Abstract

Two key features robustly describe ideological differences in society: (i) individuals persistently disagree about objective facts; (ii) individuals also disagree about which sources can be trusted to provide reliable information about these facts. We develop a model in which these patterns arise endogenously as the result of Bayesian learning with misspecification. In our model, individuals observe signals about multiple unobserved states from many information sources. Individuals learn the accuracies and ideological biases of sources with the assumption that a possibly biased reference source is independent and unbiased. With arbitrarily small bias in the reference source, individuals incorrectly learn that biased sources are more accurate than unbiased sources. Large disagreements between individuals persist, even when individuals observe the same set of information sources. We discuss implications for social network formation and media markets.

Matthew Gentzkow∗, Stanford and NBER
Michael B. Wong, MIT
Allen T. Zhang, Harvard