A Bayesian model for assessing risk using expert judgment about paired scenario comparisons

This dissertation describes a methodology for assessing rare event risks in complex technological systems using expert judgment about scenarios defined by key situational variables. Managers and decision-making authorities can use this rare event knowledge to design projects, develop policy, and allocate resources in their efforts to mitigate the greatest system risks.; Rare events inherently suffer from a scarcity of data necessitating expert judgment. The expert judgment about the rare events under examination is elicited through paired scenario comparisons that differ in only a single dimension. The results are combined with the prior knowledge about the system (typically derived from global statistics, trends, etc.) in the usual Bayesian fashion. Expert calibration is examined and a set of diagnostics have been developed based on information theoretic cross-entropy. This diagnostic information enables managers and decision-making authorities to qualify the experts and perhaps eliminate poorly performing experts.; The model identifies those situational variables that contribute the most to system risk. The model is demonstrated on expert paired comparison data taken from a study of the largest passenger vessel ferry system in the U.S. The results from this model, which accounts for a greater pool of knowledge, compares favorably with the classical results. Hence, this model and methodology might be useful for rare event risk assessment studies in other disciplines.