Zero-one inflated beta distribution about brms

[paul-buerkner]

You can now try out the zero_inflated_beta family. :-)
I will think about how to implement zero-one-inflated families, but this requires some more work on formula syntax and related features, which takes some time.

[paul-buerkner]

Currently, it is not possible, but I could manage to implement it relatively easy (just copy things from the other zero_inflated families). When you say you have some ones, you also need a "one_inflated" family, if you want to use beta, because beta is only defined on ]0,1[.
Of course this is also possible to implement but requires more time.

[alireza202]

I can definitely ignore the ones right now. There are not that many, and are of less importance. But zeros can't be neglected. It would be great to implement the zero-inflated beta distribution. I can do it myself, if it's relatively easy.

[paul-buerkner]

I will try to do it tomorrow, it shouldn't take too long.

[alireza202]

Awesome!

[alireza202]

This is great! I think this might be enough for me.

[paul-buerkner]

Ok, I am going to close this for now. A zero_one_inflated model is currently a bit far away from where I want brms to go. We would have to define 3 processes, one for the zero, one for the one, and the basic process in between. Thus, we would have to define 1 more internal variable, since main and spec are already in, referring to the main process and to the zero process (when using zero-inflated or hurdle models). I currently do not see another application of the 3rd variable outside of a zero_one_inflated_beta family. If we ever need this family or something similar, we can just reopen this issue or make a new one.

[liamkendall]

Hi Paul,

I'm currently unable to fit a zero_one_inflated_beta model, with the attached data, despite the response being strictly between 0 and 1 (and including both). Any ideas why it is not working? Thanks!

data_example.txt

`
data_example <- read.csv("data_example.csv")

example_zob_model <- brm(d ~ PC1 + (1|Id),
data = data_example,
family = "zero_one_inflated_beta")`

Error: Family 'zero_one_inflated_beta' requires response smaller than or equal to 1.

[paul-buerkner]

The example just works for me. Consider updating your brms. Perhaps that could help. If that doesn't help, I assume there is some numerical issue so consider using something like d <- ifelse(d > 1, 1, d)