Online social network platforms have been, since their rise to prominence, considered as relevant gateways to study individual behaviors. One of those realms is politics, from massive movements and demonstrations to customary events like general elections or referendums in democratic countries. In this paper, we are interested in the behavior of politicians in the United Kingdom European Union membership referendum of June 2016, colloquially known as the “Brexit referendum”. In this case study we ask how, during the final weeks of the campaign, four political actors (Jeremy Corbyn, David Cameron, Boris Johnson, and Nigel Farage) used Twitter through their official accounts regarding this popular consultation and how that action was received by their followers. To be more precise, we ask if they predominantly appealed to emotions or to rational arguments, and what was, for each of them, the impact of their tweets. Our leading hypotheses were that: 1) the appeal to emotions prevailed over rationality, 1a) particularly in the case of “brexiters”; and 2) that appeal was more effective, 2a) and again that effectiveness was more significant for those that campaigned for “Leave”. We conclude that, while the first hypothesis (and its subsidiary) was not confirmed, the appeal to emotions and the debasement of the opposing views tended to have more relevance, thus confirming the second hypothesis (and particularly its subsidiary), which is consistent with the outcome of the referendum and may be a distressing hypothesis for democracy.
Twitter studies and politics: A literature review
Protocol, dataset, and methodology
Presentation and discussion of results
Online social network sites [SNS], as defined by the widely cited introduction to a special issue of Journal of Computer-Mediated Communication on the subject (boyd and Ellison, 2007), or occasionally “social media”, or even more recently “platforms” (cf., Gillespie, 2017, 2010; Helmond, 2015), have been, since their rise to prominence, considered as relevant gateways to study individual behaviors, both within the platforms themselves and on the multiple ways they are interconnected to “real life”, i.e., to off-line social interaction.
One such case is political behavior, particularly when the issues at stake are massive — sometimes disruptive — movements and demonstrations, e.g., the so-called “Arab Spring” and the “Occupy” movements (cf., Hemsley and Eckert, 2014; also Fuchs [2017; 2014] for a skeptical approach on SNS’s role as mobilizers), but also with ordinary and regular events like general elections in democratic countries, or less common yet still institutionalized ones, as referendums.
Facebook and Twitter are, given their current indisputable worldwide dominance among social network platforms, also the ones that gather more attention, both by the media and by the academic field. Facebook due to the algorithm’s features (that allegedly lead to the emergence of “bubbles”) and also to the permeability to “fake news”, along with the concerns regarding retrieval of massive data from users for targeted political advertising, particularly after the “Cambridge Analytica scandal” of 2018. Twitter, the one in which we will be focused, among other reasons because of its claimed inherent potential for an increased sense of proximity between followers and account holders.
This appeal to proximity was already inscribed in the original concept behind Twitter (an equivalent to text messaging for cyberspace) and in its default options (accounts and tweets are public unless otherwise stipulated by the user; also by default, following does not require an approval from the follower). It may have been one of the reasons for its popularity as something distinct from Facebook (and arguably complementary to it), but also for an early distrust of the platform as a relevant (or even serious) medium, as we will briefly discuss in the next section. Meanwhile, that distrust was replaced by normalization, to a point in which it is almost de rigueur to have a Twitter account if someone is a public figure or institution, no matter how they use it, who is behind it, and how frequently they update their feed.
In this paper, we are interested in the behavior of a peculiar type of public figure, the politician, and also in a very peculiar event, the United Kingdom European Union membership referendum of June 2016, colloquially known as the “Brexit referendum”. We present a case study on how four political actors used their official Twitter accounts during the final weeks of the campaign for this popular consultation, and how that was received by their followers. To be more precise, we ask to what extent they appealed to emotions or to rational arguments, and what was, in each case, the impact of the tweets.
For reasons that will be enunciated below, in the Protocol, dataset, and methodology section, our focus is narrowed to David Cameron and Jeremy Corbyn, at the time the leaders of the main parties in the U.K. (Conservative and Labour, respectively), both in favor of “Remain”; and, on the “Leave” side, Boris Johnson (also from the Conservative Party) and Nigel Farage (at the time from the nationalist party UKIP).
Though it is not a goal the establishment of a strict statistic correlation with the results of the referendum, some trends will be extrapolated from the analysis of the available data. The objective is accomplished mostly through a quantitative approach, proposing a model loosely inspired by Clemencia Rodriguez, et al. (2015) on the usage of Twitter during the Colombian Peace Process, which also culminated with a referendum. In that paper, the authors classified the tweets of the chosen political actors according to what they called “axes” (the negotiation process, the social climate, and narratives and perceptions about adversaries), in each case subdividing them in more specific themes. “Axes” will mean something different in our modified model: we take into account a Usage Axis (which themes or categories are present and how are they presented, both quantitatively and in terms of their meaning) and an Impact Axis (to what extent did the followers engage with the posts, mainly in the quantitative sense of the average number of likes, retweets, and comments).
Our leading hypotheses were that: 1) regarding the Usage Axis, the appeal to emotions prevailed over rationality, 1a) particularly in the case of “brexiters”; and, 2) considering the Impact Axis, such an appeal to emotions (and the debasement of the opposing views) was more effective, 2a) also particularly in the case of those that campaigned for “Leave”. As we shall see, while the first hypothesis (and its subsidiary) was not confirmed, the second hypothesis (and especially its subsidiary) was indeed corroborated. Although in need of further confirmation by other empirical studies on Twitter and other platforms, this is consistent with the outcome of the referendum and may constitute, in the long run, a distressing hypothesis for democracy.
Twitter studies and politics: A literature review
Rogers (2014) explains how Twitter grew — in the minds of the users, but more importantly, though always a few steps behind, for researchers — from a Phase I when it was regarded as “‘pointless babble’ [and] ‘daily chatter’”  to a way of learning about and following events in real-time, especially disasters and emergencies (Phase II), and then to a large and analyzable data-set (Phase III):
“for researchers Twitter has evolved from a phatic and ambient intimacy machine [...] to an event-following and news machine [...]. More recently it has settled into a data set, from which researchers have made collections, and one to be archived and made available by the U.S. Library of Congress.” 
Although Rogers conflates, with this categorization of phases that are also potential usages of the platform, at least two distinct levels — a “denotative” level of activity mainly in phases I (“babble”) and II (“event-following”), and an epistemic “meta” level more prevalent in phase III (“data”) — what matters for our current purposes is that the political dimension also increased in relevance as the platform matured, enabling it as a virtual forum for political debate, thus evoking the Habermasian hypothesis that equates communicative action (and thus democracy) with the full and untethered usage of rationality .
But what is studied when the subject is the political usage of Twitter or other social media? And how has that subject been approached, in terms of methodology? Though we do not aim at a complete review of the current state of the art in studies about Twitter and politics — a task already undertaken for example by Campos-Domínguez (2017) — at least some kind of systematization is needed.
Twitter and the personalization of politics
One of those strains of research assumes that SNSs, as communication technologies, enabled new forms of interaction between politicians and electors, usually translated into a greater potential for personalization, either in campaigns or in “everyday” political activity (Vergeer and Hermans, 2013). As Enli and Skogerbø (2013) try to generalize:
“Our hypothesis is that social media, as a result of their design, affordances, their interplay with other media and the opportunities for creating intimate relations to voters, add to processes of personalization [...] thereby expanding the political arena for increased personalized campaigning.” 
