I have data with a binary outcome (success/failure) and a binary explanatory variable (treatment/control). For each subject (this is a clinical study), I have two observations, coming from two eyes. Obviously, the observations are correlated within a subject. I wonder if I should use GLMM or GEE to compare the two groups and to calculate the probability of success in each group. I wanted to ask what you think. I know the basic theoretical differences (that GLMM estimates a regression for each subject and GEE averages them all), yet I wanted to ask if anyone can specify the similarities (most of the time I get similar proportions), and the differences. How do you choose a model ? Based on what ?


Both of the models can be the correct approach. Both of them can deal with non-Gaussian distribution of the outcome variable and both can deal with the dependency (within subject) in your data. The important question would be: what specific research question are you aiming to answer? In an experimental design the individual subjects might be of interest. In that case a GLMM is preferred since it leads to an interpretation for individual subjects. If you are interested in the overall effect on the population a GEE is the model of choice, since it leads to an interpretation of regression coefficients for the population.

  • $\begingroup$ Thank you. So in this case, if the question is to show that treatment A is better than treatment B, should I use the subject specific model ? $\endgroup$ – user96870 Dec 2 '15 at 10:17
  • $\begingroup$ If you expect different effects for different subjects you could. But as I understand it, you are interested in the effect of the treatment on the group (and not on the individual), so GEE would be the way to go. Again, none of them is wrong, but interpreting the effects of GLMM as effect on the whole group would be wrong. $\endgroup$ – Ivo Dec 2 '15 at 10:32
  • $\begingroup$ I see. And one more question please. What is the difference between a GEE, and an "R Side" mixed model ? In SAS terminology, what is the difference between PROC GENMOD and PROC GLIMMIX with the residual statement? $\endgroup$ – user96870 Dec 2 '15 at 10:40
  • $\begingroup$ Look at what the abbreviations stand for. GENMOD is for Generalized linear models (no mixed effects). GLIMMIX is for Generalized Linear Mixed Models. $\endgroup$ – Ivo Dec 2 '15 at 10:50

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.