Election:
The UK and the Politics of
Emotion
Haylee Kelsall 11665424
Master’s Thesis
Graduate School of Communication
Master’s programme Communication Science
Supervisor: Dr. Bert Bakker
28
thJune 2019
Abstract
Some commentators have suggested that the recent successes of Brexit, Trump and
his far-right European populist peers can be linked to their persuasive, emotive rhetoric.
Brexit campaign sides were labelled ‘Project Fear’ and ‘Project Hate’, but was there any truth to these labels? This paper sets out to consider the use of emotional appeals by UK politicians
in their online communication. Whilst some studies have addressed the supply-side of
emotional appeals in the political context, online communications remain almost unaddressed
in the extant literature. Relying on a large-scale Tweet corpus (N = 971,571) and automated
text analysis methods, this thesis considers the influence of politicians’ Brexit position and party on their use of anger, fear and enthusiasm appeals overtime. Informed by at times
competing theories — the Affective Intelligence Theory and the Appraisal Tendency
Framework — the results highlight the urgent need for scholarship to pay closer attention to
the supply, rather than demand, side of emotional appeals in political campaigning.
Moreover, theoretical development needs to move beyond the borders of the US, and beyond candidate elections. In the case of the UK, politicians’ stance on Brexit held little sway over the emotional appeals present in their Tweets. Instead, the findings here suggest the strategic
Recent events in several Western democracies have shaken the political status-quo,
with some commentators attributing these developments to an increasingly pervasive role of
emotional rhetoric in the political sphere (Hewitt, 2017; Shrimsley, 2015; Foster, 2017;
Vasilopoulou & Wagner, 2017). During Brexit campaigning, Boris Johnson urged Daily
Telegraph readers to avoid being "taken in" by "the agents of Project Fear" (2016, para 6). He
was referring to the 'Remain' campaign, whilst his mayoral successor, Sadiq Kahn labelled Johnson and ‘Leave’ as having run "...Project Hate, as far as immigration is concerned" (as cited in Wright, 2016, para. 20). Emotionally charged political rhetoric is hardly unique to
Brexit; news-media and scholars alike often draw parallels with other actors – e.g. Trump and
his European far-right populist peers – and their frequent use of emotive strategies to
persuade the electorate (e.g. Cassidy, 2016; Gabbatt, 2016; Hameleers, Bos & de Vreese,
2016; Hewitt, 2017; Salmela & Von Scheve, 2017; Schrock et al., 2017). Brexits’ Leave
founder acknowledged this too: “Facts don’t work… you have to connect with people emotionally, it’s the Trump success” (Banks, as cited in Worley, 2016).
Where and when do actors use these emotional appeals? Jerit (2004) argued that “[we
must] begin with well-developed theories about the types of appeals that candidates are most likely to use,” (p.573). Following Jerit, scholars have considered (discrete) emotional rhetoric in campaign advertisements (Brader, 2006; Ridout & Searles, 2011), political speeches (e.g.
De Castella, McGarty & Musgrove, 2009; De Castella & McGarty, 2011; Erisen &
Villalobos, 2014), party manifestos (Crabtree, Golder, Gschwend & Indridason, 2015), press
releases (Breeze, 2018), or a combination (Kosmidis, Hoboldt, Molloy & Whitefield, 2019).
Aside from one recent exploration of use of emotional appeals in Italian party leaders’
Tweets (de Candia & Bellei, 2019), research has focussed solely on offline communication
emotional appeals by political actors beyond the United States borders, those which address
campaigning (Brader, 2006; Ridout & Searles, 2011) have not.
This thesis extends the recent work of de Candia & Bellei (2019) by investigating the
use of anger, fear and enthusiasm appeals by UK political actors in the Twittersphere,
dependent on their party and EU referendum position. Secondly, it extends work on the use
of emotional rhetoric during campaigns (e.g. Brader 2006; Ridout & Searles, 2011) through
comparatively considering dynamics of a unique electoral cycle, covering the UK’s Brexit
referendum, subsequent snap general election, and the ‘normal-business’ (i.e. non-campaign) periods after each vote. It seeks to answer the following research question:
To what extent do Labour and Conservative politicians' Brexit position and party affiliation relate to their use of anger, fear and enthusiasm appeals in Tweets, and what role does the electoral cycle play in these relationships?
This thesis employs dictionary based computational methods on a large-scale Tweet
corpus (N=971,571) to answer this question. Expectations are informed by two competing
theoretical frameworks: Affective Intelligence Theory [AIT] and the (cognitive) Appraisal
Tendency Framework, paving the way for the generation of new theoretical insights. Little
evidence was found to support the purported ‘Project Hate’ versus “Project Fear” campaigns of Brexit, rather the use of emotional appeals appears driven by party. Labour relied more
heavily on anger and fear except during the election campaign, where the Conservatives
ramped up their use of these emotions to match the opposition. Enthusiasm levels were
similar during Brexit campaigning, though in the periods which followed, a gap quickly
emerged with the Tories on top.
Taken together, this thesis provides initial evidence that UK parties are using
‘beat’, where journalists increasingly rely on the platform as a news source (see e.g. Broersma & Graham, 2012; Moon & Hadley, 2014). Coupled with the increase in Tweets
embedded within online news articles, this means that citizens need not follow a given
political actor, nor use Twitter, to be exposed to persuasive, emotional rhetoric in politicians’
Tweets. Considering these can influence opinions, attitudes and behaviours, both citizens and
journalists alike should approach politicians’ social-media messages with a critical eye. The findings offer new insight, derived in new contexts - in terms of geographical locale,
party-system, and mode of communication – of politicians’ emotive rhetorical strategies.
Theoretical Background
But what exactly is an emotion? Emotions are experienced in response to events and
involve appraisals of relevance, i.e. the event must hold relevance to the individual to elicit
an emotional response (Scherer, 2005). This paper seeks to detect expressions of emotion in
Tweets, under the presumption that such expressions intend to elicit a similar response
amongst the message receivers. Although the body of research concerning the supply-side of emotional appeals in politicians’ rhetoric remains limited in scope, some expectations can be derived then supplemented with insights provided by AIT and Appraisal Tendency
Framework.
AIT asserts that there are two cognitive states, or systems, which respond to stimuli.
Each discrete emotional experience is associated with one of these two systems - the
dispositional system and the surveillance system (Marcus, Neuman & MacKuen, 2000). The
former encourages reliance on predispositions, reinforcing existing behaviours and attitudes
(Brader, 2005; Ridout & Searles, 2011; Valentino et al., 2011). Anger and enthusiasm fall
under this dispositional system, with anger emerging when one feels their goal is blocked and
enthusiasm when it is (likely to be) achieved. The surveillance system however, operates in
heightens the attention paid to the situation, encouraging information-seeking and less
reliance on heuristic cues (Ridout & Searles, 2011). In this way, the “...AIT focuses on the
immediate, preconscious impact of emotional reactions,” (Valentino et al., 2011, p.147). Cognitive appraisal theories — such as the Appraisal Tendency Framework (Lerner &
Keltner, 2000, 2001) — instead focus on the responses to these emotions. Specifically, the
framework concerns “…the subsequent cognitive evaluations of these immediate [emotional] reactions,” (Valentino et al., 2011, p.147), thereby extending the AIT. Like the AIT, the Framework provides clear expectations of when anger as opposed to fear, should arise.