In a study combining a survey and a laboratory experiment, Kruikemeier, et al. (2013) “observed that interactive, personalized online communication has a positive effect on citizens’ feelings of having the opportunity to come into contact with politics, and citizens’ feelings of closeness to politics” . However, that increased sense of participation does not necessarily entail a change in voting behavior in favor of those actors that have a better grasp of online platforms (cf., Labella, 2012). Nevertheless, we may ask if those “feelings of closeness” foster the appeal to emotions and diminish the value of rational arguments, as stated by the hypotheses of our case study.
Twitter vs. traditional media
At least for now, stronger claims of a more effective power of Twitter over traditional media are still lacking conclusive evidence by empirical assessments on the subject. Or it may as well be simply the case that politicians are still far from mastering these new media (Graham, et al., 2013a; Zamora Medina and Zurutuza Muñoz, 2014; Gómez-Calderón, et al., 2017). There are occasional exceptions of more technologically engaged ones, usually also connoted with fringe politics (Jürgens and Jungherr, 2015; López-Meri, et al., 2017 ), that strive to use the platform to enter into dialog with their followers instead of replicating the “classical” top-down communication. But even then “there is little evidence that shows that these forms of campaigning actually have an effect” .
At best, as Campos-Domínguez and Calvo (2017) argue, a social media strategy for electoral acts can only be effective as part of a widely coordinated campaign that reinforces the messages delivered through more traditional media, and besides
“the messages [...] that achieve a higher degree of viralization are precisely those that have been broadcast first on television or those that, taking advantage of a televised event (such as an electoral debate), achieve their viralization on Twitter.” 
As we will see below, one of the proposed categories of tweets in our study is related to this articulation between Twitter and traditional media, but only in the form of future event announcements (rallies, participations in TV shows) and other calls for off-line action.
Twitter as a reflection of the political mood
Another line of analysis, rather than being focused on the immediate or long-term effects of Twitter in politics, i.e., as a modifier of behavior, assumes it as a reliable mirror of that behavior (at least provided that the adequate methodology is applied). Elections and other events that can be isolated in a time frame are the most obvious case studies to test the assumption that either this one or other SNSs are better replacements for surveys when it comes to predictability of the results, but also of the fluctuations of the “collective mood” during campaigns. Some innovative methodologies have recently been devised for that purpose, mostly dealing with sentiment analysis and AI assisted opinion mining based on natural language processing (Jahanbakhsh and Moon, 2014; Lansdall-Welfare, et al., 2016; Hürlimann, et al., 2016; Vasiliu, et al., 2016; Bovet, et al., 2018; Celli, et al., 2016; Grčar, et al., 2017; also Khatua and Khatua, 2016, based on volumetric analysis of hashtags).
While these tend to describe themselves in an optimistic fashion as predictors of the outcomes, prudence and a cautious skepticism regarding the “big data” approach that “aims to transform the social sciences into computer science”  seem to be the most appropriate stances . Or at least while the demographics of these platforms’ users (active users, we might underline) are disproportionately different from the electors . In an even more demolishing meta-analysis, we read that “such predictions so far have proven to be not better than chance, thus exposing the limits of predictability of elections by means of social media data” .
Those limits were also exposed by Cram, et al. (2017), who present a report on Twitter posts during the 2017 British general elections showing “an overwhelming dominance of pro-Labour tweeting” that clearly differs from the electoral results. Thus, for the authors,
“Twitter is of course not representative of the voting public as a whole, and therefore not necessarily a clear reflection of ‘the many, not the few’. However, whilst Twitter cannot be used to predict elections and the overwhelming support we see for Labour and Jeremy Corbyn may not be fully reflected in the ballot boxes, it is a useful tool in giving us the mood of those who are motivated enough to comment in social media.” (Cram, et al., 2017; our emphasis)
A similar acknowledgment was already stated in a study on the 2010 Australian elections, in which the authors warned that
“while the #ausvotes Twitter community very clearly does not exist in a vacuum, and is thus influenced by political events and media [...] these events and themes are filtered through the community’s own established interests and news frames, resulting in a distribution of attention that is different from that of the mainstream media or of general public debate.” 
All of these previous studies led us to opt for a similar approach, related to that broad assumption of a homology between the “Twittersphere” and the actual events, but acknowledging the caveats mentioned supra. What matters for us, at least in the scope of this paper, is not to demonstrate if and how the results could be predicted. It is rather to evaluate, regardless of causal direction, whether there was some consistency (or inconsistency, if the hypotheses are invalidated) between the activity in this platform regarding the referendum and the referendum outcome itself.
The specific case of Twitter and the “Brexit” referendum
Several papers have chosen this referendum as their research object, thus confirming that it was a particularly relevant event for what we could colloquially call “Twitter studies”.
Lansdall-Welfare, et al. (2016) aimed at detecting changes in the public mood during the campaign, correlating them with other social and political events. Khatua and Khatua (2016) did a clustering analysis of tweets trying to prove that the outcome could be predicted, while Celli, et al. (2016) claimed that the method of natural language processing allied to opinion mining is even more reliable. Real-time monitoring by Vasiliu, et al. (2016) interpreted the tweet activity as a close call but with a “slow but constant downward trend for the ‘Remain’ side”.
With the aim of evaluating “the relation between the Twitter mood and the referendum outcome, and who were the most influential Twitter users in the pro- and contra-Brexit camps”, Grčar, et al. (2017) acknowledged, resorting to the Hirsch index, that “users tweeting from the Brexit perspective have generated a larger volume of content, and are better at tagging their contributions, in order to link posts to a broader argument and wider community of support”, an hypothesis that is somewhat related to our own analysis, namely what we have named as the “Impact Axis”.
Bastos and Mercea (2019) looked at the activity of bot accounts that were created for the occasion and that disappeared after the event, detecting large cascades of retweets that lead to polarized clusters, but also identifying, more than a polarization, what the authors call “hyperpartisanism”, with a clear bias for the “Leave”. That confirms the worries of Howard and Kollanyi about “the risk of massive cascades of misinformation at a time when voters will be thinking about their options and canvasing their social networks for the sentiments of friends and family” .
Protocol, dataset, and methodology
For practical reasons, our analysis is narrowed to a case study involving four actors: David Cameron and Jeremy Corbyn, then the leaders of the main parties in the U.K. (Conservative and Labour, respectively), both in favor of “Remain”; and, on the “Leave” side, Boris Johnson (also from the Conservative Party; Prime Minister of the U.K. since July 2019) and Nigel Farage (then leader of the nationalist party UKIP; leader of the Brexit Party since March 2019).
Even if the choice of these — and not other — actors could be easily justified by the fact that they were among the most mentioned in the news, the official Twitter accounts of the parties (though both Labour and Conservatives were divided on the subject) or those of the movements for or against could also have been chosen.
However, we opted for personal accounts due to three reasons:
- these accounts existed prior to the referendum rather than being created for this sole purpose, already having gathered a significant number of followers;
- though the same can be said regarding the main British parties, these were in some cases divided on the subject while individual politicians were clearly positioned;
- considering that the role of emotional discourse is at the core of this research, it was our belief that it could emerge as more significant — or at least as well represented as the rational arguments — both in personal tweets (even if managed by communication professionals rather than the claimed owners) and in the followers’ engagement.
But ultimately, this should be taken, with its inherent limitations, as an assumed research choice; a feature rather than a bug.
Concerning the temporal scope of the corpus, the “Brexit referendum” took place on 23 June 2016, which by itself constitutes the natural upper time limit of our interval . Though the official campaign started on 15 April, we decided to focus only on the last three weeks, placing the lower time limit of our data retrieval on 1 June.
