We study the implication of biased belief updating in a model of bandit experimentation. A player faces a task for which the optimal action is fixed over time but uncertain. She learns about the optimal action over time: in each period, she tries out an action, observes its outcome, updates her belief about the optimal action, and then decides the action to try in the next period. The point of departure from the traditional experimentation literature is that the player updates in a biased way. Following Eyster-Rabin (2005), we assume that the player underappreciates the connection between the state of the world and the observed outcome. As a result, she underreacts to new information. We characterize the implication of the bias on experimentation and use the results to explain the near miss effect observed in gambling. Lastly, we show how the bias affects experimentation if, in addition, the player has a misspecified prior belief.
