Emotionally aligned and structurally sorted: How opinion groups divide on climate and asylum

Abstract

Comprehensive evidence on affective polarization stems from bipartisan political systems with well-defined partisan identities and unidimensional measures of ingroup favoritism and outgroup dislike, while neglecting shared emotions and intergroup emotional dynamics. This study addresses these gaps. Before the 2021 German federal elections, we surveyed individuals in Germany concerned about climate change (n = 2,477) or asylum policies (n = 3,177) using a large, innovative social media sample. We assessed positive and negative feelings toward supporters and opponents of progressive policies on these issues and the perceived emotional alignment with these opinion-based groups. Cluster analyses reveal two affectively polarized groups, a group of resenters characterized by negative affect and emotional misalignment, and three groups showing affective differentiation but lacking emotional identification. Compared with the other groups, the two polarized opinion-based groups perceive themselves as being more emotionally similar to like-minded citizens and more dissimilar from those with opposing views. They also experience consistent, yet distinct patterns of emotions toward like-minded and dissenting others regarding anger, disgust, contempt, and joy. The two groups are sociodemographically distinct and have less contact with people holding opposite views. These structural divisions are reflected in behavioral patterns. Polarized groups engage in more political discussions, especially with strong ties. However, only polarized conservatives tend to engage in political discussions with weak ties, such as colleagues or acquaintances. Polarized progressives are more likely to engage in collective forms of political mobilization. In essence, this work underscores the interplay among the emotional, structural, and interactional components of issue-based affective polarization.

Significance Statement.

Affective polarization—strong emotional divides between political groups—is typically studied in two-party systems with well-defined partisan identities. This study shows that even in multiparty systems, citizens form emotionally charged, opinion-based groups around key issues like climate change and asylum policy. These groups not only favor their ingroup and dislike the outgroup but also feel emotionally aligned with like-minded individuals and emotionally distant from the other group. These emotional patterns are reinforced by social sorting and influence how people talk about politics and take political action. Our findings highlight how emotions, social interactions, and political behavior interact to entrench issue-based divisions, posing challenges for democratic cohesion in fragmented political landscapes.

Introduction

Affective polarization is considered a threat to democracies and social cohesion worldwide. As a group-based and identity-driven phenomenon, it implies harboring negative affect toward outpartisans and political ingroup favoritism and is associated with attitudes, emotions, and behavioral tendencies considered problematic from a democratic standpoint (1–3). This predicament includes emotions of hate and contempt, biased perceptions of outgroup members, reduced democratic accountability, and support for political violence.

Most insights into affective polarization come from bipartisan contexts, in particular the United States, with clear and historically established group boundaries rooted in partisan identities. In multiparty contexts and in systems with weakening party identities (4–6), affective polarization is less clear to conceptualize and measure (7, 8), and it cannot be reduced to ideological blocs along the left-right spectrum (9). Recent studies suggest that issue-based affective polarization, i.e. between opinion-based groups, can be observed for particular political issues, for instance, separationism, gun ownership, vaccination, and climate change (7, 10, 11). A key assumption of this line of research is that social identities evolve in relation to the position citizens hold regarding salient issues (12). However, there is a lack of clarity on how these identities evolve, how group boundaries emerge around issue positions, and how emotions shape these dynamics. The current research addresses this gap by (i) examining polarized relationships between opinion-based groups in a multiparty system, (ii) studying emotional alignment as a novel mechanism of affective polarization, (iii) identifying intergroup emotion profiles for these groups, and (iv) assessing their associations with social structural embeddedness and political behavior.

Emotions and issue-based affective polarization

Affective polarization is likely to arise from partisan and opinion-based groups in distinct ways. While partisanship provides the grounds for robust social identities associated with a host of well-established cognitive, affective, and behavioral consequences for intergroup relationships (13, 14), it is less clear how social identities are formed and maintained in the case of issue positions. Partisan identities are rooted in comprehensive ideologies, a well-defined, relatively stable set of political attitudes, and formal criteria (e.g. party membership, casting votes) that define ingroup and outgroup. In contrast, opinion-based groups lack clearly defined boundaries, remain ideologically shallower, and and their contours are fuzzy, for instance, due to gradual alignment and multigroup affiliation (15, 16). Moreover, affective polarization between opinion-based groups is likely to be more volatile, because political issues and opinions are driven by rapid news cycles—i.e. issue salience fluctuations—and short-term shifts in political opinions (16, 17).

To account for these specific characteristics of issue-based affective polarization, the present study follows recent calls for research focusing on the role of sharing of emotions and discrete emotions in shaping affective polarization (18–21). We argue that emotions add analytical value in two respects: First, as objects of reflexive and evaluative thought, emotions matter for signaling group cohesiveness and entitativity. Social identity approaches to partisan affective polarization emphasize the cognitive aspects of self-categorization (1), but pay considerably less attention to the perception of being emotionally aligned as a further source of categorization. Although individuals from diverse ideological backgrounds may converge on similar issue positions (22), as in vaccination hesitancy during COVID-19 (23, 24), such shared positioning is not necessarily associated with established group identities, nor does it automatically contribute to their emergence. Yet, the perception that others share similar emotions toward an issue while political opponents do not—i.e. the perception of high intragroup and low intergroup emotional alignment—significantly contributes to group entitativity and the impression of group distinctiveness. This, in turn, heightens the salience of opinion-based groups and members’ self-categorization (25). Beyond observing a common issue position (i.e. cognitive consensus), the impression of emotional alignment—even if individuals do not, in fact, experience the same emotions—is likely sufficient to produce a sense of unity and togetherness, strengthening group cohesion and efficacy (26). This is corroborated by research suggesting that the perception of shared emotions within a group contributes to the development of social identities (27–29) and can serve as a motivator of collective action—even in the absence of fully formed groups (30). In more fuzzy and ephemeral groups, such as opinion-based groups that lack clear boundaries, perceived emotional alignment can become a crucial marker of affective ingroup cohesion. Ultimately, perceiving emotional alignment with ingroup members and divergence with outgroup members may deepen affective rifts between opposing camps and ignite or reinforce the “us versus them” dynamics that characterize affective polarization.

Second, emotions matter in terms of their specific affective qualities, cognitive structures, and social functions, including characteristic appraisals (31), action tendencies (32), and expressive signals (33). In particular, they matter as intergroup emotions, i.e. when they are intentionally directed at ingroup and outgroup members (34–37). Affective polarization research typically relies on a one-dimensional account in which affective valence, i.e. positivity versus negativity, defines intergroup relationships (1). However, recent research has suggested that the valence dimension only covers a limited spectrum of the politically relevant affective dynamics between groups, and that discrete intergroup emotions add conceptual depth and granularity to the picture (18–21). For example, the negatively valenced emotions “hate” and “anger” are considered activating emotions that motivate electoral participation (38), activism (22), and political conversation (39). Conversely, “disgust” and “contempt” are also negatively valenced but associated with avoidance tendencies (40, 41), withdrawal from information seeking (42), and “harmful inaction,” that is, the shirking of cooperation (43). Disgust and contempt are examples of emotions that signal group boundaries, while other emotions can indicate the presence of a common ground: disappointment (20), national pride (44), and embarrassment (39) have been shown to make shared identities salient and to promote constructive engagement. Moreover, discrete emotions are not mutually exclusive and can encompass conflicting feelings (21, 41). For instance, disappointment involves positive and negative emotions toward individuals who are simultaneously perceived as fellow citizens, i.e. ingroup, and political adversaries, i.e. outgroup (20).

To account for this two-sided relevance of emotions—perceived emotional alignment and discrete intergroup emotions—for issue-based affective polarization, the present study complements established measures of ingroup favoritism and outgroup hostility with self-reports of issue-based perceived emotional alignment and of discrete intergroup emotions.

Social structural niches and political behavior

Even though opinion-based groups lack well-developed social identities and clearly demarcated boundaries, their members may nevertheless share similar sociodemographic characteristics and patterns of political behavior, both of which have been shown to be associated with affective polarization (45–47). We therefore also seek to understand how the novel emotion-based conceptualization of polarization is related to sociodemographic characteristics and distinct patterns of political behavior.

The concept of social sorting is frequently employed in research on polarization to explain the alignment of sociodemographic characteristics with political identities, which renders political groups socially homogeneous and distinct from one another (45, 48, 49). For opinion-based polarization, the approach of describing groups in terms of these characteristics may be less useful due to the lack of clearly defined boundaries between groups. To account for such fuzzy social identities, we use the concept of social structural niches (SSNs) to refer to more or less distant clusters of citizens in social space who share similar sociodemographic characteristics, and with that interaction opportunities, and social environments, regardless of their political identity (50, 51).

A key mechanism through which inhabiting distinct SSN translates into and reproduces polarized intergroup relationships is intergroup contact. It is well established that social contact can reduce stereotyping and intergroup animosity when it is frequent, nonthreatening, and perceived as positive (52, 53), while lack of positive contact is supposed to deepen social divides. Many interventions indeed seek to reduce affective polarization by promoting intergroup contact (54, 55). However, as individuals tend to interact with people similar to themselves (56), contact across partisan- or opinion-based groups is often hampered by social distance and stratification, i.e. citizens’ socioeconomic status and class positioning (57, 58), ideological sorting (45, 48, 49), and network homophily (59, 60). We thus aim at understanding whether affectively polarized opinion-based groups are indeed located in distinct SSNs with corresponding patterns of intergroup contact.

