İki kategorili maddelerde tek biçimli değişen madde fonksiyonu belirleme tekniklerinin karşılaştırılması: Bir simülasyon çalışması
Özet
The purpose of this study was to compare Type I error rates and power for Mantel-Haenszel (MH), logistic regression (LR) and item response theory likelihood ratio (IRT-LR) tests for reference and focal groups with varying ability distributions, sample size and sample size ratios.In this study WinGen3 was used to simulate ability estimates and to generate response data. The two-parameter logistic item response model was used to generate the item parameters for both non-DIF and DIF items. The same item parameters were used for both reference and focal groups. Uniform-DIF was created by altering b parameters between the reference and focal groups. Three variables were manipulated: ability distribution differences between groups, sample size and sample size ratio while four factors were fixed: number of groups (reference and focal), number of items, DIF-conditioning (uniform DIF), and matching variable. Taken together, 18 conditions were studied. A 3 (ability distribution) X 6 (sample sizecombinations) fully-crossed design was used. Each condition was replicated 100 times. Therefore, 1800 sets of 1-0 data were generated. Type I error rates of MH, LR and IRT-LR procedures were assessed on non-DIF items, and power of these procedures were assessed on DIF items.The results showed that with equal group distribution, MH and LR techniques had similar and better Type I error rate control than IRT-LR. On the other hand, with unequal group distribution, IRT-LR-DIF had better Type I error rate control than MH viand LR DIF procedures. Also, under the same conditions inflated Type I error rates were observed in MH and LR procedures.Key words: Differential item functioning, Type I. error, power, Mantel-Haenszel, Logistic Regression, Item Response Theory Likelihood Ratio, simulation