Within the GLM procedure in SPSS, when you estimate a gamma model, the software automatically excludes cases with a dependent variable value of "0". Does anyone know how to retain these cases in a gamma model? Or, does anyone know of an alternative general linear model I can try that retains "0" values? The data are on an interval scale and are not normally distributed, so I need something besides ordinary least squares.
If the zeros are genuine values, then you can use
the zero adjusted gamma distribution (ZAGA) in the R package GAMLSS
(see also www.gamlss.org/).
This models the distribution as a gamma GA(mu,sigma) , with probability nu,
and zero with probability (1-nu).
The three parameters mu, sigma and nu can then each be modelled using explanatory variables.
I don't think ZAGA is available in SPSS.
Thanks for the feedback gentlemen, but I still have to figure out a way to actually run this gamma model in SPSS without these cases being dropped. I know both 'R" and "Stata" allow this, but unfortunately SPSS and SAS do not. If it is not possible, is there another GLM option native to SPSS that I may use for non-normal, scale data? Thanks again for your responsiveness.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS