Four item selection methods were proposed to achieve higher accurate classification by the information from class-specific latent ability estimation. Followed with four manipulated item pools, pool with item difficulty separation smaller will lead to higher accuracy on classification. In terms of accuracy, there is no evident difference except on pool 2.
1. Multidimensional latent trait in mixture model should be considered in future study, and the KL methods should be revised for use.
2. The KL or reverse KL method was not used with theory indicating which one is better. May it be possible to use the two alternatively during the CAT?
3. The study assumes only two latent classes existing and its performance in accuracy may depend on the item pool. Although the study suggests constructing item pool with more mush disparate items, it is a bit interesting to learn if three or more latent classes exist instead.
4. In classification testing, we usually set up a cut point to classify people. In the latent trait framework, what is the standard and its meaning here?