Multicollinearity's Effect on Regression Estimation Accuracy With Real Data Structures

• In Event: Division D: Multiple Linear Regression: The General Linear Model SIG

Abstract

Recommendations from popular statistics texts regarding avoidance of predictor variable multicollinearity in the use of multiple regression are considered from the perspective of the alternate purposes of explanation and prediction. For 21 data sets examined, the different effects of multicollinearity on prediction and explanation accuracy are detailed. As has been demonstrated previously, prediction accuracy (relative or absolute) is not affected by multicollinearity; but multicollinearity manifests an appreciable effect on explanation accuracy. Some consideration of the degree of effect at differing levels multicollinearity and an alternative that may partially ameliorate this loss in explanation accuracy are also considered.

Authors

• John D. Morris, Florida Atlantic University

• Mary G. Lieberman, Florida Atlantic University