An experiment on social mislearning

We investigate experimentally whether social learners appreciate the redundancy of information conveyed by their observed predecessors' actions. Each participant observes a private signal and enters an estimate of the sum of all earlier-moving participants' signals plus her own. In a first treatment, participants move single-file and observe all predecessors' entries; Bayesian Nash Equilibrium (BNE) predicts that each participant simply add her signal to her immediate predecessor's entry. Although 75% of participants do so, redundancy neglect by the other 25% generates excess imitation and mild inefficiencies. In a second treatment, participants move four per period; BNE predicts that most players anti-imitate some observed entries. Such anti-imitation occurs in 35% of the most transparent cases, and 16% overall. The remaining redundancy neglect creates dramatic excess imitation and inefficiencies: late-period entries are far too extreme, and on average participants would earn substantially more by ignoring their predecessors altogether.