The developments for cognitive diagnosis model in CAT have been growing. The item exposure control is still with a lack of consideration, thus the paper applied two item exposure control methods in CDM-CAT. One of the CDMs called Fusion model was used. As the matrix of mastery of attributes was design, the responses were collected, slipping parameter and the guessing parameter can be estimated. In CAT, the slipping and guessing parameter were assumed fixed. The aim is to measure the mastery of attributes the respondent has. For item selection, the paper adopted the Kullback-Leibler (KL) information rather than traditional Fisher information. The restrictive progressive method is trying not to choose highest information item in early stage to save available items for later stage. As CAT is going on, the method will have higher possibility to choose items with highest information. The restrictive threshold method is trying to set up a set of items within a wider interval in early stage. As CAT progresses, the interval is getting smaller. The aim is as similar as restrictive progressive method but in probabilistic approach.
For item exposure rates, the restrictive progressive outperformed other methods. All exposure rates of items were well controlled under .1. At the same time, the accuracy of attribute pattern was reduced as well.
Qs:
1. If sample size is much smaller than in this study, will the results be different?
2. It seems that the restrictive progressive method works well because of the author add a term to keep maximum exposure rate in control. But the overlapping rates may not be well controlled.
3. A new way of controlling exposure rate and overlapping rate was proposed by Chen (2010). It is worthy of applying it in CDM-CAT.
4. The important parameter is a stochastic one. It is so important in CDM-CAT?