New Developments in Quantitative Psychology - Presentations from the 77th Annual Psychometric Society Meeting

Abstract

In analyzing test data, it is often assumed that students were motivated to answer the items correctly, hence that the attribute of interest drove test performance. However, if the test is administered in a low-stakes administration condition or if students do not receive feedback, students might not put their best effort into answering the items correctly. Within the item response theory (IRT) framework, lack of motivation threatens the consistency of proficiency and item parameter estimation and therefore the usefulness of the IRT model. The goal of the current study was to explore to what extent a mixture Rasch model and the l z person-fit statistic could be used to model motivational differences in data administered in a low-stakes administration condition. In modeling the mixture Rasch model, constraints distinguished two latent classes of students: (1) a class representing “motivated” response behavior and (2) a class representing “unmotivated” response behavior. We investigated the usefulness of the mixture modeling strategy in a sample of primary-school students (N = 1,512) by comparing the posterior probabilities of the mixture Rasch model and the student’s self-reported motivation. Furthermore, the study investigated the relationship between the student’s self-reported motivation and the l z person-fit statistic.