Skip to content
Prev 363135 / 398502 Next

Odds ratios in logistic regression models with interaction

I suspect that "you can't get there from here"... a1$fit and a2$fit are not independent, so you can't work out the s.e. of their difference using sqrt(a1$se.fit^2+a2$se.fit^2). 

You need to backtrack a bit and figure out how a1$fit-a2$fit relates to coef(fit3.16). I suspect it is actually just the age times the interaction term, but since you give no output and your code uses a bunch of stuff that I haven't got installed, I can't be bothered to check....  

Once you have your desired value in the form t(a) %*% coef(...), then use the result that V(t(a) %*% betahat) == t(a) %*% vcov() %*% a  (asymptotically).

-pd
On 03 Aug 2016, at 15:08 , Laviolette, Michael <Michael.Laviolette at dhhs.nh.gov> wrote: