Confidence interval for OR with interactions

[JuanVte]

Dear all,

I am trying to compute confidence intervals for a logistic regression parameters with interactions. SAS automatically gives the intervals for the main effects but not for the interactions.

I now that OR(A*B) = exp(effect A + effect B + effect(A*B)) but what about the confidence intervals?

Regards!
Juan Vicente

[Doc_Duke]

The only way to get a CI on the interaction is to recode the data with the IA as a "main effect"
ABia=a*b;
and include it in the model as a separate variable.

Although you can technically get a CI that way, it's nearly impossible to interpret. Good luck with that. If A & B are binary, it's OK, but beyond that is a stretch.