| Summary: | 碩士 === 國立彰化師範大學 === 統計資訊研究所 === 97 === The testlet-based tests usually do not conform to the assumption of local independence in the item response theory. Bradlow, Wainer, and Wang (1999) proposed a testlet response model to include random effects for the dependence between items within the same testlets. Because of the additional random effects, the probability of a correct response to an item is reduced. Therefore, DIF items may not be identified appropriately. The purpose of this study was to analyze testlet DIF using SIBTEST program. Simulation variables of interest include: the variances of random effect within the testlet, the number of DIF items, the potential amount for bias, sample sizes and the target ability difference. The results showed that the obtained testlet DIF estimates were small and close to the theoretical values. In the type I error study, the average rejection rates were also close to the significant level .05. In the power study, powers were decreasing while the variances of random effect were increasing. However, the average powers were between .59 and .94 in the cases with 3 DIF items. For testlets with 5 DIF items, the average powers were from .75 to .96. The results demonstrated that the performance of SIBTEST in detecting testlet DIF was reliable.
|
|---|
