Very often on the web one can find discussions related to the specific form of graphic representation of the part-to-whole relationships called the Pie Chart. Almost inexplicably but with great help from developers and Vendors of reporting and / or analytical tools, pie charts are recognized as not only legitimate, but also the best way of showing parts to whole. Moreover, to make data look even nicer, pies are made with special effects like transparency and three-dimensional display. Nothing could be farther from a good way of displaying the data.

In order to briefly demonstrate the example of the pie chart shortcomings, first I made a small table with some company’s sales realization over Regions:

Graphs that follow I have made in Excel, which for all its quality has a huge flaw – it allows the user to create completely incomprehensible and non-functional graphs (see the last image).

Let’s look at representation of the share in total sales in the pie-chart:

Do you see interpretation problems? For example, if someone just showed you this chart without handing you details about the sales from the table above (and you are Sales Manager), would you know how much sales in the South region is less than sales in the West region? Or, can you discern whether sales in North or East region are better? I believe that you can not, although the relative difference of even 2%. The main reason is that the human mind and perception is not good in comparing two-dimensional surfaces. You will be able to tell which is bigger though (and it is doubtful if equal), but not by how much.

Definitely the best alternative for a pie chart is simple bar or column graph because in that case we have to compare only one dimension, length:

Answers to the above two questions are now simple enough – West Region is exactly twice as big in Sales then the South region, and we can clearly see that East Region is better then the North Region, even if the data is not sorted by Sales- which, by the way, is one of the advantages of this data presentation.

Let’s get back for a moment to the pie chart display; I will try to simulate three-dimensional view which is, unfortunately, too often represented in the reports coming from leading Business Intelligence Vendors.

Now North Region that is smaller then East Region appear to be bigger because of the wrong perspective. They are used so often probably because they look modern and somewhat cool, but in a more accurate, scientific representation of the data they have no application. I think you can not find a single scientific publication in natural sciences that has a pie chart. If you do, those scientists should be educated on how to display their results.

However, there are two reasons that at this moment I know of when pie charts can be used.

The first is that, unlike a bar graph, from pie chart at first glance you can see that categories summed up equals 100%. However, if the bar graph shows the part-to-whole relationship and it is clearly stated in title, then that information you already know, right?

The second reason is a little stronger case. If the focus is on presenting information about what regions summed up contribute over 50% of sales, pie chart is the best choice! In our example, from the pie chart it is relatively easy to read summary contribution of East and South: they are only slightly less than half of the total (<50%), while in a bar graph that is not so obvious.

In almost all other cases the best practices are clear – the pies as a display of quantitative information should be avoided. This, of course, does not apply to apple pies.

1. Matija 21/07/2011 at 17:22