| Computerized adaptive testing for cognitive diagnosis(CD-CAT)combines the advantages of cognitive diagnosis(CD)and computerized adaptive testing(CAT).Currently,most modern psychometrics focus on the single-strategy cognitive diagnosis models in CD-CAT studies,which are not capable of accommodating multiple strategies.The multiple-strategy models by Ma and Guo in 2019 not only allow different students have different strategies on the test,but also allow a student to switch strategies for different items.Therefore,this study applies multiple-strategy models to CAT(called as MSCD-CAT),and according to the framework of the GMS-CDMs,the G-DINA model can be employed to the multiple-strategy situation.One of the most important designs in the computerized adaptive test is the item selection method.The commonly used item selection methods include KL method,PWKL method,HKL method,MPWKL method,GDI method and SHE method.In this study would extend those item selection methods in single-strategy CD-CAT to multiple-strategy CD-CAT.This research is divided into three specific studies:Study 1: the expansion of item selection methods for MSCD-CAT.The traditional single strategy item selection method is extended to multiple-strategy condition.Study 2: efficacy verification and comparison of MSCD-CAT selection methods under the fixed-length test.The feasibility of different item selection methods for MSCD-CAT was discussed and compared under the condition of fixed length.Study 1was evaluated by pattern correct classification rate,marginal match rate,Chi-square index (χ~2) and test overlap rate to compare the advantages and disadvantages of these methods.Study 3: efficacy verification and comparison of MSCD-CAT selection methods under the variable-length termination condition.For evaluation criteria,Study 2 compared test efficiency of different selection methods,including minimum,maximum,mean and standard deviation for test length.The findings as follow:(1)Under the fixed-length condition,MS-GDI method and MS-SHE method are the best in the multiple-strategy,and the results of them are similar to each other.It can be noted that strategy selection parameter increased,the evaluation criteria will change.When s=10 result in a good effect regardless of test lengths or item selection method.(2)Under the variable-length condition,MPWKL method,MS-GDI method and MS-SHE method perform better in whole study 3.It can be concluded that generating model had an impact on this condition,the GMS-DINA model preforms better than the GMS-RRUM model and MS-GDINA model.The two simulation studies suggest that the MS-GDI and the MS-SHE methods are very promising as item selection methods for MSCD-CAT.Furthermore,CD-CAT in multiple-strategy situation,under the condition of fixed-length test,both of the MSGDI and the MS-SHE methods can guarantee more than 90% PCCR using 30 items regardless of strategy selection parameter,with a lower test overlap rate and a higher item bank security,which improve considerably the efficiency of testing.MS-SHE and MS-GDI methods perform better than other multiple-strategy selection methods in various indicators under the condition of variable-length test.Under the guidance of the era of big data,it is of epochal significance to combine multiple-strategy models with CD-CAT.Especially with the rapid development of the education industry,understanding the difference and diversity of students has attracted everyone’s attention. |
