Research On The Development Of Item Selection Methods In Multidimensional Computerized Adaptive Testing With Polytomously Scored Items:From Non-statistical Constraints Perspective

The Computerized adaptive test(CAT)needs not only to consider statistical optimization problems but also to meet non-statistical constraints in practical applications.Statistical optimization means that the test must ensure a high measurement accuracy,while non-statistical constraints mean that the test composition meets certain test specifications,such as content constraints,question types constraints,dimensional constraints,and item exposure control,etc.Just like traditional CAT,meeting non-statistical constraints is also an important issue that needs to be addressed in PMCAT.Multidimensional computerized adaptive test(MCAT)can not only capture the ability and trait information of the subject in multiple dimensions,but also improve the accuracy and efficiency of the participants.It is also possible to obtain diagnostic information for each dimension from the test responses.Polytomously-scored items are widely used because they can provide more information and can measure more complex abilities and skills.Therefore,PMCAT is more promising in practical applications.However,the non-statistical constraint methods are mainly used in UCAT,there are few applications in MCAT.Especially,the related research in PMCAT has not been reported in related literature.The main purpose of this paper is to explore the application of traditional non-statistical constrained selection strategies to PMCAT and develop new non-statistical constraint selection strategies(Study 1).At the same time,Monte Carlo simulation study(Study 2 and Study 3)is used to verify the performance of Study 1,non-statistical constraint selection strategy,on the one hand.On the other hand,it is used to explore the influence that the factors such as the exploration model,the number of test dimensions and the correlation between dimensions have on the measurement accuracy,the security of the question bank of the PMCAT and non-statistical constraints.The simulation results shows:(1)The RMMPI method performs best as a whole,because it can fully satisfy the non-statistical constraints on various conditions and with different models,and it has the highest measurement accuracy and good test security.The RWDM method excels the RMPI method in terms of non-statistical constraint satisfaction and test safety indicators,while the RMPI method outperforms the RWDM method in terms of test accuracy.(2)The three improved methods perform well in satisfying non-statistical constraints,and greatly reduce the number of violations of non-statistical constraints compared with the original method,and the measurement accuracy is improved.(3)The number of non-statistical constraint violations is not affected by the degree of correlation between models,dimensions,and dimensions.(4)The estimation error of each topic selection strategy becomes larger as the number of dimensions increases,and the estimation accuracy decreases.(5)The higher the correlation between the dimensions and the larger the number of dimensions,the lower the item exposure rate distribution(x~2)and the test overlap ratio(TOR),and the higher the question bank security.