This study proposed a new method to find out the appropriate Q-matrix. A flexible T-matrix would be constructed and the information from the response data can be well used. The Q-matrix is estimated by minimizing the objective function, which minimizing the distance between model and the observed information. This method is interesting but not easy for me to understand.
Questions:
1. When using other models, like DINO or NIDO, whether the construction of Q-matrix would be affected? If so, whether the model-data fit should be assessed before choosing a model?
2. What is the maximum number of missing items in the Q-matrix?
3. Given the exploratory nature of this method, it is possible that a Q-matrix is defined, but it could not be interpreted. How to deal with this situation?
4. In equation (14), both the T and p are estimated, whether the estimations of these two would be confounded?
5 If compare with the Bayesian method, what is the advantage of the proposed method in this study?