PureSpectrum - Schedule A Demo
Qualtrics: Here to Help

Why Pie Charts are Better Than Bar Charts

The implications of the research studies used to criticize pie charts are greatly over-stated. And some are junk science.

By Tim Bock

Ok, ok, this blog’s title would be a bit more accurate if the word “often” appeared in the title.  In my defense, all the anti-pie chart trolling provoked me! Troll HQ, Wikipedia, writes “Statisticians generally regard pie charts as a poor method of displaying information“.  Ouch! And a curious error of logic hides here. Let me give you a hint: who hires statisticians to design visualizations?

Before I jump into the detail of my thesis, let me jump straight to some examples, as many people that hate pie charts really just hate ugly pie charts. Below I show both an ugly and an amazing pie chart. I am sure we can all agree that one of these deserves contempt. But, are they both so bad?

Back to the debate. The redoubtable Michael Friendly has written a 14 page treatise, Save the pies for dessert, denigrating the pie chart, in which he says:

I have read every research study that I could find that tested the effectiveness of pie charts versus other means of displaying quantitative data … and have found only one advantage that can confidently be attributed to pie charts. Unfortunately, this one strength is rarely if ever useful.

Despite this denigration, businesses use them all the time. Why? Is it that business people are dumb, and that they are all making the same mistake? No. It is not. The problem is that the implications of the research studies used to criticize pie charts are greatly over-stated. And some are junk science.

Compare the column chart of the same data. Yes, it is better than the ugly bar chart. It is, however, markedly inferior to a pretty pie chart.

A simple visual experiment demonstrates the power of the pie chart.

Pie charts are better than bar charts

Look at the two bars. How long are they? There is no way you can tell without labels. You cannot even tell their relativity without a ruler. If I were to tell you that the bar above was one-quarter the length of the one below, you may well believe me. Short of using a ruler, you will never know for sure. Now look at the pie chart on the right. It is clear that the missing slice is 25%. Not 27% and not 23%. Sure you cannot tell if it is 24.5%, or perhaps 25.3%, but you can readily see that it is very close to being precisely 25%.

Pie charts tap into our instinctive ability to assess proportions when we look at things. As a result, we should consider the pie chart whenever we need to communicate proportions. There are lots of situations where proportionality is key. For example, as we can all recognize a straight line, a pie chart showing voting preferences is the safest way to communicate whether a political party has a majority or not.

Our instinctive love of pies

Our ability to interpret proportions is hard-baked into our brains. Surviving on the savanna frequently required us to look at objects and assess proportionality. How much of the apple have we eaten? How much water do we have left in the gourd? How much of the cake is left? Evolution has given us the skill to assess proportions instinctively.

We continue to train this skill teaching fractions using pie charts. This is why in the example above you get to exactly 25%, as your brain reaches back to junior high fractions and geometry. Watches and clocks require the same skill, which is why some people use watches without numbers and ticks. Most importantly, we regularly practice these skills when dividing up a pizza.

So, the first great strength of the pie is that we are really good at reading them. Of course, it is lot easier to make a bad pie chart than a bad pizza. Consequently pie charts often get a bad rap.  The biggest problem with normal pie charts is the labels. You will see in the example below, that with a bit of love (from my colleague Michael Wang), this is a solvable problem. Nevertheless, the pie chart is still far from perfect, but this one makes it easy to see that there are many browsers out there, with Chrome 48.0 dominating the market.

If you want to play around with these examples, or plot your own data so that it looks like the examples in this blog, click here.

Sorting helps

If we sort, we end with something a whole lot better. Our brain can easily work out from this chart that two browser versions, Chrome 48 and IE11, make up more than half of the market in our data. Again, we can do this instinctively, as we can see that their combined shares are bigger than a semi-circle. The only way to get that from the comparable bar chart would be to add up all the numbers. The point of a visualization is to let the viewer see the patterns, not to provide numbers that they can then add up.  Thus, the pie chart wins hands down for data like this.

Even the brands that are too small to plot are taken care of. We end up with a beautiful visual effect as they fade into obscurity on the left-side. However, you can hover over them with your mouse to see the tooltips, thus losing no information. In a bar chart, these would likely have been merged into an unhelpful “others” category.

