Pie charts, confirmation bias and generic dataviz advice
Every self-respecting dataviz educator has an opinion about pie charts, just look at the list:
And rightfully so because there is probably not another topic which is so debated and discussed in data visualisation. I expect the dark vs white background topic to pick it up at some point, but even that one will most likely never compete with this good ol’ pie chart debate.
I also want to be a self-respecting dataviz educator, so I will write an article about pie charts. But instead of just explaining why they are good or bad, or how to use them, I will use it as an example of generic dataviz advice, highlight why they might be poor, and whether you should still follow them.
So, what’s with the pie
The obvious problem – that I hope both sides of the debate will agree – that pie chart is heavily overused. There are two main reasons for that.
First, the pie chart is available as one of the default options in several charting software packages (notoriously Excel), like it is one of “basic” charts. If you don’t think what you’re doing, you will probably end up using a pie chart.
Second – it is round, it looks nice. Bars are pieces of bricks, tables are boring grids, and pies come straight from heaven to infuse your powerpoints and dashboards with cuteness and harmony. I am convinced that if Sankey charts were easy to make – they would be as overused as pies because they are also round and nice.
The debate revolves around how much overused are those pies.
One side tells that every use of a pie chart is a crime, pie chart usage should be reduced to zero. The other side tells, that if we reduce pie chart usage by 80% we’ll be good.
I see more problem with the debate than the actual pie chat. It is a classical example of confirmation bias – people have opinions, and then they look for confirmation for that opinion. A few other examples of similar discussions:
- what is the scientifically correct way to hang toilet paper roll
- what is the scientifically correct way to start your worksheet – A1 or B2
When people have opinion first, and then look for confirmation, they end up with one of the poor arguments which pop up again and again.
1. Here is an example, see?
First such argument is a constructed example which exactly supports the claim. Probably the most famous example, presented below, is made by a Wikipedia contributor. The pies are outrageously bad, and the data is fine-tuned to precisely support the claim against them. You can find many cases where poorly designed pie charts compared to well-designed bar charts.
That is just wrong – if we create a poorly designed bar chart and compare it to a well-designed pie chart, the result might give a less obvious impression.
Even Stephen Few himself acknowledged an example when pie charts are superior to bar charts. We might want to see if several parts together are bigger than the rest, and bars would struggle to provide such insight. In the example below only with the pie chart, it becomes obvious that stars starting with letter A take more than half of some metric.
There are variations of this example – we might want to see how many parts do we need to achieve 50%. Or finally – we might want to compare two parts with two or three other parts. With all seriousness, I dare you to declare that bar charts are still better at these cases!
And with all seriousness, people still do. They suggest adding axis grid to the bar chart (oh so you want to add axis now) or all other crazy solutions (about them later) just to avoid dreaded pie chart. Because pie chart just has to be avoided at all costs. Even if that cost is readability.
Provided examples only show isolated cases, sometimes in favour of a pie chart and sometimes in favour of a bar chart. They do not prove that one type of chart is superior to another in the whole universum of possibilities, only in those isolated cases.
2. Humans are bad at judging angles
Another gem is claiming that humans are bad at judging angles.
The LinkedIn post below summarizes everything nicely, someone claims that humans are bad at judging angles, and then claims that in a pie chart you cannot distinguish between 20% and 25%. While obviously this is exactly what is easy in pie charts – seeing where the value hits 25%, 50%, and 75%.
Saying that you cannot use a pie chart because humans are bad at judging angles is like saying that people cannot use analogue clock because humans are bad at judging angles.
Just imagine if a time would be represented as a bar, showing how much of the day have passed. Can you guess the hour out of these images?
What if we show the same hours as angles?
Are you suddenly better at judging angles than bars now? Humans are good enough at judging angles, I saw people wearing fancy branded watches without numbers. Another clear case for angles is a speedometer, where you are supposed to understand how fast are you going instantly. Angle is an additional clue beside the position of the arrow, which speeds up your judgement.
There is an argument that people who used to play games a lot are used to seeing energy or ammo levels as bars. It is easy for them to get whether the bar is above or below the middle. Furthermore, we use analogue clocks less and less often, replacing them by digital ones. People might be used to bars more than reading angles nowadays.
But then, by the same argument, we can say that if we use what people are used to, then we should use pie charts because people are used to seeing pie charts! Haven’t we proved the case for pie charts by insisting on bar charts?
We can make an even more ridiculous case by claiming that the phrase “you cannot use pie chart because humans are bad at judging angles” is like saying that “you cannot use pie chart because humans cannot hold breath no more than a few minutes”. How is that even related, you may ask? Exactly.
