| Summary: | 碩士 === 國立彰化師範大學 === 統計資訊研究所 === 100 === Random effects of the testlet response model usually affect the probability of a correct response to an item. Therefore, the potential testlet-based DIF may not be identified appropriately. The purpose of this study is to propose a modification of SIBTEST to reduce the influence of random effects as well as to improve the power of SIBTEST in detection of testlet DIF. The suggested method applies the linear regression on the valid subtest scores of the two examinee-groups to re-calculate DIF estimates. It also utilizes the bootstrap procedure to derive the estimation error and test statistic. The factors of the simulation study include the variances of random effects, the number of DIF items, the sample sizes, the potential amount for testlet DIF, and the group ability difference. In terms of type I error rate, the refined approach is performed reasonably well. The average rejection rates are approximated to the significant level of 0.05. From the perspective of power study, the proposed method outperforms SIBTEST. Particularly, the powers of SIBTEST are decreasing when the variations of testlet effect are increasing and the adjusted method is able to improve the power dramatically. The results show that the refinement of SIBTEST in detecting testlet DIF is reliable.
|
|---|
