Brexit: a story of two tales

Over the past few months, the British EU Referendum has been the most prominent issue in the European political debate. The most recent opinion polls show that the gap between Remain and Leave has been narrowing, in a crescendo of heated discussions that are taking place both offline and online.

In this study, we look at how the referendum has been publicly discussed in the UK on Twitter. The latter may well be considered a political arena, where supporters and opponents of Brexit express their views and back the respective campaigns. More precisely, we are interested in understanding the development of the thematic focus of the general Brexit debate as well as the behaviour of particular groups of users.

Data overview

Between 04 April and 09 June 2016, we collected 3,139,049 UK-based tweets by means of a keyword match criterion and through a Twitter “stream” application programming interface. In order to obtain an unbiased starting dataset, we performed a preliminary manual review of Brexit-related conversations prior to collection, and opted for the three campaign-neutral keywords “brexit”, “euref” and “eureferendum”.

For the purposes of this analysis, we divided the tweets into two groups using the six hashtags most often associated with the Remain and Leave campaign respectively. Taken together, these hashtags can be seen as the “battlegrounds” in which Twitter users’ conversations about Brexit take place. Results are shown in Table 1.

Table 1: Numbers of users, hashtags and tweets for Remain and Leave battlefields
Table 1: Numbers of users, hashtags and tweets for Remain and Leave battlefields
Left and right in the referendum debate

To fully grasp the dynamics behind the Twitter Brexit debate, we then examined the nature and content of tweets over time and the profile of the users contributing to the discussion. As a first step, by analyzing tweets’ metadata, we were able to infer the political preferences of users. More precisely, through natural language processing techniques we separated users who clearly identify themselves as engaged or interested in politics from those who do not.

The Leave campaign battlefield is heavily patrolled by right-wing users. It should be noted, however, that this is mostly due to the massive activity rate of Ukipatriots within the right-wing field.

We further differentiated, in the former group, between “left-wing”, “right-wing” and “others” (the latter being a residual category for anybody who could not be placed in either of the two sides of the traditional political spectrum). We then fine-tuned the analysis by breaking up the left- and right-wing groups into more specific areas: “Labour” and “Non-Labour left” (green party, trade unions, socialists) on the one hand; “Lib-Cons” (Tories and liberals) and “Ukipatriots” (UKIPpers and nationalists) on the other. Table 2 shows the size of these groups as well as data on their Twitter activity.

Table 2: Number of users, tweets and tweets/users for Left-wing and Right-wing political groups
Table 2: Number of users, tweets and tweets/users for Left-wing and Right-wing political groups

Table 3 shows user activity divided by camp. Interestingly, for all political subgroups, with the exception of Labour, activity rates are higher on the Leave side than on the Remain side. The table also shows that the Remain side is composed of a majority of left-wing profiles, whereas for the Leave field the opposite is true. If we look at the number of tweets instead, the picture is much less balanced: the Leave campaign battlefield is heavily patrolled by right-wing users. It should be noted, however, that this is mostly due to the massive activity rate of Ukipatriots within the right-wing field. Indeed, inside the Leave campaign, each Ukipatriot tweeted on average 48 times, largely outperforming Liberals and Tories (18 tweets), the non-Labour left (9.3 tweets), and Labour profiles (8.6 tweets).

Table 3: Number of users, tweets and tweet/users of Left- and Right-wing groups divided by Remain and Leave camps
Table 3: Number of users, tweets and tweet/users of Left- and Right-wing groups divided by Remain and Leave camps
Talking Brexit

We then analysed the language of tweets to detect the themes at the centre of the Brexit debate. We first sorted tweets featuring “informative” content (that is, one or more reasons for either Remain or Leave) from “propaganda” (namely tweets expressing support for one side without mentioning a clear reason). The distribution of “informative” and “propaganda” tweets turned out to be constant (35% and 60% respectively), across both political subgroups and campaign fields.

Second, we classified the “informative” tweets into three categories: “Economy & welfare”, “Sovereignty & migration” and “Other”. The first group includes discussions on jobs, the welfare state, international trade. The second includes talk about the limitations of sovereignty imposed by the EU, the Union’s democratic deficit, the EU’s bureaucracy and corruption, and the issue of migration and border control. Table 4 shows the distribution of tweets across thematic categories, cross-tabulated by user’s political position.

