I have a data set with about 2000 twin pairs and want to apply a logistic regression to it while taking account of the correlated observations (twin pairs = clusters). I use the GEE procedure in SPSS.
Looking at the QICs for independent vs unstructured working correlation matrix (WCM), the QIC is minimally smaller for the independent WCM (at the second decimal), the correlation between the twins in the unstructured WCM is .04. Therefore I decided to use the independent WCM.
When comparing the results of the GEE with independent WCM with the results of the GLM, I get the same parameter estimates, but different estimates for their CIs, and I wonder why? Shouldn't the CIs also be the same?