Announcing tidybayes and ggdist 2.1

Tidybayes is now tidybayes + ggdist

tidybayes began as a package focused on munging posteriors from Bayesian models into a format suitable for use with ggplot2. Over time, it grew an additional tree of functionality: a pantheon of geoms and stats designed for creating uncertainty visualizations.

The inclusion of this uncertainty visualization functionality in a package ostensibly focused on munging Bayesian posteriors has become increasingly awkward, especially with the addition of the stat_dist_ family for visualizing analytical distributions and the frequentist uncertainty visualization vignette it enables. So I decided to spin off the geoms and stats from tidybayes into a new package, ggdist, with fewer dependencies and without the word “Bayes” in its name. I hope this brings the slab+interval and lineribbon families of geoms to a larger audience, increasing the spread of modern uncertainty visualization techniques:

For existing tidybayes users, this change should not affect your workflow: you can keep using all the familiar tidybayes geoms and stats directly from tidybayes itself. Currently (and for the foreseeable future), tidybayes will re-export all geoms and stats from ggdist. This is because, besides the fact that I don’t want to break existing code, these functions form a core part of the tidybayes workflow. So you can continue as before and ignore the existence of ggdist if you like! Meanwhile, I hope new folks will find ggdist useful with its slimmer dependencies and a focus just on visualization.

Automatic orientation detection for all geoms and stats

All geoms and stats in tidybayes (well, ggdist) now automatically detect their orientation based on aesthetic mappings, making the h-suffixed geoms obsolete. This change is enabled by the new orientation-detection code in ggplot2 itself, written by the inimitable Thomas Lin Pedersen. You can see automatic orientation detection in action below:

library(ggplot2)
library(ggdist)
library(patchwork)

set.seed(1234)
df = data.frame(var = c("a", "b"), value = rnorm(2000, c(1,3)))
  
horizontal_plot = ggplot(df) +
  stat_halfeye(aes(x = var, y = value)) +
  ggtitle("stat_halfeye(aes(x = var, y = value))") +
  theme_ggdist()

vertical_plot = ggplot(df) +
  stat_halfeye(aes(x = value, y = var)) +
  ggtitle("stat_halfeye(aes(x = value, y = var))") +
  theme_ggdist()

horizontal_plot + vertical_plot

The main implication for existing users is that if you are using horizontally-oriented geoms (like the ever-popular stat_halfeyeh()), you can (in 99% of cases) simply delete that ugly trailing h from the function name and the same plot should be output. If you don’t, it should still work, but you will get a deprecation warning.

More notes on orientation detection:

• All h-suffix geoms are now deprecated. The h-suffix geoms have been left in tidybayes and give a deprecation warning when used; they cannot be used from ggdist directly.
• If the orientation detection fails, you can always specify the correct orientation manually using orientation = "horizontal" or orientation = "vertical". The alternate spellings (used by base ggplot2) orientation = "y" and orientation = "x" also work; I prefer the horizontal/vertical nomenclature because (1) x and y are axes, not orientations, (2) I find it easier to remember which is which with horizontal/vertical, and (3) I already implemented the orientation parameter using those names in a previous release (before automatic orientation detection was a thing :) ).
• The h-suffix point_interval() functions are also deprecated, since they are no longer needed in tidybayes nor in ggplot2::stat_summary().
• geom_interval(), geom_pointinterval(), and geom_lineribbon() no longer automatically set the ymin and ymax aesthetics if .lower or .upper are present in the data. This allows them to work better with automatic orientation detection (and was a bad feature to have existed in the first place anyway). The deprecated tidybayes::geom_intervalh() and tidybayes::geom_pointintervalh() still automatically set those aesthetics, since they are deprecated anyway (so supporting the old behavior is fine in these functions).

Stepped lineribbons

Solomon Kurz requested stepped lineribbons, which are useful for visualizing survival curves, amongst other things. You can now use the step argument of lineribbon geoms and stats (points shown for reference) to construct stepped lineribbons:

library(ggplot2)
library(ggdist)
library(patchwork)

stepped_plot = function(step, step_text = deparse(step)) {
  ggplot(data.frame(x = 1:5), aes(x = x)) +
    stat_dist_lineribbon(aes(dist = "norm", arg1 = x), step = step) +
    geom_point(aes(y = x), size = 3) +
    ggtitle(paste0("step = ", step_text)) +
    theme_ggdist() +
    guides(fill = FALSE)
}

stepped_plot(step = FALSE, "FALSE (the default)") +
  stepped_plot(step = TRUE, 'TRUE (or "mid")') +
  stepped_plot(step = "hv") +
  stepped_plot(step = "vh")

## Warning: `guides(<scale> = FALSE)` is deprecated. Please use `guides(<scale> =
## "none")` instead.

## Warning: `guides(<scale> = FALSE)` is deprecated. Please use `guides(<scale> =
## "none")` instead.

## Warning: `guides(<scale> = FALSE)` is deprecated. Please use `guides(<scale> =
## "none")` instead.

## Warning: `guides(<scale> = FALSE)` is deprecated. Please use `guides(<scale> =
## "none")` instead.

Minor changes

• ggdist now has its own implementation of the scaled and shifted Student’s t distribution (dstudent_t(), qstudent_t(), etc), since it is very useful for visualizing confidence distributions. These functions are re-exported in tidybayes as well.
• All deprecated functions and geoms now throw deprecation warnings (previously, several deprecated functions did not).