Strategic Experimentation in the Lab

This paper reports the results of experimental tests of the Nash equilibrium predictions in a one-armed bandit game with information spillover. Players learn the probability that a risky prospect pays by either taking draws from the distribution themselves or observing the outcome of another player’s choice. Our experiment is designed to learn whether players experiment strategically, anticipating the opportunity to free-ride on others’ information and doing so. While error rates exhibit a bias toward under-experimentation, we observe a significant strategic effect. Structural parameter estimates suggest the lack of experimentation observed is due to decision error and somewhat pessimistic priors, rather than risk preferences or probability weighting.