Inference About the Masking Probabilities in the Competing Risks Model

Abstract

The competing risks model considers the setting where a subject or unit is exposed to multiple risks and a failure can be caused by one of the multiple risks. When the causes of failure for an item is not precisely known but can be narrowed down to a subset of the potential risks, the cause is reported as masked. The likelihood model for a masked failure includes the masking probabilities as the chance of the cause getting masked may actually depend on the true cause of failure. This article investigates issues related to statistical inference for the masking probabilities. In terms of predicting the actual but unknown cause of failure of a masked case, the performance of model based prediction is found to be less than desirable in a limited study.