Research On P-STR Item Selection Strategies And Variable-length Termination Rules In Computerized Adaptive Testing For Cognitive Diagnosis

Computerized Adaptive Testing for Cognitive Diagnosis(CD-CAT) is a product integrating Cognitive Diagnosis(CD) with Computerized Adaptive Testing(CAT). Making good use of CD-CAT selection strategy can quickly and accurately diagnose how many attributes among the test the examinee knew and what pattern the examinee got, so CD-CAT has attracted more and more researchers and applicators those both domestic and foreign. At present, there is little research about CD-CAT items pool exposure and the termination rule, so this paper will discuss the exposure control of cognitive diagnosis CAT item selection strategies and the termination rules.This paper first proposes two item selection strategies: one is balancing ability, knowledge state of test-taker and securer items pool item selection strategy, the other is combining with the exposure control factor named KL-E item selection strategy, then presents three variable length termination rules named DFSPP, DFMPP and HM, finally do some comprehensive simulation experiments based on DINA using common selection strategies and termination rules. Monte Carlo simulation results show: first, adding three control factors such as Fisher information, exposure control and item priorities in the item selection strategy named balancing ability, knowledge state of test-taker and securer items pool item selection strategy can not only improves the measurement accuracy of item pool but also works well in balance of using items under the condition of slightly lower test speed; second, although the KL-E method is not better than the traditional KL method in the PMR index, but it has very a big advantage in the chi-squared statistic index compare with the traditional KL method and in the AMMR index the KL-E method goes near to the traditional KL method; third, comparing with the traditional method DFSPP, DFMPP and HM have their own advantages; fourth, adding three control factors such as Fisher information, exposure control and item priorities in the item selection strategy named one is balancing ability, knowledge state of test-taker and securer items pool item selection strategy, DFSPP, DFMPP and HM are better in the comprehensive effect based on the Linear attribute hierarchy under K equal six.