Hi. I have a binary outcome where the probability of
outcome is low so I want to get an adjusted relative
risk (rather than odds ratio) using glm. When using
glm with fam(bin) link(log) command, I get a warning
"parameter estimates produce inadmissible mean
estimates in one or more observation". Could I still
use the results or is there a way to get around this
problem?
I think it's time to listen to Stata. The trouble with the relative
risk is that it can produce nonsense estimates of probability in some
cases. If a risk factor has a relative risk of 2, and a case has a
predicted probability, based on other risk factors, of 0.6, then the
addition of the final risk factor will bring the predicted risk to 1.2.