In my study, I need to apply GLMM or GEE. Some days back I was pretty much sure about using GLMM, but some of my friends suggested me to use General Estimation Equations (GEE). I tried looking on the internet for the same and failed miserably. So, please help me in differentiating between them and tell me what are the conditions in which each analysis is used.
In studies using GLMM, there are given source or group that are correlated but there is no specific correlation technique or pattern. For GLMMs the value of the sources is calculated. Whereas GEE doesn’t use to make calculations regarding the sources, in fact, the model for the mean response depends only in the covariates and not on the random effects. Coming to the advantages of the analysis:
GEE: In this type of analysis, you can stipulate several types of within-group correlation, i.e., unstructured, autoregressive, exchangeable or compound-symmetry, stationary, auto-correlation, ante-dependence, and some other correlation structures.
I would suggest you use GLMM in your study.
Hey, you have asked a very important question. When I was doing my PhD, I was also confused regarding the same. I have also seen many PhD scholars getting confused regarding the same and sticking to GEE, not because of its accuracy and correctness, but because of its history. I have researched thoroughly about the difference between the two, and have been able to find a paper related to it. I found the following reference relevant:
- Lindsey, J. K. and P. Lambert (1998). On the appropriateness of marginal models for repeated measurements in clinical trials. Statistics in Medicine 17, 447–469.
You must read from the book, as it has a detailed explanation of the difference between the two.
I think, to differentiate between GLMM and GEE, one should first decide which model is used in the study, marginal or conditional model? Also, decide which model answers the research questions effectively. After deciding which model is apt for your study, it becomes easy to decide which on these model will be used. If you are using a conditional model, then, you must employ GLMM, but if, you are using a marginal model, then, you can use GEE directly and GLMM in an integrated form.