We study the effects of different punishment institutions on cooperation in a six-person prisoner’s dilemma game in which actors observe others’ cooperation with some noise (i.e. imperfect public monitoring). In the presence of noise, co-operators will sometimes be mistaken for defectors and punished, and defectors will sometimes be mistaken for co-operators and escape punishment. Both types of mistakes are detrimental for cooperation because cooperation is discouraged and defection is encouraged. By means of a laboratory experiment, we study whether this adverse effect of noise can be mitigated by consensual punishment. The more other group members have to agree on punishing a defector, the less likely will a co-operator be punished by mistake. We compare a punishment institution in which each subject decides individually whether to punish another, with institutions in which punishments are only implemented if subjects reach sufficient consensus that a particular group member should be punished.
The experiment was programmed in z-Tree (Fischbacher, 2007) and conducted at the Experimental Laboratory for Sociology and Economics (ELSE) at Utrecht University. Subjects were recruited through ORSEE (Greiner, 2015). A total of 252 subjects participated in the experiment (38% male, 86% students, 32% economics students, average age of 22.57). The number of subjects per session was either 18 or 24 and a session lasted one hour on average. A participant’s earnings averaged €11, with a minimum of €7 and a maximum of €14. We conducted twelve sessions, six with noise and six without noise. In each session, subjects participated in three sequences of 15 six-person PD games (i.e. 45 periods in total). In each round, every subject was endowed with w= 20 points and could decide whether or not to contribute the entire endowment. The total amount contributed in each round was multiplied with m= 2.4 and equally divided among the six members of a group. The groups of six subjects were disbanded and randomly formed anew after each round (i.e. random matching).
In conditions without noise, we find that cooperation and subjects’ payoffs are higher if more consensus is required before a punishment is implemented. In conditions with noise, cooperation is lower if more consensus is required. Moreover, with noise, subjects’ payoffs are lower under all punishment institutions than in the control condition without punishment opportunities.
