(Closer to)
Accessible
Data
Viz

Day 06

Prof Amanda Luby

Carleton College
Stat 220 - Fall 2026

Today

• {patchwork}
• Colorblind-friendly color palettes
• Legible text, plain-language labels, and data tables
• Writing alt-text

Small Multiples

Each plot shares aesthetics but shows different subsets of data

Compound Plots

The plots might share data, but don’t share aesthetics

Compound Plot Example

Imagine reading these three plots one after another in a report. What’s hard about it? How would you fix it?

library(palmerpenguins)
ggplot(penguins) + 
  geom_histogram(bins = 20, col = "white", aes(x = body_mass_g))

ggplot(penguins) + 
  geom_histogram(bins = 20, col = "white", aes(x = flipper_length_mm))

ggplot(penguins) + 
  geom_point(aes(x = body_mass_g, y = flipper_length_mm))

Patchwork

library(patchwork)
p1 <- ggplot(penguins) + 
  geom_histogram(bins = 20, col = "white", aes(x = body_mass_g))

p2 <- ggplot(penguins) + 
  geom_histogram(bins = 20, col = "white", aes(x = flipper_length_mm))

p3 <- ggplot(penguins) + 
  geom_point(aes(x = body_mass_g, y = flipper_length_mm))

(p1 + p2)/p3

Patchwork (layout 2)

p3 + (p1/p2)

For more complex layouts, see the “Controlling Layouts” vignette

Patchwork (layout 2)

What’s redundant in this figure?

p1 <- ggplot(penguins) + 
  geom_histogram(bins = 20, col = "white", aes(x = body_mass_g, fill = species))

p2 <- ggplot(penguins) + 
  geom_histogram(bins = 20,  col = "white", aes(x = flipper_length_mm, fill = species))

p3 <- ggplot(penguins) + 
  geom_point(shape = 21, alpha = .9, col = "white", aes(x = body_mass_g, y = flipper_length_mm, fill = species)) 

p3 + (p1/p2) 

Patchwork (with a shared legend)

p1 <- ggplot(penguins) + 
  geom_histogram(bins = 20, col = "white", aes(x = body_mass_g, fill = species))

p2 <- ggplot(penguins) + 
  geom_histogram(bins = 20,  col = "white", aes(x = flipper_length_mm, fill = species))

p3 <- ggplot(penguins) + 
  geom_point(shape = 21, alpha = .9, col = "white", aes(x = body_mass_g, y = flipper_length_mm, fill = species)) 

p3 + (p1/p2) + 
  plot_layout(guides = 'collect')

Patchwork (hiding point legend)

p1 <- ggplot(penguins) + 
  geom_histogram(bins = 20, col = "white", aes(x = body_mass_g, fill = species))

p2 <- ggplot(penguins) + 
  geom_histogram(bins = 20,  col = "white", aes(x = flipper_length_mm, fill = species))

p3 <- ggplot(penguins) + 
  geom_point(shape = 21, alpha = .9, col = "white", aes(x = body_mass_g, y = flipper_length_mm, fill = species)) +
  theme(legend.position = "none")

p3 + (p1/p2) + 
  plot_layout(guides = 'collect')

Patchwork (with annotation)

p3 + (p1/p2) + 
  plot_layout(guides = 'collect') + 
  plot_annotation(
    title = "Penguin Plot",
    tag_levels = "A"
  ) 

Patchwork (with a common theme)

p3 + (p1/p2) + 
  plot_layout(guides = 'collect') & 
  theme_minimal() &
  scale_fill_viridis_d()

But what does this have to do with accessibility?

• Output (knitted files, slides, websites, etc.) should be designed to make it as easy as possible for the user to understand your content
• There’s a cognitive load involved with scrolling or turning a physical page and trying to remember a visual from the previous page
• When plots belong together, we should put them together

Colorblind-friendly palettes

Can you tell the states apart?

Graph A, as seen by someone with deuteranopia (a common form of colorblindness): which state is which? How sure are you?

Now try graph B

Graph B: same data, same viewer. What changed? Why is it easier?

Which palette would you trust?

Rank these four palettes from best to worst for someone with colorblindness. Then we’ll check.

00:30

colorBlindness::displayAllColors(scales::hue_pal()(10))

colorBlindness::displayAllColors(rainbow(10))

colorBlindness::displayAllColors(colorblindr::palette_OkabeIto)

colorBlindness::displayAllColors(viridisLite::viridis(10))

Color scales

Use colorblind friendly color scales (e.g., Okabe Ito, viridis)

Back to graph A

What if we can’t change the colors (brand guidelines, grayscale printing, …)?

What else could we map to state so readers can tell the states apart without relying on color?

Double encoding

Use shape and color where possible

Default ggplot2 scale

Default ggplot2 scale with deuteranopia

Source: Mine Çetinkaya-Rundel, Sta313. Generated with {colorblindr}.

Without direct labeling

Graph A: Which line is California? Minnesota? New York? Try it with the right panel (deuteranopia).

Default ggplot2 scale

Default ggplot2 scale with deuteranopia

Source: Mine Çetinkaya-Rundel, Sta313. Generated with {colorblindr}.

With direct labeling

Graph B: Try again. What changed, and why is it easier?

Default ggplot2 scale

Default ggplot2 scale with deuteranopia

Source: Mine Çetinkaya-Rundel, Sta313. Generated with {colorblindr}.

Use direct labeling

• Prefer direct labeling where color is used to display information over a legend

• Quicker to read

• Ensures graph can be understood without reliance on color

Without whitespace

Graph A: In the right panel (tritanopia), where does one state’s bar end and the next begin?

Default ggplot2 scale

Default ggplot2 scale with tritanopia

Source: Mine Çetinkaya-Rundel, Sta313. Generated with {colorblindr}.

With whitespace

Graph B: Try again. What did the white lines add?

Default ggplot2 scale

Default ggplot2 scale with tritanopia

Source: Mine Çetinkaya-Rundel, Sta313. Generated with {colorblindr}.

Use whitespace or pattern to separate elements

• Separate elements with whitespace or pattern

• Allows for distinguishing between data without entirely relying on contrast between colors

Legible and clear

Graph A: can you read it?

What can you read from this graph? What can’t you?

Graph B: same graph

What changed?

Contrast of the text against the white background: Graph A 2.1:1, Graph B 21:1

Text size and contrast

• Make text big enough to read from the back of the room, and when zoomed in
• Contrast between text and background: at least 4.5:1 (WCAG AA). Large text and graphical elements: at least 3:1
• Light gray on white is the usual culprit
• Check yours: WebAIM Contrast Checker or colorspace::contrast_ratio()

Graph A: what’s this graph about?

Explain this graph to someone who can’t see it. What are the variables and units? What’s the point?

Graph B: same data

What changed? Which graph tells you what to look for?

Plain-language titles and labels

• Say what the graph shows
• Write axis and legend labels for people: “Median hourly wage (US dollars)”, not hourly_wage_median
• Include units

Can you get the exact numbers?

Using only this graph (or its alt text): what was New York’s median hourly wage in 2010?

Offer the data

State	1998	2010	2020
California	$23.95	$41.03	$56.93
Minnesota	$21.32	$34.67	$38.24
New York	$22.67	$35.15	$43.19

• Give readers a table (or a link to the data) alongside the graph or in an appendix
• Screen reader users can move through a table cell by cell, and everyone can look up exact values
• Put the data source in the surrounding text, not in the alt text

Alt-text

Who needs alt text?

With a neighbor: who might need a text description of a graph? Think of as many situations as you can.

01:00

Alternative text

It is read by screen readers in place of images allowing the content and function of the image to be accessible to those with visual or certain cognitive disabilities.

It is displayed in place of the image in browsers if the image file is not loaded or when the user has chosen not to view images.

It provides a semantic meaning and description to images which can be read by search engines or be used to later determine the content of the image from page context alone.

Source: WebAIM

Try it: describe this graph

Imagine someone who can’t see this graph. Write 2-3 sentences that tell them what they need to know.

01:30

Alt and surrounding text

CHART TYPE of TYPE OF DATA where REASON FOR INCLUDING CHART

(plus link to data source somewhere in the text)

• CHART TYPE: It’s helpful for people with partial sight to know what chart type it is and gives context for understanding the rest of the visual.
• TYPE OF DATA: What data is included in the chart? The x and y axis labels may help you figure this out.
• REASON FOR INCLUDING CHART: Think about why you’re including this visual. What does it show that’s meaningful. There should be a point to every visual and you should tell people what to look for.
• Link to data source: Don’t include this in your alt text, but it should be included somewhere in the surrounding text.

Source: Writing Alt Text for Data Visualization

Alt Text Practice

CHART TYPE of TYPE OF DATA where REASON FOR INCLUDING CHART

• A scatterplot
• of median hourly wage of RN’s by year in California, Minnesota, and New York. The data span the years 1998 to 2020.
• The three states follow the same linear increasing trend until about 2007, when New York and Minnesota begin to flatten.

