Non-Random Randomization

We provide a unifying experimental framework in which to study randomization behavior in games and individual choice questions. In each decision, subjects face twenty simultaneous repetitions of the same choice, whereby randomizing constitutes making different choices across the twenty repetitions. We find very high rates of randomization, even in questions that offer a first-order stochastically dominant option. Randomization is highly correlated across domains, while more individuals randomize in games than in analogous decision problems. Experimental treatments test theories of randomization behavior, ruling out most theoretically-based explanations. Results suggest that dominated randomization stems, in part, from a failure of contingent reasoning.