Website CV Paper
Fields
Behavioral & Experimental, Theory
Job market paper
Beyond the Exploration–Exploitation Trade-off: The Behavioral Structure of Switching between Alternatives
Many sequential decisions, such as research projects, investments and partnerships, involve choosing among alternatives whose quality is learned only through experience. Standard models predict that people switch between alternatives only when they learn something new. I test this in an incentivized one-armed bandit experiment: subjects choose repeatedly between a familiar option and an uncertain one, a reward reveals that the uncertain option is better, and outcomes of the familiar option reveal nothing. The model predicts single switching: try the uncertain option until it pays off or enough failures pile up, then stay with one option for good. About half of subjects return to the uncertain option after abandoning it with no new information about it, and 39% leave it at least once even after learning it is the better one. Those who switch this way explore inferior options longer, find good ones later, and earn 20 to 25% less than single-switchers. Part of this switching is a reaction to outcomes that carry no information, such as rewards from the familiar option. When subjects can choose plans that limit their reaction to outcomes, about half of those who deviated adopt single switching, consistent with noisy execution of a correct plan. The rest choose plans that keep switching, consistent with an incorrect model of the problem or non-standard preferences.