INTRODUCTION
Scholars and policymakers have long struggled with the question of how to reduce corruption in public institutions. While one branch of this literature focuses on the design of optimal rules to reduce opportunities for graft and increase incentives for honesty (Andvig et al., Reference Andvig, Fjeldstad, Amundsen, Sissener and Søreide2001; Klitgaard, Reference Klitgaard1988; Rose-Ackerman, Reference Rose-Ackerman1978), other work highlights the many ways in which ordinary citizens can bring about greater governmental accountability (Adserà et al., Reference Adserà, Boix and Payne2003; Besley, Reference Besley2006; Grimes, Reference Grimes2013; Olken, Reference Olken2007; Rose-Ackerman, Reference Rose-Ackerman1999; World Bank, 2004). One important mechanism highlighted by this latter stream of research involves the reporting of corruption to formal oversight agencies.
As many scholars have noted, law enforcement authorities seldom have adequate time and resources to investigate all instances of potential malfeasance (McCubbins and Schwartz, Reference McCubbins and Schwartz1984; Sunshine and Tyler, Reference Sunshine and Tyler2003; Tyler, Reference Tyler2010). Many oversight agencies must therefore depend upon citizens to sound “fire alarms” to expose corruption and provide evidence against wrongdoers. Furthermore, while political elites may have incentives to block governance reform, the same is not true for citizens, who are often corruption’s primary “victims,” and therefore ideally placed to push for change (Mungiu-Pippidi, Reference Mungiu-Pippidi2006). Indeed, case studies of successful anti-corruption campaigns have highlighted the importance of grassroots monitoring in improving accountability (Grimes, Reference Grimes2013; Manion, Reference Manion2009; Peruzzotti and Smulovitz, Reference Peruzzotti and Smulovitz2006).
Although the literature often assigns citizens (or civil society) a central role in controlling corruption, individuals’ motivations to engage in such actions (particularly in endemically-corrupt societies) remain poorly understood. Does the willingness to participate in grassroots monitoring differ between high-corruption and low-corruption societies? And does the answer depend on whether individuals in these societies have access to effective and efficient enforcement institutions?
This article explores the empirical relationship between institutions, corruption exposure, and anti-corruption monitoring. I report results from an economic experiment involving participants from the North and South of Italy. Importantly, research has shown that the level of corruption differs significantly across these regions (Banfield, Reference Banfield1958; Chang et al., Reference Chang, Golden and Hill2010; Golden and Picci, Reference Golden and Picci2005; Putnam, Reference Putnam1993), and participants who are socialized in these separate environments may have internalized different accountability norms. The experimental design allows me to study the effect of these norms by holding the quality of enforcement institutions constant. Moreover, for each population, I also vary the probability that someone reported for corruption will be formally sanctioned, and thereby test whether regional effects depend upon the effectiveness of formal oversight agencies. By comparing individual decision-making under different norms and institutions, this paper contributes to the growing literature on “bottom-up” accountability (Barr et al., Reference Barr, Lindelow and Serneels2009; Bauhr and Grimes, Reference Bauhr and Grimes2014; Cameron et al., Reference Cameron, Chaudhuri, Erkal and Gangadharan2009; Grimes, Reference Grimes2013).
RELATED LITERATURE
Several studies have examined the relationship between cultural norms and corruption (Barr and Serra, Reference Barr and Serra2010; Cameron et al., Reference Cameron, Chaudhuri, Erkal and Gangadharan2009; Fisman and Miguel, Reference Fisman and Miguel2007; Paldam, Reference Paldam2002; Treisman, Reference Treisman2000). Fisman and Miguel (Reference Fisman and Miguel2007) investigate the parking behavior of United Nations diplomats during a period in which they were immune from enforcement actions. Even in the absence of legal constraints, diplomats from low-corruption countries accumulated significantly fewer unpaid parking violations (a form of abuse of office), suggesting the importance of cultural norms in curbing opportunistic conduct. Barr and Serra (Reference Barr and Serra2010) report similar findings from a laboratory experiment in the UK: exposure to a “culture of corruption” in students’ home countries is associated with a greater propensity to bribe in the lab.
While extant studies focus on variation in corrupt behavior, relatively little research has examined how corruption tolerance varies across societies (Cameron et al., Reference Cameron, Chaudhuri, Erkal and Gangadharan2009). However, a willingness to participate in corruption oneself does not necessarily imply an acceptance of such behavior on the part of others. Indeed, ethnographic research has shown that individuals can perceive the same corruption scenario as right or wrong, depending on whether they are the beneficiaries or the victims of the transaction (Hasty, Reference Hasty2005; Olivier de Sardan, Reference Olivier de Sardan1999; Smith, Reference Jordan2010). Employing a rational-choice framework, Heckathorn (Reference Heckathorn1989) argues that, under some conditions, it may be optimal for an individual to act opportunistically, while simultaneously policing others’ behavior. Thus, it is important to study how everyday exposure to wrongdoing shapes accountability norms, as distinct from honesty norms.
