Parametric Estimation for Window Censored Recurrence Data

Many applications, in particular the failure, repair, and replacement of industrial components or physical infrastructure, involve recurrent events. Frequently, the available data are window-censored: only events that occurred during a particular interval are recorded. Window censoring presents a challenge for recurrence data analysis. For statistical inference from window censored recurrence data, we derive the likelihood function for a model in which the distributions of inter-recurrence intervals in a single path need not be identical and may be associated with covariate information. We assume independence among different sample paths. We propose a distribution to model the effect of external interventions on recurrence processes. This distribution can represent a phenomenon, frequently observed in practice, that the probability of process regeneration increases with the number of historical interventions; for example, an item that had a given number of repairs is generally more likely to be replaced in the wake of a failure than a similar item with a smaller number of repairs. The proposed model and estimation procedure are evaluated via simulation studies and applied to a set of data related to failure and maintenance of water mains. This article has online supplementary material.