This is an open access article distributed under the terms of the Affective polarisation (AP) refers to a pattern of political attitudes resulting from the tendency to feel warmly (affinity) towards one’s own political camp and coldly (animosity) towards its opponents (Iyengar et al., 2012; McCoy et al., 2018). Beyond affect towards the parties as institutions, these sentiments can be further differentiated by measuring affect towards parties’ leaders (vertical AP) or voters (horizontal AP). These various dimensions are interrelated and may even reinforce one another (Reiljan et al., 2024), affecting the way citizens act and interact, potentially reinforcing political conflicts and influencing everyday social interactions (Iyengar et al., 2019; Torcal et al., 2025).
Despite affective polarisation becoming increasingly relevant in many democracies (Gidron et al., 2020), its emotional mechanisms remain underexplored, leading researchers to call for the inclusion of emotions to expand our understanding of their dynamics, patterns and consequences (Bakker & Lelkes, 2024; Huber et al., 2015; Huddy et al., 2015; Mason, 2016). Responding to this gap in the literature, this article focuses on discrete emotions towards political elites to deepen our understanding of how emotions relate to affective polarisation. Although recent research has explored emotions towards partisan groups and their impact on partisan affect (Berntzen et al., 2024; Kretchner et al., 2026; Renström et al., 2023), the role of emotions directed specifically at leaders has been insufficiently explored. Among potential emotional targets, political leaders are especially salient, eliciting strong emotional reactions (Hogg, 2001; McLaughlin et al., 2020; Ruben & Gigliotti, 2016), which could end up serving as proxies for how people feel about their partisan supporters.
In the following pages, I examine how five discrete emotions (hope, pride, anger, fear and disgust) towards leaders correlate with partisan AP (in-partisan affinity and out-partisan animosity). This approach seeks to fill in some of the gap surrounding the connection between the vertical and horizontal dimensions of AP. Following a set of hypotheses that assume emotions towards leaders generally align with broader evaluations of their party, I show that hope and pride towards one’s preferred leader are strongly correlated with liking in-partisans, while anger and disgust towards opponents are generally associated with disliking out-partisans. At the same time, I also control for the possibility of affectively incongruent emotional structures, such as negative emotions towards liked leaders and positive emotions towards disliked leaders. In addition, I control for individuals’ ideological extremity, partisan identification and political interest, since political engagement and motivated reasoning may influence partisan affect. Also, to test whether certain social groups are more prone to polarisation through exposure to inflammatory narratives I control for age, sex, education level and subjective wellbeing.
Although previous research has studied the effects of emotions towards political leaders on partisan affect in particular cases (Bakker & Lelkes, 2024; McLaughlin et al., 2020), I extend this analysis to five multiparty contexts in traditionally underexplored regions, across Latin America and Southern Europe (Argentina, Chile, Italy, Portugal and Spain), which vary in institutional structure, levels of polarisation and party competition, but at the same time show similar signs of partisan division. Cross-national survey data (TRI-POL; Torcal et al., 2023) allows us to observe patterns across varying contexts, enhancing the generalisability of findings while also highlighting contextual variation, showing that the relationship between emotions towards leaders and AP can take various forms depending on the emotional and political landscape. Overall, based on the distinct profiles of discrete emotions, findings may help frame why affective polarisation could be considered both a threat and an asset to democratic functioning.
Theoretical Framework
Affective polarisation (AP) may be defined as a set of positive feelings towards liked parties and negative feelings towards disliked parties (Iyengar et al., 2012). Empirically, affective polarisation seeks to capture a broad range of psychologically biased political perceptions and evaluations (including like-dislike, coldness-warmth, closeness-distance and broader emotional reactions to political actors and phenomena). However, it takes its name from a concept in psychological research (core affect), which refers to a basic feeling state structured along a limited number of dimensions, most notably valence (pleasant-unpleasant), and sometimes arousal (activated-subdued), similar to mood (Russell, 2009). Resulting from this conceptual linkage, AP research has tended to rely on valence-based measures.
There is an ongoing debate in AP surrounding whether these feelings are influenced by converging/diverging policy preferences or by partisan identities per se, as when attachments belie parties’ ideology and policies (Bankert & Stone, 2025; Dias & Lelkes, 2022; Huddy et al., 2015; Orr & Huber, 2020). While these represent analytically distinct explanations, they may be seen as layers of the same phenomenon, with partisan identities encompassing policy preferences and inferred norms and values. As a result, positive identification with parties generates positive affect towards the party as a group (in-party affinity), while identities constructed in opposition to disliked parties generate negative affect (out-party animosity) (Abramowitz & Webster, 2018; Russo & Areal, 2025). Social categorisation has been shown to generate differential treatment of individuals, even when group differences are arbitrary (Tajfel, 1970), producing affective experiences that reinforce positive beliefs about one’s own group and its distinctiveness with regard to others (Brewer & Brown, 1998). In political contexts, this categorisation often occurs along party lines, driving cognitive and affective evaluations of others based on their party tag (Marcus, 2000).
