A Simulation Study to Compare the Sensitivity of Goodness of Fit Indices of Measurement Invariance: Partial Replication of Khojasteh and Lo (2015)

• In Event: Measurement Invariance, Model Selection, and Measurement Uncertainty in Structural Equation Modeling

Abstract

Measurement invariance (MI) provides valuable information to address cultural difference issues and is the key to many psychological research studies. Researchers had developed effective fit indices to evaluate MI. The current study tested the sensitivities of some fit indices (ΔCFI, ΔRMSEA, ΔSRMR, ΔGamma, and Δχ2) with Monte Carlo simulation. Experimental conditions included the number of factors/variables, sample sizes, factor loadings, and the percentage of non-invariant items. Results showed that the ΔCFI and ΔGamma are powerful and less sensitive to the sample size or non-invariance situations. There was inflation in Type I error in some models. ΔSRMR and ΔRMSEA are powerful only when the sample size is 1,000. ΔSRMR is sensitive to sample size and level of non-invariance.

Authors

• Xueying Gao, University of Iowa
  Presenting Author

• Stephen A. Sivo, University of Central Florida
  Presenting Author

• Debbie L. Hahs-Vaughn, University of Central Florida
  Non-Presenting Author

• Haiyan Bai, University of Central Florida
  Non-Presenting Author

• Laurie O. Campbell, University of Central Florida
  Non-Presenting Author