Affective polarization has also been shown to promote different noninstitutional forms of political engagement (38, 39). Several mechanisms suggest that emotions, in addition to affect, play a central role in political behavior. First, the impression of shared or aligned emotions strengthens group identification and motivates collective action (27) and participation in institutional as well as noninstitutional forms of political engagement (47). Second, the action tendencies associated with emotions can account for the mobilization of political engagement (38). Third, meta-analyses show that affect positively predicts electoral and political participation (47, 61) and that emotions motivate citizens to engage in political conversations (39). Following these lines of research, we investigate whether affectively polarized opinion-based groups are, similar to partisans, characterized by distinct patterns of political conversation and engagement (62, 63).

Taken together, by triangulating measures of affect, perceived emotional alignment, and discrete intergroup emotions with citizens’ SSNs and political behavior, this study seeks to significantly improve our understanding of the emotional dynamics of issue-based polarization in a multiparty context.

Data and measures

Sample

This study draws on unique data from a survey of politically engaged individuals we recruited through an innovative social media sampling procedure (64). Compared with traditional probability samples, social-media-based sampling systematically oversamples hard-to-reach populations. We leveraged this sampling approach to specifically target politically engaged individuals, a population central to research on affective polarization (47, 61). Respondents recruited through this procedure are substantially more active across different types of political engagement than participants in established probability surveys. Our sample is thus drawn from a population that shows higher levels of political and affective involvement than the general population, making it particularly valuable for investigating the mechanisms and relationships underlying affective polarization (64).

The survey was conducted between 2021 August 2 and 2021 September 10—2 weeks before the German federal election—and comprises a sample of n = 6,163 individuals living in Germany. Since we seek to investigate affective polarization of opinion-based groups, participants were asked to select one of three political issues based on its importance: climate change, asylum, and affordable housing. We focused on the two most prevalent issues and created two subsamples, citizens for whom climate change (n = 2,477, M_Age = 47, SD_Age = 14, Male = 0.63) and asylum (n = 3,177, M_Age = 50, SD_Age = 14, Male = 0.66) were the most pressing issues. Previous research suggests that, particularly in the domains of immigration and asylum, issue salience is a strong predictor of individuals’ issue positions (65). Therefore, our design choice likely maximizes participants’ emotional and political engagement with the issue, albeit at the expense of achieving balanced representation of different positions on the issue (66). The study involved anonymous survey research with adult participants and did not include any experimental manipulation, deception, or collection of sensitive personal data. According to institutional and national regulations, formal approval by an ethics committee was therefore not required. Permission to conduct the interviews for the purposes of this research was obtained by all respondents, who were fully informed about its purposes and how their responses would be used and stored.

Measures

To assess participants’ affect toward opinion-based ingroup and outgroup members, perceived emotional alignment, and discrete intergroup emotions, we developed highly polarizing politically progressive policy statements on climate change and asylum and asked respondents about their affective stances toward supporters and opponents of these policy positions (opponents or supporters of a “climate policy that does everything possible to stop climate change” or an “asylum policy that grants asylum to all persons in need in Germany”). Such statements feature frequently in the German political discourse and can thus serve as reference points for a continuous and potentially polarizing spectrum of affective orientations of issue-based groups (rather than serving as accurate representations of the political positions of partisans).

For the issue respondents evaluated as most important to them, (i) a thermometer measured their affect toward supporters and opponents of the policy statements (−5 = negative/cold to +5 = positive/warm). (ii) Perceived emotional alignment, i.e. the extent to which respondents evaluate their own emotional reactions to these issues as consistent with those of others supporting or opposing progressive positions, was measured on a 7-point Likert scale (1 = does not apply to 7 = fully applies). This measure draws on an established scale assessing perceived emotional synchrony in rituals and face-to-face gatherings (27). The original scale was adapted to a single-item measurement to be suitable for large-scale data collection. Our measure of perceived emotional alignment is designed to directly capture participants’ beliefs about how their own and others’ emotions match. Such beliefs about emotions have been shown to impact psychological and behavioral outcomes (67–69), apply to perceptions of collective or group-based emotions (70), and are distinct from ingroup favoritism and outgroup dislike (see Results section and Tables SI1.1 and SI1.2). (iii) Discrete intergroup emotions were measured by asking participants how strongly they felt fear, anger, disgust, shame, contempt, pride, and joy when thinking about supporters and opponents of progressive policies (1 = does not apply to 7 = fully applies). (iv) To identify respondents’ SSNs, we assessed a range of sociodemographic indicators and calculated respondents’ positions in a multidimensional Blau Space (50, 51), in which smaller distances represent higher chances of social contact. Contact with like-minded individuals and those with different opinions was measured by assessing the frequency of contact with supporters and opponents (using 5-point ordinal scales, 1 = never to 5 = every day), as well as the opinion congruency between ego and their social networks (using a 5-point ordinal scale; 1 = everyone has a different attitude than mine to 5 = all have the same attitude as me). (v) In terms of political behavior, we used 5-point scales to assess the frequency of political discussions with different types of social ties (e.g. partner, friends, etc.; 1 = never to 5 = every day) and binary items to determine whether participants engaged in different political actions (see SI Appendix for details on all measures). Taken together, our measures are specifically designed to capture affective and emotional dynamics of issue-based polarization among politically interested citizens, with issue positions being assessed as support or opposition to particular policy positions.

