Is there a possibility to absorb with GENMOD?

[margalits1]

Hello all,

I use the following code to have both fixed effects and a two-way cluster. Is there a way to not show all the f.e categories, in a similar matter to the absorb option in GLM? I could not find an answer in the documentation...

proc genmod;
class gvkey year_data;
model capx_1 = incu_post_ipo sign_inc_p sign_p sale_rank market_share ind_code_count gvkey year_data;
repeated subject=gvkey*year_data / type=ind; run;
quit;

Thanks!

[PaigeMiller]

The documentation is very clear. There is no ABSORB statement in GENMOD.

Of course, you could perform your own an ABSORB-equivalent on the data before it enters GENMOD.

[margalits1]

Thanks, what do you mean by "own an ABSORB-equivalent on the data before it enters GENMOD."?

[StatDave]

The REPEATED statement effectively has the same effect. Typically the variable(s) specified in the SUBJECT= option are considered nuisance variables and are not included in the model. Instead, those variables define the clusters and the correlation within the clusters is taken into account by the GEE estimation method and parameters do not need to be estimated for the levels of the variables themselves.

[margalits1]

Thanks, Does this mean that I need to remove these variables in the model statement?
I am interested in both fixed effects and clusters by gvkey and year... therefore I have them on both model and repeated statements.

[StatDave]

Yes. If, like with the ABSORB statement in GLM, you want to avoid estimating parameters for all of the levels in those variables, then you need to remove them from the MODEL statement and only use them in the REPEATED statement. However, it is not uncommon that a variable that defines the clusters has a continuous nature and interest is in evaluating its linear (and maybe even quadratic) effect. In such a case, you can create a copy of the variable and use the one version in the CLASS and REPEATED statements and the other only in the MODEL statement. 

[margalits1]

Got it, thank you very much