The Effect of Estimation Methods on SEM Fit Indices.

Area of Science:

• Psychometrics
• Statistical Modeling
• Quantitative Psychology

Background:

• Structural Equation Modeling (SEM) is widely used for analyzing complex relationships between variables.
• Model fit indices are crucial for evaluating the adequacy of SEM models.
• The choice of estimation method can influence SEM results and interpretation.

Purpose of the Study:

• To investigate the impact of different estimation methods (ML, ULS, DWLS) on population SEM fit indices (RMSEA, CFI, SRMR).
• To assess the robustness of these fit indices under various types and levels of model misspecification.
• To provide guidance on selecting appropriate fit indices and criteria for SEM evaluation.

Main Methods:

• Examined Maximum Likelihood (ML), Unweighted Least Squares (ULS), and Diagonally Weighted Least Squares (DWLS) estimation methods.
• Introduced misspecifications including incorrect dimensionality, omitted cross-loadings, and ignored residual correlations in factor analysis models.
• Evaluated the performance of RMSEA, CFI, and SRMR as model fit indices.

Main Results:

• Estimation methods significantly affected the Root Mean Square Error of Approximation (RMSEA) and Comparative Fit Index (CFI).
• Different cutoff values for RMSEA and CFI are necessary depending on the estimation method used.
• The Standardized Root Mean Square Residual (SRMR) demonstrated robustness across all tested estimation methods and misspecifications.

Conclusions:

• SRMR is a reliable fit index for evaluating SEM models at the population level, irrespective of the estimation method.
• Researchers should exercise caution when interpreting RMSEA and CFI, considering the estimation method employed.
• The findings advocate for the consistent use of SRMR as a robust indicator of model fit in SEM.