I briefly looked at cozigam a while ago, and I had the impression that it doesn't support the negative binomial distribution, which was a drawback for me. Also, I don't think there is much support available for it (it doesn't even seem to be available on CRAN any more either).

On the other hand, the authors of glmmADMB and MCMCglmm are both extremely active on user forums. For glmmADMB , Ben Bolker is very active on the R mixed models mailing list and stackoverflow., and for MCMCglmm Jarrod Had eld is very active on the R mixed models mailing list. There are also good online tutorials for both packages.

glmmADMB and MCMCglmm could both of of use. Take a look here:

http://glmm.wikidot.com/faq

Have you considered COZIGAM ?

The following paper shows a good application of the method to jellies....

Good luck, Chris

Hai LIU, Lorenzo CIANNELLI, Mary Beth DECKER, Carol LADD, and Kung-Sik CHAN

(2010) Nonparametric Threshold Model of Zero-Inflated Spatio-Temporal Data with Application to Shifts in Jellyfish Distribution

Journal of Agricultural, Biological, and Environmental Statistics

DOI: 10.1007/s13253-010-0044-4

http://doc.nprb.org/web/publication/project_0606_Liu%20et%20al_Nonparametric%20threshold%20model.pdf

I briefly looked at cozigam a while ago, and I had the impression that it doesn't support the negative binomial distribution, which was a drawback for me. Also, I don't think there is much support available for it (it doesn't even seem to be available on CRAN any more either).

On the other hand, the authors of glmmADMB and MCMCglmm are both extremely active on user forums. For glmmADMB , Ben Bolker is very active on the R mixed models mailing list and stackoverflow., and for MCMCglmm Jarrod Had eld is very active on the R mixed models mailing list. There are also good online tutorials for both packages.

Joseph Hilbe at the Jet Propulsion Library has written a book on Negative Binomial Regression in R. As of last fall when I contacted him, a zero-inflated negative binomial model was not available in R. Perhaps glmmADMB and MCMCglmm are the ticket.

For guidance on using ZIP and ZINB models in R see the book by Zuur etal 2012 Zero Inflated Models and Generalized Linear Mixed Models in R. These methods use MCMC by calling WinBugs in R. More recently JAGS in R can also be used. Chances are you will need to apply these methods as I guess your zooplankton data will be overdispersed so the variance will be much greater than the mean so poisson distribution is no longer valid, if not you are lucky but one source of overdispersion can be zero inflation. Negative Binomial can handle a bit more overdispersion but even then not that much more. Zuur, Hilbe and Ieno have just written a book on using JAGS for GLMs. I haven't seen this yet but assume they point you down the route of simulation using Gibbs sampling to avoid the need to apply hypothesis testing on parameters in models affected by overdispersion.

For zero inflated poisson and negative binomial, I recommend the package pscl.

Perfect!, Thanks for all those comments.

For my impression I have tried with pscl and cozigam packages and the first is better. COZIGAM always returned computational problems. In the Bayesian or at least with the MCMC the two mentioned packages glmmADMB and MCMCglmm are great!..Many thanks!

PD: The book of Zuur et al, is great!

I have used pscl package for may data set. bt every time it shows an error which is

" Error in zeroinfl(Y ~ ., data = dt, dist = "negbin") :

invalid dependent variable, minimum count is not zero"

my dependent variable is a count var and starts from zero, it is the number of antinatal care visit.

help me plz to find out the mistakes if possible.

Hi Syeda Asma,

I'm affraid that here will be impossible to answer your question. I haven't  had your problem (yet) but in similar circumstances people ask more info, like a histogram of your data (is really zero-inflated?), the session info of your work, etc.

Hope this help!,

Thanks Antonio

In the attached page the first one is my dependent variable. And I dont understand what is the mistake i am making here.

I have used censReg. It works really well with my data sets.