Table 4: Number of tweets of Left- and Right-wing political groups divided by theme
Table 4: Number of tweets of Left- and Right-wing political groups divided by theme

Generally speaking, we found the “Economy & welfare” topics scoring highest across all political groups and subgroups, establishing itself as the main thematic battleground.
Yet, if we look at the distribution of tweets between political fields, we see that in relative terms over the total, “Sovereignty & migration” discussions weigh more on the right-wing side than on the left-wing one.

Even small groups can become “agenda setters”, when acting online in a coordinated and organised fashion, especially in highly polarised debates.

We found as well differences within the two main political fields. Within both political families, Labour and Tories were relatively less prone to tweeting about “Sovereignty & migration” than the non-Labour left and Ukipatriots. Strikingly, within the left-wing field more than 70% of the tweets dealing with this topic come from “left-activists”’ profiles–a distribution even more unbalanced than on the right-wing side, where UKIPpers are responsible for “only” 58% of the tweets on these topics. At the same time, while within the right-wing area both subgroups tweeted approximately in equal numbers about “Economy & welfare”, labour users were slightly more active on this front than left-activists.

Approaching the big day

Our results show that the Leave battlefield is generally more populated in the Twitter debate on the referendum. Contents, slogans and hashtags pertaining to this side have been more actively propagated and more widely discussed. This, of course, does not necessarily translate into a higher success rate, in terms of penetration, on the part of the Leave campaign messages. Yet it indicates that even small groups can become “agenda setters”, when acting online in a coordinated and organised fashion, especially in highly polarised debates.

This is confirmed, first, by the heavy presence and activity of UKIP users (who can unambiguously be considered as Brexit supporters) in the Leave as well as the Remain battlefield, and second by the fact that, in general, all political groups of activists (with the exception of Labour) are more active in the Leave battlefield. Soon after the closing of the polls tomorrow, we will find out whether, or in any case to what extent, the social media activism of the Leave side translates into real political results.

(EuVisions, 22.06.2016)

CC Photo Credits: seth m

 

 

Corbyn’s EU-turn speech as seen from Twitter

The Labour party (and the Remain field?) might therefore have lost some support from those leftist voters who are more prone to anti-austerity rhetoric and more sensitive to the argument that the European construction has lost its solidaristic flavour.

 

On April 14, Labour leader Jeremy Corbyn held a long awaited speech outlining the party’s position with respect to the EU referendum to be held on June 23.

In committing to campaigning for the Remain side, Corbyn amended his longstanding sceptical stance towards the European Union. He warned that a victory of the Leave campaign would pave the way to an aggressive Tory government, which would endanger those worker rights that are currently safeguarded by EU legislation.

The Telegraph wrote that by focusing on “worker rights” and scaremongering about a new Tory government, Corbyn tried to make an appealing case for the Labour base to back a stay-in vote.

The British press covered extensively Corbyn’s U-turn on the integration project, which attracted a lot of criticism from the Brexiteers’ front. Although negative feedback came mostly from the right, even from inside Labour it has been argued that Corbyn’s recent support of the EU project is not genuine but hides instead a move to consolidate his leadership of the party.

From inside Labour it has been argued that Corbyn’s recent support of the EU project hides a move to consolidate his leadership of the party.

Labour backbencher and former Welfare Minister Frank Field argued that Corbyn was unable to connect with the Labour base and warned that his repositioning could lead to a massive loss of support in favour of UKIP. Given Corbyn’s longstanding Eurosceptic views it seems that it might be exactly his own constituency to be at risk here. In a recent article on The Telegraph, Ben Riley-Smith and Kate McCann even suggested that Jeremy Corbyn might decide to reconsider his support to the Remain campaign in order to re-establish a connection with the Labour rank and file.

Is Corbyn really losing support because of its new pro-EU stance? How did people react to his speech? On April 14 and 15, 25,581 unique Twitter users produced 55,357 tweets containing the word (or hashtag) “Corbyn”. By looking at these tweets, we tried to answer the above questions.

How did people react?

We used natural language processing techniques to sort tweets into different categories. After discarding irrelevant tweets, we separated attitudinal tweets – i.e. tweets containing some kind of reaction to the speech or the speaker – from news recast or merely descriptive tweets. We then separated tweets expressing a negative sentiment or disagreement with Corbyn from the rest. Table 1 shows the results of this procedure.

Table 1
Table 1

Recent poll results show that Corbyn’s announcement did not have a significant impact on the Brexit issue, and that Britons are still evenly distributed between Remain and Leave. In the Twittersphere however, the prevailing reaction was negative. As shown in Table 1, out of 28,056 attitudinal tweets, 10,095 (38%) are positive (or neutral) tweets, whereas 17,151 (62%) are negative. This confirms the view that the microblogging platform is a medium where voices of protest are more visible.

