On Twitter, false stories are 70 per cent more likely to be retweeted than true ones. Here’s whyby Philip Ball / March 13, 2018 / Leave a comment
The problem with fake news isn’t just that there’s a lot of it around, but that it gets about more effectively than real news. That’s the conclusion of a new study published in Science that compares the way fake and true news spread on Twitter.
Social scientist Dean Eckles of the Massachusetts Institute of Technology told me that this study, by Sinan Aral (also at MIT, although Eckles wasn’t involved in the work) and colleagues, is “the most comprehensive descriptive account of true and false information spreading on social media that we have to date.” It is certainly a huge effort, using 126,000 stories spread in tweets by 3m users between 2006 and 2017. Twitter supplied the information to the research team.
For the gargantuan task of figuring out which items carried true as opposed to false information, Aral and colleagues crosschecked it against six reliable fact-checking websites, such as politifact.org and factcheck.org. There’s obviously a grey area of “true-ish” news, but the researchers used items where there was more than 95 per cent agreement between their sources on the veracity or otherwise.
They prefer not to call false stories “fake news” though. That term, they point out, has been pretty much devalued by certain politicians, most notably of course by United States President Donald Trump and his administration, who now routinely apply it to any reports that don’t suit their agenda.
The researchers found that, on average, false news spreads faster than truth on Twitter and has deeper penetration, reaching more people. If a news story reached 1,500 people, it did so six times faster if it was false than if it was true. The researchers’ analysis suggested that false stories were 70 per cent more likely to be retweeted than true ones. These findings hold fast even when the team found ways of weeding out automated “bots” from among the Twitter users.
“False political news traveled deeper and more broadly, reached more people, and was more viral than any other category of false information,” the researchers say. They found that false political information increased during the 2012 and 2016 US presidential elections.
“If a news story reached 1,500 people, it did so six times faster if it was false than if it was true”
It’s not clear we should be terribly surprised by any of this. False news is generally created with a purpose: to spread disinformation. So it is tailored to catch the attention. “If one builds false news, it is much easier to introduce catchy elements,” says Walter Quattrociocchi, a data scientist at the Ca’Foscari University of Venice in Italy. It’s also probably designed to pander to the prejudices of the intended audience. You’re not likely to get much traction with a story asserting that the Clintons left a pizzeria without leaving a tip; say instead that they were there as part of a clandestine paedophile ring and you’re away.
And that’s another feature of false news right there. By analysing the language used in tweets, Aral and his colleagues found that such news tends to elicit different emotions in spreaders: not sadness, joy and trust, as in true news, but disgust, as well as surprise.
The surprise comes from “novelty,” which the researchers estimate by comparing the topics of these fake stories with those of other tweets received by a random subsection of users. They found that false news also shows more novelty, according to this measure.
Novelty might not be the only, or even the major, attribute that helps false news to spread so quickly and so far; but it’s consistent with the well-known finding in cognitive science that novel stimuli command our attention more readily.
The notorious echo chamber is surely playing a role here too, although the researchers don’t try to assess that. Typically, says Quattrociocchi, “users acquire information adhering to their preferred narrative, even if this contains false claims, and ignore dissenting information.” Ah, don’t they (we) just.
At any rate, false news is evidently shaped to spread effectively. What’s less obvious is that it seems not to owe its viral tendency to the endorsement of influential tweeters, but is more likely than true news to be spread by individuals who have only a few followers. Yet Twitter users will probably recognise this phenomenon too: aren’t those precisely the folks who send the oddest, most cranky, and often most abusive tweets?
It’s important to recognise, though, that false news differs from true not just in its content but in its sources. Real news is typically reported through big, reputable sites like those for newspapers. People seeing such stories are likely to click on the original source and then perhaps to share that, rather than just retweeting the tweet in which they saw the link. “If someone I trust shares a link to a New York Times story, I might just tweet that same link after visiting that story,” says Eckles. “But if they were to share a link to a less credible source, I might retweet them so that the information retains their (possibly implicit) endorsement.” In other words, we want to be able to say: Look at this, but don’t blame me if it’s wrong.
