Effects of Some Coding Techniques On Multicolinearity and Model Statistics

Abstract

Two known methods of coding data for analyses in the presence of multicollinearity and evaluation of model performance viz: Dummy coding and Effect coding which are alternatives to each other were considered. Efforts were made to improve on their performances by modifying them as modified Dummy coding and modified Effect coding respectively and their performances of the now coding methods compared in this paper. The results show that all coding methods significantly reduced the effect of multicollinearity. The effect coding was found to be the best coding method in remedying multicollinearity while closely followed by the dummy coding. However, the proposed modified dummy coding gave the best R-squared values as well as F-values while still reducing the effect of multicollinearity to a great extent and closely followed by modified effect coding. The dummy and effect coding methods proved very efficient in remedying multicollinearity as their observed variance inflation factor (VIF) were all close to unity.

Keywords: Dummy coding, effect coding, multicollinearity, variance inflation factor.