Results

The analysis proceeds in four steps. First, for each subsample, we conducted a k-means cluster analysis using the affect thermometer and perceived emotional alignment to determine how respondents affectively relate to and align with supporters and opponents of the two policy positions, and whether affectively polarized groups emerge from these relationships. Second, we compared these clusters in terms of discrete intergroup emotions, i.e. whether participants within the clusters actually align in reported emotions toward supporters and opponents (ingroup and outgroup members), and whether the clusters differ in this regard. Third, we inspected whether the clusters occupy distinct SSNs and, fourth, differ in terms of political behavior. In the main article, we report only essential information; for each analysis, we point to the corresponding section in the Supplementary Information.

Issue-based affective polarization

The cluster analysis was conducted separately for the two political issues, using four variables: affect thermometers toward supporters and opponents of progressive policies and perceived emotional alignment with these two groups. Based on the cluster-polarization coefficient (71), the within-cluster sum of squares, and the between-cluster sum of squares, an exploratory six-cluster solution for both datasets emerged. This solution demonstrated good stability with respect to outliers and proved robust across alternative clustering methods (SI1). Overall, we find evidence of affective polarization across opinion-based groups for both political issues, as indicated by large affective differentials and clear emotional alignment (see Table SI1.3). The six clusters can be described as follows: (i) “polarized progressives,” characterized by high positive affect and strong emotional alignment with supporters, and negative affect and low alignment with opponents of progressive policies; (ii) “polarized conservatives” with the opposite pattern; (iii) “resenters” with negative affect and low alignment toward both sides; and (iv) three “nonaligned clusters” that exhibit pronounced affective differentiation but shallower perceived emotional alignment gradients, suggesting a lack of emotional identification with both groups. Given our focus on perceived emotional alignment, the nonaligned clusters are considered as one in the following. Figure 1 illustrates these patterns, with the two polarized clusters showing steep gradients in affect (upper panels) and perceived emotional alignment (lower panels) between supporters and opponents. Descriptive statistics reported in the Supplementary Information also yield substantively meaningful insights. Consistent with prior research on issue salience (65, 66), cluster sizes differ across issues (Table SI1.3). Specifically, more participants positioned themselves closer to conservative than to progressive policy positions on the asylum issue, whereas more participants aligned with progressive than with conservative positions on the climate issue. In addition, the pattern of correlations between affect thermometers and perceived emotional alignment measures indicates that the two constructs are only moderately related and thus capture distinct underlying dimensions (Tables SI1.1 and SI1.2).

Fig. 1.

Profile plots of mean ratings for affect (top; feeling thermometer, range: −5 to +5) and emotional alignment (bottom; range: 1 to 7) toward supporters and opponents of progressive policies, shown separately for the climate (left) and asylum (right) issues. Dots represent cluster-specific means, with colors indicating cluster membership. Higher affect values denote more positive feelings; higher alignment values indicate greater emotional alignment. The vertical lines indicate the range around the mean, as determined by the positive and negative SDs. The lines connecting the mean values for supporters and opponents represent the degree of differentiation.

Polarized intergroup emotions

The first set of analyses provides more detail about these clusters by considering the discrete emotions reported by participants. It answers the question of whether the polarized clusters show distinct patterns of intergroup emotions in terms of within-cluster consensus and potential polarization of emotional experiences toward ingroup and outgroup.

