Handling Measurement Error in Predictors With a Multilevel Latent Variable Plausible Values Approach: An Illustrative Example Using Teacher Instructional Practices

• In Event: Roundtable Session 14
  In Roundtable Session: Issues in Multilevel Latent Variable Modeling

Abstract

To properly deal with measurement error in predictors, a multilevel latent variable plausible values approach is proposed. Using an illustrative example of teacher reading instructional practice measures in the Early Childhood Longitudinal Study (ECLS), the current study shows how to construct proper teacher reading instructional measures from teacher item responses in the frameworks of Item Response Theory (IRT) and multilevel latent variable modeling. The results are compared to three more traditional approaches: summed score analysis, IRT score analysis, and a single-level IRT model combined with a plausible values approach. A subsequent simulation study is conducted to compare the performance of different approaches.

Authors

• Ji Seung Yang, University of Maryland - College Park

• Michael H. Seltzer, University of California - Los Angeles