Are Campaigns Getting Uglier, and Who Is to Blame? Negativity, Dramatization and Populism on Facebook in the 2014 and 2019 EP Election Campaigns

ABSTRACT Relating to theories of dissonant public spheres and affective publics, we study negativity, dramatization, and populist content in political party Facebook posts across 12 countries during the 2014 and 2019 European Parliament Election campaigns. A quantitative content analysis of 14,293 posts from 111 (2014) and 116 (2019) political parties shows that negative emotion, negative campaigning, dramatization, and populist content has increased over this time. We show that political parties sought to evoke more negative emotions and generate more dramatization, engaged more in negative campaigning, and included more populist content in their Facebook posts in the 2019 EP election than in 2014. Further, we show that posts evoking negative emotions and dramatization and involving negative campaigning yield higher user engagement than other posts, while populist content also led to more user reactions in 2014, but not in 2019. Negative, exaggerated, and sensationalized messaging therefore makes sense from a strategic perspective, because the increased frequencies of likes, shares, and comments make parties’ messages travel farther and deeper in social networks, thereby reaching a wider audience. It seems that the rise in affective and dissonant communication has not emerged unintentionally, but is also a result of strategic campaigning.


Introduction
In the past decade, a wealth of studies has been published on how political parties and candidates campaign on social media platforms. Indeed, all political actors have adopted platforms like Twitter, Facebook, and Instagram as key outlets for their campaigns. They have explored different strategies and professionalized their online campaigns in a way that scholars argue it puts us in the midst of a fourth era of campaigning, incorporating the datadriven strategies that social media platforms afford (Roemmele & Gibson, 2020). It has become clear that social media do not level out the playing field for marginal actors with few dramatization, and populist messages. These phenomena are explored empirically through a quantitative content analysis of Facebook posts by 111 (2014) and 116 (2019) political parties across 12 countries and two consecutive election campaigns.

Dissonant Public Spheres, Affective Publics, and Elections
Against the backdrop of declining institutional trust and legitimacy, Bennett and Pfetsch (2018) argue that public spheres are becoming increasingly disconnected from traditional structures of media and journalism and marked by disruptive communication processes. In a similar vein, Bimber and Gil de Zúñiga (2020) theorize about how technological affordances of social media enable the emergence of "unedited" public spheres, with obscure information sources, deceptive communication, and manipulated social signals. Pfetsch (2018) and  have developed this argument along the lines of "dissonant public spheres." They argue that public sphere theory and related empirical studies need to depart from the "wishful thinking" that public communication builds bridges across discursive contestation, that political actors seek consensus, deliberation, and reasoned decision-making, and that forming consonance is still a dominant drive within public spheres (Pfetsch, 2018, p. 60). The argument is not purely normative, as many theoretical concepts in political communication are constructed on the inherent assumption of consonance, e.g., agenda-setting requires congruence about important topics among elite actors, or indexing assumes that media actors agree on power relations among political organizations, at least to some degree.
The idea of dissonant public spheres differs from both agonistic conceptions of the public sphere and the notion of counter-publics. Agonistic public spheres (Mouffe, 2000) depart from the idea that all conflict can or should be resolved through rational debate by acknowledging that much public communication consists of and maintains contestation, fierce opposition, rivalry, and hostility. Yet, it still rests on the normative ideal that political opponents can accept and respect their differences without resolving them, i.e., they can live and let live. Dissonant public spheres, in contrast, are marked by a cacophony of voices and an "inability to communicate across difference" (Waisbord, 2016, p. 2). In this view, political actors are unable and/or unwilling to engage in rational discourse, to foster understanding, to exchange arguments, or even to share a common perception of reality: "dissonant public spheres (and even more so during campaigns) abstain from rational arguments or references to others. They aspire neither to understand, nor deliberate or seek consensus" (Pfetsch, 2018, p. 61). Dissonant public spheres are also different from counterpublics that are built on dissent and protest challenging hegemonic positions. While counter-publics are somehow dissonant, they do not feature the disconnection, disinterest, and refusal to discursively engage that characterize dissonant public spheres.