One of the reasons for this restriction is that we wanted to analyze the more intense period of campaign just before the elections ; the other is a very practical one. Due to a lack of resources to obtain full data-sets via software tools that use the API, our method of retrieval was Twitter’s own “advanced search” function, that, provided the scope is a relatively short period of time (as was the case), returns the full list of tweets within that date range .
The 422 tweets (and their timestamps) were manually copied and classified independently by both authors, according to the following set of categories that constitute the Usage Axis. The order in which these categories are presented reflects a decreasing degree of rationality. Some illustrations are presented below.
A: “Presentation of arguments and counterarguments” [positive, rational];
B: “Appealing to third party arguments” [positive, rational, but weaker than A];
C: “Event announcements and calls for off-line action”  [neutral];
D: “Disqualification of the opposition” [negative, emotional];
Off topic: All remaining tweets that do not have the referendum as theme. Considered as an additional category for statistical purposes, it will be briefly discussed in the Usage Axis subsection of the presentation of results but discarded for the main part of the analysis.
The classification of the corpus aimed at the accomplishment of the principles of mutual exclusion (i.e., only one category per tweet) and exhaustiveness (i.e., all tweets must belong to a category) . We have determined the category of each tweet through contextual reading, given that an automatic or semi-automatic categorization based solely on the presence of specific words and phrases could lead to incorrect attributions. Besides, the emphasis on the opposition between rational arguments and the appeal to emotions is shaped not so much by isolated words but rather on the tweet itself (and on external content in those cases where the tweet contains pointers to webpages, images or videos) as a semantic unit.
A few illustrations, in Table 1 below, are helpful to understand the process of categorization.
Table 1: Examples of tweets and attributed categories. Author Content URL Category Example 1 Jeremy Corbyn We want to remain in the EU to build a real social Europe with workers at its heart. In Newquay for #LabourIn rally https://twitter.com/jeremycorbyn/status/739110222816542720 A (Presentation of arguments and counterarguments) Example 2 David Cameron Vote Remain — so that our children and grandchildren have a brighter future https://twitter.com/david_cameron/status/745870103397404672 A (Presentation of arguments and counterarguments) Example 3 Nigel Farage Mr. Cameron cannot answer why he pledged to reduce migration in his manifesto when door to the EU is completely open #InOrOut https://twitter.com/Nigel_Farage/status/738447952000192512 A (Presentation of arguments and counterarguments) Example 4 Boris Johnson Gove is dead right on the elites vs the rest. EU a vast stitch up for big business #VoteLeave #InOrOut https://twitter.com/BorisJohnson/status/738813919876255744 B (Appealing to third party arguments) Example 5 David Cameron I’ll be on @SkyNews at 8pm, explaining why Britain is stronger, safer and better off in the EU — and why leaving would be a leap in the dark https://twitter.com/David_Cameron/status/738436221010116610 C (Event announcements and calls for off-line action) Example 6 Jeremy Corbyn You have 5 days left to register to vote for the EU referendum http://www.gov.uk/register-to-vote #Eureferendum #LabourIn https://twitter.com/jeremycorbyn/status/738472086356660224 C (Event announcements and calls for off-line action) Example 7 Nigel Farage Hysterical comment that fills me with great cheer. Mr. Cameron knows he’s going to lose. #Brexit https://twitter.com/Nigel_Farage/status/739782675251220480 D (Disqualification of the opposition) Example 8 Nigel Farage Mr. Cameron said UK would do fine outside EU, now daily predictions of doom. I don’t believe a word this man says https://twitter.com/Nigel_Farage/status/741926802034040832 D (Disqualification of the opposition) Example 9 David Cameron If the pound falls because of Brexit, prices would go up & UK families would be hit. Leave’s Nigel Farage’s response to this: ‘So what?’ https://twitter.com/david_cameron/status/741918376759599104 D (Disqualification of the opposition) Example 10 David Cameron It’s becoming increasingly clear the Leave campaign don’t have a plan and are prepared to take a leap in the dark. We’re #StrongerIn the EU https://twitter.com/david_cameron/status/739479869596307457 D (Disqualification of the opposition) Example 11 Jeremy Corbyn The threat to British people is not the European Union — it is the Conservative government here in Britain #LabourIn https://twitter.com/jeremycorbyn/status/738398941343932416 D (Disqualification of the opposition) Example 12 Nigel Farage Disgusted that Bob Geldof was abusing fishermen today. They want to earn an honest living, made impossible by the EU https://twitter.com/Nigel_Farage/status/743096883527208961 D (Disqualification of the opposition)
Jeremy Corbyn’s tweet in Example 1, starting with “We want to remain in the EU to” and followed by an argument (even if very condensed for obvious reasons), signals a tweet belonging to category A. Almost the same can be said for Example 2, by David Cameron (“Vote Remain — so that our children and grandchildren have a brighter future”). In the case of Example 3, Nigel Farage’s expression “Mr. Cameron cannot answer why”, although addressing the opposition, is not disqualifying it, but rather trying to engage in a more rational line of discussion; hence, it also fits in category A.
A similar structure is present in Example 4, but in this case someone other than Boris Johnson, the author of the tweet, takes the burden of argumentation — “Gove is dead right on the elites vs the rest”. This expression of agreement with Michael Gove, another Conservative Party politician that was on the “Leave” side, places the tweet in category B.
Whenever the emphasis is on some upcoming event outside the scope of online social media, the tweets are classified in Category C. Such is the case of Example 5, which in spite of a second part slightly leaning to category A, announces David Cameron’s participation on a televised debate later on that day (“I’ll be on @SkyNews at 8pm”). Other tweets of this category are the appeals for the attendance of off-line activities related to the referendum, such as rallies, or, as is the case of Example 6, one of several pledges by Jeremy Corbyn to register to vote.
In Examples 7 and 8, Nigel Farage’s expressions “hysterical comment” or “I don’t believe a word this man says”, referring in both cases to David Cameron, are clear indicators of tweets belonging to category D. In Examples 9 and 10 (from David Cameron) and also 11 (from Jeremy Corbyn ) the focus is not only on the harmful consequences of leaving the EU — the devaluation of the pound, “leap in the dark”, “threat to British people” — but also, instead of an explanation of why that may be the case (which might have placed them in categories A or B), those outcomes are waved to blame the supporters of “Brexit”. Finally, in Example 12, a tweet by Nigel Farage with a photo of singer and activist Bob Geldof supposedly insulting fishermen in a rally, there is an outright shaming of the singer.
It is also helpful to look at a chart published by the Economist (Figure 1), which gives a good outline of the main arguments of “Remain” (left) and “Leave” (right). Tweets that have as their main subject one of these arguments very likely belong to category A, or at least B if attributed to a third party. Other claims need a closer inspection, but not being part of this list is far from meaning that they belong to category D.
Figure 1: Main arguments for and against Brexit (Source: Economist, 2016).
These four categories (five if “off topic” is included) add up to what we have labeled as the Usage Axis, a nominal qualitative variable. Relevant to our goals are the frequencies (absolute and relative) of each category for each of the four actors, but also, to a lesser extent, the textual contents of the respective tweets.
The engagement (i.e., the reactions) for each tweet was also recorded. These reactions, a set of discrete quantitative variables, constitute the Impact Axis: 1) the number of likes for each tweet; 2) the number of comments or replies; and, 3) the number of retweets. A closer look at how individual tweets performed concerning these metrics would give a detailed but fragmented portrait. In order to examine if there are global patterns and to keep the analysis manageable , we have aggregated them through a matrix that intersects each of the four actors with the categories of the Usage Axis, and in each category and for each actor the average of likes (replies, etc.) will be our metric.