Relevant to political decision-making amongst the electorate, it sheds light on how these
emotions might respectively reduce or attenuate evaluations of risk. Individuals in a state of
anxiety – in response to fear – have more pessimistic risk-perceptions, whilst angry
individuals evaluate risky judgements in a more positive fashion (Huddy, Feldman, Taber &
Lahav, 2005; Lerner and Keltner, 2000, 2001). The mechanisms behind these responses will
be explored in the following sections.
In the present study, predictions are informed by both theories. The author could not
source any literature where the AIT, nor the Appraisal Tendency Framework had been
applied to the study of politicians’ online communications. Moreover, when applied to emotional appeals in the campaigning context, the two theories offer divergent predictions
concerning fear (see below). Testing both permits an opportunity to (in)validate their
application in a new context, testing their level of generalisability and therefore abstraction.
Politics of Anger
Anger as an emotional experience can encompass numerous feelings, “…from mild irritation or annoyance to fury and rage” (Spielburger, Jacobs, Russell & Crane, 1983, p.162).
responsibility) can be attributed to a perceived ‘guilty’ party (Groenendyk, Brader & Valentino, 2011; Lerner & Keltner, 2000, 2001). A party is perceived as ‘guilty’ as anger
emerges in response to perceptions of injustice, or when one feels that “...their goals are blocked… [or] feels slighted” (Carver & Harmon-Jones, as cited in Weeks, 2015, p.700). According to the AIT, anger will push reliance on (partisan) habits as an ‘aversive’ emotion (MacKuen, Wolak, Keele & Marcus, 2010). This heightens an individual’s desire to limit
cognitive dissonance; thereby ignoring information which conflicts with their position, or
alternately, seeking information to confirm their position (MacKuen et al., 2010), i.e. it
strengthens pre-dispositions (Brader, 2006; Vasilopoulou & Wagner, 2017). White (2012)
claims that when voters are angry, rather than being apathetic they will express ‘democratic
anger’ in the form of participation and mobilisation. Indeed, anger spurs participation
amongst the less politically active (Lamprianou & Ellinas, 2018) and angry citizens engage in
more ‘costly’ modes of participation than their peers, such as financial contributions,
campaign work and attending party events (Valentino et al., 2011). The Appraisal Tendency
Framework presumes angry individuals will evaluate risks in a more positive fashion, due to
the high certainty and individual control appraisals inherent to its elicitation (Lerner and
Keltner, 2000, 2001).
How might these observations inform expectations concerning politicians use of anger
appeals in the UK? Concerning Brexit campaigning, the Appraisal Tendency Framework and
to some extent the AIT suggest the Leave camp should employ anger to a greater extent than
Remain. Anger strengthens pre-dispositions whilst limiting information-seeking; Leave
needed to pique interest and mobilise supporters, whilst limiting their considerations of
Remains’ arguments. Secondly, concerning risk-perceptions of the Appraisal Tendency Framework, ‘Remain’ represented a safer, less risky choice as it would maintain the
(Vasilopoulou & Wagner, 2017). Therefore, the ATF also assumes Leave would employ
more anger to reduce risk-perceptions around the uncertainty of the EU departure.
H1a: Leave Tweets will feature more anger appeals than Tweets from Remain
Emotional appeals however, are not stable. Rather, these will vary depend on political
events (Marcus & MacKuen, 1993). Evidence from the Italian context demonstrated “…
power turnover and elections” impacted the emotional rhetoric present in Tweets of party leaders (de Candia & Bellei, 2019, para.12). Anger has been found to increase towards the
end of campaigning in the USA — presumably for mobilising purposes (Ridout & Searles,
2011). As this study considers only each campaign in comparison to ‘normal-business’
periods, rather than considering strategy within campaigns, it hypothesises that:
H1b: Anger appeals will be employed to a greater extent during campaign periods rather than non-campaign periods
H1c: The conditional effect of time is presumed stronger for Leave than Remain
How might party affiliations influence the level of anger in Tweets? In the UK
context, Labour — as the opposition — should presumably employ anger to a greater extent
than Conservatives, pushing the electorate towards change (Frijda, Kuipers, & ter Schure,
1989), whilst reducing risk-perceptions associated with upsetting the status-quo (Lerner &
anger to be heightened during the election campaign compared to the incumbent
Conservatives (Ridout & Searles, 2011).
H2a: Labour Tweets will feature more anger than Conservative Tweets
H2b: The conditional effect of time will be strongest for Labour during the general election campaign
Politics of Fear
Fear is a negatively-valanced emotion like anger, yet it emerges in response to danger
or threat, and when experienced, an individual is in a high-state of arousal (Brader, 2006).
Fear is experienced when situations are appraised as highly uncertain, where an individual
has diminished control, and blame attribution proves difficult (Lerner & Keltner, 2000, 2001;
Marcus et al., 2000; Smith & Ellsworth, 1985; Valentino et al., 2011, Weeks, 2015).
Experiencing fear in response to political stimuli often leads individuals to a state of political
anxiety in relation to that given event and/or message. Fear increases attention,
information-seeking and learning, however exposure to skewed or biased information could have
detrimental, rather than positive effects on knowledge, opinions and behaviours (Gadarian &
Albertson, 2014). In response to ‘threats’ posed by immigration, fear can affect attitudes and behaviour “...independent of actual information about the severity of the threat…” (Brader, Valentino & Suhay, 2008, p.963), whilst it also drives biased information-seeking, with a
goal of re-affirming the threat (Gadarian & Albertson, 2014). For this reason, fear can lead to
opinion and behavioural change in the direction of available information, thus leaving
citizens prone to manipulation (Brader, Valentino & Suhay, 2008). Additionally, due to their
persuasive appeals...” which propose solutions to the threat (Marcus, Sullivan, Theiss-Morse and Stevens, 2005, p.961). Unlike anger, fear lessens reliance on partisanship for processing
when exposed to misinformation, consequently reducing instances of partisan motivated
reasoning (Weeks, 2015). Also, under the Appraisal Tendency Framework, fear heightens
risk-perceptions (Lerner & Keltner, 2000, 2001, Huddy et al., 2005), especially when a
situation holds personal significance to an individual (Butler & Mathews, 1987). This
assumption is corroborated by empirical evidence: a policy for foreign military intervention
post 9/11 garnered less support amongst those whom were fearful following the attacks
(Huddy et al., 2005). In sum, fear heightens risk-perceptions, increases citizens’ absorption
of information and enhances political learning, yet it also paves the way for persuasion and
manipulation via biased and/or incorrect information.
Applying this to Brexit campaign groups, it would be reasonable to expect Remains’
‘Project Fear’ label to ring true. Put simply, by invoking fear amongst the electorate, the risk associated with leaving the EU would be heightened in the eyes of the public. Ultimately, it
would push citizens to favour remaining with the status-quo. As with anger, a conditional
effect of time could also be expected, whereby Remain should ramp up their use of fear during their campaign. As the ‘losers’ of the referendum, for Remain, the snap election offered another opportunity to woo voters to their position, and Brexit was a key issue; many
treated it as a rerun of the referendum (Mellon et al., 2018). Therefore, Remains’ fear should
also be higher during the election campaign.
