3. Maximize the data-ink ratio, within reason (maybe)

The data-ink ratio

“Maximize the data-ink ratio, within reason.” (Tufte 1983; emphasis added)

  • Excessive decoration/color should be avoided (but see my caveat below)
    • All aspects of the graph should aid in presenting the data
      • The “data ink” ratio should be as high as possible (Tufte 1983)
        • The basic idea behind this principle: all aspects of a graphic should serve the purpose of showing the data
        • Consider whether superfluous additional pieces/colors on your graphic can be removed
    • My less strict rule: all color should convey some information

A busy, brightly colored cartoon monster covered in clashing patterns.

The data-ink ratio: my caveat

“… ‘remove clutter and strive for clean and elegant designs’… is reasonable advice. But… ‘do everything you can to remove non-data ink’…will result in poor design choices.” (Wilke 2019 p.277-278)

  • I am much less of a fundamentalist about the maximize data ink rule
    • In the age of PDFs and online journals/reports/websites, it isn’t “wasteful” to include extra color or other data ink
  • Sometimes the color displays useful information (e.g., your brain registers amount better with a bar chart than a dot plot because the blocks of color carry visual information on quantity)
  • Color can help you quickly identify different parts of the graph (e.g., the blue line shows…)
  • Sometimes the color… looks nice
    • Having aesthetically pleasing graphics is important

Dot plots

  • Dot plots can often be used interchangeably with bar charts
    • But, instead of a colored in bar, you simply place a dot at the value being plotted (would be the top of the bar)
    • Dot plots can similarly easily incorporate measures of uncertainty in the form of SEs or CIs
  • Dot plots were invented to take advantage of the knowledge gleaned from Cleveland and colleagues’ visual perception experiments and Tufte’s opinions about excess decoration

Dot plot of men's long jump records in meters for seven countries.

Meme of a startled cat with the caption: He possesses... the red dot!

Advantages of dot plots

  • An advantage of dot plots is that they waste less data ink when compared to a bar chart (Tufte 1983)
    • This also keeps them from becoming crowded and unreadable
  • A purely utilitarian reason to use a dot plot is that people will give you a hard time for having a bar chart that does not start at zero—while no one seems to care if a dot plot starts at zero
    • There are good reasons for this concern as bar charts display information both as length and area—and the area can be misleading without a zero baseline

Bluesky logo. X logo. Cartoon of a troll typing on a laptop.

use "https://tdmize.github.io/data/data/hrs_si.dta", clear
(Health and Retirement Study 1992-2022 - Cleaned for MCAP SI)
* Percent of men and of women in each category of days drinking
proportion dayweekdrink if woman == 0
Proportion estimation                  Number of obs = 108,573

--------------------------------------------------------------
             |                                   Logit
             | Proportion   Std. err.     [95% conf. interval]
-------------+------------------------------------------------
dayweekdrink |
          0  |      0.562      0.002         0.559       0.565
          1  |      0.131      0.001         0.129       0.133
          2  |      0.080      0.001         0.078       0.081
          3  |      0.062      0.001         0.060       0.063
          4  |      0.029      0.001         0.028       0.030
          5  |      0.029      0.001         0.028       0.030
          6  |      0.015      0.000         0.014       0.016
          7  |      0.092      0.001         0.090       0.094
--------------------------------------------------------------
est store drink_men
proportion dayweekdrink if woman == 1
Proportion estimation                  Number of obs = 154,553

