The Effects of Multicollinearity and Validity Concentration on Prediction Accuracy in Multiple Regression

• In Event: Topics in Multiple Linear Regression

Abstract

Morris and Lieberman’s 2013 study contrasting the two-group cross-validated classification accuracies of five algorithms, least squares, ridge regression, principal components, a common factor method, equal weighting and logistic regression, will be extended to investigate these methods, less logistic regression, on cross-validated prediction accuracies in multiple regression utilizing a continuous criterion.
From the evidence collected so far, it is validity concentration, rather than collinearity, that causes decrements in cross-validated classification accuracies for OLS and LR methods in respect to alternative methods. As well, these same decrements can be seen with higher validity concentration, even if collinearity is not present. As multicollinearity only involves predictor variables, results in predicting a continuous criterion are expected to be similar.

Authors

• Mary G. Lieberman, Florida Atlantic University

• John D. Morris, Florida Atlantic University