Why are pie charts evil?

The short answer is that they are not. However, like many things considered inherently evil, it is not the object itself that is evil, but how we use it.

One can argue that the poor reputation of the pie chart (and the misuse of many others) is a result of the technological advances in software-based charting tools. It is now extremely easy to take a set of data and flip through various chart types to see which one seems to look the best. But just because you can apply a chart type, doesn’t mean you should. Each distinctive chart type has its advantages and disadvantages. We will examine the pie chart in an effort to determine when a pie chart should be used, and when it should not.

So when is it not evil?

A pie chart is typically defined as:

A pie chart (or a circle chart) is a circular statistical graphic, which is divided into slices to illustrate numerical proportion. Wikipedia

The key word in this definition is “proportion”. Since a pie chart is comprised of a complete circle, it represents a whole, and the slices then define a portion of the whole.

This means that, in limited circumstances, a pie chart is actually very good at representing parts-to-whole comparisons. This is especially true when those parts are equal, or a common fraction such as one third, one quarter, etc.

Take a look at the example below:

Here we can easily see that each of the pie types consists of one third of the total pie sales. If that is all the information that is required, this pie chart presents a pretty clear view of this data. We are very familiar with even fractions in a pie shape, probably from eating so much pie. However, once more information is required, or more complexity is introduced, things start to crumble for our pie chart.

So really, why are they evil?

One of the main issues with reading data in a pie chart is due to our own evolution. During the eve our species, it was never especially important to be able to quickly determine the area of the field that contained the lion looking at us. It was, however, extremely important to be able to determine exactly how far away that lion was. This behavior led to something called “pre-attentive processing”. This is the instinctual ability to process information. Unfortunately one thing that humans are terrible at is estimating area.

Compare the two circles below:

If we specify that the smaller circle has a value of 1, what is the value of the larger circle? Go ahead, guess! Nope. The correct answer is 16.

This is a problem for pie charts since the value of each slice is defined by its area. If we take our initial pie and change the values a bit, you will start to see how it becomes difficult to estimate to a relatively high level of accuracy.

Without distinct proportions, it becomes difficult to make comparisons without including more information such as data labels. For example, what is the percentage of lemon pies in our pie sales? Very tough to tell.

Adding more series starts to highlight another issue that pies have, that of only being able to display a limited amount of data points. In our example above, we are adhering to one of the principles of good data visualization: visual simplicity. This is achieved by using a single colour and only varying the opacity to identify the different data points. However, by adding any more data points, it becomes increasingly difficult to differentiate between them.

With this many data points, and there are really only ten, differences are hard to make out. Additionally, the legend starts to become unwieldy. You might think we could solve this by using different colour hues. This can make things a little better, but still not optimal.

This is the default colouring from the Microsoft Chart Tool. Not only is it less aesthetically pleasing (I won’t even start on the unnecessary use of gradients), it is still difficult to differentiate specific data points. It would be very easy to confuse Lemon and Shepard’s pie without additional study of the legend. We also haven’t lost the issue of estimating area. Which sells more, Mud or Banana cream?

So how do we fight the evil?

There is no pancea for pie charts. Each set of data should be approached individually to determine what the information the data contains and what is the best way to represent it. But let’s take a look at our current example and find a better solution.

What we have been looking at essentially is a ranked list of sales of different types of pies. When looking at ranking, it is almost always best to use a bar chart.

We can now more easily see the ranking of the different pies, as well as estimate to a high degree the specific values. Another added benefit is that there no longer needs to be a legend. The data labeling is now incorporated as part of the graph, and it is also much easier to associate a label with its data point, since they are beside one another.

If more precision is required, data labels can be added and the grid lines removed.

This has the added benefit of increasing the data-ink ratio.

So the pie chart is not truly evil. It is quite effective at representing proportions of a whole for a limited amount of data points. Unfortunately it is often overused and misused. However, with a small amount of thought towards the correct choice of chart type, better insights can be provided to the viewer.

When can you use a pie chart? Every set of data is different an unique but consider these basic guidelines:

The data set has no more than 6 data points

The data points are an even fraction (thirds, quarters, fifths)

In almost every other case, an alternative data visualization is usually a better choice. Yes, the pie perhaps represents the whole marginally better, but loses it’s value in the representation of the details, and it is within the details where insight lie, like a lion in high grass.