Primary points of the argument:

1. Embellishments can be useful

There are a few cavets here.

Embellishment can work (primarily) if:

  1. The chart communicates an extremely brief message
  2. The chart contains a small set of quantitative values
  3. The embellishment is tied metaphorically to the chart’s message.

2. The debate between mimalist and embellished data viz is often based on emotion, not data (ironically)

To make the comparison more quantitative, Carl Gutwin, Aaron Genest, David McDine, and Christopher Brooks of the University of Saskatchewan conducted a study titled: “Useful Junk? The Effects of Visual Embellishment on Comprehension and Memorability of Charts”

The study was designed to test the influence of chartjunk vs. plain (“minimalist”) graphs on two effects: comprehension and recall. In the end, long term recall was far better in the embellished charts, but short term recall was practically the same between the two groups.

On the flipside, non-embellished graphs are useful in the following (primary) situations:

  1. Present nothing that isn’t needed.
  2. Represent data in a way that is easy for the eyes to perceive and the brain to interpret.
  3. Provide appropriate context for interpreting the meaning of the data.

3. Chartjunk should be defined more clearly

Chartjunk was relatively vaguely defined by Edward Tufte and excludes any kind of thematic display of data as useless. However, embellishment can primarily assist comprehension by:

  1. Engaging the interest of the reader (i.e., getting them to read the content)
  2. Drawing the reader’s attention to particular items that merit emphasis
  3. Making the message more memorable.

All of these can only work under the assumption that they “refrain from undermining the message by significantly distracting from it or misrepresenting it.”

4. Benefits of the study

  1. Properly designed embellishments can make simple messages more memorable.
  2. Encourages a more rigorous definition of “chartjunk”
  3. Sets up a good platform to scientifically test different kinds of embellishments.

General comments:

Looking at anything from the New York Times, Bloomberg graphics, The Pudding, or 538, almost every visualization they have is (at a minimum) thematically embellished. Politics take a huge role in the design of their visualizations, and they make a conscious effort to make concrete connections to their viewers and the data. The abstractness of statistics and mathematics, in general, often scares people away from trying to analyze visualizations, but mixing the visualizations with relatable and relevant imagery or icons can, as shown above, have real benefit.

Also, I’d be curious to see if you could use eye tracking (pupil dilation, or gaze fixation) and other biometrics to more accurately measure visual attention on the graph at hand.