Homogeneous Actions and Hetergeneous Beliefs: Experimental Evidence on the Formation of Information Cascades

Theories on the formation of information cascades have been tested in experimental settings in which players publicly announce binary expectations of a binary event based on private signals and preceding public announcements. We replicate and supplement the experimental data collection by privately eliciting beliefs on the signals and the event. Therefore, we are able to directly observe how players update expectations and form inferences based on the announcements of others. Past studies have focused on one implication of cascade theory--the realization of cascades. Yet the prediction of the model is stronger than simply that players will follow the example of others once a cascade has begun. In fact, new public announcements are predicted to have no effect on the subjective expectations of others, because these announcements contain no information about private signals. We see in our data that cascade behavior arises as frequently as in previous cascade experiments without belief elicitation. Yet, we find evidence of systematic heterogeneity in belief-updating rules across subjects. As a result, a Bayesian decision maker may extract information about private signals from public announcements made during a cascade. In fact, reported subjective probabilities are revised in response to announcements after a cascade has begun.