'Learning' with no feedback in a competitive guessing game

An assumption underlying current models of learning in games is that learning takes place only through repeated experience of outcomes. This paper experimentally tests this assumption using Nagel's (1995, Amer. Econ. Rev. 85, 1313-1326) competitive guessing game. The experiment consists of several periods of repeated play under alternative feedback conditions, including no-feedback conditions in which players receive no information between periods. If learning takes place only through reinforcement resulting from experienced outcomes, choices in the no-feedback conditions should not converge towards the Nash equilibrium. While less than under full information, there is convergence towards the equilibrium prediction in the no-feedback conditions. Varying priming given to subjects between periods does not affect the results. (C) 2003 Elsevier Science (USA). All rights reserved.