0
$\begingroup$

In contrast for generalized linear models (for GLM, there is e.g. AIC and null vs. residual deviance), I could not find such criteria implemented in R to judge my generalized equation (GEE) models.
Is it appropriate to look at

  • residuals vs predicted values scatterplot?
  • normal distribution of residuals (e.g. QQ plot)?

I am also happy to learn a better way to assess GEE models (available in an R package, or easily possible to calculate on my own if you show me how).
So far, I am using geepack::geeglm (or gee::gee)

$\endgroup$
1
$\begingroup$

You can use the quasi-likelihood under the independence model criterion (QIC; Pan, 2001) to assess the model fit. It's implemented in geepack. However, caution is advised when employing QIC and its use should not be routine, see Wang et al. (2015).

library(geepack)
data(ohio)
fit.e <- geeglm(resp ~ age + smoke + age:smoke, id=id, data=ohio,
             family=binomial, corstr="exchangeable")
fit.a <- update(fit,corstr="ar1")
QIC(fit.e)
QIC(fit.a)

Pan 2001: https://onlinelibrary.wiley.com/doi/abs/10.1111/j.0006-341X.2001.00120.x

Wang et al. 2015: https://onlinelibrary.wiley.com/doi/abs/10.1002/sta4.95

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.