A Monte Carlo Investigation into the Performance of Structural Equation Modeling Trees

• Activity Number: 519
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Holmes Finch*+ and Brian French
• Companies: Ball State University and Washington State University
• Keywords: Structural equation modeling ; Recursive partitioning ; Model based recursive partitioning ; SEM trees

Abstract

Structural equation modeling (SEM) is a widely used method relating observed indicators to latent factors (e.g. intelligence), and relating factors to one another. SEM trees (SEMT) is a modeling approach combining classification and regression trees (CART) with SEM. SEMT partitions a sample based on one or more partitioning variables (e.g. gender) using parameters of a SEM, concluding with multiple terminal nodes that differ based on parameters such as factor loadings and structure coefficients. Such partitioning can assist in identifying model non-invariance in the population. This Monte Carlo simulation study assesses the ability of SEMT to correctly partition samples using both categorical and continuous partitioning variables. Manipulated conditions include the number of observed indicators per factor, number of factors, model complexity, sample size, type of partitioning variables (2 categories, 4 categories, continuous), and number of partitioning variables. Results demonstrate accuracy of SEMT based on parameter estimation accuracy (i.e., bias, root mean squared error) and the correct number of identified of groups.