PROC GLIMMIX, lsmeans odds ratios at a level of a binary covariate

maggieKlein — Sat, 22 Feb 2020 00:26:50 GMT

I am using the LSMEANS statement to request specific odds ratios in PROC GLIMMIX. I have an interaction term (binary by binary interaction) that I am requesting odds ratios for but would like to see the odds ratios at each level of another binary covariate.

a; binary 0,1

b; binary 0,1

c; binary 0,1

a*b interaction between a and b

lsmeans a*b / ilink oddsratio cl diff;

gives me beautiful output but once I try using the 'at' option to specify the value of another covariate SAS gives me a warning:

lsmeans a*b / at c =1 ilink oddsratio cl diff;

WARNING: Incorrect AT specification in LSMEANS statement. c is not a covariate in the model. The statement is
ignored.

I already know that this is because variable c is in the class statement and only continuous variables can be used in the 'at' option.

Does anyone know how to create odds ratios for an interaction at a certain level of a binary variable?

Thanks!

Re: PROC GLIMMIX, lsmeans odds ratios at a level of a binary covariate

tellmeaboutityo — Mon, 09 Mar 2020 22:14:54 GMT

I’m not 100% sure, but you might be able to change the combination of ref= statements in the class statement.

Class iv1(ref=‘0’) iv2(ref=‘1’);

I would assume the two reference categories would end up as the reference combination for the interaction.

A similar option might be to actually create your interaction variable and define the reference group.

topic Re: PROC GLIMMIX, lsmeans odds ratios at a level of a binary covariate in Statistical Procedures

PROC GLIMMIX, lsmeans odds ratios at a level of a binary covariate

Re: PROC GLIMMIX, lsmeans odds ratios at a level of a binary covariate