Have I Resolved the Pie Chart Debate?

tl;dr: For decades, chart creators have disagreed on whether pie charts should be used, but I think there are a few important arguments that have been missing from the debate and that might—just might—help to bring it to some kind of conclusion.

---

I know.

People have been arguing about “the pie chart question” for literally 100 years. In fact, this topic has been so beaten to death that, in their article submission guidelines, the Nightingale editors specifically ask writers to not submit yet another $#%*ing article about pie charts (my paraphrasing). They also, however, mention that articles about beaten-to-death topics can be submitted as long as they add something new to the discussion. Well, I think I have something new to add to this particular discussion, but I’ll let you be the judge. Which I’m sure you will be.

Let’s cut to the chase: Am I for or against pie charts? Readers of my recent assassination attempts on box plots, connected scatterplots, and bullet graphs may be surprised to learn that I’m actually pro-pie-chart 😱. Yes, I believe that there are times when a pie chart is the best choice for showing the breakdown of a total. I used to be an anti-pie-charter, but, after taking a hard, honest look at my reasons a few years ago, I switched camps. Some of the key reasons why I changed my mind weren’t in any of the articles or books that I’ve read on the topic, though, so I thought I’d share them here to see if anyone else finds them to be persuasive, or has already heard these arguments from a source that I’m not aware of.

Because this is such a third-rail topic, I want to be clear about what I’m not saying:

• I’m not saying that pie charts are “better” than bar charts, stacked bar charts, or other breakdown-of-total chart types. Each of these chart types has a different set of strengths and weaknesses and each is the best choice in different situations (more on that below).
• I’m not saying that you should always use a pie chart to show the breakdown of a total. There are many situations in which I think a bar chart, stacked bar chart, or other breakdown-of-total chart type should be used (more on that also below).

I also want to set a few ground rules:

1. I’m only going to consider well-designed charts (of any type). Articles that deride pie charts sometimes feature pie charts that are poorly designed (slices not sorted, 3D effects applied, etc.). This doesn’t really prove anything because every chart type looks like a bad chart type if you show examples that are poorly designed. For the record, this is what I consider to be a well-designed pie chart (this diagram is from this article on how to format pie charts):

2. I’m only going to consider situations in which the most effective chart type is used (IMO, anyway). Articles that deride pie charts also sometimes show them being used in situations in which another chart type clearly would have been a better choice, which also doesn’t really prove anything because every chart type looks like a bad chart type if it’s used in the wrong circumstances.

When, exactly, do I consider that a pie chart should be used? Only in situations in which all of the following conditions are met:

• The values in the chart are parts of a meaningful total (not averages or other values that produce a meaningless number when summed).
• All the parts of the total are included in the chart.
• The parts (i.e., the categories) don’t overlap (e.g., if there’s a “software developer” category and a “data analyst” category, there should be no software developers who are also data analysts).
• The parts consist of values that are all positive or all negative (not a mix of positive and negative values).
• There are fewer than about six or seven parts (though this can vary based on the distribution of the parts and what, exactly, the chart needs to say about the data).
• It’s more important to show what fraction of the total each part represents, as opposed to showing precise comparisons of the parts, or cumulative subtotals of absolute values (dollars, employees, etc.) of the parts. (I’ll show examples to clarify what this means in a bit.)

BTW, these conditions are derived from a “cheat sheet” in my Practical Charts course and book:

When to use a pie chart

As it turns out, then, figuring out when to use a pie chart or another breakdown-of-total chart type is trickier than most people think, and I often see pie charts used in situations in which another chart type would have been a better choice. I suspect that this is part of the reason why many people don’t like pie charts: they often see them being used inappropriately. That doesn’t mean that pie charts should never be used, though.

With those exciting terms and conditions out of the way, let’s get into the meat and potatoes.

Since I’ve had debates with a lot of anti-pie-charters over the years, I think I have a good idea of what the main arguments against pie charts are, so I’m going to lay out my reasoning in the form of a debate between me and an amalgam of every anti-pie-charter whom I’ve debated. Presumptuous? Yes. My call because I’m the one writing this article? Also, yes 😊.

In those debates, anti-pie-charters virtually always point to bar charts as a better alternative, so their real claim doesn’t seem to be that pie charts are “bad,” but that bar charts are always a better choice than pie charts. I think the real question at hand, then, is whether there are times when a pie chart is a better choice than a bar chart.

Here goes. (Hey, Pandora, that’s a cool-looking box you’ve got there…)

Anti-pie-charter: Many studies show that audiences can’t compare values as precisely in a pie chart as they can in a bar chart. Consider the examples below. In the bar chart, it’s easy to precisely compare the lengths of the bars, but it’s harder to precisely compare the sizes of the slices in a pie chart:

Why use a less precise chart type (i.e., a pie chart) instead of a more precise one (i.e., a bar chart)?

Me: Yes, bar charts offer higher precision when it comes to comparing the parts with one another. Pie charts, however, offer higher precision when it comes to perceiving fractions of the total, and that might be more important, depending on what kind of message the chart is intended to communicate. For example, let’s say that the main message to be communicated isn’t that “Mrs. Perez donated more than Mrs. Smith,” but is instead that “Mrs. Perez accounted for a quarter of total donations.” If that’s the main message, the bar chart doesn’t make that clear at all:

What fraction of the total does Mrs. Perez represent? A third? A quarter? A sixth? Hmmm. In order to perceive that, the audience would need to imagine what all the bars would look like stacked on top of one another, and then imagine what fraction of that stack the “Mrs. Perez” bar would represent:

Obviously, this would require a great deal of cognitive effort and the audience’s estimates would probably be pretty bad. In a pie chart, however, it’s easy to see what fraction of the total each part represents with a pretty high degree of precision.

Therefore, a blanket statement like, “Bar charts are more precise than pie charts” simply isn’t true. Bar charts are more precise when it comes to comparing the parts with one another, but pie charts are more precise when it comes to perceiving fractions of the total. The type of precision that matters most in a given situation depends on what type of message the chart is intended to convey.

Anti-pie-charter: If the main message that you want to communicate is about what fraction of the total each part represents, just label the percentage values in a bar chart! That makes it easy to see what fraction of the total each part represents:

Me: Percentage labels do make it easier to see what fraction of the total each part represents in a bar chart, but still not as easy as in a pie chart. Why? Well, the graphics in this chart (i.e., the bars) don’t clearly communicate the main message in this situation, so the audience is forced to rely almost entirely on reading the (textual) percentage labels to get that message. That’s certainly doable, but mentally converting textual numbers into mental representations of quantities is slower and less informative than perceiving those quantities as graphics (shapes of different lengths or areas, darker or lighter color shades, etc.).

Anti-pie-charter: Come on… It’s not hard to read percentage numbers and convert those into mental representations of quantities!

Me: Okay, but I could then say the same thing about a pie chart with the percentage values labeled:

Ta-daa! Now, the main weakness of pie charts (i.e., the inability to compare parts precisely) disappears! The audience can just read the percentage labels and compare the parts with perfect precision!

Okay, I was being a bit facetious, there, because I wouldn’t actually make that argument regarding pie charts, bar charts, or any other chart type since perceiving quantities as textual numbers is much slower and less informative than perceiving quantities as graphics. The superior speed and informativeness of graphics are literally why we bother to show data as graphs (as opposed to tables or lists of numbers) in the first place, so this is kind of an important point.

Therefore, if the main message of a chart is about what fraction of the total the parts represent, you could use graphics that show that clearly (i.e., pie slices), or graphics that don’t show that clearly (i.e., bars). I’m not sure why you’d choose the latter.

