Bayesian Polarization
Tuval Danenberg
Published: 2025/9/2
Abstract
We study belief polarization among Bayesian agents observing public information about a multidimensional state. Baliga et al. (2013) show that divergence in the sense of first-order stochastic dominance is impossible for one dimensional beliefs, but we find that in multidimensional settings it can occur for all marginal beliefs, even with infinitely many signals. At the same time, we extend their impossibility result: divergence in the sense of multidimensional stochastic dominance is impossible. For an intermediate stochastic order, polarization may arise only in the short-run. We provide necessary and sufficient conditions on signal structures for persistent polarization and discuss implications for polarization in actions.