Corbyn losing Corbynistas?

By analyzing the tweets’ metadata, we were able to separate those coming from leftist tweeters from the rest. We identified 1,528 users as “leftists” and 1,761 as “other” (a residual category including all remaining politically-oriented profiles, ranging from Liberals to Tories and Ukippers).

Table 2
Table 2

We cross-referenced these data with the classification based on the text of the tweet: Table 2 shows the distribution of tweets between “leftists” and “other” users. Our results indicate that only relevant tweets are almost evenly distributed between the two groups. Indeed this distribution becomes unbalanced when we look at those classified as attitudinal and negative: in other terms non-leftist users become “louder”.

The Labour party might therefore have lost some support from those leftist voters who are more sensitive to the argument that the European construction has lost its solidaristic flavour.

However, and most importantly, almost one out of three negative tweets (that is 1,399 tweets) comes from the leftist area. Even more surprisingly, we found that almost 40% of the users who reacted negatively belong to the same area. In other words, per-user negative tweets were fewer on the left side of the political spectrum than for remaining users. This, in turn, implies that non-leftist users were more active in expressing their disagreement.

Left and right Brexiteers: sharing a common ground

As expected, Corbyn’s pro-EU speech provoked a lot of criticism from the right-wing Brexiteers, conservatives and UKIP supporters. However, a considerable negative reaction also came from leftist Twitter users. We tried to shed some more light on these results by processing the negative tweets through text clustering applications. In terms of content, the “leftist-negative” and the “other-negative” fields share a common ground, namely the criticism of Corbyn’s previously highlighted U-turn on Brexit. In particular, it is the political “coherence” of Corbyn — often labeled as a former “man of principles”– to be at the centre of concerns. Austerity-related groups of terms, on the other hand, appear more often among leftist users.

The Labour party (and the Remain field?) might therefore have lost some support from those leftist voters who are more prone to anti-austerity rhetoric and more sensitive to the argument that the European construction has lost its solidaristic flavour.

However and most importantly, the proximity of leftists and right-wing Brexiteers highlights the need to transcend the traditional left-right distinction when it comes to the politics of European integration. As recently stated by Raphael Behr, “we are in a strange world now where two distinct sets of politics–the old one that follows left-right lines, and a new one that operates around an EU in/out axis–are running concurrently and on top of each other”.

(EuVisions, 06.05.2016)

CC Photo Credits: Bob Peters

Renzi vs Juncker: a social media analysis

We focus on the recent war of words between Renzi and Jean-Claude Juncker, triggered by the latter’s critical remarks towards Italy during the Commission’s new year’s press conference. The Renzi-Juncker quarrel is the latest, and certainly one of the most visible episodes in what some have called a new fault line in European politics.

Over the past few weeks, a rift has emerged and deepened between the Italian government, led by Matteo Renzi, and European institutions, most notably the Commission. Disagreements between the two sides span many areas, including immigration, energy policy and banking. The latest chapter in the saga is being played out in the budgetary field, in which Italy claims the right to more flexibility (and a higher deficit) than the Commission is willing to concede.

The clash between Italy and the Commission is as communicative as it is political. From the start, the ways in which both sides have addressed one another have been as important as the substance of their respective grievances. In an unusually sharp exchange, for instance, Renzi recently reminded the Commission that the times in which Italy would be “remote-controlled” from Brussels are over. As a reply, the Commission let it be known that it actually struggles to find any institutional interlocutor in Rome.

As with any communications battle, much of the row between Italy and the Commission is taking place in and through the media, including social ones. Here we want to analyse the effects and resonance of this clash in the Twitter-sphere. In particular, we focus on the recent war of words between Renzi and Jean-Claude Juncker, triggered by the latter’s critical remarks towards Italy during the Commission’s new year’s press conference. The Renzi-Juncker quarrel is the latest, and certainly one of the most visible episodes in what some have called a new fault line in European politics.

An all-Italian affair

Given the highly personalized nature of the incident, our research strategy was to collect all tweets mentioning both Renzi and Juncker, on January 15 (the day of Juncker’s press conference) and for the following five days. Over the monitored period, a total of 8,724 such tweets were written across languages. The figure below shows their temporal flow.

