1. The meaning of the discrimination and difficulty parameters are changed from stage 1 to stage 2. For example in stage 1, b is the commonly defined difficulty parameter of getting the correct response of an item, which represent the difficulty of an item. In stage 2, b is the difficulty of revising an incorrect answer to a correct answer, which has nothing to do with the “item difficulty” any more.
2. In Figure 4, critical value of e increases as fewer review time. Does this mean that the less review time, the harder one can change an incorrect answer to a correct one, hence a stricter criterion we should use? I thought we should use a looser criterion in this case.
3. For polytomous models, the distribution of the total numbers of a certain pattern would be difficult to determine. It will not be a generalized binomial any more. So the statistical test for detecting cheating is hard to construct.