation Out of the 48 administrators in the corruption condition, 38 decided in favor of expropriation. This provides us with observations on 114 contributors from the corruption condition for our main analysis. We implemented the same distribution in the control condition and thus obtained another 114 contributors from 38 groups with a passive administrator receiving the additional compensation. This gives us a final sample comprising 228 observations from contributors matched to an administrator who benefited from contributions, either actively through expropriation or passively through the additional compensation. Our main result is that corruption impairs social capital formation. Figure 1 compares mean contributions between the corruption and the control conditions and shows 13 This sets our work apart from field studies identifying donor aversion against charities overhead cost (Gneezy et al., 2014). Because administrators received a fixed compensation of the same size as a contributor s individual endowment contributors could not reasonably interpret the expropriation of contributions as supportive to the fairness of the payoff allocation. 9

10 a pronounced negative impact of corruption on contributions to social capital. In the control condition, subjects contributed 16.9 percent of their endowment. In contrast, subjects contributed only 11.7 percent of their endowment in the corruption condition. This 30.9 percent reduction in mean contributions to social capital is statistically significant (P = 0.039, Mann-Whitney U test). 14 Figure 1 identifies how the active decision of administrators in favor of expropriation affects contributions to social capital. Given that the marginal impact of contributions on social capital is identical between conditions, the evidence suggests that the anti-social aspect of corruption is what drives the effect. This interpretation requires that in our experiment, contributors perceive the administrator s decision in favor of expropriation as reflecting her (lack of) pro-sociality. This assumption is supported in the data coming from the survey on game-related issues percent of contributors matched to an expropriating administrator stated they were dissatisfied with the administrator s behavior. Among contributors matched to a non-expropriating administrator, only 13.3 percent made a corresponding statement. This difference is statistically significant (P < 0.001, Mann-Whitney U test). 15 Figure 2 provides a more detailed perspective on our main finding. It displays histograms of contributions (bin size 10 ECU) and cumulative distribution functions of contributions for both conditions. Compared to the control condition, the figure reveals a higher likelihood of contributions up to 20 percent of endowments in the corruption condition. Contributions in the range between 40 percent and 70 percent of endowments are more likely in the control condition. Overall, we note that introducing corruption results in a substantial shift of probability mass towards lower contributions. RESULT 1: Contributions to social capital are significantly lower in the presence of expropriating administrators. The anti-social aspect of corruption is what makes individ- 14 See Panel A in Table A1 in the appendix for descriptive statistics. 15 Because control-group administrators did not make any active decision, the comparison contrasts survey responses of contributors subject to active expropriation to responses of contributors in groups with active administrator decisions against expropriation. Subjects could assess their satisfaction with the administrator s behavior on a scale between 1 (very dissatisfied) and 10 (very satisfied). The reported percentages refer to the share of subjects choosing values lower or equal to 5. 10

11 uals less inclined to behave pro-socially. Differences in average contributions are driven by a shift towards small and zero contributions in the corruption condition. From its beginnings, the literature on social capital has considered interpersonal trust a prerequisite of social capital formation (Putnam, 1993). The rationale is that only if individuals trust in others, they are willing to engage in cooperative social interactions that are at the core of the civic community (Thöni et al., 2012). To shed light on the relation between interpersonal trust and the impact of corruption on social capital, we use subjects responses to a survey question on generalized trust elicited after the experiment. 16 Specifically, we split the sample of contributors into trusting types and non-trusting types and study the heterogeneity in the response to corruption between subjects with high and low baseline levels of interpersonal trust. 17 The subsample analysis assumes that the survey-based measures of baseline levels of trust are predetermined and unaffected by the treatment itself. Our data supports this assumption: the share of trusting types is 59.7 percent in the control condition, compared to 54.4 percent in the corruption condition. This difference in the share of trusting types is not statistically significant (P = 0.42, Mann-Whitney U test). Figure 3 demonstrates that differences in baseline levels of interpersonal trust determine how subjects respond to corruption. 18 Three observations emerge. First, average contributions in the control condition are significantly higher for trusting than for non-trusting types (P < 0.01, Mann-Whitney U test). This finding illustrates that interpersonal trust is a prerequisite of social capital formation. Second, the left panel of Figure 3 shows that trusting types reduce their contributions to social capital to the level of non-trusting types in the presence of expropriating administrators. The difference between the average contribution of trusting types in the control condition (20.8 percent of endowment) and the corruption condition (10.5 percent) is statistically significant (P < 0.001, Mann-Whitney U test). Third, the difference between average contributions of non-trusting types in the control condition (11.2 percent of 16 The survey asked subjects to respond to the statement Generally speaking, people can be trusted, with response categories completely agree, agree, disagree, and completely disagree. 17 We categorize subjects as trusting types if they choose one of the first two categories, and as nontrusting types if they choose one of the last two categories. This classifies 130 subjects as trusting, and 98 as non-trusting. 18 See Panel B in Table A1 in the appendix for descriptive statistics. 11

