Comparative Performance of MM and Least Trimmed Squares (LTS) Robust Regression Methods

Abstract

This study compares the performance of MM and Least Trimmed Squares (LTS) Robust Regression methods with the Ordinary and Modified Least Squares. The data of Hawkins-Bradu-Kass (1984) were used for the investigation. This is a much referenced data of 75 observations with one response variable and three independent variables. These data are known to be quite troublesome in terms of masking and swamping. Masking refers to bad data points being camouflaged because they are clustered; while swamping refers to good data points which appear to be outliers. In the study, it was shown empirically
that LTS performs better than the other mentioned methods in terms of finite efficiency, goodness - of - fit and breakdown point.