Save the Pies for Dessert


1 perceptual edge Save the Pies for Dessert Stephen Few, Perceptual Edge Visual Business Intelligence Newsletter August 2007 Not long ago I received an email from a colleague who keeps watch on business intelligence vendors and rates their products. She was puzzled that a particular product that I happen to like did not support pie charts, a feature that she assumed was basic and indispensable. Be- cause of previous discussions between us, when I pointed out ineffective graphing practices that are popular in many BI products, she wondered if there might also be a problem with pie charts. Could this vendor’s omission of pie charts be intentional and justi fi ed? I explained that this was indeed the case, and praised the vendor’s design team for their good sense. Here sits the friendly pie chart: Its slices are upturned into an inviting smile. Its simple charm is beloved by all but a few, welcomed almost everywhere; familiar and rarely threatening. Of all the graphs that play major roles in the lexicon of quantitative communication, however, the pie chart is by far the least effective. Its colorful voice is often heard, but rarely understood. It mumbles when it talks. Nothing in the graph design course that I teach is more controversial than what I say about pie charts. People love them dearly and are often shocked to hear me speak ill of them. When students in my course don’t raise objections immediately, despite the shock on their faces, they often approach me during a break to defend their frequent use of pies. Many, I suspect, respond this way because they can’t imagine how they could possibly convince people where they work to abandon their beloved pies. Pie charts are not without their strengths. The primary strength of a pie chart is the fact that the message “part-to-whole relationship” is built right into it in an obvious way. Children learn fractions by looking at pies sliced in various ways and decoding the ratio (quarter, half, three quarters, etc.) of each slice. A bar graph doesn’t have this obvious purpose built into its design. Not as directly, anyway, but it can be built into bar graphs in a way that prompts people to think in terms of a whole and its parts. This can be accomplished in part by using a percentage scale. It is easy and natural to think in terms of various percentages in relation to Page 1 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

2 the whole of 100%. Seeing a bar extend to 25% along a quantitative scale conveys a part-to- whole relationship only slightly less effectively than a pie chart with a quarter slice, especially if the bar graph’s title declares that it displays the parts of some total (for example, “Regional Breakdown of Total Revenue”). Despite the obvious nature of a pie charts message, bar graphs provide a much better means to compare the magnitudes of each part. Pie charts only make it easy to judge the magnitude of a slice when it is close to 0%, 25%, 50%, 75%, or 100%. Any percentages other than these are dif fi cult to discern in a pie chart, but can be accurately discerned in a bar graph, thanks to the quantitative scale. Allow me to illustrate. Here is a pie chart with six slices. Notice how easy it is to determine that the value of Company C (the green slice) is 25%, one quarter of the pie. Company A Company B Company C Company D Company E Company F Now notice how that even the green slice, which was easy to read as 25% above, is no longer as easy to recognize as 25% in the chart below. Company B Company C Company D Company A Company E Company F None of the values have changed. I simply sorted the slices by size. In the earlier example, our ability to decode the green slice at 25% was assisted by the fact that the green slice began at the 6 o’clock position and extended neatly to the 9 o’clock position. Positions at the extreme top, right, bottom, and left of a circle mark 90 degree intervals from one another, Page 2 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

3 each of which forms a right angle. These four positions, as well as the 90 right angles that the intervals form, are quite familiar to our eyes and minds. In the second example, however, because the green slice now begins at a less recognizable position, the size of the slice is more dif fi cult to judge. You might argue that this problem can be easily solved by labeling the values of each slice, as shown here: 1% 7% 10% Company B 40% Company C Company D Company A 17% Company E Company F 25% Why stop here? With this pie chart, we’re forced to waste time bouncing our eyes back and forth between the legend on the right and the slices of the pie to fi gure out which slice repre- sents which company. We can solve this problem by directly labeling the slices with both the company names and the values, as shown here: Co mp any F, 1% Co mp any E, 7% Co mp any A, 10% Company B, 40% Company D, 17% Company C, 25% cult to read, we Do you realize what we have just done? Because the pie chart was dif fi added values so we wouldn’t have to compare the sizes of the slices and we added direct Page 3 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

