Changing a ‘culture’ of corruption: Evidence from an economic experiment in Italy

Abstract

Empirical evidence demonstrates that bribery, extortion and graft are often the outgrowths of a deeper ‘culture of corruption’ which has proved disconcertingly resilient in the face of public sector reforms. This article investigates whether changing collective beliefs about how ‘most people in society’ will behave can reform prevailing cultural practices. Employing an economic experiment involving northern and southern Italian university students, this study shows that (costly) honest behavior can be sustained by conditional beliefs about the honesty of others. I also hypothesize that, given southern Italians’ reputation for corruption, informing participants that they are interacting with southerners should increase the level of bribery in the experiment. However, surprisingly, I find the opposite effect: when exposed to information about the identity of their fellow participants, southern Italians are not only less corrupt, but they are also more likely to believe that their counterparts are less corrupt. I discuss several explanations which may account for these unexpected findings. Overall, the paper provides theoretical foundations and experimental support for how new cultural practices might emerge.

Keywords

Corruption culture experiment social norms Italy

Introduction

Political scientists, economists and legal scholars have traditionally attributed the causes of endemic corruption to weak legal enforcement and misaligned bureaucratic incentives (Klitgaard, 1988; Rose-Ackerman, 1999). However, in recent years, a growing body of evidence has demonstrated that bribery, extortion and graft are often the outgrowths of a deeper ‘culture of corruption’ which has proved disconcertingly resilient in the face of public sector reforms (Barr and Serra, 2010; Fisman and Miguel, 2007; Husted, 1999; Licht et al., 2007). The case of Italy demonstrates this point powerfully: although largely sharing a unified legal structure and formal institutions, Italian regions exhibit enormous diversity in the honesty and effectiveness of local government (Charron et al., 2013b, 2015; Golden and Chang, 2001; Golden and Picci, 2005; Putnam, 1993). While the ‘civic’ north is characterized by high levels of social trust that enable citizens to place their common welfare above the pursuit of narrow particularist interests, behavior in the south is exemplified by the ethos of ‘amoral familism’: ‘maximize the material, short-run advantage of the nuclear family: assume that all others will do likewise’ (Banfield, 1958: 83). Putnam (1993) has famously traced these cultural patterns to differences in political and social organization arising from the Middle Ages. The cultural roots of the Questione Meridionale run deep indeed.

This paper contributes to the literature on anti-corruption reform by analyzing this notion of culture from the perspective of formal, rational-choice theory. Rather than conceptualizing corrupt practices as arising from a set of deeply-ingrained values which are held in some societies, but not in others (see Ekeh (1975), Lipset and Lenz (2000), Husted (1999), Olivier De Sardan (1999) and Price (1974)), the rational-choice approach treats cultural differences as a contingent outcome of individual choices which, in equilibrium, are sustained through shared, mutually-enforcing beliefs about the consequences of deviating from the norm (Laitin and Zhang, 2012). For example, while corruption is generally condemned across societies (Rothstein and Torsello, 2014) an individual may be more likely to engage in corrupt practices when she believes that her fellow citizens are doing the same. In this context, bribery appears socially appropriate, and honesty achieves nothing except to leave her as the only ‘sucker’ (Persson et al., 2013). By contrast, if an individual believes that her peers are honest, she may refrain from taking advantage of corrupt opportunities in order to avoid triggering social disapproval. In this sense, the Italian south’s ‘culture of corruption’ and the north’s ‘civic culture’ can be reformulated as conditional choices, sustained by common knowledge beliefs about how other citizens will behave.

The equilibrium approach not only helps us to understand how cultural practices are maintained and reproduced over time, but also opens up vistas on the processes facilitating cultural change. Laitin (1998) shows how, in the context of Estonian language policy, changes in the economic calculus of educating children in Estonian rather than in Russian led to early moves toward national assimilation amongst the formerly monolingual Russian-speaking minority. Once parents started to believe that most other parents would choose Estonian over Russian education (lowering the relative payoff to learning Russian), they too made the shift, setting off a cultural cascade towards an all-Estonian equilibrium. Can a similar dynamic work in the corruption context? That is, if we change collective beliefs about whether ‘most people in society’ are corrupt or honest, can we reform the culture of corruption?

To address this question, I first present a game theoretic model of cultural choice. Based closely on the tipping games popularized by Laitin (2007), Mackie (1996) and Schelling (1978), the model helps to illustrate the dynamics through which individual decisions produce continuity and change in cultural practices. To test the implications of the model, I report results from a laboratory experiment involving Italian university students in a stylized corruption game, in which I attempt to induce a change in cultural practices by activating stereotypes about other participants. In particular, if it is common knowledge that ‘all southerners are corrupt’, then informing participants about whether they are playing with southerners should change both (a) beliefs how much corruption is likely to prevail within the group, and (b) participants’ own behavior.

I report some unexpected results: in the absence of information about their fellow participants in the corruption game, southerners are indeed more likely than northerners to accept bribes. Moreover, the likelihood of bribe-taking is significantly predicted by beliefs about the honesty of other participants. Yet, surprisingly, when informed that they are interacting with other southerners, southern participants are not only less corrupt, but they are also more likely to believe that their counterparts are less corrupt. In other words, the experiment confirms the dynamics illustrated in the tipping model, but the cultural equilibrium tips in the opposite direction to that hypothesized. I discuss several explanations that may account for these counter-intuitive findings, before concluding with some suggestions for future research.

An equilibrium approach to culture

Given that corrupt transactions are often extremely lucrative for the participants involved (or their immediate family or friendship circle), does it make sense to think of corruption as a collective choice? Why would individuals not take advantage of all corrupt opportunities that come their way, at least in cases where they will neither be easily detected nor harshly punished?

