The Study Of Combining With Shadow Bank And Dynamic A-Stratification Strategy

Computerized Adaptive Testing(CAT) is the result of the successful application ofitem response theory(IRT), the introduction of a powerful means of moderntechnology——computers, CAT automatically select item based on the abilities of subjects,finally, estimate subjects’ abilities. This new test is different from the traditional writtentests, which is based on different reflections of subjects to the test then actively choice itemfrom the bank to adapt to abilities of subjects in order to obtain an accurate estimate of thesubjects’ abilities. Therefore, CAT is a personalized quiz. According to the different scoringmodels, CAT models have0-1scoring model and multi-level scoring model.For CAT, selection strategy is the most critical part. Accurate, efficient and safeselection strategy has always been the goal of CAT. Classic Maximum FisherInformation(MFI) selection method with high efficiency and accurate estimate, etc., but theitem selected unevenly threaten the safety exam, while the shadow bank is a good balanceof the item selects. This paper takes advantages of these two selection method intoconsideration, get a new selection method, and pull this new method ina-Stratification(a-STR) and Maximum Information Stratification(MIS).Another multi-levelscore can provide more feature information about items, which is a CAT developmentdirection, this paper proposed the Dynamic a-Stratification method(DAS) in the Samejimagraded response model(GRM), combined with exposure control factor. Simulation resultsshow that these new methods achieve the desired effect and the results are ideal.