Examining the Use and Utility of Dominance and Relative Weights Analysis

Abstract

How should researchers judge whether one finding is more important than another? Recent statistical methods purport to assess the relative importance of variables in an unambiguous and objective way. Relative importance analysis is a collection of statistical methods that decompose R2 and attribute a share of the explained variance to each variable in a regression equation. The variables can then be ranked. However, do these tools deliver on their promise? In this study, we review the literature to see how these techniques are used and investigate if relative importance analysis techniques better asses the relative causal importance of variables as compared to OLS. We provide multiple intuitive explanations as well as a thorough Monte Carlo simulation comparing relative importance analysis techniques against regression analysis; we show that regression provides a more accurate ranking of the relative causal importance of variables across a variety of scenarios. We conclude with practical guidelines on how to assess the relative causal importance of variables using the straightforward idea of comparable investments.