Emotional consensus

The cluster analysis revealed that respondents in the two polarized clusters perceive their emotional reactions to the climate and asylum issues as highly aligned with their ingroup (i.e. with other supporters or opponents of progressive policies) and as highly inconsistent with outgroup members (i.e. those with opposite political views; for details, see SI2). To check whether perceived emotional alignment is reflected in the actual emotions of participants in the two polarized clusters, we conducted eight separate consensus analyses—one for each combination of cluster and target group (e.g. the emotions experienced by polarized progressives toward—target group—supporters of progressive policies). Following the framework of cultural consensus theory (72, see SI2.1), we performed principal component analysis (PCA) on the person-by-person correlation matrix of participants’ reported emotions. A group is considered to exhibit consensus when the first eigenvalue of the PCA is at least three times larger than the second, indicating the presence of an emotional consensus (73). We find evidence that polarized progressives and conservatives not only have the impression of shared emotions toward climate and asylum issues—as captured by the perceived emotional alignment items—but indeed converge on the emotions they experience toward their respective ingroup and outgroup. The polarized clusters show stronger emotional consensus toward their ingroups when compared with the emotional consensus of resenters and nonaligned clusters toward the same target group. Among polarized progressives, consensus in emotions toward ingroup members—i.e. supporters of progressive policies—is particularly high (first/second component ratio: climate: 20.25; asylum: 23.01), while polarized conservatives show stronger consensus in emotions directed toward the ingroup (climate: 72.75; asylum: 20.23). Consensus in emotional responses toward outgroups is comparatively lower: while polarized progressives moderately converge in their emotions toward conservatives especially on the asylum issue (climate: 6.71; asylum: 9.12), polarized conservatives show lower levels of consensus in emotions toward outgroup members, i.e. progressives (climate: 5.46; asylum: 2.97). For the resenters’ and nonaligned clusters, consensus is lower compared with the polarized clusters, especially in the case of the resenters’ cluster when considering supporters of progressive policies (climate: 3.41; asylum: 3.98) and of nonaligned when considering opponents of progressive policies (climate: 10.22; asylum: 16.82; see SI2.1).

Emotional polarization

To further add granularity to our understanding of affective polarization, we calculated individual-level differential scores for each intergroup emotion following the formula Emotion_{DifferentialScore} = Emotion_Supporters − Emotion_Opponents. These scores reflect differences in the reported intensities of each discrete emotion experienced toward supporters versus opponents, ranging from −6 (stronger toward opponents) to +6 (stronger toward supporters; see SI2.2 for mean level differences). They indicate whether the experience of a particular emotion is polarized, i.e. whether the experience of some emotions is more “skewed” toward an issue-based group, and whether this differs by political issue and cluster. We estimated seven ordinal logistic regression models with differential scores of each discrete emotion as dependent and cluster membership as independent variables. Nonaligned participants, representing those who, according to our theoretical argument, do not have clear-cut opinion-based identities, serve as the reference category. The results indicate that members of polarized clusters consistently exhibit greater emotional differences than the reference category (see Fig. 2 and Table SI2.3 for coefficient estimates and Fig. SI2.3 for predicted probabilities). This pattern holds across both issues and is most pronounced for anger, disgust, contempt, and joy. In contrast, differentials for fear, shame, and pride are comparatively smaller, suggesting weaker polarization of these emotions. Resenters also showed emotion differentials, which tended to align with the direction of polarized conservatives, albeit with notably lower magnitude.

Fig. 2.

Ordinal logistic regression coefficients and 95% CIs for each discrete emotion differential, plotted by cluster and issue (climate: n = 2,225; asylum: n = 2,849). Coefficients (x-axis) reflect the effect of cluster membership (y-axis) on the likelihood of experiencing an emotional differential in the given direction, relative to the reference category (intercept, the nonaligned clusters). Positive coefficients indicate an increased likelihood that the emotion was experienced more intensely toward supporters of progressive policies compared with the reference, and vice versa for negative values. For example, in case of the climate issue, polarized progressives have a significantly negative coefficient, i.e. emotional differential—thus suggesting more fear toward opponents than toward supporters of progressive policies. The gray vertical line at zero marks the null effect. Asterisks are used to indicate significance levels (*P < 0.05, **P < 0.01, ***P < 0.001). The significance level 0.001 is smaller than the adjusted significance levels derived from the application of a Bonferroni correction to the level 0.05 for the three null hypotheses for each of the seven models that were examined (0.05/21 = 0.002).

Social structural niches

Our findings suggest that polarized progressives and conservatives, across both political issues, perceive themselves as emotionally aligned with their respective clusters, converge in their experience of intergroup emotions, and show larger polarization regarding discrete emotions than nonpolarized clusters. Issue-based affective polarization bears a particular potential for social conflict when the opinion-based groups occupy different places in the social structure, indicating social sorting and network closures that reduce intergroup contact.