A key factor that determines the intensity of differential evaluations of others is the social identity category’s relevance (Maitner et al., 2010). Differential group treatment can take various forms: in-group favouritism may be equal to, higher, or lower than out-group derogation (Brewer, 2017; Brewer & Brown, 1998). Similarly, partisan identities in AP have different dimensions that need not be symmetrical: positive partisanship does not always generate negative partisanship, nor does negative partisanship require positive partisanship to occur (Bankert, 2022). However, competition over scarce resources (e.g. political power) emerges as a primary condition under which both in-group favouritism and out-group discrimination occur (Brewer & Brown, 1998). Indeed, perceptions of out-group threat increase under zero-sum circumstances, making group identities more salient and intensifying the associated affective responses (Ray et al., 2014; Renström et al., 2021). While differences in in-partisan affinity may depend on the strength of individuals’ party-group identity, differences in out-partisan affect could vary based on individuals’ perceptions of how competitive, effective and/or threatening the out-party is. Thus, while it is reasonable that both in-party affinities and out-party animosities are present to some extent in AP, it is important to keep these dimensions separate, lest we conflate (1) high in-party affinity and low out-party hostility, with (2) partisan indifference, or (3) strong out-party hatred in the absence of positive partisanship. Keeping these dimensions separate puts to the test the assumption that one form of bias necessarily begets the other(s).
In the context of AP, such feelings can be attributed to various political events, entities and identities (Roseman, 2025). Among these, leaders serve as important focal points that elicit clear and reportable emotional responses: political persons are salient and easily evaluated, and affective evaluations of them are stored and automatically retrieved upon exposure (Lodge & Taber, 2005; Roseman et al., 1996). Since leaders function as symbolic party prototypes (Bäck et al., 2023; Turner & Oakes, 1986), people often rely on their feelings towards them when forming broader political evaluations and interpreting political events (Glass & Glass, 2021; Ruben & Gigliotti, 2016). At the same time, these evaluations are not formed in isolation. Feelings towards leaders are influenced by their party membership (Jennings et al., 2024), suggesting that affective evaluations may reflect underlying partisan identities as much as they influence them. Politicians may thus condition and be conditioned by individuals’ evaluations of partisan groups (Bolsen & Thornton, 2021; Kingzette et al., 2021). Generally, partisan affiliation provides a heuristic enabling individuals to quickly evaluate and appraise political actors based on attributed party characteristics (Brady & Sniderman, 1985; Marcus, 2000; Nicholson et al., 2018; Orr & Huber, 2020).
Political Emotions
While AP research is generally becoming increasingly attentive to the complexity of affect, much of the literature still relies on valence as its primary affective indicator. This is understandable, given that positive and negative evaluations are intuitive for respondents and highly reproducible across survey research. While valence is arguably one of the most stable attributes of affective experience (Barrett, 2006), treating affect and valence as synonymous collapses affective states that may differ meaningfully (Redlawsk, 2023). Distinct emotions can share valence while diverging in their appraisals, action tendencies and implied orientations towards others. Thus, although measures such as the feeling thermometer remain highly effective and reliable indicators of AP (Gidron et al., 2022), they obscure variation in how political actors are experienced and how those experiences may relate to intergroup dynamics.
If there are dimensions of affect that valence alone cannot capture (Fontaine et al., 2007; Kasper et al., 2025; Russell, 1980), treating emotions as discrete states may be a practical way of recovering these dimensions. Although emotions are difficult to delimit with complete precision (Russell, 2009), political psychologists broadly converge in viewing them as multi-component response syndromes involving subjective feeling, appraisal (threat-benefit), bodily expression, and readiness to act in different ways (Marcus et al., 2000; Redlawsk, 2023; Roseman, 2025). In this sense, emotions are not simply stronger or weaker instances of positive or negative affect, but parsimoniously report on multiple dimensions of experience. Emotion System Theory (EST) captures this idea by conceptualising emotions as different strategies for coping with opportunities and threats, each associated with distinct patterns of appraisal and behavioural readiness (Roseman, 2025; Roseman et al., 1996). This makes discrete emotions particularly relevant for AP, where intergroup relations depend not only on whether evaluations are positive or negative, but also on how groups interact as a result of distinct emotional configurations.
It follows from this theoretical discussion that emotions attributed to (felt towards) political leaders should relate to AP in ways that broadly align with their (positive/negative) partisan identities. Emotional experiences that converge with partisan categorisations (positive emotions towards in-party leaders, negative emotions towards out-party leaders) are expected to correlate with higher partisan affective polarisation as a result of highlighting partisan group distinctions (Brewer & Brown, 1998). Conversely, when affective experiences conflict with prior classifications, individuals may re-evaluate membership within their partisan group (Lavine et al., 2012). However, because in-party affinity and out-party animosity are both conceptually and analytically distinct, these relationships need not be symmetrical: emotions may reinforce group boundaries by strengthening identification with the in-party, intensifying perceptions of intergroup threat, or increasing the salience of partisan identities. From this framework, I broadly expect emotions towards leaders to relate to partisan AP by mapping onto existing identity constructs and categories. Nevertheless, since emotions reflect different situational appraisals (threat-benefit; Roseman et al., 1996), their association with partisan affinities and animosities may vary.
Next, I discuss the five emotions analysed in this study: hope, pride, anger, fear and disgust. While other emotions may be relevant for distinct aspects relating to intergroup dynamics, these emotions are expected to constitute especially salient evaluations of political elites and to be closely associated with perceptions of and affect towards partisan voter groups.
Hope is a positive, future-oriented emotion associated with uncertainty about desirable outcomes (Lazarus, 1991; Roseman, 2025). Defined as a “feel-good, do-good” emotion, hope is associated with pro-social behaviour and increased willingness to engage in collective action (Cohen-Chen et al., 2020; Cohen-Chen & Van Zomeren, 2018). In political contexts, hope towards an in-leader may signal a commitment towards the goals of the in-party, suggesting an affective attachment with the in-party group.
H1: Hope towards an in-party leader is correlated with increased partisan AP.
Pride also has positively valence, but is instead associated with realised or attributed success, “ego-enhancement” and preferential in-group treatment (Lazarus, 1991). When compared with hope, pride typically signals a higher sense of agency (Roseman, 2025), and often involves affirming oneself or one’s group in order to benefit from improved social standing (Tracy et al., 2023). In political settings, feelings of pride will typically be experienced towards an in-party leader, reinforcing group distinctiveness and sharpening comparisons with out-groups, including the out-party.
H2: Pride towards an in-party leader is correlated with increased partisan AP.
Anger is a negative-valence, high-arousal emotion, typically elicited in response to perceived injustice, violation of norms, or obstruction, and is associated with perceived control over the situation, usually resulting in approach-oriented behaviours designed to confront, disarm or compel (Halperin & Gross, 2011; Harmon-Jones et al., 2014; Roseman, 2025). Anger has been widely studied in political science for its tendency to activate and mobilise individuals (sometimes violently) to correct the disliked situation (Redlawsk & Mattes, 2022), as well as for its resistance to information contradicting prior beliefs (Suhay & Erisen, 2018). In AP, anger will usually be experienced towards the out-party, and may be linked to perceptions of out-party leaders as unethical or corrupt (Huber et al., 2015; Webster, 2020).
H3: Anger towards an out-party leader is correlated with increased partisan AP.
Fear is a negative emotion associated with uncertainty and perceived threat under conditions of low control (Roseman, 2025). It may promote avoidance, vigilance and support for protective or stabilising responses, sometimes linked to support for authoritarianism (Fontaine et al., 2007; Mackie et al., 2008; MacKuen et al., 2010; Shaffer & Duckitt, 2013; Whitson et al., 2015). When attributed to out-party leaders, fear should be associated with perceptions of the out-party as threatening, possibly increasing reliance on social groups that convey security.
H4: Fear towards an out-party leader is correlated with increased partisan AP.
Finally, disgust is a negative emotion often tied to moral judgements and perceptions of intrinsic aversion (Roseman, 2025). Disgust, which may either refer to physical (disgust proper) or cognitive repulsion (contempt) of others (Redlawsk, 2023), entails perceptions of superiority, and promotes exclusion by the “othering” and dehumanisation of out-groups (Rozin et al., 2008; Taylor, 2007; van Leeuwen & Petersen, 2018). Disgust towards political leaders may therefore be closely associated with entrenched animosity towards the out-party, as well as with views of the in-party as morally superior.
H5: Disgust towards an out-party leader is correlated with increased partisan AP.
Thus, while valence is an important dimension of affect and matters for the congruence of goals with individuals’ (positive/negative) partisan identity, these expectations seek to test the idea that not all emotions necessarily relate to partisan evaluations in the same way.
Method
Sample Characteristics
TRI-POL (Torcal et al., 2023) collected panel data using quota-based convenience samples of 6,201 individuals across three waves in Argentina, Chile, Italy, Portugal and Spain, with sample sizes ranging from approximately 800 to 1,300 participants per country-wave, split evenly between men and women. Country samples differed primarily by age, with slightly older European populations; habitat size, with Latin Americans more highly concentrated in large metropolitan areas; and education level, with nearly half of the Spanish sample not completing upper secondary education, most Italians and Chileans achieving upper secondary, and higher education being most common among Argentines and Portuguese respondents. Despite the differences between countries, samples were representative of their respective populations.
The surveys were partially timed with elections in three of the five countries: Wave 1 (September-November 2021) coincided with the first round of Chile’s general election and Argentina’s legislative elections; Wave 2 (December 2021 – January 2022) covered the second round of Chile’s general election and Portugal’s snap legislative elections. While Italy’s own snap election was not covered directly, data was collected in the lead-up to it. Wave 3 took place between March and April 2022. Full details on the technical information and sample characteristics can be found in Section A of the Electronic Supplementary Materials (ESM).
This research follows a structured cross-national comparative design, highlighting the varying role of affective polarisation across multiparty settings. Although the countries vary in party systems, democratic traditions and electoral systems, they provide a robust testing ground to explore whether the relationship between specific discrete emotions towards leaders and partisan AP transcends national contexts. The study broadens the scope of AP research by focusing on multiparty systems, which introduce additional complexity that merits examination (Wagner, 2021). Including both Southern European and Latin American countries with varying levels of AP also enhances regional diversity, making the findings more comparable across different political settings.
Operationalisation
Much of the existing literature on AP focuses on two-party systems, where partisan dynamics are more straightforward, due to clear-cut in-party/out-party distinctions. In adapting this to multiparty settings, we are faced with the choice to include several or just one party (each) within the in- and out-party concepts (Wagner, 2021). I use respondents’ most- and least-liked party to measure in- and out-party affect, following recommendations to analyse parties in pairs when dealing with multiparty contexts to avoid weakened effects in AP measures (Gidron et al., 2020).
To operationalise the in-party, I used a survey question asking respondents to identify their most-liked party. For out-parties, I used a question asking participants to name the party they would never vote for (only available in Waves 2-3). If these questions were left unanswered, I selected the parties to which they respectively assigned the highest and lowest probability-to-vote (PTV) scores. Those who could not be assigned a party were excluded from the analyses. The main variables derive from this in- and out-party specification.
Independent Variables
The primary independent variables are emotions (hope, pride, anger, fear and disgust) towards the respondent’s in-party and out-party leaders. Survey questions asked respondents how often they had experienced these emotions towards various political leaders in the last 3 months during each of the waves, with responses recorded on 5-point Likert scales (where 1 = Always, and 5 = Never). I treat emotion frequency as indicative of overall emotional degree, given evidence that repeated affective experiences enhance emotional salience and strengthen memory-based associations (Madan et al., 2019; Popov & Reder, 2020). Finally, I selected respondents’ answers to the political leaders of their most- and least-liked political parties, reversing and rescaling responses such that scores ranged from 0 (“Never”) to 1 (“Always”).
Dependent Variables
In-party affinity and out-party animosity are computed using 0-100 (coldest-to-warmest) feeling thermometer (FT) ratings of partisan voters by respondents. Although AP can also be measured using survey items on partisans’ traits (Garrett et al., 2014; Iyengar et al., 2012), and items on trust in political parties (Levendusky, 2013), the feeling thermometer is the most common instrument to measure AP (Druckman & Levendusky, 2019) and has particularly been validated in multiparty systems (Gidron et al., 2022). FT ratings for the voters of each respondent’s most- and least-liked parties were selected, with the latter reversing the 0-100 ratings to reflect higher animosity. A positive coefficient for either dependent variable therefore indicates a higher degree of AP towards that partisan group.
Controls
While this study focuses on identity-congruent emotions, affective experiences may occasionally contradict people’s pre-existing partisan views. In addition to the main emotion variables, the models also include controls for affectively incongruent evaluations (e.g., negative emotions towards preferred leaders or positive emotions towards opposing leaders), as instances of identity-conflicting emotions, while rare, may generate cognitive dissonance (Festinger, 1957; Harmon-Jones & Mills, 1999) and motivate partisan realignment (Lavine et al., 2012), ultimately reducing AP.
Ideological extremity, partisan identification and political interest were included as controls for respondents’ political orientations, given their potential links to in-party attachment, political engagement and motivated reasoning, which could intensify partisan affect and emotional responses to political leaders. Ideological extremity reflects respondents’ distance from the ideological centre on a 0-10 left-right scale, based on absolute self-placement values (rescaled slightly, ranging from just over 0 to just over 5). Partisan identification is based on reported closeness to a political party, combining whether respondents feel close to a party (binary variable) and how close they feel (0 = “not at all close” to 3 = “very close”). Political interest reverses a 1-4 self-reported measure of interest in politics (from “not at all” to “a lot”).