Fig 1: Renzi-Juncker – tweets over time (click to enlarge)

As the chart shows, after a very brief moment, on January 15, in which the numbers of tweets in Italian and other languages were close, the Renzi-Juncker Twitter conversation became an almost exclusively Italian affair. Overall, of the tweets collected, 7,839 (89.9% of the total) are in Italian, 574 (6.6%) in English, 90 (1%) in Spanish, 57 (0.7%) in French and 9 (0,1%) in German. That German users ignored the Renzi-Juncker exchanges is particularly remarkable given that in these sorts of debates their country is often accused of being the Commission’s favourite—if not its puppet master.

The lack of a foreign reaction to the Renzi-Juncker clash is consistent with the view, held by many (in the first place in Brussels), that Renzi’s motives and objectives in the quarrel are largely domestic in nature. Weakened by Italy’s still underwhelming economic performance and some scandals involving his government, the argument goes, Renzi is for one thing trying to divert attention towards a “foreign enemy” and rally support around himself in time for next spring’s local elections. For another, he is preparing for a possible infringement procedure against Italy by pre-emptively shifting the blame onto the Commission.

That the quarrel was followed only in Italy is, in a sense, an advantage for Renzi as it insulated the debate from those audiences – primarily the German one – that might not be too sympathetic to Italy’s complaints. Granted, it also insulated it from (potentially) more benevolent audiences, such as the Spanish or the French one.

A gamble that backfired?

Generating a domestic debate might be a good strategy if one’s objective is simply to distract voters from other topics. It can, however, backfire if the debate ends up being dominated by negative sentiments. Following a method outlined in a previous article, we have tried to answer this question by separating, in the subset of Italian tweets, those expressing criticism of Renzi from the neutral or positive ones. Overall, 4,759 (60.7%) tweets belong to the former camp, while 3,080 (39.3%) are in the latter. For a politician well versed in the use of Twitter like Renzi, this was not a very good performance.

Fig 2: Renzi-Juncker – negative tweets word cloud

The above word cloud illustrates the frequency of terms in text-only anti-Renzi tweets (i.e. excluding re-tweets of URLs). Some recurring criticisms concern the uselessness (inutile) of Renzi’s attacks on the Commission, and the suspicion that his “song and dance” (manfrina) was largely for domestic purposes. In addition, many did not see Renzi’s fight for flexibility as credible, coming from a “puppet” (burattino) of powerful financial lobbies.

These last accusations are quite interesting because to a significant degree they are levelled against Renzi and Juncker at the same time. More generally, tweets that are negative towards Renzi are often also critical of the Commission president and the EU. If gaining some support among Italian Eurosceptics was among Renzi’s goals in the quarrel, his strategy—at least judging from the Twitter-sphere—did not work too well.

(OpenDemocracy, 26.01.2016)

Cameron’s renegotiation speech and intra-EU migration: how the web reacted

The reaction we observed was mainly directed to Cameron’s reform proposals. In particular, we noted an overwhelming presence of negative attitude with respect to what was presented by the PM: this result seems consistent with evidence from opinion polls indicating little confidence in Cameron’s ability to strike a good deal out of the renegotiation process.

On November 10, 2015 Britain’s Prime Minister David Cameron held a keynote speech at the Chatham House, launching his proposals for the reform of the European Union – a preliminary step and condition to the referendum on Britain’s stay in the Union, which will be held by 2017.

Cameron’s renegotiation proposal revolves around four political issues – corresponding to the main points of his Chatham House speech: granting equal power to Eurozone and non-Eurozone member states in EU decision-making processes; reversing the objective of an ‘ever closer Union’, written in the treaties; making the European single market more competitive; finally, easing off intra-EU migration flows.

Using natural language processing (NLP) techniques, we have analysed reactions to Cameron’s speech on the micro-blogging platform Twitter. We have focused, in particular, on the issue of intra-EU migration – Cameron’s fourth point – which is arguably the most politically sensitive and prominent in the current Brexit debate. Our objective was to see, first, if Cameron’s stance on the migration issue triggered a reaction and discussion on Twitter, and second whether such discussion was structured along clear lines of conflict.

Detecting attitudinal digital voices

Through Twitter’s ‘search’ and ‘stream’ APIs we downloaded tweets matching a set of keywords at different levels of precision,[i] starting on November 10 and for the two days following Cameron’s speech. We then analysed this corpus of tweets by means of a semi-automated approach relying on NLP, and based on the construction of classifiers to detect, at each successive stage of the research, language and meaning patterns in the examined tweets.