The obvious answer (at least to non-economists) is that we are often restrained from self-interested behavior not by the fear of legal sanctions, but rather by our sense of right and wrong. As Ariely (2013) shows, many individuals do not cheat (very much) when given the opportunity, even when it is common knowledge that the probability of detection is zero. Moreover, while sometimes our sense of justice is innately driven – for example, our private morality prevents most of us from killing ¹– in many other cases we define right and wrong according to our beliefs about how others think we ought to behave. There is nothing innately immoral about wearing shorts to the opera, or running a red light at a deserted intersection at three in the morning, but injunctive norms spell out rules of appropriateness in these situations (Cialdini et al., 1990). Behavioral conformity to normative prescriptions/proscriptions is ensured not through appeal to material costs and benefits, but rather through social rewards and sanctions: praise for following the rules, shame or ostracism for norms transgressions, and guilt at the thought of violating others’ expectations (Bicchieri, 2005; Elster, 1989; Sunstein, 1996). Once we introduce social rewards and punishments into the standard utility function, we can easily see how normative compliance leads individuals to take actions counter to their material self-interest.

Recent research shows that individuals in most societies do not internalize corruption as a legitimate part of their cultural repertoire (Rothstein and Torsello, 2014). Rather, even societies at the very bottom of cross-national corruption rankings possess injunctive norms against corruption (Miller, 2006; Persson et al., 2013). However, the force of these injunctions depends upon the underlying extent of normative violations: by observing the compliance/transgressions of others, individuals update their beliefs about the likelihood that their own actions will trigger social sanctions. For instance, if I see that most others cheat, I am likely to infer that they will not really mind if I cheat as well.² Indeed, to single me out in this environment would seem overly scrupulous, unfair or even hypocritical. Applied to the corruption context, individuals act as conditional norm compliers (Bicchieri, 2005): the more corruption they believe exists in society, the less reluctant they become to engage in corruption themselves.³ This leads to my first hypothesis, as follows.

Hypothesis 1. Although individuals understand that corruption is wrong, they will not abide by injunctive norms if they believe that such norms are commonly violated.

Economists have long recognized that corruption outcomes constitute frequency dependent equilibria. However, their explanations have tended to focus on the material mechanisms underlying individuals’ desires to coordinate with others (Andvig and Moene, 1990; Bardhan, 1997). For example, in societies where corruption is rampant, it is easier to find a corrupt partner since more people have the ‘skills’ to participate in corruption (Della Porta and Vannucci, 1999), there are more middlemen around to help match potential partners (Englebert, 2009) and there may exist informal institutions (e.g. the mafia) capable of enforcing illicit contracts (Gambetta, 2013). At the same time, law enforcement agencies may themselves be corruptible (Cadot, 1987), or prosecutors may be overwhelmed by the sheer number of corruption cases to try, leading to a lower probability that any one individual will be caught and punished (Lui, 1986).

The mechanism posited here – conditional norm compliance – represents a complement to these approaches.⁴ However, individual conformity to the modal behavior of society, rather than arising from an economic cost-benefit calculus, is sustained by the desire to receive social praise, or to avoid shame and social exclusion. This mechanism is depicted in Figure 1, which represents an individual’s decision to behave corruptly or honestly as a function of social costs. There exist two stable equilibria in which all are honest (H) or all are corrupt (C), along with a tipping point $τ$ . At any point to the left of $τ$ , because many others are corrupt, an individual does better by entering into corrupt transactions since there is likely to be little social stigma associated with the realization of illicit gains. By contrast, at any point to the right of $τ$ , an individual is better off behaving honestly because many others are honest themselves, and are therefore likely to police corrupt behavior. In this case, violations of the integrity norm trigger social shame and feelings of guilt, while honest individuals are lauded as positive role models instead of ridiculed as suckers. These social rewards and sanctions lead to the following second hypothesis.

Hypothesis 2. Independently of the material costs and benefits, the likelihood that an individual behaves honestly is directly proportional to his belief that his peers are also honest.

Figure 1.

Tipping game for corruption decision.

The tipping game neatly captures an important social dilemma: while better collective outcomes can be achieved by coordinating on H, people face strong individual incentives to engage in corruption when all others are doing so. Figure 1 thus illustrates how a ‘culture of corruption’ (represented by C) is sustained over time: individuals find it irrational to change their behavior unless they can be sure that others will switch as well; but since everyone believes that ‘society’ is corrupt, everyone behaves in a way that confirms these collective priors. Thus we can see how a so-called cultural predisposition towards corruption, while appearing natural and permanent ex post, is actually the product of complex inter-dependent choices ex ante.

By opening up the box of cultural (re)production, the tipping game also sheds light on the dynamics of cultural change. Suppose that at time t, society is at the all-corrupt equilibrium C. Now consider an individual i who (for whatever reason) believes that the population of honest people lies to the right of $τ$ . Given this belief, i will choose to act honestly (represented by point h), and thereby shift society closer to $τ$ at time t+1. However, once i realizes that h < C, this shift is not sustained in equilibrium, and society eventually falls back to C.

But suppose instead that, in addition to i, individuals j through N also believe that society lies to the right of $τ$ , and N > $τ$ . In this case, each individual’s best course of action is to choose honesty over corruption, and since N individuals make this choice simultaneously, initial beliefs become self-fulfilling. As a consequence, cultural practice ‘tips’ from C to H.

The key to precipitating such a cultural shift lies in coordinating a collective revision of beliefs about the likely behavior of others in society. In what follows, I consider the possibility that a cultural tip may be set off by mixing two populations (northern and southern Italians) with different reputations for honesty. In particular, southern Italians are often stereotyped as amoral and opportunistic (Banfield, 1958), while their northern compatriots are perceived as more virtuous and civic-minded. Furthermore, public opinion data show that these characterizations do not merely embody northern prejudice, but actually reflect how southerners think about themselves. As shown in Table 1, when locally-representative samples of Italians are asked to rank how often other citizens in their own region engage in bribery, the average standardized score for southern regions is $- 1.04$ , as opposed to $+ 0.36$ for northern regions (with positive scores denoting greater honesty).