H3a: Remain Tweets will feature fear to a greater extent than Leave Tweets
H3b: This difference will be more pronounced during the campaigns than the normal-business periods
Turning to the parties’ use of fear, earlier research, the AIT and the Appraisal Tendency Framework offer conflicting expectations. Based on the AIT, Labour should use
more fear in order to encourage voters to pay attention to new information and make them
more susceptible to persuasion (Brader et al., 2008; Gadarian & Albertson, 2014; Marcus et
al., 2005; Weeks, 2015) as the alternative to the Conservative government. Indeed, Ridout
and Searles (2011) found trailing candidates employed more fear appeals than incumbents in
campaign adverts at the individual politician level, whilst Brader (2006) found challengers
used more fear appeals than incumbents, both of which support the AIT predictions. These
assumptions however contrast to those provided by the Appraisal Tendency Framework.
According to this, the Conservatives — as long-standing incumbents — ought to use more
fear than Labour. They would strive to elicit fear to encourage voters to remain with what is
known and continue with the status-quo: their government (Lerner & Keltner, 2000, 2001;
Huddy et al., 2005). In line with this, a (limited) analysis of BBC’s readers’ targeted
Facebook adverts found Conservative adverts during the 2017 election negatively-valanced,
framing the prospect of a Labour led government as inherently risky (Bright, 2017). Due to
these conflicting perspectives, no clear expectations can be derived; if Labour make greater
use of fear appeals, this would support the AIT, whilst if it were the Conservatives, this
would support the Appraisal Tendency Framework. Similarly, whilst time will impose a
conditional effect during the election campaign between the parties, for whom this effect
would be stronger, and in which direction, proves difficult to hypothesise.
RQ1a: Which party’s Tweets rely most heavily on the use of fear appeals?
RQ1b: What role do the differing electoral periods have concerning the use of fear in the parties’ Tweets?
Politics of Enthusiasm
Although enthusiasm serves to encourage participation and mobilise much like the
other dispositional emotion — anger — it instead holds positive valence. Enthusiasm
emerges when goals are being met (rather than threatened, as with anger), or when the
prospect of certain goals being met appears likely (Brader, 2005, 2006; Marcus & MacKuen,
1993). Moreover, political actors who wish to persuade voters to solve issues via their
proposed solutions will employ enthusiasm appeals to appear confident the goal will be met
(Downs, 1972). In the UK context, enthusiasm would signal (certainty of) re-election or
winning the race. Alternatively, it may be used to drum up support for proposed solutions to
issues that — under other circumstances— would elicit angry responses amongst the
electorate.
Whilst enthusiasm increases interest in campaigns (Marcus & MacKuen, 1993;
Marcus et al., 2000; Valentino et al., 2010; Brader, 2005) it also strengthens citizen’s intent to cast a ballot (Brader, 2005). The heightened interest does not translate to information-seeking
however (Valentino et al., 2008), but it does stimulate campaign involvement – therefore
encouraging reliance on pre-dispositions, or in the political sense, partisanship. This is
expressed through both ‘cheap’ forms of participation (e.g. placement of a bumper sticker)
and more ‘costly’ forms, such as donations or working for a candidate (Valentino et al., 2010). In a nutshell, enthusiasm appeals serve to shore-up support, and encourage greater
involvement in political processes, from campaigning through actual mobilisation to the
polls. Is there evidence to suggest that enthusiasm is used in such ways by politicians though?
Surprisingly, although enthusiasm can mobilise citizens, neither of the studies which
considered its use during campaigning found it to be employed more at the end of the
campaign, where you would expect its mobilising effect to be most beneficial (Brader, 2006;
The Appraisal Tendency Framework offers little insight concerning enthusiasm, thus
based the discussion above, would Leave or Remain Tweets feature more enthusiasm?
Considering the timeframe of the dataset spans roughly three years and the perceived
likelihood of goal-attainability has fluctuated for both sides, it is difficult to arrive at a clear
expectation.
RQ2: Which Brexit campaign group - Leave or Remain – employs enthusiasm to a greater extent in their Tweets?
It becomes easier to make predictions considering the role of time. Since Leave
achieved their goal of winning the referendum, their Tweets in the period immediately
following ought to reflect this. Additionally, as previous empirical work surprisingly found
no increase at the end of campaigns to mobilise voters (Brader, 2006; Ridout & Searles,
2011), it is possible that enthusiasm only emerges once a goal has been met, rather than
anticipating that it will be met. Therefore,
H4: In the period immediately following the Brexit referendum, Leave will employ more enthusiasm in their Tweets than Remain.
Turning to the parties, Conservatives would on the one hand be presumed to include
more enthusiasm in their Tweets than Labour. They would want to shore-up partisan support
through reliance on pre-dispositions, and their incumbency and re-election in 2017 suggests
more enthusiasm as the AIT would suggest (Brader, 2006; Ridout & Searles, 2011). In the
period which immediately followed their re-election however, we would expect the Tories to
use more enthusiasm than Labour, having just won the elections.
RQ3: Which party — Conservative or Labour — feature enthusiasm to a greater extent in their Tweets?
H5: The Conservatives will use more enthusiasm in the period which immediately followed their general election win of 2017
Method
Data collection
Tweets were collected to test the assumptions outlined above. A list of MPs active on
Twitter, including their screen names, constituencies and party was obtained from
mpsontwitter.co.uk.1 All data collection, cleaning and analyses was done in R version 3.6,
using RStudio version 1.2.1335. Using rtweet (Kearney, 2019) - an R package to assist with the access and collection of data from Twitter’s streaming and REST APIs - Tweets were collected from members listed as active on mpsontwitter.co.uk on February 26th, 2019.
Limitations of the Twitter search API mean that a maximum of roughly 3,200 historical
Tweets can be collected per timeline/user (Twitter, 2019). In total, this dataset comprised
1,370,456 Tweets from 570 politicians. Due to the API limitations however, this does not
mean that Tweets were collected for each individual cover identical time-frames, as the
search API restrictions mean that for a more prolific Tweeter, their earliest Tweet would be
far more recent than a less prolific Tweeter. To verify that the accounts were correct, the
1
screen names were compared to social media contact lists obtained from theyworkforyou.com
and everypolitician.org.
Sample selection
Data was collected for all House of Commons representatives. Exclusion criteria were
developed, resulting in the exclusion 162 MPs and their 432,885 Tweets. Parties where
Tweets of less than 10 politicians had been collected were excluded, due to the small number
of observations, leaving the biggest parties - Conservatives, Labour, The Scottish National
Party [SNP] and Change UK in the dataset2. A decision was also made to exclude SNP. As a
regional party they cannot be deemed representative of the UK. Additionally, Change UK
was established 12 days prior to data collection (18th February 2019), therefore Tweets from
these MP’s were matched to their previous party, and those published whilst representing Change UK were removed from the dataset. Finally, the Liberal Democrats were excluded as
their 14,495 Tweets would have comprised only 1.6% of the remaining corpus. Such a low
comparative group sample could compromise the reliability of test results due to power
concerns (e.g. Type II errors. Table 1 presents an overview of the final sample composition.
Additional member information was collected from theyworkforyou.com and
everypolitician.org. A list of known EU referendum stances was obtained from the former,
whilst Every Politician provided demographic information such as gender, date of birth,
consecutive office terms and social media links. After stripping titles, removing initials and
converting all to lower-case, demographic information was matched to politicians’ Tweets,
using their name, party constituency and Twitter handles to validate the initial list used.
