多次元項目反応モデルにおけるプロクラステス回転を応用した共通受検者による等化法の比較

抄録

Multidimensional item response theory model (MIRT) is a sub-model of Item Response Theory (IRT). It assumes that multidimensional latent traits influence test item responses. Equating or linking, which calibrate the parameters of two or more tests, are critical to the application of IRT, because they allow for comparisons between test scores. However, little research has been conducted for MIRT equating, especially on the common examinee design. The present study describes three equating methods for MIRT, which utilizes rotation methods in factor analysis focusing on factor scores, by employing the common examinee design, and analyzes the characteristics of each method through computer simulations. The results of the simulations reveal that the Weighted Procrustes method is effective when correlations between equated scores of each dimension are expected to approximate the target ones.