1
$\begingroup$

I have fit a marginal logistic model or GEE Logistic Regression model using SAS' proc genmod to obtain estimated parameters associated with mortality (death). Using SAS, I am able to obtain subject-level predictions, $\hat{p}$. However, as I understand it, marginal models are population average models, so does it make sense to obtain these subject-level predictions? I was thinking about taking these $\hat{p}$'s and making prediction of death based on a cut-point and then performing a cross-validation with the actual known values of death as a means to validate my model. Does it make sense to do this with individual level predictions with GEE?

$\endgroup$
2
$\begingroup$

You will obtain predictions of the same type as you obtain in other standard statistical models, that is, conditional on covariates. Subject-specific predictions you would again if you could also condition on the specific outcome data of a subject, which is not possible for GEEs.

In any case, it makes sense to cross-validate these predictions to see the degree of potential over-fitting in your data.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.