이요인 명명반응 모형을 위한 다차원 IRT 척도연계 방법

초록

The full-information bi-factor nominal response (NR) model is one of the confirmatory multidimensional item response theory (IRT) models. The purpose of this paper is to present three types of scale linking methods for the bi-factor NR model, which are used under the common-item design, and investigate their characteristics and performance through computer simulations. The linking methods presented are the direct least squares (DLS), mean/least squares (MLS), and item category response function (ICRF) methods. These linking methods estimate the dilation and translation coefficients of linear functions for each of the general and specific dimensions that link ability scales defined by different examinee populations. For the DLS and MLS methods, two versions were presented to estimate the translation coefficients, one that uses only the parameters of the highest response category for each common item, and the other that uses the parameters of all but the lowest categories. The simulation study included three factors of (a) level of nonequivalence in ability between populations, (b) sample size, and (c) number of common items. Three main results were found. First, for the DLS and MLS methods, the highest-category-only version estimated the translation coefficients more accurately than the all-categories version. Second, all of the DLS, MLS, and ICRF methods estimated the linking coefficients for the general factor more accurately than the linking coefficients for the specific factors. Third, across all simulation conditions, the ICRF method performed best in recovering the true linking coefficients and the item parameters of the bi-factor NR model.

키워드