A Group Fit Statistic for the Multilevel Item Response Model

• In Event: Innovative Approaches to Item Response Theory

Abstract

Aberrant behaviors of test takers can be observed more often in certain groups. For example, in some test centers or schools, test-takers are more likely to cheat or respond carelessly than other clusters. However, research in detecting group level misfit are rare. This paper proposes a group fit statistic lz2 by extending the lz statistic to the multilevel item response model (IRT). Using true item parameters and scores, the new statistic achieves adequate power and controls the Type I error rate at the nominal level. When estimated parameter and scores are used, the new statistic still achieves acceptable detection rate and precision, though the Type I error rate is lower than expected.

Authors

• Yishan Ding, University of Maryland - College Park

• Ji Seung Yang, University of Maryland