The Establishment And Application Research On Three-Parameter Graded Response Model

Item Response Theory (IRT) is the most influential measurement theory in educational and psychological testing. The primary goal in modern IRT research is to expand the existent class of models to cover response data from any"natural"test format. To realize this goal, deeper insights into response processes and the precise way item properties and human abilities interact are needed (van der Linden et al.1997).although models would not be attached to real test fully.but only by developing and exploring new models continuously, the gap between them can be reduced and IRT would be perfected.In the existing researches of models, for polytomous item, only item discrimination and grade difficulty (or step parameters) have been considered. But in real examination, polytomous items may also have guessing. So in this study, based on samijima graded response model, guessing parameter was involved in Polytomous response model and a three-parameter graded response model(3P-GRM)is proposed ,other works related to this model also be conducted as bellow:1. Describe the three-parameter graded response model and compare it with GRM through item response function and information function, Besides, it is also proved that ignoring guessing parameter of Polytomous items may lead to inaccurate estimation of ability parameters;2. Based on MMLE/EM algorism, a related program for estimating item parameters in 3P-GRM is developed. Both simulated date and real date are used to check the program;3. Write the program of modle-data fitness to Check the fitness of the model and real data;4. Discuss how to control the test-paper quality with test information.