LS means procedure proc genmod binary dependent variable

[Abby]

I have deduced my final model using proc genmod procedure in SAS. My dependent variable is binary; and I have categorical variables as predictors in my final model. I am interested in understanding which category of a variable is driving the whole effect, and was wondering if it is possible to use LSMEANS statement in genmod. When I try using the LSMEANS statement, I get the following error:

The model does not have a GLM parameterization. This parameterization is required for

         the TEST, LSMEANS, LSMESTIMATE, and SLICE statement. These statements are ignored.

Could someone please let me know the feasibility of using this procedure when we have a binary dependent variable? Thanks.

Best regards,

Abby

[SteveDenham]

Hi Abby,

Which version of SAS/STAT are you using, as the answer may depend on that?

Can you share your PROC GENMOD code?  It should be easy to change the CLASS statement to glm parameterization, which will open up all of these options. Something like:

CLASS <your list of class variables here> / PARAM=GLM;

A quick check showed that the GLM parameterization is the default.  I only got the error you mentioned when I specified PARAM=REF.

An alternative approach, which works with the REF parameterization, would be to use PROC GLIMMIX.

Steve Denham

[msb1]

Hi there,

Is it possible that when changing from param=ref to param=glm the model experiences convergence issues? Here's my original code:

proc genmod data=in.final;
class link_id agec (param=ref ref='1') sexf (param=ref ref='1') eth_f (param=ref ref='1')
year (param=ref ref=last) poly_meds_anyF (param=ref ref="0") Cluster (param=ref ref=last)
imd_local_quintile_f(param=ref ref=last) lang_F (param=ref ref='E');
model total_cons=sexf agec eth_f Cluster countltcs Cluster*countltcs imd_local_quintile_f poly_meds_anyF year lang_F /dist=negbin;
repeated subject=link_id / type=AR(1) covb corrw;
store out=catcont;
run;

Here's the param=glm code:

proc genmod data=in.final;
class link_id agec sexf eth_f year poly_meds_anyF Cluster imd_local_quintile_f lang_F/param=glm;
model total_cons=sexf agec eth_f Cluster countltcs Cluster*countltcs imd_local_quintile_f poly_meds_anyF year lang_F /dist=negbin;
repeated subject=link_id / type=AR(1) covb corrw;

store out=catcont;
run;

Thanks very much,

Marina