Investigating the effects of testlets of an English reading comprehension test on the parameter estimation:A comparison between conventional IRT and Testlet IRT approaches

서지정보

· 발행기관 : 한국교육평가학회
· 수록지 정보 : 교육평가연구 / 30권 / 2호 / 341 ~ 362페이지
· 저자명 : 최세일

초록

Many English language tests consist of testlets in which a group of items share a common written or spoken text. This testlet format can enhance efficiency of test development but could increase inter-item dependence within a testlet, which violates the key assumption of local independence in item response theory(IRT) modeling. Theorists have proposed several models to cope with the testlet structures and warned against test scoring ignoring the violation. Recently, however, there also exist some concerns about modeling negligible testlet effects that make the scoring process unnecessarily complicated and bring about errors in item calibration. Using an English reading comprehension test, this study empirically investigated the magnitude of tetlet effects on the key parameter estimation by comparing the estimates of the regular IRT and the testlet response theory model proposed by Bradlow et al.(1999). The results showed that the testlet effects were very small. Item difficulties and person parameters were almost intact while item discrimination and guessing parameters were slightly overestimated. Test information was locally overestimated but such overestimation was due to the estimation method chosen, not to the testlet structure itself. Some implications of the results were further discussed.

참고자료

· 없음