The Research Of Adaptive Testing For Cognitive Diagnosis

There have been two major modes of adaptive testing designs: Computerized Adaptive Testing(CAT)and Multistage Adaptive Testing(MST),as combinations of cognitive diagnostic and adaptive testing,the CD-CAT and CD-MST can help us to measure knowledge structures more efficiently.And the item selection strategy or the assembly method is a fundamental component of CD-CAT or CD-MST,they are important algorithms for realization of adaptive function.This paper studied the item selection strategies based on attribute balancing in CD-CAT,item selection strategies is a fundamental component of CD-CAT,most item selection strategies didn't consider about balanced coverage of the attributes.As a matter of fact,attribute balancing can make toward adequate coverage of every attribute and improve pattern correct classification rate.The literature review revealed that Cheng(2010)studied the item selection strategies based on attribute balancing,and proposed the MMGDI method.The other existing item selection strategies based on attribute balancing was the MGCDI method,which combines modified MMGDI with CDI.The MMGDI method and the MGCDI method can only ensure that each cognitive attribute was measured by roughly similar numbers of items.This paper proposed four new item selection strategies based on attribute balancing in CD-CAT.The new item selection strategies were revised maximum global discrimination index method(RMGDI),revised maximum cognitive diagnosis index method(RMCDI),RMGDI based on standard error of attribute(SE-RMGDI),RMCDI based on standard error of attribute(SE-RMGCDI),respectively.The RMGDI method and the RMCDI method can ensure that each cognitive attribute was measured by roughly similar numbers of items,yet the SE-RMGDI method and the SE-RMCDI method can ensure that each cognitive attribute was measured with roughly similar measurement accuracy.Two monte carlo simulation studies were conducted to compared new item selection strategies with the MMGDI method and the MGCDI method.the simulation results showed that:(1)Under the fixed-length CD-CAT,all the new item selection strategies based on attribute balancing were better at pattern correct classification rate than the traditional MGCDI method except when the testing length was 12,and only RMGDI,SE-RMGDI and SE-RMCDI method performed better than the MGCDI method when testing length was 12.Compare with the traditional MMGDI method,the SE-RMGDI method and the SE-RMCDI method performed better in pattern correct classification rate when the testing length was longer than 5,instead the MMGDI method was the best item selection strategy when the testing length was shorter than 5.(2)the PCCR of the RMGDI method was higher than the existing item selection strategies based on attribute balancing under variable-length CD-CAT,the test efficiency and the comprehensive performance of four new item selection strategies were better than existing item selection strategies based on balanced attribute coverage under variable-length CD-CAT.(3)In general,the performance of SE-RMGDI method and SE-RMCDI method was the best among all selection strategies based on attribute balancing,and the SE-RMCDI method was more recommended when the length of the fixed-length CD-CAT was short or the requested precision of the variable-length CD-CAT was lenient,instead the SE-RMGDI method had a better performance when the fixed-length CD-CAT was longer or the variable-length CD-CAT requested more stringent precision.This paper also proposed several module assembly methods for CD-MST,and these methods were compared in terms of measurement precision,test security and constrain management.The module assembly methods in the study included the maximum priority index method(MPI),the revised maximum priority index(RMPI),the weighted deviation model(WDM),and the two revised Monte Carlo methods(R1-MC,R2-MC).Simulation results showed that the PMPI method was generally better than the MPI method,the R2-MC method was generally better than the R1-MC method,and the two revised Monte Carlo method performed best in terms of test security and constraint management,whereas the RMPI and WDM methods worked best in terms of measurement precision.The study is not only expected to provide information about how did these assembled methods in CD-MST perform relative to each other but also offer guidance for practitioners to assemble modules in CD-MST with both statistical and nonstatistical constraints.