1.8 Professionally Styled Plots

Learning objectives

By the end of this chapter, you can:

decompose the four theme areas—text, axis, panel, and legend—and their element_*() controls
replace legends with direct endpoint annotations to reduce interpretation effort
apply accessible RColorBrewer/viridis palettes and drive scale_*_manual() with named vectors
compose labs() with a conclusion in the title, context in the subtitle, and sources in the caption
integrate _brand.yml with Quarto typography and ggplot2 output

Prerequisite check (≤5 minutes)

Answer these questions independently; otherwise review the tidyverse prerequisites:

Check In: Prerequisites

After drawing ggplot(penguins, aes(bill_length_mm, body_mass_g)) + geom_point(aes(color = species)), move the legend to the top with one line of code.
Describe one visual difference between theme_bw() and theme_minimal().

1. Anatomy of `theme()`: four areas and `element_*()`

Just Our Opinion

Many people style plots by copying ten mysterious lines of theme(). The core distinction is that themes control non-data ink: grids, borders, and fonts. To change points and lines, use scales or geom arguments. A theme call in the wrong place accomplishes nothing.

Theme names use dot notation, area.component, with the appropriate element_*() function:

Area	Typical arguments	Element function
text / axis	`plot.title` / `axis.title` / `axis.line` / `axis.text`	`element_text()` / `element_line()`
panel	`panel.background` / `panel.grid.major`	`element_rect()` / `element_line()`
legend	`legend.position` / `legend.title`	A string for position; `element_*()` for styling

library(ggplot2)
library(dplyr)

p <- ggplot(palmerpenguins::penguins,
            aes(bill_length_mm, body_mass_g, color = species)) +
  geom_point(alpha = 0.75, size = 2)

p + theme_minimal(base_size = 12) +
  theme(
    panel.grid.minor = element_blank(),      # Remove minor-grid clutter
    panel.grid.major.x = element_blank(),    # Keep horizontal grid lines for comparison
    legend.position = "top"
  )

Workshop developments: ggplot2 4.0+

The posit::conf(2025) theme session introduced newer APIs: theme_gray(ink/paper/accent) for global colors, theme(geom = element_geom(...)) for default layer appearance, and theme(palette.colour.continuous = ...) for a default palette in the theme. Try them with ggplot2 ≥ 4.0; the main examples use established APIs.

2. Replace legends with direct labels

A legend makes readers repeatedly look between the plot and a lookup key. Put labels beside the data; for lines, the endpoint is often the natural location.

countries <- c("China", "India", "United States", "South Africa")
g4 <- gapminder::gapminder |> filter(country %in% countries)   # Subset once
ends <- filter(g4, year == 2007)              # Endpoints and label data

ggplot(g4, aes(year, lifeExp, color = country)) +
  geom_line(linewidth = 1) +
  geom_text(data = ends, aes(label = country), hjust = 0, nudge_x = 1.5,
            fontface = "bold") +
  scale_x_continuous(expand = expansion(mult = c(0.05, 0.25))) +  # Add space on the right
  coord_cartesian(clip = "off") +             # Allow labels beyond the panel
  guides(color = "none") +                    # Remove the legend
  labs(x = NULL, y = "Life expectancy")

Three supporting steps: add right-side space with expansion, disable clipping with clip = "off", and remove the legend with guides(color = "none"). Leaving out any of them can undermine the result.

Common mistake: keeping both labels and a legend

If you add endpoint labels but forget guides(color = "none"), readers must check whether the duplicated information agrees. When direct labels take over, remove the corresponding legend.

3. Color discipline: accessible palettes and `scale_*_manual()`

About 8% of male readers have color vision deficiency; design your plot with these readers in mind. Two reliable approaches:

# Option 1: qualitative RColorBrewer colors; inspect display.brewer.all(colorblindFriendly = TRUE)
p + scale_colour_brewer(palette = "Dark2")
# Option 2: viridis; trim the lightest yellow for contrast against white
p + scale_colour_viridis_d(end = 0.85)

viridis is a useful default because it is perceptually uniform (equal numeric differences produce roughly equal visual differences), accessible for color vision deficiencies, and ordered in grayscale. Dark2 and Set2 suit qualitative comparisons with ≤8 categories. Beyond eight, consider facets rather than stretching a palette (1.7 §6).

