Forward Selection Procedure for Linear Model Building Using Spearman’s Rank Correlation

Abstract

Forward selection (FS) is a step-by-step model-building algorithm for linear regression. The FS algorithm was expressed in terms of sample correlations where Pearson’s product-moment correlation was used. The FS yields poor results when the data contain contaminations. In this article, we propose the use of Spearman’s rank correlation in FS. The proposed method is called FSr. We conduct an extensive simulation study to compare the performance of FSr with FS. The proposed FSr performs better than the FS algorithm in the contaminated data. We also demonstrate a real data application of FSr.