« Parallel programming with foreach and snow | Main | Creating your own R packages: a tutorial »

August 26, 2009


Feed You can follow this conversation by subscribing to the comment feed for this post.

I recently published a paper with pie charts, and felt slightly guilty the whole time. But, intuitively, it made the data look much more attractive. I tried both dot and bar charts, and both looked somewhat wonky, whereas the trends were much clearer in pie charts (link - see the figs). Reviewers seemed to think so, too. Funny how this bias for pies has crept in. Then again, I'd argue that when trends are overwhelming, they're not such a bad thing.

Three examples wherein the pie chart shows that there is not a large difference in the groups. You can imply a trend using a bar chart... however there more than likely is no trend.
It all depends on what kind of data you are displaying, as to what gives better information.
Typically pie charts order things from smallest to largest, and because I am fairly alert, I can see the difference in A and C. Pie charts can be good for advertisement or confusing bosses - especially when you add perspective (regardless of whether it is isometric).

If my question was, "How well represented are different ethnic groups in this given population?" and I need to present that data - pie chart. It answers the logical next bunch of questions. A bar chart could be used - but if I use the bar graph A it implies a massive gap - but everyone is represented between 18-22%... pretty damned close. Most publishing would look at the above graphs and demand the y-axis cropped to save on paper space and make things visible. Net result - a scary trend.

If you are comparing linear trends - "Income against years", "coffees per day" you would be a fool to use a pie chart.... but pie charts serve a purpose, in an engaging way - and short of fetishist usage of them, they can be extremely good at making numbers tell a fair story.

I admit to hating pies. I suggest my students never use them, except in the case of a single descriptor expressed in %-of-whole terms. Even there, they're as good as a bar chart. Everywhere else, they're worse -- terrible for comparison among charts, bad at displaying longitudinal data. They're seldom as good as alternatives, and often much worse.

The attractiveness is important, but from where I sit, we're feeding a dangerous public addiction to a lousy chart format.

I think another issue with pie charts is that they suggest completion of the target data set.
Thus in jebyrnes cited paper, for example, the suggestion is that the graphs represent all possible types of invasive or extinct "things". (Sorry this is way outside my expertise).

However, this might not be true - there might be (???) others that were outside the scope of the study and hence are not represented. Certainly the target audience would know the difference here but the layman (I) would not.

However audiences are not always knowledgeable. Many presentations are made outside one peer group and in some cases this sort of graph could give a misleading impression.

To be fair, this is not solely the province of pie charts. For example, some people tend to lop off the bottom of bar charts to only include the area where differences exist (say Example A here only included 15%-25%). This gives a distorted view of the differences in the subject categories.

The comments to this entry are closed.

Search Revolutions Blog

Got comments or suggestions for the blog editor?
Email David Smith.
Follow revodavid on Twitter Follow David on Twitter: @revodavid
Get this blog via email with Blogtrottr