Robust Small Sample Inference for Fixed Effects from Restricted Maximum Likelihood

• Activity Number: 276
• Type: Contributed
• Date/Time: Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Chunpeng Fan*+ and Donghui Zhang and Cun-Hui Zhang
• Companies: sanofi-aventis and sanofi-aventis and Rutgers University
• Keywords: Empirical Covariance Estimator ; Mixed Effects Model ; Small Sample ; REML ; GEE

Abstract

Although asymptotically, the empirical covariance estimator for GEE model introduced by Liang and Zeger (1986) is consistent and robust with respect to the selection of the working correlation matrix, with small sample size, this property does not hold. We propose a small sample correction for the empirical covariance estimator in mixed effects models which is a special GEE model. Inference for the fixed effects based on the corrected covariance matrix is also derived. Simulation studies show that the proposed method outperforms other bias-correction methods including Mancl and DeRouen (2001, Biometrics), Kauermann and Carroll (2001, JASA), and Fay and Graubard (2001, Biometrics). A repeated measure ANOVA model which evaluates the effect of a CB1 receptor antagonist and a crossover design which assesses the treatment effect in a nonhuman primate allergic rhinitis model serve as examples.