The Pie Chart Challenge

Rule #1 of Presentations is this: know thy audience.

When Steve Jobs used pie charts in his keynotes, he was addressing an audience of developers, consumers, and loyalists. These were people he needed to keep impressed, and a great way to impress people is to make the data seem really BIG and compelling and outsized. (Remember that one presentation with the graph that expanded its axes and "zoomed out" in real time? That's serious data showmanship right there).

Pie charts may not be the most accurate way to represent data, but when your goal is to sell a story, they can do the job nicely. That even nontechnical consumers are used to seeing them makes them easily digestible, too; in this sense, pie charts are a form of universally recognizable data language. Pie may be junk food of a sort, but it's comfort food all the same.

You seem to believe that there is a "best" way to represent data, but therein lies the assumption that all audiences are largely homogenous. I disagree. There are certainly universal principles for good presentation design, but data representation is a tough one. The sophistication and expectations of the audience play a huge role in how you should speak to that audience.

I think some of this debate is slightly missing a key point: tables are for showing precise data, but graphics are for showing patterns. If you're expecting your viewer to read accurate values off a chart, you're probably using the wrong tool for the job. If you're trying to use the same chart type to show different kinds of pattern in the same underlying data, you're probably having a "when all you have is a hammer, everything looks like a nail" problem.

I do agree that pie charts are overused and there is often a better alternative, but do stacked sausage charts really do much better when it comes to showing differences between roughly similar values? Here's an example:

http://imgur.com/zGuwg

Sure, it's obvious which red bars are longest on the left-hand side, because they have a common edge. But in that case, it's not really the length of the bars that the viewer is comparing, it's the position of the right-hand end, just like a simple bar chart.

As soon as you stack thing up, so you're trying to compare sections that don't align at either end such as the green or orange bars here, it is still difficult to see which ones are longer. Something more drastic, such as reversing the order of the bars completely, throws off any comparison, as with the second and fifth lines.

I don't see how a series of sausage charts like this is much better than a series of pie charts if your goal is to allow your viewer to compare the sizes of each category across the series. You need to get the key values onto the same scale to do that effectively, and I would expect a better choice to be something like a multi-bar chart (with the same zero level for each bar, either grouping all categories for each point in the series or isolating each category in a chart of its own) or perhaps some sort of overlapping line chart if the gradients are interesting.

I'm not convinced at all. It took much longer for me to dissect the information in the example bar chart that it did for me to dissect the information in the example pie chart. The bar chart was confusing; the pie chart was instantly clear.

Furthermore, I disagree with the author's assertion that it is difficult to compare areas and angles, and I personally find it more difficult to compare length. The author provides no references that support his claims - without some clear research behind it, I simply cannot take his statements as anything other than opinion.

Not only does it indicate how close I am to completion, but also which categories I have completed. Additionally its pie shape makes it easier to use in this setting (where a horizontal chart would be too big)

http://www.mint.com/blog/wp-content/uploads/2010/04/trivial-...

There are some good arguments both in this thread and in your blog, and both for and against the two types of charts.

One thing that you would have to make absolutely clear to your audience when using the "sausage" chart [1] is that the length represents a whole. Pie charts obviously represent no less and no more than one whole, but that's not so obvious with a bar.

You handle this well in your examples by marking the end point with "100%", but I worry that others would not be as diligent in their labeling if these charts were to take off.

1. Is there really no better name for that? :D How about calling them snake charts or something?

I'd add that the pie charts you converted in this post are actually the sorts of pie charts you were okay with in your first blog post, where you said "pie charts are really only useful when a small number of categories of data are far, far greater than others" (http://blog.jgc.org/2009/08/please-dont-use-pie-charts.html), with "republicans" being, arguably, an exception.

The "sausage" (worst. name. ever. :-) ) solve some of the problems, but not all of them. With enough slices, it's not always clear which is larger than which. Depending on what the viewer wants/needs, often a humble bar or column chart would be more effective. The sausage and the pie both do a decent job with just a few slices (the sausage is definitely an improvement), but both break down with a lot of slices.

The bar chart with independents, republicans and democrats makes it seem like every group is equally large.

Are there as many independents as republicans?

The bias of this method towards smaller groups is no worse than the pie charts. But it's not better.

I just have to say that the bar chart you show is so small that I can hardly read the numbers, so based on that I would say the pie chart won. As you mentioned in your other article, if you can't read it (aka 3d charts) then it is pretty much useless.

"I do not believe there is a better representation than a pie chart for this: http://goo.gl/yQMve

awesome.