For that reason, false news seems to have not just a different rate and depth of spreading but also a different pattern. According to a separate study of Twitter and the Chinese equivalent Weibo by a team based in Beijing, fake news spreads along diverse, dispersed channels, while real news circulates though a dominant source. The researchers say that these differences could provide a means of quickly identifying news likely to be fake, even without analysing its content: a kind of early-warning system.
“False news seems to have not just a different rate of spreading but also a different pattern”
You might think, given these findings, that the content of a story—its “novelty,” degree of truth, emotional or moral charge and so on—might itself allow us to predict how far and fast it will spread. But Duncan Watts, a specialist on social networks at Microsoft Research in New York doubts that. “In our own work, we have consistently found that content seems to have very little predictive power over and above the origin of a cascade,” he says. “I would be willing to bet that something similar applies here: even though false and true stories had significantly different average properties, knowing that a story is true or false probably doesn’t help you much when predicting how far it will spread.”
Maybe, then, it’s not so easy after all to engineer false news: you can see what average attributes it tends to have, but can’t reliably build them into any given story. Let’s hope so—because if there was indeed a simple “formula” for making false news, it would surely be exploited. “If we understand ‘scientifically’ how to engineer viral news, this would mean a transition to an industrial era of fake news,” Quattrociocchi says.
It’s been said many times that fake news is nothing new—people have been spreading (and complaining about) it for centuries. It’s true that social media now offer these falsehoods very rapid channels through which to spread, but it’s also easier to access information that can enable us to check it out (if we feel inclined), or to send and receive corrective messages.
What’s more, the characteristics of fake news identified in this new study will sound familiar to historians: novelty, surprise, ability to incite moral outrage. Any demonised group in the Middle Ages—Jews, Waldensians, Templars, Fraticelli—knew what kind of stories would circulate about them: sodomy, sexual perversion, even the sacrifice and eating of babies. And there were receptive, uncritical echo chambers for those tales. Norman Cohn’s classic 1975 study Europe’s Inner Demons has been looking disturbingly topical recently.
More troublingly, perhaps, such stories might very well have been considered the absolute truth at the time. Even with careful fact-checking, there’s always an element of “truth” that remains contingent. “It can be strongly dependent on the social group that defines it,” says Quattrociocchi. In the seventeenth century, plenty of well-informed observers would have classified the work of Galileo on the heliocentric cosmos as fake news, he suggests.
The role that fake news played in Trump’s election or the Brexit vote remains unclear. But Watts cautions against seeing it as the cause of all that seems to be going awry in liberal democracy. “It’s definitely troubling that bad actors are deliberately manufacturing false and inflammatory information, and that ordinary social media users are uncritically consuming and sharing it,” he says. “But we also know from other studies that the vast majority of news circulating on social media and being consumed by moderates and partisans alike is produced by mainstream news organisations”—and is therefore most probably not “fake” (whatever Trump says).
“Twitter is not obviously better or worse than the web as a whole, or society as a whole,” Watts adds. “There is endless opportunity to find interesting and useful information if you’re looking for it, and also endless opportunity to find loudmouth idiots, liars, snake oil salesmen, hate speech, and people who will reinforce your own views.”
“If we look back at the techno-utopianism of the 1990s, the prevailing belief that the web would be a boon for democratic discourse now seems pretty naïve,” he says. But now the question is what can we do now to improve the integrity and quality of the information ecosystem we have created, without destroying it through a combination of paternalism and censorship. “It’s a very complicated question with no clear answers,” says Watts.
“If what we really care about is the larger societal problems such as political polarisation, suspicion of experts, and erosion of shared reality,” he adds, “it is not at all clear that fake news is even close to the most important cause.”