Last week, Jessica Ball of the American Geophysical Union’s Magma Cum Laude blog pointed out this map of natural hazards in the U.S., published by the New York Times:
The map has many of the design virtues common in graphics from the Times (clean, focused on the data, clearly labeled, small multiples), but when I first viewed it I had trouble parsing the data. It took me a while to figure out that hazard was indicated by color, and the size of each circle denoted ancilliary information (population).
It turns out that our eyes & brains perceive area more strongly than color. Here’s a list of ways to visually compare quantities (described by Bang Wong in his Points of View column for Nature Methods), in order from easiest to hardest:
- Positions on a common scale
- positions on the same but nonaligned scales
- lengths
- angles, slopes
- area
- volume, color saturation
- color hue
I suspect I’m not the only person who would consider the size of the circles more important than the color, and assume size was correlated with hazard. I had to carefully read the key to figure out the proper way to read the map. Perhaps using size to encode hazard, and color (or opacity, or icons) for population would have worked better.
I have no idea what the top map is supposed to show. Are they adding the probabilities of all the risks they enumerate? Are they taking into account autocorrelation? Are they trying to illustrate severity of the risk? (Hail can range from golfing annoyance to serious problem, but even so it’s not likely to reach the severity of hurricanes.) Are they accounting for the nature of the sites? (An earthquake at inhabited site near nuclear power plant is more “severe” than one of similar intensity in a sparsely populated place.) It looks like they are just discussing cities rather than geographical areas. If so, wouldn’t a table be the easier way to present the data?
In fact, the difficulty is here to represent two quantitative criterias. One for the population, and one for the “scale of hazard”
The choice of proportinal circles for such a quantitative data is a good one, but the choice of colors is less accurate as it’s realised here. Order in a color scale is readable for “camaieu” (the three small maps confirm) , but not between different hues, where visual perception is more qualitative. And Green/red is more percepted as an opposition !
The main question here, is in my opinion : is it accurate tou quantize population on a map titled “Where to Live to Avoid a Natural Disaster”?
For graphic semiology, it’s possible to find many good references from the french Jacques Bertin. http://www.cnig.serveur-1.net/fiches/fiche40.doc is an interesting abstract.
DK:
I think a table would be better, but requires a city of interest. I think the problem is that there’s no real correlation between hazards and city size, so combining the two in one map doesn’t make much sense.
Pascal:
Good points. I can get a little, shall we say, “animated” discussing the use of color in data visualization, so I’ve been putting it off. I’ll probably end up doing a series of posts. I’ve read parts of Bertin, and I’ll be sure to include him in another planned future series of posts about good graphics resources.
Yeah, the color is not type of hazard, it’s the unexplained “scale of hazard,” but I’m not sure it’s quantitative… looks like it might be categorical. There doesn’t seem to be intermediate colors. But then how do they rank the top and bottom? Maybe quantitative but binned. Very confusing.
To get to the point of this post though, if color was hazard type, maybe using different icons indicating different hazards, and different areas to indicate a quantitative value like population, would work better. Where does shape fit into the scale above? Seems to me it would sit somewhere around 4?
(And as an Alaskan, I have to point out that hurricane force winds are quite common in Alaska, but they aren’t classified as hurricanes. We’re proud of our hazards! And no mention of tsunami hazard?)