Social learning with case-based decisions

In decision problems with unknown payoffs, individuals use their own past experience and the experiences of others. In this article I develop a theory that links in a natural way aggregated decision processes to the underlying communication structure. In particular, I consider a binary decision problem. Since there is noise in the payoff structure agents have to learn which decision has the better performance. I analyze the long-run convergence behaviour of the social learning path for four basic communication structures. Specifically, it is shown that for complete information there is a tendency to conformity if the aspiration level of the population is low and a tendency for non-convergence if the aspiration level is high. Star communication structures, in contrast, are characterized by diversity in the long run, whereas in Delta-neighbourhood communication structures the better decision diffuses viith a constant pace. The relative performances of subpopulations or cliques are shown to depend crucially on the communication intensity and relative size of the subpopulations. (C) 1999 Elsevier Science B.V. All rights reserved.