An Approach to the Analysis of Accident Precursors

The interpretation of accident precursors or “near misses” (e. g., the space shuttle flights with damaged O-rings prior to the Challenger disaster) is often problematic, since the ambiguous nature of the evidence in such cases makes them subject to widely divergent interpretations. Various methods have therefore been developed to allow the analysis of accident precursors and incipient or partial failures. In this paper, we present a Bayesian approach to the analysis of accident precursors. According to this framework, events such as accident precursors and near misses can be analyzed in terms of two prior distributions: one for the frequency of the precursor itself, and one for the conditional probability of a severe accident given a precursor. Our primary focus is on the implications of this model for risk management and decision making. In particular, through the use of conjugate prior distributions, we are able to show that accident precursors are more likely to lead to increased rather than decreased risk estimates.