When, and why, do teams benefit from self-selection?

We investigate the effect of team formation and task characteristics on performance in high-stakes team tasks. In two field experiments, randomly assigned teams performed significantly better than self-selected teams in a task that allowed for an unequal work distribution. The effect was reversed if the task required the two team members to contribute more equally. Investigating mechanisms, we observe that teams become more similar in ability and report to cooperate better when team members can choose each other. We show how different levels of skill complementarity across tasks may explain our results: If team performance largely depends on the abilities of one team member, random team assignment may be preferred because it leads to a more equal distribution of skills across teams. However, if both team members' abilities play a significant role in team production, the advantage of random assignment is reduced, and the value of team cooperation increases.