Table 1.

Perceptions of corruption across Italian regions.

Northern regions	Bribe	Central regions	Bribe	Southern regions	Bribe
Bolzano	0.89	Marche	−0.04	Abruzzo	−0.44
Trentino Alto-Adige	0.87	Umbria	−0.16	Basilicata	−0.88
Friuli-Venezia-Giulia	0.83	Lazio	−0.54	Molise	−0.9
Valle d’Aosta	0.55	Average	−0.25	Sicilia	−1.05
Toscana	0.42			Puglia	−1.18
Veneto	0.37			Calabria	−1.21
Emilia-Romagna	0.07			Campania	−1.63
Piemonte	0.02			Average	−1.04
Liguria	−0.09
Lombardia	−0.31
Average	0.36

Data from the 2013 European Quality of Governance Index (Charron et al., 2013b, 2015). Scores are based on survey responses about the amount of perceived bribery committed by others in respondents’ area. All ratings have been standardized.

In light of these stereotypes, it may be possible to change beliefs about whether ‘society’ lies to the right or left of $τ$ by manipulating the proportions of northerners and southerners in a group. Furthermore, if the choice between corruption and honesty is conditional upon beliefs about the likely behavior of others, then as the probability of interacting with southerners increases, so should the propensity to engage in illicit behavior. This leads us to our final two hypotheses.

Hypothesis 3. Informing Italians that they are interacting with individuals from the south strengthens the belief that others are dishonest.

Hypothesis 4. Informing Italians that they are interacting with individuals from the south increases the likelihood of corrupt behavior.

In the following section, I describe an experiment designed to test these four hypotheses.

Methodology

Experiments have gained increasing currency in applied social science as a tool to help researchers understand the dynamics underlying individual decision-making. They have proven particularly useful when real-world data are scarce or difficult to observe directly. For this reason, experiments represent valuable research tools in the study of corruption, as such behavior is usually hidden due to its illicit nature (Abbink and Serra, 2012; Armantier and Boly, 2013). Secondly, experiments can help us to disentangle multiple mechanisms that produce equivalent results in observational data. As noted in the previous section, both economic incentives and social rewards and punishments are frequency dependent: by simply observing that individuals are more prone to engage in corruption in highly corrupt environments, we cannot be sure which mechanism is at work. Scholars employing experimental methods can gain analytical leverage over these issues by holding economic incentives constant, thereby isolating the effects of normative constraints. Finally, experimental techniques can directly illustrate the mechanisms represented in Figure 1 by measuring agents’ beliefs, and mapping such beliefs to individual actions.

Description of experimental tasks

The experiment described in this section embeds participants in a framed hospital setting where they are given the opportunity to offer and accept bribes, and also to report corrupt behavior on the part of other individuals.⁵ In each session of the corruption game, a maximum of 24 participants are randomly assigned to one of three roles: {nurse, early patient (EP), late patient (LP)}. Both EPs and LPs begin each round with an initial endowment of 32 tokens. Players in both Patient roles are told to imagine that they are waiting in line at a public hospital to see the doctor. Waiting is costly for Patients, but because EPs have arrived early at the hospital, they face a very short wait, and therefore maintain their initial endowment. However, LPs face a very long wait, and will lose 16 tokens as a result. LPs therefore retain $32 - 16 = 16$ tokens in the round.

However, LPs also have the possibility of skipping the line and avoiding the loss of 16 tokens. Each LP is randomly matched to one nurse, and has the option of offering this nurse a ‘gift’ worth six tokens, in exchange for being allowed to skip the line. In other words, if the LP makes an offer and the nurse accepts, the LP jumps the queue and earns $32 - 6 = 26$ tokens. Although in this case the LP pays the cost of the gift, he is much better off at the front of the queue (with 26 tokens) than at the back (with 16 tokens). However, as a consequence of the exchange between the LP and the nurse, all of the EPs must wait a little longer to see the doctor and, as a consequence, each suffers a small loss of three tokens.

Notice that offering a gift is a dominant strategy in the game. If the LP chooses not to offer a gift, he is sure to lose 16 tokens. The same result obtains if the LP offers a gift, but his offer is rejected by the nurse (however, having your offer rejected carries no additional cost). On the other hand, the LP’s earnings increase from 16 to 26 if his offer is accepted. Therefore, LPs can only gain by offering.

For their part, nurses must decide whether to accept gifts from patients. Nurses begin the game with 24 tokens, and retain this amount if they are offered no gifts, or if they refuse to accept LP offers. On the other hand, if they decide to accept gifts, and in exchange permit LPs to skip the line, they will receive six additional tokens for each gift they receive. Because LPs and nurses are randomly matched, it is possible that nurses interact with more than one LP in the round, and also that nurses interact with no LPs. However, in either case, a nurse must choose what he/she would do in case a gift is offered. In this way, I can elicit players’ full strategies while allowing them to make decisions simultaneously rather than sequentially.

The decisions that can be taken by LPs and nurses (and the effect of these decisions on all players’ earnings) are shown in Figure 2.⁶ Note that although skipping the line is mutually beneficial to both nurses and LPs, it is harmful to ‘society’. This is because there are eight EPs in each round, and therefore EPs as a group lose 3 × 8 = 24 tokens for each corrupt transaction, while the total gain for the other players is only 16 (10 for the LP and six for the nurse).

Figure 2.

Screenshots of LP and nurse decisions.

