I am making a list of disadvantages of GEE and GLMM for a correlated binary outcome. So far I know that GEE requires a relatively large number of clusters, and that it produces profile curves that corresponds to no individual. GLMM on the other hand is not robust for violation of distributional assumptions, and is numerically intensive. Can you add more disadvantages of the two models ? Thank you !
Actually it is difficult to get reference regarding the drawback of GLMM. But I have attached here what I have seen on a website. It says
The only real downsides of GLMMs are due to their generality: (1) some of the standard recipes for model testing and inference that you have learned previously may not apply, and (2) it's easy to build plausible models that are too complex for your data to support.
Sorry for forgetting the website.