Why you (probably) don’t want a pie chart: Reflections on Pi Day

Pi is an irrational number in that its decimal form never ends. Similarly, it seems like, for the go-to data visualization person, the request for pie charts will also never end. Anyone who’s ever had even a cursory introduction to the best practices of data visualization has likely (maybe?) heard a diatribe against the pie chart. Its form is bad. It doesn’t communicate clearly. And so, it really doesn’t do the job of data visualization.

Unlike pi though, the request for pie charts is not really irrational. When people ask for pie charts, they’re requesting information to gain insight about a dataset. I don’t think they actually care about pie charts in and of themselves, but rather are seeking specific information. The pie chart seems to be the default because for a long time it was what popped up after pushing the “insert chart” button. (Or, maybe people really like pie.)

But it’s time for the default to change. So on Pi Day, I’m here to not only share some of those best practices that show why you (probably) don’t want a pie chart, while also showing an alternative.

What’s in a visualization

To understand why pie charts have a bad rap, and why they deserve it, we can start by breaking down the major elements we use in data visualizations:

• Length: Bar and column charts that compare number count or percentage
• Direction: Line charts that show fluctuation
• Angle: The inner portion of a pie chart that determines a slice’s overall size
• Area/size: Bubble charts where the bigger one represents a larger number
• Curvature: Donut charts that are like pie charts but don’t have a pointy end
• Volume: Anything that’s three dimensional

The ordering of these bullet points is not accidental. In fact, it’s a loose hierarchy of the most effective ways to convey information accurately. In 1984, a couple of researchers published a study based on experiments they did about which visual elements are easiest for people to interpret. Length and direction are the easiest, while the research showed that angle, area, volume, and curvature led to the most instances of inaccurate interpretations of data. And while this research is almost as old as I am, it remains frequently cited.

Now, let’s look at two re-configured versions of pie charts I saw in the wild a couple weeks ago. The numbers and formatting are almost identical, but the labels are hypothetical and represent the leisure budget of two hypothetical households. The charts are meant to highlight the different proportions of spending on restaurants between the two households.

You will notice that these two charts, and all pie charts, only include the visual elements that research has shown are the most difficult for people to interpret. A slice of the pie includes an angle at the point, a representative size, and curvature at the slice’s edge. At least this one’s not 3-D.

Your go-to data visualization person is not being a curmudgeon when they resist the pie chart request. They’re following best practices.

Keep elements to a minimum

You’ll also notice in the charts above that they also include two additional stylistic elements that are also used to communicate information:

• Color, used either to distinguish dimensions or shaded to reflect numerical difference
• Text/data labels, used in many different charts to add specificity to proportionality

These are not only preferences. A pie chart must have different colors to distinguish dimensions, and the labels are necessary because judging proportionality between slices is quite difficult.

And that brings us to the other reason you (probably!) don’t want a pie chart: There’s just too much going on. They include three visual elements — all of them difficult to interpret — color, and text. In the examples above, the restaurants category has a white background because those are the numbers the audience is supposed to be drawn to. That’s yet another element.

Luckily, charts the like the one above are very easy to reconfigure into something that has fewer elements while also hitting a much easier way to interpret information:

This column chart contains all of the information from the two pie charts cited above, except it uses one design element (length) and two colors rather than six. Rather than including a legend to signify what the colors represent (which would be another chart element), I used the title to do that work. In a chart like this, I like data labels inside the columns rather than the y-axis percentage scale, but that would work as well — only one form of labeling is needed though.

To highlight restaurant spending, which the original pie charts and indicated are the important numbers, I put it on the left side of the x-axis. I also could have made those two columns a different color or lightened the other columns so restaurant spending stands out, or I simply could have put a box around them to draw the audience’s attention there.

Ok but when do I probably want a pie chart?

The scenario in which your go-to data visualization person will not only comply with the pie chart request, but probably select it themselves, is when you have two variables. In this situation, only two colors will be needed, and frequently only one number as well. Better yet, the complications that angle, curve, and area bring are simplified because there are only two pieces of information to take in and take apart. Going back to the pie charts we started with, I isolated restaurant spending and combined all of the other forms of revenue shares into one “other” category, since that was the part of the budget to focus on.

With two variables, there’s little room for error in visual interpretation. Because only one number is important, we don’t need to display the other one. It’s quickly evident that Household 1 spends more of its leisure budget at restaurants.

This blog post won’t cease the request for pie charts for me or anyone else. Like pi, they are eternal. But that’s okay because sometimes they are exactly what you want — even though in most cases you really, truly don’t want a pie chart.

Send chart requests or any other questions or inquiries to eric@theimpactarchitects.com