Tweets from politicians whom their EU referendum stance was unknown or undeclared were
2
eventually excluded. The decision was made to retain retweets, as rtweet returns the text field
of the retweet if available, otherwise the original tweets text. Whilst a retweet where a
politician provides no comment may not be their own words, their retweets are increasingly
viewed as endorsements. Recent examples include criticism levelled at Trump for retweeting
Britain First (BBC News, 2017a; Weaver, Booth & Jacobs, 2017), Andrew Boles retweeting
an image of Tommy Robinson as a ‘hero’ (Kahn, 2019; Norris & Bartlett, 2019), and Jacob Rees-Mogg retweeting content from Germany’s far-right AfD (BBC News, 2019). The final
dataset comprised 937,571 Tweets in total (Table 1).
Sample Characteristics
As Table 1 shows, the final Tweet corpus was evenly divided in terms of party, with
52.7% of the Tweets being from Labour politicians and 47.3% from Conservative politicians.
Although Remain MPs appear overrepresented (80.1%) compared to Leave MPs (19.1%),
this reflects MP’s actual positions, where 73.8% were Remain, 24.3% Leave and 1.8% Undeclared (calculated based on data obtained from theyworkforyou.com). Turning to
gender, 38.3% of Tweets were from women. Following the 2015 general election, during the
parliamentary term that covered the Brexit referendum, the House of Commons comprised
roughly 29% female MPs (BBC, 2015), rising to 32% following the 2017 election (BBC,
2017b), suggesting women are slightly overrepresented in the Tweet corpus compared with
the House of Commons. The average age of MPs in the corpus is 52.85 years, a figure
similar to the average age of 50 years for representatives elected in the previous two general
elections (UK Parliament, 2019). The average length of service for an MP elected in 2015
was 8.7 years (UK Parliament, 2019), whilst within the corpus this was higher, at 10.8 years.
Turning to time, the majority of Tweets followed the 2017 snap elections. The reason is
twofold: it is the period with the longest duration (9th June 2017 – 26th February 2019), and
Table 1
Sample composition and descriptive statistics
M SD Min Max % N Anger Standardised 0.00 1.00 -0.44 29.04 937,571 Unstandardised 1.48 3.39 0 100 937,571 Fear Standardised 0.00 1.00 -0.49 25.91 937,571 Unstandardised 1.87 3.79 0 100 937,571 Enthusiasm Standardised 0.00 1.00 -0.91 12.87 937,571 Unstandardised 6.63 7.26 0 100 937,571
Consecutive years in office 10.80 6.40 3 22 937,571
Age 52.85 9.69 31 82 937,571 Party Labour 52.7 494,015 Conservative 47.3 443,556 Leave 19.9 186,560 Referendum Stance Remain 80.1 751,011 Referendum Campaign 5.7 53,246 Post-Referendum 13.5 126,220 Time Election Campaign 4.0 37,594 Post-Election 76.8 720,511 Gender Female 38.3 359,535 Male 61.7 578,036
Brexit referendum account for 13.5% of the corpus, again likely due to the longer timeframe
— 23rd February 2016 through 17th April 2017. Tweets during the Brexit campaign (20th
February – 22nd June 2016) account for 5.7% of the corpus, whilst the general election
campaign (18th April - 7th June 2017) accounts for 4%. Both campaign periods had much
overall corpus. Figure 1 shows division of referendum positions amongst the parties, and the
division of party across the two positions. This shows 40.16% Conservative Tweets came
from Leave, compared to only 1.69% of Labour Tweets. This is unsurprising, considering in
total there were only 10 Labour MPs who declared their support for Leave (BBC, 2016) and
of whom only 5 have verified Twitter accounts and were still in the House of Commons at
the time of data collection.3
Sentiment/Emotion Analysis
Sentiment analysis is a broad term applied to a variety of text-based analytical
endeavours. These analyses typically attempt to identify “…opinions… evaluations, attitudes, and[/or] emotions” (Liu, 2015, p. 1) within text. Here the goal is to identify and extract scores for individual Tweets based on three specific emotions: anger, fear and enthusiasm. Using a
dictionary-based approach, with the support of the Quanteda package (Benoit et al., 2018) in
R and the NRC Emotion Lexicon (Mohammad & Turney, 2013) scores for each individual
Tweet were derived. A dictionary-based approach essentially treats each text-document as a ‘bag of words’, and subsequently matches individual words (tokens) with words listed in the dictionary. In the case of the NRC Emotion Lexicon, word-lists exist for the eight basic
emotions of Plutchik (1962) along with lists to identify positive and negative sentiment, a
total of 14,182 unigrams. Whilst dictionary methods have frequently been applied to political
texts and is a method commonly used in social science research, the approach has some
3
Although an interaction between party and referendum stance would prove interesting, unequal group sizes within the sample (and population in general) would render the effects of each variable near impossible to distinguish. Disentangling the effects would be problematic due to Leave being comprised primarily by Conservative Tweets, and Labour primarily Remain Tweets.
limitations. Negations fail to be accounted for – consider the difference between the two: “I
hate Brexit” versus “I don’t hate Brexit”. This means of analysis would score the two
sentences the same in terms of anger both recognising the word hate, and the latter failing to
pick up the negation ‘don’t’ (for a more detailed discussion see Grimmer & Stewart, 2013).
Text data — and particularly Tweets — are noisy and require pre-processing steps
before analysis. Punctuation, URLs, hashtags, usernames, rt (preceding a retweet), and
incorrectly parsed HTML elements (e.g. &) were removed first. Next, stop-words —
smaller common words with little underlying meaning on their own such as the, and, etc. —
were removed, prior to text being tokenised at the word level. These tokens were then
converted to lowercase, to allow matching with the unigrams of the NRC Emotion Lexicon
(N = 14,182), embedded within the quanteda.dictionaries R package (Benoit & Müller, 2019).
Operationalisation
Measurement of the dependent variables – anger, fear and enthusiasm – relied on the
NRC Emotion Lexicon. The unigram-lists for “anger” and “fear” were used to measure their
respective emotions, whilst the positive sentiment list was used to measure enthusiasm. Figure 1. How Tweets are distributed across the two key independent variables. Proportion of Leave and Remain Tweets within party, and proportion of Conservative and Labour Tweets within EU Referendum stance.
Although positive sentiment is not a direct measure of enthusiasm, it is logical to assume that
in the Tweets of political actors, positive sentiment does represent enthusiasm — when the
goals of an actor are being met (or are likely to be met), their Tweets should be high in
positivity, and consequently, high in enthusiasm. In this regard, higher positivity is an
indirect measure of the presence of enthusiasm.
The dictionaries were applied to the corpus with a count score for each emotion being
compiled per Tweet, whereby the number of tokens (words in this case) within one document
(the Tweet text) matched per emotion were tallied, i.e. if three words from the ‘anger’ list matched 3 words within the text, the Tweet would score a 3 for anger. Subsequently, these
scores were converted to relative scores dependent on the length of the Tweet, i.e. if a Tweet
was 25 words in length, with a raw anger score of five, the relative score was 20, reflecting
20%. These relative scores ran from 0-100. Finally, these relative scores were standardised
using z-transformations to permit easier comparisons across groups. The overall sample
means for each emotion are presented in Table 1 in both the original 0-100 scale and the
standardised version. Figure 2 shows the differences between group means per emotion,
using standard deviations.
Information for individual politicians’ characteristics collected from They Work for
you and Every Politician provided demographic co-variates (age, gender, years of service) as
well as the independent variables of interest specific to an MP; their party and EU
referendum position. Age was calculated by subtracting their year of birth from 2019,
representing the age they will become in 2019, e.g. an MP born in 1967 would score 52.
