Statistical Methods for Bivariate Failure Times Under Event-Dependent Censoring

• Activity Number: 253
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: SSC
• Author(s): Yujie Zhong*+ and Richard Cook
• Companies: and University of Waterloo
• Keywords: Event-dependent censoring ; Copula model ; Misspecification

Abstract

In studies of life history processes, multiple events may be of interest. When one or more of the events is serious their occurrence may lead to a higher risk of withdrawal resulting in event-dependent censoring. This can be ignorable when models are fully and correctly specified, but can lead to inconsistent estimators when models are only partially specified, as is the case when working independence assumptions are adopted. Several methods will be discussed for the analysis of bivariate failure time processes including marginal methods using inverse probability of censoring weights, partially conditional multi-state models, and nonparametric and semiparametric joint models based on copulas. Empirical properties of the resulting estimators will be examined with a focus on bias and relative efficiency. The robustness to model misspecification will also be examined.