Heterogeneous Tastes and Social (Mis)Learning

Abstract

How do people learn from others’ actions when those people may have differing tastes? We present data from two experiments in which properly extracting information from other people’s actions requires an observer to account for how her predecessors’ tastes may have influenced those actions. We find support for social learning that obeys some basic comparative statics predicted by the rational model. However, we also find significant and systematic departures. Participants systematically over-infer from others’ behavior when that behavior is weakly predictive of the underlying state and under-infer from others when their behavior is strongly predictive. This pattern of misinterpreting others’ actions is consistent with participants over-weighting the likelihood that others have tastes similar to their own. Information about others’ tastes does not eliminate these biases in inferences.