Alternative Robust Variable Selection Procedures in Multiple Regression

Most of the commonly used linear regression variable selection techniques are affected in the presence of outliers and high leverage points and often could produce misleading conclusions. This article proposes robust variable selection methods, where the suspected outliers and high leverage points are identified by regression diagnostics tools and then the best variables are selected after diagnostic checking. The performance of the proposed methods is compared with the classical non-robust criteria and the existing criteria via simulations. Furthermore, Hawkins-Bradu-Kass data set was analyzed for illustration.