Assessment for Catastrophic Risks

Records of nonoccurrence of catastrophic accidents from technologically based industrial operations, such as nuclear power generation and liquefied natural gas (LNG) transportation are sometimes cited as evidence of the safety of such operations. The appropriateness and inappropriateness of different types of models of accident processes for these systems are discussed. Selectivity biases in defining both die numerator and denominator for accident rates are shown to be important to the size of the estimated rate and these are illustrated by examples. Inferential approaches to estimating a rate based on zero occurrences are briefly discussed. The conclusion reached is that a record of zero occurrences is of cold comfort in ruling out catastrophic events. Experience, unless it is interpreted by strong models and assumptions, helps little to establish very small probabilities of a catastrophe.