Blau space distances

To assess whether members of the polarized clusters occupy distinct positions within the social structure, we projected participants’ sociodemographic characteristics—age, education, gender, migration history, size of residential area, weighted household income—into a Blau Space (see SI3 for details on the operationalization). Positions in a Blau Space represent structural separation, differences in economic and cultural capital, values, attitudes, and ideologies (51), which should be reflected in distinct emotional experiences. They also capture baseline homophily, with sociodemographic distances serving as a proxy for the likelihood of contact (51). We calculated the distances between all participants within each cluster and between clusters based on their sociodemographic characteristics. Our findings revealed that the two polarized clusters tend to be sociodemographically distinct. In the case of participants interested in the climate issue, polarized conservatives and progressives are on average separated by the largest average social distance (M = 1.68, SD = 0.79 with a climate sample mean of M = 1.61, SD = 0.77; see Table SI3.1) and, in case of the asylum issue, by the second-largest social distance (M = 1.77, SD = 0.77 with an asylum sample mean of M = 1.65, SD = 0.79; see Table SI3.2). In contrast, the resenters cluster exhibits the shortest internal distances in both contexts (climate: M = 1.53, SD = 0.79; asylum: M = 1.58, SD = 0.77), indicating greater homogeneity. A comparison of the demographics of polarized progressives and conservatives reveals that, across both issues, the progressives are slightly younger and have a lower household income. Furthermore, they are more likely to be female, to reside in larger cities, and—in the case of the asylum issue—to have a higher level of education than conservatives (Figs. SI3.1–SI3.6).

Intergroup contact

Is this structural separation reflected in a lack of intergroup contact? To address this question, we estimated ordinal logistic regression models using the frequency of contact with supporters and opponents as the dependent variables and cluster membership as the predictor (see Fig. 3 and Table SI4.1 for coefficient estimates and Fig. SI4.1 for predicted probabilities). The nonaligned cluster served as the reference category. For the climate issue, polarized progressives and conservatives reported significantly more contact with ingroup but not significantly less contact with outgroup members. For the asylum issue, both polarized clusters show significantly more contact with their ingroup and less with their outgroup. Resenters exhibit a distinct profile. Compared with the reference category, resenters reported significantly lower frequencies of contact with both target groups, i.e. supporters and opponents of progressive policies. Of these groups, only contact with those who opposed progressive climate policies did not reach statistical significance.

Fig. 3.

Ordinal logistic regression coefficients and 95% CIs for intergroup contact items, plotted by cluster and issue (climate: n = 2,477; asylum: n = 3,177). Coefficients (x-axis) reflect the effect of cluster membership (y-axis) on the likelihood of reporting a given frequency of contact (levels of the dependent variable: never/rarely/at least once a month/at least once a week/daily) or to have a given ego-network congruency (everyone has a different attitude than me/a majority have a different attitude than me/about half/a majority have the same attitude as me/everyone has the same attitude as me), relative to the reference category (intercept, nonaligned cluster). Positive coefficients indicate increased likelihood, while negative coefficients indicate decreased likelihood compared with the reference. The gray vertical line at zero marks the null effect. Asterisks are used to indicate significance levels (*P < 0.05, **P < 0.01, ***P < 0.001). The significance level 0.001 is smaller than the adjusted significance levels derived from the application of a Bonferroni correction to the level 0.05 for the three null hypotheses for each of the three models that were examined (0.05/9 = 0.005).

Ego-network congruency

Finally, we examined whether SSNs are reflected in congruency between the participants’ and their personal networks’ opinions on the issues. Using ordinal logistic regression (see Fig. 3 and Table SI4.1 for coefficient estimates and Fig. SI4.1 for predicted probabilities), we find systematic differences in ego-network congruency by cluster membership: Polarized progressives’ political networks are strongly congruent with their own positions on climate policies, whereas polarized conservatives’ and resenters’ positions are less aligned with their respective networks. For the asylum issue, in contrast, participants’ political networks reflect their own views more closely across all three clusters compared with the reference category.

Political behavior

Affective polarization is said to be detrimental to democracies because it undermines cross-cutting dialogue (62). Yet, when emotions run high, individuals are more likely to engage in talks with like-minded people (39) and in various forms of political action (47). We examine whether this holds for affectively polarized opinion-based groups, looking at the frequency of topical discussions with different types of interaction partners and participation in various forms of political engagement.

