[4H2-E-5] Human interface, education aid: human evaluation

[4H2-E-5-02] Computerized Adaptive Testing Method using Integer Programming to Minimize Item Exposure

〇Yoshimitsu MIYAZAWA¹, Maomi UENO² (1. The National Center for University Entrance Examinations, 2. The University of Electro-Communications)

Keywords:Computerized adaptive test, e-testing, Item Response Theory

Computerized adaptive testing (CAT) estimates an examinee's ability sequentially and selects test items that have the highest accuracy for estimating the ability.However, conventional CAT selects the same items for examinees who have equivalent ability.Therefore, tests cannot be used practically under circumstances in which the same examinee can take a test multiple times. As described herein, we propose CAT that minimizes item exposure and which adaptively selects different items for examinees of equal ability, while retaining accuracy.
This paper presents the method's effectiveness through a simulation experiment and with item pools used by actual test providers. Results confirmed that 1) the proposed method yielded the shortest test length. 2) The proposed method controls exposure and selects different items for different examinees. The non-uniformity of estimation was low. 3) The average item exposure of the proposed method was the lowest.