Comparison of DIF methods in different sub-groups according to sample sizes in rating scales

Some methods used in polytomous items were compared according to various samples and subgroups in determining the different item function (DIF), which is the first step in determining item bias, which is shown as one of the important validity problems in this study. For comparison, the 21-item short form of the 4-Likert-type Depression Anxiety Stress Scale (DASO-21) was used. The study group consisting of 2,087 university students reached online in the research was randomly divided into sub-samples of 250, 500, and 1,000 people. In these sample groups, each item was analyzed by the MantelTest, Poly-SIBTEST, Ordinal Logistic Regression and MTK-Likelihood Ratio methods, and DIF status in subgroups of gender, place of residence, presence or absence of chronic disease, and presence of significant disease. As a result of the analysis made, in the gender subgroup of DASO-21 in a sample of 1,000 people, 2 items in the Depression sub-dimension, 2 items in the Anxiety sub-dimension, and 1 item in the Stress sub-dimension; In a sample of 500 people, 1 item in the Depression sub-dimension, 2 in the Anxiety sub-dimension and 1 item in the Stress sub-dimension; In the sample group of 250 people, DIF was found in 1 item in the Depression sub-dimension, 2 in the Anxiety sub-dimension, and 1 item in the Stress sub-dimension. While DIF was not detected in any item in the sample of 1,000 people in the residential subgroup; In a sample of 500 people, 1 item in the Depression sub-dimension, 2 in the Anxiety sub-dimension, and 1 item in the Stress sub-dimension; While no DIF was detected in the Depression sub-dimension in the sample group of 250 people, DIF was detected in 1 item in the Anxiety sub-dimension and 2 items in the Stress sub-dimension. While no DIF was found in the Depression sub-dimension in a sample of 1000 people in the subgroup of whether or not they have a chronic disease, 1 item in the Anxiety sub-dimension and 1 item in the Stress sub-dimension; While DIF was detected in 2 items in the Depression sub-dimension in a sample of 500 people, DIF was not detected in any item in the Anxiety and Stress sub-dimensions; DIF was not detected in any item in the sample group of 250 people. In the subgroup of whether or not a significant illness is present, in a sample of 1,000 people, 1 item in the Depression sub-dimension, 2 in the Anxiety sub-dimension, and 1 item in the Stress sub-dimension; In a sample of 500 people, while no DIF was found in the Depression and Stress sub-dimensions, in 1 item in the Anxiety sub-dimension; In the sample group of 250 people, DIF was found in 1 item in the Depression subscale, 1 item in the Anxiety subscale, and 2 items in the Stress subscale. When the compatibility of the methods used in these analyses was examined, it was seen that DIF detection methods gave consistent results in all other methods, except for the subgroups of whether or not there is a chronic disease and whether or not there is a significant disease in a sample of 250 people. When the four methods used in the research were examined as to whether they gave consistent results according to the sample sizes, it was seen that all other methods, except the MantelTest method, gave consistent results in samples of 250, 500, and 1,000 people.