| Detailed description |
Wen fitting a beta-binomial model with glmmadmb or a zeroinflated poisson or a zeroinflated negative-binomial model, the pearson residuals are not calculated correctly. For example, when I fit a binomial model that is overdispersed and then switch to the beta-binomial, this results in nearly the same pearson residuals than from the binomial.
The actual parameter estimates seem to be correct, as I fitted the same models with glmmABMB, INLA, JAGS and the pscl-package (zeroinfl) and always got the same estimates. But calcualting the pearson residuals by hand from INLA and JAGS, I got differences to those obtained by resid(modelADMB, type = "pearson").
The models I fitted were very simple, and did not contain any random effects. Find the very ugly code for calculating the residuals from a glmmADMB-object attached.
|
|
