Thursday, September 7, 2017

In visualization, captions are as important as graphics themselves

(Updated on September 8 and 9. Go to the bottom of the post)

Data visualization isn't just about visualizing data, but also about writing headlines, intros, captions, explainers, and footnotes. I'm right now closely following the news about hurricane Irma —I live in Miami!— and feeling both amazed and terrified by the many great graphics news organizations and independent designers are publishing. As I've just tweeted, beauty is sometimes correlated with terror.

Anyway, I've just read a very good graphics-driven story in The Washington Post. This is its first map:

This is its caption:

I'm no expert in weather forecasting, but I believe that this is inaccurate. To learn why, go to minute 14:30 in my keynote at Microsoft's Data Insights summit. Here's some of what I said there:

Maps based on cones of uncertainty are quite problematic, as this article by Jen Christiansen, and this other by Robert Kosara explain. Among other reasons, some people don't see in that cone the possible range of paths the center of the hurricane can take, but the size of the hurricane itself.

This happens event to those who, like me, do know how to read this kind of map. I need to consciously struggle with my brain's inclination to see a physical object, and not a probability range. Why? I don't know for sure, but I'll make a conjecture: it's because the representation looks pictorial. The rounded shape of the tip of the cone roughly resembles the shape of a hurricane.

This map is made even more confusing if a black line is placed in the middle of the cone. Just read tweets like this. People may see that line not as a visual aid to emphasize the center of the cone (right), but as the most probable path (wrong).

Going back to the caption, the reason why it sounds wrong to me is related to something most of you probably aren't aware of: the cone of uncertainty doesn't represent the range of all possible paths the hurricane could follow, based on simulations. This excellent paper explains that the most common cone, the one by NHC, “accurately predicts the ultimate path of the tropical cyclone’s center about 2/3 of the time (J. Franklin 2005, personal communication). In other words, one out of three storm centers directly impact areas outside of the cone.” That's a 66%-33% chance.

Therefore, the caption could say something like this: “Based on predictive simulations of past hurricanes, there are 2 out of 3 chances that the path of the center of the hurricane could be anywhere within this cone, and a 1 out of 3 chance it will be outside of it.” This is longer and clunkier —I'm sure that any copy editor in the audience can improve it!— but truer to reality.

This other map shows the actual uncertainty of predictive simulations quite well; notice the faded lines, corresponding to less probable (but still possible) paths:

UPDATE: It seems that NOAA is listening. See the explanation that they have been tweeting. It ought to be published next to every single cone of uncertainty map out there:

UPDATE 2: The map below, by meteorologist Ryan Maue, is far better than any cone of uncertainty map if your goal is to inform the general public about the risks posed by wind. See it animated. The scale is predicted maximum wind speed in mph.