We argue here that dissonant public spheres do not only emerge unintentionally (e.g., through polyvocality, information abundance) but can result from strategic campaigning using negative, emotional, dramatic appeals rather than rational arguments, and through populist appeals that attack or exclude. This connects to the concept of affective publics (Lünenborg, 2019;Papacharissi, 2016), which criticizes the classic Habermasian model of public spheres based on deliberation and rational exchange as being blind to the emotional, passionate aspects of public discourse. Affective publics are "networked publics that are mobilized and connected, identified, and potentially disconnected through expressions of sentiment" (Papacharissi, 2016, p. 312). Affect is not an emotion but refers to the "intensity of emotion" (p. 318), often promoted by political storytelling narratives appealing to gut feelings, dramatization, and perhaps even perceptions of alternative realities. Social movements and election campaigns have always been affective and included emotional appeals (e.g., Derks et al., 2008;Russmann, 2018), however the increasing importance and immediacy of social media, user-generated content, mobilization through personal networks, and algorithmic curation further catalyze this development, e.g., outrage as a driver for hashtag activism.
From the view of dissonant public spheres and affective publics, research on political communication should seek to understand how political actors produce affective communication, polyvocality, discordance, and cacophony and how campaigns, especially on social media, use divisive emotional content to drive polarization and fragmentation, distorting the ability to communicate across differences and burning social and institutional bridges. Based on the observation that political actors employ divisive, negative, and aggressive campaign strategies (e.g., Nai, 2018), our objective here is to analyze strategic party communications during two EP election campaigns in order to determine whether parties are drivers of dissonant public spheres.

Negative Emotions, Negative Campaigning, Dramatization, and Populist Content in Campaigns
The perception that election campaigns are getting uglier, more dissonant, and overly negative is not new and not unique to campaigning on social media. Quintus Tullius Cicero advised his older brother Marcus, who in 64 B.C. was campaigning for consul in the Roman republic, to always put on "a good show" for the masses, adding, "It also wouldn't hurt to remind them of what scoundrels your opponents are and to smear these men at every opportunity with the crimes, sexual scandals, and corruption they have brought on themselves" (Cicero, 2012, pp. 77-78). For centuries, campaigns have been run with passion, emotion, and arguments of varying quality. Negative campaigning with adversarial messages, dramatization, and populist appeals may not be detrimental to democracy, per se, but it does breed dissonant public spheres, in which political actors yell at each other rather than engaging in constructive dialogue, point fingers at presumed wrongdoings rather than exchanging rational arguments, and treat each other as enemies rather than as legitimate competitors with diverging views, albeit all on the common ground of democratic practice.
Emotions play a vital role in campaigns, as they are key in processing political information and thereby affect voters' attitude formation and voting decisions (Brader & Marcus, 2013;Crabtree et al., 2020;Nai & Maier, 2021;Weeks, 2015). They are evoked through the ways in which parties say something: how they frame their message. Fear appeals "are very common in modern electoral campaigns" (Nai, 2018, p. 225). Especially extremist parties use negative emotions like anger, fear, or anxiety in their messages (Crabtree et al., 2020;Widmann, 2019), preferring fear to enthusiasm in campaign communications (Nai & Maier, 2021). Nai (2018, p. 241) sees negative and fear-fueled campaigns as "detrimental forces in modern democracies, fostering depressed turnout, cynicism, apathy, and a gloomier public mood." Affective communication is further displayed in negative campaigning. At its core, negative campaigning is about adversarial messages that focus on the weaknesses of opponents rather than the strengths of one's own party or candidates. Such messages argue why citizens should not vote for other parties or candidates, rather than pointing to why they should vote for their party or candidate, in an attempt to demobilize opponents' supporters and sway undecided voters. Adversarial messages can be aggressive to varying degrees. Negative campaigning is not necessarily bad for democracy, as it may convey information to voters that they otherwise would not get (Kahn & Kenney, 2000;Mattes & Redlawsk, 2015). On the other hand, scholars have argued that "going negative" can have demobilizing or corrosive effects (Ansolabehere & Iyengar, 1995;Jackson et al., 2009), alienate voters, decrease turnout, and lead to a "spiral of cynicism" (Cappella & Jamieson, 1997). Research on negative campaigning has shown that negativity is multi-dimensional, including, but not limited to, broadcasting messages of negative tonality and evoking negative emotions such as anger, fear, anxiety, or even hate (Sanders & Norris, 2005). Thus, negative campaigning often seeks to evoke negative emotions toward political opponents, yet messages designed to evoke negative emotions can also refer to other things, like economic crises or immigration. While negative campaigning can involve evoking negative emotions among voters in some cases, they are nevertheless distinct communication strategies.
The development of negative campaigning over time remains poorly understood, as does the question of whether the approach yields benefit for parties and candidates. The observation that parties do go negative implies a belief that a negative strategy can be effective, yet research on ads has shown mixed results (Lau & Rovner, 2009). In his extensive literature review, Haselmayer (2019) points out that structural features distinguish campaigning in multi-party systems, some with pre-election coalitions, from the polarized campaigning in two-party systems such as the US, and calls for more large-scale, longitudinal, comparative studies in this field. Indeed, not only party systems but also election regulations vary, and negative campaigning in multi-party systems can have networked dynamics, e.g., retaliation or attacking an "enemy's friend" (Song et al., 2019). Auter and Fine (2016) suggest that negative campaigning can be an "underdog" strategy of "punching upwards" against incumbents, governing parties, or those who lead in the polls (Gross & Johnson, 2016) and that populist parties and parties at the fringes of the political spectrum, who often present themselves as political outsiders, are more likely to go negative (Nai, 2020). Moreover, research has shown that negative emotions and populist messages increase the "likeability" of Facebook postings (Bobba, 2019;Hameleers et al., 2016).
Since ancient Greek theater, there has always been a close connection between politics and dramatization. Alexis de Tocqueville observed that democracy needs drama, a feverish state before elections. Dramatic effect is also reflected in theoretical concepts of political communication and the public sphere, such as the idea of politics being "staged," frontstage and back-stage politics, or the arena model of the public sphere. Mackenzie and Porter (2011) discuss dramatization in political theory and found that dramatization is not necessarily deplorable, especially if it "brings to life" politics and political concepts. However, the literature on campaigning connects dramatization more closely with negative aspects, such as decreasing rational discourse, increasing the "intensity of public sentiment" (Krastev, 2014), overemphasizing threats and crises, increasing risk perceptions with alarmist tonality (Klemm et al., 2016), turning serious politics into a spectacle, and giving rise to affective politics steered by emotion rather than arguments and facts. Particularly with the advent of commercial television in Europe, scholars noticed a trend toward dramatization and spectacularizing in election campaigns (e.g., Schulz & Zeh, 2005). Dramatization can be employed as a strategy to mobilize and perform protest, incentivize people to vote, or to nudge them into certain behaviors. Thus, in a democracy, political actors should know how to balance dramatization and trivialization, and when to heat up or cool down public discourse.
Populism is closely connected to negativity and dramatization -both in content and in style of messaging (Bracciale & Martella, 2017;Jagers & Walgrave, 2007). Over the years, the chameleonic phenomenon of populism has been studied as an ideational concept, a mode of discourse, a mobilization strategy, an organizational type, or a political logic (e.g., Laclau, 2005;Moffitt & Tormey, 2014). Following De Vreese et al. (2018, p. 3), populism is a communication phenomenon that manifests in the content of messages, rather than an attribute of specific actors: "the emphasis is on populist messages as independent 'phenomenon-as-such' and no longer on a particular party family or type of politician." Populist messages typically underline the moral superiority of a homogenous "people" that are threatened vertically by corrupt and dishonest "elites" (in politics, the economy, international organizations) and horizontally by ostracized groups that are rhetorically excluded from "the people," such as migrants, people of color, LGBTQI+, religious or other groups, or all of the above. Populism repudiates intermediation (Puhle, 2020), instead claiming a direct communication channel between "the people" and the populist leader, party, or movement. This is why social media are a natural habitat for populist messages, as their affordances help to sustain the narrative and imagination of the direct channel. Furthermore, there is no "epistemic editing" on social media (Bimber & Gil de Zúñiga, 2020), such as the scrutinizing of truth claims or information sources. Accordingly, scholars have studied populist communication on social media, albeit with only few comparative studies published so far (e.g., Ernst et al., 2017;Stanyer et al., 2016). Populist messages contribute to dissonant public spheres because they seek to unite and mobilize "the people" but divide society (the people versus elites versus presumed out-groups) through emotion rather than rational arguments, while emphasizing anger and fear (Hameleers et al., 2017), and seeking to disrupt the norms of mainstream political communication through calculated provocations (Pauwels, 2011).
Campaigning in hybrid media systems and democratic elections is strategic communication (Klinger & Russmann, 2017;Strömbäck & Kiousis, 2014). Political parties invest in professional communication departments, employ external PR strategists, and work directly with social media platforms to optimize their messages (Kreiss & McGregor, 2018), maximizing followers and interactions to reach and mobilize voters at large as well as targeting specific groups. As social signals of "success," social media platforms afford a variety of user engagement measures, or popularity cues -in the case of Facebook, these include the number of reactions (in 2014: likes), shares, and user comments. Parties reap double benefits from posts that generate substantial user engagement. There is no audience on social media unless users spread content within their own networks. User engagement enables messages to spread beyond the circle of immediate followers to more peripheral networks, so that more potential voters can see and engage with them. Moreover, social media are algorithmically curated environments, where algorithms attribute relevance to messages that generate engagement and amplify these messages. Thus, parties seek to get as many reactions, shares, and comments as possible, because this helps spread the message with double amplification: by users and algorithms. However, it remains an open question whether communication strategies centering on evoking negative emotions, negative campaigning, dramatization, and populist content are actually rewarded with more user engagement. Against this background, our study was guided by the following research questions: RQ 1: Has the share of posts with negative emotions, negative campaigning, dramatization, and populist content changed between 2014 and 2019? RQ 2: Do posts with negative emotions, negative campaigning, dramatization, and populist content yield more reactions, likes, shares, and comments? RQ 3: Which parties generate most reactions, likes, shares, and comments -and do they benefit from negative emotions, negative campaigning, dramatization and populist content?

Data and Method
The study includes the United Kingdom and 11 of the 27 remaining EU countries: Austria, Croatia, France, Germany, Hungary, Ireland, Italy, the Netherlands, Poland, Spain and Sweden. These countries present diversity across several important differentiating dimensions within the EU, e.g., geographic positioning, political system, media system, monetary contributors to and benefactors from the EU. Facebook is the most important social media platform in each of the countries, particularly in terms of news and political information gained from social media (AU 30%, ESP 44%, CRO 55%, FR 43%, GER 22%, HU 61%, IRE 39%, IT 56%, NL 28%, PL 65%, SWE 32%, UK 24% (Digital News Report, 2021)).
Data were collected during the two weeks before the election day in each country. The data were drawn from the Facebook profiles of 111 (in 2014) and 116 (in 2019) political parties. Our study includes parties with a Facebook page that were categorized in the Party Manifesto Project 1 or the European Electoral Project 2 (data sources for party characteristics), with the addition of very few exceptions. Thus, we include more than just the major parties, but not all fringe parties. The sample includes 6,100 party posts for 2014 and 8,193 party posts for 2019 (see Table A1 in the Appendix). Data was collected by academic-led company Sotrender.com, which continuously collected posts and their metadata during the period of study. The data was downloaded just after the campaigns ended, so any additional reactions by users produced after the elections were not taken into account. Sotrender does not control for the possibility of bots or "like farms," but makes a scan of official party profiles as they are visible to the users.