Another point to note is that, unlike LPs, for whom corruption is a dominant strategy, nurses must make their decisions in the shadow of punishment. In particular, the game contains an element of monitoring and sanctioning, implemented through the role of EPs. EPs are randomly matched to one nurse, and are told that they can ‘overhear’ conversations between this nurse and other patients, so that they will know if the nurse has exchanged any favors. EPs then have the option of reporting the nurse to the hospital administration, which may or may not act on the report (see Figure 3).

Figure 3.

Screenshot of the reporting decision.

The report is effective 50% of the time, in which case the nurse forfeits any gifts he has received, and also pays a fine of nine tokens, so that he retains only $24 - 9 = 15$ tokens at the end of the round. But the other 50% of the time, the report is ignored and the nurse is not sanctioned.⁷ However, no matter what happens to the nurse, reporting is costly to the EP, who must pay a reporting ‘fee’ of three tokens. In fact, from the EP’s perspective, reporting is never rational in a one-shot game: even if the nurse is punished, LPs who have skipped the line remain at the front. In other words, EPs still lose tokens for every LP that skips the line, irrespective of their own actions. If participants played as perfectly rational payoff maximizers, the one-shot game has a single equilibrium outcome: all LPs offer gifts, no EPs report corrupt exchanges and all nurses accept offers.

The corruption game is played for three rounds. In each round, players are randomly assigned to a (new) role, so that by the end of the game players have had the chance to play in all three roles. Players are only shown the results of each round at the end of the session. This feature, coupled with the fact that players are randomly rematched in each period, means that behavior in the game should not be influenced by concerns about reputation or retaliation. Finally, since the rules are somewhat complex and the payoffs hard to keep track of, at the moment they make their decision, players are shown a summary of their choices and the possible consequences for themselves and for other participants. In general, self-reported understanding of the game was excellent.⁸

Next, participants take part in an incentivized belief elicitation task to measure their perceptions of the relevant injunctive norms prevailing in the corruption game, as well as the extent of compliance with these norms. In particular, participants are first asked whether they personally AGREE with, DISAGREE with or have NO OPINION about the following three statements.

‘A participant in the role of LP should not offer gifts’.

‘A participant in the role of nurse should refuse gifts that are offered’.

‘A participant in the role of EP should report an exchange of favors’.

Next, participants are asked to guess the total number of individuals in their session who:

Q1: chose OFFER in the corruption game;

Q2: chose ACCEPT in the corruption game; and

Q3: chose REPORT in the corruption game.

Finally, participants are asked to guess (separately for each question pair) the total number of individuals in their session who:

Q4a and Q4b: AGREE and DISAGREE with the statement: ‘A participant in the role of LP should not offer gifts’;

Q5a and Q5b: AGREE and DISAGREE with the statement: ‘A participant in the role of nurse should refuse gifts that are offered’; and

Q6a and Q6b: AGREE and DISAGREE with the statement: ‘A participant in the role of EP should report an exchange of favors’.

In all, players make a total of nine guesses, and earn tokens based on the accuracy of their answers.

Questions Q1–Q3 are direct measures of the degree of normative compliance; however, operationalizing the injunctive norm itself is more complicated. First, answers to the separate AGREE and DISAGREE segments from Q4–Q6 are converted into percentages by dividing by the total number of participants in each session. Next, the DISAGREE segment is subtracted from the AGREE segment, yielding a variable that captures how strongly any particular individual believes that an action would meet with social disapproval. For example, if I answered that 30% of people agree with the statement ‘A nurse should refuse gifts’, and also that 30% disagree, my Q5 score would be $30 - 30 = 0$ , reflecting the fact that I have no strong beliefs about what constitutes socially appropriate behavior for nurses.⁹ On the other hand, if I believe that most people viewed the acceptance of gifts as socially inappropriate, then I would be likely to answer that more people agree with the previous statement than disagree.

After completion of the main experimental tasks, participants take part in a lottery game designed to measure their risk preferences (see the Appendix). Once players have completed all of the tasks, they are shown the results for each part of the experiment, as well as their total earnings. Participants then answer a short debrief questionnaire before receiving payment. In total, each session lasts about one hour and 10 minutes, and participants are paid about €13.50 (US$17.50) for their time.

Participants and treatment manipulation

The experiment was conducted at the University of Bologna’s Laboratory for Experiments in Social Science (BLESS) in March and April 2013. A total of 93 student participants were recruited to one of nine sessions at BLESS.¹⁰ Overall, 52 participants were female (56%) and 41 were male (44%). The average age was slightly over 25 years. Column (1) in Table 1 presents basic descriptive statistics.

Importantly, the University of Bologna enrolls a large student body, and admits students not only from the surrounding region of Emilia-Romagna, but also from across Italy, including many students from the south. Since the geographical composition of the BLESS participant pool is likely to reflect the underlying composition of the student body, I am able to recruit individuals from diverse regions of Italy with varying reputations for honesty. I then test whether providing information based on these reputations can induce a ‘shock’ to collective beliefs, and thereby ‘tip’ cultural practices.

Specifically, immediately prior to the corruption game, participants are shown one of following messages.

Primed: ‘Most participants in the session today come from the south of Italy’.

Neutral: ‘Participants in the session today come from all over Italy’.

The messages attempt to manipulate beliefs about the likely behavior of other participants in the corruption game by activating stereotypes that ‘All southerners are corrupt’.

In order to avoid deceiving participants, the manipulation of the messages could only be accomplished by actually controlling the composition of each session. Consequently, invitations to potential participants were sent on the basis of their region of origin. In three sessions, only individuals from southern regions were invited to take part, and these participants saw the primed message.¹¹ In six sessions, recruitment was open to the entire BLESS participant pool, meaning that each session contained a mix of southern and northern participants. All of these individuals were exposed to the neutral message. In total, as shown in Table 2, 35 participants came from the north, and 58 participants were from the south. Of these 58 individuals, 26 received the neutral message, and 32 received the primed message. There are no statistically significant differences in individual-level characteristics across the three groups, apart from the obvious fact that northerners tend to have lived in Bologna for a longer period of time.

