Grudge Match: Pie Chart vs. Bar Chart

pie-chart-vs-bar-chartIf you follow the data visualization cognoscenti at all you’ll know that the prevailing school of thought is that bar charts are vastly superior to pie charts. That prevailing wisdom was called into question last week as a recent study from Tufts University gained attention from Timo Elliott on the Business Analytics website, possibly reigniting the long-running debate about pie charts.

The Tufts study was ostensibly designed to “determine the effectiveness of functional near-infrared spectroscopy (fNIRS) in distinguishing differences in cognitive states that derive from visual design alone”. It compared bar graphs and pie charts to “test the viability of fNIRS for measuring the impact of a visual design on the brain. The results demonstrated that we can indeed measure this impact, and furthermore measurements indicate that there are not universal differences in bar graphs and pie charts” (emphasis added).

Research method

The study group conducted three experiments to (a) examine how participants process bar graphs and pie charts differently in their brains, (b) determine the efficacy of using fNIRS as a technique for evaluating mental workload in visual tasks, and (c) classify visual tasks that are most suited for using fNIRS in evaluation.

Participants were presented a series of slides, each displaying either a bar graph or pie chart, to view se- quentially. They were instructed to estimate the size difference to the nearest ten percent of a smaller section of the graph (marked by a red dot) in the current slide to a larger section (marked by a black dot) in the previous slide. Estimates were entered using a single keystroke on the keyboard (‘1’ for 10 percent, ‘2’ for 20 percent, etc).

Research conclusion

Our results show that changes in deoxygenated hemoglobin during the use of bar graphs in a complex task are statistically different from those observed during the use of pie charts. However, this distinction was not categorical. Instead, brain activity depended on the individual (emphasis added) and correlated with reports of mental demand in a NASA-TLX questionnaire. These differences between participants may call into question the conventional wisdom to always use bar graphs instead of pie charts.

We discovered that 14 out of 16 participants found one chart to be more mentally demanding than the other. Therefore, we reject our initial hypothesis that brain signals would indicate that bar graphs are easier to use for most people.

Subjectively, there was no indication that either bar graphs or pie charts were superior across all participants on this particular task. 7 participants reported pie charts to be more mentally demanding and 7 participants reported bar graphs to be more mentally demanding (the final 2 reported no noticeable difference). Although we did not investigate the underlying cause of this observation, we suspect that this is due to either differences in cognitive traits (e.g. spatial ability), strategies employed to complete the task, or previous experience with bar graphs and pie charts.

The full report is available for download:

Using fNIRS Brain Sensing to Evaluate Information Visualization Interfaces

 History of the Pie Chart and Its Ill Favor

This latest development got me wondering about when and why the pie chart came into such ill favor. (For those interested in the full story of the pie chart, Ian Spence’s article “No Humble Pie: The Origins and Usage of Statistical Chart,” Journal of Educational and Behavioral Statistics Winter 2005, is a great place to  start.)

Along with the line graph and the bar chart, the creation of the pie chart is credited to William Playfair, first appearing in his 1801 Statistical Breviary. According to Spence, Playfair’s statistical graphs didn’t really catch on until the early 20th century and it wasn’t long after that the pie chart came under fire. In 1914, W.C Brinton was quoted as saying “the circle with sectors is not a desirable form of presentation.” That criticism seems to have sparked a debate that continues to rage today and was followed by several volleys back and forth between proponents and detractors. The result of these debates was that by the middle of the 20th century, many statisticians had a dim view of the pie chart.

In 1977 MacDonald-Ross concluded that the bar chart was superior to the pie chart. And in 1984, Cleveland and McGill concluded that because human judgements of angle and area are inferior to our judgement of length, that it logically follows that pie charts are less effective than bar charts in estimating proportions or comparing proportions.

In 1983 Edward Tufte  wrote in his classic The Visual Display of Quantitative Information,

A table is nearly always better than a dumb pie chart; the only worse design than a pie chart is several of them, for then the viewer is asked to compare quantities located in spatial disarray both within and between charts [...] Given their low density and failure to order numbers along a visual dimension, pie charts should never be used.

