Modeling and estimation of accident rate and trend in air transport

An established approach in the evaluation of aviation accident statistics is to determine point estimates of the accident rate by dividing number of accidents by number of flights and to determine an uncertainty interval through evaluation of the underlying binomial distribution. The trend, however, is not estimated. Another established approach is to perform a regression analysis to estimate rate and trend, but then uncertainty is not estimated. In this paper we overcome these limitations of established approaches by studying the problem as one of Bayesian estimation of the joint conditional density function of accident rate and trend given accident and flight statistical data. Subsequently, a particle filter is used in order to perform numerical evaluations. The novel approach is shown to work well on commercial aviation accident data.