Various approaches to the characterization of analytic uncerta&ties in system safety assessment are contrasted. It is concluded that probabilistic techniques, interpreted in a Bayesian framework, currently offer the most practical and conceptually sound basis for the representation of uncertainty. Important distinctions exist, nevertheless, between standard problems addressed by Bayesian decision theory and topical regulatory issues of risk control and management. Certain of these distinctions are delineated and concomitant implications for the performance and the interpretation of probabilistic safety analyses are discussed. It is suggested that the potential of probabilistic methods to provide robust insights into the safety characteristics of complex systems has not yet been realized fully. Methodological developments reflecting substantial progress in the techniques of uncertainty analysis are described and their application to recent programs of the US Nuclear Regulatory Commission are cited.
'Probability' is a word that carries much freight. To disentangle the often confused semantic and technical issues of probability, I shall first address the former.
It is difficult to criticize with conviction attachment of the word probability to any mathematical quantity that satisfies the three equations defining the algebra of probabilities. This encompasses quantitative degrees of belief as modeled by the Bayesian, and the parameters governing various stochastic models. Where both types of quantity arise in a single analysis, however (I have in mind probabilistic safety assessment), then for the sake of 315 Reliability Engineering and System Safety