Populism and emotions: a comparative study using Machine Learning

Abstract This study aims to unpack the mobilization of emotions in the political discourse of populist and non-populist parties and above all, across ‘varieties of populism’ (right wing vs. left wing or hybrid). Is there an empirical connection between emotions and populism? Are all types of populisms alike with regards to the emotional appeals within their political discourse? Focusing on Italy as a crucial case for populist communication and using a novel methodological approach based on supervised machine learning, it systematically investigates the intensity and trends of specific emotions in political discourses (institutional and informal, i.e. leaders’ speeches) of all Italian political parties over the last 20 years, for a corpus of more than 13,000 sentences analysed. The findings confirm that (i) populists tend to use more (and a broader repertoire of) emotional appeals than non-populist parties; however (ii) overall, there is an increase in the use of these appeals in the Italian political party discourse over time, especially in terms of negative emotions; and, most importantly, (iii) different types of emotions are mobilized by different types of populisms. Right wing populism mainly uses negative emotions while left wing or hybrid populism employs positive emotional appeals. The communication arena (party manifestoes vs. speeches) nevertheless does matter in the degree and types of emotions mobilized by political actors. This study identifies important implications for research on emotional appeals in politics, populist communication and political campaigning, and populist contagion from an emotion-based perspective.


Introduction
This study aims to explore the presence and types of emotions in the populist political discourse, and above all, across the 'varieties of populisms': right wing, left wing or hybrid (Zulianello, 2020).Often in academia and beyond, it is noted that there is a special connection between emotions and populism (Nguyen, 2019), and that populist parties and leaders have an additional emotional ingredient among their political communication tools (Canovan, 1999) and that emotional appeals are a recurrent characteristic of populist speeches, language, and style (Alvares and Dahlgren, 2016;Bossetta, 2017;Ernst et al., 2019aErnst et al., , 2019b)).These appeals are one of the essential drivers of populism (Kenny, 2017) an aid in 'framing injustice' and in creating a populist collective identity (Cossarini and Vallespin, 2019).Many scholars have also credited emotions for the electoral success of populists.This specific type of communication is characterized by 'emotion-eliciting appeals rather than rational argument' (Hameleers et al., 2016;Wirz, 2018).Indeed, studies have shown the role of emotions in increasing 'populist attitudes' of citizens and in favouring the propensity for the vote to populist parties (Rico et al., 2017;Vasilopoulos et al., 2018Vasilopoulos et al., , 2019)).
However, compared to other subfields of populism research (e.g.definitions, causes, measurements, etc., for an overview, see Manucci, 2022), the presumed link between populism and emotional communication still merits attention, especially for what concerns a 'systematic empirical scrutiny' (Widmann, 2021).Moreover, comparative studies across varieties of populisms and different types of emotions remain scarce and are often conducted more in the normative-theoretical than empirical realm (see the theoretical framework by Salmeila and Von Scheve, 2018).
This study aims to unpack the role of emotions in the political discourse of populist and nonpopulist parties and above all, across 'varieties of populism' (right wing vs. left wing or hybrid).Is there an empirical connection between emotions and populist parties' discourse?Have nonpopulist parties increasingly adopted the populist 'emotional appeal' style of communication?Are different types of emotional appeals, generally used by different types of populisms?
Focusing on Italy as a 'crucial' case study for the analysis of populism, due to its endurance in the political party system and its varieties, and using a novel methodological approach based on supervised machine learning 1 , the study systematically investigates the presence, intensity and trends of different types of emotions (i.e.anger, distrust, hate against outgroups, hope, joy) in the discourses of populist vs. non-populist political parties over the last 20 years in Italian politics (from 2000 to 2020).More in particular, analysing the political discourse (institutional vs. more informal, i.e. electoral manifestoes vs. leaders' speeches) of all the Italian political parties (with a corpus of more than 13,000 sentences analysed), and combining quantitative and qualitative data, it shows the similarities and differences among various populists 2 , and among populists and nonpopulist parties, in terms of the use of various types of emotions in their rhetoric as well as different communication arenas.
We start from the assumption that there is an increasing strand of literature which addresses populism as a political communication style that also takes emotions into account ( just to mention a few, Crabtree et al., 2018;Nai, 2021;Widmann, 2021).Moreover, some of these studies also focus on the Italian case (e.g.Bracciale et al., 2021;Martella and Bracciale, 2022).However, they mainly focus on the social media communication (i.e.Facebook and Twitter) of a limited number of Italian leaders, during specific electoral campaigns (i.e.2018).Our study, which is an attempt to translate the Salmela and von Scheve conceptual framework into a machine learning analysis, is positioned within this growing field of research, building on it and attempting to offer an empirical and methodological contribution to it.Previous studies (e.g.Ernst et al., 2019aErnst et al., , 2019b;;Bracciale et al., 2021) stress that populists use emotional language by expressing 'discrete' positive (e.g., happiness, contentment, hope, pride, trust) or negative (e.g., anger, hate, uneasiness, sadness, fear, regret, affection) emotions.
Disentangling varying types of emotions predominant in the political discourse of a specific type of party or 'party families', such as populists, or varieties of populism, may be important given that emotions have been found to be 'key strategic styles' for attracting voters attention 1 With this approach we position our study within the strand of literature using automated text analysis to investigate sentiments and emotions in political communication.Unlike other valuable contributions (e.g.Haselmayer and Jenny, 2017;Rauh, 2018), our approach does not rely on dictionaries but on a broader emotional coding. 2 (ibid., Boussalis and Coan, 2021 3 ).Moreover, looking at the populist phenomenon in terms of its communication style with a focus on emotions seems particularly relevant as it may affect the overall political communication of a political system (ibid., Ernst et al., 2019aErnst et al., , 2019b)).For this reason, our cross time study, over a large timespan, on a 'most likely' case study for populism and populist political communication, may thus have important implications for research on emotional appeals in politics, populist communication and political campaigning, as well as studies on the 'populist contagion' (Widmann, 2021;11).
Following the literature review on populism and emotions (section 2) and a section on data and method (3), we turn to the discussion building on the empirical analysis (4, 5).In the conclusion, we problematize our findings to future research (6).We argue that not only are Italian populists consistently found to be more 'emotional' than non-populists, but also that the prominence of emotions in the overall Italian political party discourse has increased in recent years and, significantly, that the types of emotions mobilized (both positive and negative) vary considerably depending on the party ideological alignment (and the communication arena).
Theory and analytical guidelines: populism/populisms and varieties of emotions First, both populism and the role of affective politics are two prominent topics in the academic debate in recent decades (Neuman et al., 2007;Hameleers et al., 2016;Arias-Maldonado, 2017).Emotions are considered to play a role in defining and influencing cognitive processes and value judgements on politics, including citizens' electoral preferences (Bakker et al., 2021).They also have a role in mobilization, recruitment and collective political actors' endurance (e.g. on social movements, see Jasper 2011), as well as in social and political in group/outgroup identification (Sindic and Condor, 2014), an aspect particularly resonant in populism.
Second, those studies looking at the nexus between populism and emotions have underlined the role of emotions in boosting populist attitudes and an increasing propensity to vote for these parties has been emphasized (Bakker et al., 2016;Rico et al., 2017;Vasilopoulos et al., 2018;2019).Emotions, and the emotive content of political campaign messages (Utych, 2018) may influence the political behaviour (e.g.Vasilopoulos et al. (2019) on the role of fear and anger on the vote for the radical right); as well as citizens' attitudes and opinions (e.g.Vasilopoulou and Wagner (2017) on fear, anger and enthusiasm on attitudes towards European integration), since they also affect how people process information (e.g.Kuhne and Schemer (2015) on the role of fear for new voters for the populist radical right).For example, Widmann (2021), on the basis of press releases and twitter communications of all the major parties in Austria, Germany and Switzerland (from 2016 to 2018) showed that populists use significantly more negative emotional appeals than non-populist parties.Similarly, Crabtree et al. (2018), focusing on election manifestoes from the 1980s up to 2012 in various European countries, including Italy, explain the type of electoral campaign run by political parties, showing that the level of positive sentiment used depends on their incumbency status, their policy position, and the objective economic conditions (for a similar study, see Kosmidis et al., 2019).Finally, Nguyen (2019), comparing panel data from Austria and Germany, investigates the relationship between anger and anxiety on the one hand, and support for populist parties on the other; and Wirz (2018) focusing on German-speaking regions in Europe, shows that populist appeals do indeed elicit emotions and that this increases the persuasiveness of the appeals.Most of these studies offer valuable insights when examining the Italian case.We build on and update them, and moving beyond the focus of only one type of populism (i.e.right wing), while not focusing selectively on some electoral campaigns or on the Internet sphere.
Studies have also centred on the association of specific types of emotions with different political ideologies, such as 'conservative or liberal thinking' (Taber and Young, 2013).In this regard, fear, nostalgia5 and anger, are considered essential components in reactionary views and anti-liberal political parties (Hochschild, 2018), but even (especially anger) in radical left wing politics (Cossarini and Vallespín, 2019;Magni, 2017).Anger, has been connected to insecurity and seen as a reaction to citizens' vulnerabilities (Salmeila and von Scheve, 2018).Those populist messages provoking anger aim to 'deflect shame-induced anger and hatred away from the self and instead toward the political and cultural establishment and various 'Others'' (Salmela and von Scheve, 2018: 443).Insecurities or resentment, understood as complex affective states that intertwine in the web of past, present and future (Bonansinga, 2020), have been linked to feelings of social disintegration, economic insecurity, relative deprivation and powerlessness (Abts et al., 2019) and considered a strong predictor of (right wing) ethno-populism, anti-immigrant attitudes and welfare chauvinism (Abts and Kochuyt, 2013), even in times of economic prosperity (Mols and Jetten, 2016).However, through the use of anger, populists might also tap into discourses on morality and injustice (e.g. against neo-liberalism), as many left wing populist leaders do (Rico et al., 2017).
