A Bayesian method for calibration and aggregation of expert judgement

Abstract

This paper outlines a Bayesian framework for structured expert judgement (sej) that can be utilised as an alternative to the traditional non-Bayesian methods (including the commonly used Cooke's Classical model [13]). We provide an overview of the structure of an expert judgement study and outline opinion pooling techniques noting the benefits and limitations of these approaches. Some new tractable Bayesian models are highlighted, before presenting our own model which aims to combine and enhance the best of these existing Bayesian frameworks. In particular: clustering, calibrating and then aggregating the experts' judgements utilising a Supra-Bayesian parameter updating approach combined with either an agglomerative hierarchical clustering or an embedded Dirichlet process mixture model. We illustrate the benefit of our approach by analysing data from a number of existing studies in the healthcare domain, specifically in the two contexts of health insurance and transmission risks for chronic wasting disease.