Presentation and discussion of results
We start with a global description of the corpus and presentation of the trends identified through this categorization, along with some relevant highlights.
The following table (Table 2) sums up the distribution of tweets (absolute and relative) for each category and candidate. The first relative value includes off topic tweets (Figure 2a for a graphical representation); the second one excludes them (Figure 2b).
Table 2: Distribution of tweets for each category and candidate.
Note: * First relative figures represent the percentage of the grand total (including “off topic”); second figures represent the percentage excluding “off topic”.
Jeremy Corbyn David Cameron Boris Johnson Nigel Farage A: Presentation of arguments and counterarguments 28
B: Appealing to third party arguments 13
C: Event announcements and calls for off-line action 36
D: Disqualification of the opposition 10
Total excluding off-topic tweets 87
[70.7% of grand total]
[92.0% of grand total]
[90.2% of grand total]
[94.2% of grand total]
Grand total of tweets between 1 June and 23 June (until cut-up) 123 138 41 120
Figure 2: Relative distribution of tweets by category/actor during the period under scrutiny. Figure 2a (top) includes off-topic tweets; Figure 2b (bottom) discards them. Cf., Table 2, which contains absolute and relative values for each category/actor.
All actors except Boris Johnson made frequent and sustained tweets during the period under scrutiny, averaging between five and six tweets per day, with an absolute total of 123 (Corbyn), 138 (Cameron) and 120 (Farage), before discarding those that were considered off topic. Johnson was more modest, with an average of slightly less than two daily tweets, totaling 41, with a high occurrence of tweets of category B (“Appealing to third party arguments”).
The overwhelming majority of the tweets were about the referendum, with the percentages of off-topic posts being below 10 percent for Cameron, Johnson and Farage. The remarkable exception was Jeremy Corbyn, with almost a third of his messages (36 out of 123) being about other topics occurring during that period: current political issues, regrets for the murder of Labour MP Jo Cox and for the massacre in a disco in Orlando, Florida, the Euro 2016 soccer cup, etc. These and other current issues, such as the death of Muhammad Ali on 3 June, are also identifiable in the off-topic tweets of the other actors.
Table 2 also shows that, although in opposite sides of the referendum, both Cameron and Johnson had a similar behavior, often relying in other actors to legitimize their positions: i.e., in both cases category B earned the highest frequency of tweets. Corbyn privileged the communication of events (C), also a significant category for Cameron. Surprisingly, the actor that had a higher percentage of tweets classified in category A was Nigel Farage. This must, however, be paired with other data, namely the Impact Axis (see below).
For nearly everyone, D was the category with fewer tweets (“disqualification of the opposition”), with Farage having this category just slightly ahead of B, by one tweet. Again, the pairing with the Impact Axis may dye this early picture with different colors.
A preliminary graphical representation of the impact axis can be seen in Figures 3a to 3d, which, along with the absolute number of tweets for each actor, also display the average of likes, comments and retweets for each category.
Figure 3a: Jeremy Corbyn: Absolute number of tweets, raw average of likes, comments and retweets (all divided by category). Figure 3b: David Cameron: Absolute number of tweets, raw average of likes, comments and retweets (all divided by category). Figure 3c: Boris Johnson: Absolute number of tweets, raw average of likes, comments and retweets (all divided by category). Figure 3d: Nigel Farage: Absolute number of tweets, raw average of likes, comments and retweets (all divided by category).
Nigel Farage, while on a par with Corbyn and Cameron regarding the frequency of tweets and how they were distributed along the usage axis, was by far the most successful actor in two of the three variables of the impact axis (in raw numbers), with an average of likes approaching or even overcoming 1,000 per tweet. He also led in the average of retweets, in either case regardless of category. The leader in the other variable we contemplated, the absolute average of comments, was David Cameron, again for all categories.
It is also relevant to note that Boris Johnson, in spite of being the actor who resorted less to Twitter, and also the one with less followers (as we will discuss below), managed, in the impact axis, to achieve an average of likes always at least close to Cameron and Corbyn, in some categories (C and D) sometimes even coming in second place. The scenario was very similar regarding the average of retweets — near Corbyn and Johnson in categories B and C, approaching Farage in category D. In the case of comments, the performance of Boris Johnson is however only significant in category A.
If interpreted from the point of view of the categories, the most remarkable regularity is the impact of the tweets disqualifying the opposition (D), always achieving a high degree of engagement.
These figures allow a first approximation of how the engagement varied for each politician and for each category, but we wanted to crosscheck this insight with another metric for a more accurate representation of that impact. This must be considered as dependent on the amount of followers (in spite of the fact that tweets are public): an average of 500 retweets, for example, has a meaning for an account with 1,000 followers and a different one for another with 100,000. Our strategy to address this issue was to “flatten” or “normalize” the averages of likes, comments and retweets by considering not raw averages as above but the average per 100,000 followers. One hundred thousand emerged, after a few trials, as the best to achieve a scale similar to the one with the number of tweets, thus enabling an easier numeric and visual comparison. In this particular case, and although it is impossible to discern the precise number of followers at the time of the referendum, we chose the arbitrarily-defined proxy of the number of followers at the time of data retrieval. This was then rounded to the nearest 1,000, thus giving an approximation with an error margin of one percent (again considering 100,000 as the base unit for the ratio).
Jeremy Corbyn and David Cameron were almost head to head, respectively with 1,800,000 and 1,890,000 followers. Nigel Farage trailed slightly behind, with approximately 1,200,000, and Boris Johnson, the only that was not a party leader, lagged with 442,000 . Table 3, below, summarizes these results, and Figures 4a to 4d present them in graphical form.
Table 3: Absolute number of tweets, average of likes, average of comments and average of retweets per 100,000 followers, for each candidate and category.. Jeremy Corbyn David Cameron Boris Johnson Nigel Farage Approximate number of followers 1,800,000 1,890,000 442,000 1,200,000 A: Presentation of arguments and counterarguments 16
Av. likes/100kf.: 31.00
Av. com./100kf.: 4.05
Av. retw./100kf.: 26.08
Av. likes/100kf.: 69.86
Av. com./100kf.: 31.59
Av. retw./100kf.: 32.47
Av. likes/100kf.: 258.96
Av. com./100kf.: 72.15
Av. retw./100kf.: 137.62
Av. likes/100kf.: 127.53
Av. com./100kf.: 13.78
Av. retw./100kf.: 90.73
B: Appealing to third party arguments Tweets: 13
Av. likes/100kf.: 24.94
Av. com./100kf.: 3.40
Av. retw./100kf.: 15.81
Av. likes/100kf.: 24.11
Av. com./100kf.: 19.35
Av. retw./100kf.: 14.87
Av. likes/100kf.: 101.51
Av. com./100kf.: 13.95
Av. retw./100kf.: 67.26
Av. likes/100kf.: 80.25
Av. com./100kf.: 9.73
Av. retw./100kf.: 74.06
C: Event announcements and calls for off-line action Tweets: 36
Av. likes/100kf.: 15.81
Av. com./100kf.: 2.38
Av. retw./100kf.: 13.60
Av. likes/100kf.: 18.86
Av. com./100kf.: 13.59
Av. retw./100kf.: 11.06
Av. likes/100kf.: 109.50
Av. com./100kf.: 17.91
Av. retw./100kf.: 49.36
Av. likes/100kf.: 76.69
Av. com./100kf.: 8.45
Av. retw./100kf.: 43.38
D: Disqualification of the opposition Tweets: 10
Av. likes/100kf.: 31.42
Av. com./100kf.: 5.49
Av. retw./100kf.: 33.64
Av. likes/100kf.: 20.51
Av. com./100kf.: 28.62
Av. retw./100kf.: 14.24
Av. likes/100kf.: 240.05
Av. com./100kf.: 48.79
Av. retw./100kf.: 190.72
Av. likes/100kf.: 129.53
Av. com./100kf.: 20.16
Av. retw./100kf.: 99.86
Total (excluding off-topic tweets) Tweets: 87 Tweets: 127 Tweets: 37 Tweets: 113
Figure 4a: Jeremy Corbyn: Absolute number of tweets, then average of likes, comments and retweets per 100,000 followers (all divided by category). Figure 4b: David Cameron: Absolute number of tweets, then average of likes, comments and retweets per 100,000 followers (all divided by category). Figure 4c: Boris Johnson: Absolute number of tweets, then average of likes, comments and retweets per 100,000 followers (all divided by category). Figure 4d: Nigel Farage: Absolute number of tweets, then average of likes, comments and retweets per 100,000 followers (all divided by category).
