Investigation of intercultural measurement invariance with factor analytical latent class and IRT models: TIMSS 2019 Students Like Learn Mathematics Scale example

This study, it was aimed to examine the TIMSS 2019 application on the samples of Singapore, Hong Kong, Japan, Norway, Turkey, and the Republic of South Africa with factor analysis, latent class analysis, and item response models of intercultural measurement invariance. This study, which was carried out as a descriptive research type, assessed the cross-cultural measurement invariance and estimated the model structure based on the data from the TIMSS 2019 application student surveys. Initially, using MG-CFA, a method for factor analysis, it was determined whether measurement invariance had been attained at the scale level across all comparison groups. Second, it has been found that the latent variable, which is included in the latent class models, has a three-class latent structure with LCA. The variable with a three-class latent structure was then tested using MG-LCA to see if it provided measurement invariance at scale and item level across all comparison groups. To ascertain the measurement invariance at the item level with IRT-likelihood, which is covered by item response theory, DIF analyses were conducted in the third stage. As a result of the analysis of the research, strict invariance between Singapore-Hong Kong and Hong Kong-Norway cultures and structural invariance between Turkey and the Republic of South Africa was ensured according to MG-LCA; Metric invariance was detected between all other comparison groups. Using the use of MG-LCA, partial homogeneous models from 15 comparison groups were found to have a good fit between the cultures of Singapore and Hong Kong, Singapore and Norway, Singapore and Turkey, Singapore and South Africa, Hong Kong and Japan, Hong Kong and Norway, and Norway and Turkey. whereas demonstrating that the heterogeneous model's level of concordance was maintained among the other comparison groups. As a result of the MG-LCA and IRT-likelihood analyses at the item level, it was determined that all items showed DIF between the comparison groups, and in this respect, the results were consistent with each other. The participants in the South African Republic, Turkey, and the Japanese samples were found to have a particular response style, according to the results of the latent class analysis; however, a dominating response style was not seen in the samples from Singapore, Hong Kong, and Norway. MG-LCA is a partial homogeneous model resulting in the model with a 100% probability if both comparison groups do not have a dominant response style, a partially homogeneous model with a probability of 45% if one of the comparison groups does not have a response style, and a heterogeneous model with a probability of 100% if both groups have a response style. If one of the MG-CFA pairwise comparison groups showed a response style and the other did not, it resulted in metric invariance with a probability of 100%. Strict invariance with a probability of 67% if neither paired comparison group showed a response style; Metric invariance was observed with a probability of 33%. If both paired comparison groups show a response style, 67% probability of metric invariance; Structural invariance was observed with a probability of 33%.