Anti-pie-charter: The bar chart that you showed isn’t a fair comparison with the pie chart because it forces the audience to read textual percentage labels. A bar chart with a “percentage of total” scale would be a fairer comparison with a pie chart:

Me: This version still forces the audience to read textual percentage labels. In fact, seeing what fraction of the total each part represents requires even more cognitive effort in a chart like this because the audience must estimate the value for each bar by comparing its length with the (textual) scale of percentages, and then mentally convert those (still textual) estimates into mental representations of fractions of the total.

Anti-pie-charter: Okay, but at least the audience will be able to see what portion of the total each part represents more precisely in a bar chart like this than in a pie chart. For example, in the bar chart above, I can see that Mrs. Jones represents 16% (not 15% or 17%) of the total.

Me: If the main insight to be communicated requires showing what portion of the total each part represents to within 1%, a pie chart with percentage labels on the slices would still be better than a bar chart (with or without percentage labels on the bars). This is because the graphics in a pie chart provide a pretty precise idea of what portion of the total each part represents without reading the percentage labels, and the percentage labels just serve to “fine tune” those estimates. In a bar chart, however, the graphics (i.e., the bars) provide almost no idea of what portion of the total each part represents, so the audience would need to rely almost entirely on reading text labels on the bars or on the quantitative scale to form a mental representation of the values.

So, if the main message requires the audience to perceive portions of the total with a very high level of precision (which is rare in practice, BTW), they will need to read textual percentage labels regardless of what chart type is used. The choice, then, is between a chart type that provides a pretty precise idea of what portion of the total each part represents even before reading the labels (i.e., a pie chart) and a chart type that doesn’t (i.e., a bar chart). Again, I’m not sure why you’d choose the latter.

Anti-pie-charter: In your pie chart example above, it’s only clear that Mrs. Perez represents a quarter of total donations because that slice starts at the “12 o’clock” position. If that slice started at some random angle instead, it would be much harder to see what portion of the total it represents.

Me: Okay, let’s try it:

That still looks like a quarter to me. The audience might not be able to say whether it’s exactly a quarter, but they’ll be able to say that it’s very close to a quarter, probably to within a few percentage points or less. The more important point here, though, is that pie slices still show fractions of the total far more precisely than bars do. In a bar chart, fractions of the total are always hard to perceive, whether a bar is in the first position in the chart or in any other position.

Anti-pie-charter: Your example only works because the slice that’s being featured is a “familiar fraction,” like one quarter, one third, three quarters, one half, etc. It would be much harder if it were an oddball fraction like 5/12ths or 4/15ths.

Me: Audiences will perceive a “5/12ths” slice as “less than half,” and a “4/15ths” slice as “a bit more than a quarter,” which is actually pretty precise. If the audience had to estimate those fractions as percentages, they’d probably be within a few percentage points, which is comparable with many other chart types. As I mentioned a moment ago, though, the important point here is that pie chart slices show fractions of the total far more precisely than bars do. In a bar chart, fractions of the total are always hard to perceive, whether a bar represents a familiar fraction or not.

Ultimately, I don’t think it really matters how well audiences can assign fraction or percentage values to the slices in a pie chart, because they don’t need to do that in order to read a pie chart. The audience just sees slices of various sizes that represent parts of various sizes, and they don’t need to mentally come up with textual fraction or percentage numbers in order to read the chart; they’d only need to do that if they needed to describe the values in the chart to someone else (such as a data visualization researcher).

I suspect that this is one reason why so many audiences like pie charts: they show breakdowns of totals in a way that’s probably very similar to how we represent those types of quantities in our minds, so they require very little “mental translation” to interpret. Reading textual percentage labels off the scale in a bar chart, however, requires considerably more “mental translation” effort to form a mental representation of the parts of a total.

Anti-pie-charter: What about a pie chart like the one below, in which all the parts are very similar in size? It’s problematic because the audience can’t see the small differences among the parts. In a bar chart, though, you can see those small differences:

Me: If the main message requires precise comparison of the parts, then, yes, you should use a bar chart. If, however, the main message is that each donor accounted for about a fifth of total donations, the pie chart makes that point more clearly since the audience doesn’t need to read any textual numbers or count the number of categories, which a bar chart would require them to do. As usual, it depends on what, exactly, you need to say about the data.

