MIRT 등급반응 모형을 위한 공통-문항 척도연계 방법

초록

In applications of multidimensional item response theory (MIRT), the multidimensional graded response (MGR) model is often used for analyzing graded response data from the polytomous items that are intended to measure multiple latent traits. The purpose of this study is to present alternative scale-linking methods for the MGR model and investigate their characteristics and performances through computer simulations. Three scale-linking methods, based on common/linking items, are presented: (a) least squares (LSQ), (b) item-category characteristic function (ICF), and (c) test characteristic function (TCF). These methods are the extensions of those for the multidimensional two- or three-parameter logistic (M2PL or M3PL) model. Computer simulations that included three factors of (a) multivariate ability distribution for examinees, (b) sample size, and (c) number of common items between linked test forms were conducted to examinee the characteristics and relative performances of the three scale-linking methods. Overall, the simulation results suggest that the three scale-linking methods for the MGR model should properly work for estimating the rotation matrix and translation vector for linking two MIRT ability scales. As a main result, the ICF or TCF method performed best in recovering the underlying multivariate ability distributions through scale linking, whereas the ICF method performed best in recovering the item parameters on the MGR model. A computer program, called STMIRT, is introduced that was developed for MIRT scale linking, and some suggestions for future research are provided.

키워드