4 company labels so we wouldn’t have to rely on a legend. We turned the pie chart into an awkwardly arranged equivalent of a table of labels and values. Here’s the same information, arranged as a properly designed table: Companies Percentage Company B 40% Company C 25% Company D 17% Company A 10% Company E 7% Company F 1% Total 100% This information is much easier to read when presented in a table than it was when awk- wardly arranged around the periphery of the pie. So why use a graph at all? Why show a picture of the data if the picture can’t be decoded and doesn’t present the information more meaningfully? The answer is: You shouldn’t. Graphs are useful when a picture of the data makes meaningful relationships visible (patterns, trends, and exceptions) that could not be easily discerned from a table of the same data. But what if we could display this same information in a graph that is easy to read; one that adds useful meaning by allowing us to compare the magnitudes of the values without label- ing them? Here’s the same data displayed in a bar chart: Company Percentages of Total Market Share 10% 15% 20% 0% 30% 35% 40% 5% 25% Co mp any B Co mp any C Co mp any D Co mp any A Co mp any E Co mp any F Now the values can be compared with relative ease and precision, relying solely on the graph, without labeling the values. What value does this bar graph offer, compared to a table? In little more than a glance it paints a picture of the relationships between six com- panies regarding market share. Not only is their relative rank apparent, but the differences in value from one company to the next is readily available to our eyes. Could we construct this same picture in our heads from a table of the same values? Perhaps, but it would take a great deal of effort and time. Why bother when a graph can do the work for you and tell the story in a way that speaks directly to the high-bandwidth, parallel imaging processor in Page 4 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

5 your brain, which operates much faster than the part of your brain that handles text, which is needed to process tables? Few people, if any, know more about graphs than William Cleveland, who has studied them extensively. Cleveland writes: When a graph is made, quantitative and categorical information is encoded by a display method. Then the information is visually decoded. This visual perception is a vital link. No matter how clever the choice of the information, and no matter how technologically impressive the encoding, a visualization fails if the decoding fails. Some display methods lead to ef fi cient, accurate decoding, and others lead to inef- fi cient, inaccurate decoding. It is only through scienti fi c study of visual perception that informed judgments can be made about display methods. The (William S. Cleveland, Elements of Graphing Data , Hobart Press, 1994, p. 1) And what is Cleveland’s opinion of pie charts? He refers to pie charts as “pop charts,” because they are commonly found in pop culture media, but much less in science and technology media. “Pie charts [along with other forms of area charts] do not provide ef cient fi detection of geometric objects that convey information about differences of values.” ( Ibid , p. 268) I’m afraid I’ve jumped ahead a bit in our story. I intend to explore the pie charts more com- prehensibly, beginning with their history. To sum up what I’ve said so far: Countless people buy that this familiar chart is a friendly guy. This opinion, however, is far too high; he is, after all, a humble pie. The History, Use, and Workings of Pie Charts fi rst appeared in 1801 in a publication entitled The pie chart by The Statistical Breviary William Playfair. In this publication, Playfair used a variety of graphs to present geographical areas, populations, and revenues of European states. We have Playfair to thank for many of the popular graphs that we use today, including the bar graph. Although he didn’t invent the line graph, his innovative work popularized it as a means to display quantitative values across time. More than anyone before him, Playfair relied on graphical representations of quantitative data because he believed that “making an appeal to the eye when proportion and magnitude are concerned, is the best and readiest method of conveying a distinct idea” (William Playfair, The Statistical Breviary , T. Bensley, 1801, p. 4). The term “pie chart” was not coined until years later and it is not the only food metaphor that has been used to describe it. The French referred to it as using the name of their soft round cheese—camembert. Playfair used circles in various ways to represent quantitative relation- ships by varying their sizes, subdividing them into slices, and overlapping them, much like Venn diagrams. The graph on the following page, from The Statistical Breviary , displays a series of circles arranged side by side with their centers aligned. Each circle represents a geopolitical region, sometimes divided into sectors through the use of slices, such as parts of the Turkish Empire in the second circle from the left. Page 5 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