The data was coded using standardized quantitative content analysis. 13 coders (12 being native speakers) were recruited and trained in two sessions. After each, we ran a reliability test (on 150 randomly chosen posts in English), with a final outcome of Percent Agreement (.956) and Brennan and Prediger (.941).

Dependent Variables
User engagement: Kalsnes et al. (2017) argue that there are three basic types of connective affordances on social media platforms -acknowledging (reaction/likes), redistributing (shares), and interacting (comments). Following this logic, engagement is measured by the number of likes (reactions in 2019

Independent Variables
Our main content variables -negative emotions, negative campaigning, dramatization, and populist content -are each considered independently, such that all of them could be coded in a single Facebook post. All variables are binary coded and moderately correlated (see Table A2 in the Appendix).
Negative emotions: Fear and anger are the most prevalent of negative emotions that campaign communications may seek to evoke (e.g., Crabtree et al., 2020;Widmann, 2019). Following Bobba (2019), fear refers to messages presenting explicit elements related to alarm, anxiety, apprehension, confusion, nervousness, panic, pessimism, worry, danger, and unease, among others. Anger refers to messages with explicit elements related to disappointment, discontent, dislike, dissatisfaction, frustration, humiliation, malice, moodiness, shame, spite, and wrath. These can be found, for instance, in: "With such an unabashed and provocative display of contempt for basic democratic and human rights principles, one cannot simply pass over to business as usual. Vienna must not become a marching ground for right-wing extremist groups from all over Europe" (The Greens/ Austria, 2014, translated by the authors).
Negative campaigning about political opponents includes all forms of disapproval, criticism, and attack messages. Negative campaigning seeks to portray an opponent's character or position on issues as undesirable. It reflects disapproval of parties and their politicians' decisions and statements on current issues, questions their policies, records, and qualifications, and challenges their competence to govern. In doing so, they highlight politicians' and parties' shortcomings, criticize their personalities, and call their judgment into question. Parties' campaigns can also be presented in an unfavorable light in these negative posts, such as, "Did you know? Facts about UKIP Nigel Farage and his millionaire #UKIP pals are doing everything they can to dupe people into voting for them tomorrow . . . so they won't tell you about their actual policies! Please help warn voters by sharing this viral image everywhere." (UK, British National Party Facebook account, 2014) Dramatization was coded if the post contained an exceptional degree of exaggeration or emphasis in discussing politics, political actors, or the campaign. Dramatization manifests in messages presented in a sensationalizing, moralizing, pillorying, excited, exaggerating, or escalating tone. It therefore differs from posts that maintain a rather sober, factual, unbiased, descriptive, de-escalating, neutral, and non-moralizing tone. For instance, dramatization is coded in, "On Thursday you have a choice: let fear decide Britain's future, or stand up to it. The only way to stop Farage winning, is to vote Labour" (Labor-Facebook account, 2019).
Populist content was measured according to the presence of (1) references to "the people," for whom populists demand more power and claim to be the one true representative, (2) exclusion of out-groups, (3) anti-elitism, and (4) using xenophobic language. Populism is not necessarily xenophobic, but as our sample includes more right-wing and far-right populist parties than left-populist parties, we included this variable. The post was coded as populist if it contained at least one of those elements, such as, "Stop the Islamification of Britain now. Please like and share" (UK, British National Party, 2014).
Party family: Parties were grouped into party families according to their ideological standpoint, in most cases based on the typology and data provided by the Party Manifesto Project. Those 25 parties that were not included in the Party Manifesto Project data were classified using a comparable proxy of ideological and economic standpoints from the European Electoral Project (see Table A3 in the Appendix).

Control Variables
Party size classifies the two main parties in parliament as major, other parties represented in parliament as minor, and those outside of parliament as fringe. The classification is performed separately for 2014 and 2019 based on the outcome of the last election to national parliaments preceding each EP election.
Average number of posts per day records the mean number of posts that each party produced per day (daily totals range from 0 to 31 in 2014 and from 0 to 28 posts in 2019). Controlling for posting frequency, allows consideration of the fact that overly frequent posting may limit the visibility of each post and thereby affect the number of user reactions.
