Bayesci Madde Tepki Kuramı Modellerinde Ölçme Değişmezliğinin İncelenmesi
Özet
The aim of this study is to examine measurement invariance in Bayesian IRT models with Bayes factor in simulated data generated under different conditions and to examine measurement invariance in real data with MGCFA and Bayes factor. For this purpose, within the scope of Bayesian IRT models, measurement invariance was examined with Bayes factor in simulated data at different sample sizes (n) and when the difference between item difficulty parameters (dk) was 0.00, 0.10, 0.30, 0.50 and 0.70. In real data, pairwise group comparisons were performed with both MGCFA and Bayes factor. Real data application was conducted with TIMSS 2019 mathematics achievement test. Bayesian analyses were conducted with the R software, and MGCFA was conducted with Mplus. According to simulation data results, in the case of dk=0.3 and n=1500 or more, the Bayes factor evidenced that measurement invariance was not achieved. In the case of dk=0.5, the Bayes factor results were found to vary based on group size. In large samples, the Bayes factor was found to be more likely to evidence that measurement invariance was not achieved. In the case of dk=0.7, the Bayes factor results evidenced that measurement invariance was not achieved regardless of group size. According to the MGCFA results, measurement invariance was achieved at the metric invariance level in Chile-Turkey and Turkey-Singapore samples, and at the configural invariance level in Chile-Singapore samples. The Bayes factor results of Chile-Turkey and Turkey-Singapore samples were generally consistent with those of the simulated data, but the Chile-Singapore results were not.
