Approaches to robust estimation in the simplest variance components model

Abstract

The simplest case of a random effects or variance components model is given by g groups with random location. We apply several general principles which can help to obtain robust methods to this situation. Two approaches, introduced by Rocke (Biometrika 70 (1983) 303-309) and Fellner (Technometrics 28 (1985) 51-60) are described and elaborated. A small simulation study suggests that considerable gains can be obtained by using robust procedures.