Model Identification for Linear and Nonlinear Bayesian Crossed Random Effects Models

• In Event: Advances in Multilevel Modeling

Abstract

Bayesian crossed random effects models (CREMs) are used in educational research because they allow researchers to account for and estimate the impact of group membership (e.g., teachers) that changes over time. Past literature established the benefit of implementing CREMs. However, no research has determined what data conditions are necessary to empirically identify intrinsically linear and nonlinear CREMs with varying complexity, yielding accurate and precise parameter estimates. Additionally, this study extends the current literature by developing an intrinsically nonlinear CREM—a piecewise CREM with an unknown changepoint, which is useful for educational applications with longitudinal data. Preliminary results suggest that more data are needed from teachers to recover the teacher-level effects on individual outcomes. Practical recommendations for researchers will be provided.

Authors

• Corissa Taylor Rohloff, University of Minnesota
  Presenting Author

• Nidhi Kohli, University of Minnesota
  Non-Presenting Author

• Eric F. Lock, University of Minnesota
  Non-Presenting Author