Graded Response Model

In this thesis we give a research on polytomous item response theory. According to the previous research study, we choose the one-dimensional Graded response model (GRM), which analysis the responses of 200 subjects in 3 oral items rated on 5 point scaling of HSK [advanced]. The postulation of our research is that the rating of GRM differentiates the ability of examinees more precisely. The ANOTE program is utilized to estimate the original data and obtain the ability and item parameter. We transform the ability parameter to ability scores and compare to the scores of the subject test. The result turns to that the ability scores differentiate the ability of examinees more precisely. The fitness of model-data shows that some items fit the model better, while others not. With respect to the item information, as the differentiate parameter higher, the item information parameter more bigger. All the detection of the parameter shows our postulations verified to some extent.