Optimally aggregating elicited expertise : a proposed application of the Bayesian Truth Serum for policy analysis

Abstract

Uncertainty and risk are obstacles that nearly all policy-makers encounter during their careers. However, evaluating uncertainty and risk can be difficult since these concepts may be broadly defined. This may result in inaccurate estimates of risk and uncertainty. Expert elicitation is a formal, structured method of obtaining subjective expert judgment in scenarios where objective data is unobtainable. It is designed to reduce the influence of ambiguity on expert judgment, meaning that analysts may use such subjective data as if it were objectively generated. Expert elicitation methods tend to aggregate expert judgment in order to create a unified response, but determining how to combine expert opinions remains a difficult problem. In this thesis, a review of the literature and background behind defining expertise and expert elicitation will be provided. Additionally, this thesis introduces the Bayesian Truth Serum as a potential weighting algorithm for combining expert judgments. As opposed to other weighting algorithms, the Bayesian Truth Serum uses the metaknowledge of experts to create weights for aggregation. Using such information may prove superior to assuming a normal distribution of expertise or relying upon experts to provide estimates of their own expertise.