Brms iformative prior for bionomial regression

Interfaces brms
21

This absolutely terrific.
I really like the quick way probability density plots can be created with parse_dist.
Two comments about option 1

  1. Option 2 is much preferred!
  2. I think plogis() is putting values on a probability scale (ie btw 0-1) while a log odds scale could in theory go from - to + infinity
    I really appreciate your time and am possibly a bit smarter following your replies.
1 Like
22

came upon this thread when trying to set reasonable priors for an aggregated binomial model.
Regarding your 2nd point, I tried the transformation from normal to logit scale by simulating from a normal distribution with mean=0, SD=10 and using plogis() once and the logit transformation directly log(p / (1 - p)).
The result looks different indeed:

000027
000027840×519 6.59 KB

000026
000026840×519 6.94 KB