With a neighbor: which part of the formula is each bullet? What would you add or change? How does it compare to yours?

02:00

What’s wrong with these alt texts?

A. “Image of a graph.”

B. “A scatterplot of nurse wages.”

C. “California: $23.95 in 1998, $25.12 in 1999, $26.50 in 2000, $27.36 in 2001, $28.38 in 2002, $29.47 in 2003, and so on for every year and every state…”

For each one: what’s missing, or what’s too much? Use the formula (chart type, type of data, reason for including the chart) to help.

Screen reader demo: no alt text

Listen: what do you learn about this graph?

Screen reader demo (dry run first!). Turn VoiceOver on/off with Cmd+F5. Move to the image with Control+Option+Right Arrow (VO+Right Arrow). Safari tends to work best with VoiceOver.

Screen reader demo: with alt text

Listen again: what changed? What could you do now that you couldn’t before?

Accessible Visualization via Natural Language Descriptions: A Four-Level Model of Semantic Content

Alan Lundgard, MIT CSAIL
Arvind Satyanarayan, MIT CSAIL

IEEE Transactions on Visualization & Computer Graphics (Proceedings of IEEE VIS), 2021

To demonstrate how our model can be applied to evaluate the effectiveness of visualization descriptions, we conduct a mixed-methods evaluation with 30 blind and 90 sighted readers, and find that these reader groups differ significantly on which semantic content they rank as most useful. Together, our model and findings suggest that access to meaningful information is strongly reader-specific, and that research in automatic visualization captioning should orient toward descriptions that more richly communicate overall trends and statistics, sensitive to reader preferences.

Let’s try it!

CHART TYPE of TYPE OF DATA where REASON FOR INCLUDING CHART

1. Take one graph and two blank cards

2. Write an alt text description of your graph on one of your blank cards.

   • Please label with your plot number!

3. In two’s or three’s, trade alt text descriptions only

4. On your second blank card, try to draw the graph based on the alt text provided.

5. Now, look at the original graph. How’d you do?

04:00

Recap

• What was hard about writing alt text?
• Looking at the graphs, do you notice things that make it harder/easier to write alt text?
• What was hard about recreating from the alt text?

Code templates

Adding alt text to plots

Short:

```{r}
#| fig-alt: Alt text goes here.

# code for plot goes here
```

Longer:

```{r}
#| fig-alt: |
#|   Longer alt text goes here. Make sure to add line breaks ~roughly
#|   80 characters.

# code for plot goes here
```

Using Okabe Ito palette

nurses_subset |>
  ggplot(aes(x = year, y = hourly_wage_median, color = state)) +
  geom_point(size = 2) +
  ggthemes::scale_color_colorblind() +
  scale_y_continuous(labels = scales::label_dollar()) +
  labs(
    x = "Year", y = "Median hourly wage", color = "State",
    title = "Median hourly wage of Registered Nurses"
  ) +
  theme(
    legend.position = c(0.15, 0.75),
    legend.background = element_rect(fill = "white", color = "white")
  )

Double Encoding

Use both color and shape aesthetics

nurses_subset |>
  ggplot(aes(x = year, y = hourly_wage_median, color = state, shape = state)) +
  geom_point(size = 2) +
  scale_y_continuous(labels = scales::label_dollar()) +
  labs(
    x = "Year", y = "Median hourly wage", color = "State", shape = "State",
    title = "Median hourly wage of Registered Nurses"
  ) +
  theme(
    legend.position = c(0.15, 0.75),
    legend.background = element_rect(fill = "white", color = "white")
    )

Direct Labeling

Could do “by hand” with annotate(). Alternatively, use geom_text()

nurses_subset |>
  ggplot(aes(x = year, y = annual_salary_median, color = state)) +
  geom_line(show.legend = FALSE, linewidth = 2) +
  geom_text(
    data = nurses_subset |> filter(year == max(year)),
    aes(label = state), hjust = 0, nudge_x = 1,
    show.legend = FALSE, size = 6
  ) +
  scale_y_continuous(labels = scales::label_dollar(scale = 1/1000, suffix = "K")) +
  labs(
    x = "Year", y = "Annual median salary", color = "State",
    title = "Annual median salary of Registered Nurses"
  ) +
  coord_cartesian(clip = "off") +
  theme(
    plot.margin = margin(0.1, 0.9, 0.1, 0.1, "in")
    )

