R help - Jun 2008 - GAM hurdle models

Hello,

I have been using mgcv to run GAM hurdle models, analyzing 
presence/absence data with GAM logistic regressions, and then analyzing 
the data conditional on presence (e.g. without samples with no zeros) 
with GAMs with a negative binomial distribution.

It occurs to me that using the negative binomial distribution on data 
with no zeros is not right, as the negative binomial allows zeros.

Does anyone know of a package that does negative binomial GAM hurdle 
models?  Or is there another quick fix to overcome this problem (e.g. 
using quasipoisson)?

Thanks for your suggestions,
John

On Thu, 5 Jun 2008, John Poulsen wrote:
> Hello,
>
> I have been using mgcv to run GAM hurdle models, analyzing presence/absence
> data with GAM logistic regressions, and then analyzing the data conditional
> on presence (e.g. without samples with no zeros) with GAMs with a negative 
> binomial distribution.
>
> It occurs to me that using the negative binomial distribution on data with
no
> zeros is not right, as the negative binomial allows zeros.
>
> Does anyone know of a package that does negative binomial GAM hurdle
models?
> Or is there another quick fix to overcome this problem (e.g. using 
> quasipoisson)?
Packages "VGAM" and "gamlss" both contain functionality for
fitted
zero-inflated GAMs. If I recall correctly, both have ZIP and at least VGAM 
also has ZINB as well as hurdle models for both distributions (under the 
name "zero-altered" models).

I think that both require to use the same set of regressors in both parts 
of the model. An implementation of zero-inflated models and hurdle models 
without GAM effects but with separate sets of regressors is available in 
"pscl".

Further information about the packages is available from:
   http://www.stat.auckland.ac.nz/~yee/VGAM/
   http://www.jstatsoft.org/v23/i07/
   vignette("countreg", package = "pscl")

hth,
Z
> Thanks for your suggestions,
> John
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
>