Reimagining Pie Charts

4 min readDec 31, 2020

Pie charts stir emotions. I experienced this first-hand when a lecturer once brought to class what I assumed was his lunch, only to proceeded to empty the contents of the great and good pukka pie into a bin. Deeply offended, I was left puzzled at what could cause such vitriol. The following is a more diplomatic approach to explaining why, and how, proportional distributions can be communicated more effectively.

The invention of the pie chart is attributed to William Playfair. In the Statistical Breviary (1801) he used it to show the division of land within what was then the Ottoman Empire. Further reading on the origins of the pie chart is linked below.

Theory

Pie charts display proportions between categories by dividing a circle into segments. The length of each arc is indicative of the proportion of the category it represents and is most effectively used when a single category dominates.

Beyond a few categories, it becomes increasingly difficult to interpret as each slice’s size becomes smaller. It follows that accurate comparisons between pie charts are trickier still, without a common reference point to anchor our baseline. The other caveat relates to formatting with 3D animations, which is probably best left to Pixar. Including it in a pie chart, distorts proportions, which is precisely what is being evaluated.

Practice

With the theory covered let us apply these principles to an existing publication that can be improved upon with a makeover. The original concerns courses on offer at the University of Berkley, UCLA, Davis, and Stanford. The data source and the published visualisation is also linked below.

Does it jump out that Stanford offers fewer course in biology than Berkeley? No? Another clue that the visualisation can be made clearer is that the same colour is used to represent multiple categories. I count seven (biology, biochemistry, forestry, religion, agricultural engineering, veterinary science, and sociology) all colour coded orange at Davis.

Viz Makeover

Identifying biases in data collection and processing, as well as the insight generation phase is a prerequisite to creating trustworthy visualisations. As the data set is squeaky clean, though, we can move ahead with the design of the data dashboard.

First, a recap of the problem statement. I want to understand the trend in the number and type of courses on offer at UC Berkley over the last century. A stacked bar graph is an alternative for making comparisons between parts-of-a-whole. Positional encoding (orientation along an axis) can also be baked into the analysis to understand trends over time.

The natural sciences offered the most courses in 1900. By 2011 this was no longer true with the humanities growing threefold over the recorded period. There is, however, more to the story. Even as each subject’s relative importance has changed, the number of course on offer has also increased.

Design and formatting choices also serve to aid comprehension. Reducing clutter like heavy grid lines directs focus toward the data itself, while labels obviate the need for a column header. The sequencing of the dashboard with a title and subtitle at the top, followed by the main graphic and supporting information at the bottom, renders a familiar layout.

Conclusion

The use of pie charts should be handled with care. And only when it is safe to do so (i.e., when one/few slice(s) stand out). Stacked bar charts overcome challenges associated with evaluating differences in angles by segmenting a larger category into subsections. Applied with design best practice, the stacked bar chart reduces time-to-insight and dare I say, might even have your interlocutor share their pie with you.

Reimagining Pie Charts

Theory

Practice

Viz Makeover

Conclusion

Further Reading

Written by Norman Mangera