--------------------------------------------------------------
             |                                   Logit
             | Proportion   Std. err.     [95% conf. interval]
-------------+------------------------------------------------
dayweekdrink |
          0  |      0.723      0.001         0.720       0.725
          1  |      0.107      0.001         0.105       0.108
          2  |      0.054      0.001         0.053       0.055
          3  |      0.036      0.000         0.035       0.037
          4  |      0.016      0.000         0.016       0.017
          5  |      0.016      0.000         0.016       0.017
          6  |      0.007      0.000         0.006       0.007
          7  |      0.041      0.001         0.040       0.042
--------------------------------------------------------------
est store drink_wom
// 3.a - Dot plot for days drinking per week, by gender
coefplot (drink_men, label("Men") offset(-0.15)) ///
         (drink_wom, label("Women") offset(0.15)), ///
    vertical noci msize(*1.5) rescale(100) ylab(0(10)80) ///
    xlab(1 "0" 2 "1" 3 "2" 4 "3" 5 "4" 6 "5" 7 "6" 8 "7") ///
    ytitle("Percent") xtitle("Days usually drink each week") ///
    title("[3.a] Dot plot for days usually drink each week (older adults, HRS)
") ///
    subtitle("Percent of men and of women")

[3.a] Dot plot of the percent of men and of women who usually drink 0 to 7 days a week: 56% of men and 72% of women drink 0 days; at every other number of days the percent is higher for men than for women, with a small rise at 7 days (9% of men, 4% of women).

library(foreign)            # load Stata datasets
library(ggplot2)            # main graphics package
library(dplyr)              # data manipulation tools
library(cleanplots)         # Mize's graphics scheme

## Use cleanplots colors, markers, lines, and theme for all graphs ##
cleanplots_defaults()
theme_update(plot.title.position = "plot")   # titles start at left edge of graph

# Health and Retirement Study Data
hrs <- read.dta("https://tdmize.github.io/data/data/hrs_si.dta")

# Percent of men and of women in each category of days drinking
drink_pct <- hrs %>%
  filter(!is.na(dayweekdrink)) %>%
  count(woman, dayweekdrink) %>%
  group_by(woman) %>%
  mutate(pct = 100 * n / sum(n))

# 3.a - Dot plot for days drinking per week, by gender
ggplot(drink_pct, aes(x = factor(dayweekdrink), y = pct, color = woman)) +
  geom_point(size = 4.5, position = position_dodge(width = .3)) +
  labs(x = "Days usually drink each week", y = "Percent", color = "",
       title = "[3.a] Dot plot for days usually drink each week (older adults, HRS)",
       subtitle = "Percent of men and of women") +
  scale_y_continuous(breaks = seq(0, 80, 10), limits = c(0, 80))

[3.a] Dot plot of the percent of men and of women who usually drink 0 to 7 days a week: 56% of men and 72% of women drink 0 days; at every other number of days the percent is higher for men than for women, with a small rise at 7 days (9% of men, 4% of women).

// 3.b - Same information as a bar chart
coefplot (drink_men, label("Men") offset(-0.2)) ///
         (drink_wom, label("Women") offset(0.2)), ///
    vertical noci rescale(100) ylab(0(10)80) ///
    recast(bar) barwidth(0.4) ///
    xlab(1 "0" 2 "1" 3 "2" 4 "3" 5 "4" 6 "5" 7 "6" 8 "7") ///
    ytitle("Percent") xtitle("Days usually drink each week") ///
    title("[3.b] Bar chart for days usually drink each week (older adults, HRS
)") ///
    subtitle("Percent of men and of women")

[3.b] The same percents as side-by-side bars for men and women at each number of days drinking: the bars for 0 days dominate the graph, reaching 56% for men and 72% for women.

# 3.b - Same information as a bar chart
ggplot(drink_pct, aes(x = factor(dayweekdrink), y = pct, fill = woman)) +
  geom_col(position = position_dodge(width = .8), width = .8) +
  labs(x = "Days usually drink each week", y = "Percent", fill = "",
       title = "[3.b] Bar chart for days usually drink each week (older adults, HRS)",
       subtitle = "Percent of men and of women") +
  scale_y_continuous(breaks = seq(0, 80, 10), limits = c(0, 80))

[3.b] The same percents as side-by-side bars for men and women at each number of days drinking: the bars for 0 days dominate the graph, reaching 56% for men and 72% for women.

YouGov dot plot: Americans overestimate the size of minority groups and underestimate the size of most majority groups, comparing the true and estimated percent for each group.

Figure from YouGov 2022

Back to top