[R-sig-ME] hurdle modfels in glmmadmb
bbolker at gmail.com
Tue Feb 16 17:11:53 CET 2016
if you assume that the zero-causing process and the non-zero
count-determining process are independent (which is not necessarily
true, but which I think you're more or less forced to assume by
identifiability issues in the absence of more information), then you
should just be able to add the two distinct AIC values for the two
components of the model to get a single AIC value. (Since the processes
are independent, the log-likelihoods are additive, and the number of
parameters in the combined model is clearly additive too.)
You could cross-check by checking what the pscl package does (it
deals with zero-inflated GLMs too, although not mixed zero-inflated GLMs).
On 16-02-16 03:35 AM, Panagiotis Theodorou wrote:
> Dear Ben Bolker,
> I would like to thank you very much for developing the r package
> glmmadmb that I am currently using to model my zero inflated data.
> I have one small question regarding using two part (hurdle) models
> within the glmmadmb package.
> Is there a way to evaluate the fit (e.g. AIC) of a two part model as a
> whole and justify using it?
> Or the only way to justify using a hurdle model comes from prior
> hypothesis of two processes; one causing zeros versus non-zeros, and
> another process explaining the non-zero counts?
> Thank you in advance
> Panagiotis Theodorou
More information about the R-sig-mixed-models