Expit transformation

[Lidya1888]

Hi Dr. Ezra, I have one question. When our goal is to estimate a risk ratio, we use a log-link model (modified Poisson) instead of relying on the expit transformation from a logistic model. Is this because the risk ratio is not parameterized by a single coefficient in a logistic model (it depends on both b0 and b1), meaning the estimated RR will change across populations with different baseline risks? Thank you

[d-morrison]

@Lidya1888 with a logistic link function, the risk ratio depends on both coefficients and also depends on the covariate values being compared (try plugging in some values and you'll see that), so even within a single population the RR will vary.

[Lidya1888]

Thank you!