Issues of Misspecified Measurement Error Models for Survival Data with Covariate Measurement Error

• Activity Number: 77
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Ying Yan*+ and Grace Y Yi
• Companies: University of Waterloo and University of Waterloo
• Keywords: Measurement Error ; Model Misspecification ; Bias Analysis ; Test of No Treatment Effect

Abstract

Cox proportional hazards models with covariate measurement error have been studied extensively, and most methods that correct for measurement error effects are developed by assuming a classical additive error model for the measurement error process. It is not clear what the impact would be if the assumption of the classical additive error model is incorrect. In this talk, we adopt bias analysis techniques to study the impact of misspecifying the classical additive measurement error models, and find that error model misspecification could seriously bias the existing error-corrected estimators. We further use these techniques to investigate issues associated with hypothesis testing when some covariates are error-contaminated. Simulation studies and a real data example are presented.