This paper addresses the relationship between the degree of decentrali
zation in a population and the probability of coordinating on an effic
ient outcome rather than the risk-dominant outcome in a coordination g
ame. The model here extends the literature on evolutionary games by su
pposing that players interact in groups but that occasionally they may
change groups. The main innovation is the group-changing assumption w
hich allows players to 'vote with their feet.' In contrast to most of
the recent literature (e.g. Kandori et al., 1993) in which the risk-do
minant equilibrium is selected in the long run, the model here predict
s convergence to the efficient equilibrium if interaction is decentral
ized enough. Furthermore, it is shown how the model can be applied to
the problem of product standardization.