The characteristics of common items are always considered as an important factor affecting the quality of scale linking between tests. Although many studies have focused on the common item selection via the Unidimensional Item Response Theory (UIRT) models, seldom did researchers investigate the selection in the vertical scaling under the framework of Multidimensional Item Response Theory (MIRT). This study examines different common item selection methods when the correlation among proficiencies varies at different levels and when the content structures in tests are either identical or different. With respect to the recoveries of the probability matrix, item parameters and effect sizes, the results show that (1) full content coverage in the common item set is important, no matter whether the content structures are identical or not, (2) high correlation among proficiencies could partly compensate for the adverse effect caused by the common items not covering all content domains, and (3) the common item set covering all content domains with medium difficulty items yields better linking results and common items with high item-total-test correlation also perform well when the content structures are identical for both tests.
Read
