A performance comparison of some high breakdown robust estimators for nonlinear parameter estimation

摘要

While the inevitable occurrence of departures from the assumptions made beforehand can damage least squares reliability, robust estimators will resist them. A number of al-ternative robust regression estimators have been suggested in the literature over the last three decades, but little is known about their small-sample performance in the context of nonlinear regression models. A simulation study comparing four such estimators to-gether with the usual least squares estimator is presented. It is found that the MM- and T-estimators are quite efficient when the proportion of outliers in data is not too large.