The impact of the tweets becomes clearer with the help of the table and the graphics. Reading the table vertically, we get a global grasp of how each category fared for a particular actor; reading it horizontally, how a specific category can be compared across actors. For example, the averages of likes per 100,000 followers of Johnson’s tweets are outliers, regardless of the category. Conversely, the averages of comments per 100,000 followers of Jeremy Corbyn are always the lowest numbers. Another example: tweets from Corbyn or Cameron almost always generate less engagement (regardless of categories or of type of reaction) than those from the politicians that were campaigning for “Leave”.
Focusing on the likes, the behavior (i.e., category) that consistently led to a higher increase in the average of likes was the disqualification of the opposition (D), although the presentation of arguments (A) also fared good for all actors. In the case of Cameron, actually the presentation of arguments (A) was slightly more efficient in this regard, even if his arguments were most of the time as simplistic as “Vote Remain — so that our children and grandchildren have a brighter future.” (Example 2 in the “Protocol, dataset and methodology” section). The announcement of events and calls for off-line action — the dominant category in the case of Jeremy Corbyn, as mentioned above — was also the less effective one, even if this category mobilized for Farage and for Johnson a significant amount of likes.
Comments were mostly scarce, although Cameron and Johnson managed to attract, for almost all categories, a reasonable amount; Cameron inclusively gathered a higher average of comments than likes in category D, a unique case. Two illustrations, that were also those eliciting more comments in this category, are Example 9 (“If the pound falls because of Brexit, prices would go up & U.K. families would be hit. Leave’s Nigel Farage’s response to this: ‘So what?’”), with 869 comments at the time of retrieval, and Example 10 (“It’s becoming increasingly clear the Leave campaign don’t have a plan and are prepared to take a leap in the dark. We’re #StrongerIn the EU.”), with 827 comments. Apart from that, another regularity we observed was that categories A and D — the two extremes in the rational vs. emotional spectrum — performed slightly better at generating engagement in the form of comments than the other two.
As to retweets, what happened was similar to the likes. Jeremy Corbyn was the actor who achieved a poorer engagement with tweets of category A, and all four had less than remarkable (sometimes even dismal) results with tweets of category C, though with an advantage for Boris Johnson. The appeal to third party arguments, i.e., category B, led to a good outcome for Nigel Farage and even for Johnson, although not for Cameron. But the category that for almost all actors had the best turnover (just a bit less for Cameron) in terms of retweets was again D, the disqualification of opposition. An example from the “Leave” side is the tweet by Nigel Farage (Example 12 in the “Protocol ...” section), with 2416 retweets: “Disgusted that Bob Geldof was abusing fishermen today. They want to earn an honest living, made impossible by the EU”. Another, this time from the “Remain” side and from Jeremy Corbyn, is Example 11, with 1235 retweets (“The threat to British people is not the European Union — it is the Conservative government here in Britain #LabourIn”).
Some final remarks concerning the differences in the impact axis for each actor. Jeremy Corbyn and David Cameron, both supporters of “Remain”, both leaders of their parties, and both approaching two million followers, had mediocre results in the engagement metrics, as figures 4a and 4b show. This may either be a confirmation of what would be the outcome of the referendum, or simply mean that the more followers one has, the higher percentage of “unengaged” followers that seldom check their Twitter accounts or just do not bother to participate. The confrontation of the third set of figures (3a, 3b, 3c, 3d) with the fourth one (4a, 4b, 4c, 4d) seems however to favor the first interpretation: they could at least have achieved comparable results to the other two actors in our “normalized” distribution, which is not the case. In other words, on the side of the “brexiters” Boris Johnson and Nigel Farage, a small amount of tweets — and, in the case of Johnson, an even smaller amount of followers — was enough to enable a high level of impact through the acts of liking and retweeting, even if not particularly significant in the case of comments.
This case study analyzing the presence on Twitter of four political actors during the United Kingdom European Union membership referendum of June 2016 — and how that led to different forms of engagement by the followers — enabled us, first of all, to identify distinct but sometimes overlapping tweeting behaviors.
Jeremy Corbyn, David Cameron and Nigel Farage had a regular presence in the platform during that period, most of them (although less in the case of Corbyn) mentioning predominantly the upcoming electoral act. Boris Johnson had a more humble presence; besides, a significant percentage of his posts, though related to “Brexit”, delegated in a third party (often his Conservative colleague Michael Gove) the burden of argumentation. And while Cameron also relied on others’ arguments, he did it in a more balanced fashion, with event announcements and calls for off-line action coming in a close second place. For Jeremy Corbyn, this latter category of event announcements took the lead (cf., Graham, et al. ; López-García, ; Campos-Domínguez and Calvo  on the articulation between Twitter and traditional media), though the presentation of arguments and counterarguments appealing to the vote on “Remain” was also significant. Surprisingly, the presentation of arguments — for the “Leave” side, in his case — was by far the most common category in Nigel Farage’s tweets. The most “passionate” category, i.e., the disqualification of the opposition, was always (with the exception of Farage, but by just one tweet) the least common.
These findings invalidate — unexpectedly, we may add — our first hypothesis, i.e., that the appeal to emotions prevailed over rationality. Concerning the Usage Axis, the tweets of category D were in minority for every actor we analyzed, either for those that stood for “Remain” or — also invalidating hypothesis 1a) — those that were in support of “Leave”. Even if their arguments (or those from third parties that they reproduced) could be considered fallacious or based on untrue facts — an analysis which is beyond the scope of this paper — the ones pertaining to categories A and B were more frequent than those that were merely appeals to emotion through a disqualification of the opposition. This nevertheless does not invalidate previous literature regarding the affordances of social media facilitating personalization in politics (Vergeer and Hermans, 2013, Enli and Skogerbø, 2013; Kruikemeier, et al., 2013; Kruikemeier, 2014), rather calling our attention to the need to distinguish between personalization and the appeal to emotion, as the former does not presuppose the latter.
However, as we mentioned in the previous section, the Impact Axis conveys a very different interpretation of their tweets. An attempt to sum it up in a sentence could be “passion trumps reason”, something that in a slightly more nuanced synthesis may be observed in two distinct but apparently related dimensions: a) the absolute number of tweets and the absolute number of followers cannot be correlated with a higher impact; b) the category of a tweet matters, and thus may be a more accurate predictor of its impact.
