The Research Of Multistage Testing With Multidimensional Assessment:Automated Test Assembly Algorithms And Routing Rules

Most educational and psychological tests measure multiple structures or latent traits.Comparing with the unidimensional structure test,the multidimensionality of the test data is consistent with the cooperation when people completes the test with multiple capabilities.However,multistage testing(MST)as a frequently used adaptive testing design,which is mostly based on unidimensional item response theory(UIRT).Therefore,the purpose of this study is to achieve the extension from traditional MST to multidimensional MST and to explore the feasibility and effect of multidimensional MST.The basic framework of multidimensional MST is consistent with that of traditional MST.However,under the framework of multidimensional MST,the measured traits of test takers are transformed from a scalar to a vector(i.e.multidimensional traits).Therefore,it is necessary to extend several elements,such as the automated test assembly(ATA)algorithm,routing rule,from traditional MST to multidimensional MST.In brief,this paper conducts three studies as follows:(1)new ATA algorithms and new routing rules are proposed in this article to construct multidimensional MST.(2)Two simulation studies are implemented to demonstrate the feasibility and effectiveness of the proposed multidimensional MST.The simulation results indicate that:(1)NWADH-D and NWADH-A algorithms have similar parameter estimation accuracy and are acceptable.Under each condition,the ability estimation accuracy in the high dimensional case is inferior to the estimation accuracy that obtaining from the situation of two dimension.(2)The results obtaining from the multidimensional AMI routing rules performed better than that of the Multidim-proximity method and the effects of the high dimensional case is inferior to that of two dimensional case.(3)Overall,multidimensional MST has good performance under each condition.In General,the study is not only expected to provide information about how did these elements in multidimensional MST perform relative to each other but also offer guidance for practitioners to assemble modules in multidimensional MST with both statistical and nonstatistical constraints.