Topical discussions

Participants reported how frequently they discussed the climate or asylum issue with five types of social ties: family, friends, colleagues, acquaintances, and other individuals (see Fig. 4 and Table SI5.1 for coefficient estimates and Fig. SI5.1 for predicted probabilities). We estimated five ordinal logistic regression models using discussion frequency as the dependent variable and cluster membership as the independent variable. For the climate issue, polarized progressives reported significantly more frequent discussions with strong ties (family, friends) compared with the reference category. In the asylum issue, both polarized conservatives and resenters reported more frequent discussions with strong ties (family and friends). Interestingly, polarized progressives were less likely to discuss the asylum topic with family members—suggesting potential strain or avoidance in intimate settings (74). When examining frequency of discussion with weak ties, polarized conservatives and resenters were more likely to engage in topical discussions with colleagues and “other” individuals on asylum policies. Despite the coefficients being significant only at the 0.05 level, polarized conservatives in the climate domain consistently reported to discuss more frequently with weak ties. In contrast, polarized progressives showed no increased likelihood of discussing either issue with weak ties. In sum, polarized progressives are more likely to engage in discussions about the climate but not about the asylum issue with strong ties. The reverse is true in the case of polarized conservatives: they are more likely to discuss the asylum but not the climate issue with strong ties. Yet, polarized conservatives are more likely than the reference category to discuss both topics with weak ties. Once again, the behavioral profile of resenters more closely resembles that of polarized conservatives than of polarized progressives.

Fig. 4.

Ordinal logistic regression coefficients and 95% CIs for frequency of issue-related political discussions with different interaction partners, plotted by cluster and issue (climate: n = 2,477; asylum: n = 3,177). Coefficients reflect the effect of cluster membership on the likelihood of talking about the specific issue with interaction partners (levels of the dependent variable: never/rarely/at least once a month/at least once a week/daily), relative to the reference category (intercept, the nonaligned cluster). Positive coefficients indicate increased likelihood, while negative coefficients indicate decreased likelihood compared with the reference. The gray vertical line at zero marks the null effect. Asterisks are used to indicate significance levels (*P < 0.05, **P < 0.01, ***P < 0.001). The significance level 0.001 is smaller than the adjusted significance levels derived from the application of a Bonferroni correction to the level 0.05 for the three null hypotheses for each of the five models that were examined (0.05/15 = 0.003).

Political engagement

To assess whether issue-based affective polarization translates into political action, respondents indicated whether they engaged in a number of issue-related activities: (i) posting political content online, (ii) attending informational events, (iii) distributing campaign materials, (iv) signing petitions, (v) joining organizations advocating for the issue, (vi) participating in demonstrations, and (vii) engaging in civil disobedience (e.g. sit-ins, blockades). Given the binary nature of these outcomes and the rarity of some, we calculated seven ridge-penalized logistic regressions. The ridge penalty improves coefficient stability and reduces variance from sparse data (see SI6). Our results show that, compared with the reference category, issue-based affective polarization has a mixed effect on political engagement (see Fig. 5 and Table SI6.1 for coefficient estimates). Polarized progressives show very consistently significantly higher levels of political engagement on both issues. They are more likely to participate in almost all activities on both issues, especially in public and collective forms; the only exceptions are signing petitions and attending events for the climate issue. Polarized conservatives, in contrast, show lower levels of climate-related engagement—particularly in public or collective activities—but are more active when it comes to the asylum issue, especially in digital and solitary (i.e. distributing material and signing petitions) forms. Similarly, resenters show consistently lower engagement levels, particularly for climate-related activities. In the asylum issue, they only exceed the reference group in petition signing, with no significant differences observed in other areas.

Fig. 5.

Logistic regression with ridge-penalty coefficients and 95% CIs for likelihood of taking part in different forms of political engagement, plotted by cluster and issue (climate: n = 2,477; asylum: n = 3,177). Odds ratios (x-axis) reflect the effect of cluster membership (y-axis) on the likelihood of engaging in the given forms of political action (levels of the dependent variable: True = participant used this form, False = participant did not use this form), relative to the reference category (intercept; the nonaligned cluster). Odds ratios above zero points to a higher likelihood, coefficients below zero points to a smaller likelihood compared with the reference category (gray line). Asterisks are used to indicate significance levels (*P < 0.05, **P < 0.01, ***P < 0.001). The significance level 0.001 is smaller than the adjusted significance levels derived from the application of a Bonferroni correction to the level 0.05 for the three null hypotheses for each of the seven models that were examined (0.05/21 = 0.002).

Conclusion and discussion

Drawing on a large and innovative sample of politically engaged citizens, this study investigates affectively polarized intergroup relationships between two opinion groups structured around the issues of climate change and asylum. These groups are defined by participants’ support for—or opposition to—progressive policies and exhibit strong ingroup favoritism and outgroup antagonism. Importantly, they are simultaneously characterized by high perceived emotional alignment within the ingroup and distinct profiles regarding discrete intergroup emotions, dominated by anger, disgust, and contempt toward the political outgroup, and joy toward the political ingroup. Thus, affective polarization can be determined by perceived emotional alignment in addition to intergroup differentiation, thereby accounting for emotional experiences in self-categorization.

