topic Re: How to get a odds ratio in Proc Glimmix for continuous predictors with interaction in Statistical Procedures

How to get a odds ratio in Proc Glimmix for continuous predictors with interaction

timothy19 — Thu, 10 Nov 2022 17:31:46 GMT

Hi Everyone,

I am trying to perform regression analysis with binary outcome and continuous predictors with repeated measurement on subjects.

proc glimmix data=data;
where position="Post";
      class sub_id;
      model y_variable (event="1") =  Force_N10 Resistance__ohms10 Time_in_Seconds  Resistance__ohms10*Force_N10/ solution dist=binary link=logit ddfm=satterth oddsratio;
      random intercept / subject=sub_id solution;
nloptions tech=nrridg;
lsmeans Resistance__ohms10 / cl ilink plots=mean(ilink cl);
ods exclude solutionr;run;

I tried to use the code above but got the error below

I can't put the continuous variable in the class statement. Can anyone please help?

Tim

Re: How to get a odds ratio in Proc Glimmix for continuous predictors with interaction

PaigeMiller — Thu, 10 Nov 2022 18:07:43 GMT

I think the message is clear, you can't do that. It also doesn't make any mathematical or logical sense to compute LSMEANS on a continuous variable.

What is it you are trying to learn here from Resistance__ohms10? If it is what your title says "odds ratio ... for continuous predictors with interaction", maybe something like this from the GLIMMIX documentation is what you want:

model y = A x A*x / dist=binary oddsratio(at x=3);

Re: How to get a odds ratio in Proc Glimmix for continuous predictors with interaction

StatDave — Thu, 10 Nov 2022 18:07:43 GMT

If you remove the LSMEANS statement, which can't be used for continuous variables, the ODDSRATIO option that you have in the MODEL statement will provide a table showing the odds ratio for each variable in the interaction for a unit change.

Re: How to get a odds ratio in Proc Glimmix for continuous predictors with interaction

timothy19 — Thu, 10 Nov 2022 18:27:47 GMT

Thanks a lot for this. Just figured I should remove the lsmeans statement.