topic Re: multinomial ordinal longitudinal regression in Statistical Procedures

multinomial ordinal longitudinal regression

UcheOkoro — Mon, 16 May 2022 21:08:07 GMT

Hello,

I am struggling with codes for multinomial ordinal(4 levels) longitudinal regression and its interpretation. The outcome variable has 4 levels which are ordered. I used proc GENMOD and Proc GEE and the results are very different. I also need help with the interpretation. Initially I was hoping to get the differences of least squares means for each level of the outcome and predictor variable but it does not appear to give such results or am I doing it wrong. Please, I need help.

Thank you


PROC GENMOD DATA=COVERED.Demographics ;

CLASS  complete_adherence2(ref='1')  record_id ethnicity (ref='1');

MODEL complete_adherence2 = ethnicity 
/ DIST=MULTINOMIAL LINK=CUMLOGIT;
REPEATED SUBJECT=record_id/ corr=indep;
lsmeans ethnicity/  ILINK diff e cl oddsratio ;
RUN;


PROC GEE DATA=COVERED.Demographics DESCENDING;
CLASS complete_adherence2(ref='1') week record_id ethnicity (ref='1');
MODEL complete_adherence2 = ethnicity/ DIST=MULTINOMIAL;
REPEATED SUBJECT=record_id / TYPE =indep WITHIN=week LOGOR=EXCH;
lsmeans ethnicity/  ILINK diff e cl oddsratio ;
RUN;

Result from PROC GENMOD

Result from PROC GEE

Re: multinomial ordinal longitudinal regression

StatDave — Mon, 16 May 2022 22:27:04 GMT

I recommend that you start by making the code the same in both procs except for "GEE" vs "GENMOD". DESCENDING should be removed or used in both procs - it reverses the order of the response levels and will certainly change the results. You use the WITHIN= and LOGOR= options only in one of the two procs. With LOGOR, a very different model is fit, so only one of TYPE= or LOGOR= should be specified.

Re: multinomial ordinal longitudinal regression

UcheOkoro — Tue, 17 May 2022 02:56:48 GMT

Thank you for your response. I made the two models as identical as possible but it did not help.

The following is my code and the results.

Thank you.

PROC GENMOD DATA=COVERED.Demographics ;
CLASS  complete_adherence2(ref='1')  record_id ethnicity (ref='1');
MODEL complete_adherence2 = ethnicity 
/ DIST=MULTINOMIAL LINK=CUMLOGIT;
REPEATED SUBJECT=record_id/ corr=indep;
lsmeans ethnicity/  ILINK diff e cl oddsratio ;
RUN;

PROC GEE DATA=COVERED.Demographics ;
CLASS complete_adherence2(ref='1') record_id ethnicity (ref='1');
MODEL complete_adherence2 = ethnicity/ DIST=MULTINOMIAL LINK=CUMLOGIT;
REPEATED SUBJECT=record_id / TYPE =indep ;
lsmeans ethnicity/  ILINK diff e cl oddsratio ;
RUN;

The result from PROC GENMOD

The result from PROC GEE

Re: multinomial ordinal longitudinal regression

SAS_Rob — Tue, 17 May 2022 14:37:13 GMT

Could you post both the LOG and the full output from the two procedures?

Re: multinomial ordinal longitudinal regression

UcheOkoro — Tue, 17 May 2022 15:31:20 GMT

I was able to get identical results with the following code

PROC GENMOD DATA=COVERED.Demographics desc;
CLASS  complete_adherence2(ref='1')  record_id ethnicity (ref='1');
MODEL complete_adherence2 = ethnicity 
/ DIST=MULTINOMIAL LINK=cLOGIT;
REPEATED SUBJECT=record_id/ corr=indep;
lsmeans ethnicity/  ILINK diff e cl oddsratio ;
RUN;

PROC GEE DATA=COVERED.Demographics desc;
CLASS complete_adherence2(ref='1') record_id ethnicity (ref='1');
MODEL complete_adherence2 = ethnicity/ DIST=MULTINOMIAL LINK=CUMLOGIT;
REPEATED SUBJECT=record_id / TYPE =indep ;
lsmeans ethnicity/  ILINK diff e cl oddsratio ;
RUN;