2
$\begingroup$

In R, you can use quasipoisson or quasibinomial for overdispersed Poisson or Binomial GLMs. But what if you have overdispersed Gamma GLM or other continuous variables?

$\endgroup$
2
$\begingroup$

The gamma already has a dispersion parameter, so there's no "overdispersed" case. It's like the normal in that sense (where $\sigma^2$ is a parameter, so there's no "overdispersed" Gaussian).

However, the readily available estimate of the dispersion parameter from the GLM output isn't ML (more generally the ML estimates for dispersion parameters in a GLM don't work well; the Poisson is a classic example). You can find an ML estimate, though, if you need it.

(If you're an R user, the package MASS that comes with R has a function for estimating the shape parameter after a GLM fit.)

| cite | improve this answer | |
$\endgroup$

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.