Relative Variable Importance Measures for Multiple Linear Regression in Suppression Situations: A Monte Carlo Study

• In Event: Roundtable Session 41
  In Roundtable Session: Simulation Studies

Abstract

Regression is one of the most widely used statistical techniques in psychosocial research. Researchers often want to know the relative importance of independent variables in a regression model. Two widely cited methods for answering this question are the Pratt index and dominance analysis. In this Monte Carlo simulation study we investigate how these methods assign importance to predictors in complex multiple regression situations often referred to as “suppression.” Results show that although the two measures may agree in relatively simple cases, they assign importance in very different ways for more complex cases. We encourage the use of variable importance measures to supplement regression analyses but highlight the importance of addressing conceptual issues regarding these measures, particularly in cases of suppression.

Authors

• Benjamin R. Shear, Stanford University

• Oscar L. Olvera, The University of British Columbia

• Bruno D. Zumbo, The University of British Columbia