다차원 일반화부분점수 모형을 위한 척도연계 방법의 특성 및 상대적 기능 분석

초록

In item response theory (IRT), the multidimensional generalized partial credit (MGPC) model can be used for analyzing data from the partially-scored items with two or more response categories that are intended to measure multiple latent traits. The present study was purposed to present three scale-linking methods for the MGPC model under the context of common-item linking designs and investigate their characteristics and performances through computer simulations. The common-item linking methods presented are the least squares (LSQ), item-category response function (IRF), and test response function (TRF) methods. Computer simulations, including four factors of test type, multivariate ability distribution for the examinee population, sample size, and number of the common items used for linking, were conducted to examinee how well the three linking methods would recover the distribution parameters and the MGPC item parameters through scale linking. The simulation results suggested that in various linking conditions, the three methods should properly work for estimating the rotation matrix and translation vector of linear linking functions. However, overall, the relative performance of the three linking methods differed by the test type. For the test consisting of items of simple structure and items of complex structure, the IRF, LSQ, and TRF methods performed best, second best, and worst, respectively, in recovering the distribution and item parameters through scale linking. For the test consisting of items of complex structure only, the IRF method again performed best but the LSQ method performed worst.

키워드