Brand colors may need manual selection, but position-based manual scales are fragile: changing factor levels can silently change color assignments. Use a named vector, with keys matching factor levels, to separate the mapping from order.

species_cols <- c(
  Adelie    = "#0072B2",   # Okabe-Ito blue
  Chinstrap = "#E69F00",   # Orange
  Gentoo    = "#009E73"    # Green
)

p + scale_colour_manual(values = species_cols)

Three rules: the named vector may contain extra keys, but must not omit needed levels, which otherwise receive the NA color; define the palette once and reuse it throughout the document, preparing for §5; named colors remain stable when upstream factor order changes.

Check In: Two minutes of practice

Use colorspace::cvd_image() or an online CVD simulator to inspect a recent plot under deuteranopia. Which colors become hard to distinguish? Replace one with an Okabe–Ito color and simulate again. Then deliberately reorder penguins$species factor levels and verify that the named palette preserves assignments.

4. The `labs()` hierarchy: a title is a conclusion

Readers often scan title → plot → axes and legend. Put the conclusion in title to help with the first step of interpretation; a conclusion buried in caption may go unread.

Position	Purpose	Poor choice
`title`	One-sentence conclusion	“Penguin data plot”
`subtitle`	Scope, period, and definitions	Repeating the title
`caption`	Sources and explanatory notes	Hiding the conclusion
`tag`	Panel identifier such as “A”	Putting the panel identifier in the title

p + labs(
    title    = "Longer-billed penguins are heavier, with similar slopes across species",
    subtitle = "Palmer station, 2007-2009; 333 complete records",
    caption  = "Data: palmerpenguins (Gorman et al., 2014)",
    x = "Bill length (mm)", y = "Body mass (g)", color = NULL   # Remove the legend title too
  )

5. `brand.yml` × Quarto × ggplot2

Quarto ≥ 1.8 reads _brand.yml from the project root to apply colors and fonts to HTML, PDF, and slides. Align ggplot2 with the same brand so figures and typography feel coherent.

# _brand.yml: read automatically by Quarto
color:
  palette:
    brand-blue:  "#2266B1"
    brand-green: "#1B7D3A"
    brand-amber: "#B1781B"
typography:
  base: Inter   # Other brand settings include foreground, background, and headings

# Option A (no dependencies): maintain colors alongside _brand.yml and reuse one definition
brand_cols <- c("#2266B1", "#1B7D3A", "#B1781B")
# Option B (optional): install.packages("thematic") to apply brand colors and fonts to ggplot2
if (requireNamespace("thematic", quietly = TRUE)) {
  thematic::thematic_on(bg = "white", fg = "#1A1A1A", qualitative = brand_cols)
} else {
  message("Install thematic for the optional brand theme; basic plots remain available.")
}

Note

The posit::conf(2025) quarto-brand workshop calls this “bring your own brand.” Quarto reads _brand.yml natively for HTML; thematic provides the graphics-side bridge. Option A has fewer dependencies and explicit behavior. Start there for team work and adopt B when branding requirements become stricter.

6. A plot makeover: Before → After

Before: every choice asks more work of the reader.

p + scale_colour_manual(values = c("red3", "limegreen", "deepskyblue1")) +
  labs(title = "Figure 1") + theme_grey(base_size = 14)

Problems: harsh saturated colors that are not color-vision accessible, an uninformative title, competing gray background and grid, a default right-side legend, and axes without units. After: five changes using this chapter’s tools.

