American Statistical Association New York City Metropolitan Area Chapter Mailman School of Public Health Columbia University Department of Biostatistics Colloquium

Extensions of the classical Least Squares regression estimator to the case of censored responses were first proposed by Miller (1976) and Buckley and James (1979). More recently, Ritov (1990) and Lai and Ying (1994) studied more general M-estimators for censored responses. Unfortunately, these estimators require monotone estimating equations and hence may be affected by high-leverage outliers. We propose an extension of high-breakdown regression estimators to the case of censored responses that can resist the effect of high-leverage atypical observations. In particular, we develop extensions of the LMS, S-, M- and tau-estimators and discuss an algorithm to compute them. Simulation studies show that these estimators have good finite sample properties.