Marathon season is almost upon us! So gird yourself for pictures of grannies with more vigor than you and men with bloody nipples. And if you’re a runner, pray to whatever god you worship that the temperature is within a hair of 53 degrees.
This infographic by Gene Lu illustrates why. Granted, it’s not a great example of data viz, but it does reveal some interesting things. The horizontal bands on the chart show the times at which various runners finished in the New York Marathon; the larger the bubble, the more runners finished at that time. Meanwhile, the vertical strings of bubbles represent the temperature that various races were run at. So, for example, you can read the graph horizontally, to see how many runners finish at between 5:00-5:30 when temperatures range from 42-63 degrees:
If you look at the tiny, tiny little bubbles on the bottom of the chart, you see that the perfect temperature is 53 degrees — that’s when course records fall, and when the greatest number of runners finish at sub 2:30 times. But looking beyond that, you’ll see that the horizontal bubbles don’t vary much among those finishing at elite times. That’s because top runners are well conditioned to deal with heat; despite feeling more pain, they perform at their peak. The real people that suffer are the amateurs — as you can see by how much variance there is at the slower time brackets. Not that you want it to be cold — temps below 45 degrees have about the same sort of slowing effect on the field as temps approaching the 60s.
Now, you might be confused by the chart — we certainly were. It’s worth noting why: Infoviz people often hate showing relative amounts as circles. The simple reason is how circles work. Remember that formula for a circle’s area, A=?*r^2? Well, that just means that area goes up as the square of radius — so just increasing the radius of a circle by 40% raises its area by 200%.
When you apply that to data viz, circles just aren’t great for showing variances within a tight range. That’s simply because a bunch of circles might look almost the same size but their areas might actually represent strikingly different figures. Generally, a much better tactic for showing relative percentages is either just to show the numbers, or do side-by-side bar charts.