America is not that segregated!

To the editor:

I would like to simultaneously applaud and boo the recent interactive online publication, “Mapping America: Every City, Every Block.” The online web interface is impressive and provides a beautiful portal to the important newly published data from Census Bureau's American Community Survey. However, due to some cartographic errors I believe the dot map showing “Distribution of racial and ethnic groups” is quite misleading.

The first problem is that the dots on the map are incorrectly placed. In suburban and rural areas, dots frequently appear in uninhabitable zones such as lakes, rivers, canyons, and national forests. This causes readers to not only wonder how people could live in, on, or under lakes, but also to mistakenly believe that the habitable portions of these tracts are more sparsely populated than they actually are. To correctly apply these techniques the designers should consult additional geographic data sets including land cover data to help place the dots more intelligently.

The second major problem with these maps is that they greatly exaggerate racial segregation within the United States. This is a result of the design decision to place one colored dot for every fifty individuals of a particular race. This approach necessarily implies racial clustering that may or may not actually exist; using these techniques, even the most diversely populated, integrated census tracts will appear quite segregated. For example, imagine a hypothetical suburban census tract with exactly 50 white residents and 50 black residents. The software would place two dots randomly (likely separated by a large distance) within that tract, one for the white residents, one for the black. Unfortunately, even though in reality this tract could be quite integrated, your map would depict it as severely segregated – that’s a problem.

Readers tend to trust maps more than other media, so it is particularly important that cartographic depictions of sensitive topics, such as racial distribution, are given sufficient attention to detail. Unfortunately, in this case that attention has not been paid, and the results misinform your readers about American settlement patterns.



Design first, cartographic common sense second.

I'm a huge fan of the New York Times graphics division; they routinely amaze me with their interactive visualizations. However, the most recent product.... not so much. It overestimates segregation effects and makes a mockery of the dot density mapping technique.

The work is clearly inspired by Bill Rankin's excellent work that depicted the racial composition of Chicago that had also inspired Eric Fisher to employ the same technique in other American cities.


Rankin's map does well to use block level (the finest resolution of census data) in an incredibly densely populated metropolis, while the NYT map attempts to apply the same technique using tract level (much coarser) data to the entire nation - urban, suburban, and rural. Some times things that are good for one thing aren't good for another.

Dot density maps, when made correctly present wonderful depictions of complex spatial distributions. However, to create these maps responsibly, er accurately, it takes a LONG time, requires many data sets, and demands human intelligence. Fortunately, but more likely unfortunately, the wonderful folks at ESRI (the microsoft of mapmaking software) have provided cartographers a chance to make really-bad, innacurate dot-density maps in a fraction of the time it would take to make a good one. This enables cartographers to take great GIS data, and turn it into a steaming pile of visually elegant misinformation easily understood by human beings. The consequence is that instead of helping the audience discover new -correct- knowledge about the world they live in, you help them discover new incorrect knowledge about the world. The only thing worse than no information is misinformation.

In general census tracts cover the entire USA, so, no matter whether you're in Brooklyn or you're in Death Valley, you're in some census tract. In rural areas, tracts are much bigger, more likely to include uninhabitable zones (e.g. lakes, forests, deserts, canyons) , AND population is much more sparse. The main symptom of bad dot maps is that they provide misleading information when dots are dispersed because there is little or no chance of finding the mapped phenomenon at the location of a particular dot.

Simply put the problem with the NYT map of American population is that is grossly inaccurate; check this out:

According to this map, about 450 white people live in, on, or under beautiful Lake Lansing. The problem is that the cartographic robot that placed these dots did so in a completely random way that has absolutely nothing to do with human settlement patterns. The consequences are many.

Just as troubling is that by grouping racial characteristics with different hues symbolizing different races/ethnicities the map presents a much more segregated view of the US. For example, one blue dot in a rural area implies that there is a cluster of 50 black people together, when in reality we all know that these individuals are distributed much differently across the landscape. See below:

This, if not dishonest, is a horribly inept presentation of the human settlement; it implies much more racial segregation in suburban and rural america than actually exists. Sorry. This method necessarily divides up races into small teams of 50 players, then places dots randomly throughout the tract. Super job.


The simple way to prevent this is to use ancillary data such as land cover to help place dots where people actually live...

Here's how one textbook (Slocum et al 2009) describes how to make a dot map:

The first step is to delineate regions in which within which the phenomenon being mapped (human beings in this case) is located; ideally this step considers ancillary information that can assist in determining appropriate locations for dots.

This is pretty much where it all went wrong. The NYT used census tracts as the regions or containers within which people are found. Unfortunately, tracts are:

1) highly irregularly shaped/sized political divisions generally containing between 2,500 and 8,000 people

2) much larger in rural areas, and much smaller in big cities like NYC - the jets are awful by the way (go pats!).

3) inclusive of inhabitable places such as lakes, deserts, canyons, etc. (place where humans don't live)

In the end the product is slick but misleading. It's a map of data, not geography.


Andrew File System Retirement

Andrew File System, which hosts this address, will be ending service by January 1, 2021. Learn about the retirement process, managing your existing files, and alternative services at the Andrew File System Retirement Information Page.