Forecasts or fortune-telling: When are expert judgements of safety risk valid?

Abstract

Safety analysis frequently relies on human estimates of the likelihood of specific events. For this purpose, the opinions of experts are given greater weight than the opinions of non-experts. Combinations of individual judgements are given greater weight than judgements made by a lone expert. Various authors advocate specific techniques for eliciting and combining these judgements. All of these factors – the use of experts, the use of multiple opinions, and the use of elicitation and combination techniques – serve to increase subjective confidence in the safety analysis. But is this confidence justified? Do the factors increase the actual validity of the analysis in proportion to the increase in subjective confidence?

In this paper, by means of a critical synthesis of evidence from multiple disciplines, we argue that it is plausible that expert judgement deserves special standing, but only for well understood local causal mechanisms. We also conclude that expert judgements can be improved by using appropriate elicitation techniques, including by combining judgement from multiple experts. There is, however, no evidence to suggest that fuzzy, neural network, or any other form of complicated processing of expert judgement has any advantage over simple combination mechanisms.