ROBUST CRITERION FOR VARIABLE SELECTION IN LINEAR REGRESSION

In this paper, the variable selection problem in multiple linear regressions is considered. Variable selection method is widely used in data analysis. Many subset selection methods are available in the literature, but majority of methods are based on the least squares estimator of regression coefficients. However, when data contain an outlier observation, the performance of least squares estimator is poor. Consequently, method based on this estimator tends to select a 'wrong' subset. In this article, we propose a simple and general method called Kp-criterion for variable selection in the linear regression. The main feature of the proposed criterion is that it can be used any type of estimator of the regression coefficients without any modification in the proposed criterion. The method is illustrated with numerical examples.