In recent years, Stephen Few has become perhaps the most vocal critic of pie charts. In his August 2007 paper, Save the Pies for Dessert, Few argues convincingly in favor of using bar charts over pie charts in most situations. UPDATE: Few recently posted an article that addresses the Tufts study. See A Pie in the Face of Visualization Research on his site. As an aside, I hope this blog post of mine isn’t the one that Few is referring to that “cited my negative opinion of pie charts and then pointed to this paper as potential evidence of my error.” That was not my intent at all. I merely wanted to post some background about the debate.

So what is one to make of this latest research from Tufts? I suspect this won’t be the definitive finding that convinces pie chart detractors to amend their position. This debate has been raging for decades so I think we’ll likely see some shots fired from the bar chart proponent camp, detracting from the Tufts research, while the pie chart lovers will use this as proof to further exploit the graph beyond what it is meant to do. But, to ignore the report, or dismiss it out of hand  would be unjust. In his 1994 book, The Elements of Graphing Data, William Cleveland says:

It is only through scientific study of visual perception that informed judgments can be made about display methods.

With that in mind, I hope we’ll see additional studies using brain scanning methods to evaluate the merits of various forms of visual representations. I wonder if we may eventually find that the question of bar vs pie is more nuanced than we’ve allowed, that it might depend to some degree on individuals and context, at least to a greater degree than has been recognized.

About Jeff Bennett

Owner & Designer at Digital Splash Media. Multimedia Web Content: Visual Explanations | Animated Explanation Videos | Infographics, Data Visualization, & Diagrams

,

6 Responses to Grudge Match: Pie Chart vs. Bar Chart

  1. Alberto Cairo February 3, 2013 at 3:05 am #

    I have doubts about how firm the conclusions you can extract from this study can be. I appreciate the efforts, and I am certainly open to review my ideas about what is appropriate or not, but there are things that puzzle me, such as:

    “Participants were presented a series of slides, each displaying either a bar graph or pie chart, to view se- quentially. They were instructed to estimate the size difference to the nearest ten percent of a smaller section of the graph (marked by a red dot) in the current slide to a larger section (marked by a black dot) in the previous slide. Estimates were entered using a single keystroke on the keyboard (‘1’ for 10 percent, ‘2’ for 20 percent, etc).”

    “To the nearest 10 percent” is a really rough estimate. I have to read this paper again, but if we’re so lax, almost any graphic form will be equally valid for comparisons. I can estimate quite effortlessly “to the nearest 10 percent” in a bubble chart, but that doesn’t mean that this estimate is equally accurate to one made with a graph based on a single 0 baseline (bar graph, dot plot, lollipop graph, etc.)

  2. Jeff Bennett February 3, 2013 at 7:01 am #

    Thanks for your thoughts Alberto. Great to get some input from a leader in the field.

  3. Alberto Cairo February 4, 2013 at 6:31 am #

    I have other issues: The lack of randomization, the small sample, etc… I think that they simply go way too far in their conclusions. You cannot generalize when your sample is so small and not randomized.

  4. Jeff Bennett February 6, 2013 at 3:40 pm #

    Stephen Few has weighed in about the article: http://www.perceptualedge.com/blog/?p=1492

    I’ve updated the main article here to include that.

  5. Stephen Few February 6, 2013 at 3:55 pm #

    Jeff,

    Rest assured that your post was not the one that I was referring to in my review of this study by Tufts University. I was referring to Timo Elliott’s post. Timo took great joy in citing this study as one that seemed to challenge my position regarding pie charts. Why? Because I’ve challenged some of Timo’s statements in the past and I’ve often exposed problems in SAP’s products. Timo works in marketing at SAP.

  6. Jeff Bennett February 6, 2013 at 4:35 pm #

    Good to hear. Thanks Stephen. Good to read your article. Well-thought out as always.