Gender was scored 0 if a politician was male and 1 if a politician were female. Years of
service was calculated based on the number of consecutive terms a politician had been in
office. In most cases, where an MP had assumed office through a general election, their score
assumed office during a parliamentary term — e.g. if a previous elected member did not fulfil
their term due to resignation, passing away etc. — the score was calculated by subtracting the
year which they joined parliament from 2019. For example, an MP elected for the first time in 2015 would score ‘4’, whilst one elected consecutively since 2001 would score ‘18’. Concerning the (demographic) independent variables of interest, Tweets of politicians whom
declared themselves Leave, prior to the Brexit referendum, were scored ‘0’ whilst Remain were scored ‘1’. For the party dummy, Tweets of Conservative politicians were scored ‘0’, and those of Labour scored ‘1’.
Although Tweets have the date they were posted attached, establishing how best to
operationalise time-periods throughout the corpus required some consideration. To address
Figure 2. Party and referendum position group means standard deviation difference from overall mean,
this, exploratory analyses were conducted whereby each emotion was plotted over-time to aid
in identifying trends. From these plots it became clear that including time as a trend (i.e.
continuous variable) made little sense and that instead it would need to be included as
dummies for different periods. In their analysis of campaign adverts, Ridout and Searles
(2011) used a similar method, whereby they created time dummies for each campaign month
in their models. In the present study, based on information garnered from the initial
exploratory time-effect plots, four distinct time periods were identified: the Brexit
referendum campaign (20th February – 22nd June, 2016), the post-referendum ‘normal business’ period (23rd February 2016 - 17th April 2017), the general election campaign (18th
April-7th June, 2017), and finally the post-election ‘normal business’ period (9th June 2017 –
26th February 2019). These were treated as dummies, so if a Tweet had been posted during the Brexit referendum, it would score a ‘1’ for this period and a ‘0’ in all others. Further details of the initial exploratory work concerning time can be found Appendix A.
Modelling
To test the hypotheses, OLS models were used. Linear regression has a number of
assumptions, which if not met, can bias the estimations. In this case, some important
assumptions were violated, such as heteroskedasticity and non-normality of residuals.
However, after conducting further diagnostic checks (such as using binary logistic regression
instead of OLS and estimating OLS models with robust standard errors included) and seeing
little change in the directional effects and/or significance, it is assumed that these have not
influenced the results (see Appendix B for an overview).
Models were built in the same way for each emotion. First, a main effects model was
estimated. The dependent variable in each case was emotion, with party, EU referendum
Additional interactive models were created, one where EU referendum stance was
interreacted with each of the time dummies, and another with party interacted with the time
dummies. For formal tests between groups of the interactive models, post-hoc Tukey’s tests
were used. One important point to note is that the decision was made – due to the large
sample size – to set the level of significance at .005, rather than the traditional .05. With such
a large sample, it is easy to achieve significance at .05, therefore to reduce Type II errors, it
was set at .005, a level which Benjamin et al. (2018) present a strong case for.
Results
Angry campaigns?
The results of the nine OLS regression models are presented below in Table 2. As
H1a expected, the Tweets of Leave politicians were significantly angrier than the Tweets of
their Remain counterparts (Model 1). On average, Remain Tweets featured -0.02 standard
deviations less anger than Leave Tweets (SE = 0.003, p<.001), when holding other variables
in the model constant. Support was also found for H1b as shown by the time-period
co-efficients of Model 1. Anger featured more heavily in Tweets during the two explicit
campaigning periods than the non-campaign periods. The average election campaign Tweet
was 0.03 standard deviations (SE = 0.01, p<.001) angrier than one posted during the
referendum campaign (the reference category), whilst the periods which followed the
referendum and general election featured significantly less anger; -0.04 (SE = 0.01, p <.001)
and -0.02 (SE = 0.005, p<.001) standard deviations respectively.
But was this time-dependent relationship stronger for Leave (H1c)? The first panel of
Figure 1, Panel A plots the predicted average anger for Leave and Remain contingent on
time, derived from Model 2. The overlapping confidence intervals of each time-point
H1c. Post-hoc Tukey’s pairwise comparisons confirmed this. There was no significant
difference between Leave and Remain during the referendum campaign (p = .99), nor in the
election campaign (p =.13). The point estimates however do show an increase in anger for
both groups at the election time, and this was confirmed in hoc tests. Full details of
post-hoc tests, including t-statistic and SE can be found in Appendix C for all emotions.
Regardless, the conditional effect of time was not stronger for either campaign group, in the
referendum campaign nor the election campaign therefore H1c cannot be confirmed4.
Turning to the second set of anger hypotheses, it was assumed that Labour Tweets
would feature more anger than those of their Conservative peers (H2a). The result of the
main effects model (1) confirms this assumption: the average Labour Tweet had 0.13
standard deviations (SE = 0.002, p<.001) more anger than a Tweet from the Tories, when
keeping all other variables constant. Panel B of Figure 1 plots the point estimates garnered
from Model 3, concerning the campaign-time effect outlined above, which assumed it would
be stronger in Labour, rather than Conservative Tweets (H2b). Contrary to this expectation
however, Labour were consistent in their use of anger regardless of the time-period —— the
point estimates show little variation and the confidence intervals clearly overlap. Post-hoc Tukey’s tests further confirm this (Appendix C), indicating no conditional effect of time for Labour and therefore no evidence found in favour of H2b.
4
One interesting difference worth noting is that Leave Tweets during the referendum campaign contained -0.05 standard deviations less anger than their election campaign Tweets (p<.001), whilst for Leave, no difference was found between these periods (p=.06). However, this may be due to the lower number of actors for whom Tweets could be collected during the referendum in comparison to the election campaign. Indeed, this may also provide some explanation why no difference was found in Leave Tweets in the pre- and
post-referendum periods, as moving from this time into the election through to the post-election period, time did appear to moderate their use of anger.
A different pattern emerges with the Tories, indicating a more strategic use of anger.
Their Tweets during Brexit campaigning were angrier than for the following period (b =
-0.06, SE = 0.01, p < .001), and the election campaign was angrier again (b = 0.14, SE = 0.01,
p <.001), reaching the same level of anger that Labour consistently maintained (p=.91),
before reducing use in the post-election period (b = -0.12, SE = 0.01, p <.001). Unlike Labour
Tweets, which were angrier on average and showed no variation, the Conservatives’ relative
anger peaked during campaigns and tapered off during the non-campaign or, ‘normal
business’ periods. This conditional effect of time suggests that the parties’ social media strategies are very different with regards emotive rhetoric, this will be returned to in the
discussion.
From an angry party to fearful politics
Remain were expected to rely on fear more than Leave (H3a). However, Model 1
shows Remain Tweets featured less fear (b=-0.01, SE = 0.003, p<.001). However, the effect
size is marginal at best, and with such a large sample should be interpreted with some
caution. Perhaps the difference between the two camps was contingent on time as H3b
predicted. Turning to the predicted fear of each group over time (Figure 1, Panel C), much
like with anger, there is little difference in their use of language linked to fear in their Tweets,
at any point in time. What is particularly striking is the near identical point estimates for the
referendum campaign and post-referendum period – neither camp appear to have adjusted
their use of fear. Indeed, the only instance where the level of fear present changed was for
Remain following the general election, where fear increased by 0.03 standard deviations
(p<.001). No evidence was found that Remain would ramp up their use of fear during either
Expectations concerning parties’ use of fear were conflicting. An expectation
informed by AIT would see Labour use more fear, whilst the Appraisal Tendency Framework
would expect the Tories to do so. The main effects model (4) lends support to the AIT as
Labour used 0.15 standard deviations more fear in their Tweets than the Conservatives (SE =
0.002, p<.001) on average.
