Regularization for Variable Selection in Multidimensional Factor and Network Models

• In Event: Methodological Innovations in Structural Equation Modeling

Abstract

This study investigates the efficacy of regularization techniques for variable selection in multidimensional data, specifically comparing regularized network models and regularized factor analysis models. While network modeling and factor analysis offer distinct interpretations, both methods use regularization to simplify models and prevent overfitting. Network modeling examines partial correlations to identify key nodes and pathways, whereas factor analysis uncovers underlying factors explaining observed covariances. Despite the prevalence of factor models in psychometrics, the actual underlying structure of empirical data often remains ambiguous, raising questions about the applicability of network models in such contexts. By conducting a comparative analysis of model fit and variable selection, this research aims to determine if network analysis can effectively handle data generated under factor model assumptions.

Authors

• jiaying chen, University of Arkansas

• Xinya Liang, University of Arkansas

• Jihong Zhang, University of Arkansas at Fayetteville