Individual learning in normal form games: Some laboratory results

We propose and test a simple belief learning model. We find considerable heterogeneity across individual players; some players are well described by fictitious play (long memory) learning, other players by Cournot (short memory) learning, and some players are in between. Representative agent versions of the model fit significantly less well and sometimes point to incorrect inferences. The model tracks players' behavior well across a variety of payoff matrices and information conditions. (C) 1997 Academic Press.