Age, sex, education and subjective wellbeing were included as sociodemographic controls, to assess whether certain social groups are more prone to in-party affinity or out-party animosity, potentially due to exposure to polarising narratives rooted in salient social cleavages. Age is numeric (18+), while sex is coded as 1 for male and 2 for female. Education was harmonised across countries into three categories, broadly applicable across contexts: “Secondary or less” (reference), “Upper secondary”, and “Higher education”. A reversed four-point item measuring difficulty living on household income (1 = “live comfortably” to 4 = “many difficulties”) was used as a proxy for subjective wellbeing, capturing perceived financial strain.
All numeric independent and control variables were normalised to a 0-1 scale to aid interpretation, and demeaned by wave to control for time-invariant individual characteristics and focus on within-individual variation. Demeaning isolates the impact of changes in the predictors over time, helping address endogeneity by reducing bias from unobserved confounders. Categorical variables such as sex and education level were treated as factors.
Model
To test my hypotheses, I use a linear mixed-effects model (LMM), regressing emotions towards leaders on partisan affect, with waves nested within individuals. While the long-format dataset includes repeated observations for each respondent, the main models do not explicitly estimate lagged temporal effects, as doing so would substantially reduce the available sample size (ESM Tables D7-D16 show descriptive results). Nevertheless, including a random intercept for individuals accounts for unobserved heterogeneity and the non-independence of repeated observations within respondents over time. Fixed effects for key predictors control for between-person differences and observable confounders. These models thus test correlational expectations; while the study makes no causal claims, this multilevel approach helps reduce the influence of unobserved, time-invariant confounders that might otherwise bias the observed associations (Bell et al., 2019). However, note that prior research suggests polarisation may generate self-reinforcing dynamics due to the causal arrow pointing both ways at once (Somer, 2001; Torcal & Thomson, 2023): while some factors generate strong affective responses, intense affect may also close people’s minds to information contradicting prior beliefs, following patterns observed in the cognitive dissonance literature (Festinger, 1957; Harmon-Jones & Mills, 1999).
Results
I plot the coefficients for emotions towards leaders on levels of in-partisan affinity and out-partisan animosity by valence (Figure 1) and by discrete emotion (Figures 2 and 3) (full models ESM Tables C1-C2). I include multiple emotional states towards in- and out-party leaders to reflect the idea that multiple emotions can be felt simultaneously, even when seemingly contradictory (Roseman, 2025). The Electronic Supplementary Materials also include regressions pooling data from all five countries (ESM Table C3) and plots for the identity-conflicting emotions (ESM Figures C1-C2).
Before turning to the main results, Figure 1 shows the association between the valence of emotions felt towards in- and out-party leaders with partisan affect. These models provide a general overview of how the five emotions studied here should be expected to correlate with partisan affect based on their valence, by country. Minor deviations in the association of discrete emotions with respect to the general pattern support the view that discrete emotions can offer additional insights into the complex nature of affect, despite the valence of an emotion indicating whether experiences are consistent with one’s motives (Roseman, 2025).
Figure 1
Emotions Towards Leaders (Valence) and Partisan Affect
Note. Regression coefficients for 0-100 FT ratings of partisans.
Overall, these models show that positive-valence emotions towards in-party leaders are consistently associated with in-partisan affinity, but only conditionally linked to out-partisan animosity. The latter may depend on the degree of distinctiveness or explicit rivalry between individuals’ most- and least-liked parties (Nicholson et al., 2018), which may also respond to regional considerations within countries. Similarly, negative-valence emotions towards the out-leader are systematically associated with out-partisan animosity, but only with in-partisan affinity in the case of Spain and in the pooled sample (but note that the larger N may inflate statistical significance, and results should be taken as descriptive rather than inferential).
Figure 2 disaggregates positive emotions towards in-party leaders into hope and pride. Results show that in-partisan affinity is consistently associated with both hope and pride towards the in-party’s leader, although effects are negligible for out-partisan animosity, mirroring findings in Figure 1.
Figure 2
In-Leader Hope and Pride and Partisan Affect
Note. Regression coefficients for 0-100 FT ratings of partisans.
While coefficients appear similar for affinity, and are only weakly related to animosity, a small positive effect of in-leader hope on in-partisan animosity is found in the pooled sample. Although this may partly be due to the large sample size obtained by pooling, this effect is not found in the case of pride. This may indicate that hope in in-party leaders is inherently bounded, as it does not extend pro-sociality to out-partisans; if it did, one would expect lower levels of out-party animosity – if anything, it may even be weakly associated with higher out-party dislike. Conversely, the association between in-leader pride and out-partisan animosity appears more mixed, with a non-significant pooled effect nearing 0. Indeed, results are only significant in Italy, where the negative coefficient suggests that, in some cases, individuals who experience pride towards political leaders may tend to have a more positive view of their political landscape, thereby correlating with lower out-party animosity; or, put differently, those who evaluate out-partisans more positively exhibit lower levels of in-leader pride.