In theory, greater exposure to malfeasance in daily life may socialize citizens into a “culture of corruption,” and thereby increase acceptance of wrongdoing (Barr et al., Reference Barr, Lindelow and Serneels2009). This may be especially true if individuals come to believe that bribery is a routine strategy employed by all “normal” citizens to gain access to public services (Cameron et al., Reference Cameron, Chaudhuri, Erkal and Gangadharan2009; Miller, Reference Miller2006; Persson et al., Reference Persson, Rothstein and Teorell2012). In this context, to sanction someone for accepting a “gift” would seem overly-scrupulous, hypocritical, and insensitive to “the way things work.” Thus beliefs about the ubiquity of illicit payments serve to justify and excuse such behavior, thereby weakening the norm of accountability.
Cross-national empirical evidence would seem to support this argument. Figure 1 graphs the relationship between corruption tolerance and cross-country corruption levels, as measured by Transparency International (TI)’s 2013 Corruption Perceptions Index. Higher scores along the x-axis indicate a more “honest” society. The y-axis displays the percentage of individuals who indicate that they would be willing, hypothetically, to report an incident of corruption. The data are drawn from TI’s 2013 Global Corruption Barometer, and are available for over 100 countries. The figure shows that fewer citizens in high corruption societies are willing to report malfeasance, suggesting a direct relationship between corruption exposure and corruption tolerance.
Figure 1 Willingness to Report Corruption by Country Corruption Ranking.
However, this relationship is complicated by the fact that individuals in different societies also face different institutional constraints. In particular, while citizens can report corruption, they cannot directly enforce the law. Instead, they must depend upon formal oversight agencies to act upon their complaints and sanction the perpetrators (Grimes, Reference Grimes2013). However, in countries where corruption is pervasive, such offices may be lacking, ineffectual, or themselves deeply corrupted (Bauhr and Grimes, Reference Bauhr and Grimes2014). Thus, citizens’ apparent tolerance of corruption may not arise from moral lassitude, but rather from the perception that “sounding the alarm” is futile (Persson et al., Reference Persson, Rothstein and Teorell2012).
In principle, economic experiments can help to disentangle the influence of institutional and normative factors on the willingness to blow the whistle on corruption. By directly controlling the “rules of the game,” experiments can isolate the effect of normative constraints, as well as simulate different institutional conditions. Yet, the few studies that have adopted this approach have produced inconclusive results. For example, Cameron et al. (Reference Cameron, Chaudhuri, Erkal and Gangadharan2009) compared participants from four societies (Australia, India, Indonesia, and Singapore) in terms of their propensity to both engage in and punish bribery. While some results accord with our prior intuitions (e.g., Australians are more critical of corruption than Indians), other findings are rather surprising (e.g., Singaporeans tend to be more tolerant of bribery than Indonesians). Given these mixed results, more research examining how exposure to corruption affects accountability norms is needed. The experiment described below contributes to filling this gap.
METHODOLOGY
Setting
The experiment was conducted in two locations. A first set of laboratory sessions was implemented at [Northern Italian University] (NIU) in the Spring of 2013, with a follow-up in Summer 2016. These sessions involved both (regionally-native) Northern Italian students, as well as Southern Italians who were also enrolled at NIU.Footnote 1 However, in comparing these two groups, it is not possible to rule out that Southerners who choose to attend NIU may be different from Southerners who remain in their home regions. This self-selection may present inferential problems if, for instance, individuals decide to migrate precisely because they are frustrated with the level of corruption prevailing in the South (Casari et al., Reference Casari, Ichino, Michaeli, de Paola, Scoppa and Marandola2017). To address this possibility, a second set of sessions was conducted at [Southern Italian University] (SIU) in 2016. The full sample is thus composed of three subgroups: (a) Northern Italians enrolled at NIU, (b) Southern Italians enrolled at NIU, and (c) Southern Italians enrolled at SIU. All participants were recruited via ORSEE (Greiner, Reference Greiner2015), and the experiment was programmed in zTree (Fischbacher, Reference Fischbacher2007). In total, 20 sessions were conducted.Footnote 2 On average, each session lasted around 1 hour and participants earned approximately 13 euros (USD 17.50).
