kenjioman
commented
2 days ago Perhaps also confusing (as seen with the next line, even): there is an alpha parameter for the negative binomial likelihood, but our model also has an $\alpha$ parameter. The text shown here is using the Greek $\alpha$ in both instances, even though they are referring to two different concepts.
I think it should say the larger alpha, the smaller the variance, because $\mu^2 / \alpha$ becomes smaller. In the limit, when $\alpha \to \infty$, the variance is $\mu$ like in the Poisson case.