Circle Circle Dot Dot... What Some Dashboards Should Have Not

My name is Jon and I’m a circle addict, and chances are you are too. Humanity is addicted to circles. Understandably so, as circles are everywhere; in our vehicles, our food, even our faces. It’s no wonder we try to include them in our data visualizations, but should we? Ever since we were young and started learning about fractions we have had circles associated with numbers and data visualization. Remember the pizza and pie analogies your teacher would make?

Pie charts can be great for a comparison of two, three, four, even five values. Moving past that however, they tend to get a little confusing. It becomes harder for us to distinguish between the size/area differences than if we were to display the information in a bar chart for example. Data visualization should allow us to easily and quickly understand the data, so we can ensure we display the information in a way that it can most easily be consumed.

While the circle may be ingrained in our cultural psyche (performing rituals, sitting around the hearth, a symbol of unity, and has been regarded as the perfect shape), cognitively it is harder for us to precisely distinguish size/area with respect to circles than it would be with a bar or column chart. Our brains simply haven’t evolved as well to easily recognize area differences as it has to recognize differences in height or length.



Figure 1

One of the biggest weaknesses of circles has to do with the size. The bigger the circles are the harder it is to visually tell the difference between them. A circle with a diameter (d) of 2 units and a circle with a diameter of 4 units will appear to be double the size, yet a circle with d=10 and one with d=12 are going to be appear around the same size. Additionally the larger the circle the bigger the area and the number it is representing, even a small change in the diameter can mean a big difference in the area. Take the example above; the difference between the diameters of the smaller circles on top is 20 pixels, this results in a difference of area of about 942 pixels. Visually the two sizes are noticeably different. Now take a look at the larger two circles. Visually they appear to be roughly the same size, there is only a 10 pixel difference between their diameters. However, the difference in area between them is about 3220 pixels, over 3 times that of the smaller circles.

Let’s take a look at the image below.

Figure 2: Causes of death in the 20th century (source: Information is Beautiful)

The image is depicting the causes of death in the 20th Century. Here we can see that non-communicable diseases accounts for the most deaths and is followed by infectious diseases. Now, take the numbers away: can you tell what the numerical difference is between the two? How many causes were responsible for over 500 million deaths?

If we set this up as a stacked bar chart we could see the differences, as well as the subcategories within it, much more easily. What if we allowed the viewer to compare causes using filters? Enable drill down to further investigate the general categories? Or what if we set this up as a standard bar chart that went in sequence of category, subcategory, sub-subcategory, and color coded the bars to denote which category they belonged to? Then we could easily see how everything stacked up against each other. We could see that the sub category of cardiovascular disease accounted for more deaths than that of most of the categories themselves. In the above image this information is muted slightly because of the coloring.

Stephen Few has detailed this issue with circles in his article, Our Irresistible Fascination with All Things Circular. In this article he shows a chart (Figure 3) published by Bloomberg in January 2009. Then offers up an alternative to the chart (Figure 4), which do you find easier to understand?

Figure 3



Figure 4

I love circles just as much as the next person, but I love being informed even more. So when I have the choice between a visualization that may seem “cool” but is harder to understand, and a visualization that is easier to understand but visually less “cool” I’ll choose the less cool option. After all, the coolest thing about any data visualization should be the data itself.