What are the assumptions of the generalized estimating equations GEE?
GLM...
1. Non normal stochastic
2. Non linear relationship
See, the article on my profile"Preprint - Board Characteristics and Financial Reporting Quality"
The assumptions of the longitudinal (panel) GEE model are related to the cross-sectional GLM model:
- (1) Responses to your dependent variable (Y1, Y2, …, Yi) are correlated, i.e. there is no independence of responses in cases over time. Therefore, also the standard errors are correlated;
- (2) Yi must not be normally distributed. Alike, the relationship between Yi and covariates does not need to be linear. However, the relationship between a transformation of Yi (described by the link function, e.g. as poisson, gamma, logit, ect.) and covariates must be linear;
- (3) Homogeneity of variance must not be satisfied;
- (4) Within-subject covariance has some structure that must be specified a priori (e.g., independent, exchangeable, unstructured).
I think the decision about the structure of covariance is the main challenge with GEE. Often independence is unrealistic, although it is the easiest way to get your model converge (since correlations are all set to zero). To be on the safe side, however, I would probably model covariance as unstructured. Additionally you can specify the Huber-White estimator to model heteroscedasticity-robust standard errors, unless your design is heavily unbalanced (number of responses across individuals is largely different).
GLM...
1. Non normal stochastic
2. Non linear relationship
See, the article on my profile"Preprint - Board Characteristics and Financial Reporting Quality"
See this link for a good discussion:
https://www4.stat.ncsu.edu/~davidian/st732/notes.html
Best, David Booth
This is not an assumption as such but a limitation
that is GEE estimates population average values rather than cluster specific; if you estimate random effects model you can get both
see my long answer to this question
https://www.researchgate.net/post/Where_can_I_find_good_material_on_the_difference_between_mixed_models_and_gee_models
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS