In this study, the SME location estimates were used to replace the statement parameters estimated by MML methods. When applying the SME location estimates in the ZG IRT models for UPPs, results showed that, using SEM and MML estimates came up with highly correlated trait scores; and the SEM rating errors did not have much adverse effect on scoring when using in the adaptive testing context.
It seems the raters are proficient to provide accurate estimates about the location of the items in this study. In reality, as the true value of item parameters would never known, how to insure the quality of the raters’ estimates? How to take into account the variance across estimates from different raters?
In this study, the sample items are under single dimension and they are ordered in difficulties. The probability of choosing a statement based on the distance between the person’s location and the locations of the statements. The two statements are paired to act as a single item. If these two statements are presented separately to the person, what would happen? As the location of the statements are in the same scale with direction. What the difference is between presented them together to the person, and presented them separately?