12 endowment) and contributions of non-trusting types in the corruption condition (13.1 percent) is not significant (P = 0.32, Mann-Whitney U test). The pronounced heterogeneity in the treatment response for trusting types suggests that the overall effect of corruption on social capital formation works through a specific channel: corruption diminishes the capacity of interpersonal trust to raise individuals contributions to social capital. The evidence implies that corruption triggers a decline in individuals pro-sociality and allows us to attribute this decline to trusting types who behave rather pro-socially in an environment without corruption. RESULT 2: The effect of corruption on social capital formation works through a specific channel: corruption diminishes the capacity of interpersonal trust to raise individuals contributions to social capital. 4 Robustness and Discussion We performed a number of additional checks to see if our results are robust. Estimating the treatment effect by means of a linear regression and controlling for individual characteristics leaves all our results unchanged (see Table A2 in the appendix for details). The estimates suggest that corruption reduces contributions to social capital by 29.2 percent relative to the average contribution in the control condition. The effect is statistically significant and very close to the estimate of a 30.9 percent reduction based on a simple comparison of unconditional means. Table A2 also shows that our results are only marginally affected if we employ a Tobit specification, thereby accounting for the fact that contributions are censored at zero. All those robustness checks support the internal validity of our results. The next issue is whether the insights from our experiment generalize to realworld settings. This is, of course, a difficult question, and providing a conclusive answer goes beyond the scope of this paper. A straightforward but preliminary test of external validity based on a large population of survey respondents is displayed in Figure 4. The figure uses data from the World Values Survey (WVS) and shows the percentage of respondents who contribute to social capital formation in terms of 12

13 volunteering. 19 To capture variation in perceived corruption, we use a question that asks subjects to indicate, on a scale from one (low corruption) to ten (high corruption), how widespread they think that corruption is within their country s government. We distinguish between subjects with low perceived corruption (values lower or equal to five) and subjects with high perceived corruption (values larger than five). 20 Figure 4 shows that among individuals with low perceived corruption, 37.5 percent contribute to social capital formation by volunteering. Among individuals with high perceived corruption, the share of contributors is only 26.8 percent. The difference is statistically significant (P < , n = 7716, Mann-Whitney U test) and implies that subjects with high perceived corruption are 28.5 percent less likely to volunteer than individuals with low perceived corruption. 21 The correlation in real-world data on social capital formation and exposure to corruption thus mirrors the causal effect identified in the experiment. We therefore interpret Figure 4 as supporting the external validity of our experimental findings. This conclusion is also in line with Knack and Keefer (1997), who use cross-country data and find that measures for social capital positively correlate with indicators of good governance. A tentative conclusion is that the negative link between social capital and corruption observed in real-world data likely reflects a causal relationship. 5 Conclusion This paper demonstrates that corruption reduces individual contributions to social capital. The analysis reveals that the anti-social aspect of bad governance is what affects social capital formation. Our findings have important implications for assessing the social costs of bad governance. Several studies have demonstrated that a lack of social capital is associated 19 The survey question reads: Approximately how many total hours a month were you active in voluntary organizations?. The response categories are None, 1-2 hours, 3-5 hours, 5-10 hours, and More than 10 hours a month. Figure 4 displays the percentage of respondents with a strictly positive number of hours. 20 The set of countries were both the social capital and the corruption questions were asked comprises Algeria, Bahrain, Iraq, Kuwait, Lebanon, Tunisia, and Yemen (n = 7716). 21 The correlation from Figure 4 remains highly significant if we employ a regression framework controlling for country fixed effects and individual characteristics such as gender, age, educational attainment, and income. 13

