conditional_effects and points for binomial model

[avehtari]

The following plots the conditional effects as probability but the data points as counts. I was not able to figure out how to have both either in probability scale or count scale. If it's possible, then documentation could be updated, if it's not possible then additional argument to make it possible could be added

library(brms)
options(brms.backend = "cmdstanr", mc.cores = 4)
ggplot2::theme_set(bayesplot::theme_default(base_family = "sans", base_size=16))

df_bioassay <- data.frame(
  dose = c(-0.86, -0.30, -0.05, 0.73),
  batch_size = c(5, 5, 5, 5),
  deaths = c(0, 1, 3, 5)
)

bfit1 <- brm(deaths | trials(batch_size) ~ dose,
             family = binomial(),
             data = df_bioassay,
             refresh = 0)

plot(conditional_effects(bfit1), points=TRUE)

[paul-buerkner]

You can make the predictions both on the count scale by adding to conditional_effects: conditions = list(batch_size = <your value>). This will of course only make it comparable with points that have that batch_size. It could be feature to auto-scale points in some way but I am not sure I would know how to safely achieve that.

[avehtari]

It seems
conditions = list(batch_size = <your value>)
would work only if the batch_size is the same for all observations. In the above example, it happens to be true, but not in general. For example, if we change the data to be

df_bioassay <- data.frame(
  dose = c(-0.86, -0.30, -0.05, 0.73),
  batch_size = c(5, 5, 5, 6),
  deaths = c(0, 1, 3, 6)
)

I don't expect that plot(conditional_effects()) would be updated to do what I want, but maybe the documentation could mention the issue for binomial models.

ps. marginaleffects::plot_predictions() fails for this, too

[paul-buerkner]

Good point. I will adjust the doc