A discussion on the use of judgment in representing uncertainty in PRAs

This paper is a review of some aspcets of current practice in the analysis of uncertainties in Probabilistic Risk Assessments (PRAs). The use of judgement in performing a PRA is unavoidable due to the complexity of the analysis and the many uncertainties which surround it. In most PRAs performed to date, uncertainties have been characterized by using the subjective theory of probability to calibrate the analyst's degree of belief in various hypotheses, and the degrees of belief from different parts of the analysis have been combined to represent an overall measure of uncertainty in the bottom line results. While this may be useful in some circumstances, it may obscure some valuable insights as discussed in this paper. Historically it has been assumed that certain probabilities in the analysis of accident progression relate to stochastic variability. An example is given where a consideration of the analysis framework leads to an indication that this may be erroneous and that it may be more appropriate to regard some probabilities as characterizing modeling uncertainty. (orig.)