Figure 3 distinguishes between anger, fear and disgust towards out-party leaders. Despite similarities with Figure 1 – negative emotions towards the out-leader correlate strongly with out-partisan animosity but only weakly with in-partisan affinity – disaggregating the emotions yields distinct patterns. As Figure 3 suggests, while these are all negative-valence emotions and responses to perceived threat (Renström et al., 2023; Stephan & Stephan, 2000), we intuitively differentiate them beyond just valence (Scherer, 2003).
Specifically, while out-leader disgust is consistently associated with out-partisan animosity across all cases, the relationship appears more context-dependent for anger, and especially for fear. Figure 3 shows that anger is significantly associated with out-partisan animosity in all countries except Chile, whereas fear is only significant in Chile. This pattern suggests that different political contexts may condition which threat-related emotions become most salient. One possible explanation lies in the distinction between fear and anger identified by Lerner and Keltner (2001), particularly regarding perceptions of control and temporal orientation: future threats and uncertainty may trigger fear, while violations already committed may trigger anger. Fieldwork in Chile occurred roughly a year after the referendum, when the conditions of the new constitution were under negotiation; as such, political discourse centred on the uncertain and contested constitutional reform process, potentially framing partisan conflict as a future-oriented and institutionally mediated threat: “Facing immoral but overwhelmingly powerful authorities tends to elicit emotions such as fear … that sap anger” (Roseman, 2025, p. 12). By contrast, disgust remains comparatively stable across contexts, suggesting that moralised rejection of political opponents may operate more independently of shifting political conditions and institutional contexts. Moreover, despite having received scant attention in the AP literature, out-leader disgust is also linked to in-party bonding, as evidenced by higher levels of in-partisan affinity in Spain.
Figure 3
Out-Leader Anger, Fear and Disgust and Partisan Affect
Note. Regression coefficients for 0-100 FT ratings of partisans.
Figures 2 and 3 thus resemble but also qualify the general patterns observed in Figure 1: (1) positive-valence emotion appeared to correlate with higher out-partisan animosity in Spain, Chile and in the pooled data, but results were only significant for hope in the pooled regression; (2) the correlation between negative emotions and out-partisan animosity was generally non-significant for fear, which was hidden in the valence model; (3) the valence model showed a correlation between negative emotions and in-partisan affinity in Spain and in the pooled data, which the discrete models only confirmed for disgust in Spain. In addition, the valence models obscured the finding that in Italy in-leader pride was correlated with lower out-partisan animosity. Although differences were minor, AIC and BIC comparing these two model specifications (ESM Tables D7-D8) tended to support the discrete model over the valence model, especially in the case of out-partisan animosity, where discrete emotions behaved more heterogeneously.
Beyond the main results, additional findings are worth reporting. First, exploratory analyses of identity-incongruent emotions (ESM Tables C1-C3), included as controls, suggest that emotions contradicting partisan identities are in some cases linked to lower levels of AP, aligning with research on partisan ambivalence (Lavine et al., 2012). Specifically, fear and disgust towards one’s in-leader were associated with lower in-partisan affinity and out-partisan animosity both, although effects were context-dependent. Likewise, out-leader hope was linked to lower in-partisan affinity in Argentina. In addition, anger towards in-party leaders shows no relationship with partisan evaluations; and out-leader pride, a seemingly contradictory concept, was generally unrelated to partisan affect, with the notable exception of Chile, where out-leader pride was associated with higher out-partisan animosity, potentially reflecting a temporary misalignment between party labels and leaders’ positions during the referendum, such that leaders of disliked parties may have temporarily elicited goal-congruent reactions among their opponents. Finally, a consistent negative effect of hope towards the out-leader on animosity towards out-partisans was discovered in the regression coefficients for every single country, suggesting that differences in hope towards political leaders are a key factor in (partisan) AP.
Second, control variables behaved as expected, but highlight different dynamics across dimensions of AP. Ideological extremity and partisan identification are both strong and consistent predictors of in-partisan affinity, with coefficients sometimes even exceeding those of emotions towards in-party leaders. By contrast, only ideological extremity shows a somewhat consistent relationship with out-partisan animosity, with smaller coefficients. This suggests that while strong in-party attachment is concentrated among ideological extremes, out-party dislike is more evenly distributed across society, as partisan identification is not linked to out-partisan animosity. Finally, although both models did an adequate job predicting partisan AP, the out-partisan animosity model had a slightly weaker explanatory power, meaning that much of the variance is still unaccounted for.
I also estimated several alternative specifications to assess robustness. Identity congruence models (ESM Tables D1-D2) included only emotions aligned with partisan expectations, and produce similar results but tend to inflate some of the findings, likely due to omitted emotion categories that reflect partisan ambivalence. Identity-targeted models (ESM Tables D3-D4), included for completeness, only estimated emotions towards the relevant leader, performed less well and yielded less stable estimates. Ultimately, the main models (ESM Tables C1-C3) provide the best balance between parsimony and explanatory power, capturing variation that more constrained models overlook. Thus, while cross-party correlations remain weak, how we feel about one party’s leader(s) may be, albeit minimally, linked to how we feel about the supporters of another party, even in multiparty systems, where parties may be less directly opposed.
