Variance Component Adjustment for Latent Growth and Multilevel Models in the Structural Equation Modeling Framework

• In Event: Roundtable Session 15
  In Roundtable Session: Nuanced Methods for Latent Growth Models

Abstract

Linear mixed models (LMMs) are a popular statistical model that can be easily adapted to the structural equation modeling (SEM) framework. One limitation of the SEM implementation of LMMs is that most SEM software does not feature restricted maximum likelihood estimation (REML) but only traditional maximum likelihood (ML). It is well known that ML estimates of variance components are downward biased with smaller samples, an issue REML is able to remedy. Although attempts have been made to apply REML to SEM, these attempts are not without issue and have yet to be integrated into common SEM software. This proposal advances a simple algebraic adjustment to ML estimates that reduces the bias present in variance components with smaller samples.

Author

• Daniel McNeish, University of Maryland