One Potential Reason to Use a Pie Chart

One debate that never dies is the use of pie charts. It seems people love them or hate them. Because the human eye doesn't excel at comparing areas, we typically avoid them when we can. But the other day I was in a visual analytics training course that brought up an interesting counter-point. Let me explain:

Comments

#1

Submitted by Show Me The Data (not verified) on

Users? I think you mean followers.

#2

Submitted by Timcost (not verified) on

I use sometimes use a pie chart as a high level look at a measure on a dashboard. I try to limit the number of wedges in the pie to something like 3-5. they can make good filtering elements for dashboards. Click a wedge to drill down to a more precise viz.

#3

Submitted by Ben J. on

Hi Daniel - you bring up a great point. Due to its popular usage, there is no need to explain to someone that a pie chart shows a part-to-whole relationship, and thus the *purpose* of the chart is "preattentive" like the icons in your example.

How well the pie chart accomplishes that purpose is another matter, though.

For a small number of slices, a (2D only!) pie chart can give a general sense of relative size, but not a very precise one (e.g. is the red or green pie slice bigger and by how much? To cheat, just glance up at the bar chart for a crystal clear answer).

In this example, perhaps a single stacked bar with four colored sections would be even better since the part-to-whole is immediately obvious and the ability to compare sizes accurately is easier on the brain.

I will say this, I appreciate your bravery in coming to the defense of the much maligned pie chart. They take a beating out there.

#4

Submitted by Daniel H. on

Tim, that's a good method, using the pie as an entry point for the whole dashboard.

Ben, while I agree on a single-stacked bar (I just used this the other day in a presentation), my experience is that most people still don't immediately understand that the bar represents a whole--and it requires explanation which is time consuming. There's something about pies that everyone immediately understands.

I use a single-stacked bar with people who understand what I'm doing (particularly internally), but if I were to do an external presentation, I might, might consider using the pie instead for this very reason.

#5

Submitted by Andy Kriebel (not verified) on

Daniel,

I couldn't disagree more. If you want to represent parts to whole AND be able to compare the categorical data, use a bar chart that shows percent of total.

I'm shocked that someone from Tableau would ever recommend a pie chart. Actually, I'd say I'm disappointed.

#6

Submitted by Jeff Johnson (not verified) on

THANK YOU! I've been trying to convince people of this for a long time. We are taught how to read a pie chart from primary school, and taught that the value is represented by angle, not area. The basic pie chart takes advantage of a taught skill, and that makes it worth using for that purpose.

Andy, your point is precisely what's at issue here: Daniel's example is a case where you only want to compare parts to the whole; there's no second dimension represented. When the radius is used to add a second dimension (either within the pie or in a second comparison chart) things get very confusing very quickly, and 3-D pies distort the angles that we want people to read. You're right that those are bad uses of pies, but those aren't the use we're considering here.

#7

Submitted by Andy K. on

Hi Jeff. I know exactly the point Daniel is trying to make and it's still better to represent parts to whole with bars than pies. The whole idea is to make comparisons quickly. It's always quicker to compare lengths of bars over slices of pies.

Also, his pie is in the wrong order. People read pie charts like they read a clock, therefore the slices should be arranged largest to smallest starting at 0 degrees moving clockwise.

You also make another good point...that we're taught to use pie charts in primary school. We should be teaching proper data display at the beginning so that bar charts are the norm, not pies. I would suspect that a great majority of teachers are ignorant to data viz best practices, and continuing to teach suboptimal methods only perpetuates the problem.

#8

Submitted by Daniel H. on

Andy,

I do not recommend pies, and I haven't done so here. I don't even really use them myself (as is clear since I have the slices in the wrong order, which conveniently also invokes long-term memory of a clock). I've called them a case of anti-best practices and pointed out their flaws.

The point of this post was to ask about a trade-off that I never considered, in any circumstance, and that is the power of pre-association and long-term memory. Maybe in many cases there is no trade-off, and that long-term memory is irrelevant. Is it possible though that there are people who cannot understand that a single-stacked bar represents parts of a whole without some serious explanation? Might it be worth the precision to first use a pie and follow up with, "In the future, I'm going to use this bar, which represents the same thing"?

I'm just asking a question. It's a potential reason. I never qualified it as a good or bad one.

#9

Submitted by TimCost (not verified) on

I want to go on the record to say that I don't advocate the use of pie charts as a best practice, but neither would I go as far as to say that they should never be used. The world isn't binary. There are times when a pie chart can be useful. If a pie helps tell the story of the data in a way that doesn't distort the truth or cause confusion I have no problem using a pie. It's rare, probably very rare ... but it happens.

#10

Submitted by Andy K. on

I apologize David. I misunderstood the tone of your blog post. Honestly, I'm disappointed that Tableau gave into the pressure of their customers and included them in the first place. It'd be better to spend resources on a desktop version that runs on a Mac rather than including charts that we all know will get misused and abused.

I do appreciate the discussion, even if it does get me fired up. :-)

#11

Submitted by Andy Cotgreave (not verified) on

I agree with Dan - pies have their place. In strict terms, it's not about "preattentive" features, but about the schema that we develop as we learn the world around us. Pies do not need to be explained and thus can be shown to a wide, untrained audience in order to give them understanding quickly.

Remember that every visualisation is a compromise - when you draw a heatmap (eg your Birthday rank chart) you're making a decision about what you're missing out, too. Sure, you show relative colour, but I can't see any individual value without resorting to the tooltip.

