`rate()` modifies dispersion in negbinomial() models but help file suggests it's equivalent to an offset

[colinorourke]

The help file for rate() says:

“Using rate(denom) is equivalent to adding offset(log(denom)) to the linear predictor of the main parameter…”

This appears to be true for Poisson models, but in negative binomial models, rate() also causes the dispersion parameter (shape) to be scaled by the denominator. This results in different variance behavior than when an offset is used explicitly.

This isn’t mentioned in the help file, and it can be surprising if you're expecting rate() to only affect the mean structure. It would be helpful if the documentation clarified this distinction.

Thanks for all your work on brms.

[paul-buerkner]

Could catch. will fix the doc soon.