This paper introduced two kinds of item selection methods and proved them to be better performed in CD-CAT than some other methods. These two methods both based on the PWSL information index, one is deterministic and the other is probabilistic by employing the information interval. The full version of fusion model is selected for this study.
Qs and Fs:
1. Fusion model is one of most complicate models in CDM family, and it needs to estimate the ability profile and an additional latent ability parameter simultaneously. The additional latent ability in fusion model is defined as the abilities not specified in the Q-matrix, for a good fit item, this part should be as small as possible. I am wondering how to define and estimate this part of ability reliably in practice? What is the relationship between the ability profile and the additional latent ability?
2. What is the design to generate the additional latent abilities and difficulties in the simulation study? As there is no further description for this part.
3. How to select the first item for fusion model when using CAT?
4. When calculating the information for one item, whether the information index is separated into two parts, one for the attributes defined in Qmatinx and the other for the additional difficulty? Then how to select the next item according to the information index? Based on the maximum information summing up two parts?
5. For further study, some more straightforward CDM models could be used, and then varying the conditions, like varying the correlation between attributes, or variable-length CAT. This kind of design may be more closed to the practical situation.