PROC GLM

noetsi

Fortran must die
#1
PROC GLM is able to do more with categorical predictor variables than PROC REG (which lacks a class statement). But many assumption diagnostics exist in PROC REG that do not as best I can tell in PROC GLM.

So my question is can you specify a model in PROC GLM with dummy variables and then specify the same model in PROC REG, using the parameter estimates from PROC GLM and the diagnostic tools in PROC REG.

I have doubt this works, because PROC GLM treats dummy variables as dummy variables and PROC REG treats them as (I think) an interval variable with two levels.
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
I think they both kick out the exact same coefficient estimates. I think glm has more options then you think (isn't there a diagnostic ods graphics). There is also glmselect, which is supposed to be their cadillac.
 
Last edited:

noetsi

Fortran must die
#3
I am pretty sure when I researched this last year I found the diagnostics don't exist in PROC GLM. That they generate the same results is great to know. Thanks.

I have never heard of glmselect. I will have to look this up and see if we have that PROC.
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
Yeah I think I looked up glmselect after posting that and it is really directed toward model selection and selection criteria. So Cadillac for selection.