To be more specific, Boris Johnson gathered a highly significant amount of likes and of retweets, surpassing in absolute numbers the impact of Corbyn and Cameron (not so much for comments, where Cameron took the lead). That in spite of being the less active of the four politicians and by far the one with less followers. All actors had, on average, a better performance — again with Cameron as a near exception — with posts that merely disqualified the opposition without a rational or at least rational-like argument behind it. And while this was true for both “remainers” (Corbyn, and Cameron regarding comments) and “leavers”, in the case of the latter the impact of these “passional” tweets was distinctly much more pronounced.
For these reasons, our second hypothesis, regarding the Impact Axis, was confirmed due to the high impact (i. e., engagement) of the tweets from category D, though it is a weak confirmation given that the impact of the rational tweets of category A is also considerable. The appeal to emotions (here identified with the disqualification or even debasement of the opposing views) was thus very effective. Tweets classified in category D induced consistently, on average, a higher degree of engagement (either in the form of likes, of comments, or of retweets) from the audience than those belonging to other categories. Hypothesis 2a) was also confirmed, and with a higher degree of confidence, given that such a disproportion was much more remarkable for those that campaigned for “Leave”.
The conjunction of these dimensions does not necessarily entail that the result of the referendum could be predicted via Twitter, in spite of some claims we have presented in the literature review section (e.g., Khatua and Khatua, 2016; Celli, et al., 2016), or even less that the elections were won through Twitter (e.g., as warned by Graham, et al., 2016; Kruikemeyer, 2014). It is, nevertheless, consistent with that very same outcome (and in line with Cram, et al., 2017, Vasiliu, et al., 2016; Grčar, et al., 2017), and it is also consistent with other studies that relied primarily on metrics of engagement (Barberá and Rivero, 2015; Zelenkauskaite and Balduccini, 2017).
As we have mentioned earlier, ours was not a particularly sophisticated methodology concerning statistical or digital methods tools, and phenomena such as the eventual concerted action of bots and trolls (cf., Bastos and Mecrea, 2019; Lllewellyn, et al., 2018), or for that matter all forms of misinformation (Howard and Kollanyi, 2016) had to be disregarded for simplicity’s sake. It gave us, nevertheless, a clear picture of what happened in Twitter during the campaign and particularly of its reception by users that followed it through that platform. Perhaps even more important, it gives us a warning sign concerning the usage of online social networks as platforms for political action. While their supposed great potential for the rational exchange of ideas, thus strengthening the tenets of democracy, still stands as a valid working hypothesis, the opposite view — i.e., that they may constitute instead a danger for that very same democracy — cannot be dismissed. Not necessarily because of the discourse within Twitter, which at least in this case study turned out after all to be more rational than expected, but eventually due to the way that some attitudes and discourses that transcend Twitter are amplified (or muted) by the particular dynamics of the platform.
About the authors
Jorge Martins Rosa is an Assistant Professor in the Communication Sciences Department and researcher of ICNOVA, both at NOVA FCSH (Lisbon, Portugal). He teaches, among other courses, “Cyberculture” and “Pop Culture”. He is Principal Investigator of the project “Fiction and the Roots of Cyberculture” (2010–2012) and currently of “PINBook PT”, on the political participation on Facebook.
E-mail: jmr [at] fcsh [dot] unl [dot] pt
Cristian Jiménez Ruiz completed a Master’s degree in communication sciences at NOVA FCSH on Twitter and referendums. Cristian is currently a Ph.D. student in Complexity Sciences at ISCTE-IUL (Lisbon, Portugal), with research interests in Internet studies and digital methods.
E-mail: ccjrz [at] iscte-iul [dot] pt
1. Rogers, 2014, p. 2.
2. Rogers, 2014, p. xxi.
3. Broadly speaking, any technology (e.g., a social networking platform such as Twitter) benefits democracy whenever it facilitates the rational exchange of ideas and arguments, and endangers it when it promotes the opposite behaviors. As much as we must avoid falling into this simplistic kind of technodeterminism, it still is pertinent to observe the extent to which some particular technologies in some particular occasions may lean toward one of those antithetical political “horizons”. A fairly complete and accurate digest of both the utopian and dystopian visions regarding this debate on the potentials or perils for democracy of online social networking platforms may be found in van Dijk (2012). Regarding Twitter, cf. Fuchs (2017) that, in a chapter about Twitter and democracy (pp. 227–230), identifies two main authors on the optimistic side: Clay Shirky and Zizi Papacharissi, and three others in the critical side: Jodi Dean, Malcolm Gladwell, and Evgeny Morozov.
4. Enli and Skogerbø, 2013, pp. 757–758.
5. Kruikemeier, et al., 2013, p. 60.
6. One of the papers mentions the Pirate Party movement in the German elections of 2009 (Jürgens and Jungherr, 2015), the other Pablo Iglesias, the leader of the Spanish party Podemos (López-Meri, et al., 2017).
7. Kruikemeier, 2014, p. 131.
8. Original in Spanish; our translation: “aquellos mensajes [...] que resultan más viralizados son precisamente los mensajes mediáticos, aquellos que han sido primero difundidos en televisión o los que, aprovechando un evento televisivo (como el debate electoral) consiguen su viralización en Twitter.”
9. Campos-Domnguez and Calvo, 2017, p. 112. A paper comparing the tweets of the main political actors in the British and Dutch electoral campaigns of 2010 also acknowledges that “spikes in activity correspond with the televised party leader debates [...] and the final two days of the campaign” (Graham, et al., 2016, p. 773). Although out of the scope of our paper, this may indicate a subsidiary — if not outright parasitical — usage of Twitter regarding more traditional media (cf., also Graham, et al., 2013b; López-García, 2016).
10. Fuchs, 2017, p. 57.
11. Cf., Barberá and Rivero (2015) relying on classic demographic variables combined with metrics such as the number of retweets or replies; also Zelenkauskaite and Balduccini (2017) even though it studies the readers’ comments in a news portal, rather than in a social network platform such as Twitter.
12. Cf., Mellon and Prosser (2017) for a few illustrations of these gaps, though Twitter fares slightly better than Facebook.
13. Gayo-Avello, et al., 2011, p. 490.
14. Bruns and Burgess, 2011, p. 45. Another phenomenon to consider is that the flow of information in these networks tends to be polarized (Primario, et al., 2017), or even “balkanized” (Aragón, et al., 2013), something that the mainstream media also came to acknowledge through the popularization of concepts such as “echo chambers” or “filter bubbles”. What all these concepts intend to signify is that clusters emerge from the interaction between those who share similar ideologies, who then either do not come into contact at all with opposing views or follow them only to debase them and “troll” them instead of engaging in a rational conversation (Llewellyn, et al., 2018). Speaking of trolls, the potential interference patterns of bots, botnets, and fake accounts was discussed by Bastos and Mercea (2019) and again by Llewellyn, et al. (2018), both about the “Brexit” referendum. The latter gathered a large amount of data between 29 August 2015 and 3 October 2017, pinpointing a shift in the day of the referendum “from generalised disruptive tweeting to retweeting each other in order to amplify content produced by other troll accounts”. Though we did not consider the role of bot accounts in our analysis, it may be a relevant path to follow in future research.
15. Howard and Kollanyi, 2016, p. 5.
16. With an additional cut-up on this day when tweets begin to mention the results; i.e., these were also outside the scope of analysis.
17. Cf., Jürgens and Jungherr (2015), who in spite of resorting to API collected data made a similar temporal restriction: “In order to keep the amount of data within reasonable dimensions for both researchers and computers, we furthermore decided to limit our investigations to the ‘hot phase’ of the last month prior to the election.” (p. 473).