The corruption game
The experiment simulates petty corruption in a public hospital setting.Footnote 3 This setting was selected because the medical sector is regarded as among the more corruption-ridden institutions in Italian society.Footnote 4 Furthermore, while we may doubt that ordinary citizens have encountered corruption in other scenarios (e.g., public procurements), participants are likely to have a more concrete idea of how corruption in the health sector operates. Finally, since hospitals in Italy are public institutions, individuals should expect impartial treatment, and any personal favoritism is likely to be understood as corruption. Overall, the framing brings a measure of realism, and affords greater confidence that behavior in the lab will more faithfully reflect choices in real life.
Interactions take place between participants assigned to one of three roles: Nurse, Early Patient (PE), or Late Patient (PL). Patients are told to imagine that they are waiting in line to see the doctor, but face different wait times depending on whether they are in the role of PL or PE. Waiting is costly for PLs, but they can potentially skip the line by offering the Nurse a “gift” in exchange for faster service. However, PEs are harmed by this transaction, and must decide whether to punish corrupt Nurses by reporting them to the hospital administration. The experimental manipulation, described below, relates to the efficiency of this reporting mechanism.
Overall, the experiment draws inspiration from the designs employed by Barr and Serra (Reference Barr and Serra2010) and Cameron et al. (Reference Cameron, Chaudhuri, Erkal and Gangadharan2009).Footnote 5 However, in contrast to these studies, the present paper is primarily interested in how social norms and institutions affect the decision to blow the whistle on corruption (as opposed to the decision to engage in bribery). Therefore, in my analysis, I focus attention only on the behavior of PEs, and the main dependent variable under consideration is the willingness of PEs to report corrupt transactions.
Participants were provided information about the various roles as follows. PLs begin with an initial endowment of 32 experimental currency units (ECU). Each PL is randomly matched to one Nurse, and has the option of offering this Nurse a “gift” worth 6 ECU in exchange for being allowed to jump the queue.Footnote 6 If the offer is accepted, the PL transfers 6 ECU to the Nurse, but avoids a waiting cost of 16 ECU, and thus earns 32−6 = 26 ECU. However, as a consequence of having been skipped over, all PEs must now wait longer in line, and each loses 3 ECU. By contrast, if the PL does not offer a gift, or if his offer is refused by the Nurse, the PL pays the full waiting cost of 16 ECU (and therefore earns only 32−16 = 16 ECU), but the payoffs of PEs and Nurses are unaffected.
PEs begin with an initial endowment of 32 ECU, and are also randomly matched to one Nurse. Before receiving any information about the Nurse’s actions, PEs must first decide whether they would, in principle, be willing to report a corrupt Nurse to the hospital administration, at the cost of a reporting “fee” of 3 ECU. In case a report is filed, the hospital administration may or may not impose a fine on the Nurse, depending on the treatment condition (described below).
Importantly, the PE’s expression of a “willingness to report” results in an actual report only if the Nurse has, in fact, accepted a gift. By contrast, even if the PE is willing to report, a report is not filed if the Nurse has chosen to refuse gifts, and/or was not offered any gifts. However, so long as a report is made, the PE must pay the reporting fee irrespective of whether the Nurse is actually punished. Moreover, even if the Nurse is fined, corrupt PLs still remain at the front of the line, meaning that PEs lose 3 ECU for every PL who jumps the queue, irrespective of their own decisions. This feature ensures that punishment conveys no economic benefit to PEs.
Finally, Nurses begin with an initial endowment of 24 ECU, and are randomly matched to any number of PEs and PLs.Footnote 7 Without knowing the decisions of the other players, Nurses must decide whether they would be willing to accept gifts from PLs matched to them, or whether they would, in principle, refuse such offers. If Nurses are not open to accepting gifts, then matched PLs must pay the full waiting cost of 16 Tokens, but no PEs are harmed.Footnote 8 An exchange of favors takes place only if the Nurse indicates a willingness to accept gifts, and at least one of the matched PLs offers one. In this case, any (matched) offering PL earns a final payoff of 26 ECU, but all PEs lose 3 ECU for each PL who skips the line.
A Nurse who refuses gifts earns a certain payoff of 24 ECU. In contrast, corrupt Nurses’ payoffs depend on both the treatment condition and the decisions of PEs. Specifically, the experiment is implemented under two different conditions, which simulate varying levels of institutional effectiveness. In the “strict enforcement” version, a corrupt Nurse who is reported is sanctioned 100% of the time. In this case, he forfeits any gifts he has received, and also pays a fine of 9 ECU, so that he retains only 24−9 = 15 ECU at the end of the round. However, in the “lax enforcement” version, the PE’s report results in the imposition of a sanction only 50% of the time. The other 50% of the time, the Nurse pays no fine and keeps whatever gifts he has received. Importantly, although the incentives facing Nurses change across the two conditions, the monetary payoffs facing PEs remain identical: any report costs 3 ECU, regardless of whether punishment is actually imposed. The one-shot simultaneous game has a single equilibrium outcome: all PLs offer gifts, no PEs are willing to report corrupt exchanges, and all Nurses are willing to accept gifts.
