Dominance Analysis for Latent Variable Models: Comparison of Methods With Categorical Indicators and Misspecified Models

• In Event: Linear, Nonlinear, Latent, and Machine Learning Modeling for Educational Research

Abstract

Dominance analysis (DA) is used for ordering independent variables in a regression model based upon their relative importance in explaining variability in a dependent variable. DA has recently been extended to latent variable structural equation models. Research demonstrated that this new approach yields accurate results for latent variable models with normally distributed indicators and correctly specified models. The current study was designed to compare this approach to a method based on observed regression DA and DA for a 2-stage least squares estimator, with categorical indicators and/or model misspecification. Results indicated that DA for latent variable models can provide accurate ordering of latent independent variables and correct hypothesis selection when indicators are categorical and models are misspecified.

Authors

• W. Holmes Finch, Ball State University

• Maria E. Hernandez Finch, Ball State University