Mental Models and Learning: The Case of Base-Rate Neglect; Sevgi Yuksel (UC Santa Barbara)

Abstract

We study whether suboptimal behavior can persist in the presence of feedback and examine the role that incorrect mental models play in this persistence. Focusing on a simple updating problem, we document using a laboratory experiment the evolution of beliefs in response to feedback. We compare a baseline treatment, in which a majority of subjects display base-rate neglect (BRN) in initial beliefs, to a control treatment that does not allow for BRN as a mental model but in which learning from feedback is similarly possible. Learning is slow and partial in the baseline, such that after 200 rounds of feedback, beliefs in this treatment are farther from the Bayesian benchmark relative to the control treatment. The treatment effect is linked to partial attentiveness to feedback by those subjects who initially display BRN in the baseline. Presenting subjects with evidence that unequivocally challenges their beliefs by summarizing feedback up to that point improves the accuracy of beliefs substantially and eliminates base-rate neglect. Finally, we find evidence that learning from feedback can generate insights (for example, that the base-rate should be considered in the belief formation process) that can be partially transferred to new settings.

Click here to view the paper.