(part two of two)
In Glass Benches part one, I argued that it’s worth spending ink to help readers interpret unusual graphs. Plenty of designers have come to the same conclusions, and how-to-read sections aren’t uncommon in data visualization. See this heatmap from the Financial Times:
The more intricate the visualization, the more complex the how-to-read becomes, like this cloud of mascaras from Sonja Kuijpers:
Novel charts need explanation. But to be honest, I find how-to-read sections alienating. They take me away from the content of the visualization. Traditional legends doesn’t teach me anything about the topic at hand—they just teach me about the design of the visualization.
I’m a designer. I love design! But information design doesn’t mean information about design. It means designing to inform. I prefer active legends that demonstrate how to interpret the visualization by using part of the actual visualization. I also suspect that active legends are easier to remember than external legends. The reader uses an active legend to interpret real information, rather than passively observing it in a hypothetical situation.
In a visualization of oil imports from Saudi Arabia, The Washington Post used how-to-read instructions, but placed them within the chart itself, along with explanations and examples of how to read some of the stand-out countries.
A YouGov visualization of supermarkets does away with legends entirely. Instead, axis titles are replaced by instructions, and a single cell in the heat map is explained in words.
The format of that explanation can be used across the entire visualization: XX% of people who do most of their grocery shopping at YYYY also shop regularly at ZZZZ.
The New York Times takes active legends a step further in a story about a Chinese disinformation campaign on Twitter:
The graphic doesn’t explicitly say that each dot is a tweet. Instead, individual dots are labelled with representative tweets that illustrate the account’s evolution over time. The color coding is explained by highlighting the names of languages in the text of the article. The only explicit explanation is for bubble size: there’s no easy way to indicate that more retweets equals bigger bubbles. That explanation is placed at the bottom of the beeswarm, close to the part of the timeline where posts were retweeted more often.
My reading of this graphic is colored by my own experience with beeswarm charts. I immediately know that one dot is one case of something, placed along a scale and sized based on another variable. This graphic makes sense to me, but might be under-explained for someone who spends less time with graphs. If a chart shows multiple variables, I tend to prefer a small traditional legend in combination with an active legend inside the chart. But almost all the ink in this chart tells me something about @HKpoliticalnew, rather than telling me about the chart’s structure.
For me, external how-to-read sections are most effective when a visualization follows a common genre convention (like bigger = more) or has some element that maps very literally to the data it conveys. It’s much harder for me to apply a how-to-read section when every part of the visualization is novel and unintuitive. Two pieces by Federica Fragapane illustrate this.
Space Junk shows debris in orbit, grouped by their distance from Earth. I can move easily between Space Junk’s how-to-read section and the visualization itself, because several of the encodings make intuitive sense. Distance from the x-axis corresponds to distance from Earth. Thicker lines indicate greater mass. Other details escape me at first glance, but I can use the more intuitive mappings to scaffold my understanding of color and circle size.
In contrast, her work on the Social Progress index is very difficult for me to interpret:
There are a ton of different visual elements, none of which have an easy semantic link to what they depict. Bigger does generally indicate more, but “more” is sometimes good and sometimes bad. “Freedom of Religion” points up from the center along with “early marriage” and “corruption,” while “community safety net” and several types of tolerance move in the same direction as “discrimination and violence against minorities.” There’s a ton to interpret here, and I have to memorize a diagram of a hypothetical country to understand it.
I am picking on Social Progress and using it as an extreme example by removing it from its original purpose. Fragapane describes it as “artwork” in her Behance portfolio, and that is accurate! As data-driven artwork, it’s lovely: ridiculously well-balanced and pleasing to the eye. But taken as a piece designed for communication, it’s an example of barriers between the reader and the information being visualized.
Contrast this with an admittedly simpler piece from Bloomberg about Brexit voters:
This is also a visual puzzle. It pieces together tiny icons, without easy visual metaphors for what they represent. However, the traditional legend outside the chart combines with an active legend to orient readers to the unusual design. And that active legend also informs readers about voters’ changing opinions on Brexit.
Visualization for communication means getting information to your readers. And readers do need help! It’s a good idea to spend ink on increasing a reader’s understanding. Rather than providing a skeleton chart in an external legend, consider leaving the meat on the bone and using an active legend instead.
Hat tip to Andy Kirk’s “Little of visualization design” series for helping me to find examples for this post, and to RJ Andrews for suggesting it!