Not specifying trials in logistic regression with y = 0 has trials set to 0

[lauken13]

Bit of an edge case (and this is depreciated) but if you don't set the trials and have all 0s in your outcome, the trial number is set to 0 and the model returns the prior.

Probably doesn't require a fix (perhaps a warning), but I thought I should add an issue in case! :)

mod2 <- brm(y |trials(1) ~ 1, family = binomial(link = "logit"), data = data.frame(y = rep(0,1000)), backend = "cmdstanr")

vs

mod2b <- brm(y ~ 1, family = binomial(link = "logit"), data = data.frame(y = rep(0,1000)), backend = "cmdstanr")

[paul-buerkner]

Thanks! I think it is just about time to disallow not specifying trials() in binomial families altogether. It has been deprecated long enough.

[paul-buerkner]

I now no longer allow not specifying trial in binomial models to avoid this and similar issues.