Day of campaign counts the number of days remaining before Election Day and ranges from 14 to 1, as the Election Day itself is excluded from the analysis.
Community size is logarithmically calculated from the number of followers (those who "liked" the page) that each party had amassed by the day before Election Day. Country is added as a fixed effect variable with France as a reference, as negative campaigning is legally prohibited in France.

Results
The aim of this study is twofold. First, to examine from a longitudinal perspective the use of affective and dissonant communication by political parties on Facebook during European elections. Second, to analyze the level of user engagement achieved by parties employing varying communication strategies.
Our results for RQ1 show that all four communication strategies -evoking negative emotions, dramatization, negative campaigning, and populist content -increased in frequency from 2014 to 2019. This particularly applies to appeals to negative emotions. Nevertheless, all four strategies remain relatively infrequent when considering the percentage of all posts in which they feature: negative emotions increased from 4,7% (N = 285) in 2014 to 8% (N = 649) in 2019; negative campaigning increased from 5% (N = 325) to 7% (N = 587); dramatization increased from 4,7% (N = 289) to 6% (N = 488); and populist content increased from 4% (N = 247) to 6% (N = 505). The data further reveal that negative campaigning, dramatization, and populist content tend to decrease in the lead up to election days, whereas negative emotional appeals increase during this critical period (for more details please refer to Figure A1 in the Appendix).
Negative binomial regression models (Tables 1 and 2, see also Table A4 and Table A5 in the Appendix providing full statistical models) were run to partially answer RQ2 and RQ3. With regards to RQ2, we found that negative emotions, negative campaigning,   (Table 1), but could not be statistically connected with user engagement in 2019 (Table 2). With regard to RQ3, Figure 1 shows the relation between party ideological leaning (with centrist parties as reference) and their employment of various content characteristics, as indicated by the regression models (Table A6 and A7 in the Appendix provide full statistical models).
Taking both elections into account, Nationalist parties seek to evoke negative emotions more often and use more negative campaigning, dramatization, and populist content in their posts than the other party families. Over time, the occurrence of dramatization within Nationalist messages has doubled (2014: OR = 2.012 (p < .001); 2019: OR = 4.672. (p < .001)), whereas their use of both negative emotions and negative campaigning decreased from 2014 to 2019 (2014: negative emotions OR = 6.301 (p < .001), negative campaigning OR = 4.069 (p < .001); 2019: negative emotions OR = 3.026 (p < .001), negative campaigning OR = 2.464 (p < .001)). The data show no changes for populist content.
The most noticeable change between the elections relates to Conservative parties, which hardly employed negativity, dramatization, and populist content in their Facebook posts in 2014, * p < 0.10, ** p < 0.05, *** p < 0.01. Facebook affordances are not indicated due to space limit, full model is available in Appendix. yet used even more negativity and populism than Nationalist parties in 2019. In the 2014 EP election, Conservative parties' Facebook posts did not significantly differ in these characteristics from those of Centrist parties, while in 2019 they outperform them by two to five times (Conservatives, 2019: negative emotions OR = 3.225 (p < .001), dramatization OR = 2.957 (p < .001), negative campaigning OR = 3.946 (p < .001), populist content OR = 5.636 (p < .001)). Parties on the left side of the ideological spectrum show less change over time. Parties grouped as Left Socialists posted more populist content on Facebook in the 2019 election than in the 2014 election (2014 OR = 3.263 (p < .001), 2019 OR = 4.141 (p < .001)). The other three content characteristics show a significant increase in terms of the difference between left parties and centrist parties, but did not change substantially over time. Ecologist parties employ substantially more dramatized messages in 2019 (OR = 3.796 (p < .001)) than other parties on the left side. An important change between the two analyzed EP elections is revealed for Special Issue parties. Their messages became more dramatized (2014 OR = 1.852 (p < .001), 2019 OR = 3.464 (p < .001)), used more negative campaigning (2014 OR = 2.656 (p < .001), 2019 OR = 3.382 (p < .001)), and included more populist content (2019 OR = 2.579 (p < .001)). Figure 2 displays the proportion of likes and shares of a post, with the number of comments indicated by the size of the bubble. From 2014 to 2019, the correlation between likes/reactions and shares (r = .44 to r = .66, p < .001) as well as between shares and comments (r = .32 to r = .52, p < .001) increases substantially. We can also observe that posts by nationalist parties (black bubbles) were among the most liked and shared posts in 2014 and gained further in popularity by 2019.
