Maps are interesting, where the same data can be used to tell many stories. Displaying United States census data as numerical counts or as density portray American population distribution dramatically differently, as a country that is half empty, or half full.
On the left is a segment of a map by Nik Freeman featured by the Washington Post; on the right is a map by Kaitlin Duck Sherwood constructed for io9. Both maps use census data, marking places with zero population in green and places with any population in white. So, why do they look so dramatically different? Freman's map is broken down to the smallest unit of census data, a single block, while Sherwood's maps are aggregate to census tracts.
The maps are in different projections, so the continental outlines don't exactly line up. They also use data from different census years — Freeman uses 2010 data, Sherwood uses 2000 data. While the population of the United States has changed in the intervening decade, the basic patterns stay similar enough to compare the maps. Finally, Freeman's map marks populated locations in white, while Sherwood's are in cream. Overall, these are small differences, paling in comparison to the impact of data aggregation levels. (Both sets of maps include Alaska and Hawaii; I've clipped to just the continental United States for aesthetic convenience. Click the image-credit links for full maps.)
Populated census blocks, with zero-population blocks in green. Image credit: Nik Freeman.
Census blocks are the very smallest unit where data is collected. They're artificial constructs defined by boundaries like roads, creeks, or property boundaries. Freeman points out that the zero-population green blocks identify areas where building is difficult (mountain slopes, swamps, deserts), or living is prohibited (national parks, military bases). Since zero-population is a much different statement than undeveloped, this isn't a map of wilderness, but also a map of highways, commercial districts, steep mountain slopes, and flowing rivers. From the map analysis writeup, Freeman has calculated that the 4.8 million zero-population census blocks work out to 47% of the landmass of the United States, although I'm not sure if this number excludes the 0xxx blocks that are entirely marine.
Populated census tracts, with zero-population blocks in green. Image credit: Kaitlin Duck Sherwood
Blocks are collected together into creatively-named block groups. They're clumped together until they hit a population of between 600 and 3,000 people, never crossing state boundaries. Block groups are then aggregated into census tracts. Tracts are fairly stable agglomerations with generally-consistent borders between census years, with populations between 1,200 and 8,000 people, optimized to 4,000 people. Thus, by definition, census tracts are almost uniformly populated. It makes sense that a map of populated census tracts is almost uniformly white, with very few zero-population bright green blocks. In fact, in the year 2000 census data, fewer than 2,000 census tracts in the entire country have a population of less than 1,000 people, and just a handful with no population at all.
But that doesn't mean the country is all full up with cities everywhere. Instead of population counts that are highly sensitive to the aggregation level, let's consider population density.
Population density per census tract, scaled from zero (green) to 400 (cream) people per square mile. Image credit: Kaitlin Duck Sherwood
Sherwood worked up a few more maps for us using population density. This one is aggregated to census tracts again, but she also put together one using county-wide populations and areas. The maps follow the same basic colour scheme, so green is zero people per square mile, stepping up to 400 people per square mile in cream. Looking at it that way, the United States is almost entirely empty of people, a wilderness speckled with cities. A dense core of cities marks the northeast; the interior cities trace along a road network so tightly you can almost trace the highways. Population spreads out in the west, with the deserts and the Rockies inhibiting large population centers until reaching the far west coast.
But even looking at density depends on how you group the data. Instead of looking at density per census tract, check out population density by state.
Population density per state, scaled from zero (green) to 400 (cream) people per square mile. Image credit: Kaitlin Duck Sherwood
Suddenly, New York pops out as the home of almost everyone, while the major population centers of the Los Angeles basin and the San Francisco Bay Area are diluted by the low-population-density swaths of the agricultural central valley, or the ghost towns in the Sierra Nevada mountains. Texas looks a lot more like cowboy country, with barely any population for the vast ranch lands, and the interior looks downright uninhabited.
All these maps use the same data, yet each tells a different story. Cartographers are storytellers, artists and scientists who shape numbers to suit their needs. A well-done map can have a huge impact, but it's important to remember that any particular way of looking at that data involves choices that change your interpretations.
For more maps, check out Freeman's tumblr, or Sherwood's mashups. If you want to experiment with making your own maps, Sherwood has some tools to build custom Google Maps mashups. A similar map of Canada runs into problems with census data. Of course, I have an ongoing love affair with historical maps of the Earth or Moon. Apparently we're having a US-Population-Maps themed day — here's a very strange demographics-over-time map. Update: Freeman has joined the discussion if you have any specific questions.