Computerized adaptive testing (CAT) presents a tradeoff dilemma involving item exposure bias and measurement accuracy. To resolve this dilemma, we propose a new two-step CAT mechanism to balance exposure bias and measurement accuracy. Using zero-suppressed binary decision diagram (ZDD), the proposed method first selects and presents an optimal item from an equivalent item pool, which uniformly divides the whole item pool. The first step rapidly provides a roughly approximated ability estimate of an examinee. The second step produces a more accurate ability estimate of the examinee. Specifically, the second step selects an optimal item from the whole item pool and presents an item with a difficulty parameter value that is approximately equal to the examinee ability estimate. Numerical experiment results underscore the effectiveness of the proposed method.
4 February 2020 in Advances in Intelligent Systems and Computing