Rare event techniques applied in the Rasmussen study

The Rasmussen Study estimated public risks from commercial nuclear power plant accidents, and therefore the statistics of rare events had to be treated. Two types of rare events were specifically handled, those rare events which were probabilistically rare events and those which were statistically rare events. Four techniques were used to estimate probabilities of rare events. These techniques were aggregating data samples, discretizing ''continuous'' events, extrapolating from minor to catastrophic severities, and decomposing events using event trees and fault trees. In aggregating or combining data the goal was to enlarge the data sample so that the rare event was no longer rare, i.e., so that the enlarged data sample contained one or more occurrences of the event of interest. This aggregation gave rise to random variable treatments of failure rates, occurrence frequencies, and other characteristics estimated from data. This random variable treatment can be interpreted as being comparable to an empirical Bayes technique or a Bayesian technique. In the discretizing event technique, events of a detailed nature were grouped together into a grosser event for purposes of analysis as well as for data collection. The treatment of data characteristics as random variables helped to account for the uncertainties arising from this discretizing. In the severity extrapolation technique a severity variable was associated with each event occurrence for the purpose of predicting probabilities of catastrophic occurrences. Tail behaviors of distributions therefore needed to be considered. Finally, event trees and fault trees were used to express accident occurrences and system failures in terms of more basic events for which data existed. Common mode failures and general dependencies therefore needed to be treated. 2 figures