Finally, I examine temporal dynamics using lagged models (ESM Tables D7-D16). Due to data limitations (lagging drops first-wave observations and respondents with limited panel participation) and the potential for sample bias, these models are presented as robustness checks rather than core analyses to avoid overfitting. While results should be interpreted with caution, they are directionally consistent with the main findings, suggesting that earlier emotional responses may also predict subsequent partisan evaluations.
General Discussion
This study examined affective polarisation through discrete emotions attributed to political leaders. I show that discrete emotions may refine valence-based accounts of AP, although valence remains a useful indicator of AP: it captures what is arguably the most stable attribute of affective experience (Barrett, 2006). Consistent with this, the valence-based models broadly reflect the main patterns observed in the data; however, they also gloss over important differences by collapsing distinct emotional states of similar valence into a single evaluative continuum. Even though affect is generally understood as a complex phenomenon composed of multiple dimensions (Fontaine et al., 2007; Marcus, 2000; Marcus et al., 2000), theories considering conceptual dimensions such as valence, arousal, attribution, action tendencies, and so on, may remain overly abstract for survey instruments designed for lay audiences. Conversely, discrete emotions remain intuitive categories that capture these dimensions and which respondents can easily grasp (Scherer, 2003). While positive-valence experiences towards in-party leaders consistently correlated with in-partisan affinity and negative experiences towards out-party leaders correlated with out-partisan animosity, modelling emotions according to discrete states reveals additional structure. Rather than overturning valence-based findings, emotions qualify them, offering a more fine-grained account of how AP operates.
Within these broader patterns, different emotions point to different types of intergroup dynamics. Hope, associated with beneficial appraisals of uncertain future circumstances, is linked to affinity but remains sensitive to whom it is attributed to (Cohen-Chen & Pliskin, 2025): in-leader hope relates to stricter partisan boundaries, while out-leader hope relates to softer boundaries. Pride, a “self-enhancing”, “feel-good, do-bad” emotion, is usually associated with hierarchical views of society that favour the in-group/party (Cohen-Chen et al., 2020; Tracy et al., 2023); even when experienced towards unexpected actors, it does not seem to improve evaluations of outsider groups, reflecting recent actions without necessarily altering future expectations. Anger, associated with high control and approach-oriented responses (confrontation and conflict), is generally linked to partisan animosity (Webster, 2020), and is unrelated to partisan evaluations when attributed to one’s own leader. Fear, associated with uncertainty and perceived low control, aligns more closely with avoidance, and seems conditionally linked to animosity: the threat eliciting the emotion should be seen as beyond the individual’s ability to control (Roseman, 2025). Disgust, in turn, stands out as a particularly robust predictor of out-partisan animosity across countries, independent of institutional contexts. This is notable given that disgust remains relatively understudied in the AP literature, although it has played an important role in other literatures, such as moral intergroup conflict (Hodson & Costello, 2007), given its consequences for moral judgement, social exclusion and “othering” processes (Redlawsk, 2023; Rozin et al., 2008). Unlike anger (a response to past transgressions), disgust is tied to perceptions of who/what others inherently are (Roseman, 2025), making it especially relevant for more social aspects of AP: it is associated with avoidance-oriented responses similar to fear, but also with dehumanisation and entrenched intergroup boundaries. In particular, the importance of each specific emotion may vary according to which aspects of AP are being considered. Taken together, these findings suggest that similar valence of affect may correspond to qualitatively different forms of AP.
This paper has also empirically highlighted the idea that different elements factor into partisan affinities and animosities, consistent with research showing that positive and negative partisanship are related but distinct (Abramowitz & Webster, 2018; Bankert, 2022). This is firstly shown in the role of different emotional states, which show almost no cross-effects: positive emotions towards in-party leaders are strongly linked to in-partisan affinity but have little bearing on out-partisan animosity, while negative emotions towards out-party leaders primarily map onto hostility towards out-partisans. Similarly, the relevance of key political covariates varies depending on which dimension of partisan AP is considered, with in-partisan affinity being more strongly related to individuals’ political extremity and degree of partisan identification. These asymmetries reinforce the need to model the two forms of partisanship separately (Bankert, 2022).
