3 layer donut chart in R

Question

I am trying to recreate this image in R, however I am unable to work out how to have 3 layers to a donut chart - everything I find (for instance, webr::PieDonut) only allows 2. Using ggplot I am also unable to re-create it.

A MRE is:

library(ggplot2)
library(webr)
library(dplyr)

lexicon <- data.frame("Level1" = c(rep("Flavour", 11), rep("Appearance", 4)),
                  "Level2" = c(rep("Misc", 6), rep("Pungent", 5), rep("Colour", 4)),
                  "Level3" = c("Fresh", "Refreshing", "Soapy", "Minty", "Nutty", "Milky", "Peppery", "Sharp", "Horseradish", "Mustard hot", "Spicy", "Colourful"," Fresh Green", "Dark Green", "Bright Green")
)

PieDonut(lexicon, aes(Level1, Level2), title = "Salad Lexicon", showRatioDonut =FALSE, showRatioPie = FALSE)

ggplot(lexicon, aes(Level2, Level3, fill = Level1)) +
  geom_col() +
  scale_fill_viridis_d() +
  coord_polar("y")

While the PieDonut works for 2 levels (not shown), it doesn't allow the final level to be included. The ggplot approach also does not work, as seen in the figure below.

How can I get this style of chart in R? Either with ggplot or base plotting.

Putative

Answer 1

I think a nice alternative is to use geom_rect here after some data manipulation. Using the fill, color, and alpha scales can help improve the differentiation of categories. I would also use geom_textpath here, though I might go for circumferential labels if there is room to do so:

lexicon %>%
  mutate(top_level = Level1) %>%
  pivot_longer(1:3) %>%
  group_by(name, value) %>%
  mutate(width = n()) %>%
  unique() %>%
  arrange(name) %>%
  group_by(name) %>%
  mutate(ymid = as.numeric(sub("\\D+", "", name)),
         ymax = ymid + 0.5, ymin = ymid - 0.5,
         xmin = c(0, head(cumsum(width), -1)),
         xmax = cumsum(width),
         xmid = (xmax + xmin) / 2) %>%
  ggplot(aes(xmid, ymid, fill = top_level)) +
  geom_rect(aes(xmin = xmin, xmax = xmax, ymin = ymin, ymax = ymax,
                alpha = name, color = top_level)) +
  geomtextpath::geom_textpath(aes(y = ymid + 0.25, label = value, 
                                  group = value)) +
  scale_alpha_manual(values = c(1, 0.3, 0.1)) +
  scale_fill_manual(values = c("#cd9900", "#00817e")) +
  scale_colour_manual(values = c("#cd9900", "#00817e")) +
  scale_y_continuous(limits = c(-0.5, 3.6)) +
  coord_polar() +
  theme_void() +
  theme(legend.position = "none")

Teredo

Really nice Allan. Tried with circumferential labels too but failed with geom_col. Will keep the trick with geom_rect in mind. – Agnusago

Thank you for this. This looks really good. Would you mind telling me how to have the text radial rather than circumferential? I have tried a few things but they don't seem to work, and I am unfamiliar with ggplot, let alone these functions. – Putative

@Putative use angle = 90 inside geom_textpath – Teredo

@AllanCameron. Upon further inspection, when using duplicated values (for instance 'Earthy' is both a flavour category and aftertaste-> taste in the original plot. This seems to break this approach - do you know how to avoid this (without changing the values)? I have tried a few approaches, such as trying to use an UID for the combination but that doesn't seem to work – Putative

The graph looks impressing, BUT I get the following error: Error in namespaceExport(ns, exports) : undefined exports: GeomLabelabline, GeomLabelcontour, GeomLabelcurve, GeomLabeldensity... What kind of library could be throwing this? – Corinnecorinth

@Corinnecorinth try library(geomtextpath) and re-run your code – Teredo

Answer 2

One option would be to reeshape your data to long and do some manual aggregating before passing to ggplot. Additionally I use geomtextpath::geom_textpath to add the labels:

library(ggplot2)
library(dplyr)
library(geomtextpath)

lexicon <- data.frame("Level1" = c(rep("Flavour", 11), rep("Appearance", 4)),
                      "Level2" = c(rep("Misc", 6), rep("Pungent", 5), rep("Colour", 4)),
                      "Level3" = c("Fresh", "Refreshing", "Soapy", "Minty", "Nutty", "Milky", "Peppery", "Sharp", "Horseradish", "Mustard hot", "Spicy", "Colourful"," Fresh Green", "Dark Green", "Bright Green")
)

lexicon_long <- lexicon |>
  mutate(fill = Level1) |>
  tidyr::pivot_longer(-fill, names_to = "level", values_to = "label") |>
  mutate(label = forcats::fct_inorder(label)) |> 
  count(fill, level, label) |>
  group_by(level) |>
  mutate(pct = n / sum(n))

ggplot(lexicon_long, aes(level, pct, fill = fill)) +
  geom_col(color = "white") +
  geom_textpath(aes(label = label, group = label),
                position = position_stack(vjust = .5),
                upright = TRUE, hjust = .5, size = 3
  ) +
  scale_fill_viridis_d() +
  coord_polar("y") +
  theme_void() +
  guides(fill = "none")

Agnusago

Nice to see a geomtextpath solution :) – Teredo

This seems to have broken the structure of the data. In each ring the values are all alphabetical, rather than relating to the relevant layers. – Putative

My bad. If you want a specific order you have to use a factor. I just have made an edit and use fct_inorder after reshaping so that the order of your categories is or should be preserved. – Agnusago