Mental Models and Learning: The Case of Base-Rate Neglect
American Economic Review (Forthcoming)
Are systematic biases in decision making self-corrected in the long run when agents are
accumulating feedback informative of optimal behavior? This paper focuses on a canonical
updating problem where the dominant deviation from optimal behavior is base-rate neglect.
Using a laboratory experiment, we document persistence of suboptimal behavior in the presence
of feedback. Using diagnostic treatments, we study the mechanisms hindering learning from
feedback. We investigate the generalizability of these results to other settings by also studying
long-run behavior in a voting problem where failure to condition on being pivotal generates
suboptimal behavior. Our findings provide insights on what types of mistakes should be expected
to be persistent in the presence of feedback. Our results suggest mistakes are more likely
to be persistent when they are driven by incorrect mental models that miss or misrepresent
important aspects of the environment. Such models induce confidence in initial answers, limiting
engagement with and learning from feedback. These results have implications for how policies
should be designed to counteract behavioral biases.