An expert opinion elicitation method based on binary search and Bayesian intervals

This paper proposes an indirect method to elicit experts' opinions. The approach is an attempt to balance accuracy and time consumption (efficiency) in elicitation exercises. Each step of the proposed method involves two types of questions: first, the expert is asked about which of two subintervals most likely holds the unknown of interest (a procedure similar to the binary search algorithm); then, the expert assigns a credibility level on which he bases his decision. A set of Bayesian intervals of the unknown is thereby obtained, and the probability distribution underlying expert's beliefs is approached. The usefulness of the method is analysed by means of: 1) a comparison with established methods from the literature (e.g., the direct and bisection methods); 2) its application to a real world case within the context of probabilistic risk analysis.