关键词: 篇章, 项目功能差异, 题组DIF, 效应值

Abstract: Structuring items around reading passages (the passage-based format) is enjoying more and more popularity in reading and language testing. The passage-based format is a typical example of testlets, groups of items that exhibit dependencies due to a common prompt, so statistical techniques used to analyze passage-based items, such as DIF detection, must account for the testlet effect. Current methods to deal with the testlet effect fall into two categories: the testlet model-based approach and the alternative/indirect approach. The former is more methodically sound, but is in its infancy and is not ready for practical purposes.
This study pursues the second path and accounts for the testlet effect in DIF detection by introducing alternative hypothesis testing for testlet DIF and indices its effect size. The first part is a comparison study between traditional DIF and the alternative testlet DIF, which demonstrates the necessity and advantages of the new approach.
The second part employs four testlet DIF procedures derived from the alternative approach to conduct hypothesis tests and investigates three indices to measure the DIF effect size. These indices can provide helpful information for DIF decision-making in testlets.
The results indicate that these alternative testlet DIF procedures are more technically sound than the non-testlet DIF procedures for passage-based testing.

Key words: passage-based, DIF, testlet DIF, effect size