Test Boyutluluğunun Değişen Madde Fonksiyonuna Etkisinin Farklı Koşullar Altında İncelenmesi

Abstract

The aim of this study is to determine how DIF is affected by different conditions in two dimensional and three dimensional simulation and two dimensional real data sets. Whether the tests are simple or complex, medium or high level of DIF effect size, sample sizes ranging from 600-1200-2400-3600-4800-6000, and having a focal and reference group ratio varying between 1: 1 and 1:2 are research conditions. Type one error and power values obtained from conventional SIBTEST and multi-dimensional multiple causes multiple indicators (Multi-MIMIC) methods were compared to two-category data sets produced in 2D and 3D models. In the analyzes performed with the real data set, the number of items with DIF determined according to DIF methods were compared. In order to obtain simulation data sets, item parameters were generated by using ITEMGEN program. Real data sets from PISA 2012 Turkey sample was selected as respectively 600, 1200 and 3600 and focus (girls) and reference (male) ratio of 1: 1 and 1: 2. The FACTOR 10.8 program was used to check the suitability of the data sets. SAS 9.4 and MPLUS 7 programs were used in the analyzes performed for simulation and real data sets. 100 replicates were made for simulation data sets. As a result of the research, type 1 error decreases as the sample size increases, DIF effect level increases and the structure of the test goes from simple to complex. Although power ratios do not differ significantly depending on sample size, decrease due to the increase in DIF effect level and increase with the complexity of the test structure. Parallel results were also found in the 3D data, but the effect of the test structure was not significant. In the 2D and 3D data, the Multi-MIMIC method, which is a multi-dimensional model instead of the classical method, has lower type 1 error and higher power ratios. In the real data set, the number of items with DIF did not change depending on sample size and focus-reference group ratio, but when the Multi-MIMIC method was chosen as the DIF method, it was found that there was a decrease in the number of items with DIF. As a result, the tests applied to sufficient sample sizes should be in simple structure and Multi-MIMIC method is recommended for DIF analysis. Keywords: multi-dimensionality, differential item functioning, SIBTEST, Multi-MIMIC