×

An examination of a method for marginal inference when the cluster size is informative. (English) Zbl 1433.62201

Summary: Generalised Estimating Equations (GEE) are a popular method to fit marginal models to clustered data. When the total number of members in the cluster is informative, then inference may be for a typical member of a typical cluster or the population of all cluster members. Applying the GEE with independence working correlation provides inference for the population of all members, and with additional weighting by the inverse cluster size gives inference for the population of typical members. In earlier work an adaptation of GEE termed modified within-cluster resampling (MWCR) was proposed to give unbiased inference for the population of typical members with increased efficiency by recognising the correlation between measurements. We describe how bias can arise when MCWR is used, a potential that was not clear when the method was proposed. We present conditions on the data structure and on the choice of the working correlation that, if satisfied, allow consistent estimation from MWCR. We illustrate the method with an application to a dataset of AIDS-related condition events from the Delta trial of HIV therapy.

MSC:

62J12 Generalized linear models (logistic models)
62H12 Estimation in multivariate analysis
62H30 Classification and discrimination; cluster analysis (statistical aspects)
PDFBibTeX XMLCite
Full Text: DOI