Experimenting with Contests for Experimentation

We report an experimental test of alternative rules in multi-stage innovation contests when success may not be feasible and contestants may learn from each other. Following Halac et al. (forthcoming), the planner can vary the prize allocation rule from Winner-Take-All in which the rst successful innovator receives the entire prize to Shared in which all successful innovators during the contest duration share in the prize. The planner can also vary the information disclosure policy from Public in which at each period, all information about contestants' past successes and failures is publicly available, to Private, in which contestants only know their own histories. In our setting, the theoretically optimal contest design depends on the probability of successful innovation, given that innovation is feasible. Under some parameters the designer will prefer a WTA-Public contest; while, under others he will prefer Shared-Private. Our experiments provide evidence that Private disclosure contests behaviorally dominate Public disclosure, regardless of the prize allocation rule, and moreover that Shared-Private contests dominate WTA-Private contests.