A map of hate speech on Twitter — sort of

The Twitter Hate Map, unzoomed (top) and zoomed in on the “hotspot” at the Kentucky-Indiana border (bottom). Red spots mark areas with more homophobic hate speech on Twitter. (Geography of Hate/Monica Stephens)

If you’ve seen a certain map about hate speech and Twitter recently, you may have come to the conclusion that pretty much every state east of the Dakotas is vilely racist, homophobic and disability-prejudiced.

In reality, that’s not quite true: Zoom in several levels—like, four or five—and you’ll see that most of the “hate tweets” compiled by researchers at Humboldt State University fall in tiny clusters around the country. In other words, there are pockets of people tweeting horrible words all across America, but few patterns (or conclusions) to draw from them.

“Ultimately, some of the slurs included in our analysis might not have particularly revealing spatial distributions,” Monica Stephens, a geographer at Humboldt State, writes in an introduction to the map on the Twitter geography blog Floating Sheep. “But, unfortunately, they show the significant persistence of hatred in the United States,” particularly on Twitter.

That may be true, and it’s interesting on those merits. The project recorded, for instance, some 150,000 slurs tweeted between June 2012 and April 2013—in Stephens’ words, “a sufficiently large number to be quite depressed about the state of bigotry in our country.”

But ultimately, the “Hate Map” seems to serve more as an illustration of the vagaries of Twitter research than as a useful analysis of hate speech. As Stephens herself points out, she’s working with a very narrow slice of the population: not just “people on Twitter,” which would already cut most of the country out, but “people on Twitter who have enabled geotagging”—a mere 1.5 percent of all Twitter users, and a pool that could act much differently than Americans as a whole.

There’s another problem here as well: the ability of a few Twitter users to skew the results for a larger population. The hate tweets per county are normalized against the total number of tweets per county, Stephens points out—a method that essentially adjusts for population variants. But in a county with very few total tweets each carries more value, and even a few bad tweets could color the whole area red. (Related problem: The hotspots in small counties overlap, which is why the East Coast looks so terrible versus the west.)

Bottom line? The map doesn’t necessarily teach us anything new about where racism, homophobia and disability-prejudice are prevalent in the U.S. But it does start a dialog on that very important issue, and Stephens has promised to follow up with more research.

In the meantime, it does remind us, to quote Stephens, “that the feeling of anonymity provided by Twitter can manifest itself in an ugly way.”