Stable Randomization

We design a laboratory experiment to identify whether randomization behavior represents a stable "type" across different choice environments. In both games and individual choice questions, subjects face twenty simultaneous repetitions of the same choice. Randomization constitutes making different choices across the twenty repetitions. We find that randomization preferences are highly correlated across domains, with a sizable fraction of individuals randomizing in all domains, even in questions that offer a first-order stochastically dominant option. For some mixers, dominated randomization is responsive to intervention. Our results are inconsistent with many preference-based models of randomization, leaving open a role for heuristics and biases.