Candian researchers have created a “hate map” that splashes a U.S. map with color-coding showing the proportion of all Tweets that contain common hateful words about race, sexual orientation, or disabilities.
The group responsible, based at California’s Humboldt State University, previously faced some criticism for their methodology in mapping racist Tweets concerning President Obama; in that project the researchers identified racist terms algorithmically, meaning the context might sometimes be misconstrued. They say they answered past criticism in part by analyzing tweets by hand to make sure certain words were really used in a derogatory way. In the new study, human readers reviewed 150,000 Tweets in the study sample to confirm the context was derogatory. The results are here.
While the new effort may accurately find hate in Tweets, there are reasons to doubt that the results accurately map hate in the United States. For starters, the concentration of Twitter users is not even across the country, and not all Tweets are geotagged (they only looked at the geotagged ones). Moreover, a lot of Twitter accounts are marketing-based and wouldn’t contain hate words, so those would drown out the signal. Some Twitter accounts are phony. Perhaps most of all, the number of times an individual person Tweets can vary greatly. One guy spewing 1,000 hateful Tweets will make his county look pretty bad.
Overall, the map puts the midwest and southeast in a generally harsh light. By contrast, it says there seems to be very little hate in California, Colorado or Alabama — but lots of it in angry red splashes such as one between Davenport and Iowa City, Iowa. My skepticism is tempered by the unbroken corridor in light blue (denoting “some hate”) generally along the New Jersey Turnpike. When I’m stuck there I don’t like my fellow man either.
To make up your own mind, read more about the methodology here.