비동등집단-개별가교검사 설계에서 IRT 능력척도 유지를 위한 다집단 가교-고정 추정과 대안적 추정의 비교

초록

When test equating is conducted based on an item response theory (IRT) calibrated item pool, a new test can be administered under the nonequivalent groups separate-anchor test design. Under this design, two or more nonequivalent examinee groups take the same operational test items but, by group, different sets of anchor items that come from the item pool. The purpose of this study was to present four IRT methods for using the anchor items to estimate the parameters of the operational test items on the common ability scale of the item pool and to investigate their relative performances through computer simulations. The four IRT methods included a multiple-group fixed-anchor estimation method and other three ‘0-1 estimation and scale-linking’ methods. Four simulation factors were considered: (1) type of IRT model, (2) nonequivalence level between two ability distributions, (3) sample size, and (4) number of anchor items. Main results were as follows. Except for the conditions in which the three-parameter logistic (3PL) model was used for analysis and two ability distributions much differed from each other, the four IRT methods almost equally performed in accurately recovering the true parameters of the operational test items. However, in the exceptional conditions, the multiple-group fixed-anchor estimation method outperformed the other three methods for the recovery of the item parameters.

키워드