Information diffusion in networks through social learning

We study perfect Bayesian equilibria of a sequential social learning model in which agents in a network learn about an underlying state by observing neighbors' choices. In contrast with prior work, we do not assume that the agents' sets of neighbors are mutually independent. We introduce a new metric of information diffusion in social learning that is weaker than the traditional aggregation metric. We show that if a minimal connectivity condition holds and neighborhoods are independent, information always diffuses. Diffusion can fail in a well connected network if neighborhoods are correlated. We show that information diffuses if neighborhood realizations convey little information about the network, as measured by network distortion, or if information asymmetries are captured through beliefs over the state of a finite Markov chain.