Many countries around the world have serious corruption problems at the expense of public welfare. An experimental economic study now identifies conditions that encourage leaders to accept bribes instead of sanctioning free-riders. Possible anti-corruption strategies can have positive effects, fail or even backfire.
Humans are champions of cooperation1, far beyond what has been seen in the animal world2: numerous studies suggest reasons and mechanisms for why we should be cooperative. However, every day the newscast uncovers the dark side of human social behaviour — terrorism, civil wars, drug dealing, corruption, and so on. Transparency International3 lists corruption scores of countries around the world (Fig. 1) and concludes that more than 6 billion people live in countries with a serious corruption problem. The costs of corruption reduce the welfare of people, especially in countries where strong leaders gain more from being bribed than from their official income. Corruption can even kill, for instance, when bribing leads to compromising the quality of buildings, which become prone to collapse in earthquakes4. Corruption appears to correlate with institutional, economic, and cultural factors. The causal interconnections among these factors remain, however, largely elusive.
In this issue of Nature Human Behaviour, Muthukrishna, Francois, Pourahmadi, and Henrich5 develop a clever experimental paradigm that includes most of the ingredients that may affect the level of corruption as well as potential anti-corruption devices. They study their causal relationships with anonymous human subjects who have come from countries with varying levels of corruption.
Muthukrishna et al. developed their basic design from a classic public goods game in which 4–7 players can contribute from their endowment. In each period their total contribution to the public pool is multiplied by the pool multiplier (here the economic potential, for example, 1.20), divided by the number of group members and is redistributed equally among all group members, regardless of contribution. The pool multiplier is small enough that it is in every player's best interest to let others contribute, contributing nothing himself. However, the overall gain of the group is highest if everybody contributes maximally. For example, if each of 6 players contributes 10 points, each gains 12 points, therefore, a net gain of 2 points. If one player free-rides, he has a net gain of 10 points because he did not contribute, whereas the net gain of each contributor is 0.
Centralized ‘pool punishment’6 of free-riders after each period helps to maintain contributions to the public pool. From 12 points endowment that each group member receives at the start of each of 10 periods, 2 points are subtracted automatically as tax. One group member randomly chosen before each period is the leader, who behaves as a normal group member, contributes to the pool, and pays 2 points tax. After all players have contributed, the leader is shown each player's decision and the number of tax points available for punishing. For each player, he decides either to do nothing or use tax points to take away points with a multiplier (the leader's punitive power of, for example, 1.5 per tax point used). Unused tax points and points taken away go to the experimenter. The leader and a group member are treated (and can earn) the same, except that the leader is not punished. Each group member is informed about the leader's actions only towards her (not towards other group members) and her own payoff.
The ‘bribery game’ that Muthukrishna et al. developed is the same as the basic game, except that each player can bribe the leader. Now the leader can see both each player's contribution to the pool and also points each player contributed to the leader. For each player he can decide to do nothing, accept the contribution to him, the leader, or punish the player by taking away points. Any points given to the leader and not accepted will be returned to the group member. Group members see only the leader's actions towards them (not towards other group members) and their payoff. Group members might bribe to avoid being punished by the leader. Compared with the basic game, the corruption possibility causes a large (25%) decrease in public good provisioning.
The authors found that those subjects who had grown up in more corrupt countries were more willing to accept bribes. Stronger leaders (having a higher punishment multiplier) were about twice as likely to accept bribes. Thus, leaders accept bribes instead of disciplining defectors, a good basis for testing the effect of anti-corruption strategies.
In the next treatment ‘partial transparency’ is added to the bribery game. Group members not only see the leader's action towards them, but also the leader's own contribution to the group pool. They do not see the leader's actions towards other group members. In the bribery game with ‘full transparency’ the additional information is that each group member can see the leader's decision for each decision made by each of the other group members (contribution to the pool, contribution to the leader and the leader's action).
When applied, corruption mitigation effectively increases contributions (though not to the control level) when leaders are strong or economic potential is rich. When leaders are weak, that is their punitive power is low, and economic potential is poor, the apparent corruption mitigation strategy, full transparency has no effect and partial transparency further decreases contributions to levels lower than that of the standard public goods game with pool punishment and the bribery game with no transparency. Therefore, corruption mitigation strategies sometimes help but can also cause the situation to deteriorate and can backfire.
The results of this study suggest that as economic potential grows, less government intervention is needed to enforce cooperation and increased power may be misused, requiring greater anti-corruption efforts. In contrast, when economic potential is poor, strong government intervention is most effective at decreasing free-riding, as long as this intervention is paired with strategies to mitigate corruption. These findings are important because the potential effect of isolated influences is experimentally shown while keeping the rest of the simulated corruption scenario constant, instead of inferring such effects from big, descriptive real-world datasets. This implies that we know now the actors that facilitate or mitigate corruption and the way they interact.
The authors admit that although these experimental results begin to offer insights into the causal effect of corruption on cooperation, extending such experimental findings demands great caution, as is the case with all results from behavioural economic experiments. In addition, the present study can provide only a proof of principle, as it leaves out the complexity of the full natural situation. Future studies need to add further ingredients of the real situation to the experiment, one at a time, so that the complexity of reality is approached in a stepwise manner — a long way to go but worth the effort.