EMPIRICAL STUDY ON ROBUST REGRESSION ESTIMATORS AND THEIR PERFORMANCE

Abstract

Regression Analysis is statistical technique to model data. But the presence of outliers and influential points affect data modelling and its interpretation. Robust regression analysis is an alternative choice to this. Here we made an attempt to study different robust estimators and propose a new robust reweighted Sn covariance based regression estimator. We have evaluated the performance empirically and the simulation study shows our proposed estimator is preferable to OLS and other robust regression estimators in terms of the MSE criteria. Also, proposed robust Sn covariance regression estimator produce outperforming results for regression equivaraince and breakdown criterion. Robustness of the proposed estimator is proved empirically. The proposed method is innovatively used to model fluid data. R software is used for simulation and study