The performance of item selection methods in multidimensional computerized adaptive testing has only been studied using an independent cluster multidimensional structure. The goal of this study is to examine the effect of four different item selection methods on test utilization and measurement accuracy under more complex multidimensional data structures. The Kullback-Leibler information method, the minimum angle method, the volume method, and a method that minimizes the error variance of the linear combination were included in the study as item selection methods. We simulated four two-dimensional factor structure conditions: (a) independent cluster, (b) approximate simple, (c) complex, and (d) general factor, while varying the magnitude of the correlation among the dimensions. In general, it was found that the type of data structure played a major role, the magnitude of correlation played a moderate role, and the type of item selection method played a minor role on the research outcomes. The results show the importance of considering more complex data structures in operational MCAT applications.
