Iterative Linking with the DifferentialFunctioning of Items and Tests (DFIT) Method: Comparison of Testwide and ItemParameter Replication Critical Values
Jocob Seybert & Stephen Stark
In the differential functioning of itemsand tests (DFIT) method framework for detecting DIF, a noncompesatory DIF(NCDIF) index was used to flag DIF items. The index was tested to besignificant or not using a chi-square test. In previous studies, a ‘testwide’critical values which are determine by the test characteristic wererecommended. However, the testwide critical values are too sensitive to samplesize, hence, a critical value for individual items are computed on the numberof response categories. The item-specific critical values can be obtained via anew item parameter replication (IPR) method. The present study compared the TypeI error rate and power of DFIT method using testwide critical values withitem-specific critical values obtained using the IPR method under variousconditions.
Results indicated that testwide andIPR-based item critical values provided similar power for detecting DIF items.However, item-specific method with IPR had generally lower power when DIF wasdue to discrimination parameters. IPR-based critical method provided as good asor better control of Type I error under most conditions, and using two-stageand iterative linking will improve DIF detection.
Question and Comments:
1. BILOG-MG3 was used to estimateparameters in dichotomous items, which implements EMD method in detecting DIFitems.
2. The IPR method with nominalalpha of 0.01 was very conservative, which was not mentioned in the study.