Introducing Hate, the Map

In a lot of ways, the Geography of Hate affirms what we already know: Americans are fucking racist. Homophobic and ableist, too.

But while that may not come as any great surprise, the map reveals a startling bigotry coursing beneath our preconceived notions of just where in the US hate is harbored most. Americans, it turns out, fall racist and homophobic and ableist, and are apparently vocal enough about it to spout off bigotry on social media, in no real discernible pattern, though it's often where we least expect bigotry that we find it rearing its ugly head.

The visualization comes way of Humboldt State University's Dr. Monica Stephens and the Floating Sheep--the same group that made a map of post-election Twitter hate speech. It comprises 150,000 geo-coded hate tweets flagged between June 2012 and April 2013 for including the word "chink," "gook," "nigger," "wetback," "spick," "cripple," "dyke," "fag," "homo," or "queer". At first blush it's awfully depressing, a real day ruiner, or worse. Click around and most slurs--not all, but most--see the intercontinental US pocked by deep reds, the research team's translation for "most hate." Jesus Christ. Is it 2013? It can't be 2013.

But, really? That can't be right, can it. Surely something's off. How can we be sure "positive" uses of an otherwise hateful slur (e.g., “dykes on bikes #SFPride”) weren't inadvertently swept up in the Geography of Hate? Contextualiztion is crucial--is everything, really. Did Stephens' team allow for it?

They did. In fact, this is why they used humans (read: Humboldt State students), not machines, to analyze the entirety of the 150,000 offending tweets, all drawn from the University of Kentucky's DOLLY project. (It was also very much the reason the project got underway in the first place, as the Floating Sheep got a fair deal of flak over whether their post-election map contextualized hate rigourously enough.) It was a matter of avoiding "any algorithmic sentiment analysis of natural language processing," the researchers write, "as many algorithms would have simply classified a tweet as ‘negative’ when the word was used in a neutral or positive way. The students were able to discern which were negative, neutral, or positive."

As such, the map only includes those tweets used in explicitly negative context. Like so much of modern life, it's an uncomfortable truth perhaps best summed up by the late George Carlin.

To further tighten their project, and to further deflect trolls, Stephens and crew normalized their data set to correct for tweet density. In other words, the scale "accounts for the total Twitter traffic in each county so that the final result is something that shows the frequency of hateful words on Twitter," as Reuters put it. This implicitly adjusts for population density. And it's here that somewhat unexpected hate begins bubbling to the surface.

Take "nigger," pictured above. That's a lot of red--like, half the country--but note how scaled out the view is. (This is the furthest out I could reach using multiple browsers.) Scroll in, and what you start to see are the places, the counties that many of us would expect, rightly or wrongly, to cultivate deep-seated prejudices not even registering a blip, and vice versa. For example, Chicago, which is far and away one of the most segregated major cities in the US (if it isn't the most segregated major city in the US), saw no instances of the n-word being tweeted. Same goes for much of the Deep South, a surprising swath of which bears no shading whatsoever, which is to say use of the slur sits below the national average. But then look at the Quad Cities, the Upper Peninsula, and Austin, among other places, and you get a far diferent picture. I could go on with more examples using other slurs, but I won't. You get the idea. Poke around for yourself.

But let's talk about sample size. In the big scheme, what's 150,000 tweets? Hardly much of anything. Netting, say, 4 or 5 million tweets, if we set aside manpower for a moment (honestly, how many undergrads would it take to closely analyze that many tweets?), would doubtless reveal a far more accurate portrait of explicit hatred. Besides, with a certain level of assurity it can be presumed that many folks in Montana or Arkansas, say, who are particularly apt to voice discriminating views aren't going to be registered on Twitter. For all its insights, the Geography of Hate is far from perfect. "Ultimately," Stephens admits, "some of the slurs included in our analysis might not have particularly revealing spatial distributions."

Even still, the preliminary returns of what could be the start of a truly massive guage on hate speak are valuable in that they illustrate both "the significant persistence of hatred in the United States" and the "ways that the open platforms of social media" have been taken to in service of spreading around these ideas. Maybe in another two years, should the Geography of Hate project carry on, we'll see a shift. Until then, it's safe to assume that too many Americans will still be dicks.

Reach Brian at brian@motherboard.tv. @thebanderson