|Abstract: ||Most of the format for computer-based testing in Taiwan is multiple choice. In spite of the advantage of automated grading, the process of solving strategies can not be analyzed. In addition to that, the possibility of guessing among the respondents is likely to happen. On the other hand, the regular test of constructed-response questions will be able to provide the respondents process of solving strategies. Likewise, correcting and grading will be a heavy task for the teachers. In this study, students’multiple-strategies, skills, and the type of error were diagnosed and classified by using the example of different denominator subtraction in math. The automated analysis processes for constructed-response questions could be formed and the cognitive diagnostic analysis was conducted by using DINA and MS-DINA modes.
The results of this study show,
1.Two or more than two solving strategies for the same question could be found. The ratio of using mixed fraction is higher than that of using improper fraction.
At the same time, the type of error is more diverse and there are lower numbers of occurrences of type of error.
2.In terms of multiple-strategies, skills, and the type of error, the coherence of experts’ grading and automated analysis processes for constructed- response questions can reach the percentage of ninety-five. Therefore, the automated analysis processes for constructed- response questions is effective.
3. Pattern correct classification rate and attribute correct classification rate in mode 2, 3, and 4 are better than those of in mode 1. Pattern correct classification rate and attribute correct classification rate in MS-DINA mode are better than those of in DINA mode.
Key word：automated analysis processes, multiple-strategies, cognitive diagnosis models, different denominator fraction subtraction, constructed-response questions