Here's a dry wikipedia page about schema:
http://en.wikipedia.org/wiki/Schema_(psychology)

#12

Submitted by Dan Murray (not verified) on

I would think the big blue slice was Facebook in in the pie. ;-). Also..the biggest problem with pies are the sliver slices...the smaller outliers are just too small to see...so the pinwheel effect starts with the related cramped labeling. At least with a bar you can label the tiny nubs more easily. The scatter plot is conveying the most meaning...very clearly.

#13

Submitted by Ed (not verified) on

Daniel,

You can only make this point because your bar chart graph is pretty lame. Two reasons for this:
- you use different colors for encoding basically the same information, which distracts the reader and makes the graph confusing. Using the same color for the same measure (sales) would be more appropriate ;
- as Andy already said, you should not show sales as amounts, but as relative percentages
- a third improvement could be to display the absciss on top of the bar chart, instead of below (might be easier to read).
Why don't you (re)read Stephen Few's "Show me the numbers" ? The demonstration is irrefutable.
I am also shocked that a Tableau blogger makes such claims, plus with poorly designed graphs. Pie charts might have some effectiveness when used on maps, but this is the only somewhat useful use I can think of for them.

#14

Submitted by Daniel H. on

Ed,

I am not making any claims, points, or other recommendations of any kind. My goal with this post is to ask a question and start a discussion--not about visual best practices (my opinion on pies and its flaws are evident in the post), but about audience.

If you're presenting to an audience who does not understand other ways of visualizing part-to-whole relationships besides pies, is it worth it then, to forego best practices and precision for the sake of your audience understanding the data in a familiar manner? If so, why, and how do you decide? And if no, never do such a thing, why?

We can discuss my use of double-encoding and nominal sales at a later date (feel free to e-mail me), though in general I agree with your points there. For what it's worth though, here is one claim I will make: I think both of those pie alternatives, in terms of data visualization, are still superior than that pie chart. But that's a strawman to the actual discussion I'm trying to ask about.

#15

Submitted by Gregory L. on

Hi Daniel,

As you can see, pie charts are definitely a passionate topic amongst the Tableau community. Personally, I keep hoping that Tableau will add 3D, animation, reflection, bevels, and some rainbow gradients specifically for Pies in Version 8...LOL (c'mon Kriebel, walk away from the light, come back to us...)

OK, all kidding aside, someone once told me, if your audience wants you to "build stupid, you build stupid"...but in that compromise, you must also take the time to build out a better way to articulate the information and explain why. So, when this occurs, build a pie chart, make it the best you can, and follow the principles which will aid to inform your customers even with an inferior chart. Then you can build it your way, don't tell them you're going to do it, don't tell them it's better than a pie, just show it to them with 30 seconds of explanation, they will either get it, or they won't, and no amount of convincing will help you. You can grumble to us other DataViz geeks about why it was sooooo wrong later :)

Getting to the final few sentences of your post, it reminds me of another friend of the DataViz community, Edward Tufte who recommends that you trust the intelligence of your audience to understand the material you're presenting, in all its glorious complexity and detail. In other words, never present to the lowest common denominator in the room!!

To answer your question Daniel, if you feel that your audience needs it I would not hesitate to use a pie chart if it is the right tool (stay with me folks). In your example, it is the right tool for 1 reason, education. Instant recognition of the pie allows you to quickly transition your audience to the bar chart described by Andy Kriebel, further allowing you to discard pies from the balance of the work.

Enjoy...

OK everybody, fire away!!

#16

Submitted by Daniel H. on

Hi Gregory,

Thanks for your comment. That's an interesting way to go about it, build what your audience wants, but use it as an educational springboard to move into stronger visualizations afterwards. There'd be a tension between your audience continuing to press for how they want to see visuals, but as you said, perhaps there needs to be an element of trusting that your audience will also see the value in non-pie chart presentations.

#17

Submitted by Jeffrey Shaffer (not verified) on

I'm in 100% agreement with Andy Kriebel on this one (congrats on the new job by the way). The debate goes on, but it is clear that there are better ways to show part-to-whole relationships than using a pie chart. Yes, people are taught how to visualize data with this tool at a young age, but that doesn't mean it's the best tool and should continue to be used. If you want to simply answer "which is more?" and it's a large gap then yes, a pie chart will answer that question (and so will a bar chart). If you are trying to be more specific in detail about a comparison then a bar chart is so much better and if you are trying to show the difference between 18% and 16% then a pie chart is useless. I can understand why Tableau offers it because a user may require this chart in a spec giving the designer no choice. However, I would like to see Tableau add a warning box before allowing users to use it!

#18

Submitted by chris pagel (not verified) on

I disagree that this use of pie charts makes sense because people have "preattentive" associations. Put more simply, just because people are used to a bad visualization does not mean it should continue to be used, especially by professionals. Let's take the time to use the bar chart and explain when asked why we are using it. Change is good.

#19

Submitted by Peter Shwarzinger (not verified) on

I agree with Mr. Pagel, everyone has a different point of view.

#20

Submitted by Dan M. on

I don't consider pie charts to be evil. But they are very abused. The best use case I've found for them is when the following conditions exist:

1. I have 2 to 4 dimensions.
2. My dashboard space is limited.
3. I will not have any tiny slivers in the pie.
4. Use the pie for filter actions (space efficient).

non-humans click here