Effects of dimensionality and covariate on items with DIF in mixture models

OpenAlex · İngilizce

The aim of this study is to determine the differential item functioning (DIF) with a mixture model when the data set is multidimensional. The differences in determining the number of items with DIF and the source of DIF according to the status of considering dimensionality and adding the covariate to the analysis were examined. In this context, a total of 28 items of mathematics and science answered by 7965 individuals in the 3rd booklet of the electronic Trends in International Mathematics and Science Study (eTIMSS) 2019 were found to have a multidimensional structure, and the variable with the highest correlation with the data structure was determined and included in the model as a covariate. In order to select the most appropriate models for the data set, models with different numbers of latent classes belonging to the mixture model and multidimensional mixture model including the covariate were compared. Descriptive statistics of the latent classes created with the selected models were created, item parameters were examined and DIF analysis were conducted. In the light of the findings, it was determined that the number of items with DIF decreased as the model became more complex. In the model with the best knowledge criterion index, it was found that the items with DIF at the knowing level generally differed in favor of the focal group, while the items with DIF at the application and reasoning levels differed in favor of the reference group.