Sampling and Measurement Error Impacts on the Estimation of L2 X-Y Relationships in MLMs

• In Event: Advances in Multilevel Modeling for Complex Data and Study Designs

Abstract

Level 2 predictor-outcome coefficient bias in traditional multilevel models has been shown to stem from sampling and measurement error in L1 variables used to form the L2 predictor. However, the prior research on this issue was based on the assumption that the outcome (Y) was measured without error. Using simulations, we re-evaluate this problem under conditions in which sampling and measurement error are present in both the predictor and outcome. We consider methods of obtaining L2 aggregate variables that include doubly manifest aggregations, doubly latent aggregations, and their hybrids. Preliminary results show that the doubly manifest approach performs well only when X and Y are both very reliably measured, and that the doubly latent approach is better for lower reliabilities.

Authors

• Timothy R. Konold, University of Virginia

• Elizabeth A. Sanders, University of Washington