Why Generalized estimating equation (GEE) is not popular? I am a newbie in econometric analysis.
When I was in Master and PhD course of public health field, I learned mainly about mixed model and GEE for panel data analysis.
I learned GEE is robust, does not require strict assumptions, very easy to use. So I use GEE for a while.
However, the more I studied and read articles, I felt GEE is not really popular in econometric analysis. 
Is there any weakness for GEE compared to Fixed model or GMM?
 A: I don't think the reason for GEE not to be popular is "software availability/implementation". Almost all software I am aware of and have been working for a while (STATA, R, SPSS) can do GEE.  I wonder if there is any software that can do mixed and never(or difficult to do) GEE. 
As explained here "Mixed Model Versus GEE estimates and which to use" in a little bit more detail, I don't think it is neither the interpretation that matters. 
Even some authors (Hubbard AE et al, 2010) argue mixed model suffers from " unverifiable assumptions on the data-generating distribution". 
Literature is full of the idea that "GEE allows robust inference even if a choice of correlation model is wrong or misspecified" and in case of mixed model "SE not robust to model misspecification". 
I personally think GEE is computationally exhaustive than mixed; whether in hand or in software, if you do it nice and clean; that may be.  
A: The first reason is almost certainly software availability/implementation. You do not find GEEs implemented in all software for all (or at least a wide selection) of the cases, where it could be applied. In contrast, a lot of software makes it quite easy to specify the (generalized linear) mixed (effects) model for the same problem (when applicable). As a result mixed model alternatives are quite widely used and people feel familiar with them.
A second reason might be that the results of a GEE may need to be interpreted differently than those of a mixed model. (Note: I write "may", because in all the cases I am aware of this is the case, but I cannot exclude the possibility that there are exceptions.) E.g. if you have multiple observations for the same person in an experiment (comparing intervention A vs. B) and account for the correlation of these via a GEE or a GLMM, then the GEE will estimate a population-level effect for the intervention and the GLMM a person-based effect. By this I mean that the GLMM will provide an estimate of "What is the effect of A vs. B has for a person?", while the GEE will give an estimate of "What is the average effect on a population of people with the covariate distribution like in my experiment?". Very often we are more interested in the first question and could also still derive the answer to the second question from the answer to the first one. Thus, one might argue that the GEE approach is not so useful here.