Time influenced the parties’ use of fear in contrasting ways. As the graph of party predicted means (Panel D) shows, during the election campaign, the Tories increased the
level of fear in their Tweets, whilst the opposite was true for Labour. Likewise, Labour
Tweets during non-campaign periods would increase in fear, whilst the Tories toned down
these appeals outside of explicit campaigning. Tukey post-hoc tests found that after the
referendum, neither the Conservatives (p=.04) nor Labour (p=.77) reduced their fear appeals
compared to the campaign period, although a different pattern emerged moving into the
election campaign. Here the Conservatives increased their fear by 0.08 standard deviations
(p<.001), whilst Labour reduced their fear by -0.03 standard deviations (p=.003). These
alternate directional effects mean that during the election campaign, the parties’ Tweets had the same level of fear (p=.63), whereas at all other times, Labour relied more heavily on these
appeals. Although the use of fear was conditional on time, the (opposing) directional effects
indicate the parties differed in their strategic use of the emotion. During the general election
campaign, Labour reduced their appeals to fear whilst the Conservatives increased theirs.
Enthusiastic victories and enthusiastic incumbents?
Enthusiasm, much like anger, can shore up (partisan) support and get citizens to the
polling booth (Brader 2005, Valentino et al., 2010). Providing an answer to RQ2, in terms of
Tweets, the main effects model (7) shows that Remain were more enthusiastic than Leave (b
Table 2
OLS Regression models of relative emotion per Tweet, with conditional effects of time depending on EU referendum stance and party
Anger Fear Enthusiasm
Main
Effects Leave*Time Party*Time Main Effects Leave*Tim e Party*Time Main Effects Leave*Tim e Party*Time 1 2 3 4 5 6 7 8 9 Remain -0.02*** 0.01 -0.02*** -0.01*** -0.01 -0.01*** 0.04*** 0.05*** 0.04*** (0.003) (0.01) (0.003) (0.003) (0.01) (0.003) (0.003) (0.01) (0.003) Post-Referendum -0.04*** -0.02 -0.06*** -0.005 -0.01 -0.01 0.04*** 0.06*** 0.06*** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Election Campaign 0.03*** 0.06*** 0.08*** 0.02* 0.04 0.07*** 0.0002 0.04** 0.03*** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Post-Election -0.02*** -0.002 -0.03*** 0.04*** 0.04*** 0.02*** 0.02*** 0.03*** 0.04*** (0.005) (0.01) (0.01) (0.005) (0.01) (0.01) (0.005) (0.01) (0.01) Labour 0.13*** 0.13*** 0.10*** 0.15*** 0.15*** 0.11*** -0.09*** -0.09*** -0.04*** (0.002) (0.002) (0.01) (0.002) (0.002) (0.01) (0.002) (0.002) (0.01) Female 0.001 0.001 0.001 0.004 0.004 0.004 -0.001 -0.001 -0.001 (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Years in Parliament 0.001*** 0.001*** 0.001*** 0.0005 0.0005 0.0004 0.0004 0.0004 0.0004 (0.0002) (0.0002) (0.0002) (0.0002 ) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) Age 0.0004* 0.0004* 0.0004** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** (0.0001) (0.0001) (0.0001) (0.0001 ) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Remain*Post Referendum -0.03 0.004 -0.01 (0.01) (0.01) (0.01) Remain*Election Campaign -0.04 -0.02 -0.06 *** (0.02) (0.02) (0.02) Remain*Post-Election -0.03 0.002 -0.02 (0.01) (0.01) (0.01) Labour*Post Referendum 0.03 ** 0.03 -0.04*** (0.01) (0.01) (0.01) Labour*Election Campaign -0.09 *** -0.09*** -0.08*** (0.01) (0.01) (0.01) Labour*Post-Election 0.03*** 0.05*** -0.05*** (0.01) (0.01) (0.01) Constant -0.07*** -0.09*** -0.06*** -0.16*** -0.16*** -0.14*** -0.04*** -0.05*** -0.06*** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) R2 0.004 0.004 0.005 0.01 0.01 0.01 0.002 0.002 0.002 F Statistic 523.52*** (df = 8; 937562) 381.48*** (df = 11; 937559) 394.66*** (df = 11; 937559) 713.01* ** (df = 8; 937562 ) 518.89*** (df = 11; 937559) 537.77*** (df = 11; 937559) 189.67** * (df = 8; 937562) 139.03*** (df = 11; 937559) 141.52*** (df = 11; 937559) Note:
*p < .01, **p < .005, ***p < .001. Reference categories: Leave, Pre-referendum, Tories, Men, Leave*Referendum Campaign, Labour*Referendum Campaign. Dependent variables were scored as relative emotion words per Tweet and subsequently standardised via z-transformation. For all models, N = 937,571.
Figure 3. The conditional role of time for each emotion, by referendum stance and party. Predicted means for each time period, by EU referendum stance (left) and party (right). Predicted means are based on models from Table 2 which include the interaction terms. Confidence intervals are set at 0.995.
Were Leave at least more enthusiastic in the period immediately following the
referendum, considering they had achieved their goal (H4a)? The point estimates plot of
Figure 2, panel E shows that their enthusiasm did increase following the referendum, and a
post-hoc Tukey’s test confirmed that this was an increase of 0.06 standard deviations from
their referendum campaigning (p<.001). Indeed, their enthusiasm remained steady in the
election campaign (p = .98), and the period after the election (p=.006). This was quite a
different pattern to Remain, where the level of enthusiasm in Tweets fluctuated between each
respective campaign and non-campaign period – surprisingly increasing after their campaign
losses. Regardless, in the post-referendum period it was Remain who were most enthusiastic;
post-hoc Tukey’s tests showed their Tweets on average included 0.04 standard deviations
more enthusiasm than Leave Tweets (p<.001). Although Leave did increase their enthusiasm
following the referendum, H4a must ultimately be rejected, as it was not Leave, but rather
Remain, who’s Tweets were most enthusiastic in the post-referendum period.
Turning to the parties’ use of enthusiasm, RQ3 asked which party would rely more on enthusiasm. Evidence provided by the main effects model confirms that Conservative Tweets
were on average 0.09 standard deviations more enthusiastic than Labour Tweets (SE = 0.002,
p<.001). The final hypothesis expected the Tories use more enthusiasm following their
victory at the general election (H5), yet this was not the case. Panel F of Figure 1 suggests
that their enthusiasm was at its height following the Brexit referendum, rather than the
general election as expected. Post-hoc Tukey’s tests confirmed that Conservatives enthusiasm
was 0.06 standard deviations greater in the post-referendum rather than the campaign itself
(p<.001) and 0.02 standard deviations greater than the post-election period (p<.001), whilst
there was no difference between the election campaign and the time thereafter (p=0.02). Although the Conservative’s did not increase their enthusiasm after winning the general election, their Tweets in the post-election period were more enthusiastic than Labour’s (b =
0.08, p<.001), which supports the (party) ‘winners’ enthusiasm hypothesis — H5, as the Conservative’s use of the emotion was stronger than Labour’s.
Discussion
Brexit: Project “identical?”