Table 2.

Summary statistics and covariate balance.

	Total	North-neutral	South-neutral	South-primed
	(1)	(2)	(3)	(4)
Male	0.441	0.514	0.308	0.469
	(0.499)	(0.507)	(0.471)	(0.507)
Age	25.161	25.629	24.808	24.938
	(3.896)	(4.023)	(3.007)	(4.428)
Years in Bologna	7.086	10.800*	4.731	4.938
	(7.100)	(9.728)	(2.523)	(3.959)
Previous participation in experiments	0.667	0.714	0.615	0.656
Previous participation in experiments	(0.474)	(0.458)	(0.496)	(0.483)
Risk preference	3.753	3.800	3.654	3.781
	(1.487)	(1.511)	(1.623)	(1.385)
N	93	35	26	32

Note: Means with standard deviations in parentheses. Asterisks indicate whether differences in sample means across treatment groups are statistically significant in t-tests. *p-value < 0.05 (adjusted p-values).

In summary, the experiment compares the behavior of a cultural group (southern Italians) under two conditions: a primed condition where individuals believe that they are interacting with other southerners (who have a reputation for dishonesty), versus a neutral condition where participants believe that they are interacting with an ‘average’ Italian. The expectation is that, given common knowledge stereotypes, we should see an increase in corruption levels in the primed condition, as well as stronger beliefs that other participants are also dishonest. Finally, the inclusion of northerners in the experiment also allows us to compare north–south behavioral differences, and provides an opportunity to examine the extent to which the stereotype that ‘southerners are corrupt’ holds in reality.

Results

Overall, as shown in Figure 4, 73% of LPs offered gifts, 34% of nurses accepted and 45% of EPs reported corrupt exchanges. Clearly, outcomes in the corruption game differ from what we would expect under the assumption that players are completely rational payoff maximizers. Can we attribute this divergence to concerns about complying with anti-corruption norms? If so, how do normative beliefs and behavior change between the message treatments?

Figure 4.

Differences in behavior between northerners and southerners: neutral message.

To address these questions, I first consider behavioral differences between northerners and southerners in the mixed sessions. This will allow me to build a case for the validity of common stereotypes. Next, I examine the extent to which behavior tracks social beliefs and expectations as predicted by the tipping model. Finally, I compare the play of southerners in the mixed sessions who received the neutral message to southerners who received the primed message.

Baseline north–south differences

Figure 4 shows the differences in behavior between northerners and southerners receiving the geographically neutral message. I find that there are no statistically significant differences in terms of how participants play the patients’ roles: LPs offer gifts (Z-test p-value = 0.630) and EPs report exchanges of gifts (Z-test p-value = 0.791) at roughly the same rates; however, a large and statistically significant north–south difference emerges in the behavior of nurses: southerners are almost twice as likely to accept gifts as northerners (Z-test p-value = 0.046).

These results give us some confidence that observations and perceptions about the level of corruption in real-world groups can be replicated, at least partially, in the lab (see Barr and Serra, 2010). Moreover, the fact that a north–south gap only appears in the behavior of nurses may reflect a specific definition of corruption in the Italian context: simply, it may not be considered an act of corruption for an individual in his private capacity to offer a gift in order to circumvent the formal rules, but it would be corrupt and improper for a person in a position of public responsibility (in this case, the nurse) to accept one. As one participant stated in a debrief questionnaire:

It is normal that LPs try to jump the line by offering gifts to nurses, but it is then up to the nurse to do the right thing for the other patients… (my emphasis)

This distinction between appropriate public and private responsibilities highlights the fact that a corrupt transaction is a two-sided affair, and its success depends upon the willingness of the bribe-taker to circumvent official rules. The south’s ‘culture of corruption’, and the north’s ‘civic culture’, may in fact characterize only a narrowly professional, rather than a widespread popular, ethos.¹²

Social appropriateness and illicit behavior

To what extent is the behavior of nurses linked to individuals’ perceptions of the social appropriateness of accepting a gift? We can address this question using data from the belief elicitation task. As noted above, cross-national survey evidence has shown that people in almost all societies view corruption as wrong. This result is replicated amongst the experimental participants. Figure 5 shows the distribution of answers to ${Belief}^{Disapprove}$ , which measures the difference in the participants’ guesses about the number of individuals who agreed versus disagreed with the statement ‘A participant in the role of nurse should refuse gifts that are offered’ (converted into percentage terms). We see that the vast majority of participants (more than 80%) score > 0 on this question, meaning they understand bribe-taking to violate injunctive norms.

Figure 5.

Distribution of beliefs about the social acceptability of corruption: neutral message.

However, is knowledge of injunctive norms alone sufficient to ensure honest behavior? Column (1) in Table 3 presents estimation results from the following basic logit model

{Accept}_{i} = β_{1} {Belief}_{i}^{Disapprove} + ε_{i}

where ${Accept}_{i}$ is a dummy outcome variable denoting whether participant i accepted a gift, ${Belief}_{i}^{Disapprove}$ is a measure of i’s understanding that bribe-taking violates an injunctive norm, and $ε_{i}$ represents an individual-specific error term. We see that the coefficient, while correctly signed, is not significant at traditional levels.

Table 3.

Logit results: Beliefs and nurses’ behavior.

