Online Program Home
My Program

Abstract Details

Activity Number: 522 - Contributed Poster Presentations: Biometrics Section
Type: Contributed
Date/Time: Wednesday, July 31, 2019 : 10:30 AM to 12:20 PM
Sponsor: Biometrics Section
Abstract #304197
Title: Prediction Accuracy and Robustness to Non-Normality of Two Methods of Predicting Random Effects in Linear Mixed Effects Models: Empirical Bayes vs. Quadratic Inference Functions
Author(s): Zhiwen Wang* and Francisco Diaz and John D Keighley and Jianghua (Wendy) He and Jo Wick
Companies: University of Kansas Medical Center ASA Student Chapter and The University of Kansas Medical Center and University of Kansas Medical Center and The University of Kansas Medical Center and University of Kansas Medical Center
Keywords: Best linear unbiased predictors (BLUPs); Linear mixed effects model; Cross-validation; Distribution misspecification; Estimating equations; Mixtures of normals

Several methods for predicting random effects in linear mixed effects models have been proposed. The performances of these methods have not been thoroughly investigated when the normality assumption for the random effects is violated, except for the empirical Bayes (EB) approach, as well as comparisons of the methods. This study compared the prediction accuracy of the EB approach with that of an approach based on quadratic inference functions (QIFs) under different distributional assumptions for the random effects, using a longitudinal linear model that included a random intercept and a random slope for time. The simulations revealed that the EB approach was generally superior to the QIF approach in predicting the random effects, even under non-normal distributions for the random effects, except in some scenarios with very large error variances. In addition, the EB approach is mathematically and computationally less complex. Thus, our study suggests that the EB approach is more recommendable as the first choice in statistical practice, even if non-normal random effects are suspected. An application to the prediction of individual benefits of an anti-depressan drug was considered.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program