Hi All,
I was trying to analyze some counting data using proc glimmix. I did have a lot of zeros in my counting data, and I specified that the distribution of the counting data follows negative binomial. And then I got negative estimates from the SAS output. I was wondering why specifying negative binomial distribution would give me negative estimates? Hope someone can help me out. I'll really appreciate it!
Thanks,
Jing
This is likely related to the chosen link function. By default, predictions are made on the link scale level and the link function by default for "negbin" is "log". Use "ilink" to make sure the predictions are on the same scale as the outcome (ex: output out=pred pred(ilink)=pred)
See : https://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_glimmix_a0000001419.htm
Alexia is right. You can also exponentiate the estimates (i.e., raise e [2.72] to the power of the estimates) to derive the multiplicative effect of your independent variables on your dependent variable.
Hi guys,
I guess this is really a naive question, but I wanna make sure that I understand everything correctly. Let's say, if the proc glimmix gave me a negative estimation (estimation of LS mean) of my counting data, e.g. -3, can I say that the estimation of the average number I got from my counting data is e^-3?
Thanks!
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics