This study extended the hierarchical rater (HRM) model by replaced a different signal detection theory (SDT) model. Its structure likes the HRM which includes two levels, the SDT model functioned in the first level and the item response theory (IRT) model functioned in the second level, and called the HRM-SDT model. That is an essay produced by an examinee is viewed as belonging to one of the categories defined by the scoring rubric, and the (true) category that the essay belongs to is in turn an indicator of the examinee’s proficiency. The difference between the two models is that different SDT model was used. For the HRM model, there is a limitation is that the SDT model cannot distinguish the types of rater effects. The proposed model resolved this problem and can be found in the results.
Comments & Questions:
1. The maximum likelihood estimator and posterior mode estimator were introduced in estimation of parameters in this study. But for setting of prior distribution for the posterior mode estimator, I don’t understand why using the Dirichlet distribution. Is there a familiar distribution can replace it?