	Dependent variable: Accepting a gift
	(1)	(2)	(3)	(4)
Belief^Disapprove	−1.407		−1.780
	(0.129)		(0.089)
Belief^Accept		8.423 ^***		8.864^***
		(0.000)		(0.000)
Male			−0.199	0.347
			(0.788)	(0.715)
Age			−0.0950	−0.0195
			(0.350)	(0.893)
Lotto			1.056	1.099
			(0.149)	(0.113)
Constant	0.116	−6.003^***	2.525	−6.150
	(0.727)	(0.000)	(0.321)	(0.079)
Observations	61	61	61	61
Pseudo R²	0.049	0.275	0.088	0.302

All standard errors clustered within sessions. p-values (in parentheses).

^* p < 0.05, ^** p < 0.01, ^*** p < 0.001.

In column (2), I estimate instead

{Accept}_{i} = β_{1} {Belief}_{i}^{Accept} + ε_{i}

where ${Belief}^{Accept}$ measures a participant’s best guess about the percentage of other participants who accepted a gift. We see now that the coefficient of interest is highly significant. Since smaller values of ${Belief}^{Accept}$ represent expectations of greater honesty, the positive association with ${Accept}_{i}$ indicates that individuals will be more likely to engage in corruption themselves if they believe others are doing likewise.

This result is exactly what we would predict from Hypotheses 1 and 2, which hold that even though corruption is understood to be wrong, illicit behavior in the context of widespread normative violations are likely to generate little social disapproval. Rather, individuals behave as conditional norm compliers, choosing honesty only when they believe that others are also honest. As shown in columns (3) and (4), these results are robust to the inclusion of individual-level controls for gender, age and risk-preferences as measured in the lottery task.

The relationship between an individual’s own behavior and her expectations concerning others’ behavior is also substantively meaningful: on average, participants believe that 61% of their counterparts would accept gifts (SD = 22%). Moving from half a standard deviation below this mean to half a standard deviation above the mean is associated with a 37% increase in the likelihood that any individual will accept a gift. This is a large effect, considering that the average gift-acceptance rate is only 33%.

In summary, we see from Table 3 that it is not so much an understanding that ‘one ought not to be corrupt’ that predicts honest behavior, but rather a belief that few others are corrupt. Figure 6 presents a graphical illustration of the relationship between social beliefs and individual corrupt behavior. The key point to note is that amongst those participants who understood accepting a gift to be wrong (represented by points on the righthand side of the plot), many were nonetheless willing to take bribes if they expected illicit behavior on the part of others (represented by dots in the upper-righthand corner of the plot).

Figure 6.

Social beliefs, others’ actions and own behavior: neutral message.

Messaging effects

So far, we have seen that behavior in the corruption game follows the predictions of the tipping model. Next, I investigate whether a message about the proportion of southerners in the room can tip behavior even further towards the all-corrupt equilibrium. In particular, given the common knowledge belief that ‘southerners are corrupt,’ exposure to the primed message should lead to an increase in the perceived social acceptability of corruption, and also to a higher level of gift-acceptance in the corruption game.

Figure 7 shows the differences in nurses’ behavior between northerners, southerners who received a neutral message, and southerners under the southern prime. Surprisingly, we see that the likelihood of accepting a gift actually falls when southern participants are exposed to the primed message: acceptance rates drop dramatically from 54% to 25% (Z-test p-value = 0.024). In fact, the behavior of southerners under the prime is statistically indistinguishable from northerners (Z-test p-value = 0.741).

Figure 7.

Nurses’ behavior: acceptance of gifts across groups.

Moreover, as shown in Figure 8, individuals exposed to the southern prime are also more likely to believe that their counterparts are more honest, compared to southerners in the neutral message condition. A Wilcoxon rank-sum test confirms that this difference is statistically significant around the 5% level (p-value = 0.054). In summary, it appears that exposure to the primed message does indeed alter collective beliefs and ‘tip’ group behavior, but these effects go in the opposite direction to that predicted by Hypotheses 3 and 4. In the next section, I present several conjectures that may account for these counter-intuitive results.

Figure 8.

Change in ${Belief}_{i}^{Accept}$ across message conditions.

Discussion

What social mechanisms may produce the surprising results we observe in response to the primed message that ‘Most participants are from the south of Italy?’. Here, I consider three possible explanations: (i) selection, (ii) in-group favoritism and (iii) social stigma. First, it may be the case that the southerners in this experiment constituted a self-selected sample: perhaps only southerners who are frustrated with corruption in their home regions migrate to the north. In response to the southern prime, these individuals then become especially scrupulous in their behavior. However, the fact that southern participants in the non-primed sessions behaved more corruptly than any group in the experiment suggests that selection is unlikely to drive the messaging effects we observe.

A second possibility is that the messaging effects are attributable to in-group favoritism.¹³ Recall that the structure of the corruption game represents a social dilemma: bribe-taking is individually beneficial, but harmful to the group as a whole. Thus, when informed that they are interacting with other members of their in-group, southern participants may be more willing to engage in pro-social behavior (here: not taking a bribe) as compared to a situation where they are interacting with ‘anonymous’ Italians. Furthermore, southerners’ positive feelings towards their fellow southerners may also explain why ${Belief}^{Accept}$ shifts in the direction of greater honesty between the neutral and primed conditions.

A third possibility may be that southerners, especially those living in the north of Italy, are all too aware that they are looked down upon by their compatriots, in part because of their reputation for dishonesty. Such stigmatization may become especially salient in the context of a framed corruption game in which southerners are shown an explicit message priming their southern identity. Consequently, southern participants may attempt to rebel against these common stereotypes, thereby trading off small monetary losses against significant gains in personal pride, as well as the possibility of upgrading their social status.¹⁴ In other words, individuals may be willing to pay six or more tokens in order to state a defiant belief (‘We are not all corrupt!’) from which consistent behavior (‘…and I’ll be honest to prove it to you!’) naturally follows. Such a dynamic may explain why we observe a shift in both the probability of accepting a gift and in ${Belief}^{Accept}$ between the neutral and primed conditions.

