Learning in Canonical Networks

Abstract

Subjects observe a private signal and then make an initial guess; they observe their neighbors’ guesses and guess again, and so forth. We study learning dynamics in three networks: Erdös-Rényi, Stochastic Block (reflecting homophily) and Royal Family (that accommodates both highly connected celebrities and local intearctions). We find that the Royal Family network is more likely to sustain incorrect consensus and that the Stochastic Block network is more likely to persist with diverse beliefs. These aggregate patterns are consistent with individuals following DeGroot updating rule.