18. As an example, the output of our search in Jeremy Corbyn’s official account matches the content of the following URL: https://twitter.com/search?l=&q=from%3Ajeremycorbyn%20since%3A2016-06-01%20until%3A2016-06-23&src=typd; homologous URLs can be generated for the other three actors by replacing the string between “from%3A” and “%20since” with their Twitter handles.
19. Medina e Muñoz (2014) call it “meta-campaign”.
20. In a first round of categorization, there were two “negative” and “emotional” categories — “Distortion of the opposition’s arguments” and “Satanization of the opposition”. An amount of undecided tweets (around 20 percent) remained in this first pass, either because each coder assigned them provisionally to multiple categories or because those differed between coders. All these undecided cases and discrepancies in classification were then discussed between the authors to achieve consensus and comply with the principles of mutual exclusion and exhaustiveness. Given that most of the remaining undecided cases were between two initial “negative” categories — “Distortion of the opposition’s arguments” and “Satanization of the opposition” — the authors opted to merge these into a new category, “Disqualification of the opposition” [D], thus allowing to meet those conditions.
21. Although later in the video that is embedded in the tweet some rational arguments are presented, its most prominent topic is the idea of a “threat” coming from the opposition.
22. The 422 tweets that constitute our universe multiplied by these three variables would amount to 1,266 individual metrics, which would still need to be cross-checked with the five cases of the Usage Axis. Though that may be used in a future analysis, it is outside the objectives of this paper. The same goes for a content analysis of replies and retweets, as these may either endorse or be against the content of the tweet itself.
23. Meanwhile, with his appointment as leader of the Party and Prime Minister, the number of followers has risen to approximately 1.5 million.
Pablo Aragón, Karolin Eva Kappler, Andreas Kaltenbrunner, David Laniado and Yana Volkovich, 2013. “Communication dynamics in Twitter during political campaigns: The case of the 2011 Spanish national election,” Policy & Internet, volume 5, number 2, pp. 183–206.
doi: http://doi.org/10.1002/1944-2866.POI327, accessed 8 January 2019.
Pablo Barberá and Gonzalo Rivero, 2015. “Understanding the political representativeness of Twitter users,” Social Science Computer Review, volume 33, number 6, pp. 712–719.
doi: http://doi.org/10.1177/0894439314558836, accessed 8 January 2019.
Marco T. Bastos and Dan Mercea, 2019. “The Brexit botnet and user-generated hyperpartisan news,” Social Science Computer Review, volume 37, number 1, pp. 38–54.
doi: http://doi.org/10.1177/0894439317734157, accessed 8 January 2019.
Alexandre Bovet, Flaviano Morone and Hernán A. Makse, 2018. “Validation of Twitter opinion trends with national polling aggregates: Hillary Clinton vs Donald Trump,” Scientific Reports, volume 8, article number 8673.
doi: http://doi.org/10.1038/s41598-018-26951-y, accessed 23 February 2020.
danah m. boyd and Nicole B. Ellison, 2008. “Social network sites: Definition, history, and scholarship,” Journal of Computer-Mediated Communication, volume 13, number 1, pp. 210–230.
doi: http://doi.org/10.1111/j.1083-6101.2007.00393.x, accessed 8 January 2019.
Axel Bruns and Jean Burgess, 2011. “#ausvotes: How Twitter covered the 2010 Australian federal election,” Communication, Politics and Culture, volume 44, number 2, pp. 37–56, and at https://search.informit.com.au/documentSummary;dn=627330171744964;res=IELHSS, accessed 8 January 2019.
Eva Campos-Domínguez, 2017. “Twitter y la comunicación política,” El Profesional de la información, volume 26, number 5, pp. 785–793.
doi: http://doi.org/10.3145/epi.2017.sep.01, accessed 8 January 2019.
Eva Campos-Domínguez and Dafne Calvo, 2017. “La campaña electoral en Internet: Planificación, repercusión y viralización en Twitter durante las elecciones españolas de 2015,” Comunicación y Sociedad, volume 29, pp. 93–116, and at http://www.comunicacionysociedad.cucsh.udg.mx/index.php/comsoc/article/view/6423, accessed 8 January 2019.
Fabio Celli, Evgeny Stepanov, Massimo Poesio and Giuseppe Riccardi, 2016. “Predicting Brexit: Classifying agreement is better than sentiment and pollsters,” Proceedings of the Workshop on Computational Modeling of People’s Opinions, Personality, and Emotions in Social Media (PEOPLES), pp. 110–118, and at http://aclweb.org/anthology/W16-4312, accessed 8 January 2019.
Laura Cram, Clare Llewellyn, Robin Hill and Walid Magdy, 2017. “UK general election 2017: A Twitter analysis,” arXiv 1706.02271 (7 June), at https://arxiv.org/abs/1706.02271, accessed 8 January 2019.
Economist, 2016. “A background guide to ‘Brexit’ from the European Union: The Economist’s ‘Brexit’ backgrounder” (24 February), at https://www.economist.com/blogs/graphicdetail/2016/02/graphics-britain-s-referendum-eu-membership, accessed 8 January 2019.
Gunn Enli and Eli Skogerbø, 2013. “Personalized campaigns in party-centred politics: Twitter and Facebook as arenas for political communication,” Information, Communication & Society, volume 16, number 5, pp. 757–774.
doi: http://doi.org/10.1080/1369118X.2013.782330, accessed 8 January 2019.
Christian Fuchs, 2017. Social media: A critical introduction. Second edition. London: Sage.
Christian Fuchs, 2014. OccupyMedia! The Occupy Movement and social media in crisis capitalism. Alresford: Zero Books.
Daniel Gayo-Avello, Panagiotis T. Metaxas and Eni Mustafaraj, 2011. “Limits of electoral predictions using Twitter,” Proceedings of the Fifth International AAAI Conference on Weblogs and Social Media, pp. 490–493, and at https://www.aaai.org/ocs/index.php/ICWSM/ICWSM11/paper/viewFile/2862/3254, accessed 8 January 2019.
Tarleton Gillespie, 2017. “The platform metaphor, revisited,” Digital Society Blog (24 August), at https://www.hiig.de/en/the-platform-metaphor-revisited/, accessed 8 January 2019.
Tarleton Gillespie, 2010. “The politics of ‘platforms’,” New Media & Society, volume 12, number 3, pp. 347–364.
doi: http://doi.org/10.1177/1461444809342738, accessed 8 January 2019.
Bernardo Gómez-Calderón, Sergio Roses and Francisco-Javier Paniagua-Rojano, 2017. “La campaña en 140 caracteres: Empleo de Twitter por parte de los candidatos de los partidos mayoritarios ante las elecciones generales de 2016 en España,” El Profesional de la Información, volume 26, number 5, pp. 816–923.
doi: http://doi.org/10.3145/epi.2017.sep.04, accessed 8 January 2019.
Todd Graham, Dan Jackson and Marcel Broersma, 2016. “New platform, old habits? Candidates” use of Twitter during the 2010 British and Dutch general election campaigns,” New Media & Society, volume 18, number 5, pp. 765–783.
doi: http://doi.org/10.1177/1461444814546728, accessed 8 January 2019.
Todd Graham, Marcel Broersma and Karin Hazelhoff, 2013a. “Closing the gap? Twitter as an instrument for connected representation,” In: Richard Scullion, Roman Gerodimos, Daniel Jackson and Darren Lilleker (editors). The media, political participation and empowerment. London: Routledge, pp. 71–88.
doi: http://doi.org/10.4324/9780203381113, accessed 8 January 2019.
