A Twitter update intended to encourage thoughtfulness on the platform in the run up to the presidential election has hit Joe Biden’s campaign – and barely affected Donald Trump at all.
Twitter implemented the changes over several hours on 20 October, saying they would stay in place until “at least the end of election week in the US”. Previously, posts would be recommended to a user’s feed if they were “liked” by people they followed, even if the user did not follow the original “poster”. That feature has been turned off, meaning users now only see tweets if they follow the author, or if they are retweeted by someone they follow. Twitter also changed the retweet feature, pushing users to “quote tweet”, meaning they add their own thoughts rather than instantly retweeting the post in its original form.
Twitter said in a statement that the changes to the visibility of “liked” posts would “encourage more thoughtful and explicit amplification”, while the retweet update would “encourage everyone to… consider why they are amplifying a tweet”. But experts warned the move could instead result in polarising American users further.
In order to measure the early impact of the move, the New Statesman data team collected and analysed the “likes” and “retweets” racked up by the two presidential candidates in the days leading up to and following Twitter’s update.
The analysis found that before the change was implemented, Biden’s Twitter engagement was almost as high as Trump’s – even though Trump had almost eight times as many followers – and on a number of occasions Biden significantly outperformed Trump. However, the update has apparently hit Biden much harder than his rival, his number of “likes” having fallen while Trump’s have risen.
What does the data show?
The New Statesman data team started by tracking tweets across two periods: from 18-19 October, just before the update was implemented, and from 20-21 October, after the update took effect.
It tracked all the original tweets (not retweets or replies) posted from the Twitter accounts @JoeBiden and @realDonaldTrump – a sample size of between 27 and 45 tweets each – measured at different intervals, to track their growth across the 23 hours after posting.
Before the update, Biden’s tweets were doing much better, with a third of his 29 tweets crossing 100,000 likes. Two tweets that went “viral” broke the 500,000 barrier in under eight hours and went on to receive more than a million likes, with one reaching a million in under 12 hours.
Trump’s tweets performed slightly better before the update, with half crossing 100,000 likes, though none received more than 300,000 likes. This is to be expected, and is actually remarkable on Biden’s part, bearing in mind he has more than 11 million followers on Twitter, while Trump has almost eight times that figure.
After the update Biden had no similarly viral tweets, with his most popular posts climbing at a slower rate and “topping out” earlier, while Trump’s growth stayed consistent.
To get a longer-term picture, the New Statesman team worked with the Sloan School of Management at the Massachusetts Institute of Technology (MIT) to analyse the overall engagement with Trump and Biden tweets backdated to 10 October, and up to 27 October.
We looked at the winsorized average of retweets and likes – a kind of average that limits the effect of extreme outliers – and found Biden had suffered a fall-off on both counts since the Twitter changes. Trump’s average retweets had fallen after the changes, though by a smaller proportion than Biden’s, while his likes increased.
What does this tell us?
It is important to recognise that this is a limited dataset, with some important caveats. Counting retweets and likes on Twitter is by nature an inexact science, and there are multiple factors that determine how “viral” a tweet may be. The topic of a tweet, or whether it is about (or even simply coincides with) a significant event such as a debate or big news story, or even the time of day of posting, can all have a bearing on the number of tweets and retweets that a user receives. That “disruption” to the data is likely to become more intense as election day nears.
It is reasonable to suggest Biden was seeing such high engagement before the changes, despite having fewer followers than Trump, because he was benefiting more from Twitter features allowing users to engage beyond their followers. That would also make him more vulnerable to the changes.
Sinan Aral, a professor at MIT and author of The Hype Machine: how social media disrupts our elections, our economy, and our health – and how we must adapt, says the new Twitter update will hit accounts with fewer followers the hardest.
“Typically, information moving from person to person, cascading through the network, is helpful when you don't have as many followers, because it can be shared on more than one hop away from you,” he said. “The more that a Twitter policy throttles that avenue of information diffusion, the more people with a high number of direct followers are going to retain their audience than somebody with fewer followers.
“If the polling is correct, and Biden is leading nationally, and if Trump has so many more followers than Biden, then I think it's a relatively simple logical conclusion that a smaller fraction of Biden's supporters follow Biden on Twitter.”
Aral says the problem with Twitter’s new update, as well as other updates across WhatsApp and Facebook that aim to reduce the spread of information, is that they are slowing down this spread indiscriminately rather than targeting “bad” information.
“They're trying to put in place policies to slow the rate of diffusion of information on social media in general, in the hope that this will allow the truth to catch up with falsity, because falsity travels so much faster on social media,” he said. “But it could have plenty of unintended consequences, for example, a policy being disadvantageous to people with fewer followers and therefore favouring one candidate over another. It's a difficult position to be in.”
What does this mean for the election and beyond?
The question isn’t simply about whether the information flow on social media leads people to vote for Biden over Trump, or Trump over Biden. It’s also about whether they vote at all.
“It’s highly unlikely that it's going to significantly affect vote choice,” says Aral. “But there is good evidence that social media messaging affects voter turnout.”
Foreign voter interference via Facebook in the 2016 election targeted voter suppression, and similar messages have been tracked this year. Aral worries that Twitter’s new policy is not supported by enough scientific evidence. “It was four years ago that we should have started dealing with this situation. That would have given us enough time to study what works and what doesn't work, and to really implement good policies that would harden our democracy to foreign interference as well as the manipulation, and misinformation,” he says. “While I welcome platform changes that are being made, it's too little too late.”
David Rand, also a professor at MIT, who studies the science of misinformation and how to combat it, says that while the new policy might have been set up to target misinformation, it could have a fundamental flaw.
“My guess is what they're trying to prevent is misinformation sites that don't themselves have very many followers, posting sensational or crazy or baiting kind of stuff, then some people start engaging with it, and then other people get exposed to it, and then it sort of takes off,” he says. “But a lot of people haven't caught up to the fact that the biggest source of misinformation right now is Trump. It's not random people sharing random things on Facebook. It's Trump using his platform to directly broadcast as much misinformation as he can.”
Measures such as this one, then, in trying to reduce the spread of misinformation, may actually be causing more damage by increasing Trump’s relative reach among an audience of followers, one less likely to be exposed to alternative viewpoints.
In one sense, this is a limited problem: how should social media approach a high-stakes election in which partisan feelings are running high? But there is also a deeper issue that will not simply dissolve when the ballot boxes close. Aral’s fear is that the changes, if maintained, could reinforce “echo chambers” and lead to further polarisation, regardless of the result. If the election is contested, he says, that could have terrifying consequences.
“What happens if there [are] completely different world views, by different sides of the aisle, on who the legitimate president is?” he asks. “What is that going to do to the potential for violence?”
[See also: Why US democracy as we know it may soon be over]
Katharine Swindells is a data reporter for New Statesman Media Group