This paper introduced the presented response time model. Most of current models implied a speed-accuracy tradeoff relationship. However, this phenomenon just appeared in a within -person condition. in the population level of person, it shows a positive correlation between speed and accuracy. Therefore, van der Linden proposed the hierarchical framework model in this study. The first level is the IRT model and lognormal response time model. In the second level, one is the covariance matrix for the person between ability and slowness parameter, and the other is the covariance matrix for the item between response and response time parameter. The advantage of this model is we can use any other more appropriate model to substitute for the IRT or lognormal response time model. The model is flexible in this framework.
1) We think about the method of application for CAT. because in CAT the item parameter is known, the lognormal response time model is almost the same with IRT model. Therefore, the item selection rule will be similar as two dimensional CAT, but the one of the dimension is substituted by response time model.
2) I expect that we will select the items with larger time discriminate parameter when we apply this framework in CAT. That means the items with smaller variance of response time will be administered easily. I still think whether this phenomenon is reasonable and necessary. Perhaps, the advantage of this rule is we can have more confidence to set time limit because the variance of time is small for those selected item. However, the disadvantage is the more difficult items will be select easily because test takers often stably consume many time in those items.