Multilevel Item Response Theory Models: Allowing the Discriminations to Vary Across Levels

• In Event: Advances in Item Response Theory Models

Abstract

When cluster effects are present in educational and psychological test data, item response theory (IRT) models formulated within a multilevel framework are commonly used to psychometrically analyze the data. With these multilevel models for cluster effects, a common practice is to constrain the item discriminations to be equal across the person (Level 2) and cluster (Level 3) levels. That is, a single set of item discriminations are estimated, and this single set is applied at the person and cluster levels. Including such constraint when unwarranted could affect the accuracy in the estimation of various psychometric properties. These implications have not yet been fully explored, and the goal of this study is to fill this void in the psychometric literature.

Author

• Ken Fujimoto, Loyola University Chicago WTC