This dissertation consists of three theoretical chapters. In the first chapter, I study an evolutionary model with a finite population of boundedly rational agents, who do not have access to the same amount of information. Time is discrete, and in each period two agents are paired to play a 2 × 2 symmetric coordination game. Each player can cross paths with two kinds of opponents: Neighbors or Strangers. If a player faces a Neighbor, she can access some information about her opponents past plays, and plays using a myopic best-response. But if she faces a Stranger, she does not have access to any information, and therefore plays according to a case-based decision rule. I show that in the short run, segregated localities emerge, to finally disappear in the long run, in favor of the Pareto Efficient convention. The main contribution of this chapter is that I show that agents coordinate in an evolutionary framework on an efficient outcome, even when information is asymmetric, without assuming any pre-play communication or mobility of the agents.

In the second chapter (with Alexander Matros) we consider K finite populations of boundedly rational agents whose preferences and information differ. Each period agents are randomly paired to play some coordination games. We show that several special (fixed) agents lead the coordination. In a mistake-free environment, all connected fixed agents have to coordinate on the same strategy. In the long run, as the probability of mistakes goes to zero, all agents coordinate on the same strategy. The long-run outcome is unique, if all fixed agents belong to the same population.

The last chapter (with Alexander Matros) extends the study of survival of Altruism in a public good game similar to the one by Eshel, Samuelson and Shaked [14] to other circulant graphs. We describe short run outcomes and find sufficient conditions for the survival of Altruism in the long run.