Coordination with Cognitive Noise

We experimentally study how cognitive noise affects behavior in coordination games. Our key testable prediction is that equilibrium behavior depends on context – which we define as the distribution from which games are drawn. This prediction arises from players efficiently using their limited cognitive resources; furthermore, this prediction distinguishes cognitive noise from a large class of alternative behavioral game theory and learning models. Experimentally, we find that the frequency with which subjects coordinate depends systematically on context. We argue that cognitive noise can microfound the global games approach, while also generating novel predictions that we observe in our experimental data.