Regression Analysis with Clustered Data: Applications of Survey Methodology
B.I. Graubard
and E. L. Korn
1994 Statistics
in Medicine
13, 509-522
Contact: B. I.
Graubard, National Cancer Institute
6130 Executive
Boulevard, Bethesda, MD 20892
Keywords:
Population average, cluster specific, logistic regression
Abstract
Clustered data
are found in many different types of studies such as repeated
measures, inter-rater agreement, household survey, cross-over and
community randomized studies. Analyses based on population average
(PA) and cluster specific (CS) models are two commonly used
approaches for estimating treatment/exposure effects with clustered
data. This paper gives conditions involving marginal balancing of the
covariates and the treatment exposure variable under which the PA and
CS analyses will agree. A weighted PA analysis is proposed which will
estimate CS treatment effects when marginal balancing does not hold.
Methods for variance estimation which draw upon survey methodology
are proposed for the weighted PA analysis.