Statistical methods for the use of accident precursor data in estimating the frequency of rare events

Abstract In estimating the frequency of rare events, the number of observed events is usually too small to support the development of accurate estimates by means of the usual statistical estimator, i.e. the number of events divided by the years of experience. Data on accident ‘precursors’ can help in obtaining a reasonably accurate estimator. However, past attempts to use precursor data have been problematic. In the work described here, the authors assess the problems associated with various precursor-based estimators of rare event frequencies, and propose a new estimator that seems suitable for use in practice. However, the results suggest that the competing goals of achieving low noise and low bias are inherently incompatible. Therefore, the authors recommend possible directions for future work to explore this tradeoff in more depth.