I did a glm analysis with quasipoisson family instead of poisson due to an overdispersion of my data.
With this, i'm not able to use the fonction dredge or step in R to select the best fit model because we don't have the AIC value with quasipoisson.
I will appreciate if you have any suggestion to me.
Use a negative binomial instead, you will be able to use the AIC criteria.
You can use prediction error using cross-validation to find the best suitable model
I am dealing with the same problem, did you find any solution? Thank you
if you gonna compare it with some quasi-family, you can use quasi-AIC. But if you gonna compare it with others like binomial negative regression, you can read this reference.
https://digitalcommons.unl.edu/cgi/viewcontent.cgi?article=1141&context=usdeptcommercepub
hopefully thiscould help u.
These two references below also may be helpful. The first is a kind of 'How to..." note by prof. Ben Bolker. It briefly explains, why there is no AIC in glm(..., family = 'quasipoisson') output and how to deal with it. While the second (again, by Ben Bolker) discusses why it might be (and might not) appropriate to use F-test to compare quasilikelihood models.
https://cran.r-project.org/web/packages/bbmle/vignettes/quasi.pdf
https://stats.stackexchange.com/questions/418832/why-the-absence-of-probability-distribution-for-using-quasi-likelihood
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity