The Wrath of Candidates. Drivers of Fear and Enthusiasm Appeals in Election Campaigns across the Globe

Abstract

Little is known about why candidates decide to make use of emotional messages when campaigning for a political office, and under which conditions this is more likely to happen. We focus on the use of fear and enthusiasm appeals and assume that these are a function of profile of candidates and the nature of the context in which the election takes place. We use a new large-scale comparative dataset which includes information about campaigning strategies for 636 candidates having competed in 133 presidential and parliamentary elections in 101 countries between June 2016 and March 2020, based on judgments of 2000+ domestic and international experts. Our results show that candidates benefitting from a comparative advantage (incumbents and frontrunners) tend to rely on enthusiasm appeals, more extreme candidates prefer fear to enthusiasm, and more competitive races tend to foster the use of fear appeals. These findings have important implications for electoral competition, communication theory, and political marketing. All data and materials are openly available for replication.

Supplemental data for this article is available online at https://doi.org/10.1080/15377857.2021.1930327 .

1. Introduction

As already remarked by Ted Brader in his 2005 landmark study, it is rather surprising how little we know about the dynamics of emotional appeals in electoral campaigns – especially in light of the fact that emotionally charged messages to mobilize and energize the masses are very frequent in election campaigns. When it comes to the role of emotions in politics, consistent evidence shows their centrality for exposure to (and processing of) political information (Redlawsk 2006; Valentino et al. 2008; Nai et al., 2017), cognition (Kühne et al. 2011; Miller 2011), political attitudes (Vasilopoulos, Marcus, and Foucault 2018), and voting choices (Marcus 2000; Marcus, Neuman, and MacKuen 2000; MacKuen et al. 2010). This being said, much less evidence exists from the supply side; to what extent do politicians rely on emotional messages when competing for public office? And under which conditions are they more likely to do so?

Candidates have clear incentives to use appeals intended to stir an emotional response in the audience (Jerit 2004; Brader 2005, 2006; Crigler, Just, and Belt 2006) – for instance in terms of mobilizing the crowds by drumming up the enthusiasm, or in persuading undecided voters by triggering anxiety on relevant issues. Using emotions to give messages a distinctive “branding” and to “affectively connect” with voters is, of course, not a new approach. Political marketing textbooks provide countless examples of more or less successful emotional brandings, from Giscard d’Estaing’s “You don’t have a monopoly on the heart, Mr. Mitterrand” during the 1974 French presidential election to Donald Trump’s repeated fear appeals in his 2016 successful bid (e.g., Lees-Marshment et al. 2019, 112–114). And, yet, beyond some notable exceptions (e.g., Kaid and Johnston 2001; Ridout and Searles 2011), very little energy has been provided to try and understand the reasons why candidates decide to opt for an emotional strategy when campaigning for a political office – and under which conditions this is more likely to happen. Even more dramatically, the existing studies focus on one very specific case – American elections – and look into rather specific sets of drivers: the candidates’ political role, their competitive standing, and the timing of the campaign. In this article, we expand on the existing literature by shifting our attention also to the ideological differences between candidates (in terms of left-right positioning and “extremism”). Much attention has been provided in recent years to the use of a certain type of emotional appeals (fear and anxiety) by a certain type of political actors (far-right candidates and parties; Bennett 1988; Wodak 2015; Allen 2017) and, yet, little systematic evidence exists that the ideology of competing candidates drives their political rhetoric in a broad way.

Even more importantly, we present what is to the best of our knowledge the largest comparative investigation about the drivers of emotional messages in election campaigns worldwide. We rely on an original comparative dataset (Nai 2020) that contains systematic information about campaigning strategies of 636 “top” candidates (party leaders or presidential candidates) having competed in 133 presidential and parliamentary elections in 101 different countries between June 2016 and March 2020 – covering virtually all national elections that happened worldwide in that period, from the USA to Eastern, Northern and Western Europe, the Middle East, the African continent, Eastern and Southern Asia, Oceania, and Central and Latin America. Elections covered span across all types of electoral and party systems, and vary considerably in terms of competitiveness, closeness of the results, and media coverage. According to our rough estimation, at least 2.5 billion citizens took part in these elections. The dataset contains information about the campaigning strategies of a wide selection of “top” candidates across all ideologies, genders, incumbency statuses, and electoral fortunes. The list of 636 candidates includes world key players such as Donald Trump, Emmanuel Macron, Vladimir Putin, Angela Merkel, Matteo Salvini, Benjamin Netanyahu, Justin Trudeau, Shinzo Abe, Theresa May, Jair Bolsonaro, Narendra Modi, Rodrigo Duterte, Boris Johnson, and many more. Such a large-scale comparative database allows us to simultaneously circumvent the curse of extreme geographical uniformity of existing studies, test empirically how differences at the contextual level drive candidates’ strategies, and, ultimately, provide a significant contribution to the developing field of comparative political communication. The dataset is based on a systematic survey distributed to samples of national and international scholars with expertise in elections and politics in the surveyed country, and includes ratings provided by 2082 scholars.

We proceed as follows: In the next section, we briefly describe the concept of emotions and explain how enthusiasm and fear are processed from the perspective of the Affective Intelligence Theory (AIT). We claim that the likely consequences of enthusiasm and fear appeals for political judgments set a strong incentive for candidates to go emotional in their campaigns, explore the role of candidate and context characteristics in more depth, and formulate some hypotheses. Our results show that candidates benefitting from a comparative advantage (incumbents and frontrunners) rely on enthusiasm, that fear is more likely to be used by more extreme and right-wing candidates whereas enthusiasm is more likely from the left and less extreme candidates, and that competitive elections foster the use of fear appeals; elections fought in countries where proportional representation (PR) is the norm are less likely to contain fear appeals.

All data and syntaxes are available at the following Open Science Foundation (OSF) repository, for replication purposes: https://osf.io/xjrnc/.

2. Emotions and emotional campaigns

2.1. Emotions in politics

There is of course substantial disagreement about what exactly an emotion is, how many (relevant) emotions exist, and how they are structured (e.g., Ekman and Cordaro 2011; Russell 1983). With respect to the definition, the lowest common denominator seems to be that an emotion is a temporary physiological arousal induced by a specific stimulus. Emotions are connected with a positive or negative affect toward the stimulus, which can vary in strength (e.g., Zajonc 1984; Scherer 2005). More importantly, emotions have a structuring effect on cognitive processes (Damasio 1994). Indeed, “these unconscious evaluations are far more active, and hence far more important, than conscious cognitive processing” (Marcus 2000, 231), to the point that “without the ability to reference emotions people remain incapable of making so called rational decisions” (McDermott 2004, 694). Starting from this postulate, the AIT (Marcus and MacKuen 1993; Marcus 2000; Marcus, Neuman, and MacKuen 2000; MacKuen et al. 2010) argues that people rely on previously held predispositions in familiar circumstances, but rather tend to engage in more effortful cognitive processes when facing novel or uncertain circumstances. The surveillance system “generates anxiety/unease or relaxation/calm as incoming information suggests it is either safe or potentially unsafe to go about one’s business as usual” (Brader 2006, 60); anxious citizens are likely to pay more attention to news and campaigns (Steenbergen and Ellis 2006). Because they do so by relying less on their political preferences (Marcus, Neuman, and MacKuen 2000), they are easier targets for persuasion (Nai et al. 2017). On the other hand, the disposition system “generates enthusiasm/satisfaction or depression/frustration as incoming information reports that the execution of one’s plans either matches or does not match expectations (or success)” (Brader 2006, 60); enthusiastic citizens are more likely to get invested and participate (Marcus and MacKuen 1993), but they do so by relying strongly on their predispositions (Brader 2006; Marcus, Neuman, and MacKuen 2000) – reason why enthusiasm can also have a more nefarious consequence in the entrenchment of political fanatism and the fostering of uncritical citizenship. All in all, knowing the potential importance of emotional experiences for citizens while making up their mind, it is unsurprising that candidates have clear incentives to go emotional (Brader 2006; Crigler, Just, and Belt 2006).

What is described in the previous paragraphs is, for sure, a rather simplified view of the role of emotions in politics. First, other models exist to describe the role of emotions in cognition beyond the AIT. For instance, the Appraisal Theory of emotions (Lazarus 1991) also focus on distinct emotions and their individual experience but claims that emotions are elicited by evaluations (appraisals) of personally relevant specific situations and events (Roseman and Smith 2001; So, Kuang, and Cho 2016). Second, even if it is currently the “dominant affective theoretical model” in political science literature (Ridout and Searles 2011, 441), the AIT has attracted over the years some critiques (e.g., Ladd and Lenz 2008, 2011), mostly addressed to the role of anxiety for decision making. Third, the AIT model itself is often expanded to include a third emotion – anger (or rage and aversion) – which has been shown to follow a distinct set of dynamics than fear/anxiety (e.g., Lerner et al. 2003; Lerner and Keltner 2001; Steenbergen and Ellis 2006; Redlawsk, Civettini, and Lau 2007) and stems from a reaction “to a diminishment to one’s well-being or an attack on one’s beliefs and goals for which others are perceived to be responsible” (Steenbergen and Ellis 2006, 113). Yet, even with these considerations in mind, fear and enthusiasm are the two central emotional appeals when it comes to modern election campaigning (Brader 2005, 2006).

2.2. Why candidates “go emotional”

The centrality of emotions for political and electoral behavior creates strong incentives for competing candidates to “go emotional” in their communications - that is, use appeals intended to trigger these emotional responses in the public (Jerit 2004). We focus here on two types of emotional appeals. On the one hand, “fear appeals” are intended to trigger fear and anxiety in those exposed to them. They are a prime example of negative emotionality, and are a preferred tool in public health campaigns against, e.g., smoking or drug use. In contrast, “enthusiasm appeals” are intended to stir a positive emotional response, to drum up satisfaction and a general “feel good” sentiment toward the sponsor of the message. Why, and under which conditions, are candidates deciding to use one rather than the other type of emotional appeals in their campaigns?

2.2.1. Candidate profile

Due to the opposite cognitive effects shown in the literature for enthusiasm and anxiety (Marcus 2000; Marcus, Neuman, and MacKuen 2000), we argue that enthusiasm appeals are used by candidates wishing to consolidate their advantageous position, whereas fear appeals are used by candidates trying to improve their competitive standings by persuading additional voters to vote for them. According to the AIT (Marcus and MacKuen 1993; Marcus 2000; Marcus, Neuman, and MacKuen 2000; MacKuen et al. 2010), enthusiasm reinforces predispositions, whereas anxiety and fear uncouple individuals from their predispositions and make them more exposed to political persuasion (Nai et al. 2017). Thus, candidates campaigning from an advantageous position –most notably, incumbents (Cox and Katz 1996) – face incentives to consolidate such advantage, and thus incentives to drum up enthusiasm in the masses; inversely, candidates facing an unfavorable standing (tailgating in the polls, or having a challenger status) need to convince more voters to support them, including voters that initially were not supporting them; anxiety appeals have the power to swing allegiances by reducing the importance of partisan attitudes, and should thus be the preferred emotional weapon for them. Ridout and Searles (2011) advance a similar expectation. This expectation could also be supported by the idea that, very much like negative campaign messages (Lau and Pomper 2004), fear messages might be efficient to convince undecided voters but face, at the same time, a risk of “backlash effects” (Shapiro and Rieger 1992; Roese and Sande 1993). As discussed in Ridout and Searles (2011), the risk inherent in running “negative” political ads (that is, ads that attack the opponents instead of promoting the candidates’ own policies and programs) can be assumed to be at play also with negative emotional ads, in that these might be viewed unfavorably by the public.

H1. Incumbents are more likely to use enthusiasm appeals, whereas challengers are more likely to use fear appeals

Second, more extreme candidates – that is, candidates whose ideology is the most distant from the “center” – are more likely to use fear appeals (and less likely to use enthusiasm appeals), regardless of their ideological position on the left-right spectrum. Spatial models of electoral competition (Merrill and Grofman 1998; Tomz and Van Houweling 2008; Großer and Palfrey, 2014) assume that extreme or “fringe” parties and candidates need to disrupt the partisan loyalties in place to increase their electoral chances – which, as suggested by the AIT, is efficiently done through fear appeals that uncouple voters from their partisan preferences.

H2. More extreme candidates are more likely to use fear appeals and less likely to use enthusiasm appeals

Third, we could expect that the ideological profile of candidates drives their campaigning choices. Evidence exists that parties and candidates on the right-hand side of the ideological spectrum, and especially far-right parties, are more likely to rely on campaigns with a negative emotionality, and especially on fear appeals (e.g., Bennett 1988; Wodak 2015) for instance on partisan issues, such as immigration or crime (Charteris-Black 2006; Allen 2017). At the individual level, the difference between liberalism and conservatism is rooted in differential preferences for approach or avoidance motivations; political conservatism “is avoidance based; it is focused on preventing negative outcomes (e.g., societal losses) […, whereas liberalism] is approach based; it is focused on advancing positive outcomes (e.g., societal gains) and seeks to regulate society via activation (interventions)” (Janoff-Bulman 2009, 120). More importantly, hard conservatism and especially right-wing authoritarianism can be associated with “a view of the world as a ruthlessly competitive jungle in which the strong win and the weak lose” (Duckitt 2006, 685), and where the values of personal security and social dominance are promoted (Heaven and Bucci 2001), which is likely to make individuals seeing social interactions through the prism of anxiety and fear. It seems thus safe to expect that the use of fear appeals is particularly likely for candidates on the (hard) right side of the political spectrum, and vice-versa for enthusiasm.

H3. Candidates on the (far) right-end of the ideological spectrum are more likely to use fear appeals, and less likely to use enthusiasm appeals

2.2.2. Election context

Turning to the context in which the elections take place, we focus on two elements that are likely to shape the way campaigns are fought, above and beyond the profile of competing candidates: the conflict inherent in the race, and the “political culture” inherent of the political system in which the election takes place.

First, emotional campaigns should be expected during particularly challenged elections. When the results of the election are uncertain, for instance during very competitive elections, candidates have, we argue, incentives to use both fear and enthusiasm appeals. Appeals that generate fear and anxiety are theoretically more likely to be persuasive (Nai et al., 2017) and can thus be expected to be particularly useful to grasp the last undecided voters and convince some ambivalent or weakly aligned citizens to turn out for the candidate using fear appeals. At the same time, enthusiasm appeals are likely to boost the participation of partisans (Marcus and MacKuen 1993; Brader 2006; Marcus, Neuman, and MacKuen 2000), and can thus be particularly useful to make sure that the partisan potential in the electorate remains mobilized up to the moment of the vote.

H4. Fear and enthusiasm appeals are more likely during competitive elections

Second, and beyond the conjunctural dynamics of the election itself, the institutional and political environment is likely to shape the content of campaigns. It is not a mystery, for instance, that the use of harsh campaigns is especially endemic to certain political cultures; that negative campaigning is “as American as apple pie” (Scher 1997, 27), whereas in the Netherlands, for instance, it is substantially less frequent (Walter and van der Brug 2013). We argue that the use of negative emotions should be less likely in electoral systems characterized by PR. In PR countries electoral competition tends to be more inclusive, more parties occupy the ideological space, and governance tends to reflect compromise and concertation (Lijphart 1999; see also Valli and Nai 2020). As discussed by Walter, Van der Brug, and van Praag (2014) with regards to campaign negativity, in proportional settings political actors have strong incentives for cooperative behavior due to the likelihood of post-election coalition bargaining. Because post-election governance is often shared among multiple actors, potentially belonging to different ideological families, the use of excessively harsh campaigns could likely alienate coalition allies. Furthermore, the multi-partisan nature of electoral competition in proportional environments is likely to create additional incentives for competing actors to mobilize their own base – on top of incentives to demobilize support for their opponents – which is what enthusiasm appeals are intended to secure (Brader 2006).

H5. Enthusiasm appeals are more likely and fear appeals are less likely in countries where proportional representation (PR) is the norm

3. Data and methods

3.1. The dataset

We test our expectations via an innovative dataset (Nai 2020) that contains information about the campaigning strategies of 636 “top” candidates (party leaders or presidential candidates) having competed in 133 presidential and parliamentary elections in 101 different countries between June 2016 and March 2020 – covering virtually all national elections that happened worldwide in that period, excluding some micro-states with less than 100,000 inhabitants (e.g., Cayman Islands, Palau, and Bermuda) or noncompetitive elections (e.g., Turkmenistan or the uncontested election in Singapore of September 2017). The list of all candidates and elections covered is in Appendix A, available in the OSF repository (https://osf.io/xjrnc/).

The dataset contains information based on a systematic survey distributed to election-specific samples of national and international scholars¹ with expertise in elections and politics in the surveyed country. Experts were contacted in the direct aftermath of each election (often, the very next day), and asked explicitly to rate the campaign of up to 10 of the most important “top” candidates – that is, party leaders in legislative elections (e.g., Angela Merkel in Germany, Justin Trudeau in Canada) or candidates in presidential elections. In this sense, the expert survey is very much designed as a post-election survey, and not a general expert survey detached from the electoral calendar (such as, e.g., the Chapel Hill Expert Survey; Polk et al. 2017).

Considering elections for which at least three experts have provided their ratings, the dataset includes responses of 2082 experts. On average, 15.7 experts per election provided their answers (Table A1, Appendix A). On average experts lean rather to the left (M = 4.37/1–10, standard deviation [SD] = 1.81), 74.9% are domestic (that is, work in the country for which they were asked to evaluate the election), and 30.8% are female. Overall, experts declared themselves very familiar with the elections (M = 8.06/0–10, SD = 1.71), and estimated that the questions in the survey were relatively easy to answer (M = 6.57/0–10, SD = 2.34). Table B1 (Appendix B) shows the composition of the expert sample for each of the 133 elections.

Using experts to measure the content of political campaigns might seem unorthodox. Traditional approaches usually rely on content analyses of secondary data, for instance party manifestos (Curini 2011), TV spots (Benoit 2007), leadership debates (Maier and Jansen 2017), and so on. These approaches are unrealistic in a large-scale comparative setting as the one discussed in this article, where elections in more than 100 countries are compared. To the enormous costs to retrieve and code the materials are added massive language barriers. For instance, analyzing media coverage of candidates across the world raises severe complications in terms of, e.g., which comparable media to code across these extremely diverse countries. The same goes for the content of election campaigns per se, which are implemented rather differently across the globe; for instance, TV ads are common in the USA but are forbidden in other countries. In other words: because of the extreme diversity and sheer amount of cases covered in our study, traditional approaches to content analysis are unlikely to result in comparable and reliable measures. Experts, on the other hand, can be asked to assess the content of campaigns on the whole. Their opinions about campaign content can be uncoupled from the medium; even assuming that a comparable campaign channel exists across all countries investigated here (which is unrealistic), results of a content analysis of that channel would yield a partial figure, dependent on the nature and limitations of that specific channel. Experts can be asked to provide more holistic ratings, thus moving beyond problems of comparability across channels (e.g., Walter and Vliegenthart 2010).

Of course, aggregated expert ratings should be taken carefully. On the one hand, it is naïve to assume that experts are completely objective when it comes to the assessment of political candidates, even if they perform of course much better than the public at large (see, e.g., Nai and Maier 2019); we return to this issue later on. On the other hand, the confidence we can have in expert assessments is an inverse function of their agreement – that is, how similarly different experts evaluate any given candidate (e.g., Vazire 2006, 2010). One way to ascertain inter-expert agreement is to look at average SDs within each expert sample on aggregated ratings per candidate; the lower the SD, the more all values in the expert sample are close to the average rating, suggesting that raters tend to agree with each other. In our case, the average SD for the expert assessment of fear and enthusiasm appeals for the 636 candidates in our dataset are, respectively, 0.75 and 0.67 – relatively low considering the range of the original variables (0–10; see below). In other terms: experts tend to agree with each other quite consistently when it comes to the use of fear and enthusiasm appeals by the competing candidates.

Due to the large-scale scope of the dataset, it is unfortunately impossible to validate our measures against other data – quite simply, there is no other dataset that content the tone of candidates’ campaigns across the globe. Nonetheless, as we report elsewhere (Maier and Nai 2020), we were able to confirm the validity of expert assessments of campaign content by comparing the tone of candidates’ campaigns during the 2018 US Midterms elections as assessed by selected experts with two independent measures: the tone of their campaigns on social media (Twitter), and the content of TV ads in their favor. Even controlling for several factors at the candidate and state levels, our expert measure was significantly and positively correlated with the negativity on Twitter and in TV ads. This suggests that experts should be able to pick up the content of candidates’ campaigns, beyond the specificities of different campaign channels (Walter and Vliegenthart 2010); we have no reasons to believe that this should not be the case outside of this specific case as well.

All in all, expert ratings represent a systematic, reliable and easily scalable alternative to traditional methods of content analysis. No other method currently available is able, in our opinion, to provide comparable measures of campaign content for very large comparative datasets as the one we use here, that includes cases as different as, e.g., Afghanistan, Australia, Armenia, Austria, Argentina, Albania, and Algeria.

3.2. Measuring the use of emotional appeals

Experts were asked to assess the extent to which each candidate used fear and enthusiasm appeals. We provided experts with some selected examples of both fear and enthusiasm appeals², and asked them to rate each candidate on a scale ranging between 0 (“very low use of fear/enthusiasm appeals”) and 10 (“very high use”). Expert assessments were averaged for each candidate to provide the measures of fear and enthusiasm appeals. On average, the candidates in our dataset made a slightly stronger use of fear (M = 5.14, SD = 1.82) than of enthusiasm appeals (M = 4.43, SD = 1.57).

The use of the two types of emotional appeals is negatively and significantly associated, r (634) = 0.45, p < .001, that is, the likelihood of using fear appeals is a negative function of the use of enthusiasm appeals, and vice-versa. Figure 1 illustrates the use of both emotional appeals for all 636 candidates in our dataset and identifies a few selected examples.

Figure 1.

Fear and enthusiasm, all candidates. Note: Both axes vary between 0 “Very low” and 10 “Very high”. Selected candidates highlighted.

The graph shows for instance that Geert Wilders, leader of the Dutch far-right Party for Freedom (PVV), scored the highest across all candidates in terms of the use of fear appeals, while scoring among the lowest in terms of enthusiasm appeals – in line with his public reputation of often displaying a “controversial attitude and aberrant political style” (De Landsheer and Kalkhoven 2014) and of “not trying at all to be agreeable”³. Very similar scores of high fear and low enthusiasm appear for other far-right populist candidates as well, such as the leader of the Alliance for Germany (AfD) Alexander Gauland (“gloomy and pessimistic”⁴, often using an “inflammatory rhetoric”⁵ and known for his “controversial, headline-grabbing statements”⁶), The leader of Italy’s Lega Matteo Salvini, the leader of Spain’s far-right Vox party Santiago Abascal, and Donald Trump. On the other end of the spectrum, we find more mainstream and consensual candidates, such as Angela Merkel or New Zealand’s PM Jacinda Ardern, which have been evaluated as having used more positive emotional appeals during their campaigns.

3.3. Campaign loathing

Emotional appeals are, by far, not the only stylistic components – that is, related to rhetoric form – of modern election campaigns. Widespread research has highlighted the importance of additional dimensions, most notably related to the “tone” and incivility of the messages conveyed (e.g., Lau and Pomper 2004; Brooks and Geer 2007). Our dataset includes two measures for these additional dimensions. First, experts were asked to assess the “tone” of the campaign used by competing candidates in the election, that is, to what extent they “talked about the opponents in the race by criticizing their programs, attacking their ideas and accomplishments, questioning their qualifications, and so on” instead of “talking about one’s own accomplishments, qualifications, programs and ideas by praising them” (quoted directly from the questionnaire). We provided experts with a scale ranging from −10 to +10, where −10 meant a “very negative” campaign and +10 a “very positive” one (see Nai 2020 for a detailed development on this variable). Second, expert had to evaluate whether candidates “mostly used policy or character attacks in their communication and campaign events” (from 1 “exclusively policy attacks” to 5 “exclusively character attacks”), which reflects the focus of candidates’ negative campaigning (e.g., Brooks and Geer 2007; Hopmann, Vliegenthart, and Maier 2018).

Emotionality, negativity, and focus are intrinsically related, as shown in Table 1 via zero-order correlations. Campaigns that include more fear appeals tend to include fewer enthusiasm appeals (as discussed above), but also tend to be more negative in tone and focusing more on character attacks. As discussed in Gerstlé and Nai (2019), emotionality, negativity, and focus are associated to form the more general dimension of campaign “loathing”, that is, the use of a harsher rhetoric based on political attacks and fearmongering. A factor analysis (PCA) on the four variables (fear appeals, enthusiasm appeals, campaign tone, focus) reveals indeed the existence of one single underlying dimension (Eigenvalue = 2.73, 68% of explained variance). High positive scores on this variable reflect high loathing (negativity, character attacks, and fearmongering), whereas high negative scores reflect low loathing (more “positive” campaigns and enthusiasm appeals). We will replicate our analyses using this variable of campaign loathing as well, to put the results for the two specific emotions into perspective.

Table 1.

Campaign dimensions; zero-order correlations.

		Campaign tone	Focus of attack	Enthusiasm
Campaign tonea	r
	p
	N
Focus of attackb	r	−0.66
	p	0.000
	N	633
Enthusiasm	r	0.62	−0.23
	p	0.000	0.000
	N	636	633
Fear	r	−0.80	0.60	−0.45
	p	0.000	0.000	0.000
	N	636	633	636

^aVaries between – 10 “very negative” and +10 “very positive”.

^bVaries between 1 “exclusively policy attacks” and 5 “exclusively character attacks”.

3.4. Covariates

To measure the ideology of candidates, lacking again a comprehensive and systematic repository covering all the candidates in our data, we relied on information provided by the Wikipedia pages for each candidate (or party); those pages present, in the information box in the top-right corner, a standardized classification of the actor’s “political position” (far left, left, center-left, center, center-right, right, far right), which we have transformed into a seven-point scale ranging from 1 “far left” to 7 “far right”. Although not ideal, Wikipedia has been shown to provide quality factual information when it comes to electoral results and party competition (Brown 2011). Indeed, as we discuss elsewhere (Nai 2020), however, our measure strongly correlates with three other measures that exist for subsamples of candidates only – respectively, the measures included in the Chapel Hill Expert Survey (Polk et al. 2017), in the study by Benoit and Laver (2007), and in the Manifesto Project Dataset (Volkens et al. 2016). Table A2 in the Appendix indicates the ideology on this seven-point scale for each candidate in our database. The left-right variable is, then, folded on itself to create the “extremism” variable, which takes the value 0 for low extremism (this includes candidates from center-left to center-right), 1 for moderate extremism (left and right candidates) and 2 for high extremism (far left and far right candidates). All models also control for age and gender at the candidate level and for the number of competing candidates, type of election, and geographical region⁷ at the contextual level. We measured competitiveness of the election via a question in the expert survey that asked experts to evaluate how much they agree that “the race was not competitive, the winner was clearly known beforehand” (from 0 “very low competitiveness” to 4 “very high competitiveness”). We used the formula proposed by Laakso and Taagepera (1979) for the effective number of parties (ENPP) to measure the total effective number of candidates; this measure takes into account the differences in candidates support and yields a number to be interpreted as the number of competing candidates with a similar strength. We then used a binary variable that sorts countries with a PR electoral system (including Mixed Member Proportional, value 1) from countries with a plurality/majority system (including Mixed Member Majoritarian, value 0; see Gallagher 2014), and simple binary variable to distinguish between presidential (2) and legislative (1) elections. Descriptive statistics for all variables are presented in Table 2.

Table 2.

Descriptive statistics (missing values excluded).

Level	Variable	N	Mean	Std. dev.	Min	Max
Candidate	Fear appeals**	636	5.14	1.83	0.50	9.77
	Enthusiasm appeals**	636	4.44	1.57	0.33	9.00
	Campaign Loathing**	633	0.00	1.65	−4.84	4.15
	Incumbent	636	0.17	0.38	0.00	1.00
	Left-right position	636	4.17	1.56	1.00	7.00
	Extremism	636	0.47	0.68	0.00	2.00
	Female	636	0.14	0.35	0.00	1.00
	Age	636	54.89	11.58	23.00	93.00
Country and election	Election competitiveness	133	2.17	1.05	0.00	4.00
	Electoral system: PR	133	0.62	0.49	0.00	1.00
	Effective number of candidates	133	4.11	1.93	1.02	12.71
	Presidential election	133	1.41	0.49	1.00	2.00

^**Dependent variable

4. Results

4.1. Fear and enthusiasm appeals

To what extent are the differences in the use of fear and enthusiasm appeals driven by the profile of the candidates? To test for this overarching question we regressed the “emotions” scores on their profile and characteristics of the context via two linear hierarchical regressions where candidates are nested within elections. Results are presented in Table 3, respectively, for the use of fear (M1) and enthusiasm appeals (M2).

Table 3.

Determinants of fear and enthusiasm appeals.

	Fear			Enthusiasm
	M1			M2
	Coef	Sig	Se	Coef	Sig	Se
Incumbent	−0.06		(0.16)	1.50	***	(0.13)
Left-right position	0.30	***	(0.04)	−0.14	***	(0.03)
Extremism	0.83	***	(0.09)	−0.70	***	(0.08)
Female	−0.07		(0.18)	0.04		(0.15)
Age	0.01		(0.01)	−0.01	*	(0.00)
Competitiveness	0.26	**	(0.09)	−0.04		(0.08)
PR	−0.50	*	(0.21)	0.23		(0.18)
Effective N of candidates	−0.07		(0.05)	0.05		(0.04)
Presidential election	−0.37	†	(0.20)	0.33	†	(0.17)
MENAa	1.02	***	(0.30)	0.02		(0.26)
Sub-Saharan Africa	0.49		(0.32)	0.57	*	(0.28)
Latin America & Caribbean	1.32	***	(0.34)	0.42		(0.29)
Central & South Asia	0.38		(0.48)	0.45		(0.41)
East & SE Asia	0.90	**	(0.33)	0.64	*	(0.28)
Eastern Europe	1.01	***	(0.28)	−0.15		(0.24)
Southern Europe	0.98	***	(0.27)	−0.02		(0.23)
Constant	3.05	***	(0.59)	4.77	***	(0.50)
N (candidates)	636	–	–	636	–	–
N (elections)	133	–	–	133	–	–
R2	0.271	–	–	0.303	–	–

Note: All models are random-effect hierarchical linear regressions (HLM) where candidates are nested within elections. Models run only on candidates evaluated by 3 experts or more. Dependent variables vary between 0 “very low use” and 10 “very high use”.

^aFor all regions the reference category is “The West” (Western and Northern Europe plus the USA, Canada, Australia, and New Zealand).

^***p < .001;

**p < .01;

*p < .05, and;

†p < .1.

Comparison between the two models suggests that candidates benefitting from a comparative advantage tend to rely on positive emotions, as incumbents are significantly more likely to use enthusiasm appeals. Candidates having a comparative advantage over the rivals know that this can be sustained via campaigns that drum up the enthusiasm of the crowd. We do not find however support for the inversed mechanism for fear appeals. Our first hypothesis (H1) is thus only partially confirmed, for enthusiasm but not for fear. Models M1 and M2 confirm however the expectation that more extreme candidates – perhaps because less sensitive to potential backlash effects – are more likely to use fear appeals and less likely to use enthusiasm appeals (H2). Similarly, we find support for our third hypothesis (H3), according to which fear appeals are more likely among right-wing candidates – and, more specifically, are a function of an increasingly right-wing ideology; the reverse is true for enthusiasm appeals.

Turning to the contextual drivers, our models show that fear appeals are more likely during competitive elections, which is in line with our expectation (H4), according to which competitive elections are likely to incite the use of campaigning strategies able to persuade the masses and capture the last ambivalent or unaligned voters – in our case, the use of messages designed to incite fear and anxiety in the public. We do not find a similar effect for enthusiasm; we expected competitive elections to provide incentives for candidates to use appeals intended to mobilize their partisans to make sure that their support in principles is transformed into actual turnout; much evidence exists that enthusiasm appeals are particularly efficient to do so (e.g. Marcus, Neuman, and MacKuen 2000, Brader 2005). Our results do not confirm this trend. Broadly, our results suggest that competitive elections call for more “negative” forms of emotional campaigning, in line with research showing that competitiveness is associated with a greater use of political attacks and negative campaigning in general (e.g., Lau and Pomper 2004; Fowler, Franz, and Ridout 2016). The table also shows that campaigns in systems with PR are less likely to contain fear appeals, partially confirming H5.

4.2. Campaign loathing

Table 4 replicates the main models discussed above, but estimates the level of campaign “loathing” – the underlying campaign dimension that, broadly speaking, reflects high levels of fear appeals, low levels of enthusiasm appeals, and the use of more negative messages and character attacks (Gerstlé and Nai 2019).

Table 4.

Determinants of campaign loathing.

	Campaign loathing
	M1
	Coef	Sig	Se
Incumbent	−0.51	***	(0.15)
Left-right position	0.28	***	(0.04)
Extremism	0.77	***	(0.08)
Female	−0.06		(0.16)
Age	0.01		(0.01)
Competitiveness	0.15	*	(0.07)
PR	−0.50	**	(0.17)
Effective N of candidates	−0.06		(0.04)
Presidential election	−0.49	**	(0.16)
MENAa	1.02	***	(0.24)
Sub-Saharan Africa	0.53	*	(0.27)
Latin America & Caribbean	1.20	***	(0.28)
Central & South Asia	0.63		(0.40)
East & SE Asia	0.57	*	(0.27)
Eastern Europe	1.13	***	(0.23)
Southern Europe	1.20	***	(0.22)
Constant	−1.60	**	(0.49)
N (candidates)	633	–	–
N (elections)	133	–	–
R2	0.320	–	–

Note: All models are random-effect hierarchical linear regressions (HLM) where candidates are nested within elections. Models run only on candidates evaluated by 3 experts or more. Dependent variable varies approximately between −4.8 “very low loathing” and +4 “very high loathing”.

^aFor all regions the reference category is “The West” (Western and Northern Europe plus the USA, Canada, Australia, and New Zealand).

^***p < .001;

**p < .01;

*p < .05, and;

†p < .1.

The table suggests that harsher campaigns – that is, campaigns that score higher on the loathing dimension – are, ceteris paribus, less likely for candidates with a comparative advantage (incumbents and frontrunners), and more likely for candidates on the right and, especially, more extreme candidates. Election competitiveness increases the chances that candidates campaign in a harsher way, and elections under PR generally tend to be “gentler” (lower loathing), in line with what discussed in Valli and Nai (2020).

Taken together, these results show that the effects for the use of emotional appeals – fear and enthusiasm – are part of a broader set of dynamics at play in terms of campaign content, and that emotionality and campaign tone tend to go hand in hand (see also Crigler, Just, and Belt 2006).

4.3. Robustness checks

A series of robustness checks (Appendix C) replicate the main results discussed above. First, we replicated the models only on candidates for which at least 10 experts provided their ratings, which excludes 165 candidates and 42 elections; Table C1 and C2); results are largely in line with the ones in the main models.

Second, a frequent critique against expert surveys is that experts are ideologically biased (e.g., Curini 2010; Wright and Tomlinson 2018). With this critique in mind, we replicated all analyses controlling the models for the composition of the expert samples, in terms of the experts’ average left-right position, self-assessed familiarity with elections in the country, perceived simplicity of the questionnaire, percent domestic experts (that is, percent of experts living in the country where the election took place), and percent females (Tables C3 and C4). Results show some direct effects of the experts’ average profile; for instance, when the sample tends to the right on average candidates are more likely to be seen as having used enthusiasm appeals. More importantly, the main trends discussed above exist also when controlling for the composition of expert samples. This being said, these models also show significant interactions between the ideology of the candidate and the average ideology of the expert samples (see, e.g., models M2 and M4 in Table C3). For instance, as illustrated in Figure C1 (top panel), samples whose average expert ideology lean toward the right are less likely to believe that the campaign of right-wing candidates included many fear appeals; the inverse is at play for enthusiasm appeals (bottom panel). If the magnitude of these effects is not extremely high, it nonetheless suggests that we ought to proceed with caution when assessing the effects of candidate ideology. With this in mind, Table C5 replicates the main models for fear and enthusiasm but use instead “adjusted” dependent variables that eliminate all partisanship effects. Inspired by a procedure described in Walter and Van der Eijk (2019), we have computed adjusted measures by (1) regressing the value of fear and enthusiasm appeals on the difference between the average expert left-right position and the candidate’s ideological profile, and (2) keeping the regression residuals as adjusted measures of fear and enthusiasm that are independent of the effect of the (average) expert left-right position. Results in Table C5 show much weaker effects for candidates’ left-right position (due to the fact that the adjusted variables voluntarily tamper down any ideological effects), but show robust results for all other components, including extremism.

Third, for a subsample of countries (27 out of 101) more than one election is included in our dataset (see Table A2); to exclude potential confounding effects due to this clustering, Tables C6 and C7 replicate the main models but include only one election per country (the first chronologically); results are again robust.

Finally, Tables C8 and C9 include interaction effects between the candidate profile and the type of election (presidential or legislative); in general, there are not substantial differences between the two types of elections, even if winners and female candidates tend to use more enthusiasm appeals in presidential elections (Table C8, M2).

5. Discussion and conclusion

5.1. Results at a glance

Messages intended to stir emotions are a central component of modern communication. Anti-drug ads promote abstinence by comparing “your brain on drugs” with an egg in the frying pan. Doomsday electoral ads show little girls playing with daisies suddenly turned to ashes by a nuclear apocalypse. Pro-gun lobbies hype up patriotic enthusiasm with ads tapping into national pride stereotypes. And presidential candidates use campaign slogans suggesting that hope is worth living for because, yes, change is possible. Emotions are powerful drivers of social and political behaviors (Marcus 2000; Marcus, Neuman, and MacKuen 2000; Redlawsk 2006; MacKuen et al. 2010; Miller 2011) and is thus unsurprising that the effects of those emotional messages are well studied, both in politics and beyond (e.g., Brader 2005, 2006). This is however not the case for the other half of the equation: why, and under which circumstances, are emotional messages used in electoral campaigns? The evidence in this sense is surprisingly scant (for an exception, see Ridout and Searles 2011). Even more, comparative evidence – that is, outside the US case – is virtually inexistent.

We presented here the first systematic cross-national and large-scale comparative study about why candidates competing in elections worldwide used of two types of emotional messages – fear and enthusiasm appeals – during their campaigns. We did so by relying on an original comparative dataset (Nai 2020) that contains systematic information about campaigning strategies of 636 candidates having competed in 133 presidential and parliamentary elections in 101 countries worldwide between June 2016 and March 2020. We were able to point to several trends: incumbents tend to rely on enthusiasm appeals; fear is more likely to be used by more extreme and conservative candidates, whereas enthusiasm is more likely from the left and less extreme candidates. If results for left-right should be considered cautiously, as discussed above, those trends are broadly consistent with previous research on emotional campaigning (Ridout and Searles 2011; Wodak 2015). The comparative nature of the dataset also allowed us to test for the first time the extent to which the context plays a role. We expected effects for both the competitiveness of the race and the nature of the political system. Our results show that competitive elections foster the use of fear appeals, whereas elections fought in countries where PR is the norm are less likely to contain fear appeals.

