Fear and loathing: Tracking threatening language online

x Sponsored content on Research Live and in Impact magazine is editorially independent.
Find out more about advertising and sponsorship.

The world feels increasingly unstable as the years roll by. A pandemic, the war in Ukraine and an increasingly toxic political environment are just some of the factors that have raised anxiety levels and decreased optimism across large swathes of the population. Portents of doom abound online, decrying the latest supposed threat to humanity.

What is the cumulative impact of this torrent of bad news? Does it change people’s outlook on life, society, their fellow human beings and country? One recent study by researchers at the University of Maryland and Stanford University looked to understand the implications of threats broadcast through mass communication channels in the media and online. They used a computational linguistic tool to index threat levels from text, allowing the researchers to examine how the prevalence of threatening language in the media over the past century has impacted cultural, political and economic shifts.

Michele Gelfand, professor in cross-cultural management and of organisational behaviour at Stanford University, and one of four lead researchers on the project, says that the study came about because of a dearth of available tools and evidence on the subject. “We have been studying the impact of threat on human psychology for some years, but we realised there were few methods developed to track threat in real time in public media and its impact on human groups,” she explains. “We wanted to also develop this tool so that people can use it in their own lives – for example, to see how much threat talk one is exposed to when online or when listening to speeches or reading the newspaper.”

The research team created a ‘threat dictionary’ to help develop a semantic measure that tracks the form, frequency, and magnitude of communicated threats. This involved scanning documents for the presence of select keywords, identified through a technique to determine how words related to the concept of ‘threat’ cluster together across different platforms, such as Twitter, Wikipedia and raw webpage data website Common Crawl. The researchers identified 240 words, including: attack, crisis, destroy, disaster, fear, meltdown, outbreak, suffer, tension, toxic, unrest, unstable, and violent.

“Developing a threat dictionary can help us detect threats in many settings in real time and track it across many years,” Gelfand explains. “It can be used in a variety of settings, from analysing the impact of threat on the stock market and examining how leaders use threat to motivate, to understanding how threat spreads across platforms online, among many other uses.”

In this study, researchers used the dictionary to examine trends across US newspapers over the past 100 years, and found a general reduction in the prevalence of threatening language over time, with spikes for various major events, such as wars, pandemics and global catastrophes. For example, World War I and II both caused a significant spike in the model, and similar trends were seen for events such as the terrorist attack on the World Trade Center in New York in 2001. But even with these three events, threatening language did not rise above that seen at the turn of the 20th century, reflecting a broader march towards a more peaceable existence.

The prevalence of threatening language in the media has other impacts, too. “Threat talk, whether real or imagined, has predictable consequences,” explains Gelfand. “It tightens cultural groups and makes them more ‘group oriented’; it relates to conservative shifts in voting and xenophobia, and it relates to lower stock-price performance and lower innovation.”

Gelfand says that the relationship between cause and effect in these cases could be blurred, with threatening language sometimes a cause of societal tensions and sometimes an effect of major crises, such as World War II or the 2020 pandemic. For example, poor stock market performance could be triggered by news of threats, while market downturns could also cause an uptick in threatening language appearing in the media.

In an increasingly more dangerous world, what about the next few decades? Will we see a continuation of positive trends from the 20th century, or will there be rising tension across the world and, as a result, an upsurge in the use of threatening language?

The research also modelled responses for two decades into the future – broadly until 2040 – using an autoregressive integrated moving average (ARIMA) statistical model, which uses previous trends to make predictions about the future prevalence of language contained within the threat dictionary. The result? A gradual rise in threat levels, closer to those seen in the first half of the 20th century, reflecting an upwards trend in threat levels from the past eight years. It should be noted that the trend is uncertain and should be interpreted carefully; there is no guarantee of a more dangerous future ahead. But given recent world events, rising tensions look a safer bet than usual.

What’s next for the research team? “We’re working to see how threat talk is motivating extremists in online settings,” Gelfand says. “We’re also interested in analysing how news programmes differentially use threat talk and how this affects their viewers. We’d like to expand the dictionary so it can be applied to other languages.” With plenty of economic and political turbulence on the horizon, there is undoubtedly scope for further investigation.

Virginia K. Choi, Snehesh Shrestha, Xinyue Pan and Michele J. Gelfand, ‘When danger strikes: A linguistic tool for tracking America’s collective response to threats’, PNAS, Vol. 119, No. 4, January 2022

This article was first published in the January 2023 issue of Impact.