Guessing-testlet response model

摘要

The psychometric standard of fairness can be violated if the guessing effect is improperly handled. The two most common ways of handling guessing effect are through item design and guessing effect modeling. Items with lower priori guessing probability helps to reduce guessing effect. This paper proposes a two-parameter logistic guessing-testlet response model to model such items. The proposed model is an extended testlet response model where items of the same guessing priori guessing probability are grouped in the same testlet. To reduce the priori guessing probabilities, the items are designed to have multiple-correct responses and the number of correct responses is varying across items. Simulation result shows that the proposed model outperforms the two-parameter logistic item response model in model fit. The proposed guessing-testlet merits ability with no guessing but penalizes ability with guessing.