So-called positive emotions, like pride and joy, are often assigned by populists to the notion of honesty of the ordinary man (Bonansinga, 2020).The populist left would especially relay on them in its political communication, due to its ideal of participation from below and collaborative sentiments which enable the transformation of shame into high-energy, active emotions, such as also indignation and anger (ibid., 446).
While repressed shame, resentment, anger and hate targeted at other generic groups would lead to, and be mobilized mainly by right wing populism; acknowledged shame, indignation, but also joy and pride, would be favoured and mobilized more typically by the populist left (Salmeila and Von Scheve, 2018: 449).This differentiation has also been applied to understand the emergence and success of different types of populisms in structurally and socially similar conditions (e.g. of modernization, globalization, etc.) related to crises (Brubaker, 2017).In this regard, a concept elaborated, though not tested empirically, is the 'emotional opportunity structure' (Salmeila and Von Scheve, 2018: 438), which shows how structural factors may interact with (individual and collective) emotions to set the terrain for populist entrepreneurs, who, in turn, often exploit such opportunities for their political gain.Indeed, conditions refer not only to structural factors, but rather are 'constituted and maintained by social and cultural processes that render (certain) emotions more visible, desirable, and acceptable than others ' (ibid., 439).This analytical lens seems particularly fruitful in investigating the Italian case, as its (similar) contextual 'emotional opportunities' can be exploited in different ways by different varieties of populism.In this study, we look at the actors who may connect the macro level causes with the macro level outcomes.
Against the background of this scholarship in this study we will focus on five main ('positive' and 'negative') emotions (see also Bracciale et al., 2021;Nai, 2021) generally attributed to populism: fear and insecurity; anger; hate against outgroup (i.e.generic others) 6 ; indignation against neoliberalism; joy and pride (for the definition and operationalization of the specific emotions we analyse, see the method section and table A in the appendix), elaborating the following assumptions: i. that the populist Italian political discourse is more emotional than the nonpopulist one (hyp.1);ii. that all in all, Italian political parties use of emotions in their political discourse will increase over time (hyp.2, or the 'populistisation' of the political discourse with regards to emotions); and, finally, that, iii.different types of populist parties (i.e.right wing vs left wing/hybrid) will tend to mobilize different types of emotions in their political discourse (hyp.3).
Some of these assumptions are not new (i.e.see the reflections on the difference between Widmann's results and the Italian context, e.g.Bracciale et al., 2021;Martella and Bracciale, 2022), however we expanded on them by focusing on a broad(er) timeframe, on varieties of populists and communication arenas (institutional vs. informal).In particular, while our first and third expectations are based on insights derived from the literature review illustrated above, the second is a broader interpretation of the so-called 'contagion effect' of populismin terms of emotions.Populist parties are considered to have a contagion effect on the mainstream parties, polarizing or radicalizing the content of the political debate (Iyengar et al., 2019), beyond increasing the salience of specific issues (Akkerman, 2015), so that mainstream parties would react to the success of populist parties by, among other things, adopting a populist rhetoric (Schwörer, 2021).Finally, although this is an exploratory research effort, we are also guided by the hypothesis that we should expect populist features to grow (i.e. the 'emotional appeal style') due to the presencesince 2008of an 'exogenous shock' such as the financial and economic crisis.Moreover, in this regard (and beyond the general trend), as emotions can be influenced by the (cultural) context (Lim, 2016), Italy seems to provide a perfect setting to clarify, beyond the context (i.e.'ceteris paribus'), specificities (in the 'agency') of varieties of populisms.

Method and sources
Italy appears as a paradigmatic case to such a study on populist communication, as it offers a great variety of populisms throughout its history -(i.e.left wing and right wing, even centrist; 'fourth types', Biancalana, 2020;Verzichelli and Castaldo, 2020) and has been considered a 'laboratory' and 'showcase' of populism, for its success and endurance (Albertazzi et al., 2018;Hamdaoui, 2021)  7 .
To investigate the presence and types of ('complex', although discrete) emotions mobilized in the political discourse by populist actors in Italy, we analysed the national party manifestos and leaders' speeches of all the political parties (populist and non-populist) in the country (for details on the parties included in the analysis, see table B and B2 in the appendix), across the leftright wing ideological spectrum, over the last two decades (2000-2020)8 .We took into account both the institutional (i.e.manifestoes) and more informal political communication (i.e.leaders' 6 Also Wirz (2018), stresses that 'an important element of populist anger appeals is the attribution of blame to the elite or out-groups, as anger is elicited when an adverse situation is caused by others (Ellsworth and Smith, 1988) '. 7 Even the mainstream, center-left Partito Democratico (PD) had a populist phase at the time Matteo Renzi was Prime Minister and Secretary of the party (Hamdaoui, 2021).More recently, in the 2018 national elections, the 5SMafter its astonishing electoral debut in 2013 (winning 27% of electoral votes)obtained 32% of the votes.In the same year, the right wing League (formerly the 'Northern League'), refashioned in its core-ideology and leadership, achieved an unprecedented 17% (and 34% in the 2019 European elections) (Biorcio and Natale, 2018).Together, they formed an 'all populist government', the choice of populism for Italian politics.Since the fall of that coalition and the subsequent alliance between 5SM and the PD, the 5SM has been facing a sharp and steady electoral decline while Matteo Salvini, the leader of the Lega, has been the main figure of populist politics in Italy, up until the elections in September 2022.speeches during electoral campaigns, party rallies and congresses, talk shows) 9 .Our unit of analysis is the sentence 10 comprising a final corpus of 13,530.
The sentiment analysis approach usually relies on dictionary-based approaches (e.g.Widmann, 2021) 11 .However, this approach may be insufficient to intercept complex emotions, like those in which we are interested and that can be considered the 'features of the populist communication styles' (Ernst et al., 2019a)  12 .
In supervised machine learning, still scarcely applied to measure aspects of populism (see Di Cocco and Monechi, 2022), the researcher in contrast 'trains' an algorithm, simulating coders' activity (for more information on this 'training', see appendix table C) 13 .In order to train the algorithm, we derived emotional scores per sentence using a Random Forest (RF) classification algorithm (Breiman, 2001)  14 , adopting a bag-of-words approach after text pre-processing.In terms of our procedure, first of all, we went through a manual coding process involving 15% of the Italian corpus (totalling 1917 sentences), randomly selected, in which the presence of an emotion was assessed and a type attributed 15 .
Specifically, based on the extant literature, we focused on five broad emotions particularly relevant to populism (see table A in the appendix): a. 'hate against outgroup (i.e.generic others)', mainly referred to the hostility towards immigrants, ethnic and social minorities (e.g.homosexuals, Roma people, Muslims); b.'anger' towards political opponents, judicial and intellectual elites, or the 'others' in the Manichaean perspective of the society; c. 'fear and insecurity', referred to uncertainty, dissatisfaction and insecurity with the economic dimension and wellbeing; d. 'joy and pride', which include any emotional feelings connected with pride and cheerfulness regarding political topics or achievements and the Italian political situation; and, finally, e. 'indignation and victims of neoliberalism', which refers in particular to the feeling of losing out as a result of globalization and neoliberalism, but also to the opposition of these economic processes.
Secondly, on the remaining part of our dataset we implemented the automated coding process based on machine learning.We trained five different models, one for each emotion.The goal was to obtain classifiers to assign class labels to chunks of texts where different classes represent different emotions.We used 70% of the corpus sentences for training the model, 20% for validating 9 The leaders' speeches were retrieved though data mining, by using key words (e.g.'name of the leader' + 'election year'-only the first page of each Google/You Tube search in our time frame-2006/2018 was analysed assuming that videos exhibiting higher viewing rates have also broader public coverage and greater importance for all party supporters).They mainly consisted of: political speeches during party rallies; political meetings; electoral campaigns; speeches to the nation and TV talk shows.Each speech was manually transcribed and classified according to the party and year.We did not include speeches drawn from social media (e.g.Twitter, Facebook) or party blogs having only become more popular in recent years, and not all leaders or parties rely on them extensively for their political communication. 10 Differently than other studies measuring populism (e.g.Hawkins, 2018), our unit of analysis is not the speeches, rather the sentences.This enabled us to have more units of observation, even when analysing a fewer number of speeches per leader (e.g. 10 on average), offering to the training a significant number of units of analysis as well as covering, probabilistically, a broader spectrum of emotions.

11
The idea underpinning dictionary-based sentiment analysis is that the sentiment of a text is determined by the sentiment of the words in that text.However, the presence of a combination of words can convey specific emotions when it recurs in populist textssomething that supervised machine learning can grasp.12 E.g. a specific kind of hostility that addresses outgroup members like minorities or migrants; not merely fear, but fear and insecurity related to economic regression or dissatisfaction.Not just 'anger', but anger against specific categories of enemies typical of populism, including, above all, political opponents and elites (vertical dimension); not only in happiness, but in the joy and pride of belonging to a specific political and social community (the 'Us' versus 'Them' typical of the populist Manichaean vision of society) 13 RF also gives the advantage of accurate predictions in the case of non-linear relationships.This feature supported its use for predictions within many topics, e.g.voting behaviour, partisanship and political sentiments, parties' populism (Ansari et al., 2020).

14
We have also tested an alternative algorithm, namely, the Naïve Bayes, obtaining the same results.and tuning the model parameters, and the remaining 10% to build the score and test its predictive power.The models predict whether a sentence could be classified as including that specific emotion (Y = 1) or not (Y = 0).We finally obtained our emotional scores measuring the probability that a specific emotion is found over the years.To validate the results, we measured the accuracy on the validation test (see table E and F in the appendix).For all the types of emotions we obtained good average precision scores and AuROCs, considering the complexity of the emotions classified and the standard positive-neutral-negative classification-which makes our approach particularly effective (performing better than existing alternatives to measure emotions and populism) 16 .
Empirical analysis: 'emotional populism', emotional times When looking at the presence and intensity of emotions (i.e.aggregate emotional level) in the political discourse of Italian political parties over the last two decades, our data suggest that, overall, the populist political communication is more emotional than the non-populist.(Figure 1a and b).
Populist parties (and leaders) generally make more frequent (and broader) use of emotions than non-populists, especially since the 2008 economic crisis 17 .This trend is constant over time, with minimal fluctuations and it also holds true when distinguishing among arenas of communication (electoral manifestoes vs. speeches)although, on the whole, the informal arena of speeches is characterized by higher emotional levels for both the populist and non-populist parties.However, as a comparison, non-populist Italian parties have generally increased their use of emotional tones over time in their electoral manifestoes (especially since 2008), almost on a par with the populist levels by the end of our timeframe 18 .These findings suggest that studies on emotions and political discourse should be based on different sources, to better intercept variations of the political style and to be in line with current research on the populist communication and social media, which emphasizes how informal arenas of communication are particularly conducive in the mobilization of emotions (Aalberg et al., 2017).
Moreover, looking at Figure 2, which shows the presence and levels of specific emotions in the populist and non-populist political discourse, we see that, for both actors there is an increase in the use of all types of emotions analysed over the years (i.e.hate against outgroup, anger, fear and insecurity, joy and pride, indignation).
The levels of emotions are, overall, higher between 2016-2020 than at the beginning of our analysis (2000) 19 .However, since the majority of the emotions we analysed are negative, this result seems to corroborate the relationship between populism and polarization of the political debate.
In particular, hate (i.e.hostility against various types of 'outgroups', such as social and ethnic minorities, immigrants, gypsies, homosexuals) 20 increases over the years, although with 16 For instance, in a paper classifying sentiments in positive-neutral-negative, the accuracy of measurements is around 75% (Bahrawi, 2019).17 The overall emotional score for all the years analysed is 0.102 for Italian populist parties vs. 0.050 for the non-populists', which means that political discourses containing one of our emotions analysed are double (10% vs. 5%) for populist vs. nonpopulist political communication.In terms of 'style of communication' the party Foza Italia obtains a low score in 2004 and 2009, for what concerns the use of emotions (respectively 0.5 and 1 on a scale of measurements 1-4, on the basis of indicators including resentment; direct and non-institutional language; strong emotions such as fear and enthusiasm) (Caiani and Graziano, 2016: 260).The Lega can be considered as a populist party not before 2008under the leadership of Salvini (Albertazzi et al., 2018).

18
The emotional score in electoral manifestoes during the years 2016-2020 is 0.115 for populists (i.e.11.5% of the sentences analysed) vs. 0.114 (11.4%) for non-populists (whereas differences between the two were greater for the speeches over the same period.i.e. 18.5% for populists vs. 12.4% for not populists).
19 Accounting for 12.2% (score 0.122) of 'emotional' sentences out of all sentences in the period 2016-20 for populists and 11.9% for non-populists.

20
This emotion has been much stronger among the populists since 2008, when it accounted for more than 15% of their codified sentences.fluctuations (after a peak in 2008 it slightly decreases, at first, for both types of actors before increasing again) always remaining higher in the populist political discourse.On the other hand, anger against other generic groups 21 (i.e.mainly political elites and adversaries) increases over the years for all actors, but particularly since 2008 in the populist political discourseremaining constantly higher than among the non-populists 22 .Finally, fear and insecurity, which also appear more frequently mobilized in the discourse of the populist Italian parties23 , are significantly prevalent among the non-populist actors too in the most recent years.
In sum, our data point out that overall the Italian political parties' use of emotions increases over time and that the discourse of populist parties seems to be characterized by some recurring emotions: they are in particular anger against political opponents and hostility towards minorities (hate).The new populist era, at least in the Italian case, seems to be mainly populated by negative emotions, especially among the populists.Moreover, this picture remains true when we differentiate by communication arenas (as figure A in the appendix shows).
With a closer look from our qualitative analysis, we can see that hate usually assumes an immigration-oriented dimension.For instance, in his speeches Salvini frequently reiterates the Note: The scores showed in the figures represent the percentages of 'emotional sentences' per year, and have to be considered as a proxy of a political actors' emotional level.For example, a score of 0.122 (i.e. the emotional score of populists in 2016-20 fig.1a) means that 12.2% of the codified sentences of the Italian populist political discourse in 2016-2020 contained emotions (at least one among the five broad emotions analysed in this study).

21
Present in populist parties in more than 19% of their codified sentences vs. about 10% of non-populists in the overall period aggregated.

22
In contrast, emotions such as fear and insecurity and indignation against neoliberalism (namely the opposition to neoliberalist policies and the idea that there are winners against losers from globalization), although generally on the increase over time, still remain higher for populists with some fluctuations.hostility towards migrants (e.g. the 'law and the control of borders', 'the expulsion of hundreds of thousands of illegal immigrants', 2018, speech) 24 , mobilizing this emotion by framing the political discourse around the idea that some outgroups receive more attention than neglected in-groups (i.e.deserving vs. undeserving people).He often stresses 'the unequal and unfair treatment of Italian citizens (election speeches, 2013, 2018), intercepting a resentment strongly connected to feelings of economic insecurity and powerlessness (Abts et al., 2019).However, the targeted outgroups have changed over time.For example, between 2006-2008, when the Northern League was a federalist and separatist political party, its speeches mainly focused on 'Padania's independence' and opposition to 'Southern Italian regions and people'.Conversely, in 2013, and even more in 2018, under Matteo Salvini's leadership, the enemies were no longer from the South but were now 'foreign people' ('concrete problems are work, food (…) not the discussions about the citizenship for immigrants' children'; 2013, TV speech). 25Indeed, during the 2018 electoral campaign, Salvini often stresses the need to 'campaign on truth, honesty and seriousness.No miraculous promises (…)', but instead towards 'a total control of borders, with mass expulsions for the hundreds of thousands of illegal immigrants in Italy' 26 .Note: In the figures, the label hate refers to 'hate against outgroup'; indignation' to 'indignation against neoliberalism'.The scores showed in the figures represent the percentages of 'emotional sentences' per year, and have to be considered as a proxy of a political actors' emotional level related to the specific emotion.https://youtu.be/_sCfwh7W1SM.The outgroup hostility emotion, however, is not only a prerogative of right-wing populists.For instance, Luigi Di Maio, a former leader of the Five star Movement, talking about immigration, stressed in some speeches that his party 'closes the ports to those who do not respect laws (…)', has the goal of having 'a bill on judicial cleansing at sea and those boats' ('let us take it straight to Parliament', 2018, source: https://youtu.be/Jiz0GpFK0KI), is against 'a government who authorised indiscriminate landings of immigrants in Italian ports for a few pennies' ('it is shameful!' 'they have transformed Italy into the biggest port in Europe', 2018, source: https://youtu.be/Jiz0GpFK0KI).
Anger often includes anger towards political opponents and the government.Critical examples of this emotion are the speeches of Beppe Grillo's 27 , in which he delegitimizes the Italian political class as 'the people who go there [in parliament] to fuck us again' (2013)'28 , who 'have occupied the Parliament all their lives' (2013) 29 .However, it also included frequent claims over the years referring to anger towards an ambiguous, not specified or easily identifiable 'Other' (e.g.'they'/ 'them'), typical of the populist 'conspiracy' discourse in various European countries (Pirro and Taggart, 2022).
Fear and insecurity target both the national and the European level of governance, and even the global levels, in the appeals of populist leaders.For example, Salvini often emphasizes that European treaties 'condemn Italy to hunger and poverty' and that the solution is 'to stop financing the EU', which is for Italians 'only a damage'30 , a 'sacrifice policies enriching few and ruining many'. 31Different leaders, however, tend to mobilize this emotion by making reference to different aspects of it.For example, sometimes (as seen in many speeches of the leader of Fratelli d'Italia party, Giorgia Meloninot classified as a populist party in this study) fear and insecurity are coupled with a feeling of 'political powerlessness' or 'inefficiency' (e.g.talking about Fornero laws, the 'dichotomy between first-class workers in Italy and second-class workers' is stressed, or a 'hyper-protected generation and one that will never be guaranteed', 2013)32 .At times, this emotional category includes more insecurity about economic instability (especially in left wing populist discourses).Less common, do we see fear and insecurity linked to feelings of nostalgia, where a rhetoric of fear-insecurity-evoking' is prevalent.
Claims conveying joy and pride are present in the Italian political debate too (although accounting for less than 10% of the total sentences of populists' communication).Examples are discourses which emphasize 'pride', 'enthusiasm', 'faith and values', '[our young people]', 'being passionate', 'courageous', 'grateful', 'excited', etc. (for examples see speeches of the radical right-wing politician Daniela Satanchè in which this emotion is quite present, e.g.200833 ), or the electoral campaign in 2013 of the Democratic party ('we are able to put on our shoulders a concrete alternative to the right!' speech of Bersani). 34Beppe Grillo and his party has appeared to mobilize this emotion more frequently, as in the case of one of his famous speeches where 'the discovery' of 'the path towards a better future' is stressed, which is described as 'honest, concrete, supportive and happy' (2013, speech). 35s for indignation related to neoliberalism (e.g.'victims of..'), Italian populist political parties and leaders seem to mobilize it mainly as a collective 'action frame', namely by a framing of the political discourse where the economic and political elites are blamed, and a call for political action (usually 'new', from the people) is madereferring to acknowledged shame and allowing for self-identifying as victims of neoliberalism ('this old man is ashamed, he feels like a failure for losing his job.…36 ').In the same vein, there was a strong emphasis placed on 'the effects of globalisation' on the Eurozone, in the 2018 5SM electoral manifesto, a process that 'escaped the initial good intentions, betraying its own founding values', which embeds 'clearly distorting elements of a globalising mechanism', 'widens the gap between institutions and citizens'.

Varieties of populism(s) and different types of emotions
Our data over 20 years of Italian political discourses, show that specific types of emotions which distinguish populist parties leaning towards different ideologies (left wing or hybrid vs. right wing37 ).Indeed, as Figure 3 shows (see also fig.B in the appendix), if we consider populists depending on their (hosting) ideology, outgroup hostility is mainly mobilized by right wing populist parties and leaders, in line with the literature on right-wing populism, which link its emergence and success to the cultural dimension crisis (Kriesi, 2018).
As we see in many discourses of Salvini addressing the EU this sentiment is frequently evoked, stressing that '[regarding immigration policies] another Europe is necessary which is independent from Brussels', 'crazy' and 'criminal' on the management of these issues, 2013, speech. 38nger is mainly used by left wing and hybrid populists, similarly to joy and pride.On the contrary, when looking at fear and insecurity, the picture appears less clear cut among different types of populisms.The usage of this emotion in political discourses seems divided equally among various actors: before 2012 it features mainly in right wing populism, and afterwards in left wing (and hybrid) populism.Looking more in depth at our qualitative data we see however that this emerging emotion does not often exclude a reference to anger against others (for example the 'political caste', as in many claims of the 5SM) or indignation against neoliberalism.As always, the empirical reality is more complex than any classification.For instance, during the 2013 national electoral campaign, the leader of the 5SM, Beppe Grillo often emphasized that 'banks cannot fail because it is an oxymoron and a lie, while thousands and hundreds of families fail with their Note: In the figures, the label hate refers to 'hate against outgroup'; indignation' to 'indignation against neoliberalism'.The Italian party 'Potere al Popolo' has been classified as Left wing populism (see also Di Cocco and Monechi, 2022); the party Lega as Right wing populism; and, finally, the 5SM party as hybrid (i.e.'other') (see also Mosca and Tronconi, 2019).The scores showed in the figures represent the percentages of emotional sentences per year, per emotion, per type of populism, and have to be considered as a proxy of a political actors' emotional level.
savings'. 39Similarly, Di Battista (5SM), talking about access to educational opportunities in the country, emphatically stresses the dimension of insecurity and fear for the lower classes. 40.
Finally, when considering feelings of indignation against neoliberalism, although this emotion is (in absolute terms) less mobilized than the others in the Italian political discourse, it is above all found in the discourse of left wing (and partly hybrid) populism, such as the 5SM and Potere al Popolosupporting the expectation in literature that left wing populists are more joyful than other parties along the ideological spectrum.For example, talking about the 2018 election polls, Carofalo (representative of the party Potere al Popolo) emphatically stresses that his party is 'building something huge today (…), with assemblies in 150 cities in Italy', and was 'optimistic' to enter the Parliament. 41Similarly, the 5SM political discourse is often characterized by the rhetoric of joyful 'rebellion', positive 'expectations', 'hope', as well as 'pride of being part of a sovereign people who can promote a change' in the society (e.g. the 'Let's take our lives together democratically, rebel', speech). 42n sum, as our qualitative analysis indicates, in the emotional appeals of right wing populists fear is mainly related to the discourse on powerlessness and déclassement, hate against outgroups while anger is directed against the corrupt political and cultural elites, failing to protect peoples' interest and favouring outgroups at the expense of the ingroup or against those who have a 'good life' without having to work hard (such as politicians and managers, welfare recipients and refugees, the long-term unemployed, as well as ethnic, cultural, and sexual minorities).On the other hand, in left wing populist discourses, anger is more often framed towards the domestic and foreign political establishment implementing austerity politics for instance or as being responsible for enforcing politics that increases injustice, inequality, and precariousness.

Conclusions
Since the 1990s, many social scientists have considered Italy as a fertile testing ground for populist communication or even a populist 'paradise' (Tarchi, 2015).In this article, we looked at this case in terms of emotions and populism, with a descriptive effort (across types of emotions, type of populisms, time and types of communication arenas) that seems to offer an empirical and methodological contribution to this field of enquiry.In sum: Why is populist communication intrinsically emotional?And what are we still lacking in terms of our theoretical understanding about the links between populism and emotions?Our study includes some of these answers in particular in terms of the different emotions used by different types of populisms, and the role of different platforms, or communication arenas, where the political debate takes place.
