INFLUENCES OF ITEM NUMBER CONSTRAINT AND ITEM EXPOSURE CONTROL ON MEASUREMENT PRECISION OF ABILITY ESTIMATION AND ITEM EXPOSURE RATES IN MULTIDIMENSIONAL COMPUTERIZED ADAPTIVE TESTING
Chen, P.H., National Yunlin University of Technology and Wang, W.C., National Chung-Cheng University, Taiwan
The influences of item number constraint and item exposure control on the measurement efficiency of ability estimation and item exposure rates in multidimensional computerized adaptive testing (MAT) are investigated through simulation studies. Three levels of item number constraint and three levels of item exposure rater control were manipulated under four kinds of correlations between latent traits. Two kinds of item banks, between-item multidimensional and within-item multidimensional, were formed, which leads to between-item multidimensional adaptive testing (BMAT) and within-item multidimensional adaptive testing (WMAT). The dependent variables are measurement precision (i.e., reliability) and item exposure rate. Results show that the measurement precision of ability was not affected by the item number constraint nor by the item exposure control in the BMAT and WMAT situations under the long-test condition, but was reduced by these two constraints in the WMAT situation with the short-test condition. With respect to item exposure rate, the item exposure control worked well across all the conditions. The item number constraint made the exposure rates of items across dimensions similar. MAT yields more precise measures of ability than unidimensional CAT by utilizing the correlations between latent traits. For multidimensional items, only MAT is applicable. MAT is suitable in between-item multidimensional tests, such as test batteries, multiple achievement or aptitude tests, personality inventories, and within-item multidimensional tests, such as performance-based tests, and tests with items measuring more than one dimension simultaneously. This study sheds some light on future studies of MAT on theoretical and practical issues.