Inference on the Miss Distance in a Conjunction

Abstract: Over the last quarter-century, spacecraft conjunction assessment has focused on a quantity associated by its advocates with collision probability. This quantity has a well-known dilution feature, where it is small when uncertainty is large, giving rise to false confidence that a conjunction is safe when it is not. An alternative approach to conjunction assessment is to assess the missed detection probability that the best available information indicates the conjunction to be safe, when it is actually unsafe. In other words, the alternative seeks to answer the question of whether unknowable errors in the best available data might be especially unlucky. A proper implementation of this alternative avoids dilution and false confidence. Implementations of the alternative use either significance probabilities (p-values) associated with a null hypothesis that the miss distance is small, or confidence intervals on the miss distance. Both approaches rely on maximum likelihood principles to deal with nuisance variables, rather than marginalization. This paper discusses the problems with the traditional approach, and summarizes other work that developed the alternative approach. The paper presents examples of application of the alternatives using data from actual conjunctions experienced in operations, including synthetic scaling to highlight contrasts between the alternative and the traditional approach.