I have a panel cost data with some zero values (not missing values but rounded to 0). How should I handle zeros when I use SAS Proc Glimmix with Gamma distribution? Maybe change zero to very small values? Or is there an option for it (which I couldn't find) in PROC GLIMMIX?
Janne
Best Answer
I think you should consider a Poisson model instead. There was discussion of when it works and when it does not as an approximation to a continuous model on Stata blog. If the values are restricted to a range from 0 to 100, you can also do binomial regression with it.