Research

Job Market Paper

Social Learning with Coarse Communication

Abstract: This paper studies social learning on networks when communication is coarse. Each period, agents update a log-odds opinion by combining (i) their own past opinion, (ii) a network-weighted aggregate of others’ communicated opinions, and (iii) a new private log-likelihood increment. However, when agents share the opinion, they cannot transmit real-valued beliefs: instead, each agent encodes the current belief into one of K messages and receivers decode messages into real-valued reconstructions, so the socially transmissible map is a step function and social influence operates only through threshold crossings. This discreteness creates a bandwidth-mixing mismatch: dense social exposure collapses cross-sectional dispersion in opinions, yet threshold crossings require dispersion. Along with this fast-mixing, we show that with positive probability a sufficiently cohesive set enters an interior bin whose robustness band lies strictly inside that bin; thereafter, communicated content becomes constant, private evidence cannot dislodge the group, and an information ceiling measured by mean-squared error remains bounded away from zero even as time grows. Furthermore, on segregated networks, a locked-in seed can replicate across communities, so architecture amplifies whichever bin is selected. Finally, we derive a resolution requirement: preventing interior traps on fast-mixing networks requires a sufficiently fine reporting alphabet.

Working Papers

Information Overload Under Rational Learning
with Sudipta Sarangi, and Hector Tzavellas

Learning Networks
with Promit K. Chaudhuri, Sudipta Sarangi, and Hector Tzavellas

Working in Progress

A Citizen Candidate Model with Public Funding
with Euncheol Shin