Donald Trump’s meteoric rise from political outsider to president of the United States surprised nearly everyone – not least political analysts and scientists. Many are hoping for an easy explanation from the Mueller report, including evidence of heavy Russian interference in the campaign. Mueller has indicted numerous Russians in this regard, though more details will emerge when the report is published in the coming days.
In our new paper, which has been published in the PLoS One journal, we have covered similar ground via Trump’s stronghold: Twitter. By sampling some 250,000 accounts, we found a powerful new group of Trump supporters emerged during the election and effectively usurped the Republican Party on the social network. But very much to our surprise, very few bots or Russian accounts were involved. This suggests that if the Russians were acting to influence the election, the effect at least on Twitter may have been much more limited than has been claimed.
We identified three kinds of Twitter accounts that were particularly relevant to the election: a Republican Party group; a Trump group; and a group of more extreme alt-right adherents. We classified accounts into these groups based on who they followed and the hashtags and other words that they used in posts. The Trump group often used #maga, for instance, as in “make America great again”, and also “Trump supporter”. Mainstream Republicans often used #tcot or #tgdn, respectively “top conservatives on Twitter” and “Twitter gulag defence network”; while the far right used the #altright hashtag and words like “white” and “nationalist”.
What happened on Twitter
When we looked at how these three groups had developed over time, we found the Republican accounts mainly dated from the Tea Party marches following Barack Obama’s first election victory in 2008, and also the 2012 Obama vs Romney campaign. Conversely, the Trump and alt-right groups had largely emerged during the 2016 election campaign.
By late 2016, very few new accounts were being opened that fit the characteristics of our Republican Twitter group. We found a big shift in following behaviour as well, with existing Republican accounts becoming more likely to follow accounts in our Trump group rather than other mainstream Republicans. This reflects the way in which Trump suddenly jumped ahead of a crowded field in the Republican primaries. When you combine the followers of the three Twitter groups, they amount to some 57m unique users: this almost certainly made the difference in an election where the margin of victory was so tight – remember Trump beat Hillary Clinton in the electoral college, but without winning the popular vote, winning key marginal states by only a few tens of thousands of votes.
But what drove support for this shift? Staying with following behaviour, we found that members of all three groups tended to follow people who came under the same group, while those that we identified within the Trump and Republican groups frequently followed one another. But while members of the alt-right group followed those in the Trump group, this was not reciprocated to the same degree. This suggests that the widely held idea that the far right were very influential in the growth of support for Trump may be an exaggeration.
To estimate Twitter bots, we used a US tool called Botometer, which scores each account on the likelihood that it is automated. We concluded that Twitter bots and foreign accounts were certainly part of Trump’s Twitter community, and played a role in spreading his message, but were vastly outnumbered by the massive groups of real-life supporters who suddenly started joining Twitter and following one another after Trump announced his election campaign. In fact, we found more automated accounts in the Republican Party’s group than in Trump’s group. Our findings match other recent research, which found that fake news was not nearly as pervasive on Twitter and Facebook as previously feared; in the case of Twitter, for instance, 80% of fake news appeared on only 1.1% of users’ newsfeeds.
Our point is not that foreign-owned bots generating fake news didn’t interfere with the election, but rather that they probably had less influence than various other factors – particularly Trump himself, his group of highly motivated supporters and the US media. Trump’s supporters did not coalesce around an army of bots – they do appear to have been a grassroots movement of previously disengaged voters. Trump’s victory seems more driven by his own particular style of campaigning, galvanising his followers into a political backlash against “Washington elites”.
These kinds of movements certainly aren’t unknown. Political analysts are very familiar with the concept of the Overton Window, in which the political centre ground shifts in response to pressure from disenfranchised and frustrated groups on the fringes. In Trump’s case, the shift was surprisingly rapid. In only a few months, a relatively small group had grown to the point it was able to subsume the traditional Republican Party.
Predicting the future
Our era will long be remembered for the populist swings that took place in politics – not only Trump but elections in the likes of Hungary and Japan, and also the UK’s Brexit referendum. These results frequently surprised politicians and the media, prompting much discussion about problems with the tools with which we have tracked people’s voting intentions.
We think that our method, or one derived from it, can be a valuable addition to the toolbox in future. By following the development of political groups on Twitter, you can observe what is happening in real time. In future, this could help identify disenfranchised voter groups amenable to populist candidates and better understand their behaviour and the issues that motivate them. Our method might also make it easier to determine when extremist political minorities, massively amplified by the global reach of Twitter, might be exerting a disproportionate level of influence.
Studying Twitter allows you to observe these things at a speed that traditional polling and analysis can’t match. Hopefully by studying the world of online political discourse in a more rigorous and systematic way like this, we can finally start to catch up with the breakneck speed of modern political change.