Hi Tasneem,
Could you give more information about your data set? Is the response a survey score or an econometric variable? Why is it zero inflated? What is the independent ordinal variable and which values does it take between 0-2? Do you have other covariates in the model? Have you tried any transformation on the response to normalize the data?
For now, I can mention "zero inflated Poisson" and "zero inflated negative binomial regression" (the latter for over-dispersed responses) if transformation is not helpful. However, you say you are dealing with a mixed model here, and therefore you must be careful how you model the zeros. If you have time-series with consecutive counts of zeros, then both the random effects as well as the binary part of the zero inflated model will try to capture the zeros and that usually makes the model unnecessarily complicated. Maybe a simple mixed Poisson or mixed negative binomial model could already produce good estimates.
There are various packages that could handle such models, both frequentist, e.g. "pscl" package with zeroinfl function, as well as Bayesian, e.g. MCMCglmm setting faimily to zippoisson. See:
Zuur, A., and I. Saveliev. "Zero Inflated Models and Generalized Linear Mixed Models with R (Highland Statistics Ltd.)." (2012).
--Ehsan
------Original Message------
Dear All,
I have a longitudinal clustered data set with response as count variable (zero inflated) and the independent variables are measured on an ordinal scale of (0-2, higher score is better). What is the model you would suggest to fit and whats the best package to do so in R, if any?
Any suggestions would be highly appreciated.
Best,
------------------------------
[Tasneem] [Zaihra]
[Post Doctoral Fellow]
[McGill University]
------------------------------