Pie chart haters: I'm listening
October 16, 2016 8:53 PM Subscribe
I understand why pie charts give dataviz folks agita, and I certainly get that there are many bad applications for pie charts. But I'm having a hard time with the maxim that there are no good applications for pie charts. In fact, I think I have a scenario where not just a pie chart but an exploding pie chart is ideal ... but if someone wants to talk me out of it, I'm listening.
I need to represent a situation where we have finite money that can be put in one of two buckets -- or, it can be in limbo, i.e. not yet assigned to either bucket. The major takeaway of this chart is the relative proportion of "unbucketed" money. I think a pie chart that emphasized just that piece of the pie communicates this very effectively.
A three-bar bar chart -- A, B, U -- doesn't seem as effective. A stacked bar chart with two stacks -- A+B and U -- doesn't seem as effective. A triple-stacked bar chart -- A+B+U -- is just a vertical pie chart, and it's not clear to me that this relative proportion is more easily understood when linear and not radial.
... But I understand this is a subject about which folks that care have really strong feelings. I want to do the best thing, and you don't think this is it, I'd like to hear why.
I need to represent a situation where we have finite money that can be put in one of two buckets -- or, it can be in limbo, i.e. not yet assigned to either bucket. The major takeaway of this chart is the relative proportion of "unbucketed" money. I think a pie chart that emphasized just that piece of the pie communicates this very effectively.
A three-bar bar chart -- A, B, U -- doesn't seem as effective. A stacked bar chart with two stacks -- A+B and U -- doesn't seem as effective. A triple-stacked bar chart -- A+B+U -- is just a vertical pie chart, and it's not clear to me that this relative proportion is more easily understood when linear and not radial.
... But I understand this is a subject about which folks that care have really strong feelings. I want to do the best thing, and you don't think this is it, I'd like to hear why.
Pie charts are primarily disliked because it's hard to compare the size of individual pieces. If you're showing what is effectively one or two slices of the pie (vs. the unallocated empty pie tin), then it's less of a big deal.
posted by mcav at 9:15 PM on October 16, 2016 [1 favorite]
posted by mcav at 9:15 PM on October 16, 2016 [1 favorite]
second If only I had a penguin on all points, except sometimes, very occasionally, I do go WHOA that's like a THIRD when confronted with a visual explanation.
posted by wym at 9:29 PM on October 16, 2016 [4 favorites]
posted by wym at 9:29 PM on October 16, 2016 [4 favorites]
You're trying to understand proportions, then a pie chart is exactly the kind of chart you want.
Except everybody thinks that's what pie charts are good at, but they're actually very bad at. Humans seem to be lousy at comparing the relative sizes of wedges- even small numbers of wedges. They're good at comparing the heights of similarly sized rectangles.
A three-bar bar chart -- A, B, U -- doesn't seem as effective.
Define 'effective'. If you mean 'makes people what I want them to see', then fine. But if you mean 'gives people are truly accurate sense of the relative sizes of different amounts', then no.
I get the sense you're trying to make something look bigger or smaller or more important than it really is in purely numerical terms, and the fact that this doesn't pop out in a bar chart but does in a pie chart is another argument against pie charts, not for them.
posted by obiwanwasabi at 10:00 PM on October 16, 2016 [17 favorites]
Except everybody thinks that's what pie charts are good at, but they're actually very bad at. Humans seem to be lousy at comparing the relative sizes of wedges- even small numbers of wedges. They're good at comparing the heights of similarly sized rectangles.
A three-bar bar chart -- A, B, U -- doesn't seem as effective.
Define 'effective'. If you mean 'makes people what I want them to see', then fine. But if you mean 'gives people are truly accurate sense of the relative sizes of different amounts', then no.
