1 paper
Ken Akira Fujimoto, George Karabatsos
Typical IRT rating-scale models assume that the rating category threshold parameters are the same over examinees. However, it can be argued that many rating data sets violate this…