A little over a year ago, I sat in a ballroom at Rstudio::conf, in stunned awe of the keynote talk I had just seen, given by renowned visualization researchers Fernanda Viegas and Martin Wattenberg. The talk was a tour de force of data visualization, presenting totally novel dataviz methods that were both technically astounding and beautifully designed. As the applause died down, questions started rolling in, and right off the bat came a zinger, “Why are you, as visualization experts, using pie charts?“
I suspect that the question was at least partially tongue-in-cheek, but I was still disappointed that after such an inspiring talk, some people were just interested in parroting one of the most tired dogmas in dataviz: no pie charts.
Dogmas arise as shortcuts. As a beginner in a new field, there’s so much new stuff to learn, so we use rules to help: do this, don’t do that. Easy. Simple. Clean. These rules can indeed be helpful, but they should be thought of as starting points, not the be all end all. The problem comes when you never move past the rules. To quote Andy Kirk:
Far too often in visualisation, things that are supposed to be considered as general guidelines get marked up — and preached to others — as rigid rules when they can’t be universally applied. Don’t label your axis? Might not need it. Don’t offer a colour legend? The colours may be immediately obvious in their association. Using red and green? Maybe your audience has no colour-blind members.
Today we’ll explore three dogmas in dataviz and design to show where they fall flat, why you should always think deeper, and why you should never have blind faith in any rule.
Let’s return to Rstudio::conf and that question about pie charts. Martin Wattenberg was quite gracious, and simply laughed and gave an honest answer: “I think pie charts are hugely underrated. In cases where you only have a small number of categories and you’re interested in comparing the sums of different slices, I think they have a lot of advantages over other charts. If you look at the history of research into pie charts, since the 1920s, people have been trying to prove that they’re bad, and they never really have.”
There’s a couple of lessons here: first is that no rules are universal. Pie charts get a bad reputation because of poor use, people showing pies with 50 slices, each as unintelligible as the last. But like Martin says, when used with a small number of categories, and for specific purposes, they perform quite well. The second is that you should always be skeptical and look at the evidence yourself before blindly believing something. As Martin pointed out, people will often say research shows pie charts are bad, but if you read the papers yourself, they’ve never actually managed to show that in a convincing way. This lesson was echoed by visualization researcher Steve Haroz:
Letting design rules limit how you communicate is an unfortunate mistake. As many so-called rules for visualization are little more than popular opinion with no empirical backing, they can cause many missed opportunities. One way to avoid this problem is to seek evidence to support a rule and to ensure that the evidence goes beyond a mere anecdote or the insistence of an “authority” in visualization. If you can’t find the evidence, the rule is just an opinion. And your opinion is as valuable as anyone else’s.
The golden ratio is an irrational number, approximately 1.618. It’s most often expressed as a rectangle with sides of length 1:1.618, but can be captured in a multitude of forms. This number has captivated artists and designers the world over. It appears all over nature, and has been repeated in architecture, art, and logos we see every day. Many young designers are taught that this number is the key to aesthetically pleasing layouts and beautiful typography. Just do a quick google search and you can find hundreds of articles and youtube videos explaining how to use the golden ratio to craft the perfect design.
But the reality is much the same as pie charts: years of research has shown no effect between the golden ratio and an inherently pleasing aesthetic. It’s true that you can find the golden ratio all over nature, but this is likely a case of finding what you’re looking for. Measure a bunch of distances on the human body and you’ll find just as many 1.2’s or 1.3’s as 1.618’s. If you want the full history of where this misbelief and subsquent dogma originated, there’s a whole article about it. But all you really need to know is spelled out by a simple experiment. For years, Keith Devlin, a mathematics professor at Stanford, has asked hundreds of psychology students to pick their favorite rectangle out of a group of possibilities. If there was any merit to the golden ratio, we would predict that a higher proportion of students would pick the golden ratio rectangle, but that’s not what happens. Instead, students pick rectangles essentially at random.
But even if it weren’t for the lack of evidence, this is a case where you should always be wary of seductive shortcuts. It sounds so easy, just use this ratio and your designs will look nice. But of course that’s not the case. Even if it were a real effect, the golden ratio would be nothing more than a starting point, and by no means a guaruntee of good design.
It’s hard to overstate the weight of bar charts in data visualization. As Amanda Cox famously said, “There’s a strand of the dataviz world that argues that everything could be a bar chart. That’s possibly true, but also possibly a world without joy.”
I tend to agree, but I also see a flaw in the original argument for why everything could be a bar chart. The reasoning most proponents give is that a bar chart is the most accurate representation of data. Or rather, that it’s the most easily and quickly perceived. This entire line of reasoning can be traced back to a single research paper: “Graphical Perception and Graphical Methods for Analyzing Scientific Data”, 1985. In this paper, Cleveland and McGill showed participants a variety of different chart types and measured how accurately they could determine the numerical values behind these visual encodings. They found that encodings like position along an axis or length (as used in bar charts) were the most accurate, while encodings like area and color were less so.
This result was repeated through so many foundational dataviz books, courses, and tutorials that it became ingrained in us that length is more accurate than angle, which is more accurate than color.
The problem in this case isn’t that I doubt the paper, but that the results have been misinterpreted and applied without critical thought. What the paper measured–ability to accurately determine the precise number behind a visual encoding–is not the purpose of a chart. Charts are meant to give a visual overview of a dataset, to help us synthesize many values into a quickly readable format, and help us make comparisons. If you want to communicate the precise values in a dataset, use a table. The lesson being that instead of blindly following a set of rules, you need to think deeply about the purpose of what you’re designing: who is your audience, what are their needs, what is the desired outcome, and how does your design achieve that?
Work is barrelling forward on the dataviz membership program I’m developing with some friends. If you expressed interest in this last time I mentioned it, you’ve likely received a big email with a survey and updates on our plans. Actually, you likely received a whole bunch of duplicate emails because we had some technical glitches… my apologies for that. If you weren’t aware of this but you’re intrigued, feel free to reach out by responding to this email. The gist is that it’s a paid membership where a group of five information design experts will provide all kinds of tutorials, goodies, and exclusive content for members, along with providing a community and career support. The team includes myself, Alli Torban, Duncan Geere, Gabrielle Merite, and Jane Zhang. I’m really excited about this and think it’s gonna be absolutely killer!
I decided that for now I would stop streaming my dataviz project about The Office. My work at Axios is awesome, but also exhausting, and I simply don’t have the energy at the end of the day to work on camera. But I’ve decided to keep working on the project off-screen. Things are moving forward, and I’m excited about where it’s going!