We study a model of policy experimentation in organizations. Membershave a common objective but differ in their prior beliefs about a risky policy.Current members decide whether to experiment with the policy. Agents in thewider population possess resources of use to the organization and can enterand leave the organization freely, taking their resources with them. We showthat for a wide range of parameters there is too much experimentation in equi-librium relative to the social optimum. The potential change in membershipand control due to experimentation lowers the incentives to experiment. At thesame time, self-selection into the organization plays a countervailing role: whenexperimentation is unsuccessful, only the staunch optimists most supportive ofexperimentation choose to remain. For some parameters this force dominates,yielding equilibria with overexperimentation. We apply the model to decision-making in cooperatives, civil rights organizations and for-profit firms.