"Pie Charts are a bad dataviz type and should not be used"
"You should never use pie charts, they are bad"
"Pie Charts they are work of the devil!"
The Myth's Origin
For years now the pie chart has been the whipping boy of the dataviz tool set. It gets a bad reputation because in a lot of cases it gets used badly. Whether it is this classic example from Fox News, where the segments don't actually add up to 100%
Or as in this one where you have 50 segments :
Then you have the pie chart in 3D, which distorts the areas making it impossible to compare the segment sizes.
And if that wasn't bad enough, some people then decide to blow the thing up!
Or where a multi segment pie chart is being used to look at changes over time
All of these examples show the pie being used in the wrong way, in the way its not designed to be used. In each case it plays to the weakness of the pie, and the fact that you are not good at comparing area sizes, so it makes it very hard to accurately compare area sizes. For this reason a bar chart is often sighted as being the better alternative.
Debunking The Myth :
The Myth of the pie chart being bad has built up over time, and like most myths there is an element of truth in it, but over the years that has been forgotten and instead replaced with an almost religious fanaticism. We have been told that humans are not good at comparing areas, we are better at comparing lengths. In his blog post https://eagereyes.org/criticism/in-defense-of-pie-charts Robert Kosara looks at some of the research and notes that :
They’re almost always considered the wrong choice by those supposedly in the know. But how do we know that this is true? What evidence do we have to support this? The truth is, not much. And when we start digging for proof, it turns out that pie charts are much better than we want to admit.
3 of the best dataviz experts have written about the problem with pie charts, Edward Tufte says “the only thing worse than a pie chart is several of them.” Stephen Few says “save the pies for dessert“ and Cole Nussbaumer says “Death to pie charts.”
One of the most common examples of the myth is that a bar chart is far superior to accurately compare values. Well that is correct, but what about if we just want a general feeling of the data, we don't need that accuracy.In Stephen Few's article he came pretty much firmly on the side of Pies = Bad, with one exception. One thing that a pie chart does very well is allowing you to combine more of more adjacent slices against another slice, ie how does A+B compare to C+D and E. This is much harder to see with a stacked bar. We can see this in practise below
Where we have a large number of segments its gets very hard to see the difference in size between them, if we want to get precise comparisons then the bar chart at the top right gives us that information more more accurately. But what if we want to know where 50% of the sales come from? Well with the pie chart we can see that easily, and that is highlighted below
We don't have to do any calculations, we don't need to look at any axis, all the work is done for us by the pie. We can see that 4 segments add up to around 50%. We could do the same with the stacked bar, but that still takes more work, unless all the segments are contiguous.
The other use of pie charts that gets short shrift is comparing them when you are looking for changes, for example over time, when a better choice is a bar or line chart. However, in this example below of a simple date set, which conveys the information quicker?
Again, we are not looking for accuracy, we are looking at the general overview of the data, and in that case, the pies give a better result
Any dataviz must be able to be interpreted and understood by your audience, if the viewer cannot understand the viz then it has failed its job of conveying information. Its a challenge with some of the more sophisticated chart types out there, you end up taking up more time and space explaining how to read the chart than actually reading the data itself. This is the Pie Charts greatest strength.
Everyone knows how to read a pie chart.
You are an expert at reading a pie chart, no-one has to tell you that its showing a part to whole relationship, that the angle determines the segment size and therefore its contrubution to the whole. And the reason for this is very simple. Since your very first birthday, you have been exposed to a pie chart in the form of your birthday cake. And you know very well if your slice is smaller than everyone else's. You know that your slice is small slice of the overall cake, you know how many slices could be cut from that cake and how your slice is slightly smaller than that of your siblings. That inbuilt understanding of how to read the chart makes it one of, if not the most useful chart types in our toolbox for rapid interpretation by our audience.
I don't believe that there is a data viz type that is inherently bad. There are however, bad implementations of a dataviz type, and the pie is probably the biggest victim of that. At the right time, with the right dataset and the right content, the pie chart is a great chart to use. The problem for the pie, and the reason for the myth is that it has been abused and misused so badly that not anyone's reaction to seeing a pie chart is negative. When it comes to showing part to whole relationships though, its probably the best viz type to use. There are a few rules though that really should be followed to ensure make the most of the Pie strengths.
Keep the segments to a minimum. Its very difficult to compare sizes of slices together, thats not what the pie chart excels at. Any more than 5 and a pie isn't the right choice, you are much better of going for something like a bar chart so you can compare the lengths.
Never use a 3D pie chart. Adding 3D distorts the sizes of the segments, making an already difficult task night on impossible.
Segments at the front appear larger than those at rear, even though those segments might actually be the same size, or reversed.
Do all of the parts make up a meaningful whole? Does it make sense to talk about the data as segments?
Don't use pie charts to show changes over time, its incredibly hard to compare the size of a segment across different pie charts.
When used correctly Pie charts can be a good choice, they are easily understandable by the widest range of people without needed detailed explanations. They are however probably the easiest chart type to get wrong so extra care must be used with them.