In addition to the payoff structure, participants are also informed that they will play the corruption game for three rounds. Participants are randomly assigned to a role in round one, and will rotate through the remaining (unplayed) roles in random order in rounds two and three. This ensures that roughly one-third of participants are assigned to each role in every round.Footnote 9 Participants are also rematched in every round, and feedback on the outcome of interactions in all rounds is provided only at the conclusion of the session. Finally, participants are provided with a summary of these rules when making their decisions.Footnote 10
Participants
The data reported in this paper are drawn from 371 participants: 156 Northerners at NIU, 120 Southerners at NIU, and 95 Southerners at SIU.Footnote 11 Within each group, institutional treatments were randomly assigned at the level of the experimental session, with half of the sessions being selected to implement the “strict enforcement” condition, and the remainder implementing the “lax enforcement” version. Participants were unaware that there were two versions of the experiment.
In the overall sample, 51.8% of the participants were male (NIU: 53.6%, SIU: 46.3%), and the median age was 23 years (NIU: 24, SIU: 21). 56.9% of the overall sample is composed of triennale students (NIU: 52.5%, SIU: 69.5%), while the remainder is made up of magistrale students.Footnote 12 Table 1 compares demographic characteristics across treatment conditions for the three subgroups: Northerners at NIU, Southerners at SIU, and Southerners at NIU. Aside from the proportion of triennale students among Southerners enrolled at NIU, there are no statistically significant differences across treatment conditions.
Table 1 Covariate Balance across Subgroup Samples and Treatment Conditions
Note: p-values derived from two-sided t-tests. z-statistics from non-parametric tests-of-proportions (for Male and Triennale) and Wilcoxon rank-sum tests (for Age) are also displayed. The difference in the proportion of Triennale students is significant at the 5% level for Southerners enrolled at NIU (Panel C).
Research questions
If all individuals are perfectly selfish, nobody would report corruption in either of the institutional conditions, since reporting leaves PEs strictly worse off. However, if PEs are motivated by a norm of accountability, they may choose to denounce corrupt Nurses despite the monetary disincentives. The literature also shows that accountability norms can differ across societies (Cameron et al., Reference Cameron, Chaudhuri, Erkal and Gangadharan2009) in ways that may be related to the institutional environment (Bauhr and Grimes, Reference Bauhr and Grimes2014; Persson et al., Reference Persson, Rothstein and Teorell2012). The experiment therefore addresses the following research questions:
1. Are participants from societies with higher levels of corruption less willing to report bribery in comparison to participants from societies experiencing lower levels of corruption?
2. Do these effects depend upon the quality of enforcement institutions (i.e., the probability that these reports will be acted upon)?
RESULTS
Overall, 220 out of 371 participants (59.3%) indicated a willingness to report a corrupt Nurse. As a preliminary step, we can break this number down in two ways. First, pooling both institutional conditions, we observe very little difference between the subgroups: 57.1% of Northerners indicate a willingness to report, compared to 57.9% of Southerners at SIU, and 63.3% of Southerners at NIU. Second, pooling all three subgroups, we find evidence of an institutional effect: while only 51.8% of participants in the lax enforcement treatment are willing to report, this number rises to 67.8% in the strict enforcement treatment.
Next, I consider the possibility that Northerners and Southerners may behave differently depending upon the institutional condition to which they have been assigned. This interaction effect is illustrated in Figure 2. Moving from lax to strict enforcement increases reporting from 52.9% to 62.3% among Northerners, from 50.0% to 65.3% among Southerners at SIU, and from 51.6% to 75.9% among Southerners at NIU. These latter results suggest that individuals from high-corruption societies are not “culturally” predisposed to tolerate malfeasance. Rather, when facing the same institutional environments, Southern Italians appear to be just as vigilant as their Northern counterparts, if not more so.
Figure 2 Willingness to Report by Sample Subgroup and Treatment Condition.
