Dear Statlist,

I am trying to determine whether there is a difference in the success proportion between two groups of subjects who are accomplishing a certain task.
In both groups, there were 20 subjects, coded as ID in the dataset, who each did the task 3 times. To account for the non-independence between repeated measurements, I would like to fit a GEE model assuming an exchangeable correlation structure.

I came across two possibilities to do this:

1. xtgee success i.group, family(binomial) i(ID) link(logit) corr(exc) robust
2. xtgee success i.group, family(binomial ID) link(logit) corr(exc) robust

While these models both specify the same group variable, link function, family and correlation structure, the results are different:




Does anyone know why there is a difference, and which model to use?

As a second question, I realized that the results are identical when I use an independent correlation structure. This essentially specifies that there is no within-subject correlation, which does not seem a realistic assumption. I know that GEE is robust against misspecification of the correlation structure when the "robust" option is selected, but I still would not expect that the results are identical. Any ideas?

Any help is greatly appreciated,

Thank you and best regards,

Patrick