Consider a population of agents who play a game through repeated interactio
ns, and adapt their behavior based on information about other agents' previ
ous behavior. The standard way of modeling such a process is to assume that
everyone in the population is governed by the same adaptive rule, e.g., be
st response, imitation, or the replicator dynamic. This paper studies heter
ogeneous populations of agents in which some agents are best responders, ot
hers are conformists (they do what the majority does), and still others are
nonconformists (they do the opposite of what the majority does). Unlike de
terministic best reply processes, which in 2 x 2 games converge to Nash equ
ilibrium, these heterogeneous processes may have limit cycles; moreover lim
it cycles may exist even when the proportion of non best responders is arbi
trarily small. We show how to analyze the asymptotic behavior of such proce
sses through a suitable generalization of Bendixson stability theory combin
ed with stochastic approximation theory. Journal of Economic Literature Cla
ssification Numbers: C44 C73, D83. (C) 2000 Academic Press.