6 Although Playfair usually made wise judgments about graphical images and their use for fi ts of the large collection of research presenting data, he did not in 1801 have the bene that has since examined the effectiveness of various graphical devices. He understood the intuitive ability of a sub-divided circle to represent part-to-whole relationships, but not the perceptual problems that we encounter when trying to compare its parts. Pie charts encode quantitative values primarily by two means: two-dimensional areas of the slices and the angles formed by the slices as they radiate out from the center of the pie. We now know that neither of these visual attributes is easy to compare. Our eyes are great at comparing differences in 2-D location and differences in line length, but not 2-D areas and angles. In the example below, the graph on the left uses the 2-D locations of data points and the one on the right uses line lengths (in this case, the lengths of the bars) to encode values. As you can see, both make it very easy to see and interpret differences between values. 30% 0% 5% 10% 15% 20% 25% 35% 40% 10% 15% 40% 35% 30% 25% 0% 5% 20% Company B Co mp any B Co mp any C Co mp any C Co mp any D Co mp any D Company A Co mp any A Company E Co mp any E Co mp any F Co mp any F Page 6 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

7 You might be inclined to believe that you can do a better job than most in judging differences in values that are encoded as 2-D area. Here’s a simple test. If the area of the small circle below equals a value of 1, what is the area of the large circle? ? ? 1 1 Not easy, is it? When I ask students to guess the size of the large circle, I get answers rang- ing from around 6 to 50.The area of the large circle is actually 16 times the area of the small circle. Stephen Kosslyn writes: The systematic distortion of area is captured by “Steven’s Power Law,” which states that the psychological impression is a function of the actual physical magnitude raised to an exponent (and multiplied by a scaling constant). To be precise, the perceived area is usually equal to the actual area raised to an exponent of about 0.8, times a scaling constant...In contrast, relative line length [such as the lengths of bars] is per- ceived almost perfectly, provided that the lines are oriented the same way. (Kosslyn, Stephen, , Oxford University Press, 2006, p. 40) Graph Design for the Eye and Mind Perhaps you object to the fact that you had to rely on relative 2-D areas alone to discern the differences above, without the bene fi t of relative angles as well, which play a role in pie charts. Here’s another test, this time using an actual pie chart. Look at the pie chart below and try to place the slices in order from largest to smallest. Cabernet Sauvignon Sangiovese Prosecco Chardonnay Syrah Tempranillo Pinot Grigio Having trouble? As you can see, comparing the angles of the slices doesn’t make it any easier. Page 7 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

8 Naomi Robbins writes: We make angle judgments when we read a pie chart, but we don’t judge angles very well. These judgments are biased; we underestimate acute angles (angles less than 90°) and overestimate obtuse angles (angles greater than 90°). Also, angles with horizontal bisectors (when the line dividing the angle in two is horizontal) appear (Naomi Robbins , Creating More Effective larger than angles with vertical bisectors. , Wiley, 2005, p. 49) Graphs If a chart is doing its job, you shouldn’t have to struggle. Look at how easy it is to compare the percentages using the bar graph below, which displays the same values: Total Revenue by Product 0% 5% 10% 15% 20% 25% 30% Sangiovese Pinot Grigio Prosecco Cabernet Sauvignon Chardonnay Syrah Tempranillo Pies in All Their Modern Glory The pie chart, like all graphs that use the position, length, or area of objects to represent quantity, includes an axis with a quantitative scale, only it is never shown. The axis and scale of a pie chart is not linear, as it is with most graphs, but circular, for it is located along the circumference of the circle. Here’s what a pie chart would look like if its axis and scale were visible: 100% | 0% 95% 5% 90% 10% 85% 15% 20% 80% 25% 75% 30% 70% 65% 35% 40% 60% 55% 45% 50% Page 8 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

9 People love dressing up their pie charts today to look mouthwatering. Why stick with a simple 2-D pie chart when you can add a third dimension of depth to the picture and throw in some lighting effects and contoured edges while you’re at it, as shown here? It’s pretty and eye-catching, but is it more meaningful or easier to interpret? Actually, by cult to interpret. The adding depth to the pie and changing its angle, we’ve made it more dif fi green slice now appears greater than it actually is, because of the depth that’s been added. The slices are now more dif fi cult to compare, because the angle skews their appearance. For those of you who can’t resist tilting your pies (after all, pie tilting is an ancient and re- spected sport among cultures known for their talent with pastries), let me illustrate the effect that you’re creating. Below are three pie charts that are exactly the same, except that the one above is 2-D and the other two are 3-D and tilted. Notice how different the relationships between the slices appear from one version to the next. Page 9 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

10 Why stop here? Based on what many software vendors advertise, we’re encouraged to tweak them out with abandon. For example, with Excel and several other products, you can now easily manipulate the transparency of the pie, creating utterly useless charts such as this: Believe it or not, this is the same pie chart as the one above. The only difference is that now you can see through it. Aren’t you happy Microsoft added this nifty feature? Trust me when I say that I am not doing anything here that isn’t marketed with pride by countless software vendors. Here’s a Crystal Xcelsius pie chart that’s been polished to a high-gloss gleam: I pulled this example from a dashboard. One of the objectives of a dashboard is to present information in a way that can be quickly read and easily understood. If you glance at this too quickly, however, you’re liable to think that it contains three slices. This misperception is a result of the simulated re fl ection of light on the shiny surface of the pie. When light re fl ects like this off of objects in the real world, we fi nd it annoying. We have to squint to block the glare in an effort to see the object clearly. Why would we ever want to reproduce this annoy- ing and misleading effect on a computer screen? Sometimes, despite their obvious limitations, absurd demands are placed on pie charts to show a great deal more than usual. On the following page there is an example from Advizor Analyst/X (an otherwise good product), which attempts to show two levels of part-to-whole relationships at one time: one per country (the slices) and one per product type (the circular bands of color within each country). It would be impossible to compare quantities of a prod- uct type between countries, given how differently they are shaped. Page 10 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

11 Let’s examine another ineffective use of pie charts. Edward Tufte once said that “the only worse design than a pie chart is several of them, for then the viewer is asked to compare The quantities located in spatial disarray both within and between pies” (Edward Tufte, Visual Display of Quantitative Information , Graphics Press, 1983, p. 178.) I share Tufte’s opinion that this is an ineffective way to compare multiple part-to-whole relationships. 2006 2007 2005 2004 Co mp any A Co mp any B Co mp any C Co mp any D Co mp any E Co mp any F Try to follow the changes of these various companies and how they compare to one another through time. It is nearly impossible. Notice how easily you can do it, however, using the following display: 2007 2006 2005 2004 50% 40% 30% 20% 10% 0% 20% 30% 40% 50% 0% 0% 10% 20% 30% 40% 50% 10% 0% 10% 20% 30% 40% 50% Co mp any C Co mp any E Co mp any B Co mp any D Co mp any A Co mp any F Page 11 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