First, confirming our easiest hypothesis (n.1), our study showed that Italian populist parties, like pre-existing ones in other countries and on social media (Béland, 2020), adopt a more emotional political discourse than non-populists (consistently over time and regardless of the communication arena).This is however coupled with the evidence that, as similarly suggested by studies on the populism communication in social media (e.g.Engesser et al., 2017), the more informal arena of speeches seems more conducive, overall, to the use of emotions (see similarly Widmann (2021) on Italy regarding social media messages of political parties).
Second, if this is the characteristic of (Italian) emotional populism, confirming and integrating Bracciale et al. (2021) (who focused on Italian populist communication on social media), we also found that emotions are increasingly adopted by Italian political parties and leaders as communicative tools to command attention and gain consent.Another important finding of this study (confirming our hypothesis n.2) is indeed that non-populist Italian parties have also increased their emotional appeals over time.We can therefore argue that the (Italian) political field in general is becoming more emotional, with mainly populist political speeches increasing the general level of hostility against various relevant 'outgroups' in the Italian political discoursein line with the polarization of the political debate potential of populism (also in line with evidences on emotional rhetoric in German state Parliaments (Valentim and Widmann, 2023)).These findings might also align to previous studies which emphasized for the Italian case how the 'exogenous shock' of the 2008 financial and economic crisis, seems to have favoured the growth of populist features among the Italian political parties (Caiani and Graziano, 2016).
A third relevant finding is the broad correlation found between different types of emotions and types of populisms, in line with our third expectation.In particular (partly in contrast to Bracciale et al., 2021), we have found a strong association between negative emotions and right wing populism and positive emotions and left wing populists.These findings confirm the idea that politicians use different emotions systematically for different purposes (Widmann, 2021).However, beyond this general trend, our qualitative analysis also showed some nuances: Left wing populist parties may also mobilize negative emotions (e.g.fear or anger) but in a different way; joy and pride are also used with right wing populism.Indeed, as the qualitative analysis showed, anger is mainly characterized by anti-elitism, whereas fear and insecurity are often both linked to the socio-economic dimension and the cultural backlash.
Although limited in scope, these findings can also be read as complementary to previous work theorizing different causes behind different types of populism (Kriesi, 2018).If different crises (economic, political or cultural-backlash) are 'performed' (Moffitt, 2015) and amplified by (different types of) populisms (Pirro and Taggart, 2022), our findings may suggest that emotions can be understood as the 'mechanisms' (Jasper, 2011) linking macro causes to macro effects.
As for the Italian case, against the background of the same context of political and 'emotional opportunities', a different framing of different types of crises by varieties of populists can be helped by the different emotions mobilized.Understanding the link between populism and emotions could therefore provide not only important insights into the activation of populist attitudes (Hawkins et al., 2018), but also illuminate a crucial mechanism for the way in which populist parties (of different types) maintain support over time.
Beyond providing an updated trajectory of the phenomenon in Italy, our study could elucidate similar trends in other countries, as populist actors and leaders, as shown by previous work, tend to behave similarly with regards to political campaign styles and emotions in particular, across various regions of the World (Gerstlè and Nai, 2019).
Furthermore, methodologically, our original approach to emotion detection based on supervised machine learning, which combines the strength (and richness of nuances) of manual and automated method for text analysis typical of natural language processing, is a useful test exercise for future studies, as it can capture emotions related to populism in a systematic and accurate way.What this method is unable to do, and what in fact we did not do in this study was, beyond differentiating which party was more emotive-populist in which periods, was to explore the 'why'.Future qualitative and historically driven studies should explore, perhaps starting from our systematic empirical overview, how the populist emotive communication has 'travelled' and materialized historically in the defined time-loops over the last 20 years in Italy, and how it has been shaped by specific political trajectories of Italian politics.
Finally, at the end of the story, we can also re-read all these results under a different light.Moving from a radical democratic theory perspective, we could contend that by using a Laclauian framework which sees politics as equal to hegemony as equal to populism, one can conclude that populist actors are no different from other political actors: emotions and affects are always central to any political identity (Eklundh, 2019).In this regard, we could argue that the division between the emotional and rational in politics serves to sediment exclusionary practices against newcomers and challengers of the status quo (ibid.).

Figure 1 .
Figure 1.The overall emotional level (i.e.presence of emotions) in the populist vs. not -populist party political discourse in Italy (2000-2020).(a) aggregated.(b) by type of communication arena (electoral manifestoes vs. speeches).

27
See also 2013 UDC (Unione Democratici di Centro) election manifesto, stressing that the diseases of Italy are '… the selection of the ruling classes entrusted to random, oligarchic and masterly mechanisms; the worsening of the crisis between representation and territory' (UDC, 2013, https://manifesto-project.wzb.eu/datasets).