Brexit’s Leave and Remain campaigns were (purportedly) built on fear and anger respectively. The Leave campaign slogan — ‘Take back control’ — attributed blame to the
EU (high certainty) for taking the power away (an injustice) from citizens and national
representatives, appealing to anger (Groenendyk et al., 2011; Lerner & Keltner, 2000; Weeks,
2015). The name of the official Remain campaign -- “Britain Stronger in Europe” -- implied
uncertainty and danger should the UK leave, a clear appeal to fear (Brader, 2006). Yet, the
results of the present study suggest a politicians’ party, rather than Brexit position, holds more sway over the emotional appeals present in their Tweets. Given the media-hype and
accusations levelled by campaign-group members of ‘Project Hate’ versus ‘Project Fear’, it’s
surprising that little evidence was found to support these claims. Before accounting for the
role of time, there were some differences observed between the two camps — i.e. Remain
were more enthusiastic and relied less on anger and fear — however these differences were
marginal at best, and even with the level of significance set at p<.005, should be interpreted
with caution due to the large sample size. After addressing the use of each emotion overtime,
strikingly similar patterns of use emerge between the two camps, and taken together, these
findings suggest little difference between groups in the use of any of the emotions concerned.
In the Twittersphere at least, the differences between camps is not evident, and whilst it may
be that emotions were employed differently on other campaign- fronts, further research
What might explain these apparent (lack-of) findings? Not all politicians whose
Brexit position was known were included. Several significant Brexit campaigners were
omitted, and their inclusion could have altered the results5. Moreover, theoretical
development has focused on the American context (Marcus et al., 2000; Lerner & Keltner,
2000; 2001), and the campaigns of representatives (Brader, 2006, Ridout & Searles, 2011),
neither of which really reflect a referendum in the UK. If anything, the unexpected
similarities highlight the need for further theoretical development concerning the supply-side
of emotional appeals — beyond both the US border and candidate elections.
It may simply be that referendum stance just does not matter, and that neither side
uses a distinct emotional campaigning strategy on Twitter. This seems plausible, as the
parties, unlike the Brexit campaign groups, do have remarkable differences in their use of
emotional appeals. Labour relied far more heavily on the negatively-valanced emotions (fear,
anger) than the Tories, with the only exception being during the general election, where the
level of these emotions present within a Tweet was equal between parties. The Conservatives
employed enthusiasm more than Labour, and although initially (during the Brexit campaign)
the parties used it to a similar extent, a significant gap emerged between the two immediately
afterwards, and continued thereafter.
Labour used more anger and fear in general, in line with predictions based on the AIT
(Brader, 2005, 2006; Marcus et al., 2000; Ridout & Searles, 2011). However, the
Conservatives spike in use of these appeals in the general election campaign is of particular
interest. Their sudden increase in anger is unexpected, as it should encourage voters to desire
5For example, a number of leading Remain MPs are no longer in the House of
Commons (e.g. David Cameron and George Osborne), or were excluded due to their regional party membership (e.g. Nicola Sturgeon), whilst prominent Leavers included MEPs, rather than MPs (e.g. Nigel Farage and Daniel Hannan).
change (Frijda et al., 1989) and become less risk-averse (Huddy et al., 2005; Lerner &
Keltner 2000, 2001) with the goal of upsetting the status quo – i.e. successive Tory
governments. However, increasing fear to heighten perceptions of risk and therefore
encourage voters to stick with the status-quo would be expected of the incumbent Tories, by
the Appraisal Tendency Framework (Huddy et al., 2005; Lerner & Keltner 2000, 2001). It is
worth noting that this sudden increase only applies to the negative emotions – their
enthusiasm after the Brexit vote was fairly stable – which suggests it is not a general increase
in emotional rhetoric. Instead, this is suggestive of negative campaigning tactics, which in its
broadest sense, implies political actors ‘attacking’ an opponent during a campaign (Geer,
2006). If indeed this were the case, an explanation for the Tories sudden increase in
negatively-valanced emotions and, with the exception of Brexit campaigning, relatively
stable enthusiasm, may lie in their concern of generating voter-backlash (Garramone, 1984;
Ridout & Searles, 2011). Whilst the Tories may recognise potential value in anger and fear
appeals, they may also harbour concerns that their over-use could damage their appeal to
supporters. Therefore, during non-campaign periods, they rely on enthusiasm to shore-up
support (Marcus et al., 2000; Brader, 2005), and reserve appeals to anger and anxiety for the
election campaign, to minimise potential backlash – i.e. their supporters may turn against
them due to perceptions of over-negativity (Ridout & Searles, 2011). Indeed, this could be an
interesting avenue for future research to explore.
Whilst a reasonable explanation, why would Labour not follow a similar pattern, and
instead keep their use of fear and anger fairly constant? Some scholars have suggested that
social media are the new site of the permanent campaign (e.g. see Elmer, Langlois, &
McKelvey, 2012), and indeed Labour’s consistency versus Conservatives’ spike is suggestive
of a party-difference in the utilisation of Twitter as a campaigning tool. If both parties were
campaign space. However, its only Labour that use the negative emotions to a similar degree
overtime. The Conservatives increase in the election period could suggest that they view
Twitter differently: it functions as a true-campaigning space only during periods where their
governing status is at stake, i.e. during the election campaign. If offline Labours’ use of these
negative emotions fluctuated more than the near-consistent Twitter patterns observed here,
and the Conservatives followed a similar pattern offline as in their Tweets, this would
indicate a clear difference in perception. Namely, that the Tories treat Twitter as an explicit
campaign-period communication tool whilst Labour perceive it as a ‘permanent campaign’
platform. Regardless, the observations here concerning negative-valanced emotions ‘election
effect’ amongst the Tory Tweets, and consistent use amongst Labour Tweets, is certainly worthy of further investigation.
Affective Intelligence Theory and the Appraisal Tendency Framework
The findings of this study in some instances lend support to the AIT and in others the
Appraisal Tendency Framework. Generally speaking, Leave used more anger, confirming
expectations of both, yet their higher use of fear goes against each theory. Remains’ generally
higher enthusiasm would also go against the AIT’s predictions. Looking closer at the group differences overtime, neither theory finds support – the groups’ patterns of use are
near-identical. Turning to the parties, Labours generally higher use of anger and fear again
confirms the AIT, but only partially the Appraisal Tendency Framework: Conservatives
should have used more fear, as a more risk-averse electorate would wish to maintain the
status-quo. The AIT did find more support in the Tories’ greater use of enthusiasm, as
incumbents their electoral goals are being met. However, again looking over time, the sudden
spike within the Tories for both negative emotions is not explained by either concerning
anger, and the Framework only partially explains their increased fear. As both theories
supply-side concerning parties. Moreover, although the Appraisal Tendency Framework has been
validated for policy-support on specific issues, such as foreign military intervention (Huddy
et al., 2005), these mechanisms may operate differently in a scenario where a citizen should
act on their position by casting a ballot – either in a referendum or candidate election. These
mixed-results demonstrate that there is an urgent need for scholars to continue theoretical
development for the supply-side of emotional appeals; especially those online.
Limitations
There are a number of limitations to the conclusions drawn here. Firstly, the Twitter
API that permits Tweet collection has a limit of 3,200 Tweets per user. This means that the
more prolific Tweeters are not represented in all three periods, which may have biased the
results as the more content, the more likely emotional appeals will be used (Ridout & Searles,
2011). Further, bag-of-words analyses can be criticised as they fail to account for context. It
would be interesting to apply alternate means of analysis (e.g. supervised machine learning
using labelled data) to see if similar conclusions can be drawn. Indeed, another criticism may
be levelled at the use of positive sentiment to measure enthusiasm; perhaps a better-defined
measure would have captured volatility amongst the Tories similar to the negatively-valanced emotions. The development of a more precise ‘enthusiasm’ dictionary would be valuable for studying emotional campaign rhetoric, since enthusiasm is a key emotion in the literature.
