| Summary: | 碩士 === 輔仁大學 === 統計資訊學系應用統計碩士班 === 101 === Nowadays, all the achievement test, psychological scales and questionnaires, which are scored in the type of Likert scale, aim to classify the answer of the subjects into the corresponding category. This research uses generalized sequential probability ratio test to solve the problem of multi-categories. Generally, the result of research revealed that the classification accuracy of consequence of GSPRT is approximately 80% in circumstance without limiting the number of item. When the maximum number of test is getting less, the classification accuracy will descend, the reason is that likelihood value in most of subjects doesn't reach to stopping boundary in circumstance with limiting the number of item. If we categorize by the maximum likelihood value of GSPRT, it’s effective in enhancing the classification accuracy. In circumstance of fixed length tests, the classification accuracy will grow by higher test length. However higher test length will raise the cost of test than circumstance of variable-length test.
|
|---|