I get the sense you're trying to make something look bigger or smaller or more important than it really is in purely numerical terms, and the fact that this doesn't pop out in a bar chart but does in a pie chart is another argument against pie charts, not for them.
posted by obiwanwasabi at 10:00 PM on October 16, 2016 [17 favorites]
(This isn't to say people can't abuse bar charts as well - for example, exaggerated y axis, use of colour and labels.)
posted by obiwanwasabi at 10:01 PM on October 16, 2016
posted by obiwanwasabi at 10:01 PM on October 16, 2016
I always thought pie charts were the place to go when you want to give the quantities for certain categories, but don't want to spell out the exact percentages. That way, you're not insulting your reader by both telling them the percentage and showing it to them in a picture format.
posted by klausman at 10:29 PM on October 16, 2016
posted by klausman at 10:29 PM on October 16, 2016
A three-bar bar chart -- A, B, U -- doesn't seem as effective.
There are, of course, people who define a chart's effectiveness in terms of what it masks rather than what it shows- if A is 80%, B is 5% and U is 15% - and if you want to play down the difference between the size of B and U - then a pie chart can be a good way of doing this - even more so when there are more small categories.
I think pie charts work best when you could mimic them with an actual pie and still convey your message at a glance. It sounds like that might be a good use case for you - but even with a small number of criteria a pie chart could mask small differences.
posted by rongorongo at 10:55 PM on October 16, 2016 [1 favorite]
There are, of course, people who define a chart's effectiveness in terms of what it masks rather than what it shows- if A is 80%, B is 5% and U is 15% - and if you want to play down the difference between the size of B and U - then a pie chart can be a good way of doing this - even more so when there are more small categories.
I think pie charts work best when you could mimic them with an actual pie and still convey your message at a glance. It sounds like that might be a good use case for you - but even with a small number of criteria a pie chart could mask small differences.
posted by rongorongo at 10:55 PM on October 16, 2016 [1 favorite]
Does it even make sense to compare U to A, B, or A+B at all? Could it be more of an hourglass or thermometer metaphor where U is the amount remaining from the total amount of money possible to allocate? If A vs B is important to represent on its own, a pie would be fine I think for just two things. You know, so you get to pie it up at least a little bit.
posted by rhizome at 11:39 PM on October 16, 2016
posted by rhizome at 11:39 PM on October 16, 2016
I don't have specific suggestions for your question, but in general, the Junk Charts Trifecta Checkup made by Kaiser Fung has served me well. It is worth a read. Also, here are several examples of what not to do with regard to pie charts.
posted by thewildgreen at 11:44 PM on October 16, 2016 [3 favorites]
posted by thewildgreen at 11:44 PM on October 16, 2016 [3 favorites]
I understand this is a subject about which folks that care have really strong feelings.Remember, the reason that dataviz folks have feelings about pie charts is that they've been shown in experiment to obscure the information in them. If you care about conveying information then that's a problem. If you don't care about conveying the information and just want to say something strongly independent of what the data says, then picking the right visualization doesn't matter because your goal isn't visualizing the data, is it?
I would try a side-by-side pair of three-bar bar charts (so, ABU with one allocation, then ABU with a different allocation).
.. .. .. ..
.. .. .. .. .. ..
A1 B1 U1 - A2 B2 U2
A A
A A
U U
B U
B B
Either way, this is few enough numbers that just presenting a single chart of the numbers is probably best anyway:
A1, A2
B1, B2
U1, U2
T1, T2 (showing the constant total)
Highlight U2 in bold red and you should be good to go. That's the presentation that I'd bet would be most clear while staying most true to the data.
posted by daveliepmann at 12:58 AM on October 17, 2016
If your goal is to cleanly inform people of information as simply as possible, then a bar chart is what you want. If you want to wow a non technical audience with a specific figure and don't mind that they may get an inaccurate picture of the truth, then a pie chart will probably work fine.
That sounds rather pointed and snarky, but in some contexts you may want to do the second one. Just be aware that that is what you'll be doing.
posted by Cannon Fodder at 2:40 AM on October 17, 2016
That sounds rather pointed and snarky, but in some contexts you may want to do the second one. Just be aware that that is what you'll be doing.
posted by Cannon Fodder at 2:40 AM on October 17, 2016
You might be interested in reading "Pie charts: seldom or never?"
posted by pemberkins at 4:47 AM on October 17, 2016 [1 favorite]
posted by pemberkins at 4:47 AM on October 17, 2016 [1 favorite]
True story: I am a full time data visualization person at a research university.
