|
Activity Number:
|
44
|
|
Type:
|
Invited
|
|
Date/Time:
|
Sunday, July 29, 2007 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #307878 |
|
Title:
|
Finite Sample Bias Corrections to Sandwich Covariance Estimators for Longitudinal and Clustered Data
|
|
Author(s):
|
John Preisser*+ and Bing Lu and Bahjat Qaqish
|
|
Companies:
|
The University of North Carolina at Chapel Hill and Brown University and The University of North Carolina at Chapel Hill
|
|
Address:
|
Department of Biostatistics, Chapel Hill, NC, 27599-7420,
|
|
Keywords:
|
Cluster trials ; Correlated binary data ; Generalized estimating equations ; Intraclass correlation ; Sandwich estimator
|
|
Abstract:
|
Empirical sandwich covariance estimators are widely used in the regression analysis of biomedical longitudinal and clustered data owing to the desire to avoid modeling complex correlation structures often considered secondary to the parameters of interest. Unfortunately, when the number of clusters, or subjects in a longitudinal study, is small, sandwich estimators may underestimate the true variances of estimated regression coefficients and lead to undercoverage of confidence intervals. We review some bias corrections proposed for the first-order GEE sandwich estimator, and show that similar adjustments for modeling intracluster association lead to improved finite sample properties. The proposed methods are illustrated using biobehavioral data from a nested pretest-posttest cross-sectional cluster intervention trial on reducing underage drinking.
|
