Study 1 investigated the recovery of item location by ZG model and SME method. The results showed that the scores estimated from ZG model and SME have high correlation; Study 2 investigated the random error of item parameters produced by estimation used in CAT setting. The item location parameters estimated from the ZG model and SME method were used in CAT.
1. The ZG model is interesting and appealing. It does not obviously consist of slope parameter but bear the property. As we know, the slope parameter is a tricky mean to make ipsative theta become comparable to other thetas. I did not see any excuses of explaining why the model can produce comparable thetas.
2. Ignoring why the theta of the model is comparable, it is interesting to extend this model to multidimensional one.
3. The CAT for the ZG model is a crucial application. The item selection methods have been developed for decades. We can apply several methods such as KL information or minimum-variance selection to compare the effectiveness of these methods in ZG model.
4. Although it is explained that the concept of ideal point was used in ZG model, We can see that the ICC of ZG model does not show a single peak. For information function, it does not show bimodal curve as well.
5. Even the estimates of thetas of the ZG model and those of SME have high correlation, it seems not a evidence the estimates of thetas consist of little random error. It may happen we used the inappropriate model (e.g., ZG model) to estimate the parameters.
6. In CAT study, no weird item location parameters were involved within SME conditions. That is, the estimates of location parameters are very different to each other between ZG model and SME. However, it actually happened in study 1. It may be worthy of consideration.
7. Although the SME method sounds interesting, it may be a more annoying to train the experts to be reliable. As we learned from standard setting, there are many problems about using the experts’ judgments such as inconsistence or central tendency.