Figure 1 shows our classification pipeline. Broadly, we built and trained classifiers to sort tweets on the basis of: 1) their relevance vis-à-vis Cameron’s speech, 2) the presence of an attitude towards the immigration issue, and 3) the sentiment being expressed. The performance of our classifiers is detailed in table 1 (click pictures to enlarge).

Open vs. closed borders

The classification procedure left us with 8,568 tweets expressing an opinion on the intra-EU migration issue. Overall, such tweets show a negative reaction to Cameron’s stance on the matter. Looking more closely, we were able to separate groups expressing different types of criticism: 3,982 (46.5%) tweets attack Cameron while defending the principle of open borders in the EU, whereas 4,077 (47.5%) oppose the PM by arguing for a closed borders scenario. This balanced picture chimes with results of a recent study by ICM, showing that Britons are evenly split on the topic of the EU referendum.

Figure 2 shows the hourly distribution of tweets containing a given attitude. The closed borders discourse (blue line) displays three hikes (where it surpasses the open borders one—orange line): the first in the morning hours of the November 10—exactly when Cameron was presenting his demands at the Chatham House—and two more on the next day. The two November 11 hikes were mostly due to an intense retweet activity around two specific tweets by Nigel Farage (the UKIP leader accused Camreon of ‘fiddl[ing] with migrant benefits’, whilst ‘EU migration surges to record high’, and suggested to ‘leave the EU to control borders’) and the sharing of a Telegraph article by William Hague, in which the former foreign secretary defined the migration crisis as a ‘gust of the hurricane that will soon engulf Europe’.

On November 12, conversely, the open borders supporters were louder. This increase in activity was mostly due to the Official statistics watchdog’s confuting of the figures used by the PM during his speech. The Independent, in particular, covered the news in two separate stories. This case, together with that of Hague’s article, exemplifies the important role played by news releases in animating Twitter discussions. Table 2 shows the most shared media contents for both groups over the three days of collection. The most shared tweet in the open border group reminded people that ‘EU migrants pay more than they take out in benefits’, whereas the second most shared news story by the closed-borders group focuses on the absolute numbers of newcomers to the UK from Romania and Bulgaria.

Figures 3 and 4 highlight the two discourses’ semantic core, which we obtained by selecting the 50 most shared tweets of, respectively, the closed and open borders groups, and removing from each corpus a set of common words, such as ‘david’, ‘Cameron’, ‘EU’, ‘migration’, and so on.

Both word clouds show the centrality of the welfare benefit theme in the broader discussion on intra-EU migration flows. They also show, however, some distinctive traits. The importance of news sharing appears quite clearly in the closed borders group, which tweeted words such as ‘hurricane’, ‘gust’ ‘mere’ and ‘engulf’, directly derived from the Telegraph story. On the open borders side, it was words such as ‘statistics’, ‘stats’, ‘watchdog’ and ‘figures’ – echoing the clash between statistic authorities and Cameron – that appeared more frequently.

Twitter as a battleground

Our analysis provides interesting information about the nature of Twitter and its reaction to Cameron’s proposals. In the first place, it is clear that Cameron’s speech triggered an attitudinal discussion on Twitter. Approximately 40 thousand tweets (roughly half of the tweets relevant to Cameron’s speech) can be interpreted as containing an attitude of some sort. Of these, more than a fifth, as shown, focused on the intra-EU migration issue.

The reaction we observed was mainly directed to Cameron’s reform proposals. In particular, we noted an overwhelming presence of negative attitude with respect to what was presented by the PM: this result seems consistent with evidence from opinion polls indicating little confidence in Cameron’s ability to strike a good deal out of the renegotiation process.

It is also interesting to see how the ‘offline’ event – Cameron’s Chatham House speech – functioned as a trigger for a much wider discussion on the general principle of free movement inside the European Union, in particular with reference to the welfare benefit debate. In this sense, Twitter appears as a ‘battleground’ for users holding different political views. In such a battleground, active and engaged profiles tend to operate as amplifiers of political leaders’ messages and news items.


[i] Extracting Twitter data on a specific topic means facing a trade-off between ‘precision’ and ‘recall’ (i.e. comprehensiveness): a very precise collection strategy will likely result in tweets that are mostly on topic, but will also miss important conversations. The opposite is the case for more inclusive collection strategies. We opted for a balanced set of keywords and hashtags, containing very specific terms (e.g. #euref, #cameronletter) as well as generic ones, like ‘Cameron’.

(OpenDemocracy, 04.01.2018)