Liang and Zeger (1986) introduced generalized estimating equation or GEE models as a way of using the ideas of generalized linear models when repeated measures are available on subjects or clusters of subjects, but normality of the response may not be appropriate. They have shown that consistent estimates of regression coefficients are obtained even if the within cluster or subject correlation is misspecified. In this article we show that under certain common circumstances, Fisher consistent estimates of regression coefficients will be obtained even if the link function in the generalized linear model is misspecified.
Bibliographical noteFunding Information:
* Corresponding author. ~This work was supported by National Science Foundation grant DMS-9208819.