What is the best way to visualize percentage information (a set of numbers that add up to 100%)? The obvious answer would be a pie chart. You see it everywhere right. But is it that simple?
Pie charts: why are they so common and so bad?
I was aware through one of my stats undergrad class that pie charts are not well seen in the scientific community in general. In the visualization community , it’s something you won’t touch with a 10 foot pole! Quoting Edward Tufte from “The Visual Display of Quantitative Information” :
Tables are preferable to graphics for many small data sets. A table is nearly always better than a dumb pie chart; the only thing worse than a pie chart is several of them, for then the viewer is asked to compare quantities located in spatial disarray both within and between pies, as in the heavily encoded example from an atlas. Given their low data-density and failure to order numbers along a visual dimension, pie charts should never be used.
These are quite strong words against pie charts. I totally agree that using several pie charts for comparison purposes is useless; there are much better charts that allows us to do that. But for a single comparison of percentages, why not?
It turns out that us humans are not that good at estimating angles but conversely we find rounded shapes attractive.
Silvia el al.[SIL09] do a user study to verify (or reverify) that humans are supposed to prefer curved shapes to one with lots of sharp angles. Their results confirm the long held beliefs that we prefer rounded corners. So we like pie charts as they are nicely rounded. Eventually to appeal to the general public, they are often used in the press.
One of the reason Tufte does not like pie charts is their “failure to order numbers along a visual dimension”. Being circular, we can’t apply the usual cartesian coordinates but they do make sense in the polar coordinate system. However, whether humans are any good at estimating angles has been the subject of lots of discussions for a long time with studies counter proving each other. Spence [Spe05] argues that pie charts are not that bad and points out mistakes in prior user studies. I’ll go with Kosara [Kos10] on that (because he has a larger sample size for his study) where in his study using Amazon’s mechanical turk, he find that people are not too good at estimating angles and stacked bars for that matter. Squared pie charts (coming to that down) are better. Spence also makes the point why bar charts are not good for representing percentages.
Let’s now take a look at the alternatives. All the charts below show exactly the same information.
These two above are basically variants of each other. The main difference of the doughnut chart is the hole in the middle. These two are usually our default choice for percentages. They look nice with their rounded shapes and show that the different components make up one whole. But it’ll take a brave man to estimate the values of each slice in here.
The bar chart is sometimes proposed as an alternative that I do not like. I don’t think it does a good job at showing percentages. With the pie chart, we know the different slices are part of a whole which is not the case for a bar chart.
The treemap and Waffle chart (no wikipedia link here: someone needs to create an article for it!!! Why not me??? Well …) look similar. Waffle charts are sometimes referred to as squared pie charts but they eliminate angles which is supposed to be the weak point of pie charts! I think the treemap is the prettier one of the two but the waffle chart has one advantage: a correctly built waffle chart is 10×10 squares where each square represents 1%. So it is fairly easy to quickly estimate which the quantity of each component. Also treemaps comes with baggage. They are often used to represent hierarchies and so users familiar with it might be wondering what the hierarchy is here.
So the elements that I want the visualization to have are:
- it does not mislead reader and allow them to correctly estimate the values
- it shows that the different components are parts of a whole and sum up to 100%
- be reasonably aesthetically pleasing (I don’t want a table here)
The most effective one according to me is the Waffle Chart. It solves the issue of poor angle estimation, does not have any other meaning (like representing a hierarchy that a treemap has) and it shows that the individual components are part of a whole (unlike a bar chart). Moreover, at a glance we can guess what the values of each component and we can even count the squares if we need more accuracy.
Some other nice links on the same subject:
- The worst chart in the world
- In defense of pie charts
- Why Tufte is flat out wrong about pie charts
- 9 ways to visualize proportions
- Square Pie Charts
- Waffle Charts (code to build waffle charts : not the square one but the code can probably be fixed for that)
- [Kos10] Robert Kosara and Caroline Ziemkiewicz. 2010. Do Mechanical Turks dream of square pie charts?. InProceedings of the 3rd BELIV’10 Workshop: BEyond time and errors: novel evaLuation methods for Information Visualization (BELIV ’10). ACM, New York, NY, USA, 63-70. DOI=10.1145/2110192.2110202 http://doi.acm.org/10.1145/2110192.2110202
- [Sil09] Silvia, P. J., & Barona, C. M. (2009). Do people prefer curved objects? Angularity, expertise, and aesthetic preference. Empirical Studies of the Arts, 27, 25-42.
- [Spe05] Ian Spence. 2005 No Humble Pie: The Origins and Usage of a Statistical Chart. JOURNAL OF EDUCATIONAL AND BEHAVIORAL STATISTICS 2005 30: 353. DOI: 10.3102/10769986030004353