It seems at the first glance that to determine the size of the slice we have to compare angles. But then there is the donut chart which misses the middle part, and it is ridiculously harder to determine the angle. However, it is not much more difficult to determine the value. What is happening here?
The answer is that a pie chart is a complex beast, definitely not a basic chart. Pie chart and donut chart encode their value in combination of angle, area size and arc length as claimed by Robert Kosara in his research. In case the angle is missing – area and arc length can still be used. That is why ability to read angle has not much to do with the ability to read a pie chart.
This is an example when one claim which is mostly true (humans are not terrific at reading angles, I must agree) used as a supporting argument for another claim, but there is no solid connection between those two claims. That is just lazy argument. Conspiracy theories are developed like that.
3. Silly alternatives
One way to prove that something is not good and should be avoided is to offer a better alternative. And here comes all the silly alternatives to pie charts.
Some suggested alternatives are good, but then there is a treemap. Treemap is great for showing complex hierarchies when we don’t need to know exact values because it is quite challenging to guess the values from irregular shapes. And still, it is suggested as an alternative to pie charts because in a pie chart it is quite challenging to guess the values. I’m not saying that pie chart is better. I’m saying that treemap is not better.
Further, the Pareto chart suggestion pops up occasionally, solving the problem mentioned before – how to see whether two or more values add up to a certain percent? The goal here is not to provide a simpler, more elegant and more readable solution, the goal is to get rid of pie chart. Are people really able to understand the Pareto chart? Maybe they are, but for me, this seems like using a spaceship to go to a grocery store – a really complex and difficult to read thing is used instead of a pie chart because a pie chart has to be avoided at all costs, even if that cost is readability. Again – Pareto chart is not better than pie chart.
Finally, we have waterfall charts. They are suggested because pie charts cannot show negative numbers. Here I must agree, they cannot and in such cases we should use waterfall charts or other alternatives. But to use waterfall charts for all part to whole visualisations is not a good practice because data is represented as floating bricks – it is hard to compare values which do not have a common baseline. Try judging in the image below – which has bigger value – Earth or Venus? While it is also hard on pie chart – in both cases we rely on sorting. Waterfall is just not better.
In conclusion, suggesting an alternative chart which does worse than the original chart is not an argument against using the original chart. It’s the opposite.
4. Space argument
There is one more common argument, I call it the space argument. Bar charts take less space than a pie chart, which takes a square and has to be big to be readable. If you don’t have enough squared space in your canvas, it is better to use bars, but what if you do? What if the area you have is exactly squared, and one bar in your chart takes 75% of the value? It might be more readable if we just make it a pie chart.
Once more, I’m not saying that pie charts are better than bars. I’m saying that if you have a square, you can fit a circle inside. And the big amount of ink to draw one might be even better for readability.
What is one common problem with all this advice and arguments?
I guess we agree that pie charts are overused, their use cases are limited, but they are literally everywhere and often used wrongly. The question is what to do about that?
The answer lies in understanding the people.
There are people who are just starting their data visualisation journey, maybe they were asked to do a PowerPoint for the first time, or they usually do data engineering, and now they have to also create a dashboard. For them, all those nuances about choosing the right colours, choosing the right number of slices, providing intuitive labelling just will be too hard to tackle. I entirely agree that for such people the advice should be – avoid pie charts at all costs, and I do give such advice myself. Simple advice which guarantees good enough result.
There are people who are experienced presentation makers, or dashboard developers, they know the basics, and they are looking for better more engaging ways to represent data. And then they get a request to add a pie charts somewhere in the dashboard. Should they stubbornly resist? Or should they look for profound explanations on all the best-practices on using a pie chart? There they will find more advanced advice which guarantees good enough result.
Giving generic advice while ignoring the audience will get you weird results. If you say that pie charts are OK – beginners will abuse them in outrageous ways. If you say that pie charts are BAD – experts will attack you for not being completely precisely right.
Solution
Solution is this simple mantra: If you know what you’re doing – do it. If you’re not sure – follow the best practice.
If you can explain why are you adding this additional colour, this additional chartjunk or this additional pie chart, and you know that you are taking risks here – do it. You won’t be surprised if there will be backlash from dataviz gurus, some people might not like your work or find it confusing. But maybe you’re up to something extraordinary here.
If you are not sure – “I will use this chart because it is cool” – please don’t. All things in dataviz are about deliberately encoding, being able to explain very well why you are using a certain shape or graphical element to display a number. If you cannot explain – stick to basics, stick to some instruction, or stick to your favourite data visualisation expert.
And you will be good.
Probably, my final advice about pie charts is that it is safe to never ever use a pie chart. Same with Comic Sans, dashboards with dark background, travelling to other countries, trekking in mountains, diving, skiing, and driving a car.
If you know what you’re doing – do it. If you’re not sure – follow the best practice.