This morning, a friend emailed me a link to this story at the Daily Mail, which contains a number of beautiful maps on America’s infrastructure networks. Go check it out; I’ll wait here while you spend fifteen minutes going “ooh!” and “ahh!” at all the images.
While they are beautiful, they are not without problems. Lovely visuals, deep conceptual errors. Let’s start with the one that stuck out to me, the visualization of job losses. At least, I assume it’s job losses. The caption at the Daily Mail reads, “Unemployment: The number of job losses in the U.S. chronicled in this stunning image.” Unemployment is different than job losses, so I’m not sure which is being mapped, but I suspect the latter.
(EDIT: I’ve lately found the video that these maps come from. This is actually a map showing the distribution of manufacturing jobs in the 1990s. In the video, the dots wink out — turn black — to show the decline of the manufacturing sector. Seen at about 12 min, 50 sec in the video here: http://www.pbs.org/america-revealed/episode/4/)
A dot density map is approximately the worst way to look at a data set like this. The author(s) took the number of jobs lost in each state, converted that to a certain number of dots, and then scattered those dots all throughout the state. The result is misleading in several ways. First off, this is really a map of job loss density. The brightest areas are where the dots are clustered, which means a lot of jobs were lost in a tight space. Texas lost a lot of jobs, probably as many as Ohio or Illinois (I’m not going to count every dot, but they probably have similar numbers). But the latter two states look worse off, because they have about the same number of dots crammed into a smaller physical area. Actually, it’s not even a map of job loss density, because the areas of the states are distorted due to curvature of the Earth in this view. Washington, for example, is a lot smaller than if we were looking at it from directly above, and its unemployment picture therefore looks worse. So, this isn’t really a map of “job losses” so much as it’s a map of “job losses divided by the size of each state, if you distort the sizes of the states a lot.”
So, reading the dot concentrations will simply mislead us. But what about the number of dots? If we painstakingly counted them, we could certainly find out at least how many jobs were lost in a state without having to worry about this density issue. Well, the second big problem is that these data aren’t normalized to population. There are a lot more dots in Illinois than in Wyoming. This is because Illinois is a very populous state, whereas no one lives in Wyoming. No account was taken, seemingly, of population differences. Some states are being hit harder by the recession than others, but all you can tell from this map is that places that had more people lost more jobs. I quickly found this table prepared last year and it points out that Nevada has been hammered by the downturn, losing 8% of its jobs. Ohio, on the other hand, lost only 2.6%. But when you look at the map, which state looks worse off?
Lastly, jobs weren’t lost evenly throughout each state, so why scatter the dots evenly? Probably because the author(s) only had state-based data, but making some account for population locations would be nice. Why show a sea of lost jobs in eastern California, which is mostly desert, mountains, and unpopulated forests? The exact locations of the dots are meaningless anyway, since they’ve been distributed randomly in each state.
This map should have been a choropleth of job losses per state divided by population. It’s not nearly as sexy, but it’s also not seriously misleading.
(EDIT: As said, this critique was based on an inaccurate description of what this map is about, and I apologize for not doing my research. Much of the critique still stands — showing dots still seems odd for this data set. It’s a map of “manufacturing jobs divided by the size of each state, if you distort the sizes of the states.” The need for normalization is probably less, as well, though it couldn’t hurt.)
The dot density map was what galled me. On the other hand, my friend Chris, who sent me the link, was bothered by the map of wireless access towers:
I can’t say for certain without hearing the accompanying narration (EDIT — see below), but the author(s) very likely received a data set which had point locations for towers and broadcast power, and simply made circles proportional to the power. That’s all quite reasonable, but the map very much looks like it’s trying to portray the actual signal coverage areas, and that’s a very different thing. Most any electromagnetic signal coming from a tower is not going to move in a perfect, even circle away from the transmission point. It gets distorted by a lot of things — buildings get in the way, so do mountains and other terrain features, and the Earth’s magnetic field also affects it. And if it’s been broadcast by a directional antenna, the signal starts out stronger in some directions than others. Presenting these transmissions as circles is an overly idealized view of how they work. Even if the authors were only using circles as a symbol for power, rather than suggesting these are areas of coverage, a lot of people are going to misunderstand this as the latter. Every reader brings their own interpretations to the map, and it’s not always the one the author wants. The best you can do is try and head them off.
(EDIT: Again, I’ve now looked up the relevant section in the video to figure out what they map is actually of. It’s found at 40:20 in this video: http://www.pbs.org/america-revealed/episode/4/ — the narrator doesn’t actually say what the map is of; it just shows while he talks about how your wireless signal is bounced between towers, so I’d say the critique above stands as is. If anything, I’d say my point about misunderstanding the map is even stronger, because the program really does leave it up to you to figure out what’s going on.).
All the maps on this site have a pseudo-realistic appearance, and they’re even discussed as though you’re “seeing what the nation looks like from the skies.” My colleague Marty Elmer pointed out to me that this realism means an increased expectation of accuracy. If you’re telling me that I’m floating above the US, really seeing the job losses or broadcast signals, I’m likely to believe that this is really how things are distributed on the ground. That the signals really are circles, or the jobs really were lost in the woods of northern Michigan. I’m less likely to take the thematic data as an abstraction because the base map, with its fancy lighting effects and clouds, doesn’t look very abstracted. Generalization is not just about redrawing your linework to the right scale; it’s about credulity. This was one of the biggest points I would emphasize to my students back when I lectured on the subject. Visual abstraction needs to match data abstraction. Readers seeing a highly simplified visualization will assume the data are likewise telling a highly simplified story. If they see a very realistic and detailed basemap, they’re likely to assume that the data have been treated similarly.
In the end, we’re left with beautiful, but potentially (or sometimes outright) misleading images, and that’s a travesty. These are for a television special, and besides going in front of millions of eyes on PBS, these works will very likely go around the Internet to millions more. The maps are lovely to look at, and this means they’ll get a chance to misinform many, many more people. It’s a shame, because it’s a squandered opportunity to inform people about actual facts. Imagine pairing quality visuals with well-thought-out data treatment and map concepts, and how far that would go. But infotainment isn’t about the substance, just the style. See the headline to that Daily Mail link, for example — “Secret corpse flights,” as though this transport of bodies were illicit, rather than a routine movement of your loved ones to their desired resting place. It makes a good story, even if it’s not true.
Also: poor Alaska and Hawaii. They remain, as ever, second-class states.
One Nice Thing: I could go on for a long time about how excellent these look. Maps need to be beautiful if you want people to look at them and spend the time to learn something off them. I’d love to see the author(s) continue to do this great work, just with better data and ideas for portraying it.