An experimental study of prisoners' dilemma and stag hunt games played by teams of players

We compare behavior in six variations of prisoners' dilemma and stag hunt games played by 5-member teams to behavior in the same games played by individuals. The experimental design is informed by a theory of team equilibrium that generates hypotheses about systematic differences between teams and individuals in these games. We also compare three different team collective choice procedures: majority rule; majority rule preceded by a poll; and majority rule preceded by chat. In all six games, we observe significant bandwagon effects that drive consensus in the poll and chat treatments, but voting procedures have no significant effects on team decision frequencies except for the poll treatment of the stag hunt games. Surprisingly, teams do not always defect more than individuals. In prisoners' dilemma games with higher incentives to cooperate, teams cooperate more than individuals. A one-parameter logit specification of the team equilibrium model provides a good fit to the prisoners' dilemma data across all treatments. Teams coordinate more successfully than individuals in the stag hunt games. Additional data are gathered using Amazon Mechanical Turk, including data for much larger (25-member) teams. The observed MTurk behavior differs quantitatively from laboratory behavior, but with similar qualitative comparative statics.