External evidence consistent with this interpretation of the laboratory results can be found in a historical case study of anti-corruption reform in early 20th century Boston. At the time, the proliferation of electoral corruption was largely blamed on the voting patterns of immigrants, poorly schooled in the responsible practice of democracy, whom many believed were firmly in the pockets of corrupt machine bosses (Brownell, 1983; Cornwell, 1964; Huthmacher, 1962; Reid and Kurth, 1992). In order to escape this social stigma and prove themselves the political equals of Boston’s native-born Anglo elite, the city’s Irish leaders organized their communities in a political fight against patronage, bosses and party machines (Connolly, 2009). To a large extent, this effort was successful in shifting the Irish vote (particularly amongst the middle-classes) in favor of progressive municipal reforms.

Drawing from this experience, it is plausible that a similar reaction against social stigma may explain our laboratory results. That said, I currently do not have the data to conclusively distinguish the stigmatization hypothesis from in-group favoritism. I would argue, however, that pro-social in-group feelings are unlikely to account for the experimental results. While norms of in-group solidarity certainly prevail in the south of Italy, they exist primarily within closely-knit personal circles (Banfield, 1958). In fact, southerners are as likely to engage in rivalry with people from other southern regions as they are with northerners, suggesting that no special pro-social feelings are generated by simply knowing that others are from the south. On the other hand, since negative stereotypes apply to all southern regions, the uniform reaction of southerners is more consistent with the social stigma hypothesis. A formal test to disentangle these two explanations, however, must unfortunately await a future date.

Conclusion

The general failure of formal, institutional-design-based approaches to reduce corruption has engendered an updated understanding of corruption as primarily a cultural phenomenon (Andrews, 2013). This article, through the lens of a tipping model, provides a new perspective on the processes underlying cultural stability and change. Rather than painting corrupt practices as legitimate components of specific cultural repertoires, formal equilibrium theory understands cultures of corruption and honesty as the products of conditional choices, formulated in the shadow of shared beliefs about the likely behavior of others. This approach implies that cultures are not static, immutable, or hard-coded. Instead, when the collective beliefs of a society change, that society may tip towards a new set of culturally-prescribed behaviors.

The experiments reported in this paper provide support for the behavioral mechanisms underlying this tipping model. They also show that attempts to revise collective beliefs can produce some surprising results: when informed that they are interacting with other southern Italians, southern participants are not only more likely to believe that their counterparts are honest, but they are also more likely to behave honestly themselves. I have discussed three possibilities that may explain these counter-intuitive findings, although clearly more work needs to be done to understand the mechanisms at work. Nevertheless, the fact that we can reduce levels of malfeasance by changing beliefs suggests some exciting new possibilities for tackling entrenched cultures of corruption.

Footnotes

Appendix

Acknowledgements

Many thanks to Giulia Andrighetto, Maria Bigoni, Stefania Bortolotti, Marco Casari, Diego Gambetta, Johanna Gereke, Judith Goldstein, Stephen Krasner, David Laitin, Hayagreeva Rao and participants at the SONIC, BEELab and UPenn PPE meetings for helpful suggestions and comments. I also thank Gianandrea Lanzara and Stefano Rizzo for outstanding research assistance. Finally, I gratefully acknowledge the generosity of Professor Marco Casari in allowing me access to the Bologna Laboratory for Experiments in Social Science.

Funding

This work was supported by the Stanford University Vice Provost for Graduate Education’s Graduate Research Opportunity (GRO) Fund and the Stanford University Freeman Spogli Institute (FSI) Global Underdevelopment Action Fund.

Notes

References

Abbink

Serra

(2012) Anticorruption policies: Lessons from the lab. In: Serra

Wantchekon

(eds) New Advances in Experimental Research on Corruption. Bingley: Emerald Group Publishing, pp.77–115.

Andrews

(2013) The Limits of Institutional Reform in Development. New York: Cambridge University Press.

Andvig

Moene

(1990) How corruption may corrupt. Journal of Economic Behavior & Organization 13(1): 63–76.

Ariely

(2013) The Honest Truth about Dishonesty: How We Lie to Everyone–Especially Ourselves. New York: Harper.

Armantier

Boly

(2013) Comparing corruption in the lab and in the field in Burkina Faso and in Canada. The Economic Journal 123(573): 1168–1187.

Banfield

(1958) The Moral Basis of a Backward Society. Glencoe: Free Press.

Bardhan

(1997) Corruption and development: A review of issues. Journal of Economic Literature 35(3): 1320–1346.

Barr

Serra

(2010) Corruption and culture: An experimental analysis. Journal of Public Economics 94(11): 862–869.

Bicchieri

(2005) The Grammar of Society: The Nature and Dynamics of Social Norms. New York: Cambridge University Press.

10.

Bicchieri

Xiao

(2009) Do the right thing: But only if others do so. Journal of Behavioral Decision Making 22(2): 191–208.

11.

Brownell

(1983) Interpretations of twentieth-century urban progressive reform. In: Colburn

Pozzetta

(eds) Reform and Reformers in the Progressive Era. Westport, CT: Greenwood Press.

12.

Cadot

(1987) Corruption as a gamble. Journal of Public Economics 33(2): 223–244.

13.

Charron

Dijkstra

Lapuente

(2013) Regional governance matters: Quality of government within European Union member states. Regional Studies 48(1): 68–90.

14.

Charron

Dijkstra

Lapuente

(2015) Mapping the regional divide in europe: A measure for assessing quality of government in 206 European regions. Social Indicators Research 122(2): 315–346.

15.

Cialdini

Reno

Kallgren

(1990) A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places. Journal of Personality and Social Psychology 58(6): 1015–1026.

16.

Connolly

(2009) The Triumph of Ethnic Progressivism: Urban Political Culture in Boston, 1900–1925. Cambridge, MA: Harvard University Press.

