Catching my eye during my daily romp through the Newseum today was this front page by the Press-Enterprise of Riverside, Calif., circulation 112,084.
Anytime I see a bunch of charts on page one, I typically slow down to take a look. All too often, unfortunately, it really is like rubbernecking at a fender-bender. Because all too often, the charts really don’t do the job they’re supposed to be doing.
That was the case with these pie charts today. Click for a closer look.
The biggest problem here is that we have three values for each candidate: One from November, one from earlier this month and one reportedly current, in the largest pie. In a “horse-race”-oriented graphic like this, the reader would presumably like to see who is moving up and who is moving down, in relation to each other.
But in this case, the reader has to find, say, Mitt Romney‘s current number and then look for the corresponding numbers on the other two pie charts. Is Romney moving up or down in the polls? The numbers suggest he moved up but then fell a bit lately. but it’s difficult to get a sense of that from these pie slices.
Rick Santorum, in fact, has made a huge rally in California over the past few months. Once you realize he’s the purple-blue slices, you can see he has a whole quarter of the big pie but only a tiny little slice of the November pie.
But again, you have to move your eyes back and forth and process the numbers in your head in order to make sense of it all. It’s a very sexy graphic. But it just doesn’t tell the story efficiently.
Just as a talking point, I threw together a sample of how we might redraw that graphic in order to help the reader understand how each candidate’s polling results have changed since November. The tool to use in this case, is not a pie chart, but a fever chart.
Again, click for a larger view.
Right away, you can see Romney’s number spike and then sink a bit. You can see the enormous gains made by Santorum. You can also see how Ron Paul‘s been on an uptick. You can easily see that Newt Gingrich has lost nearly half of his support since November.
Each of these six fever graphs are to the same scale — the little grey lines running behind the chart, in fact, show 10 percent increments. I didn’t bother to label the scale because I was labeling each data point anyway. If you’re labeling each data point, then scale labels are redundant.
Also notice how I’m emphasizing the most recent polling number with a black highlight box. I’ve used horizontal spacing to group the candidates’ data, so color coding is no longer necessary.
We want to show “undecided” and “other,” of course, so I charted both of those data sets onto the same grid. The only thing I’m dissatisfied with in my new chart is the intro copy appearing on the right. I’d much rather it appear on the left. But I want to lead with Romney’s data, not “undecided” and “other.”
Possibly the biggest problem with this chart — and the original by the Press-Enterprise — is that there is no label describing the margin of error of the polling data. Or the origin of the polling data, for that matter. If I were the graphics editor at that paper, I’d insist on adding a small note about each. If we can’t find either of those, then I’d insist on killing the chart.
While my version might not be quite as colorful as the original version, I think it tells the story better.
Let’s pay attention to how we handle polling data this political season, shall we?
A few pointers:
- Never run polling data without citing the source.
- Never run polling data without showing the margin or error in percentage points.
- Never force your reader work hard to find the most important data in your chart.
- Rather than decorate your chart, find ways to tell the story more efficiently.
- If you’re showing a single point of data change over time — in this case, a poll number — then don’t use a pie chart. Go with a fever graph.
- Pies are only good for showing the components of a whole. They’re not so great on comparing those components.
That front page image is from the Newseum. Of course.