Comparing Predictors in Multivariate Regression Models: An Extension of Dominance Analysis

Abstract

Dominance analysis (DA) is a method used to compare the relative importance of predictors in multiple regression. DA determines the dominance of one predictor over another by comparing their additional R2 contributions across all subset models. In this article DA is extended to multivariate models by identifying a minimal set of criteria for an appropriate generalization of R2 to the case of multiple response variables. The DA results obtained by univariate regression (with each criterion separately) are analytically compared with results obtained by multivariate DA and illustrated with an example. It is shown that univariate dominance does not necessarily imply multivariate dominance (and vice versa), and it is recommended that researchers who wish to account for the correlation among the response variables use multivariate DA to determine the relative importance of predictors.

Understanding the Results of Multiple Linear Regression

“…First, it is possible that the study of some psychological phenomena requires multiple criteria as well as multiple predictors, leading to a complex canonical prediction problem (e.g., Azen & Budescu, 2006;LeBreton & Tonidandel, 2008;Nimon, Henson, & Gates, 2010). Second, another frequent concern is the reliability or stability of importance-weight estimates across independent samples to which a regression model is supposed to generalize (e.g., Azen & Budescu, 2003;Johnson, 2004).…”

Protocol for a study of the psychosocial determinants of health in early childhood among children with cystic fibrosis

“…The psychosocial variables will be hierarchically entered into the regression model according to their empirically and theoretically prescribed utility as predictors of child health outcomes. By entering predictors in this manner, statistical power will be optimized for the more important predictors (Bedrick & Tsai 1994, Afifi et al 2004, Azen & Budescu 2006. Intra-clinic clustering in the data will be controlled by using a sandwich estimator for the standard errors (Rabe-Hesketh et al 2005).…”

Assessing the contribution of variables in feed forward neural network

“…The IQR method along with the connection weight approach is implemented on a real life data set that was used earlier by Barber [27] and Azen and Budescu [28]. The data set containing a total of 689 records was used to determine the effects of parental indicators (measures of mother's and father's parenting styles) on youth outcomes (measures of psychological adjustment).…”