Four item selection methods based on KL information were proposed in this study for CAT based on the mixture Rasch model. Unlike the previous research, where either classification or ability estimation was the main focus, interest in this study was to consider both classification and ability estimation simultaneously. Simulation studies suggested the different methods were recommended under specific conditions.
Q:
1. In equation 1, the subscript is “jc”, indicating the ability/difficulty has different values for a specific person/item in different class. And then, what is the meaning of the ability and difficulty parameter? For a person, whether ability values in different classes are comparable?
2. The difficulty parameters for simulation were generated differently for latent classes, whether they share the same meaning on the same latent continuum? It is confusing to me that, for CAT, the item pool should be clean, but if difficulties are allowed to be varied across groups, whether this also suggests this item is not so good?
3. It is suggested that, the latent class membership may be caused by variables other than the intended latent variable, like speediness. Whether a multidimension model can be considered, which treats the intended measured variable as a dimension, and then the other grouping variables (speediness) as second dimension?