An Analysis of Essay Examination Data Using Graded Response Model

The purpose of this study is to investigate the reliability of essay test scores when the scores are expressed by using several rating categories and applied to graded response model. The data were obtained by assigning an essay examination to high school students. The 165 essays were evaluated holisticaly and given marks between 0 to 100 points. The marks were then classified into the 2 to 5 categories according to the two criterions : percentile and standard deviation. It was found that the reliability of the 5-category data (standard deviation criterion) was almost as good as that of the original 100-point data, which indicates validity of categorical rating. The characteristics of some of the raters were also described by their information functions