I came across a startling map on Maps Mania today. Back in May 2013 Monica Stephens, and colleagues, published online maps showing twitter data that contained hateful words related to homophobia and racism. The message communicated by the maps is quite strong and shocking – racism and homophobia are extremely strong in the eastern half of the United States. The problem however is that these maps lie and are artifacts of poor cartographic methodology.
The maps have been online for over a year and a half with many critiques stating that something may be wrong with the maps and the similarity to population distributions. The authors even refer to XKCD’s population map but still refuse to consider that something may be wrong. This would all be excusable from an amateur cartographer yet these maps are promoted by geography professors. In their replies to critiques they state that they have normalized their data correctly (hateful tweets divided by total tweets – per county) and therefore are not susceptible to the problem of representing population density as in XKCD’s example. Their error is ignoring the density of counties.
Below is their map of all hateful homophobic terms:
Map showing the density of counties in the US while reporting to make a statement about homophobic distributions.
The red mass on the eastern half of the U.S. is the result of a density calculation being applied to the county centroids. The problem is counties are smaller and closer toghether for the eastern half of United States. Points closer together aggregate and accentuate ‘hate’ ratios/weights to give deeper red colours. The same data values spread out would not yield the same density! The authors of the maps explain that they normalize hateful tweets by total tweets but they then apply a spatial density evaluation.
We can easily illustrate their error by creating a density map of county centroids (or a hate map with the same rate of hate tweets for each county).
This map shows County density. If each county had the same rate of homophobic tweets this is what their map would look like. Looks familiar?
This looks exceedingly similar to the homophobia map above in terms of distribution. How would the map look if we normalized the values by the county area?
This is the equivalent map but normalized by area of county. The bias is gone. Using density still doesn’t make sense for this type of data.
We now see more homogeneous purple on the East Coast with redder blotches on the West Coast. Regardless of this being better, using a density map to represent centroid mean values incorrect and reflects a poor understanding of spatial analysis. Location matters for density maps. The location of individual events more importantly. The manner in which the data should have been mapped to begin with, using the same data (joined back to counties) is fundamentally more boring:
Darker areas have higher hate tweet ratios. Note the large amount of grey where no data is available. This map contains the same data as the one presented by the ‘Geographies of Hate’ but mapped correctly. Pretty boring no? This won’t get you any media attention.
This is a serious error and big embarrassment to geographers and academics. These maps gained a lot of attention and were widely reported by many mainstream news sites (The Guardian, Time, ABC, The Verge, NBC, … ). The authors should have applied some critique introspectively before sharing these strongly suggestive maps to begin with. As professional geographers they should have noticed this error. That these maps continue to be displayed is disgusting. It’s propagating a false myth that the East Coast is more racist and homophobic than the West.
Cartographers fear the spread of the casual cartographer due to mapping’s complex spatial concepts being misapplied. The media would question a casual cartographer, unfortunately this rag of a map came from careless academics spreading, and continuing to do so, a completely false message.
Update
I have added an interactive map that shows the difference between the homophobia map and county density.