Donuts are even better when you have lots of categories

We can also scoop out the middle of the pie to create a donut, and use the new-found space to add more labels, if we have the need.

As shown at the beginning, we can add even more clarity by nesting a pie chart within the donut. This final visualization allows us to quickly see that Chrome is more than half the market, and that the lion’s share of this is achieved by Chrome 48.

If you want to play around with these examples or plot your own data so that it looks like the examples in this blog, click here.

Originally posted here

Please share...

8 responses to “Why Pie Charts are Better Than Bar Charts

  1. Hi Tim, we can talk about this more in Sydney 🙂 But a better title might be there exist situations where a well-designed pie chart is the right chart?

    In terms of your pie and doughnut (which would be unreadable if produced using many software products) – might it be a better communication if the percentages in the body were rounded to whole numbers (e.g. Fireox and IE would both be 14%), and if the percentages of all the variants were not included in the labels, so the label “IE 10.0: 1.1%” becomes “IE 10” – if the purpose of the pie chart is to communicate visually the shape of the data, I would leave out the numbers. If the communication is supposed to be a source of numbers, I would probably use a table, perhaps with graphical design to help emphasise the numbers.

  2. As an example of data that does not work well in a pie chart. consider the Inverness, Nairn and Lochaber (UK Parliament constituency) and the 1992 General Election. The results were:
    LibDem 26.0%
    Labour 25.1%
    SNP 24.7%
    Conservative 22.6%
    Green 1.5%

    Since the UK uses a first-past-the-post electoral system, Russell Johnston (the LibDem) won with just 26% of the vote. In pie charts it is not easy to assess non-adjacent angles/areas of similar sizes – so the gap between, say Labour and Conservative would not be obvious. In this case the key messages (IMHO) are who won, that 4 parties were close, and the order between the parties (the order matters because that is often the basis for tactical squeeze messages at the next election – for examples the LibDems would target Conservative voters with a message that the only way to avoid a leftwing candidate winning would be to vote LibDem.

    In this case a series of columns would illustrate 4 nearly equal columns, and show the gaps – indeed this is a point you made in a video for me a few years ago 🙂

  3. Nice piece. But it is worth taking a look at the graphical perception and psychophysics research done at Bell Labs and elsewhere. It provides the empirical evidence on the inferiority of pies with respect to quantitative judgements.

  4. Research in visual perception indicates that people are much better at comparing lengths than comparing angles, which suggests that bar graphs are always superior to pie charts.

    Donut charts are even harder because the actually remove the angle information, making you rely on accurately judging the area to understand it, which is even harder visually than judging angle. Yes, you get more room for labels, but you lose the point of having a visualization – for people to understand the story in your data without relying on the labels. If you have to rely on labels you might as well just have a table.

    The donut over the pie chart is very challenging to interpret because each item now has multiple % associated with it.

  5. Completely useless debate. We MUST always decide what is the best chart for the information we want to convey. For some data sets it’s a pie, for others a bar (e.g. the political first-across-the-line example), or any other type that might be more useful. Bad use of chart types is still bad use no matter what chart type. So it’s to us to make sure we use the best available to bring our message across.

  6. Surely the comparison of your pie/donut combo to a straight forward column chart is a bit unfair. It would be more natural to compare to a stacked column which had the value for browser family total (e.g. 56.2%) and the individual versions segmented. This would allow you to easily compare overall browser family dominance and the relative proportions of versions within a browser (what I think are probably the two key things people want to know). That said I do find your visualisation fairly easy to read and have seen similar charts used effectively (e.g. for showing the usage of hard disk space).

    I agree it is easy to show dominance when it exists from a pie (but then it is usually clear from bar charts) and understand proportions when the segments are very close to multiples of one another (1-3x) or match very closely specific proportions (e.g.25%).

    In short, never say never but often there is just as good if not a better way to demonstrate and it depends very much on the actual data you are reporting (both in terms of type and results) so decisions need to be made on a case by case basis.

  7. Pie charts are always better to use and they depict a much better result of any statistical form. A bar chart simply shows you the frequency of occurrence of the event. The pie chart can depict much more as stated above.

  8. I always love me a pie chart. I doesn’t only sound delicious, it also provides you with a deliciously organized information. It’s much easier to see distributions via pie chart.

Leave a Reply to Ray Poynter Cancel reply