Anti-pie-charter: What about the one below, where some of the parts are tiny? I can barely even see what those parts are in a pie chart!

Me: A few comments, here:

1. Tiny values like these are also hard to see clearly in bar charts and in pretty much every other chart type, so this isn’t a weakness of pie charts specifically.
2. As I mentioned earlier, if it were important to communicate these values very precisely (e.g., to communicate that Mrs. Jones accounted for precisely 1% of donations, and not 2%, or 0.5%), then percentage labels are going to be necessary, regardless of whether you’re using a pie chart, bar chart, or any other chart type. In a pie chart, at least, the audience gets a pretty precise idea of what fraction of the total each part represents before they read the labels.
3. If the main message of this chart is to show that Mrs. Jones and Mr. Schultz accounted for only tiny fractions of the total, the pie chart above communicates that message effectively (and better than a bar chart would), so it’s not inherently “broken” and could work just fine as-is, depending on what the chart is intended to communicate.

Anti-pie-charter: I still don’t like the fact that you can’t precisely compare the parts with one another in a pie chart. It’s always important to be able to do that.

Me: I think that the examples in this article illustrate that sometimes it’s more important to compare the parts precisely, but sometimes it’s more important to show fractions of the total precisely, depending on what the main message of the chart is.

Anti-pie-charter: Okay, maybe it’s not always more important to compare the parts precisely, but, in practice, that’s far more common than needing to show what fraction of the total each part represents.

Me: In my experience, both scenarios are common. For example, fraction-of-total insights like, “A third of our revenue comes from one customer,” or “Operations and Administration accounts for almost half of our expenses” are pretty common.

---

I hope I’ve illustrated that bar charts don’t communicate fraction-of-total insights as effectively as pie charts do, but it’s also worth noting there are other things that bar charts don’t do well, such as showing cumulative subtotals of absolute values. In those situations, a stacked bar chart would be a better choice than a bar chart (or a pie chart):

Stacked bar charts also have their weaknesses, though. For instance, they don’t allow the parts to be compared as precisely as bar charts do, and they don’t allow fractions of the total to be perceived as precisely as pie charts do.

My point is that all of these chart types do certain things well and other things poorly, and bar charts are no exception. Bar charts also have another downside, which is that they don’t make it visually obvious that a chart is showing the breakdown of a total. This is because bar charts are kind of a unique chart type in that they can show pretty much any kind of data, but the problem is that they look the same regardless of what kind of data they’re showing:

When the audience sees a bar chart, they have to read the chart title, axis labels, etc., and then figure out what kind of data the chart is showing before they can begin to interpret it. This is certainly doable, but it does make the chart slower and more effortful to read, and this becomes especially noticeable in a report, dashboard, or presentation that contains many charts that show different kinds of data. When the audience sees a pie chart (or a line chart, or a map, etc.), however, they instantly know what kind of data they’re looking at—even before they’ve read the chart title or axis labels—which increases the speed and ease of interpretation.

What do you think? Have I resolved the pie chart debate? If you’re an anti-pie-charter, did I change your mind, or are there flaws in my arguments? Have you heard these specific arguments before? Let me have it on LinkedIn or Twitter [covers head with arms].

And here’s an Easter egg to see if people responding on social media have actually read this whole article: If you post a response (which I heartily encourage you to do, especially if you disagree with me), please add a second period to the end of the last sentence in your response, like this..

BTW…

This article is based on an excerpt from my brand-new Practical Charts book. Visit the official book page to learn more.

A huge thank-you to the following reviewers (several of whom are or were anti-pie-charters), who provided valuable feedback on a draft of this article: Bryan Pierce, Jeff Elmendorf, Kevin Flerlage, Jennifer Stanier, Chris Tauber, Stephen Redmond, Jan Weilers, James Thomson, Jami Dennis, Paul Verweij, Mark Bradbourne, and Matthias Giger.