5.2. Limitations

These results come of course with some caveats. First, some might feel that measuring the content of election campaigns via expert ratings is rather unorthodox. Although fairly common, expert ratings face still some skepticism in the literature; a common critique is that experts are unlikely to be completely objective when rating political dynamics (e.g., Curini 2010; Wright and Tomlinson 2018), which is an understandable critique in light of the rather liberal skew of academia (Maranto and Woessner 2012). Experts can be motivated reasoners as well, and have their own preferences. Robustness checks discussed above have indeed shown that, broadly speaking, liberal and conservative experts tend to partially disagree on the use of fear and enthusiasm campaigns by liberal and conservative candidates. Whether this ultimately suggests biased aggregated estimations is however another matter altogether. It goes without saying that expert ratings cannot reach the degree of sophistication and nuance that coding of textual materials can achieve. Nonetheless, as discussed earlier, experts have much to contribute when it comes to large-scale cross-sectional studies.

Second, due to the nature of the data used, our analyses provide a static image of campaign dynamics, a snapshot of election campaigns across actors and space. Yet, of course, campaigns are highly dynamic matters (e.g., Blackwell 2013), and much evidence exists that across adjust their campaigning strategies based on, e.g., the evolution of their competitive standing or the proximity of election day (e.g., Haynes and Rhine 1998; Nai and Martinez i Coma 2019). More research is needed in the future to assess to what extent the “static” results shown here, and especially the strong role of candidates’ ideology, matters from a dynamic standpoint.

Third, the trends described here are limited to two broad types of emotions – fear and enthusiasm. Although central in the contemporary political narrative and most modern campaigning efforts, they are by no means comprehensive of all the possible “emotional” strategies existing. Much evidence, for instance, exists about the different effects of anger from anxiety (e.g., Lerner et al. 2003); given the centrality of anger appeals in contemporary politics, our results clearly do not account for all emotional nuances of modern campaigning.

Finally, fourth, a point could be raised as to whether the dynamics described here are radically different from dynamics of campaign negativity, on which an abundant literature already exists. If it is true, as we have shown, that campaign emotions are associated with negativity, we believe that strong reasons exist to focus more specifically on campaign emotionality per se. On the one hand, theoretically, if fearmongering and negativity are rather proximate (but clearly not overlapping) concepts, the same cannot be said for enthusiasm and positive campaigns, which represent very different phenomena. On the other hand, empirically, some results strongly suggest that emotions – even negative ones – and campaign negativity face different dynamics. For instance, in our results the incumbency status of the candidate is not associated with the use of fear appeals, whereas it represents one of the strongest drivers of campaign negativity (e.g., Nai 2020).

5.3. Implications

Too little is still known about to what extent the profile of candidates and the contextual constraints drives the content of election campaigns, and more research is needed toward a general model of emotional campaigning and toward a broader understanding of comparative political communication in general. The results presented in this article have important implications in this sense. First, from a theoretical standpoint, the expand the existing evidence on the reasons why competing candidates decide to “go emotional.” Our results do confirm that some strategic considerations come into play – for instance in terms of competitive standing and incumbency status, in line with what discussed, e.g., in Ridout and Searles (2011) – but also show the importance of the ideological profile of the competing candidates, in terms of left-right orientation and extremism. Due to the potentially “risky” nature of negative and harsh campaigning strategies (e.g. Shapiro and Rieger 1992; Roese and Sande 1993), it is not surprising that much attention has been granted to the strategic components of campaign behavior. Our results suggest that it would be wrong to discount the identity of the candidates themselves, in line with recent research showing the centrality of the personal and attitudinal profile of candidates for their campaign choices (e.g., Nai and Maier 2020).

Second, from a methodological standpoint, our results contribute to the growing literature on measurement challenges in comparative political communication (de Vreese, Esser, and Hopmann 2017; Esser 2019). Especially in the fields of negative campaigning and the use of emotions in campaigns, comparative research remains elusive, mostly due to the great costs associated with the study of a discursive phenomenon across space and cultures. New advances in “big data” and machine learning approaches, able to digest increasingly large corpuses of textual data, are likely to reshuffle the cards in the near future (e.g., van Zoonen and van der Meer 2016). Yet, for the time being, no large-scale comparative dataset about the campaigning strategies of candidates worldwide exists. Our dataset fills an important gap in this sense and allows for new and detailed tests about the causes and consequences of political communication across space and – as the NEGex dataset expands and includes repeated elections in given countries – across time.

Third, from the practitioners’ standpoint, our results are of acute relevance for campaign managers and other spin doctors looking for ways to enhance the profile of their candidates, but also to predict the behavior of their opponents. As we showed elsewhere (Gerstlé and Nai 2019), the use of emotional appeals is indeed likely to matter for electoral results, as both fear and enthusiasm appeals can, under certain circumstances, drive a better electoral performance. Good reasons exist furthermore to believe that candidates (or their consultants) and news media are into a mutually beneficial relationship when it comes to the content of electoral campaigns. We have witnessed in recent decades a shift toward “infotainment journalism” (Brants 1998) with an increasing preference for “hype” over substance (Fox, Angelini, and Goble 2005). The use of emotional messages is “consistent with the media’s preference for drama and excitement in news reporting” (Jerit 2004, 563), and candidates have then strong incentives to “go emotional” to respond to the media imperative of framing news stories in an entertaining way – and are likely to be rewarded with a greater coverage (Maier and Nai 2020).

Funding Statement

A. Nai author acknowledges the support of the Swiss National Science Foundation (Grant ref. P300P1_161163).