label_ends <- palmerpenguins::penguins |> tidyr::drop_na() |> group_by(species) |>
  summarise(bill_length_mm = max(bill_length_mm) + 3,   # Just beyond the right edge of each species band
            bill_depth_mm = median(bill_depth_mm), .groups = "drop")
ggplot(palmerpenguins::penguins,
       aes(bill_length_mm, bill_depth_mm, color = species)) +
  geom_point(alpha = 0.75, size = 2) +
  geom_text(data = label_ends, aes(label = species), fontface = "bold") +  # Step 1: direct labels
  scale_colour_manual(values = c(Adelie = "#0072B2",    # Step 2: named Okabe-Ito colors
    Chinstrap = "#E69F00", Gentoo = "#009E73")) +
  guides(color = "none") +                              # Step 3: remove the legend
  labs(title = "Three species occupy separate bill-shape bands",        # Step 4: conclusion-led title
       subtitle = "Palmer station, 2007-2009; colors checked for color vision accessibility",
       caption = "Data: palmerpenguins (Gorman et al., 2014)",
       x = "Bill length (mm)", y = "Bill depth (mm)") +
  theme_minimal(base_size = 12) +                       # Step 5: a restrained theme
  theme(panel.grid.minor = element_blank(), plot.title = element_text(face = "bold"))

Makeover order follows the dependencies

Set colors and annotations first (steps 1–3, the information layer), then labs and theme (steps 4–5, presentation). Changing colors later can force another theme adjustment. Wait until the information layer is stable before decorating.

Practice Exercise 1 (copy)

Reproduce the endpoint-labeled plot in §2 for Rwanda, Botswana, Uganda, and Kenya, using gapminder lifeExp from 1962–2007. Some countries experienced declines: can all labels still sit at line ends? Show your approach, using hjust, small adjustments, or an endpoint geom_point() if useful.

Practice Exercise 2 (adapt)

Apply §3 to the dumbbell plot in Chapter 1.7. Use Okabe–Ito blue and orange (#0072B2 / #E69F00) for female and male endpoints. Write two comment lines explaining their suitability for readers with color vision deficiencies. Deliberately change factor-level order and verify that named colors remain stable.

Practice Exercise 3 (create · AI integration)

Round 1 (AI off): choose the least satisfactory plot in one of your old reports, or the result from 1.7 Exercise 3. Apply a complete makeover using only this chapter’s checklist: direct labels, an accessible palette, named color vectors, conclusion-led labs(), and a restrained theme(). Round 2 (AI allowed): show before and after to Posit Assistant and ask only: “Which three remaining style choices reduce readability?” Accept or reject each suggestion and connect each decision to a section of this chapter.

Capstone

Task: create a submission-ready figure. Choose a conclusion from penguins or gapminder and deliver a final figure plus design decisions: endpoint or in-plot labels, an accessible palette defined by a named vector, the full labs() hierarchy, a custom theme(), and a draft _brand.yml with an explanation of its relationship to the plot palette. Deliver a one-page Quarto report: the figure and no more than ten lines of design explanation.

Dimension	Meets expectations	Good	Excellent
Information	Direct annotations present	Positions checked for overlap	Label order guides reading
Color	Accessible palette	Named vector and contrast/range adjustments	CVD simulation checked and documented
Hierarchy	All four labs components	Title is a testable conclusion	Subtitle explains definitions and limitations
Consistency	Plot and document colors do not clash	Maps `_brand.yml` to the palette in a table	Optionally implements thematic and compares results

SOURCES · Source mapping

Section	Material	Use
§1: theme anatomy and ggplot2 4.0 note	posit::conf(2025) ggplot2 workshop `slides/4_theme_guide.qmd` (Thomas Lin Pedersen, Teun van den Brand · README states CC-BY 4.0; LICENSE.md states CC-BY-SA 4.0)	Adapted
§5: brand.yml workflow	posit::conf(2025) quarto-brand workshop (Isabella Velásquez, Sara Altman · CC-BY-SA 4.0)	Adapted
viridis, RColorBrewer, and Okabe–Ito rationale	Official package documentation; Okabe & Ito (2008)	Referenced
Endpoint labels, manual-scale discipline, labs hierarchy, makeover, and rubric	This project	Original

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