Research On Online Calibration Design Of MCAT Based On Fisher Informatio

As the education sector is increasingly computerized and cognitive diagnostic(CAT)tests become increasingly popular,online calibration technology has received increasingly attention.Online calibration design is one of the important components of online calibration technology,and it is the core technology of item calibration under CAT scenario.At present,most researches on online calibration are computerized adaptive testing(UCAT).Its disadvantage is that test can only be examined from a single dimension.Therefore,this paper studies online calibration design based on Fisher information in the context of multidimensional computerized adaptive testing(MCAT).The main content of study 1 is to expand the four kinds of online calibration design based on Fisher information,D-optimal design,D-VR optimal design,A-optimal design and A-VR optimal design,from single dimension to multi-dimension.Then,through the simulation experiment,the estimation accuracy of four kinds of adaptive online calibration design and random design under the experimental conditions of the correlation between different sample size,test length and ability was explored.The results show that compared with random design,the four adaptive on-line calibration designs can significantly improve the estimation accuracy of project parameters.The larger the sample size,the longer the test length and the smaller the correlation between abilities,the higher the estimation accuracy of the item parameters of each online calibration design.Under the condition of small sample size,with the increase of the correlation between abilities,the influence of the increase of test length on the accuracy of estimation decreases gradually.With the increase of test length,the influence of the increasing correlation between abilities on the estimation accuracy gradually decreases.Moreover,the estimation accuracy of the discrimination parameter is less than that of the difficulty parameter.The main content of study 2 is based on the optimal design of A-VR with the best performance of project parameter estimation accuracy in Study 1,which extends the research from focusing on all project parameters to focusing on some project parameters online calibration design of MCAT.The main conclusions are as follows: Consistent with the content of the first study in this paper,the estimation accuracy of item parameters of A-VR optimal,AVRS1 optimal and A-VRS2 optimal online calibration designs gradually increases with the increase of sample size and test length and the decrease of correlation between capabilities.In order to estimate OR accurately alone,A-VRS2 optimal design should be adopted,but the disadvantage is that the optimal design has the lowest estimation accuracy.For accurate estimation alone,A-VR optimal design should be adopted,but the disadvantage is that the optimal design has the lowest estimation accuracy.In general,A-VRS2 optimal design should be adopted only for the accurate estimation of discrimination parameters.If only to accurately estimate difficulty parameters,the A-VR optimal design should be used.