Humans prefer their weather in something of a Goldilocks zone—a not-too-hot, not-too-cold temperature window which not only affects our physical comfort, but also our mood. During heat waves or deep freezes, tempers fray, patience wears thin, and behavior can suffer. Now, in The Lancet Planetary Health, has found that this holds true not only in our in-person interactions, but online too. As temperatures rise or fall above or below a comfort zone of 54ºF to 70ºF (12ºC to 21ºC), online hate speech in the U.S.—at least on Twitter—increases accordingly.
The research team, led by Leonie Wenz, working group leader at the Potsdam Institute for Climate Impact Research (PIK), took a deep dive into online speech to reach their conclusions, beginning by vacuuming up more than 4 billion geolocated tweets posted in the U.S. over a six-year time frame from May 2014 to May 2020. They programed an artificial intelligence algorithm to scan the tweets for hate speech, which they defined, according to United Nations standards, as any communication that “attacks or uses pejorative or discriminatory language with reference to a person or a group on the basis of … their religion, ethnicity, nationality, race, color, descent, gender or other identity factor.”
That’s a broad definition and the algorithm could sometimes be flummoxed by it. While the researchers could train the program to recognize hateful words and terms, some have multiple meanings. Notably, the software had to learn the meaning of the N-word. One variant of the word, which ends in “-a”, for example, has been “reappropriated as a type of endearment in some communities,” the authors wrote. In that case, they taught the software to look for surrounding words that were “aggressive or derogatory.”
Overall, just over 75 million tweets—or 2% of the four billion total in the six-year window—analyzed by the algorithm qualified as hate speech. But exactly when the tweets occurred and where they originated could vary widely. The study geolocated the source of the tweets containing hate speech to 773 different U.S. cities and cross-referenced that information with what the temperature was in those places on the date the tweet was posted
In general, the study did not find that any one city or region produced more hate tweets than any other; the critical variable they did find was all about the thermometer. The fewest hate tweets occurred in a narrow six-degree temperature range of 59ºF to 65ºF (15ºC to 18ºC), within the identified broader comfort zone. Outside of that 54º F to 70º F sweet spot, things could vary widely. On extremely cold days, for example—more common in the northern tier than elsewhere in the country—when temperatures ranged from 21ºF to 27ºF (-6ºC to -3ºC), hate tweets increased by 12.5%. On extremely hot days—especially in the desert southwest—when temperatures maxed out between 108ºF and 113ºF (42ºC to 45ºC)—hate tweets rose by 22%.
“Even in high-income areas where people can afford air conditioning and other heat mitigation options, we observe an increase in hate speech on extremely hot days,” said Anders Levermann, head of complexity science at PIK and a co-author of the study, in a statement accompanying its release. “There are likely limits of adaptation to extreme temperatures and these are lower than those set by our mere physiological limits.”
That’s not to say we don’t adapt at all. The study divided the 773 cities from which the tweets originated into five different climate zones: cold, hot-dry, mixed humid, hot-humid, and marine (or, coastal). Broadly, they found that increases in hate tweets varied, with, say, people in the cold region—which covered most of the northern part of the 48 contiguous states—showing less of a bump in online misbehavior during an extreme cold snap than people in the hot-humid region, who would not be as accustomed to sudden thermometer plunges.
“This could suggest that the hate tweet increases are dependent on the temperatures we are used to,” the authors wrote.
Limitations in the study didn’t allow researchers to use geographical information to tease out any differences in weather-related hate tweeting depending on socioeconomic status, faith, race, political party membership or more. “Groupings based on income, religion and partisan [affiliation] are not perfect since cities are never perfectly homogenous,” they wrote; their geolocated data, however, did not control for those factors.
While it may have been hard to tease out exactly which demographics were doing the hate-tweeting, it was not hard to determine who the targets were. The study cites existing research showing that 25% of Black people and 10% of Hispanic people have been subjected to race-based online harassment. These communities are also among the most vulnerable to the impacts of extreme weather, made worse by climate change. Members of the LGBTQ community are also four times more likely to report online harassment than others, the study found. These same groups, the authors warn, are the likeliest to suffer from all hate tweets, including temperature-related ones—and that poses a danger to their well-being.
“Being the target of online hate speech is a serious threat to people’s mental health,” said Annika Stechemesser, a doctoral researcher at PIK and a co-author of the study, in a statement. “The psychological literature tells us that online hate can aggravate mental health conditions especially for young people and marginalized groups.”
That threat will only grow, the authors warn, as human-caused climate change worsens and temperature extremes become more common. “Assuming little adaptation and similar communication patterns,” they write, “this would mean that hate expressed online could increase under future global warming.” Hate is a uniquely human quality, and climate change is one of our most regrettable handiworks. Together, they make for a very nasty pair.
More Must-Read Stories From TIME