Summary
This article can be summarized as an attempt of using the Rasch model to estimate the effect of guessing behaviours in a multiple choice test and try to come up with a model tailored model to correct the biased item and person parameters under the effect of guessing behaviours. By incorporating the ideas of the Andersen’s theorem, two different analyses were done: first on full range of response data; the other on a sub-sample of the response data. It is expected that those person of lower proficiency will exhibit random guessing behaviours than those persons of high proficiency. And it is expected that more difficult items are more prone to receive guessing responses than easier items. By comparing the ICC, the tailor analysis is found to produce a very close solution as the 3PL model.
Reflections
Besides, the traditional 3PL modeling, the article has presented an alternative to analysis of data via an adapted Rasch analysis (tailored analysis). But it does not provide indicative guidance on deciding the situations where such a tailored analysis has advantages. The selection of the ‘cut-point’ in the tailored analysis seemed to be rather arbitrary. There is a lack of a strong evidence that differentiating items with ‘guessers’ and items without ’guessers’ b by the expected correct percentage less than .3 is valid. A better solution may be obtained by altering the 3PL model rather than the Rasch model.