14 with poor economic performance (Knack and Keefer, 1997; Zak and Knack, 2001; Guiso et al., 2008). Guiso et al. (2008), for instance, consider the long-run impact of social capital in Italy and estimate that a one standard deviation decrease in social capital decreases per capita income by 21 percent. Combining this evidence with the insights from our experiment, it is likely that corruption has a substantial negative impact on economic performance working through social capital formation. We conclude that corruption might be much more costly than previously thought. In fact, poor institutions could induce a vicious circle with corruption leading to low levels of social capital, and a lack of social capital preventing improvements in institutional performance (Putnam, 1993; Nannicini et al., 2013). We leave for future research the question how the legitimacy of administrators affects individuals responses to corruption. Based on related research on sanctioning mechanisms (Baldassarri and Grossman, 2011), we conjecture that corruption by elected officials with a high legitimacy has stronger effects on social capital formation than corruption by administrators or bureaucrats with relatively low legitimacy. It also seems worthwhile to investigate intercultural differences in how social capital responds to corruption, thereby linking the discussion to the literature on how groups manage to maintain cooperation in the presence of free-riding incentives (Henrich et al., 2006; Herrmann et al., 2008). References ABBINK, K., IRLENBUSCH, B. and RENNER, E. (2002). An experimental bribery game. Journal of Law, Economics, and Organization, 18 (2), and SERRA, D. (2012). Anticorruption policies: Lessons from the lab. In D. Serra and L. Wantchekon (eds.), New Advances in Experimental Research on Corruption, Research in Experimental Economics, vol. 15, Emerald, Bingley, pp ACEMOGLU, D., JOHNSON, S. and ROBINSON, J. (2001). The colonial origins of comparative development: An empirical investigation. American Economic Review, 91 (5),

15 , REED, T. and ROBINSON, J. A. (2014). Chiefs: Economic development and elite control of civil society in Sierra Leone. Journal of Political Economy, 122 (2), ALESINA, A. and FUCHS-SCHÜNDELN, N. (2007). Goodbye Lenin (or not?): The effect of communism on people. American Economic Review, 97 (4), BALDASSARRI, D. and GROSSMAN, G. (2011). Centralized sanctioning and legitimate authority promote cooperation in humans. Proceedings of the National Academy of Sciences, 108 (27), BANERJEE, R. (2016a). Corruption, norm violation and decay in social capital. Journal of Public Economics, 137, (2016b). On the interpretation of bribery in a laboratory corruption game: moral frames and social norms. Experimental Economics, 19 (1), BARR, A. and SERRA, D. (2009). The effects of externalities and framing on bribery in a petty corruption experiment. Experimental Economics, 12 (4), BECKER, S. O., BÖCKH, K., HAINZ, C. and WÖSSMANN, L. (2015). The Empire is dead, long live the empire! Long-run persistence of trust and corruption in the bureaucracy. Economic Journal, forthcoming. BEEKMAN, G., BULTE, E. and NILLESEN, E. (2014). Corruption, investments and contributions to public goods: Experimental evidence from rural Liberia. Journal of Public Economics, 115, BOHNET, I., GREIG, F., HERRMANN, B. and ZECKHAUSER, R. (2008). Betrayal aversion: Evidence from Brazil, China, Oman, Switzerland, Turkey, and the United States. American Economic Review, 98 (1), and ZECKHAUSER, R. (2004). Trust, risk and betrayal. Journal of Economic Behavior & Organization, 55 (4), BOURDIEU, P. (1977). Outline of a theory of practice. Cambridge: Cambridge University Press. 15

