Masked Failure Data: Competing Risks

Abstract

The competing‐risks model considers a setting when a system is exposed to multiple risks and a system failure can be caused by one of the multiple risks. The case of masked causes arises when the causes of failure for some of the systems are not completely identified but are narrowed down to a subset of the potential risks. This article provides a selective review of statistical models and methods for analysis of competing‐risks data with or without masking in the context of reliability applications. Latent failure times and cause‐specific hazards provide two parallel approaches to competing‐risks models; identifiability issues in these models in both nonparametric and parametric settings are discussed. For the case of masked data, recent developments in both likelihood‐based and Bayesian approaches are reviewed. A special emphasis is given to the treatment and models for the masking probabilities. The article concludes with a review of different reliability applications.