I've been fitting a GEE on a longitudinal study with a complete case analyis.
First problem arises that every working correlation gives the same standard errors and convergence is reached after 1 iteration. The standard errors should at least deviate slightly depending on the structure in the data and the used working correlation.
Convergence after 1 iteration might also slightly be weird.
A second problem can be seen in the working correlation matrix, which is identical for all types of working correlation:
Are there any ideas to what is causing this weird behaviour?
An example of one of the models used in R:
model1 <- gee(Binary_HC ~ log_time + male + age + cardio + reject + HC0, id = factor(id), data = renal_transformed, na.action = na.omit, family = binomial(link = "logit"), corstr = "independence")
Best Answer
The easiest working correlation matrix to use for GEE is based on the independence assumption, for which all diagonal elements of the matrix are ones, and all off-diagonals are zero. Your working correlation matrix only has one diagonal element set to one, so that's probably what's causing the issue.
It look's like you're using R, so look up in the package you are using how to specify the "Independence" type of working correlation matrix.