On gee-based regression estimators under first moment misspegification (Q4241672)

Generalized estimating equations [\textit{K.Y. Liang} and \textit{S.L. Zeger}, Biometrika 73, 13-22 (1986; Zbl 0595.62110)] are considered in the regression model under model misspecification. The question addressed is what is being estimated (what is the estimand) when the procedures are used with longitudinal and other clustered data. Proposals are identified as GEE1, GEE2, and EGEE (the first E for extended). The kind of conclusion the authors arrive at is, for example: ``When observations are grouped according to measurement occasion, GEE1 and EGEE-based estimators of group means estimate the average of the true means of the observations. When observations are grouped according to some time-invariant criterion, a weighted average of the true observations means is estimated\dots{} In the examples considered the quantity that is estimated by solving a GEE2 equation is often not appealing''.