I agree with most of the critiques of pie charts given above. I am not wholly opposed to pie charts as long as you meet two simple criteria:
1. You have at most 4 segments of your pie.
2. It is not necessary for your audience to be able to extract the raw counts or proportions from your chart.
If you don't meet either of these criteria, then I recommend a bar chart.
If you don't want a bar chart, you can also do a treemap, which can sometimes give you the option to label each section. Wikipedia states that treemaps require hierarchical data, but you can do it with pie-style data easily.
I like to use pie charts if you're trying to make a point that one or two or your pie segments (again as long as you have 4 or fewer) is either consuming waaaay too much of the pie or waaay too little.
posted by rachelpapers at 6:18 AM on October 17, 2016 [3 favorites]
I agree with most of the critiques of pie charts given above. I am not wholly opposed to pie charts as long as you meet two simple criteria:
1. You have at most 4 segments of your pie.
2. It is not necessary for your audience to be able to extract the raw counts or proportions from your chart.
If you don't meet either of these criteria, then I recommend a bar chart.
If you don't want a bar chart, you can also do a treemap, which can sometimes give you the option to label each section. Wikipedia states that treemaps require hierarchical data, but you can do it with pie-style data easily.
I like to use pie charts if you're trying to make a point that one or two or your pie segments (again as long as you have 4 or fewer) is either consuming waaaay too much of the pie or waaay too little.
posted by rachelpapers at 6:18 AM on October 17, 2016 [3 favorites]
I agree with rachelpapers with one extra criterion: the sum of your data is naturally 100%. Ie, you don't care about the size of the pie itself, just the proportions of the pieces. This rule becomes particularly important when presenting multiple pie charts together. Pie charts emphasize the proportions of the categories, not their absolute magnitude.
I agree with obiwanwasabi that human perception of wedge size can be tricky. Some folks recommend a donut chart instead, a pie chart with a blank hole in the middle. You can stuff text labels in the hole if you want. Some folks suggest you can nest a second donut chart inside the first; I find that terribly confusing in most cases.
If you cut a donut chart and unroll it, it becomes a stacked bar chart. So there's a continuum here from pie to donut to bars.
posted by Nelson at 7:02 AM on October 17, 2016 [1 favorite]
I agree with obiwanwasabi that human perception of wedge size can be tricky. Some folks recommend a donut chart instead, a pie chart with a blank hole in the middle. You can stuff text labels in the hole if you want. Some folks suggest you can nest a second donut chart inside the first; I find that terribly confusing in most cases.
If you cut a donut chart and unroll it, it becomes a stacked bar chart. So there's a continuum here from pie to donut to bars.
posted by Nelson at 7:02 AM on October 17, 2016 [1 favorite]
Whatever the kind of chart, the legend, labels, color, etc are very important. Personally, I would not be greatly concerned with the "hard to compare the size of individual pieces" problem if the percentages were clear on the chart.
In the specific situation that you describe, I'm not sure a chart is the best way to go. I knew one guy who probably would have suggested cartoons of a couple of pails for the two destination buckets, a stack of bills for the unallocated funds, some arrows and a cartoon of a guy looking confused. All appropriately labeled, of course.
posted by SemiSalt at 7:43 AM on October 17, 2016
In the specific situation that you describe, I'm not sure a chart is the best way to go. I knew one guy who probably would have suggested cartoons of a couple of pails for the two destination buckets, a stack of bills for the unallocated funds, some arrows and a cartoon of a guy looking confused. All appropriately labeled, of course.
posted by SemiSalt at 7:43 AM on October 17, 2016
Waterfall charts offer another kind of alternative. They're not perfect either, but can be better for showing multiple small entries with subgroupings. I find then useful to walk through a logic model sequentially. They can be useful to tell a story left to right, for example.