To check the statistical significance of these findings, Table 2 presents results from linear probability models regressing the willingness to report on dummies for the enforcement condition (Strict) and sample subgroup. I report both heteroskedasticity-robust standard errors, as well as p-values derived from pairs cluster bootstrapped t-statistics (clustered at the session level) to account for the number of sessions (Cameron et al., Reference Cameron, Gelbach and Miller2008; Harden, Reference Harden2011).Footnote 13
Table 2 Regression Results: Willingness to Report
Note: heteroskedasticity-robust standard errors in parentheses. p-values derived from pairs cluster bootstrapped t-statistics are also reported in italics.
Column (1) confirms that participants are more willing to report under the strict enforcement condition: the coefficient on Strict is more than twice the size of the robust standard error, and adjustment for clustering results in a p-value <0.01. Column (2) shows that there are no baseline differences in reporting rates between the three subgroups, holding the institutional environment constant.
Column (3) adds interactions between Strict and the two South subgroups to test for differential responses to the change in institutional conditions. Additionally, Column (4) pools South-SIU with South-NIU and considers whether the treatment effect differs between Northerners and Southerners in general. Finally, I address the fact that the largest difference in reporting rates in Figure 2 appears among Southerners at NIU. This observation is consistent with the self-selection of individuals who are fed-up with corruption to attend university outside of the South. Accordingly, Column (5) tests whether the treatment effect differs between the South-NIU subgroup (e.g., migrants) and participants who remain in their home regions.
Overall, none of the interactions in Columns (3)–(5) is statistically significant, indicating that the size of the treatment effect is similar across various partitions of the sample. However, a comparison of the coefficient on Strict across Columns (3) and (4) reveals an additional aspect of the main treatment effect. Specifically, Column (4) indicates that better institutions increased reporting by 20.1% among Southern participants, while Column (3) estimates that the corresponding treatment effect for Northerners is only 9.4% (n.s.). Thus, even though it is not possible to statistically distinguish the size of the treatment effect across subgroups, the results taken together suggest that the main treatment effect is driven most prominently by the behavior of Southern participants.
Additional robustness tests are reported in the Online appendix. Briefly, I show that the main treatment effect does not differ between the 2013 and 2016 waves of the experiment, and also that there are no “carry-over” effects from decisions taken in the PL and Nurse roles in previous rounds. I also drop subgroups one at a time to ensure that the results do not depend upon the inclusion of any particular subgroup. The main findings remain substantively unchanged across all specifications.
DISCUSSION AND CONCLUSION
In sum, the experimental results offer little support for the idea that a “culture of corruption” underlies the tolerance of illicit practices (at least among Italians). When faced with the same institutional environment, Southerners are not more “culturally” predisposed to tolerate malfeasance as compared to their Northern counterparts. Rather, the experiment shows that individuals from a “high-corruption” society can indeed be engaged in grassroots monitoring, provided that the right institutional arrangements are in place. These results thus highlight the importance of institutional quality in shaping accountability norms.
The ability to distinguish between the institutional versus normative drivers of bottom-up accountability has important policy implications. If citizens are socialized into a “culture of corruption,” then institutional reforms are unlikely to unleash a wellspring of popular action, and greater accountability most likely arises from more vigilant top-down monitoring. By contrast, if (as suggested by this article) citizens in highly-corrupt societies are responsive to institutional incentives, then it may be possible to harness this popular indignation in the fight against corruption, so long as the necessary institutional tools are available.
More generally, the findings suggest that “bottom-up” and “top-down” enforcement efforts may be mutually reinforcing. In particular, at the outset, enforcement authorities can demonstrate their credibility by acting upon citizen reports and punishing high-profile perpetrators. These actions then serve to strengthen the belief that citizens are now facing a “strict enforcement” regime, and thereby generate more frequent “fire alarms” from the public. Finally, the loop is closed as greater civic engagement multiplies the investigatory and prosecutorial capacities of enforcement authorities, resulting in even higher punishment probabilities.Footnote 14
Yet while the experimental results suggest that such a virtuous cycle is indeed possible, they also raise several questions about the scope conditions under which such a process might occur. How representative are North and South Italy of “honest” and “corrupt” societies more generally? How might these results depend upon the specific situational context (i.e., the hospital setting) examined? And given that whistleblowing in real life is rarely anonymous, how might social considerations (i.e., publicly playing the “hero” or the “rat”) influence individual decision-making in different societies?
As Cameron et al. (Reference Cameron, Chaudhuri, Erkal and Gangadharan2009) note, the relationship between corruption exposure and accountability norms is extremely complex, and this paper is one of the first to study this phenomenon with an eye towards incorporating institutional effects. However, more research on a wider range of societies with differing levels of corruption and institutional effectiveness is needed to fully resolve these outstanding questions.
SUPPLEMENTARY MATERIALS
To view supplementary material for this article, please visit https://doi.org/10.1017/XPS.2017.26
