On Robust Linear Regression with Incomplete Data

Abstract

In this paper, we use recently developed methods of very robust regression to extend missing value techniques to data with several outliers. Simulation experiments reveal that additional outliers may be imputed if one ignores the outliers already in the data. The combination of the forward search algorithm for high breakdown point estimators and the EM algorithm or multiple imputation for missing values can avoid problems of this kind. Some multiple deletion diagnostics for linear regression with incomplete data are also discussed.