The Study Of Item Selection Strategy-automatic Control For The Role Of Discrimination Index

Based on the IRT，Computerized adaptive testing (CAT) is the product ofcontemporary computer technology applied to the field of psychometrics. CAT can beconducted in different areas and times and can select appropriate items for examineesintelligently to achieve tailored and personalized tests.In the meantime CAT cangrade according to examinees’ answers and make precise and estimation of examinees’ability by using parameter estimation algorithms such as Expected a Posterior(EAP)and Maximum Likelihood Estimation(MLE).Besides CAT has higher testingefficiency. IRT is divided into binary scored model and multilevel scored model.For CAT item selection methods are the most critical part and the reflection ofintelligence. Exposure control is very important for CAT because it directly affects thesecurity and fairness of tests. A good item selection method should trade off betweenthe accuracy of ability estimation and the uniformity of item exposure. At presentthere are lots of exposure control strategies. This paper make a deeper discussion onthe exposure control factor which is proposed by Cheng et al(2011). On the basis ofthe exposure control factor, this paper adds the test length of fixed tests and testinformation of variable length tests into the new item selection strategy. This papercombines two indexes-the percentage of length (PI) and the percentage of information(PI) with exposure control factor proposing a new item selection strategy which cancontrol the role of discrimination index automatically in the process of item selection.Besides, the new strategy doesn’t need to divide the item pool into several stratums.The results of Monte Carlo simulations show that the new approach proposed in thispaper is more ideal in the performance of item exposure control and test efficiencyand other indexes.