12 This is a vast improvement over the series of pie charts above, but is still not as easy as it ought to be to see the up and down changes in part-to-whole percentages from year to year. Nothing shows change through time better than a line. Here’s the same data, this time displayed as a single line graph: Company Percentages of Market Share by Year 50% 45% 40% Company C 35% 30% 25% Company D 20% 15% Co mp any E Co mp any A 10% Co mp any B 5% Co mp any F 0% 2004 2005 2006 2007 Now it is easy to see the changes and to compare the magnitudes of the parts in a given year. What we can’t see in this example, however, are market revenues as a whole and how they changed from year to year. For this view, we can use an area graph (see below): Total Company Revenues by Year U.S. $ 20,000 Co mp any F Co mp any A 16,000 Co mp any D 12,000 Co mp any B Co mp any E 8,000 4,000 Co mp any C 0 2006 2007 2004 2005 Keep in mind that any individual graph might serve a particular purpose exceptionally well, but suffer from problems if used for other purposes. This area graph, for example, does a good job of showing how total revenues changed from year to year and also gives a sense of each company’s portion of total revenues in a particular year. You would not want to use it, however, to discern how each company’s portion of the whole changed from year to year. This is true for two reasons: 1. The unit of measure is dollars, not percentages, so an increase in dollars of a single Page 12 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

13 company’s revenues from one year to the next could actually represent a decrease in its percentage of the whole. 2. Other than Company C, which is positioned at the bottom of the graph and has a fl at baseline, you cannot trace the top of each country’s colored area to determine how its revenues changed, but you must focus on the height of the area in any one year from its bottom to its top, which is more dif fi cult to judge. The Secret Strength of Pies You might be wondering, aside from the natural way that pie charts suggest a part-to-whole relationship, do they have anything else going for them? I have read every research study that I could fi nd that tested the effectiveness of pie charts versus other means of displaying quantitative data, beginning with one that was done in 1926 by W. C. Ells, and have found only one advantage that can con fi dently be attributed to pie charts. Unfortunately, this one strength is rarely if ever useful. Ian Spence of the University of Toronto has been involved with several of these studies from 1989 on. In 1991, Spence and Stephan Lewandowky conducted a series of experiments to compare the relative effectiveness of different means to display the same part-to-whole relationship—a pie chart, a bar chart, and a table—for a variety of tasks. The study was titled “Displaying Proportions and Percentages,” which was published in (Volume 5, pages 61-77). For each task they measured the Applied Cognitive Psychology speed and accuracy of test subjects’ responses. The one slight advantage of a pie chart over a bar graph involved a task that required subjects to answer the question, “Which is greater, A+B or C+D?”, while examining one of the three charts below: A D B C AB CD ABCD 10 25 7 26 Page 13 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

14 Sometimes the parts that the subjects were asked to sum were adjacent to one another (for example, A+B) and sometimes they were not (for example, A+C). It is not dif cult to fi believe that it is somewhat easier to sum the areas of slices in a pie than it is to imagine the combined heights of bars stacked on one another. I wonder, though, if the advantage that pie charts had over bar graphs would have been eliminated when the parts that were being summed were not adjacent to one another and the bar graph display had been complete, with its quantitative scale (see below). D C A B 30 25 20 15 10 5 0 ABCD Regardless, the fact remains that a comparison of two sets of summed parts is rare in the real world. But, by all means, should you ever need to display data for this purpose, a pie chart would serve you well. Otherwise, save the pies for dessert. About the Author Stephen Few has worked for over 20 years as an IT innovator, consultant, and teacher. Today, as Principal of the consultancy Perceptual Edge, Stephen focuses on data visualiza- tion for analyzing and communicating quantitative business information. He provides training and consulting services, writes the monthly Visual Business Intelligence Newsletter , speaks frequently at conferences, and teaches in the MBA program at the University of California, Berkeley. He is the author of two books: Show Me the Numbers: Designing Tables and Graphs to Enlighten and Information Dashboard Design: The Effective Visual Communica- . You can learn more about Stephen’s work and access an entire library of tion of Data articles at Between articles, you can read Stephen’s thoughts on the industry in his blog. Page 14 of 14 Copyright © 2007 Stephen Few, Perceptual Edge

Related documents