Hi Guido.

Why not to try to use a zeroinflated poisson models instead of gamma GLM? These models, if I am right, should allow you to handle data inflated by zeros, also allowing you to study how your inidpendent variables affect the zero component and the counts. it basyccally decompose your data in a biomial and a poisson distribution (which then use a lof llink function, which somehow is the same of the gamma distribution). these models are available in the ‘pscl’ package in R.

don't know if it fits your scope!

cheers 

lorenzo

Thank you for your responses. My data are not integers, so a poisson distribution can't be fitted. I would try to add a small values to 0. 

Guido,

The R package glmmTMB will handle Gamma regressions with zero inflation, if you don't want to add a little bit to every value. The package is set up for mixed models, but it will solve fixed effect only models as well. Good luck!

Jack