Evidence aggregation in expert judgments

Abstract

Experts are asked for percentiles of the seismic capacity of a particular structural component. The problem is to aggregate separate bodies of evidence. It can happen that experts characterize by different schemes of reasoning, provide interactive estimates, have non-comparable levels of competence, rely upon accumulated knowledge that hides superpositions. Conventional Bayesian approach may not allow to comprehend all the complexity of the situation. Or the corresponding formal tools would become too cumbersome. Conversely, belief functions offer ground for consistent treatment of the several sources of imprecision and uncertainty. Three cases of practical interest are considered and models for aggregating evidence are developed accordingly. First, is the case of equivalent and independent experts; second, are non-equivalent and independent experts; finally, is the situation where experts are viewed as dependent though equivalent. Clearly, the general case where experts are both non-equivalent and dependent would be treated as the combination of the last two ones. Some of the merits of the models for aggregating evidence are discussed. Specifically, it is shown how the measures obtained could relate with probabilities and set upper and lower bounds.