Cities With More Hateful Tweets Have More Hate Crimes, Study Finds

High levels of race-based discrimination online correlates with offline violence in cities across America, new study finds.
Twitter bird on a screen
Image: LEON NEAL/AFP/Getty Images

Hate speech on social media has long been considered an internet problem, one for users to tolerate and content moderators to deal with. But a new study connects online and offline discrimination, linking hate speech on Twitter to rates of real-life hate crimes.

Researchers from New York University’s Tandon School of Engineering and College of Global Public Health studied tweets from 100 U.S. cities, finding that higher rates of race-based hate speech on Twitter related to higher rates of hate crimes in cities.


Using machine learning, researchers used Twitter’s API to analyze 1% of public tweets from 2011 to 2016, a total of 532 million tweets. They designed an algorithm to sort out two types of tweets using race-based discriminatory language, which they called targeted—those that directly espoused discriminatory views—and self-narrative—which simply described or commented on discrimination.

Researchers found that cities that had more targeted discriminatory tweets had higher rates of hate crime, while more self-narrative tweets predicted lower rates of hate crimes. While stressing that these results don't imply a causal relationship, researcher Rumi Chunara said that there’s a lot of room for future study.

“I think there’s a sentiment in the targeted tweets that is likely related to fostering an environment for these crimes,” Chunara told Motherboard in a phone call. “Meanwhile, having productive conversation might actually improve culture and outcomes.”

The study was limited to tweets in English using discriminatory language based on race, ethnicity, and national origin. Researchers chose cities from all 48 states that varied widely in size, socioeconomic makeup, and urbanization. Some discrimination was common across all cities, while others were more city specific.

This research could be applied to discrimination surveillance and content moderation, the study said. While the findings were presented at the Association for the Advancement of Artificial Intelligence Conference on Web and Social Media and submitted for peer review, the study has not yet been published.


"We need to consider the online world as part of our environment"

Twitter has been criticized for being slow to ban white supremacists groups and for allowing the proliferation of targeted harassment campaigns on the platform. Twitter told Motherboard last month that it is still researching whether it should continue to allow white nationalism and supremacy on the website.

At the time, Vijaya Gadde, Twitter's head of trust and safety, legal and public policy, told Motherboard that Twitter believes "counter-speech and conversation are a force for good, and they can act as a basis for de-radicalization, and we've seen that happen on other platforms, anecdotally."

Twitter spokesperson Raki Wane said that the company has been working to improve its content moderation and that they “always do look at the studies.”

“We always look at context when it comes to reviewing content,” Wane told Motherboard, noting that they have users and teams all over the world, tweeting in many different languages and cultural contexts.

Chunara said that the team analyzed tweets over a five-year period in order to try to standardize the results. It remains hard to establish a causal relationship between tweets and hate crimes because of general national trends like increasing hate crime rates, discrepancies and changes in the way different cities report hate crimes, and changing Twitter practices that make analysis difficult. These differences also makes it harder to link specific hate crimes with instances of hate on Twitter. Despite these complications, the trend held across the range of U.S. cities.

In future research, Chunara hopes to widen the scope to include health outcomes in addition to violence. She noted that not all discrimination necessarily leads to crime, but it could lead to “risky coping behavior” and negative mental health outcomes. Chunara connected the public health idea of the “built environment” to this latest research. “We need to consider the online world as part of our environment,” she said.