Eliciting informative priors by modeling expert decision making

Abstract

There are significant limitations to current methods for eliciting the prior beliefs of experts. To combat some of these limitations, this paper proposes an alternative approach that infers an expert’s prior beliefs about an uncertain event, A, from the expert’s past decisions. We show that an analyst can use past information on an expert’s decision-making task, contingent on an expert’s prior of A, to model the decision-making process and infer an approximation of the prior for A. This concept is illustrated by an application to recidivism. We conclude this work by highlighting important directions for future research. Funding: J. R. Falconer’s research is funded through the University of Waikato Doctoral Scholarship.