16 BREHM, J. and RAHN, W. (1997). Individual-level evidence for the causes and consequences of social capital. American Journal of Political Science, 41 (3), CAGALA, T., GLOGOWSKY, U., GRIMM, V. and RINCKE, J. (2015). Cooperation and trustworthiness in repeated interaction, manuscript, University of Erlangen-Nuremberg. CAMERON, L., CHAUDHURI, A., ERKAL, N. and GANGADHARAN, L. (2009). Propensities to engage in and punish corrupt behavior: Experimental evidence from Australia, India, Indonesia and Singapore. Journal of Public Economics, 93 (7), CAMPOS, E., LIEN, D. and PRADHAN, S. (1999). The impact of corruption on investment: Predictability matters. World Development, 27, CUBITT, R. P., GAECHTER, S. and QUERCIA, S. (2015). Conditional cooperation and betrayal aversion, CESifo Working Paper Series No DASGUPTA, P. and SERAGELDIN, I. (1999). Social capital: A multifaceted perspective. Washington D.C.: World Bank. DI TELLA, R. and PÉREZ-TRUGLIA, R. (2015). Conveniently upset: Avoiding altruism by distorting beliefs about others. American Economic Review, 105 (11), DUFWENBERG, M. and KIRCHSTEIGER, G. (2004). A theory of sequential reciprocity. Games and Economic Behavior, 47 (2), FEIGENBERG, B., FIELD, E. and PANDE, R. (2013). The economic returns to social interaction: Experimental evidence from microfinance. Review of Economic Studies, 80, FISCHBACHER, U. (2007). z-tree: Zurich toolbox for ready-made economic experiments. Experimental Economics, 10 (2), FUKUYAMA, F. (1995). Trust. New York: Basic Books. GLAESER, E. L., LAIBSON, D. and SACERDOTE, B. (2002). An economic approach to social capital. Economic Journal, 112 (483),

17 GNEEZY, U., KEENAN, E. A. and GNEEZY, A. (2014). Avoiding overhead aversion in charity. Science, 346 (6209), GREIF, A. (1994). Cultural beliefs and the organization of society: A historical and theoretical reflection on collectivist and individualist societies. Journal of Political Economy, 102 (5), GREINER, B. (2015). Subject pool recruitment procedures: Organizing experiments with ORSEE. Journal of the Economic Science Association, 1 (1), GUISO, L., SAPIENZA, P. and ZINGALES, L. (2008). Long term persistence, NBER Working Paper HENRICH, J., MCELREATH, R., BARR, A., ENSMINGER, J., BARRETT, C., BOLYANATZ, A., CARDENAS, J. C., GURVEN, M., GWAKO, E., HENRICH, N., LESOROGOL, C., MARLOWE, F., TRACER, D. and ZIKER, J. (2006). Costly punishment across human societies. Science, 312 (5781), HERRMANN, B., THÖNI, C. and GÄCHTER, S. (2008). Antisocial punishment across societies. Science, 319 (5868), KAUFMANN, D. (2005). Myths and realities of governance and corruption. In Global competitiveness report , World Economic Forum, pp KNACK, S. and KEEFER, P. (1997). Does social capital have an economic payoff? A cross-country investigation. Quarterly Journal of Economics, 112 (4), LEVI, M. (1998). A state of trust. In V. Braithwaite and M. Levi (eds.), Trust and governance, New York: Russel Sage Foundation. and STOKER, L. (2000). Political trust and trustworthiness. Annual Review of Political Science, 3, MAURO, P. (1995). Corruption and growth. Quarterly Journal of Economics, 110 (3),

18 NANNICINI, T., STELLA, A., TABELLINI, G. and TROIANO, U. (2013). Social capital and political accountability. American Economic Journal: Economic Policy, 5 (2), OECD (2013). Tax and development: What drives tax morale? Organization for Economic Cooperation and Development. OLKEN, B. A. (2006). Corruption and the costs of redistribution: Micro evidence from Indonesia. Journal of Public Economics, 90 (4-5), (2007). Monitoring corruption: Evidence from a field experiment in Indonesia. Journal of Political Economy, 115 (2), and PANDE, R. (2012). Corruption in developing countries. Annual Review of Economics, 4, OSTROM, E. (1990). Governing the commons: The evolution of institutions for collective action. Cambridge: Cambridge University Press. PUTNAM, R. D. (1993). Making democracy work: Civic traditions in modern Italy. Princeton: Princeton University Press. (1995). Bowling alone: America s declining social capital. Journal of Democracy, 6 (1), RABIN, M. (1993). Incorporating fairness into game theory and economics. American Economic Review, 83 (5), REINIKKA, R. and SVENSSON, J. (2004). Local capture: Evidence from a central government transfer program in Uganda. Quarterly Journal of Economics, 119 (2), and (2005). Fighting corruption to improve schooling: Evidence from a newspaper campaign in Uganda. Journal of the European Economic Association, 3 (2-3), THÖNI, C., TYRAN, J. R. and WENGSTRÖM, E. (2012). Microfoundations of social capital. Journal of Public Economics, 96 (7-8),

