# Close Races

In Close Races, Kaiser of Junk Charts writes about the “racetrack” charts used in the New York Times to compare candidates’ performances in different demographic categories. Specifically, the charts compare votes received by Clinton and Obama in cities, suburban regions, and rural areas. The New York Times charts are donut charts, or pie charts with different circumferential bands representing the different demographic regions. A portion of the New York Times chart is shown here:

I got this chart from Kaiser’s post, and he got it from a “recent issue of New York Times magazine”. The problems with this kind of chart have been described in Kaiser’s recent post as well as earlier posts on the Junk Charts blog. The problems stem from the data being depicted as angular measures, when the relative arc length and area of the ring segments varies greatly even for segments with similar angles.

Kaiser proposed an alternative chart, a two-sided bar chart, similar to a population tornado. My version of Kaiser’s chart is shown below. I’ve added vertical gridlines to help compare bars on either side of the midpoint which are near 50%

The appeal of this chart is that it shows a bar as wide as the voting block, skewed to the side of the winning candidate. You can easily imagine a see-saw, where the longer side outweighs the shorter side. In Wisconsin, the bars are fairly evenly balanced, while in Tennessee, they range from strongly Obama-sided to strongly Clinton-sided.

I made a clustered bar chart in an attempt to improve the comparison of Clinton’s vote totals to Obama’s.

This chart is good for showing comparisons in close races, but I felt it only confused the widely varied results from Tennessee, which the two-sided chart showed more elegantly.

I made one more attempt, this time creating a dot plot from the data.

This works better than the clustered chart if direct comparisons are needed, although I think the two-sided bar chart is better at showing the leanings of the voting populations.

I’ve followed up with some changes and additions here: Close Races 2.

1. alderaic says:

any chart would have been better than the one they do have on the NYT.

It would take the average reader 3 minutes at least to understand how to read the chart, and then draw some conclusions about the numbers (and even then the numbers would not be precise)

I dont know if it was designed to be compact but it sure wasnt designed to be understandable by 99% of the readers

2. @alderaic: I agree that this NYT chart is not easy to extract precise information from. I think the Times graphics team should be credited with trying lots of visualization approaches, some of which (like this one) don’t work too well, but some are very effective.

3. alderaic says:

I agree that they somtimes do have very nices charts. the problem is that the reader needs to understand the chart and they definitely got off track on this one.

Anyway now when I design new charts I get three different person to look at it and tell me what they can get from the chart within 30 seconds… if they dont get anything or the wrong conclusion then the chart is incorrect and people wont look at it.

4. Good approach. Who do you use to beta test your charts? Co-workers, family members?

5. Colin Banfield says:

Jon, I’ve never particularly liked Tonado charts because it’s harder to compare differences among multiple series. However, it does allow you to see differences between values within a single series AND make comparisons between two different series, especially when the differences are not close. Intonaco’s chart is a good solution, although it appears that the difference is Obama minus Clinton and not the other way around. The negative values on the chart is also a problem. How can Clinton lead by a negative percentage?

Note that the dot plot, as shown, is misleading because it draws lines between points having no relationship. I find myself comparing lines instead of individual points, which is what the dot plot is intended to do.

6. If Andreas used the trick to display only positive numbers on the axis of his chart, it would look fine. Andreas, if you’re following, use a custom number format like this:
0;0;0;@

Tornado charts are harder to read than the reader thinks. It’s difficult to compare actual values on either side of the center line. But you do get a good view of the shapes of the distributions.

The lines in the dot plot are helpful to keep track of which series goes where, that is, to see the shape of the distribution. It would have been better to use lighter colors for the lines, so they detracted less from the markers which show the real data in the chart.

7. alderaic says:

to test my charts, I use first my VP (planning etc…) so if he doesnt get it I know no one will, then I use an administrative assistant, if she gets it, everyone will, then I use another random person.

and it works fairly well, when I started in the company I am working for two years ago, they were not using any charts, just printing huge reports with so many columns of numbers that they would spend the whole monday analyzing the previous week, and another one or two days discussing it. Now they are using a lot more charts and do spend at best the whole monday analyzing and discussing the data (I still hope for quick monday morning but it is already a lot better than it was!)

8. Nice work! All three charts look quite professional.