مدلهای تجربی انتخاب گسسته و به روز رسانی اعتقاد به آزمایش های یادگیری مشاهده ای

Subjects in economics experiments are often asked to choose an action from a set of discrete choices. The logit-QRE approach to analyse these data places strong restrictions on how subjects in information cascades experiments extract information from observed outcomes and how they update beliefs in response to new information. We add a belief elicitation procedure to the experimental design that allows us to measure directly both the inferences drawn from publicly announced decisions and how beliefs are updated in response to new information. The reported beliefs tend to be well calibrated to frequentist probabilities and also predict individual choices. Contrary to previous conclusions, we find that respondents do not tend to overweight private information when updating beliefs. Our analysis suggests that the earlier findings arise because identification of the discrete choice model relies on a misspecified model of belief updating in response to preceding announcements.