Relative Importance of Predictors in a Multilevel Model

• In Event: Multilevel Modeling Methods and Applications

Abstract

The Pratt index has been shown to be a useful and practical strategy for day-to-day researchers when ordering predictors in terms of importance in a multiple regression analysis. However, variable ordering cannot be used in multilevel models because conventional multilevel techniques do not have an R-squared equivalent to the one in multiple regression, and within- and between-level correlations are not available. The newly developed multilevel regression approach based on a structural equation modeling (SEM) framework makes it possible to apply the Pratt index in multilevel models. The purpose of this study is to demonstrate the capacity of the Pratt index to assess the relative importance of predictors in multilevel model analysis.

Authors

• Yan Liu, The University of British Columbia

• Bruno D. Zumbo, The University of British Columbia

• Amery Dai Ling Wu, Paragon Testing Enterprises