In practice the unidimensional CAT has been manipulated in cognitive test by some test organizations. It is not necessary to let the item selection cross the different dimensions in cognitive test, but The non-cognitive test is appropriate to use multidimensional CAT. This study introduced four previous item selection method and one newly proposed item selection method in multidimensional CAT. Then, it set several conditions to investigate the factors affecting the effect of item select procedure, such as item pools, distributions of examinees, test length, and content balance method. For evaluating the five methods, the several indices including absolute bias, correlation, time waste, test reliability, and item usage etc., were observed. However, the result showed that the new method didn't perform better than the previous methods at correlation and absolute bias. At the item usage, the new method used the most items in pool. I think those results implied the new method didn't choose the items which are well adaptive for ability estimate.
comment:
1) We want to find the reason of this bad result, but it seems the author didn't explain why the V2 with the optimized weight cause the selection badly. Actually, the mathematics proposed by Yao 2010 is too difficult for me to reason the influence on item selection. Then, the future study didn't mention the reason should be investigate first before extending to the item exposure controlling.
2) There are really huge amount of information in Tables and Figures. Perhaps I should pay more time to read for fully understanding the trend of the interaction between those conditions and methods.