To answer RQ3, the negative binomial regression models (Tables 1 and 2) are employed to identify relations among party characteristics, content characteristics, and user engagement. In both elections, being a Nationalist party is related most strongly to attracting user engagement. The coefficients change only slightly between 2014 and 2019 for predicting shares (possible increase) (2014: β = .826 (p < .001), 2019: β = 1.308 (p < .001)) and receiving comments (decrease) (2014: β = 1.302 (p < .001), 2019: β = .970 (p < .001)). In general, the data for the 2019 EP election show that posts by parties on the fringes of the political spectrum are more likely to attract user engagement. This marks a change from the 2014 EP election, when, for instance, posts of Left Socialist parties attracted shares but hardly any likes and comments, while posts of Conservatives received comments but hardly any shares (Table 1). In 2019, being a Left Socialist or Conservative positively relates to receiving reactions, shares, and comments ( Table 2). The greatest shift in user engagement is observed for Ecologist parties, a classification that attracted likes (β = .132 (p = .10)) and shares (β = .489 (p < .001)) in 2014, but was negatively correlated with reactions and comments in 2019, while remaining positively correlated to shares, albeit less so than in the previous election (β = .277 (p < .001).

Discussion and Conclusions
Our study shows that political parties across 12 countries sought to evoke negative emotions and dramatization more frequently, engaged more in negative campaigning, and included more populist content in their Facebook posts in the 2019 EP election than in 2014 (RQ1). Further, we show that posts evoking negative emotions, dramatization, and negative campaigning yield more user engagement than other posts (RQ2). Populist content also increased user engagement in 2014, but not in 2019. Going negative and exaggerating as well as sensationalizing messages makes perfect sense from a strategic perspective, because parties using these tactics capture more user attention and thus reap more reactions, likes, shares, and comments, thereby making their messages travel farther and deeper in social networks. But it is not only strategic party behavior that may explain the development we show in the analysis. The Facebook Papers published by whistleblower Frances Haugan have pointed to the underlying algorithmic curation as another possible explanation. The platform does not award equal weight to all types of reactions, but gives "angry" reactions five times the weight of an ordinary "like," so that consequently, posts that evoke anger have more reach and visibility (Merrill & Oremus, 2021). A document from the Facebook Papers provides evidence that European parties have complained to Facebook that the platform structurally incentivizes them to post more angry and aggressive messages -and that the parties feel pressured to communicate more negatively to not have a disadvantage (Sabur, 2021). Of course, the impact of algorithms is hardly accessible for empirical study, due to lack of data access.
By increasing their usage of emotional, non-rational messages, attacking rather than advertising, and heating up public discourse instead of cooling it down, political parties are generating content which seems to contribute to more affective publics and more dissonant public spheres. Our study suggests that parties are increasingly making use of these strategies, even though still on a low level, and thereby it empirically supports Bennett and Pfetsch's argument (Bennett & Pfetsch, 2018) that public spheres are increasingly marked by disruptive communication processes -at least for Facebook. With attention focused on one particular actor type in dissonant public spheres -political parties -our results suggest that dissonance, a shifting away from sober, rational debate is not only to be blamed on trolls, aggressive inauthentic networks, or general noise on platforms, but also on the communication strategies of core actors in democracies, who opt to exploit the platform's preference for affective, angry messages to gain more user engagement. However, we do not claim a correlation or causal link between the tone of campaigning and that of general public discourse.
