Interpretation of relative risk ratio in -mlogit-

[Sonnen Blume]

Hi,

I'm using a multinomial logit model, and see that the results are given as either coef or RRR. There are not many articles that report RRR. So I am wondering if its similar to RR, or if it can be interpreted as likelihood e.g. 12% more likely.

Thank you for your suggestions!

[Maarten Buis]

A relative risk ratio is not an Risk Ratio. A relative risk is the risk if choosing an outcome relative to the reference outcome, which is the ratio of these two risks. The ratio of two risks is an odds*. A relative risk ratio is thus an odds ratio.

---

* There are some differences between disciplines in the exact definition. In some an odds refers only to binary choices, while in others an odds exists also for choices with more than two choices. Both definitions are somewhat awkward: In the former you need to invent a different word for an odds ratio in the case of more than two outcomes, and in the latter you need to explicitly state the reference outcome. StataCorp chose the former definition, which is why we ended up with RRR for coefficients that should (IMHO) have been called OR. https://www.stata.com/statalist/arch.../msg00085.html