Using ABSORB statement on unbalanced panel data

[rjz970]

Hello there. 

I want to estimate a fixed effect model by using PROC GLM on an unbalanced data set. I have two questions.

My first question is, if I am even allowed to use the ABSORB statement with an unbalanced data set?

My second question. I have an unbalanced panel data set, but my dependent variable theoretically should be constant over time, but not across age. Therefor my question is, if the PROC GLM statement is valid to use, if I do not include a time-variable in my model specification? I mean to ask, does SAS know that I'm regressing on panel data? This is my code:

PROC GLM DATA=PANEL_FE;
ABSORB ID;
CLASS AGE;
MODEL OPSPARING=DISP_IND DISP_IND_SQ AGE /SOLUTION;
RUN;

Other commands than PROC GLM will not run because of insufficent memory, and I can't increase the memory footprint on the external computer I am using. 

Your help is much appreciated. 

[StatsMan]

That is one approach you could take. That would approximate ID as a fixed effect in your panel model. An issue here is that any covariate that is measured at the ID level would be confounded with the ID effect and would not be estimable.

Do not give up on other procedures, though. If you want to treat ID as a random effect and if ID is a numeric variable in your data set, then PROC MIXED might work for you. Do not put ID on the CLASS statement, but do sort your data set by ID before running PROC MIXED. Then use a RANDOM statement like

   random int / subject=id;

Even with hundreds of thousands of levels to ID, this code will run in a reasonable amount of memory. To further improve performance, add the DDFM=BW option to the MODEL statement to simplify the degree of freedom calculation for this model. BW actually gives the DF you have in mind when looking at a panel model. 

Most statisticians are happier working with PROC MIXED for this type of model.