The effect of matching mechanism on learning in repeated 2x2 games with limited information Nick Feltovich * Department of Economics, University of Houston Houston, TX 77204–5019, USA nfelt@mail.uh.edu Atsushi Iwasaki Department of Intelligent Systems Graduate School of Information Science and Electrical Engineering, Kyushu University Motooka 744, Nishi-ku, Fukuoka, 812-8581, Japan Sobei H. Oda Faculty of Economics, Kyoto Sangyo University Kamigamo, Kita–ku, Kyoto 603–8555, Japan April 26, 2007 Abstract We conduct an experiment in which human subjects repeatedly play each of six 2x2 games under one of two matching mechanisms. Under fixed–pairs matching, each subject is matched to the same opponent in all rounds of a game. Under random matching, each subject is equally likely to be matched to each potential opponent in any round. In both treatments, the games are played under limited information: subjects are never shown the payoff matrices for the games they are playing, nor are they given information about opponent payoffs. We find that subject behavior, while similar in early rounds in the two treatments, diverges over time. By and large, outcomes are better under fixed–pairs matching than under random matching: coordination on a pure–strategy Nash equilibrium is more likely, and payoffs are higher. Convergence toward equilibrium play is faster under fixed pairs than under random matching in all games except Prisoners’ Dilemma. Journal of Economic Literature classifications: D83, C72, C73. Keywords: experiment, learning, behavioral economics, fixed pairs, random matching. * Corresponding author. Oda and Iwasaki are supported in part by Grants–in–Aid of Scientific Research (#17310029 and #17700155) and by the Ministry of Education, Culture, Sport, Science and Technology (Japan). Any errors are the result of the fixed matching between the three coauthors.