topic Re: Proc glm does not provide me estimates in Statistical Procedures

Proc glm does not provide me estimates

HEB1 — Mon, 18 Dec 2023 11:59:20 GMT

I run the following code ,

with fixed effects of ID year and industry:

proc glm data=library.modify1;
class two_digit_SICnum gvkey fyear;
model roa =dummy two_digit_SICnum gvkey fyear/solution ;
run;
quit;

however, the program does not provide estimates and my memsize is set to 4!

Thank you

Re: Proc glm does not provide me estimates

PaigeMiller — Mon, 18 Dec 2023 13:20:29 GMT

Please show us the output so we can see what you are seeing. Please use the "Insert Photos" icon to include a screen capture of your output in your reply; do NOT attach files.

Re: Proc glm does not provide me estimates

HEB1 — Tue, 19 Dec 2023 15:14:16 GMT

I can’t even produce the output now, the program get stack.
Basically what I want to run is dependent variable on a dummy and fixed effects.
The dummy equals 1 for almost 2,000 obs and equals 0 for 75,000 obs.
I think that problem starts when I use ID fixed effects. I have approximately 20,000 distinct ID.
Any idea how can I run this analysis?

Re: Proc glm does not provide me estimates

PaigeMiller — Tue, 19 Dec 2023 15:21:40 GMT

No need to create dummy variables for PROC GLM. Use the CLASS statement instead of dummy variables.

Running with 20,000 distinct ID seems like it will cause memory related problems and may fail. Why do you need to have 20,000 distinct IDs in your model?

Re: Proc glm does not provide me estimates

ballardw — Tue, 19 Dec 2023 15:26:56 GMT

You say that you have about 2,000 observations where your dummy equals 1 and 20,000 ids. That tells me that you have about 18,000 (possibly more) IDs that all the values are 0. So there is no variability for those ids unless there is a lot more going on in your code than you have provided. No variability, no error typically means no "estimate".

At best, 18,000 of those IDs are not going to have any prediction for the value=1 because they never see one.

As an absolute minimum you should share the code you are attempting and the LOG with the code and any messages.

Re: Proc glm does not provide me estimates

jiltao — Tue, 19 Dec 2023 15:27:43 GMT

You might consider using PROC MIXED and treating ID as a random effect.

If you must treat ID as a fixed effect, you might consider using the ABSORB ID; statement in PROC GLM and take ID out of the MODEL statement. Make sure your data is sorted by ID. More information on the ABSORB statement can be found in the documentation -

https://go.documentation.sas.com/doc/en/pgmsascdc/9.4_3.5/statug/statug_glm_syntax02.htm

Thanks,

Jill