Robust Estimation of a Linearized Nonlinear Regression Model with Heteroscedastic Errors:A Simulation Study

Abstract

A simulation study is used to examine the robustness of some estimators on a linearized nonlinear regression model with heteroscedastic errors, namely the Linearized Ordinary Least Squares (LOLS), Transformed Generalized Least Squares (TGLS) , Linearized Reweighted Least Squares (LRLS) and Transformed Linearized Reweighted Least Squares (TLRLS). The latter is a modification of Reweighted Least Squares (RLS) based on Least Median of Squares (LMS). The empirical evidence shows that the first three estimators are not sufficiently robust when the percentage of outliers in the data increases. That is, they do not have a high breakdown point. On the other hand, the modified estimator (TLRLS) has a higher breakdown point than the other three estimators.