1.7 Non-Standard Geometries

Learning objectives

By the end of this chapter, you can:

match a question to an appropriate geometry instead of defaulting to geom_point()
construct heatmaps of categories and numeric values with geom_tile()
visualize changes across distributions with ggridges::geom_density_ridges()
differentiate interval bands with geom_ribbon() from stacked composition with geom_area()
build slope graphs and dumbbell plots by hand without adding dependencies
amplify any geometry into small multiples with facet_grid()

Prerequisite check (≤5 minutes)

Answer these questions independently; otherwise review the tidyverse prerequisites:

Check In: Prerequisites

Draw ggplot(penguins, aes(bill_length_mm, species)) + geom_boxplot() and explain the x and y mappings.
Explain the difference between fill and color: one typically controls interiors, the other outlines.

1. Beyond points, lines, and bars: let the question choose the geometry

Just Our Opinion

Scatterplots, bars, and lines are general-purpose answers, which can make them inefficient for specific questions. Three boxplots for distribution shifts or grouped bars for endpoint differences are not necessarily wrong, but they can waste the reader’s time. Identify the kind of question before choosing the geometry. That is the central skill of this chapter.

Question type	Typical question	Preferred geometry	Avoid as a default
Intensity on a grid	Which combinations are high or low?	`geom_tile()` heatmap	Thirty bars
Multiple distributions	How does the whole distribution shift?	`geom_density_ridges()`	A row of boxplots
Intervals and composition	How wide is the band? How do shares change?	`geom_ribbon()` / `geom_area()`	Forced error-bar layouts or sequences of pies
Endpoint change	Who changed by how much from A to B?	Hand-built slope or dumbbell plot	Grouped bars
Any of these, by another group	Does the pattern hold within groups?	`facet_grid()`	Everything squeezed into one panel

One criterion matters: the graphic should answer the reader’s question through its structure, without requiring mental arithmetic.

2. `geom_tile()`: heatmaps

Two categorical axes and a continuous fill value turn table lookup into visual comparison of colored cells.

library(ggplot2)
library(dplyr)

gapminder::gapminder |>
  filter(year %in% c(1952, 1972, 1992, 2007)) |>
  group_by(continent, year) |>
  summarise(mean_life = mean(lifeExp), .groups = "drop") |>
  ggplot(aes(x = factor(year), y = continent, fill = mean_life)) +
  geom_tile(color = "white", linewidth = 1) +   # White gaps separate cells
  scale_fill_viridis_c() +                      # Perceptually uniform scale for continuous values
  labs(x = NULL, y = NULL, fill = "Life expectancy",
       title = "Life expectancy by continent over fifty years")

Note

geom_tile() and geom_raster() serve similar purposes; geom_raster() renders regular, evenly spaced grids faster. Category-by-category counts, such as species by island, work too: map fill to n.

3. `ggridges::geom_density_ridges()`: ridgelines

To compare one variable across groups or periods, ridgelines stack distributions with vertical offsets. Shifts become visible along with shape information that boxplots cannot show.

# install.packages("ggridges")   # The only additional package needed for this chapter
library(ggridges)

gapminder::gapminder |>
  filter(year %in% seq(1952, 2007, by = 5)) |>
  ggplot(aes(x = lifeExp, y = factor(year), fill = year)) +
  geom_density_ridges(scale = 1.2, show.legend = FALSE) +
  scale_fill_viridis_c(option = "mako", begin = 0.2, end = 0.9) +
  labs(x = "Life expectancy", y = NULL,
       title = "Global life expectancy: shifting right, with a narrowing left tail")

Two useful arguments: scale controls overlap (1 means neighboring ridges just touch; >1 allows overlap), and rel_min_height, such as 0.01, trims low tails to reduce clutter.

Common mistake: relying on alphabetical group order

For categorical y values, default ordering can place groups alphabetically and disrupt time order. Set the order with factor(year, levels = ...) before plotting, keeping the earliest period at one end.

4. `geom_ribbon()` and `geom_area()`: bands and stacks

Both draw filled regions but answer different questions: ribbons show intervals or ranges; areas show changing composition.

# ribbon: global median life expectancy and 10%-90% quantile band
gapminder::gapminder |>
  group_by(year) |>
  summarise(
    med = median(lifeExp),
    lo  = quantile(lifeExp, 0.10),
    hi  = quantile(lifeExp, 0.90)
  ) |>
  ggplot(aes(year, med)) +
  geom_ribbon(aes(ymin = lo, ymax = hi), alpha = 0.25, fill = "steelblue") +
  geom_line(linewidth = 1)

# area: changing population composition by continent
gapminder::gapminder |>
  filter(continent != "Oceania") |>
  count(year, continent, wt = pop, name = "pop") |>
  ggplot(aes(year, pop / 1e9, fill = continent)) +
  geom_area(alpha = 0.85) +
  labs(y = "Population (billions)", fill = NULL)

Note

Stacked area charts suit situations where the total matters and composition is the focus. If comparing continents’ growth patterns, consider normalized shares with position = "fill" or separate facets instead.

5. Build slope graphs and dumbbell plots by hand

Both compare two time points or states. Packages such as ggalt and ggdist offer convenience functions, but building the plot with core ggplot2 gives complete control and avoids another dependency.

