Recognizing Uncertainty in the Q Matrix via a Bayesian Extension of the DINA Model.docx
The study investigated the performance on recovery when the some elements of Q-Matrix are treated as random rather than fixed. The authors conducted simulations of eight conditions to explore the phenomena aforementioned. The results illustrated that the Bayesian approach could highly improve attributes recovery when some elements were treated as random for other attributes are correctly specified. On the contrary, attributes are wrongly specified would affect the recovery of the Q-matrix. A real data was analyzed in the end. Several comments here I have: the first, it is nice that the authors applied signal decision theory in the DINA model. It’s interesting that slip parameter represents missing and guessing parameter represents false alarm in SDT. As the authors mentioned, the issue of attributes exists or not and attributes are wrongly specified or not can be mixed in the future study. It’s usually we manipulate some incorrect attributes specified and to evaluate the performance of recovery. Can we do in opposite direction? That’s once we make sure the specified proportion of correct attributes at least is correct, thus we can use the information to help attributes recovery. In the end, if the Q-matrix may misspecifed by experts, how can we know the “true” Q-matrix in reality not in simulation?