Pie charts are tricky. In my experience I agree with most of the above. People do not read areas well as proportion, so the charts provide weak explanation. People do see linear distance as proportionality much better, which is why bar charts and their variants are often recommended instead. That said, pie's aren't the worst choice if you have to divide a known quantity into a small (3-5) number of sub groups, the majority of which are substantial (say >10% of the total). And never, ever, compare pie charts with each other for absolute counts (e.g. using chart size to show growth in a budget from year to year).
Because of the area/length problem human perception has, I don't find treemap style graphs much more helpful than their round counterparts. Just my opinion, but there it is.
posted by bonehead at 8:21 AM on October 17, 2016
Pie charts are tricky. In my experience I agree with most of the above. People do not read areas well as proportion, so the charts provide weak explanation. People do see linear distance as proportionality much better, which is why bar charts and their variants are often recommended instead. That said, pie's aren't the worst choice if you have to divide a known quantity into a small (3-5) number of sub groups, the majority of which are substantial (say >10% of the total). And never, ever, compare pie charts with each other for absolute counts (e.g. using chart size to show growth in a budget from year to year).
Because of the area/length problem human perception has, I don't find treemap style graphs much more helpful than their round counterparts. Just my opinion, but there it is.
posted by bonehead at 8:21 AM on October 17, 2016
In this case, for example, a waterfall could show costed expenditures then the spent subtotal then the unspent portion (ideally in another colour to make it pop) then the total.
You can then walk your audience through that noting first the individual assigned monies (A and B), showing how that combines to the total assigned (A+B), then here's the "unbucketed" funds, U, which together add up to the total amount, T=A+B+U. This kind of implies the the "unbucketed" funds are a budget line item for future unforeseen costs, say, to provide organizational flexibility.
Alternatively, depending on how you want to present the info, do the individuals (A&B), the total assigned (A+B), then the total amount (T), with the "unbucketed" monies (-U) falling out as the difference between the two. This emphasizes that the unbucketed funds are a gap between planned projects and the total.
Same story, but presented as a "planned unallocated" vs. "here we are now" kind of scenarios.
posted by bonehead at 8:29 AM on October 17, 2016
You can then walk your audience through that noting first the individual assigned monies (A and B), showing how that combines to the total assigned (A+B), then here's the "unbucketed" funds, U, which together add up to the total amount, T=A+B+U. This kind of implies the the "unbucketed" funds are a budget line item for future unforeseen costs, say, to provide organizational flexibility.
Alternatively, depending on how you want to present the info, do the individuals (A&B), the total assigned (A+B), then the total amount (T), with the "unbucketed" monies (-U) falling out as the difference between the two. This emphasizes that the unbucketed funds are a gap between planned projects and the total.
Same story, but presented as a "planned unallocated" vs. "here we are now" kind of scenarios.
posted by bonehead at 8:29 AM on October 17, 2016
Response by poster: Thanks for all of the great answers, guys. Plenty to think about.
posted by blueshammer at 10:45 AM on October 19, 2016
posted by blueshammer at 10:45 AM on October 19, 2016
This thread is closed to new comments.
I would say I tend to dislike pie charts because they don't typically add much, especially in the case where there are only a few categories. I mean pie chart wedges represent proportions/percentages. Proportions and percentages are already pretty damn clear. If you see 33% of the money has no assigned purpose, I don't need a pie chart to help me grasp what that means. It's not like the I look at the pie chart and go "WHOA, that's like a third... I never thought of it that way!"
If there are a lot of categories then maybe it can give you a little more visual insight ("Look, that big category is about the size of those ten categories combined!") that you might not get from just reading the percentages. But if all you want to do is make clear what proportion of money is unbucketed, I don't see what a chart gives you beyond "33% is unbucketed."
So yeah, if you must have a chart, then you want a pie chart. But must you have a chart, really?
posted by If only I had a penguin... at 9:01 PM on October 16, 2016 [1 favorite]