19 TOCQUEVILLE, A. and BENDER, T. (1981). Democracy in America. New York: Modern Library. USLANER, E. M. (2003). Trust, democracy and governance: Can government policies influence generalized trust? In M. Hooghe and D. Stolle (eds.), Generating social capital: Civil society and institutions in comparative perspective, New York: Palgrave McMillan. WEI, S. (2000). Local corruption and global capital flows. Brookings Papers on Economic Activity, 2, ZAK, P. J. and KNACK, S. (2001). Trust and growth. Economic Journal, 111 (470),

20 Figure 1: Corruption Impairs Social Capital Formation 20 Contribution to Social Capital (Mean Contribution in % of Endowment) Control Corruption Note: The figure shows mean contributions to social capital in the corruption and the control condition. The difference in means between conditions is significant at the 5% level (P = 0.039, n = 228, Mann-Whitney U test). Error bars show the mean ± the standard error of the mean. 20

21 Figure 2: Histograms of Contributions and Corresponding CDFs Percent Corruption: Relative Frequency CDF Control: Relative Frequency CDF Contribution to Social Capital (Contribution in % of Endowment) Note: The figure shows histograms of contributions and corresponding cumulative distribution functions (CDFs) of contributions for corruption (red) and control (blue) conditions. The bin size for the histograms is 10 ECU. 21

22 Figure 3: Heterogeneity of Treatment Effect by Baseline Trust 25 Trusting Types Non-Trusting Types Contribution to Social Capital (Mean Contribution in % of Endowment) Control Corruption Control Corruption Note: The figure shows mean contributions to social capital in the corruption and the control condition for individuals with a high baseline level of interpersonal trust (left panel) and individuals with low baseline trust (right panel). Subjects are categorized as trusting if they choose agree or completely agree in response to the statement Generally speaking, people can be trusted, and categorized as non-trusting if they choose disagree or completely disagree. For non-trusting types, the difference in means between conditions is insignificant (P = 0.32, n = 98, Mann-Whitney U test). For trusting types, the difference in means between conditions is significant at the 1% level (P = ), n = 130, Mann-Whitney U test). Error bars show the mean ± the standard error of the mean. 22

23 Figure 4: Corruption and Social Capital Formation in the World Values Survey Contribution to Social Capital (Percent Active in Voluntary Organizations) Low Corruption High Corruption Note: The figure shows the percent of respondents to the World Values Survey wave 6 ( ) stating they are spending a strictly positive number of hours per month working in voluntary organizations, by perceived corruption among government officials. The left bar shows activity for voluntary organizations for respondents choosing an index value lower or equal to five (on a scale from one to ten) when assessing corruption within their country s government (low corruption). The right bar shows the respective figure for respondents choosing an index value larger or equal to six (high corruption). The difference in means between low and high corruption is significant at the 1% level (P < , n = 7716, Mann-Whitney U test). Error bars show the mean ± the standard error of the mean. The set of countries were both survey questions were asked comprises Algeria, Bahrain, Iraq, Kuwait, Lebanon, Tunisia, and Yemen. 23

24 Appendix (Not For Publication) Figure A1: Experimental Design: Timing of the One-Shot Game Contributor Corruption Administrator Control Stage 1 Decision: selection of association Decision: expropriation (yes/no) Random draw: additional compensation (yes/no) Stage 2 Information: outcome of administrator decision/ random draw Decision: contribution to social capital - - Time - Administration of payments 24

25 Figure A2: Number of Contributors Conditional on Administrator Decisions and the Random Draw Administrator Decision (Corruption Condition) Random Draw (Control Condition) Expropriation/ Additional Compensation Note: The figure shows the number of contributors conditional on administrator decisions (corruption condition) and the random draw (control condition). We derive our main results from comparing the 114 contributors facing an administrator who decided in favor of expropriation with the 114 contributors in groups where the random draw assigned an additional compensation to the administrator. The symmetry in the distribution between conditions comes from the fact that we set the probability for administrators receiving the additional compensation in the control condition to be equal to the empirical propensity of administrators choosing expropriation in the corruption condition. 25

