Estimation of the Cox Model Parameters Through Kernel Resampling

• Activity Number: 38
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Haiyan Bai*+
• Companies: University of Central Florida
• Keywords: proportional hazards regression ; Cox model ; Cox regression ; bootstrap ; kernel resampling

Abstract

The accuracy of parameter estimation of Cox regression models (Cox model) has been a concern. To improve the accuracy of the estimation, the bootstrap was implemented; unfortunately, prior research revealed inconsistent findings with the application of the bootstrap to a Cox model. The current study applies a new resampling method, the kernel resampling technique (KRT), to the Cox model. An empirical dataset is employed to compare the accuracy and stability of the estimation results through multiple replications from KRT with those from the bootstrap as well as the conventional asymptotic method. The study results revealed that KRT outperformed the bootstrap and the conventional asymptotic method in terms of improving the accuracy of the parameter estimations.