Cooperation may be in everyone's long-term interest, but in any single social interaction each individual may have an incentive to defect from cooperative behavior. This is the essence of social dilemmas. When individuals are paired in an indefinitely long-term relationship, the threat of triggering non-cooperative responses by others to a non-cooperative act may be sufficient to sustain cooperative behavior. However, individuals may live in a community for a long time, frequently interact with various members of that community, but have only short-term, intermittent, relationships with any particular member of that community. In such circumstances, it is not be possible for the victim of an opportunistic act to directly punish the defector. Nevertheless, Kandori has shown that, under certain conditions, a social norm of cooperative behavior can be sustained by the same kind of history-dependent strategies that can sustain cooperative behavior in long-term relationships between fixed pairs of individuals. Kandori's theorem is an extension of the Folk Theorem for infinitely repeated games. Like the Folk Theorem, it only proves that cooperation can be sustained by history-dependent trigger strategies. Whether, and under what conditions a group will develop a social norm of cooperative behavior are the questions this study addresses. Specifically, a behavioral experiment has been designed to investigate the degree to which cooperative behavior in an indefinitely repeated prisoners' dilemma stage game is influenced by both the nature of the matching technology and the character of the information transmitted about the play of one stage game to the players of the next stage game. The study will compare the extent of cooperative behavior observed when fixed pairs play an indefinitely repeated Prisoners' Dilemma stage game with that observed in a fixed group that is anonymously and randomly matched after each stage game. In the random matching condition, the amount of information that players have concerning their opponent's past behavior will also be varied so as to determine whether more information leads to increases in the frequency of cooperative behavior.
The question of what kinds of social interaction promote or inhibit cooperative (competitive) behavior in environments where agents have both incentives to cooperate over time and incentives to opportunistically compete at any particular moment of time is of obvious importance in many settings where cooperative arrangements must be self-enforcing. These include such diverse settings as the design of test-ban treaties, the design of relative performance pay systems, and the administration of anti-trust laws to deal with implicit collusion. Beyond these matters of practical importance, the results of this experiment are also of methodological importance for experimental economists who often use random matching protocols to study learning behavior in strategic situations that are meant to represent one time encounters. .