26 Table A1: Descriptive Statistics: Behavior of Contributors Panel A: Contributions by Treatment Control Corruption Contributions in % of Endowment (18.9) (16.3) Number of Observations Panel B: Contributions by Treatment & Level of Baseline Trust Control Corruption Trusting Non-Trusting Trusting Non-Trusting Types Types Types Types Contributions in % of Endowment (19.3) (16.9) (15.7) (17.1) Number of Observations Note: The table shows mean contributions and standard deviations (in parentheses). The sample consists of all contributors who were matched to an administrator who benefited from contributions, either through active expropriation (corruption condition) or passively through the additional compensation (control condition). Panel A displays figures differentiated by condition. Panel B shows mean contributions by condition and baseline level of interpersonal trust. The trust measure comes from a survey conducted after the experiment. Subjects were asked to respond to the statement Generally speaking, people can be trusted, with response categories completely agree, agree, disagree, and completely disagree. We categorize subjects as trusting types if they choose one of the first two categories, and as non-trusting types if they choose one of the last two categories. 26

27 Table A2: Conditional Treatment Effects OLS Tobit (1) (2) Effect of Corruption (2.30) (2.15) Log likelihood Number of Observations 228 Average Contribution in Control Group 16.9 Note: Column (1) shows the effect of corruption using a simple OLS regression. As 34.6 percent of observations are left-censored at zero, Column (2) reports the corresponding effect from a Tobit regression (there is no right-censoring because no subject contributed the full endowment). The Tobit estimate shown in Column (2) is the average partial effect of changing from the neutral condition to the corruption condition. Standard errors (robust for OLS, Delta Method for Tobit) are reported in parentheses. Both regressions account for subjects gender, field of study (economics vs. other), and a dummy for experience in laboratory experiments. All results are robust to including further controls, including subjects age and more disaggregated dummies for field of study. In the survey conducted after the experiment, one subject stated that she was confused and did not understand how contributions would translate into funds going to the associations. If we exclude this subject, results become somewhat stronger: the OLS estimate becomes 5.40, the Tobit estimate is

28 INSTRUCTIONS OF CORRUPTION AND CONTROL TREATMENTS (Control treatment instructions replace highlighted text with grey text in brackets) Welcome to the experiment, we are grateful for your participation. Please read the instructions carefully. If you have any questions, please raise your hand. One of the experimenters will answer your questions. You are not allowed to communicate with other participants of the experiment. Please turn off your mobile phone. During the experiment it is not allowed to take notes. This is an experiment on economic decision making. You can earn money with your participation. You will receive 4 Euro as a participation fee. During the experiment you can earn additional money. Your additional earnings depend on your behavior and the behavior of the other participants. During the experiment, money will be displayed in ECU (Experimental Currency Units) with an exchange rate of 1 Euro = 10 ECU. Your entire earnings will be paid to you in cash at the end of the second part of the experiment. Participants will neither be informed about the identity of other participants, nor about others role in the experiment or earnings. The data will be analyzed anonymously. EXPERIMENT ROLES Every participant is assigned a role, either contributor or administrator. The roles are randomly assigned at the beginning of the experiment and do not change during the experiment. All participants are treated equally in role assignment. Every participant will be informed about her role at the beginning of the experiment. GROUPS All participants are randomly assigned into independent groups. Each group consists of three contributors and one administrator. Groups remain the same throughout the entire experiment. PROCEDURE The experiment consists of six steps. You decide only once, there are no repetitions. Step Contributor Administrator 1) Receipt of endowment Receipt of fixed compensation 2) Selection of an organization 3) Information about the administrator s decision (additional compensation) 4) Payment to organization - Decision on expropriation of payments (Random selection of additional compensation) 5) - Administration of payments 6) Calculation of payoffs of all participants, organizations and payment - STEP 1: RECEIPT OF ENDOWMENT / FIXED COMPENSATION Contributors receive an endowment of 100 ECU. The administrator receives a fixed compensation of 100 ECU for the administration of payments. 28