Geometries for diverging bar plots"
In ggstats: Extension to 'ggplot2' for Plotting Stats

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(ggstats)
library(dplyr)
library(ggplot2)
library(patchwork)

Note : if you are looking for an all-in-one function to display Likert-type items, please refer to gglikert() and vignette("gglikert").

New positions

Diverging bar plots could be achieved using position_diverging() or position_likert().

position_diverging() stacks bars on top of each other and centers them around zero (the same number of categories are displayed on each side).

base <-
  ggplot(diamonds) +
  aes(y = clarity, fill = cut) +
  theme(legend.position = "none")

p_stack <-
  base +
  geom_bar(position = "stack") +
  ggtitle("position_stack()")

p_diverging <-
  base +
  geom_bar(position = "diverging") +
  ggtitle("position_diverging()")

p_stack + p_diverging

position_likert() is similar but uses proportions instead of counts.

p_fill <-
  base +
  geom_bar(position = "fill") +
  ggtitle("position_fill()")

p_likert <-
  base +
  geom_bar(position = "likert") +
  ggtitle("position_likert()")

p_fill + p_likert

By default, the same number of categories is displayed on each side, i.e. if you have 4 categories, 2 will be displayed negatively and 2 positively. If you have an odd number of categories, half of the central category will be displayed negatively and half positively.

The center could be changed with the cutoff argument, representing the number of categories to be displayed negatively: 2 to display negatively the two first categories, 2.5 to display negatively the two first categories and half of the third, 2.2 to display negatively the two first categories and a fifth of the third.

p_1 <-
  base +
  geom_bar(position = position_diverging(cutoff = 1)) +
  ggtitle("cutoff = 1")

p_2 <-
  base +
  geom_bar(position = position_diverging(cutoff = 2)) +
  ggtitle("cutoff = 2")

p_null <-
  base +
  geom_bar(position = position_diverging(cutoff = NULL)) +
  ggtitle("cutoff = NULL")

p_3.75 <-
  base +
  geom_bar(position = position_diverging(cutoff = 3.75)) +
  ggtitle("cutoff = 3.75")

p_5 <-
  base +
  geom_bar(position = position_diverging(cutoff = 5)) +
  ggtitle("cutoff = 5")

wrap_plots(p_1, p_2, p_null, p_3.75, p_5)

New scales

For a diverging bar plot, it is recommended to use a diverging palette, as provided in the Brewer palettes. Sometimes, the number of available colors is insufficient in the palette. In that case, you could use pal_extender() or scale_fill_extended(). However, if you use a custom cutoff, it is also important to change the center of the palette as well.

Therefore, for diverging bar plots, we recommend to use scale_fill_likert().

wrap_plots(
  p_1 + scale_fill_likert(cutoff = 1),
  p_null + scale_fill_likert(),
  p_3.75 + scale_fill_likert(cutoff = 3.75)
)

Improving axes

You may also want have centered axes. That could be easily achieved with symmetric_limits().

You could also use label_number_abs() or label_percent_abs() to display absolute numbers.

wrap_plots(
  p_3.75,
  p_3.75 +
    scale_x_continuous(
      limits = symmetric_limits,
      labels = label_number_abs()
    )
)

New geometries

To facilitate the creation of diverging bar plots, you could use variants of geom_bar() and geom_text().

geom_diverging() & geom_diverging_text()

Let's consider the following plot:

ggplot(diamonds) +
  aes(y = clarity, fill = cut) +
  geom_bar(position = "diverging") +
  geom_text(
    aes(
      label =
        label_number_abs(hide_below = 800)
        (after_stat(count))
    ),
    stat = "count",
    position = position_diverging(.5)
  ) +
  scale_fill_likert() +
  scale_x_continuous(
    labels = label_number_abs(),
    limits = symmetric_limits
  )

The same could be achieved quicker with geom_diverging() and geom_diverging_text().

ggplot(diamonds) +
  aes(y = clarity, fill = cut) +
  geom_diverging() +
  geom_diverging_text(
    aes(
      label =
        label_number_abs(hide_below = 800)
        (after_stat(count))
    )
  ) +
  scale_fill_likert() +
  scale_x_continuous(
    labels = label_number_abs(),
    limits = symmetric_limits
  )

geom_likert() & geom_likert_text()

geom_likert() and geom_likert_text() works similarly. geom_likert_text() takes advantages of stat_prop() for computing the proportions to be displayed (see vignette("stat_prop")).

ggplot(diamonds) +
  aes(y = clarity, fill = cut) +
  geom_likert() +
  geom_likert_text() +
  scale_fill_likert() +
  scale_x_continuous(
    labels = label_percent_abs()
  )

geom_pyramid() & geom_pyramid_text()

Finally, geom_pyramid() and geom_pyramid_text() are variations adapted to display an age-sex pyramid. It uses proportions of the total.

d <- Titanic |> as.data.frame()
ggplot(d) +
  aes(y = Class, fill = Sex, weight = Freq) +
  geom_pyramid() +
  geom_pyramid_text() +
  scale_x_continuous(
    labels = label_percent_abs(),
    limits = symmetric_limits
  )