Dealing with uncertainty arising out of probabilistic risk assessment

: The Nuclear Regulatory Commission (NRC), like many other regulatory bodies, often must decide whether some system or some overall technology satisfies a prescribed safety goal. The NRC makes these decisions on the basis of probabilistic risk assessments (PRAs). The objective of a PRA is to quantify the risk of occurrence of some undesirable event, such as a reactor core melt, or of some undesirable consequence of an event, in this example, death. The quantified risk is expressed as a probability, e.g. , one core melt expected per 10,000 reactors per year. Estimates of risk are subject to uncertainty. This uncertainty arises from variability in available data on failure rates, difficulties in predicting the effects of unusually stressing events external and internal to the system being assessed, and an insufficient data base on human errors in uncommon situations. Therefore, the actual risk that a part. a system, or a plant will fail may be greater or lesser than the PRA's best estimate of the risk. Indeed, there is associated with each possible risk value a probability that that value is the right one, and it is thus possible to construct a frequency distribution of probabilities over all possible risk values. Past PRAs have equated the best estimate of the risk, that is, the value to be used in comparing the risk to the safety goal, with the median risk value in the frequency distribution, i.e. the median of the set of all possible risk values when weighted by their associated probabilities. It has been suggested that the mean risk value be used instead, because the mean is usually larger than the median in risk-frequency distributions, which generally include some relatively high values with nonnegligible probabilities.