Applying Design-Based, Model-Based, and Maximum Confirmatory Factor Analysis Modeling on Multilevel Measurement Data

• In Event: 63.085 - Poster Session 13
  In Poster Session: 63.085-4 - Modeling Nonindependent Data

Abstract

We compared the robustness of model-based, design-based, and maximum models in analyzing multilevel measurement data with level-varying factor loadings under equal level structures. The purpose of this simulation study was to test what model specification can detect factor loading that are trivial or of little importance in a practical sense in the within level. The true parameters were specified based on the empirical MCFA analysis of 120 unbalanced 3rd graders’ Harter competence scale. Results showed maximum models was robust to level-varying factor loadings while the design-based and the misspecified model-based approach produced conflated results and statistical inferences across simulation. The practical suggestion of constructing adequate models on empirical multilevel measurement data will be provided in the presentation/poster in annual meeting.

Authors

• Jiun-Yu Wu, National Chiao Tung University

• Jr-Hung Lin, National Chiao Tung University

• Yuan-Hsuan Lee, National Taichung University of Education