Differences of Opinions
This paper presents a generalization of the DeGroot learning rule in which social learning can lead to polarization, even for connected networks. I first develop a model of biased assimilation in which the utility an agent receives from past decisions depends on current beliefs when uncertainty is slow to resolve. I use this model to motivate key features of an agent's optimization problem subject to scarce private information, which forces the agent to extrapolate using social information. Even when the agent extrapolates under "scientific" assumptions and all individuals in the network process and report their private signals in an unbiased way, the possibility of biased processing or reporting leads agents to process social signals differently depending on the sender. The resulting solution to the agent's problem is a heterogeneous confidence learning rule that is distinct from bounded confidence learning rules in that the agent may actually move her beliefs away from, and not only discard, signals from untrustworthy senders.
Aliprantis, Dionissi. 2016. “Differences of Opinions.” Federal Reserve Bank of Cleveland, Working Paper No. 16-04. https://doi.org/10.26509/frbc-wp-201604