The inclusion of assertions of confidence in conjunction with positive-sentiment (Downs,
1972) might more accurately capture enthusiasm, unfortunately exploratory measurement
development was outside the scope of this study.
Overall, the evidence here suggests that on Twitter, it is a politician’s party affiliation driving their use of emotional appeals. But why should we be interested in this? Although the
seeks information from which to base well informed decisions, this idealistic image is far from reality. The ideal suggests citizens’ decisions are based on logical reasoning alone, ignoring the fact affective and cognitive reasoning mechanisms operate in tandem (Redlawsk,
2002). Both are crucial in understanding message reception, processing, subsequent decisions
and/or actions taken as a result of these communications. As Twitter becomes a normalised
newsbeat for journalists, and Tweets are more frequently embedded verbatim within news
stories, actors have the opportunity to get an unedited message to the public. When these are
laden with persuasive emotional rhetoric, both journalists and citizens should carefully
consider the motivation behind such a message. There is certainly evidence that Twitter is a
tool for campaigning, and this study provides the first consideration of this in the UK context.
Ultimately, an electorate lacking the knowledge necessary to navigate the complexities of
political phenomena should be aware that political actors can and do exploit this. They use
this to their advantage by strategically employing emotive appeals to persuade voters and
influence their opinions, attitudes and behaviours.
References
BBC News. (2015, May 8). Election 2015: Number of women in Parliament rises by a third.
BBC News. Retrieved from https://www.bbc.com/news/uk-politics-32601280
BBC News. (2016, June 22). EU vote: Where the cabinet and other MPs stand. Retrieved
from https://www.bbc.com/news/uk-politics-eu-referendum-35616946
BBC News. (2017a, November 29). Donald Trump retweets far-right group’s anti-Muslim
videos. BBC News. Retrieved from
https://www.bbc.com/news/world-us-canada-42166663
BBC News. (2017b, June 10). Election 2017: Record number of female MPs. BBC News.
BBC News. (2019, April 1). Brexit: Jacob Rees-Mogg defends tweet of far-right AfD clip.
BBC News. Retrieved from https://www.bbc.com/news/world-europe-47770959
Benjamin, D. J., Berger, J. O., Johannesson, M., Nosek, B. A., Wagenmakers, E. J., Berk, R.,
... & Cesarini, D. (2018). Redefine statistical significance. Nature Human Behaviour,
2(1), 6. Retrieved from
https://www.nature.com/articles/s41562%20017%200189%20z
Benoit, K., Kohei W., Haiyan W., Nulty, P., Obeng, A., Müller, S., & Matsuo, A. (2018)
Quanteda: An R package for the quantitative analysis of textual data. Journal of Open
Source Software. 3(30), 774. https://doi.org/10.21105/joss.00774.
Benoit, K. & Müller, S. (2019). Quanteda.dictionaries: Dictionaries for Text Analysis and
Associated Utilities. R package version 0.22. Retrieved from
https://github.com/kbenoit/quanteda.dictionaries
Brader, T. (2005). Striking a responsive chord: How political ads motivate and persuade
voters by appealing to emotions. American Journal of Political Science, 49(2),
388-405.
Brader, T. (2006). Campaigning for hearts and minds: How emotional appeals in political
ads work. University of Chicago Press.
Brader, T., Valentino, N. A., & Suhay, E. (2008). What triggers public opposition to
immigration? Anxiety, group cues, and immigration threat. American Journal of
Political Science, 52(4), 959-978.
Breeze, R. (2019). Emotion in politics: Affective-discursive practices in UKIP and Labour.
Bright, S. (2017, May 31). The rise of the Tory attack ads on Facebook [Blog Post].
Retrieved from https://www.bbc.com/news/blogs-trending-40059846
Broersma, M., & Graham, T. (2012). Social media as beat: Tweets as a news source during
the 2010 British and Dutch elections. Journalism Practice, 6(3), 403-419.
Butler, G., & Mathews, A. (1987). Anticipatory anxiety and risk perception. Cognitive
Therapy and Research, 11(5), 551-565.
Cassidy, J. (2016, June 23). What Do the Brexit Movement and Donald Trump Have in
Common? The New Yorker. Retrieved from
https://www.newyorker.com/news/john-cassidy/what-do-the-brexit-movement-and-donald-trump-have-in-common
Crabtree, C., Golder, M., Gschwend, T., & Indridason, I. H. (2015). Campaign sentiment in
European party manifestos. Retrieved from
https://pdfs.semanticscholar.org/93d8/b89d65ecd0c7bd1d88b6b81b93a79d1b83cc.pdf
de Candia, M., & Bellei, C. (2019, May 14). What emotions do politicians express with their
Tweets? The case of Renzi, Berlusconi, Salvini and Di Maio [Blog post]. Retrieved
from
https://blogs.lse.ac.uk/europpblog/2019/05/14/what-emotions-do-politicians-express-with-their-tweets-the-case-of-renzi-berlusconi-salvini-and-di-maio/
de Castella, K., McGarty, C., & Musgrove, L. (2009). Fear appeals in political rhetoric about
terrorism: An analysis of speeches by Australian Prime Minister Howard. Political
Psychology, 30(1), 1-26.
de Castella, K., & McGarty, C. (2011). Two leaders, two wars: A psychological analysis of
fear and anger content in political rhetoric about terrorism. Analyses of Social Issues
Downs, A. (1972). Up and Down with Ecology: The Issue Attention Cycle. The Public
Interest 28, 38-50.
Erisen, C., & Villalobos, J. D. (2014). Exploring the invocation of emotion in presidential
speeches. Contemporary Politics, 20(4), 469-488.
Elmer, G., Langlois, G., & McKelvey, F. (2012). The permanent campaign: New media, New
Politics. New York: Peter Lang
Foster, R. (2017). Nasty, British and Short: an emotional election. In E. Thorsten, D. Jackson
& D. Lilleker (Eds.), UK Election Analysis 2017: Media, Voters and the Campaign
Early reflections from leading academics (p. 105). Poole: The Centre for the Study of
Journalism, Culture and Community. Retrieved from http://www.electionanalysis.uk/
Frijda, N. H., Kuipers, P., & ter Schure, E. (1989). Relations among emotion, appraisal, and
emotional action readiness. Journal of personality and social psychology, 57(2), 212.
Gabbatt, A. (2016, June 25). Trump and Brexit: parallel campaigns built on fear, anger and
charisma. The Guardian. Retrieved from
https://www.theguardian.com/us-news/2016/jun/25/donald-trump-nigel-farage-us-election-brexit
Gadarian, S. K., & Albertson, B. (2014). Anxiety, immigration, and the search for
information. Political Psychology, 35(2), 133-164.
Garramone, G. M. (1984). Voter responses to negative political ads. Journalism Quarterly,
61(2), 250-259.
Geer, J. G. (2008). In Defense of Negativity: Attack Ads in Presidential Campaigns. Chicago:
University of Chicago Press.
Grimmer, J., & Stewart, B. M. (2013). Text as data: The promise and pitfalls of automatic
