LEAST MEDIAN OF WEIGHTED SQUARES IN LOGISTIC-REGRESSION WITH LARGE STRATA

One aim of robust regression is to find estimators with high finite sa mple breakdown points. Although various robust estimators have been pr oposed in logistic regression models, their breakdown points are not y et known. Here it is shown for logistic regression models with binary data that there is no estimator with a high finite sample breakdown po int, provided the estimator has to fulfill a weak condition. In logist ic regression models with large strata however, a modification of Rous seeuw's least median of squares estimator is shown to have a finite sa mple breakdown point of approximately 1/2.