Todd Graham, Marcel Broersma, Karin Hazelhoff and Guido van ’t Haar, 2013b. “Between broadcasting political messages and interacting with voters: The use of Twitter during the 2010 UK general election campaign,” Information, Communication & Society, volume 16, number 5, pp. 692–716.
doi: http://doi.org/10.1080/1369118X.2013.785581, accessed 8 January 2019.
Miha Grčar, Darko Cherepnalkoski, Igor Mozetič and Petra Kralj Novak, 2017. “Stance and influence of Twitter users regarding the Brexit referendum,” Computational Social Networks, volume 4, article number 6.
doi: http://doi.org/10.1186/s40649-017-0042-6, accessed 8 January 2019.
Anne Helmond, 2015. “The platformization of the Web: Making Web data platform ready,” Social Media + Society (30 September).
doi: http://doi.org/10.1177/2056305115603080, accessed 8 January 2019.
Philip N. Howard and Bence Kollanyi, 2016. “Bots, #StrongerIn, and #Brexit: Computational propaganda during the UK-EU referendum,” Project on Computational Propaganda, Working Paper, 2016.1, at https://comprop.oii.ox.ac.uk/research/working-papers/bots-strongerin-and-brexit-computational-propaganda-during-the-uk-eu-referendum/, accessed 23 February 2020.
doi: http://dx.doi.org/10.2139/ssrn.2798311, accessed 8 January 2019.
Kazem Jahanbakhsh and Yumi Moon, 2014. “The predictive power of social media: On the predictability of U.S. presidential elections using Twitter,” arXiv 1407.0622 (1 July), at https://arxiv.org/abs/1407.0622, accessed 8 January 2019.
Pascal Jürgens and Andreas Jungherr, 2015. “The use of Twitter during the 2009 German national election,” German Politics, volume 24, number 4, pp. 469–490.
doi: http://doi.org/10.1080/09644008.2015.1116522, accessed 8 January 2019.
Apalak Khatua and Aparup Khatua, 2016. “Leave or remain? Deciphering Brexit deliberations on Twitter,” IEEE 16th International Conference on Data Mining Workshops (ICDMW), pp. 428–433.
doi: http://doi.org/10.1109/icdmw.2016.0067, accessed 8 January 2019.
Sanne Kruikemeier, 2014. “How political candidates use Twitter and the impact on votes,” Computers in Human Behavior, volume 34, pp. 131–139.
doi: http://doi.org/10.1016/j.chb.2014.01.025, accessed 8 January 2019.
Sanne Kruikemeier, Guda van Noort, Rens Vliegenthart and Claes H de Vreese, 2013. “Getting closer: The effects of personalized and interactive online political communication,” European Journal of Communication, volume 28, number 1, pp. 53–66.
doi: http://doi.org/10.1177/0267323112464837, accessed 8 January 2019.
Luis Izquierdo Labella, 2012. “Las redes sociales en la poltica Española: Twitter en las elecciones de 2011,” Estudos em Comunicaç;ão, number 11, pp. 149–164, and at http://www.ec.ubi.pt/ec/11/pdf/EC11-2012Mai-07.pdf, accessed 8 January 2019.
Thomas Lansdall-Welfare, Fabon Dzogang and Nello Cristianini, 2016. “Change-point analysis of the public mood in UK Twitter during the Brexit referendum,” IEEE 16th International Conference on Data Mining Workshops (ICDMW).
doi: http://doi.org/10.1109/ICDMW.2016.0068, accessed 8 January 2019.
Clare Llewellyn, Laura Cram, Adrian Favero and Robin L. Hill, 2018. “For whom the bell trolls: Troll behaviour in the Twitter Brexit debate,” arXiv 1801.08754 (26 January), at https://arxiv.org/abs/1801.08754, accessed 8 January 2019.
Guillermo López-García, 2016. “‘New’ vs ‘old’ leaderships: The campaign of Spanish general elections 2015 on Twitter,” Communication & Society, volume 29, number 3, pp. 149–168.
doi: http://doi.org/10.15581/003.29.3.149-168, accessed 8 January 2019.
Amparo López-Meri, Silvia Marcos-García and Andreu Casero-Ripollés, 2017. “What do politicians do on Twitter? Functions and communication strategies in the Spanish electoral campaign of 2016,” El Profesional de la Información, volume 26, number 5, pp. 795–804.
doi: http://doi.org/10.3145/epi.2017.sep.02, accessed 8 January 2019.
Jonathan Mellon and Christopher Prosser, 2017. “Twitter and Facebook are not representative of the general population: Political attitudes and demographics of social media users,” Research & Politics (13 July).
doi: http://doi.org/10.1177/2053168017720008, accessed 8 January 2019.
Simonetta Primario, Dario Borrelli, Luca Iandoli, Giuseppe Zollo and Carlo Lipizzi, 2017. “Measuring polarization in Twitter enabled in online political conversation: The case of 2016 US presidential election,” 2017 IEEE International Conference on Information Reuse and Integration (IRI), pp. 607–613.
doi: http://doi.org/10.1109/IRI.2017.73, accessed 8 January 2019.
Richard Rogers, 2014. “Debanalising Twitter, The transformation of an object of study,” In: Katrin Weller, Axel Bruns, Jean Burgess, Merja Mahrt and Cornelius Puschmann (editors). Twitter and society. New York: Peter Lang, pp. ix–xxvi.
doi: https://doi.org/10.3726/978-1-4539-1170-9, accessed 23 February 2019.
Jan van Dijk, 2012. “Digital democracy: Vision and reality,” In: Ig Snellen, Marcel Theans and Wim van de Donk (editors). Public administration in the information age: Revisited. Amsterdam: IOS Press, pp. 49–62.
doi: https://doi.org/10.3233/978-1-61499-137-3-49, accessed 23 February 2019.
Laurentiu Vasiliu, Ross McDermott, Manel Zarrouk, Manuela Hurlimann, Brian Davis, Tobias Daudert, Malek Ben Khaled, David Byrne, Sergio Fernández, André Freitas, Frederico Caroli, Siegfried Handschuh and Angelo Cavallini, 2016. “In or out? Real-time monitoring of BREXIT sentiment on Twitter,” Semantics 2016, at http://ceur-ws.org/Vol-1695/paper31.pdf, accessed 8 January 2019.
Maurice Vergeer and Liesbeth Hermans, 2013. “Campaigning on Twitter: Microblogging and online social networking as campaign tools in the 2010 heneral elections in the Netherlands,” Journal of Computer-Mediated Communication, volume 18, number 4, pp. 399–419.
doi: http://doi.org/10.1111/jcc4.12023, accessed 8 January 2019.
Rocío Zamora Medina and Cristina Zurutuza Muñoz, 2014. “Campaigning on Twitter: Towards the ‘personal style’ campaign to activate the political engagement during the 2011 Spanish general elections,” Communication & Society, volume 27, number 1, pp. 83–106.
doi: https://doi.org/10.15581/003.27.1.83-106, accessed 8 January 2019.
Asta Zelenkauskaite and Marcello Balduccini, 2017. “‘Information warfare’ and online news commenting: Analyzing forces of social influence through location-based commenting user typology,” Social Media + Society (17 July).
doi: https://doi.org/10.1177/2056305117718468, accessed 16 October 2019.
Received 11 January 2019; revised 23 January 2020; revised 4 February 2020; accepted 6 February 2020.
This paper is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
Reason vs. emotion in the Brexit campaign: How key political actors and their followers used Twitter
by Jorge Martins Rosa and Cristian Jiménez Ruiz.
First Monday, Volume 25, Number 3 - 2 March 2020