Generalized Estimation Equation (GEE)

[Carolina Gracitelli]

Dear Statalists,
I am trying to use GEE for cross-section study and I am a little bit in doubt about the command because I performed in several ways and the results is slight different. So I have two question:
1) Anybody nows if it is possible to generate in Stata a number for power of association in GEE (as we have the R2 in linear regression). Because p value give us if there is significance or not but p value don't give us the power of association as R2. I might think the QIT is kind of different.
2) I am using GEE because of the dependence of the two eyes (right and left). I didn't use the variable time in my analysis as visit number or years or month because I just have one measure for each eye for each patient. I am trying to associate two different continues variables for 70 eyes in 35 patients. And I have two groups (patient and healthy control group). So, I plotted:
xtgee variable1 eye group variable2 , fam(gaus) link(iden) i(eye) t(eye) corr(exc)
But the results doesn't make sense.
Thank you
Carolina

[Clyde Schechter]

On the one hand, you really need to show the output and tell us specifically what about it doesn't make sense.

But, in any case, specifying i(eye) and t(eye) seems like a problem. It seems more likely that you want i(person) t(eye), where person is a variable that identifies distinct people.

[Carolina Gracitelli]

Dear Clyde, thank you for your answer. The problem is that as I am doing GEE for a cross-section study (I am just using GEE because of the dependence of the two eyes), I don't have any variable for time, visits etc. So I might have to drop the t(eye). Right? Thank you
My second question is about the R2 for GEE. It seems that it is impossible to get a value for R2 for GEE analysis right?
thank you
Carolina

[Joseph Coveney]

It looks like you're going to want something like

xtgee outcome i.eye i.group . . ., i(patient_id)

using the defaults for distribution family, link function and working correlation structure. As Clyde mentioned, eye is not the cluster. Patient is.

I don't know about R2. I rarely pay attention to it with regress and never beyond.

[Carolina Gracitelli]

Thank you Coveney. I will try your suggestion.