Post hoc test after multinomial regression

[akgh84]

Hello,

I have a dataset with a categorical outcome variable (strategy) that includes four groups. I also have three continuous (per_capita, unemployment, and hospitalsize) and three dummy variables (ownership, teaching, and membership) as predictors of the probability of using being in each of the four strategies. My data is longitudinal and I am using the Generalized Estimation Equation model. Here is my basic model:

proc gee data=paper2.strategy;
Class code year ownership (ref="0") membership (ref="0") teaching (ref="0");
model strategic_group1= per_capita unemployment hospitalsize ownership teaching membership / dist=multinomial link=glogit;
repeated subject= code/ within=year;
run;

My question is how can I conduct post hoc analyses to compare the resulted coefficients among four strategies for both continuous as well as dummy variables?

Thank You!

[StatDave]

If you have specific hypotheses (defined as linear combinations of the model parameters) that you want to test, then you can use the ESTIMATE statement. If you want to make comparisons among the levels of the categorical predictors, the LSMEANS statement is available.

[akgh84]

Thank you for your response.

I am aware of the estimate and the lsmeans statement but I am trying to compare two levels (pairwise comparisons) within my four level outcome and if I understand correctly the estimate or lsmeans statement is for the predictor variables not the outcome. Please let me know if this is not correct.

[StatDave]

Sounds like you might have the second situation mentioned in the opening sentence of this note. If so, several approaches are presented in the corresponding section of the note.