The purposes of exams and tests are tools for teachers to evaluate the progress of students, and to measure the level of their understanding on the subject. However the traditional tests are not only limited by the duration and location, but also time consuming and labor-intense for the grading teacher. Therefore, this study proposes a model in series of computerized diagnostic tests called “Leverage”, for Natural Science on grad-school level. The types of tests include multiple choices and constructed-response questions with the automated analytic models for auto-grading. Using online-exams with the automatic grading system, the teachers can record the students’ response on answering patterns, and analyze the patterns with “Cognitive Diagnostic model”, and extrapolate their level of understanding on the conception of sub-skills and recognize the types of error.The following are the finding of this study:1.The computer based diagnostic questions prepared in this study can effectively analyze the students' concept of sub-skills, and types of error. Using massive amount of data storage, Computers can store large array of complex response questions, and construct automated analysis base on those questions.2.The recognition rate of automatic analysis on concept of sub-skills and the types of error are 100%.3.The analysis using DINA, DINO's probability model for concept of sub-skills and the types of error, is working well.4.By combing the automated analysis of constructed-response question based on the DINA, DINO model, one can detect more patterns of recognition concepts and types of errors.As conclusion, the computerized diagnostic model proposed by this study can assist teachers to improve their teaching effectiveness on various understanding levels of students. It’s a significant enhancement to students’ learning.