In addition, while emotions are known to be context-dependent, varying across national contexts (Russell, 2009), I conclude that institutional changes may potentially upset partisan affective dynamics. For instance, contrary to findings in all other countries, in the context of Chile’s 2021 plebiscite for constitutional reform, where some leaders supported options contrary to policy positions typically associated with their party, out-partisan animosity was associated with fear rather than anger, suggesting a shift in the type of threat perceived, potentially moving from confrontation to avoidance-based dynamics. Salient threats here may have been more institutionalised and future-oriented, with uncertainty and powerlessness playing a larger role than past grievances (Roseman, 2025). The temporal and institutional aspects of political appraisals also appear to differentiate experiences of hope (uncertain, future-oriented) and pride (recent actions) towards out-party leaders in Chile, suggesting that despite leaders sometimes acting in ways that are consistent with rival voters’ motives, out-party categorisations and feelings remain stable, and require more consistent positive expectations of future actions to reduce negative party evaluations. The case of Chile suggests that political emotions and affective polarisation may be vulnerable to institutional dynamics.
In this article, I have argued that measurements of AP that rely only on valence may be mired in overly abstract views of this phenomenon. This partly relates to the ongoing discussion around whether AP poses a threat to democratic functioning (Kokkonen & Harteveld, 2025): while researchers may be more sophisticated than lay audiences, measurements of AP that consider a fuller view of affective experiences will have an easier time disentangling the negative and positive aspects of polarisation. Specifically, in this research it becomes clear that disgust matters for AP, but likely more for evaluations of partisan voters than it would for voting behaviour, where anger and hope likely remain more relevant. Thus, research concerned with the social aspects of polarisation and how these may relate to cohesion and other elements central to democratic functioning (Torcal & Thomson, 2023; Torcal et al., 2025) should be concerned with disgust as problematic and disregard in-leader anger, which is wholly unrelated to partisan evaluations. Views of hope as a potential strategy for depolarisation should bear in mind that it likely needs to be attributed to out-parties, and this may require more broad, cross-cutting appeals rather than one-sided discourse and policies.
Ultimately, the findings underscore that political emotions are not interchangeable. Even when they share valence, they reflect different appraisals of threat and opportunity, and imply different behavioural orientations. By incorporating discrete emotions into the study of affective polarisation, this paper shows that we can move beyond simple like-dislike evaluations in order to gain a clearer understanding of how partisan divisions are experienced, maintained and, potentially, mitigated.
Limitations and Future Research
This study has several limitations. First, it relies on self-reported emotions, which may not fully capture how people actually experience and act on emotions. Respondents may simplify or misreport their feelings, and theory has not yet caught up on all aspects of emotion (Russell, 2009). That said, self-reports remain widely used in political psychology (Roseman, 2025), and there is good reason to believe they reflect feelings stored in memory and retrieved when making political judgments (Madan et al., 2019; Popov & Reder, 2020; Roseman et al., 1996). A related concern is that the results may partly reflect general response tendencies rather than specific emotional effects. Because both emotions and partisan evaluations are self-reported, respondents may answer in consistent ways (e.g. liking co-partisans and disliking opponents), blurring distinctions between discrete emotions. In such cases, the discrete model does not always provide substantially more information than the valence model. However, the observed patterns are not fully consistent with a purely valence-driven account: emotions sharing the same valence do not behave identically (e.g. anger and fear), and only some identity-incongruent emotions are linked to lower animosity. Future research could address this by measuring action tendencies more directly to differentiate between the emotions more clearly.
The choice of emotions also matters, and this is an area that requires further theoretical refinement within political science. For example, disgust may overlap with contempt when applied to political leaders (Moran & Eyal, 2025), even though the two are theoretically distinct (Redlawsk, 2023). While this study represents an early attempt to include a wider range of emotions, other emotions may also be important for understanding AP. Emotion System Theory (EST) considers a longer list of relevant emotions (Roseman, 2025). In particular, AP research may be interested in studying surprise to disentangle mixed findings for fear; and other emotions such as frustration, relief, regret, guilt and shame. Along similar lines, resentment (not included in EST), may bear particular relevance for AP research, given its link to polarising and ultra-conservative discourses. Given the expanding list of possible emotional states, it may be useful to allow respondents to opt in/out of which emotions they feel to reduce survey fatigue and improve measurement quality.
Another limitation is that emotions were measured only towards political leaders. Emotions can also be attributed to partisan and/or social groups (Kretchner et al., 2026; Renström et al., 2021, 2023), or broader political events (Renström & Bäck, 2021), and each of these may have different implications. Future research could examine emotions towards these various targets collectively to disentangle which political stimuli elicit the strongest emotions, with a special focus on salient events framed as societal threats in elite discourse, which may additionally differentiate policy-based and identity-based polarisation.
Finally, the analysis is correlational, which limits causal claims. Although lagged models were estimated, the available panel data restricts the ability to draw strong conclusions about directionality. Future research should use longer panel designs or experimental approaches to better identify causal relationships between emotions and affective polarisation. Similarly, this research focuses on a handful of multiparty democratic systems, and findings may not be applicable to other institutional configurations.