Learning from Shared News: When Abundant Information Leads to Belief Polarization
(joint with Danil Dmitriev, Simone Galperti)
Abstract: (the draft is attached)
We study social learning via news sharing. Each period agents receive the same quantity and quality of first-hand information and have the opportunity to share it with
friends. Some agents (possibly a few) share information selectively. Selective sharing generates heterogeneous news diets across agents, who, however, are aware of it and
update beliefs via Bayes’ rule. We show that, contrary to standard learning results, agents’ beliefs can diverge in this environment. This occurs if and only if agents hold
misperceptions (even minor) about friends’ access to first-hand information and if its quality is low. We show that abundant information can exacerbate belief polarization.
That is, when the quantity of first-hand information grows indefinitely agents can hold opposite degenerate beliefs. Intuitively, polarization worsens with misperception
and imbalance of news diets. Polarization can also worsen when information quality rises or when the agents’ social networks expand, despite providing them with more
information. Information aggregation can mitigate, and even eliminate, polarization.