comments:
1) the first five items are randomly selected. I think this setting is not necessary because it actually didn't have a good performance in exposure rate control. Even the other exposure control methods were suggested in the discussion. This setting just was a kind of placebo for exposure control. I expect the situation without this setting will be similar with this setting.
2) The speedness behavior is a kind of latent class. I ever think that we can add this condition by setting the only last several items with latent class model. The problem is we don't know which items will be select at the last several demonstrations. However, is it reasonable or available that the estimation and information for latent class only are used at the last several items?
3) The items that are very easy for one class but very difficult for the other class will provide much more KL information for latent class classification. That means the best correlation between item difficulties for the two latent classes is -1. I am curious which example of situation in reality would yield this kind of item parameter.
4) When the latent classes are ordered in terms of ability, I think using standard setting method associating with traditional CCT will give examinee more clear classification and illustration of ability level.