Taste Projection in Models of Social Learning

Abstract

This paper studies the implications of taste projection—the tendency to overestimate how similar others’ preferences are to our own—within social-learning environments. Individuals sequentially choose among two options with payoffs dependent on an unknown state of the world and one’s idiosyncratic, privately-observed taste. Learning about the state from others’ choices requires individuals to assess whether uncommon actions were likely provoked by atypical tastes or private information contradicting the public belief. Taste projectors over-attribute these actions to information, which prevents some (and perhaps all) from ever learning their correct action. A player’s long-run beliefs are determined by her own taste and the extent to which she and all others project. When each thinks her taste is most common, all players inevitably choose the same option and each grows certain this choice is optimal. But if some acknowledge they have an uncommon taste, social beliefs and behavior may perpetually cycle—history never provides a clear message about the optimal choice. Note: Several ideas in this paper have been subsumed by `Quality is in the Eye of the Beholder: Taste Projection in Markets with Social Learning’ (AER, 2024). A new version is in progress that emphasizes how even small degrees of bias can lead to learning failures.