Multilevel Graded Response Model in Longitudinal Study: Formulation and Illustration

• In Event: Advances in Longitudinal Modeling

Abstract

There are several statistical modeling techniques to investigate change in latent traits over time and correlates of these changing trajectories. Multilevel item response theory model is one recently developed alternative. This study presents the formulation of multilevel graded response model, and demonstrates its application by examining the change of mastery goal orientation when students are repeatedly formatively assessed in a 9-week instruction period, and correlates of the change at student and class level. Two estimation procedures are compared: Bayesian estimation and two-step procedure, in which ordinary graded response model is used to obtain latent scores on mastery goal orientation at three time points, and then these latent scores are used as outcome variable modeled by hierarchical linear model.

Authors

• Feifei Ye, University of Pittsburgh

• Ting Xu, University of Pittsburgh

• Wenyi You, Pearson