An application of copulas to accident precursor analysis

Data on accident precursors can help in estimating accident frequencies, since they provide a rich source of information on intersystem dependencies. However, Bayesian analysis of accident precursors requires the ability to construct joint prior distributions reflecting such dependencies. For example, the failure probabilities of a particular safety system under normal and accident conditions, respectively, will generally not be identical (because of the effects of the accident), but will almost certainly be correlated (since both failure probabilities reflect the performance of the same components, with the same inherent levels of reliability). In this paper, we explore the use of copulas (a method of representing joint distribution functions with particular marginals) to construct the needed prior distributions, and then use these distributions in a Bayesian analysis of hypothetical precursor data. This demonstrates the usefulness of copulas in practice. The same approach can also be used in a wide variety of other contexts where joint distributions with particular marginals are desired.