Odds ratio against a reference level with surveyfreq (with more than two levels)

[JacquesR]

I am performing a stratified/clustered analysis using proc surveyfreq.

It is a study in various districts (strata) at various hospitals (clusters) randomly assigned to one of three study arms (control, Experimental1, Experimental2).

I would like to get the odds ratio for the comparison of E1/E2 to Control, but I do not want to specify an analysis using only those hospitals (i.e., one analysis with E1 vs Control, another with E2 vs Control), as that would most certainly affect the overall variance, and defeat the purpose of using Surveyfreq. Is there a way to get the odds ratio in this instance?

[SteveDenham · Accepted solution]

Take a look at PROC SURVEYLOGISTIC.  If you fit a generalized logit link, with CONTROL as the reference group, you should get the OR's you are looking for. It includes both CLUSTER and STRATA statements, so that the information you presented all should be incorporated. The example here should help you get started.

SteveDenham