In computerized adaptive testing (CAT), aberrant responses such as careless errors and lucky guesses may cause significant ability estimation biases in the dynamic administration of test items. We investigated the robustness of the 4-parameter logistic item response theory (4PL IRT; Barton & Lord, 1981) model in comparison with the 3-parameter logistic (3PL) IRT model (Birnbaum, 1968). We applied additional precision and efficiency measures to evaluate the 4PL IRT model. We measured the precision of CAT with respect to the estimation bias and mean absolute differences (MAD) between estimated and actual abilities. An improvement in administrative efficiency is reflected in fewer items being required for satisfying the stopping rule. Our results indicate that the 4PL IRT model provides a more efficient and robust ability estimation than the 3PL model.
關聯:
SOCIAL BEHAVIOR AND PERSONALITY 卷: 40 期: 10 頁數: 1679-1694
