The Rasch model only includes two parameters, person proficiency parameter and item difficulty parameter, and not includes guessing parameter. The item difficulty will be underestimated when guessing exists. Based on the idea of the Anderson’s theorem, they proposed a statistic for testing whether item estimates are affected by guessing. Defining guessing is a function of relative person proficiency and item difficulty and expected that the item difficulty estimate will not be affected by random guessing based on the tailored studys. The results show that, (1) the tailored study recovered the difficulty parameter estimates; (2) the proposed statistic can test the significance of the effect of guessing on the item estimates.
Question:
1. How to set an appropriate cut-point when implementing this method? According to this paper, the cut-point is set to be -1.0 logit, however, in my opinion, if guessing is from the less proficiency it should be set to -2.0 logit or -3.0 logit. Whether different cut-point will cause the performance for detecting guessing?
Answering Sandy’s Questions:
1. On mine understanding, there are two major points in this paper, Anderson’s theorem and tailored study. Applying the Anderson’s theorem to detect the item estimates which are affected by guessing, and conducting the tailored study for getting the more accurately item estimates than that in which guessing is not considered. With regard to the Anderson’s theorem which is used in this paper, only described in page 9 – 10. It may be too short to can’t catch the point, and this is also my problem when I reading this paper.