Waffle charts for visualizing proportions

library(tidyverse)
library(waffle)
library(viridis)

# fix seed value for reproducibility
set.seed(123)

theme_set(theme_void())

Waffle charts

{waffle} provides a {ggplot2} implementation of waffle plots. The typical workflow consists of preparing the data by tabulating in advance and then plotting it with {ggplot2} and geom_waffle().

ImportantRead this before you start

You will estimate the same three proportions twice – once from a pie chart, once from a waffle chart – before computing the true values. Do not compute the answer early, and do not revise an earlier estimate after seeing a later chart. Wrong guesses are the point; they are your data.

Round 1: estimate from a pie chart

Demonstration: Run the chunk below to draw a pie chart of penguins by species.

penguins |>
  count(species) |>
  ggplot(mapping = aes(x = "", y = n, fill = species)) +
  geom_col(color = "white") +
  coord_radial(
    theta = "y",
    reverse = "theta",
    expand = FALSE
  ) +
  scale_fill_viridis_d(end = 0.8) +
  labs(
    title = "Penguins by species",
    x = NULL,
    y = NULL,
    fill = NULL
  ) +
  theme_void() +
  theme(legend.position = "top")

Your turn: Looking only at the pie chart, estimate what percentage of the penguins belong to each species. Do not count anything and do not write any code – just read the chart. Your three estimates should sum to about 100%.

Species	Estimate from the pie chart
Adelie	%
Chinstrap	%
Gentoo	%

Basic waffle chart

Demonstration: Prepare the penguins data frame to visualize the number of penguins by species.

# add code here

Demonstration: Use the prepared data to draw a basic color-coded waffle chart

# add code here

Improve the waffle chart

Your turn: Adjust the waffle chart to use a fixed aspect ratio so the symbols are squares. Rotate the chart so the squares are stacked vertically.

# add code here

Demonstration: {waffle} will draw all observations on the chart. For larger datasets, this is problematic. Instead, we might want to visualize the proportion of observations in each category. Use geom_waffle() to represent the data as proportions instead.

# add code here

Your turn: Adjust the waffle chart to use a better color palette and move the legend to the top.

# add code here

Round 2: estimate from your waffle chart

Your turn: Now read the same three proportions off the waffle chart you just built. Again, do not count the squares one by one and do not write any code – read the chart the way a reader would.

Species	Estimate from the waffle chart
Adelie	%
Chinstrap	%
Gentoo	%

Round 3: compute the true proportions

Your turn: Now compute the actual percentage of penguins in each species. Use count() to tabulate and mutate() to convert the counts into percentages.

# add code here
penguins |>
  count(species) |>
  mutate(prop = n / sum(n))

    species   n      prop
1    Adelie 152 0.4418605
2 Chinstrap  68 0.1976744
3    Gentoo 124 0.3604651

Compare

Your turn: Fill in the table using your two sets of estimates and the values you just computed. Error is your estimate minus the truth, so a positive number means you overestimated.

Species	Pie estimate	Waffle estimate	True %	Pie error	Waffle error
Adelie
Chinstrap
Gentoo

Your turn: Which chart produced smaller errors for you? Was the gap the same for all three species, or larger for some than others?

Add response here.

Your turn: In lecture we ranked perceptual tasks by how accurately people read them: position on a common scale is most accurate, then length, then angle, then area. A pie chart asks you to judge angle; a waffle chart asks you to judge counts of discrete squares. Do your errors line up with that ranking? If they do not, what else about these charts might explain it?

Add response here.

Your turn: Compare your results with the person next to you. Where your errors disagree, what about how each of you read the chart might account for the difference?

Add response here.

One more comparison: donut charts

Demonstration: A donut chart is a pie chart with the center removed.

penguins |>
  count(species) |>
  ggplot(mapping = aes(x = 2, y = n, fill = species)) +
  geom_col(color = "white") +
  coord_radial(theta = "y", reverse = "theta", expand = FALSE) +
  xlim(0.5, 2.5) +
  scale_fill_viridis_d(end = 0.8) +
  labs(
    title = "Penguins by species",
    x = NULL,
    y = NULL,
    fill = NULL
  ) +
  theme_void() +
  theme(legend.position = "top")

Your turn: Removing the center removes the vertex of every angle. Based on what you measured above, would you expect estimates from the donut chart to be better or worse than from the pie chart? Explain your reasoning.

Add response here.