ESTIMATING THE RELATIVE IMPORTANCE OF VARIABLES IN MULTIPLE REGRESSION MODELS

摘要

Organizational scholars continue to be interested in examining relative importance of variables in multiple regression analysis (cf. Eby, Durley, Evans, et al., 2006; Luo, 2005; McAllister, Kamdar, Morrison, et al., 2007; Schleicher, Venkataramani, Morgeson, et al., 2006). Although the search for the proper measure of importance has been conducted for decades (e.g., Engelhart, 1936), variable importance in multiple regression contexts has traditionally been examined using bivariate correlation coefficients, standardized regression coefficients (i.e., beta-weights), squared standardized regression coefficients (i.e. , squared beta-weights), or product measures (i.e., the products of the correlation and standardized regression coefficients; Budescu, 1993; Johnson & LeBreton, 2004). However, these measures of relative importance are problematic when researchers are testing the importance of correlated variables, which is often the case in the organizational sciences (Johnson & LeBreton, 2004). More specifically, when used as measures of relative importance in regression models with correlated predictors, these indices do not correctly partition criterion variance (LeBreton, Ployhart, & Ladd, 2004) and therefore provide distorted estimates of predictor importance.