# Slope graph: European life expectancy, 1952 to 2007; highlight the largest gain
top_gain <- gapminder::gapminder |>
  filter(continent == "Europe", year %in% c(1952, 2007)) |>
  select(country, year, lifeExp) |>
  tidyr::pivot_wider(names_from = year, values_from = lifeExp,
                     names_prefix = "y") |>
  mutate(gain = y2007 - y1952)

champion <- top_gain |> slice_max(gain, n = 1) |> pull(country)

gapminder::gapminder |>
  filter(country %in% top_gain$country, year %in% c(1952, 2007)) |>
  ggplot(aes(x = factor(year), y = lifeExp, group = country)) +
  geom_line(color = "grey75", linewidth = 0.6) +    # Background: all countries
  geom_point(color = "grey75") +
  geom_line(                                         # Highlight the country with the largest gain
    data = \(d) filter(d, country == champion),
    color = "#D55E00", linewidth = 1.2
  ) +
  labs(x = NULL, y = "Life expectancy",
       title = paste(champion, ": Europe's largest life-expectancy gain over half a century"))

# Dumbbell: mean female/male body mass by species; distance shows the difference
palmerpenguins::penguins |>
  tidyr::drop_na(sex) |>
  group_by(species, sex) |>
  summarise(mass = mean(body_mass_g), .groups = "drop") |>
  tidyr::pivot_wider(names_from = sex, values_from = mass) |>
  ggplot(aes(y = species)) +
  geom_segment(aes(x = female, xend = male, yend = species),
               color = "grey70", linewidth = 2) +
  geom_point(aes(x = female), color = "#0072B2", size = 3.2) +
  geom_point(aes(x = male),    color = "#E69F00", size = 3.2) +
  labs(x = "Mean body mass (g)", y = NULL,
       title = "Males are heavier in every species, but the gaps differ")

Check In: Three minutes of practice

Change the dumbbell plot to compare male and female bill_length_mm. Which species has the smallest relative difference? Then inspect bill_depth_mm: is the conclusion the same? If the conclusions differ, which plot would you show readers, and why?

Common mistake: using a continuous palette for dumbbell endpoints

The endpoints represent states, such as female/male or before/after. Use two distinct qualitative colors, such as Okabe–Ito blue and orange (#0072B2 / #E69F00), rather than scale_colour_gradient(). A gradient implies quantity and encourages the wrong interpretation.

6. `facet_grid()`: extend comparisons with small multiples

Faceting adds another comparison dimension to any geometry. Compared with facet_wrap(), facet_grid() assigns explicit meanings to rows and columns. It suits comparisons with few levels in both categorical variables, such as ≤4 × ≤3. Facets share scales by default, supporting comparison; avoid scales = "free" without a reason such as different measurement units.

palmerpenguins::penguins |>
  filter(!is.na(sex)) |>
  ggplot(aes(bill_length_mm, bill_depth_mm)) +
  geom_point(aes(color = species), alpha = 0.7, size = 1.6) +
  facet_grid(rows = vars(species), cols = vars(sex)) +
  scale_colour_viridis_d(end = 0.85) +
  labs(x = "Bill length (mm)", y = "Bill depth (mm)")

Practice Exercise 1 (copy)

Follow §2 to draw a species × island sample-size heatmap using penguins |> count(species, island). Use viridis fill, white gaps between cells, and a title framed as a question the reader can answer directly.

Practice Exercise 2 (adapt)

Adapt §3 to compare continents in 2007 with x = lifeExp and y = continent. ① Use forcats::fct_reorder() to set y-axis order by descending median lifeExp; ② try two viridis option values, select the more readable one, and explain why.

Practice Exercise 3 (create · AI integration)

Round 1 (AI off): use gapminder to answer “Which countries climbed the most in GDP-per-capita rank from 1997 to 2007?” Choose a slope graph, dumbbell plot, or a better alternative yourself. Build it by hand without installing another package. Round 2 (AI allowed): show the code to Posit Assistant and ask only: “Does the geometry fit the question? Could a design with less ink be easier to interpret?” Record its alternative, render it beside your version, and explain the tradeoff.

Capstone

Task: three questions, three plots. Using penguins or gapminder, formulate questions of three different types: distribution shift, changing composition, and endpoint change. Choose a suitable geometry for each, with facets in at least one. Deliver a one-page Quarto report with three plots, two lines defending each geometry choice, and a question-to-geometry decision table.

Dimension	Meets expectations	Good	Excellent
Questions	Three different types	Specific and answerable	Form a developing narrative
Geometry fit	All three fit	Each has a justification	Justification considers the reader’s interpretation effort
Implementation	Code runs and plots are readable	Hand-built slope/dumbbell without unnecessary dependencies	Facets, colors, and ordering support comparison
Reflection	Includes decision table	Offers an alternative for one plot	Experiments with the tradeoff between alternatives

SOURCES · Source mapping

Section	Material	Use
§1: question-first teaching structure; §6: faceting	posit::conf(2025) ggplot2 workshop sessions/slides (Thomas Lin Pedersen, Teun van den Brand · README states CC-BY 4.0; LICENSE.md states CC-BY-SA 4.0)	Adapted
`geom_density_ridges()` arguments and `geom_raster()` comparison	Official ggridges and ggplot2 documentation	Referenced
Heatmap/ribbon/area/slope/dumbbell examples, geometry decision table, and rubric	This project	Original

This chapter is published under CC-BY-SA 4.0.

1.7 Non-Standard Geometries

Learning objectives

Prerequisite check (≤5 minutes)

1. Beyond points, lines, and bars: let the question choose the geometry

2. geom_tile(): heatmaps

3. ggridges::geom_density_ridges(): ridgelines

4. geom_ribbon() and geom_area(): bands and stacks

5. Build slope graphs and dumbbell plots by hand

6. facet_grid(): extend comparisons with small multiples

Capstone

SOURCES · Source mapping

2. `geom_tile()`: heatmaps

3. `ggridges::geom_density_ridges()`: ridgelines

4. `geom_ribbon()` and `geom_area()`: bands and stacks

6. `facet_grid()`: extend comparisons with small multiples