Ernst Fehr and Simon Gächter in "Altruistic Punishment in Humans," Nature, Vol. 415, January 10, 2002, pp. 137-140, article examine the effects of altruistic punishment of defectors on human cooperation in a public good game setting. Altruistic punishment means that the punishment is carried out, although it is costly and yields no material gain. The study illustrates that the cooperation rate is high if altruistic punishment is possible, and breaks down if it is not available. These results do not require people being related as in some evolutionary theories, nor they require repeated interactions for possible reputation building as in indirect reciprocity or costly signalling models.

As noted on other pages of this handbook section, the free riding problem causes serious inefficiencies in provision of public goods. However, if there is an opportunity to punish those who free ride on cooperation of others, cooperation becomes a more feasible outcome. Yet, a new problem appears: Who will bear the cost of punishing the free riders? Clearly, everybody will be better off when the free riding is prevented, but nobody has an incentive to punish the free riders, since such action is costly to onself. Thus, the second-order public good problem emerges. It can be solved if high enogh proportion of cooperators have a tendency for altruistic punishment, i.e., if they are willing to punish free riders even if it is costly and yields no direct material benefits.

The authors examined whether subjects engage in altruistic punishment and whether voluntary punishment affects the ability of sustaining cooperation in a public goods experiment with real monetary payoffs. The experiment consisted of two treatments: punishment and no punishment. In both treatments, groups consisting of four members played the following VCM public goods game.

Each player in each group received an endowment of 20 money units (MUs) that he or she could either keep or contribute to a group project. Each MU kept earned exactly 1MU only for the person who kept it, the rest of the group earned nothing. For every MU contributed to a project by an individual, every member of the group earned 0.4MUs, regardless of one's own contribution. Since there are four members in each group, the group return equaled 1.6 MUs, comparing to investor's 0.4 MUs. Thus, it was socially optimal to have all members invest everything. However, from the investor's point of view, it was optimal not to invest at all. Yet, if all members invested all their endowments, each subject would earn 0.4 x 80 = 32 MUs what is more that each person would get in case when all group members kept all MUs privately (20 MUs for each subject). All transactions took place anonymously, subjects were not informed about the composition of their group. The investment decisions were made simultaneously. Once the investments were made, each subject in a the group was informed about the contributions of others.

Punishment Condition
The only difference between the two treatments was that in the punisment condition, subjects could punish other members of the group after they have learned their investments. If a subject decided to punish somebody, she could assign between 0 and 10 points (0 meaning no punishment) to another subject within the group. Each point assigned to the punished player cost 3 MUs and also cost the punishing player 1 MU. Again, all the punishment decisions were made simultaneously. The AB-BA design was applied, that is, a portion of randomly chosen subjects played the punishment treatment first with the no punishment treatment following right after. The rest of subjects played the two treatments in a reversed order.

The subjects played six rounds of the above described public goods game with composition of group changing after every round (no subject met another subject more than once) to eliminate the possible motives of direct reciprocity and reputation. Since the punishment and cooperation were costly and yield no monetary benefits, the subjects with self-regarding preferences would never cooperate nor punish. Hence, the selfish motives linked to indirect reciprocity and costly signalling as an explanation of cooperation and punishment can also be ruled out.

A total of 240 subjects participated in ten experimental sessions. Altruistic punishment was quite frequent: subjects punished other group members 1270 times. 84.3% of the subjects punished at least once during the six rounds they participated in, 34.3% punished more than five times, and 9.3% punished more than ten times. As the authors note, "punishment also followed a clear pattern. Most (74.2%) acts of punishment were imposed on defectors (that is, below-average contributors) and were executed by cooperators (that is, above-average contributors), and punishment of the defectors was harsh. For example, if a subject invested 14-20 MUs less than the average investment of the other group members during periods 5 and 6, the total group expenditures for punishing this subject were almost 10 MUs. Moreover, the more a subject's investment fell short of the average investment of the other three group members, the more the subject was punished. The pattern and strength of punishment was also stable across time." (pp. 137-138).

The altruisic punishment served its purpose because the cooperation level increased in treatment when the punishment was available and declined when there was no opportunity to discipline the non-cooperators. The punishment of free riders substantially increased the contributions towards the public good. For example, in the five sessions that started with the punishment treatment, the average investment of 94.2% subjects was higher when punishment was an option. In fact, the average investment in this case was higher in each session and in each period than the average investment in any periods and sessions in no-punishment treatment. Also the cooperation increased overtime in the punishment treatment and decreased rapidly when the punishment was not available. For comparison, in the final round of the punisment treatment, 38.9% of the subjects contributed their whole endowments and 77.8% contributed 15 MUs or more. Contrary, in the last round of the no-punishment treatment, 58.9% subjects kept the whole endowment (that is contributed zero) and 75.6% contributed 5 MUs or less. Similar pattern was observed for the portion of subjects who participated in the no-punishment treatment first. In the punishment treatment, 91.4% subjects invested more than in the no-punishment treatment. Again, the contributions decreased overtime when the punishment was not available, but increased rapidly in the punishment treatment. Because of the patterns the punishment was executed, one can also conclude that it was not only the punishment opportunity, but actually the punishment itself that raised the level of cooperation.

The findings suggest that the lack of cooperation and free riding on others causes negative emotions, such as anger and dissapointment, and is also anticipated by the subjects. Such emotions then trigger altruistic punishment. The above described results are consistent with such emotional pattern for the following reasons:

1. If negative emotions trigger punishment, one would expect the above-average contributors to execute punishment on the below-average contributors. In the experiment, 74.2% of all punishments are of this nature.
   
   
2. Punishment is stronger for below-average contributors the bigger the difference in contributions from average because negative emotions become more intense.
   
   
3. "If negative emotions cause punishment, the punishment threat is rendered immediately credible because most people are well aware that they trigger strong negative emotions when they free ride." (pp. 139) That is exactly what can be noted from the results - immediate impact of the possibility if punishment.