Depending on the context, you could stratify on A and then have just B A*B in the model. We have had to do that on occasion when the treatment group (A) did not meet PH. You'll still need to look at A*B for PH, but it may finess the problem.
Also, some violations of PH just make the test more conservative. If so, and A*B may still be a useful test and estimand.
A is infection status (control/infected) and B is time in study (early/late) A and B dont make sense if considered alone in this model so it is the interaction that is most important - this ends up comparing control/infected early in study and control/infected late, so stratifying on time in study (B) is feasible. Will that work even though i have a clustering factor of farm (so an ID Farm statement) in the model?
I hadnt considered that some PH assumptions make the test more conservative. How would i know if this is the case with this violation?
Message was edited by: reb