17.

Cornwell

(1964) Bosses, machines, and ethnic groups. The Annals of the American Academy of Political and Social Science 353(1): 27–39.

18.

Della Porta

Vannucci

(1999) Corrupt Exchanges: Actors, Resources, and Mechanisms of Political Corruption. New York: Aldine de Gruyter.

19.

Eckel

Grossman

(2002) Sex differences and statistical stereotyping in attitudes toward financial risk. Evolution and Human Behavior 23(4): 281–295.

20.

Eckel

Grossman

(2008) Forecasting risk attitudes: An experimental study using actual and forecast gamble choices. Journal of Economic Behavior & Organization 68(1): 1–17.

21.

Ekeh

(1975) Colonialism and the two publics in Africa: A theoretical statement. Comparative Studies in Society and History 17(1): 91–112.

22.

Elster

(1989) Nuts and Bolts for the Social Sciences. New York: Cambridge University Press.

23.

Englebert

(2009) Africa: Unity, Sovereignty, and Sorrow. Boulder, CO: Lynne Rienner Publishers.

24.

Fehr

Fischbacher

(2005) The economics of strong reciprocity. In: Gintis

Bowles

Fehr

(eds) Moral Sentiments and Material Interests. The Foundations for Cooperation in Economic Life. Cambridge, MA: The MIT Press, pp.151–193.

25.

Fischbacher

(2007) z-tree: Zurich toolbox for ready-made economic experiments. Experimental Economics 10(2): 171–178.

26.

Fisman

Miguel

(2007) Corruption, norms, and legal enforcement: Evidence from diplomatic parking tickets. Journal of Political Economy 115(6): 1020–1048.

27.

Gambetta

(2013) Why is italy so disproportionately corrupt? A conjecture. Keynote Address delivered at the second Anticorrp General Meeting: European University Institute, San Domenico di Fiesole, Italy, 8 May 2013.

28.

Golden

Chang

(2001) Competitive corruption: Factional conflict and political malfeasance in postwar Italian christian democracy. World Politics 53: 588–622.

29.

Golden

Picci

(2005) Proposal for a new measure of corruption, illustrated with Italian data. Economics & Politics 17(1): 37–75.

30.

Greiner

(2004) The online recruitment system ORSEE 2.0 – A guide for the organization of experiments in economics. University of Cologne Working Paper Series in Economics 10(23): 63–104.

31.

Heckathorn

(1989) Collective action and the second-order free-rider problem. Rationality and Society 1(1): 78–100.

32.

Husted

(1999) Wealth, culture, and corruption. Journal of International Business Studies 30(2): 339–359.

33.

Huthmacher

(1962) Urban liberalism and the age of reform. The Mississippi Valley Historical Review 49(2): 231–241.

34.

Klitgaard

(1988) Controlling Corruption. Berkeley, CA: University of California Press.

35.

Laitin

(1998) Identity in Formation: The Russian-speaking Populations in the Near Abroad. Ithaca, NY: Cornell University Press.

36.

Laitin

(2007) Nations, States, and Violence. New York: Oxford University Press.

37.

Laitin

Zhang

(2012) Political culture. In: Krieger

(ed.) The Oxford Companion to Comparative Politics. Oxford University Press, pp.223–227.

38.

Licht

Goldschmidt

Schwartz

(2007) Culture rules: The foundations of the rule of law and other norms of governance. Journal of Comparative Economics 35(4): 659–688.

39.

Lipset

Lenz

(2000) Corruption, culture, and markets. In: Harrison

Huntington

(eds) Culture Matters: How Values Shape Human Progress. New York: Basic Books.

40.

Lui

(1986) A dynamic model of corruption deterrence. Journal of Public Economics 31(2): 215–236.

41.

Mackie

(1996) Ending footbinding and infibulation: A convention account. American Sociological Review 61(6): 999–1017.

42.

Mazar

Amir

Ariely

(2008) The dishonesty of honest people: A theory of self-concept maintenance. Journal of Marketing Research 45(6): 633–644.

43.

Miller

(2006) Corruption and corruptibility. World Development 34(2): 371–380.

44.

Nannestad

Svendsen

Dinesen

. (2014) Do institutions or culture determine the level of social trust? The natural experiment of migration from non-western to western countries. Journal of Ethnic and Migration Studies 40(4): 544–565.

45.

Olivier De Sardan

(1999) A moral economy of corruption in Africa? The Journal of Modern African Studies 37(1): 25–52.

46.

Persson

Rothstein

Teorell

(2013) Why anticorruption reforms fail: Systemic corruption as a collective action problem. Governance 26(3): 449–471.

47.

Petersen

(1989) Rationality, ethnicity and military enlistment. Social Science Information 28(3): 563–598.

48.

Price

(1974) Politics and culture in contemporary Ghana: The big-man small-boy syndrome. Journal of African Studies 1(2): 173–204.

49.

Putnam

(1993) Making Democracy Work: Civic Traditions in Modern Italy. Princeton, NJ: Princeton University Press.

50.

Reid

Kurth

(1992) The rise and fall of urban political patronage machines. In: Goldin

Rockoff

(eds) Strategic Factors in Nineteenth Century American Economic History: A Volume to Honor Robert W. Fogel. Chicago: University of Chicago Press, pp.427–445.

51.

Rose-Ackerman

(1999) Corruption and Government: Causes, Consequences, and Reform. New York: Cambridge University Press.

52.

Rothstein

Torsello

(2014) Bribery in pre-industrial societies: Understanding the universalism-particularism puzzle. Journal of Anthropological Research 70(2): 263–288.

53.

Schelling

(1978) Micromotives and Macrobehavior. New York: Norton.

54.

Sunstein

(1996) Social norms and social roles. Columbia Law Review 96(4): 903–968.