Test-Based Interval Estimation Under the Accelerated Failure Time Model

Abstract

The accelerated failure time (AFT) model is an important regression tool to study the association between survival time and covariates. Semiparametric inference procedures have been proposed extensively in the literature. Recently, Zhou (2005a) proposed to use a model-based empirical likelihood approach to interval estimation for the AFT model. However, comparison was not made with more standard approaches based on score test and Wald-type test. In this article, we conduct extensive simulation studies to evaluate their relative performance in small samples. In addition to these methods, we also devise and consider model-free empirical likelihood approach for the comparisons. Our simulation results suggest that these empirical likelihood-based methods are not better than the standard approach based on the score test.