Direct Labeling

First, filter the data to include the endpoints only. Use the label aesthetic to map to the label in your data (in this case, state). geom_label by default will use the x and y aesthetics defined in ggplot()

nurses_subset |>
  ggplot(aes(x = year, y = annual_salary_median, color = state)) +
  geom_line(show.legend = FALSE, linewidth = 2) +
  geom_text(
    data = nurses_subset |> filter(year == max(year)),
    aes(label = state)
  ) +
  scale_y_continuous(labels = scales::label_dollar(scale = 1/1000, suffix = "K")) +
  labs(
    x = "Year", y = "Annual median salary", color = "State",
    title = "Annual median salary of Registered Nurses"
  ) +
  coord_cartesian(clip = "off") +
  theme(
    plot.margin = margin(0.1, 0.9, 0.1, 0.1, "in")
    )

Direct Labeling

(Here’s what it looks like if we don’t filter to the endpoints)

nurses_subset |>
  ggplot(aes(x = year, y = annual_salary_median, color = state)) +
  geom_line(show.legend = FALSE, linewidth = 2) +
  geom_text(
    aes(label = state)
  ) +
  scale_y_continuous(labels = scales::label_dollar(scale = 1/1000, suffix = "K")) +
  labs(
    x = "Year", y = "Annual median salary", color = "State",
    title = "Annual median salary of Registered Nurses"
  ) +
  coord_cartesian(clip = "off") +
  theme(
    plot.margin = margin(0.1, 0.9, 0.1, 0.1, "in")
    )

Direct Labeling

hjust=0 means “left justified”, or make the label start at the x-y coordinate you gave it. size = 6 makes the label bigger

nurses_subset |>
  ggplot(aes(x = year, y = annual_salary_median, color = state)) +
  geom_line(show.legend = FALSE, linewidth = 2) +
  geom_text(
    data = nurses_subset |> filter(year == max(year)),
    aes(label = state), 
    hjust = 0, 
    size = 6
  ) +
  scale_y_continuous(labels = scales::label_dollar(scale = 1/1000, suffix = "K")) +
  labs(
    x = "Year", y = "Annual median salary", color = "State",
    title = "Annual median salary of Registered Nurses"
  ) +
  coord_cartesian(clip = "off") +
  theme(
    plot.margin = margin(0.1, 0.9, 0.1, 0.1, "in")
    )

Direct Labeling

nudge_x = 1 “nudges” each label one unit in the x-direction (so each label is a small distance away from what it’s labeling). show.legend=FALSE tells ggplot not to include the aesthetics for geom_text in the legend

nurses_subset |>
  ggplot(aes(x = year, y = annual_salary_median, color = state)) +
  geom_line(show.legend = FALSE, linewidth = 2) +
  geom_text(
    data = nurses_subset |> filter(year == max(year)),
    aes(label = state), 
    hjust = 0, 
    size = 6,
    nudge_x = 1,
    show.legend = FALSE,
  ) +
  scale_y_continuous(labels = scales::label_dollar(scale = 1/1000, suffix = "K")) +
  labs(
    x = "Year", y = "Annual median salary", color = "State",
    title = "Annual median salary of Registered Nurses"
  ) +
  coord_cartesian(clip = "off") +
  theme(
    plot.margin = margin(0.1, 0.9, 0.1, 0.1, "in")
    )

Direct Labeling

Finally, we have to tell ggplot not to trim the plot, and leave room in the right margin for the labels themselves

nurses_subset |>
  ggplot(aes(x = year, y = annual_salary_median, color = state)) +
  geom_line(show.legend = FALSE, linewidth = 2) +
  geom_text(
    data = nurses_subset |> filter(year == max(year)),
    aes(label = state), 
    hjust = 0, 
    size = 6,
    nudge_x = 1,
    show.legend = FALSE,
  ) +
  scale_y_continuous(labels = scales::label_dollar(scale = 1/1000, suffix = "K")) +
  labs(
    x = "Year", y = "Annual median salary", color = "State",
    title = "Annual median salary of Registered Nurses"
  ) +
  coord_cartesian(clip = "off") +
  theme(
    plot.margin = margin(0.1, 0.9, 0.1, 0.1, "in")
    )

Add whitespace

Set the color aesthetic to white

nurses_subset |>
  filter(year %in% c(2000, 2010, 2020)) |>
  ggplot(aes(x = factor(year), y = total_employed_rn, fill = state)) +
  geom_col(position = "fill", color = "white", linewidth = 1) +
  labs(
    x = "Year", y = "Proportion of Registered Nurses", fill = "State",
    title = "Total employed Registered Nurses"
  )