Research On IRT Equating For Multiple Item Type

The research on test equating is very important for fairness of examnation, construction of item pool, assessment of teaching quality and computerized adaptive test.In the framework of IRT, the item response models in common use are Logistic Model of dichotomous scoring and Graded Response Model (GRM) or Generalized Partial Credit Model (GPCM)of polytomous scoring. As the gradual enrichment in the forms of educational assessment, more test papers have adopted dichotomous scoring items and polytomous scoring items. This kind of paper was called Multiple Items Paper, for short, Multiple Items Type. That is often used in the tests of this country, e.g. the Doctor Qualification Test. In the past, the domestic test equating of multiple items type treated Logistic Model as the particular one of GRM or GPCM. To ascertain reasons, we assumed that the polytomous scoring items fit GRM.Because the GRM does not contain the element of guessing but the Logistic Model does, the approach ignores the element. To solve this problem, the thesis put forward a Mixed Model extended from 3PLM and GRM, which has solved the ignorance of guessing. Moreover, a relevant equating program has also been developed for the Mixed Model.In addition, in order to measure the error by the ignorance of guessing in a test, this thesis gives specific methods and procedures to get the divergence of equating by the IRT characteristic curve method.On the error of the computation of equating divergence as the measurement criterion, it has carried a lot of experiments of Monte Carlo Simulation and compared equating results based on the Mixed Model and GRM using the same data in the simulation. The finding shows, if GRM is used by the ignorance of the guessing in a test, there will be great divergence in equating and obvious difference compared to the Mixed Model mostly.Moreover, the divergence in equating will increase with the guessing increasing.In the end, taking into account polytomous scoring items may contain the element of guessing,the thesis has studied the equating of 3PL-GRM, a new model of IRT. It reviews the efficiency and adaptability of 3PL-GRM on five different equating criterions. The results showed that, when the value of the coefficient A is between 0.5 and 1.4, the performance of SLcrit is the best, Hcrit is proper when 1.5