Computerized Adaptive Testing

Random selection

Normally, we will treat RA as the baseline of performance when we make comparison of CAT with other approaches. For pure RA, it indeed can result in little gain. However, in many application, RA is not replacable; for example, to prevent item exposure, optimal selection rule must work with RA to prevent the same item be selected repeately during a short period.

During my current study of MCAT, I am considering that if RA can help the multidimensional selection rules work in a better way. For MCAT, selection rule seems more stubborn, sticking on specific items always when they are considered acceptable.

posted by Shunkai Fu @ 8:40 PM