Item preknowledge has been a serious threat to test validity because it results in some examinees having an unfair advantage and undermines the validity of interpretability of test results for all. A number of different approaches have been developed for statistically detecting preknowledge, most of which entirely rely on item response data. In addition to item response data, using responses time data may enhance the detection of examinees with item preknowledge because response times convey important aspects of problem-solving such as strategy and motivation. When two or more subgroups that show different response behaviors exist in a population, a common statistical approach for distinguishing such behaviors is the use of a mixture model.

This dissertation proposes a mixture model that extends van der Linden's (2007) model to incorporate both item responses and response times within a hierarchical framework for detecting examinees with item preknowledge. Simulation studies were carried out to evaluate the performance of the mixture model for the detection of item preknowledge. In addition, the performance of the mixture model was compared with other mixture models. Throughout simulation studies, it was shown that the proposed mixture model performs well to differentiate preknowledge examinees from honest examinees. To illustrate how to apply the proposed mixture model in practice, a real data set from a credentialing exam was analyzed by the mixture model.