Most interesting is the fact that it's not just the "usual suspects" driving this change toward more affective and dissonant communication (RQ3). In 2014 it was actors at the fringes of the political spectrum employing these strategies, predominantly the Nationalist parties, and to some extent Left Socialist parties. Our findings regarding Nationalists are in line with previous conclusions that these parties, for instance, use negative emotions like anger and fear more often than other parties (Crabtree et al., 2020;Nai & Maier, 2021;Widmann, 2019), as with negative campaigning (Nai, 2020). Indeed, the Nationalist parties constitute the only party family that did not change its Facebook communication strategy between the two elections, with no substantial change in the responsiveness of their community. This corroborates the recent finding from Koc-Michalska and Lilleker (2020, Table 5) that Nationalist parties have loyal and active Facebook communities, and it seems that not much had changed over the preceding years. Nationalist parties appear to be persistent and rather stable players on the European electoral scene.
Driving the overall shift toward negative emotions, dramatization, negative campaigning, and populist content between the elections, however, was most notably the increased engagement in these strategies by Conservative parties, who were rewarded with more stable responses from their communities, especially with more reactions and shares. Ecologist parties also shifted their 2019 campaign toward more dramatization. Overall, our study suggests shifts toward a) more affective and dissonant communication by parties and b) increased user responsive to these types of communication.
Mapping likes/reactions, shares, and comments for all party posts across two election campaigns, the glaring finding is that Nationalist parties were already very successful in generating user engagement in 2014, and that this pattern became even more pronounced in 2019. The increasingly strong linear correlations between likes/reactions, shares, and comments suggests more strategic and integrated deployment of Facebook in parties' campaign communications, yet the possibility that this is "just" an effect of changes to Facebook's algorithms cannot be excluded given currently accessible data.
As with all studies, our study has limitations. Although we analyzed the most important social media platform when it comes to disseminating news to citizens as well as gathering political information (Digital News Report, 2021), our analysis is nevertheless restricted to a single social media platform. Future studies need to take platforms such as YouTube, Instagram, and TikTok into account, in recognition that Facebook's dominance in election campaigns might decrease over time with the rise of newer platforms. Similarly, our comparison included only a selection of EU countries: Ideally, future research will take all EU countries into account. Longitudinal studies are always confronted with changes and this is especially true for social media, where changes happen constantly and often fast. For instance, Facebook's algorithms changed between 2014 and 2019, affecting user reactions, and the like button was updated in 2016. While in 2014, users could only generically "like" something, in 2019 they had more response options through six different reactions (like, love, haha, wow, angry, sad). In our study, we compared the 2014 "like" to 2019 "reactions" (all six reactions), but of course, they do not necessarily mean the same thing. Regarding the categorization of parties into party families, the Party Manifesto Project counts Hungarian Fidesz as a Conservative party and Irish Sinn Fein as a special issue party, when both are borderline cases that could also have been classified as Nationalist parties -which would have enhanced the patterns presented. From here, it could be insightful to incorporate our results with qualitative interviews with the political parties to address the why-question, i.e., understanding parties' strategic decisions.
Are campaigns getting uglier? To conclude, yes . . . on Facebook, they are. Moreover, it seems that affective and dissonant communication does not emerge purely unintentionally, as the strategic use of negative emotions, dramatization, negative campaigning, and populistic content within party campaigns yields benefits in the form of likes, shares, and comments. After all, the aim of strategic communication is to build relationships with stakeholders. But likes, shares, and comments do not directly translate into votes. Indeed, the parties who are most successful in gaining reactions on Facebook, namely Nationalists and Special Issue parties, are far from the most successful in winning votes. However, political parties are drivers of dissonant public spheres with this communication strategy that does not really win anything for them and additionally is not doing any good for contemporary democracies. Notes 1. manifesto-project.wzb.eu. 2. sciencedirect.com/science/article/pii/S2352340920308623. 3. In 2016, Facebook diversified "likes" into 6 categories of reactions (like, love, haha, wow, sad, angry). For 2014 we collect likes, for 2019 we add all reactions.