We study a model of sequential learning with naive agents on a network. The key behavioral assumption is that agents wrongly believe their predecessors act based on only private information, so that correlation between observed actions is ignored. We provide a simple linear formula characterizing agents' actions in terms of paths in the network and use this formula to determine when society learns correctly in the long-run. Because early agents are disproportionately influential, standard network structures can lead to herding on incorrect beliefs. The probability of mislearning increases when link densities are higher and when networks are more integrated. When actions can only communicate limited information, segregated networks often lead to persistent disagreement between groups.
↧