Skip to content

Network analysis of top EU referendum tweeters

Social media has come to play an ever more important role in politics; not least during large-scale, public events such as the EU referendum. Last year we showed how important Twitter has become for domestic politics through analysis of the General Election. In this short blog series, we will take a look at the aftermath of Brexit, now that some of the dust has settled.

The referendum was Twitter-ific. As such, many researchers have capitalised on the resulting glut of social media data - analysing how content influenced people and promoted dissemination of ideas. One good example is Vyacheslav Polonski’s analysis of social media voices using Instagram hashtags, splitting #brexit and #bremain camps.

In this post, like Polonski, we focus on one particular angle - the people that defined the Twitter debate.

Despite large scale public engagement in online debates, there was clearly a hard-core of Tweeters who were particularly prolific in the run up to the vote. On 23rd June (the day of the vote) colleagues at the University of Sheffield collected over 1.9 million tweets on the referendum. Of these Tweets, 714,912 were original (authored by the user who posted them - excluding retweets and replies). We identified 164,000 tweets advocating Leave and 127,000 advocating Remain.

Within these two sets, we identified the 100 most prolific users who tweeted for Bremain and the 100 most prolific for Brexit. We also took a set of 100 random users involved in the online discussion, for comparison.

Next, for each of these users, we downloaded their Twitter profiles, their followers, and Twitter accounts they follow, and most recent tweets, which we then analysed. We ended up with a series of ego networks - simple models that represent a person’s social network - they allowed us to look at the social relationships that a prolific Tweeter has with other Twitter users.

When we looked at the dynamics of these networks, initially, we noticed that the average number of people in a network varied dramatically dependent on their political view towards the referendum - Brexiters tended to have smaller networks, whilst the random sample and Bremainers has significantly larger average network sizes.

Figure 1 Average Network size for Leave, Remain and Random samples on Twitter

Figure 2 Average size of social circles for Leave, Remain and Random samples on Twitter

Average sizes of social circles show a significant structural difference in the networks of the 100 most prolific Bremain users, compared to the Brexiteers. Bremainers  seem to use Twitter more socially, in that they maintain larger networks, with an active network size that is more similar to the sample of randomly selected users.

Among Brexiteers, on the other hand, there is a higher number of accounts that do not show social usage of Twitter. These accounts do not maintain direct social contacts with other users, but rather create only plain tweets (i.e. tweets not directed to other users).

Interestingly, after filtering out these non-social accounts, Bremainers still show significantly larger networks. Figure 3 shows the percentage of Bremainers, Brexiteers, and random users who have at least one social circle. These significant differences indicate a tendency of Brexiteers to be less socially active than Bremainers, and more focused on actions such as posting and re-tweeting information rather than maintaining social relationships with others. There are a number of potential explanations for this trend. It may be that campaigners were using Twitter for astro-turfing (presenting a marketing or public relations campaign in the guise of unsolicited comments from members of the public), or perhaps Brexiteers joined Twitter specifically for the referendum - hence the presence of early stage, or small networks.

Figure 3 Percentage of users that have at least one social circle in their network

This has implications for the way that campaign managers and politicians organise their social media messaging around such divisive issues, and their use of social media platforms  to reach, or mobilise diverse audiences. Specifically, this evidence shows that there are clear patterns of use that tend to correlate with certain views. It would make sense, therefore, for campaigners to target groups with small networks through flagship accounts or hashtags, rather than relying on the effective spread of information through follower/ followee networks.

These findings will hold significant for debates that follow party lines - unlike the EU Referendum - a particular view on an issue, or individual characteristics, like political orientation, will likely relate to specific patterns of social media use.

We will come on to explore some of the reasons behind these trends in subsequent blogs in this series, including an evaluation of follower/ followee networks - developing an understanding of who follows who, and how this varies based on expression of views or sharing of certain ideas.

We investigated Tweeter characteristics further across all Brexiteers and Bremainers in our dataset over time. Figure 4 shows that over time, Brexiteers and Bremainers were posting roughly the same number of original tweets, with Brexiteers marginally out-Tweeting Bremainers in the final throes of the run up to the referendum. When retweets are considered (Figure 5), however, Brexiteers are clearly  much more active than Bremainers, particularly on referendum day itself.

Figure 4 Original tweets posted by Leave/Remain supporters

Figure 5 Re-tweets posted by Leave/Remain supporters

From the characteristics of Twitter accounts, we can see that Bremainers are more popular than Brexiteers  (i.e. they have more followers). But both Bremainers and Brexiteers are, on average, more popular than the accounts in the random sample.

When we compare the average number of Twitter users these accounts follow, Brexiteers follow many more accounts than Bremainers and random users. This could indicate that Brexiteers were more active in the monitoring and diffusion of information in the network.

And when other metadata is considered, such as average account age, the percentage of accounts with a text description, and the percentage of accounts with location field, we see very little difference between the three groups of users.

Figure 6 Topics discussed by Leave/Remain supporters

However, when it comes to the frequency of topics discussed - Figure 6 shows a much more marked difference between Brexiteers and Bremainers. Most interesting are the disparities between mentions of topics like employment, and democracy.

We know that much rhetoric around the Leave campaign centred on ‘retaking control’ of UK politics from Europe, in this sense, it is unsurprising that democracy, law and the justice system, and foreign affairs were more commonly referenced by Brexiteers than Bremainers. On the other hand, children and young people, schools and public health were more commonly mentioned in Bremain tweets, perhaps in response to debates on public spending - with reference to the future of the UK - suggested through a focus on children, schools, and young people.

We will be delving into these topics in greater depth in subsequent blogs - looking at change and difference over the campaign period, and across the UK.

The EU Referendum acted as a catalyst - allowing important insights into public sentiment on a number of big debates. Our analysis of social media - now the dust has begun to settle - will provide granular detail on these debates. This will be important for politicians, policymakers and the public to get a sense of what matters in the UK, to negotiate a relationship with Europe that best represents the views of the nation.


George Windsor

George Windsor

George Windsor

Senior Policy Researcher

George was a Senior Policy Researcher in the Creative and Digital Economy team.

View profile

Valerio Arnaboldi

Valerio Arnaboldi is a researcher at the Ubiquitous Internet group of the Institute for Informatics and Telematics (IIT), at the National Research Council of Italy (CNR). He works in...