New Model Evaluation Tools in Component-Based Structural Equation Modeling

• In Event: 15.059 - Fairmont Royal York Roundtable Session One
  In Roundtable Session: 15.059-7 - Specialized Structural Equation Modeling Applications: Heterogeneous Cross-Lagged Data, Generalized Structured Component Analysis, and Item Wording Effects

Abstract

In SEM, the ongoing debate comparing the merits of a theoretical confirmatory approach and a practical exploratory approach continues, demonstrating that the specification of a true model cannot yet be proved. In line with this goal, this paper focuses on model evaluation in generalized structured component analysis (GSCA) that is a component-based approach to SEM. Despite its growing popularity, GSCA currently relies on only a handful of descriptive measures for model evaluation. We introduce three new variable selection criteria (NVSC) that are based on bias and residual variance measures, and also demonstrate the usefulness of three NVSC in GSCA using children’s social skill data from an early childhood longitudinal study – Kindergarten: 2011.

Authors

• Ji Hoon Ryoo, University of Southern California

• Heungsun Hwang, McGill University