A month ago, respected marketer Seth Godin opened a can of worms when he recommended the use of pie charts over bar charts in powerpoint presentations. Pie charts, he argues, are more obvious and to the point, even though they carry less data.
The reaction of the data visualization community was immediate and harsh. Many graph experts do not use pie charts ever. In Show me the numbers, Stephen Few writes: “I don’t use pie charts, and I strongly recommend you abandon them as well.”. This sentiment is widely spread, because pie charts, which require readers to compare angles, are more difficult to decipher than, say, bar charts.
This is an example adapted from Naomi B. Robbin's Creating More Effective Graphs. Can you tell which wedge is biggest?
My feeling is that pie charts are actually quite good to show 1 data point, that is to show the proportion of one item to a whole. However I wouldn’t use them for more than 1 data points, let alone for 10 or more. And the worst thing you can do with two busy pie charts is putting them side by side and hope that people will able to see changes.
That being said the reactions to Seth’s post were very strong.
Jorge Camoes wrote: No bar charts – a caricature. Jon Peltier: “Pie charts are also abused and overused, and if not forbidden, should be placed on probation.”. Junk Charts: “The advice of No Bar Charts is misguided.. Zach from Juice Analytics: “Pie charts are the most frequently mis-used charts in my experience.” Andreas from More information per pixel wrote that “[...] you would rule out Seth’s pie chart, and use the bar chart in the appropriated business context.”. And Stephen Few himself wrote a post called “Godin’s Silly Rules for Great Graphs”.
All above commenters provided constructive criticism and most often, better versions of the examples that Seth had put up.
However, I sympathise with Seth’s position, because he’s a great speaker and because I feel the world of data visualization and that of presentation, while both belonging to the same greater family of visual thinking, evolve in completely parallel dimensions. Data visualization experts edict laws that are supposed to have an universal reach, but which don’t quite work in some specific contexts. Presentations gurus, apparently, do the same.
But sometimes they coincide. A friend of Seth’s, Garr Reynolds, published this year a fantastic book, Presentation Zen. In it, among very insightful presentation advice, I had the surprise to see a chart from the OECD Factbook 2007 redone!
Garr Reynold's version of the obesity chart.
To be honest, we at the OECD never published the “before” version, partly because we know our name is not OCED. The excel version can be found here. Yet, the new version struck me as an eye-opener. I had overseen the design of most of the original FB charts, and I thought they were clear and legible enough, with the 30 or so countries present on each graph ranked by value. But clearly, not as clear as Reynold’s ! the message is obvious, dramatic and direct.
On that chart, there is absolutely no element that doesn’t support the message, that doesn’t guide the audience to agree with the speaker. In my opinion, this is the role of charts in presentations: show relevant data that support an argument, not show more data or better data or glitzier data.
This before/after slide has been an inspiration for this blog, where I try to provide simpler graphs than those usually found in our publications. That’s why I bring that whole story.
While I cannot support the argument that pie charts do the work better than bar charts even in the presentation context, I agree with Seth that many presenters feel that using bar charts which are often the textbook-correct form, shouldn’t excuse graph authors from other design flows, the most common of which being to try to plot too many series on one single space.
At the end of the day, Seth commented back on Stephen’s blog and both came to an understanding. Seth admitted that if all presenters could do bar charts like Stephen, the world would be a much better place.
August 1, 2008 at 2:53 pm
Jerome, I would suggest that the OECD chart and Garr’s have different roles and can’t be compared. OECD, like Eurostat, must comply to some guidelines, and probably one of them is to display the data for all member states. Garr, on the other hand, is free to select the data that supports his point. His chart is an argument, OECD’s is a description.
By the way, the chart could be improved. A in-cell bar chart, next to the table, probably would be good enough.
August 1, 2008 at 5:49 pm
Hi Jorge,
I agree: both charts have different functions. That’s the reason why Seth’s position, which is really focused on presentations, caused quite a turmoil.
our graph remains very neutral. It’s made so that people can look up the value for their country. But the other one supports a clear message.
That being said, when we designed our graph, a few years back, (which for the record looks like it because it had to fit under a table on a page) it didn’t occur to us that there could be a simpler graph.
this is why the reflexion on graphs in presentation appeals to me. I don’t think we can just take the graphs of our publications and put them online as is. They have too many datapoints, so they cannot be understood at a glance, like Garr’s. And, they are static, which is another problem for the web. Those are aspects that guys like Seth and Garr have been thinking about. which doesn’t mean that graphs on blogs should be like graphs in powerpoint, with really big font and all. But their angle is interesting.
August 1, 2008 at 7:43 pm
Yes, but are you free to remove datapoints (countries)? Do it and some countries will skin you alive… even if no one reads the factbook.
If you don’t, I see two options: graphical tables (with sparklines, for example) and scatterplots, where the more data points the better.
August 4, 2008 at 9:26 am
Plotting all countries is more a tradition than a rule – In the book, there are quite a few charts where only selected countries are plotted. I have a problem personally with ranking 30 countries according to an inaccurate variable, like GDP, or here obesity which measurement varies from country to country, and this is what bar charts do.
But back to that chart. The problem is that for all factbook indicators, we agreed on a template with one table that takes about 2/3rds of a page, and one chart which occupies the bottom third. The table is ordered alphabetically, so you can look up a country. So the chart is often a bar chart, which allows readers to find the “top”country/ies.
But that function could have been solved by the table. The cell with the highest value for each year, for instance, could be slightly highlighted, likewise the cells with the top-decile or top-quartile data could also be singled out. There are elegant ways to do that.
So, we could have done something more interesting with the charts…
April 13, 2009 at 12:01 pm
[...] Pie charts, he argues, are more obvious and to the point, even though they carry less data ( http://oecdfactbook.wordpress.com/2008/08/01/seth-godin-on-graphs-in-presentation/ [...]