The Improved Maximum Priority Index And Its Application In Computerized Adaptive Testing With Cognitive Diagnosis

Item Selection Strategy constitutes the key content of Computerized Adaptive Testing (CAT), which not only makes tests intelligent but also exerts a direct impact on the quality of test. Traditional CAT may easily lead to safety risk and it's even more difficult to handle because of its constraints on content. The Maximum Priority Index Method (MPI), put forward by Cheng Ying in 2009, though able to meet the non-statistical constraints to a certain extent, still gives rise to a violation of constraints and tendency to choose items with relatively less constraints. In this thesis, MPI is improved and 4 new item selection strategies are proposed by means of Maximum Fisher Information Method as follows: the improved MPI, exponential method, additive method and weighted method. The Monte Carlo simulation suggests that the use of the improved MPI won't result in cases violating constraints but will be more accurate in estimating ability. Furthermore, exponential method and additive method turn out to be better in estimating examinee ability as they have taken the advantages of Maximum Fisher Information Method. Last but not least, weighted method outperforms MPI by reducing weight at the beginning of test.Compared with IRT-CAT, Cognitive Diagnostic Computerized Adaptive Testing(CD-CAT) can not only obtain examinees'mastery of skills but also understand the knowledge states and cognitive structures among various individuals, which makes the study on the latter increasingly hot. Actually, the diagnosis of cognitive diagnosis is pattern recognition. In this thesis, a new kind of CD-CAT item selection strategy Modified Maximum Global Discrimination-- Cognitive Diagnosis Model Information Index Method(MGCDI)is proposed by applying the improved MPI to CD-CAT, which can both meet balancing attribute coverage and handle similar attribute patterns effectively. Through the Monte Carlo simulation, merits and demerits of MGCDI, Modified Maximum Global Discrimination Index Method (MMGDI) and Cognitive Diagnosis Model Information Index Method (CDI) are compared under 4 different attribute hierarchical structures. The results were shown as follows: 1) MGCDI not only has the balancing attribute coverage of MMGDI and the advantages of CDI when dealing with similar attribute patterns, but also enjoys a higher pattern match ratio and marginal match rate. 2) MGCDI turns out to be best, followed by MMGDI, and CDI performs worst. 3) All of the 3 item selection strategies work best when handling the unstructured type and turns out to be worst in dealing with the divergent type.