On gee-based regression estimators under first moment misspegification
Titel:
On gee-based regression estimators under first moment misspegification
Auteur:
Hall, Daniel B.
Verschenen in:
Communications in statistics
Paginering:
Jaargang 28 (1999) nr. 5 pagina's 1021-1042
Jaar:
1999
Inhoud:
tors that solve an equation. In common special cases, estimating equation-based estimators have appeal because fhev correspond to the maximum oi minimum of an objective function. In such cases, an intuitively reasonable criterion for estimation, such as minimizing the Euclidean distance between the observation vector and the fitted value, motivates the procedure. In general, however, solutions of estimating equations need not minimize an objective function. Therefore, when the assumed model for the data is inaccurate, it is unclear what aspect of the data is being described by an estimating equation-based estimator. Since the landmark article of Liang and Zeger (1986), there has been considerable interest in using estimating equations for longitudinal and other clustered data. In this paper we examine the form of the regression parameter estimand under model misspecification when estimating equations related to Liang and Zeger's generalized estimating equations (GEEs) are used for model fitting. Closed form expressions are presented for these estimands in simple examples. These results indicate that the estimands for GEE1 (Liang and Zeger, 1986) and extended GEE (Hall and Severini, 1998) are intuitively reasonable, but estimands based on GEE2 (Prentice and Zhao, 1991) in some cases are considerably more difficult to justify than their GEEl counterparts.