Testing Group Mean Differences of Latent Variables in Multilevel Data Using Multiple-Group Multilevel CFA and...

Area of Science:

• Social Sciences
• Psychology
• Statistics

Background:

• Group comparisons are fundamental in social science research.
• Analyzing multilevel data requires specialized methods for group comparisons.
• Existing methods may not adequately address complex multilevel structures.

Purpose of the Study:

• To evaluate two latent group mean testing methods for multilevel data: MG ML CFA and ML MIMIC.
• To investigate method performance under varying conditions of variance heterogeneity and intra-class correlations.
• To provide guidance for selecting appropriate multilevel multiple-group analysis approaches.

Main Methods:

• Conducted three Monte Carlo simulation studies.
• Examined multiple-group multilevel confirmatory factor analysis (MG ML CFA).
• Investigated multilevel multiple-indicators multiple-causes modeling (ML MIMIC).
• Manipulated factor and residual variances for between-group heterogeneity.
• Explored six model specifications for within-level multiple-group analysis, focusing on intra-class correlations.

Main Results:

• Both MG ML CFA and ML MIMIC demonstrated adequacy for multiple-group analysis with multilevel data.
• No significant performance differences were observed between the two methods in latent group mean testing.
• Simulation results generally supported the robustness of both approaches across tested conditions.

Conclusions:

• MG ML CFA and ML MIMIC are suitable and comparable methods for latent group mean testing in multilevel data.
• The study provides practical insights and guidelines for researchers applying these techniques.
• Further research can build upon these findings for advanced multilevel modeling.