A Mixture Response Model for Identifying Item Preknowledge

Abstract

As one of the three broad types of test cheating, item preknowledge has always been a severe threat to test validity and is widespread in various testing programs, especially within some high-volume certification testing programs. While many response-based methods have been proposed to identify and handle item preknowledge, most require the assumption that the compromised items are known. Moreover, few studies have considered that both examinee and item characteristics might affect this cheating behavior, and even fewer have considered the relationship between ability and such behavior when estimating ability. Therefore, this article proposes a mixture model with less strict assumptions on the compromised items. By modeling cheating behavior with a latent response approach, the model takes into account the effect of both characteristics at the item-by-examinee level and assesses how a person’s ability relates to such behavior. Two simulation studies demonstrate that the parameters of the proposed model can be effectively recovered, and when data contains item preknowledge, the model generally produces more accurate ability estimates than existing models. Finally, an empirical example based on a licensure test dataset illustrates the applicability of the new model.