Pie charts did nothing to deserve how you’re treating them

The humble pie chart is no miracle worker, but surely it doesn’t deserve the constant barrage of hate mail and hatchet-job pieces it’s been subjected to. Take this one that has been making the rounds:

What is wrong with this picture? Critics point out that for the same data, the pie charts look similar, even though the bar charts look different.

Look, pie charts are simple creatures that tell simple stories. Those stories are about composition. Is my total made of many categories, or only a few? How are they distributed? Are they evenly distributed, or do a few categories dominate?

With multiple pie charts, it’s about comparing these features. How similar are the shares of these categories? Do the same categories dominate in each instance, or do they change? That’s all there is to it.

When you look at three pie charts and can’t tell much of a difference, the story isn’t that pie charts are bad, it’s that in those three instances, the composition is pretty similar! The poor pie chart did its simple and thankless job, you ingrates. No more, no less.

Unconvinced? Let’s look at the cold, hard numbers:

The largest overall difference is from graph A to graph C, and the largest shift in composition that occurred in a category is a whopping 6%!

In conclusion, the compositions remained very similar, so the pie charts look similar. The story that the pie chart told us holds up.

Of course, the aspiring intellectual will point out that sometimes, small changes matter too. A modest proposal: if you really want to show difference, why not plot difference?

I’m no artiste, but after two minutes with Excel, this chart seems much more effective at conveying small differences. With better than the bar charts too!

However, looking only at this chart, you could be lead, perhaps, to believe that these five categories changed a lot in absolute terms from situation A to situation C. After all, look at how long the red bar is! Our ever-loyal friend, the pie chart, informs us otherwise.

What we’re seeing here is a design continuum. It’s a metaphorical line along which we can move, one way or another, to trade off one benefit for another. It’s up to us to decide the tradeoff we want to make to serve the purpose we want best served.

The pie charts say that nothing changed much overall from A to C. The difference bar chart (I made this name up) says that red and black changed the most, but in opposite ways. Both of these are different statements derived from the same data, and both are true at the same time. The regular bar charts in the middle say a bit of both, but tells each story a bit less clearly than the others.

Of course, if you want to compare categories across separate instances, you might as well put the categories you want to compare closer to each other:

Three reds look odd side by side, so I colored them differently here.

Here, it’s easier to see that the Red category went up, the Black category went down, and the Green category didn’t change much from A to C.

I’d place this version similarly to the middle barchart, but closer to the left on the continuum — it’s easier to compare the same category across A, B and C, but at the expense of cross-category comparisons within each instance.

Another point to be made is that perceptual troubles plague all kinds of charts, just in different ways. Much like Cardinal Richelieu, show me any chart and I can probably find a way to condemn it as confusing. We can effectively misuse any type of chart depending on the characteristics of the data. Take, for example:

I definitely can’t tell whether red or black is larger, and I deleted the data, so no one will ever know. Another great (and well-known) example is electoral maps:

Ray Grumney / Star Tribune

The red looks huge, the blue looks tiny, but we know that overall Obama won in 2012. There’s ways to combat this by trading off geographical accuracy for size accuracy, like .

Also, there are some things that mostly just have a net benefit. Consider:

If your comparison of charts affords every beneficial feature to one type of chart, but removes common sense ones for others, that’s just not fair play.

Here’s a situation in which pie charts might be OK. Imagine a company that’s holding elections for a leader across three locations.

Looks like Person A was somehow competitive in both Boston and California. The race in Boston was very close, but in California there was another strong competitor, so there it was very close among three people. Somehow Person C had no influence in Boston, like person A and B did. Albuquerque just did its own thing, mostly unaffected by outsider candidates from Boston and California.

So how come pie charts worked well in this situation, but not the other? I think it has to do with the kinds of differences we’re trying to show. Here we’re showing characteristic differences —we might expect that there’s some strong relationship between people and the locations they influence, and thus good variance when we look across these. If we have data where everything mostly looks the same, it’s not going to work out.

Here’s another situation where this works well: when looking at questionnaire results across categories.

We see that of the people who enjoy outdoor activities, most are dog owners. Simple and clear.

So if you want to complain about something, complain about the rampant misuse of charts instead. No chart type is good at everything, and charting tools can’t read our minds yet. Pick the chart type that tells your story best, and ameliorates the pitfalls that you want to avoid.

Pie charts are OK. They existed before most of us, and they’ll probably outlive all of us. When we finally read the climate-driven death sentence of our planet in a 100-page report, it will probably have pie charts. So enjoy your artisanal donut charts and your home-baked square pie charts while you can, hipsters. Just kidding. Till then though, you can find me putting gold stars on every good little pie chart I encounter.

Lay off it, friends!