Buckley-James Type Estimator for Censored Data with Covariates Missing by Design

摘要

The Buckley-James estimator (BJE) is a well-known estimator for linear regression models with censored data. Ritov has generalized the BJE to a semiparametric setting and demonstrated that his class of Buckley-James type estimators is asymptotically equivalent to the class of rank-based estimators proposed by Tsiatis. In this article, we revisit such relationship in censored data with covariates missing by design. By exploring a similar relationship between our proposed class of Buckley-James type estimating functions to the class of rank-based estimating functions recently generalized by Nan, Kalbfleisch and Yu, we establish asymptotic properties of our proposed estimators. We also conduct numerical studies to compare asymptotic efficiencies from various estimators.