beta regression is not supported yet

[yuryzablotski]

Hi guys, beta regression can't be processed with tbl_regression(). Would be nice if it worked:

d <- mtcars %>%
mutate(mpg = mpg/100,
cyl = factor(cyl))

library(betareg)
m <- betareg(mpg ~ cyl, data = d)

plot_model(m, type = "pred")

library(emmeans)
emmeans(m, pairwise ~ cyl, type = "response")

library(gtsummary)
tbl_regression(m, add_pairwise_contrasts = T)

The error message on my computer is:

Error in dplyr::left_join():
! Join columns in x must be present in the
data.
✖ Problem with variable.
Backtrace:

1. gtsummary::tbl_regression(m, add_pairwise_contrasts = T)
2. dplyr:::left_join.data.frame(...)

Thanks upfront! Love gtsummary!!!

[ddsjoberg]

Things that work

broom::tidy()
model.matrix()
model.frame()
terms() I didn't inspect the internals of the object though

Things that need to be addressed

• The (phi) is being incorrectly categorized as the intercept
  
• The error is thrown when tidy_add_reference_rows() is run.

[larmarange]

I explored quickly beta regression.

By the way, a nice introduction: https://www.andrewheiss.com/blog/2021/11/08/beta-regression-guide/

Beta regression have something similar with zero-inflated Poisson models, there are multi-components models, as you model both the mean and phi, each could have their own coefficients. Here, (phi) is properly interpreted as an intercept, but this is the intercept of another component.

I will therefore explore it and include betareg support in #233. Next week I'll be on leave but will explore it later after the 12 of August.

A related issue for gtsummary is ddsjoberg/gtsummary#1540

Adding pairwise contrasts for such model is another question. Let's start with basic support and then see if pairwise contrast could be added or not.