These emotional divides are embedded in broader social structures. Polarized individuals occupy distinct positions within a multidimensional Blau space, reflecting their alignment with specific sociodemographic niches. They experience less contact with individuals holding opposing political views and are embedded in more politically homogeneous networks. This structural segregation is further reflected in behavior: polarized individuals are more likely to engage in topical political discussions, particularly within close social ties, such as family and friends.

Our findings contribute significantly to the understanding of issue-based affective polarization. Supporting prior research (7, 10, 11), we show that citizens form affectively polarized opinion-based groups around salient issues and—extending beyond previous works—that affective polarization not only involves ingroup favoritism and outgroup dislike but also the belief that emotions are socially shared within groups. Perceived emotional alignment is a key component of self-categorization (25). It reinforces a sense of ingroup cohesiveness and fuels “us” versus “them” dynamics. Through personal experience and media-based representations, individuals may observe the typical emotional displays and appraisals of supporters and opponents on salient political issues and come to perceive them as their own. Emotions therefore serve not only to express social identities (75) but also contribute to creating and reinforcing group boundaries that subserve social identities (25) by communicating which political events and symbols are relevant to the ingroup. The present research demonstrates that this mechanism extends beyond situations of physical co-presence, e.g. at rallies (27) and experimental manipulation (25), to dispersed, opinion-based groups.

The analysis of intergroup discrete emotions further highlights the moral dimension of polarization, suggesting that polarized individuals may be characterized by emotional blends that combine contrasting approach and avoidance tendencies (21). On the one hand, polarized individuals report high levels of anger—an approach-oriented emotion linked to political mobilization (22, 32, 38, 75) and discussions with like-minded others (39). On the other hand, they report high levels of disgust and contempt, emotions associated with moral judgment and social avoidance (40, 41). These latter emotions indicate that outgroups are not merely perceived as political adversaries but as violating core moral norms. This emotional blend—marked by concurrent approach and avoidance tendencies—is consistent with the political, behavioral, and intergroup contact pattern we identified: polarized individuals tend to express anger in political conversations with strong ties and to concentrate their interactions within the ingroup, while feelings of disgust and contempt toward the outgroup are associated with reduced contact across group boundaries.

In line with our theoretical argument and recent evidence (47, 61), our findings demonstrate that affective polarization not only fosters ingroup contact but also spurs political engagement concerning climate change and asylum issues (30). However, this association is markedly stronger for the polarized progressive clusters in our sample, who are more active in particular concerning collective and public-facing noninstitutional activities. The polarized conservative clusters, in contrast, are more embedded in politicized interpersonal networks and engage more frequently in political discussions with both strong and weak ties. These differences point to asymmetric modes of political behavior across political boundaries. Such behavioral asymmetries have received limited attention in prior research (76) and may help explain the mixed evidence regarding the effects of affective polarization on noninstitutional forms of political engagement (47, 61).

Although it has positive effects on political participation, affective polarization becomes particularly problematic for democratic cohesion when aligned with nonpolitical divisions—such as income, education, or religion—a phenomenon known as social sorting (45, 48, 49). Our findings indicate such sorting: conservatives and progressives occupy distinct sociodemographic niches. These divisions increase the intensity of emotions (45) and, consequently, the likelihood of political conflict. They also decrease opportunities for cross-cutting contact. Social, affective, and political sorting thus appear mutually reinforcing, contributing to the stability and entrenchment of issue-based identities by creating the perception of aligned emotional responses to political events. This may also explain the limited effectiveness of many interventions targeting affective polarization: brief positive intergroup contact is unlikely to counteract the long-term effects of persistent exposure to homogeneous social environments (54, 55).

However, the results should be interpreted with caution. Our sample is not representative of the general population, and the policy statements used to distinguish supporters from opponents reflect politically progressive, clear-cut positions rather than symmetric “progressive” versus “conservative” partisan camps. Consequently, the differences we observe between supporters and opponents should not be interpreted as reflecting population-level polarization between conservatives and progressives. Taken together, these limitations define the scope of our conclusions but do not undermine the study's capacity to shed light on the mechanisms underlying affectively polarized intergroup relationships. Indeed, the study offers valuable insight into the interplay between emotional, structural, and interactional components of affectively polarized intergroup relationships among politically engaged individuals, a subgroup that is theoretically and empirically central to research on affective polarization (47, 61).

Within these limitations, our findings show that affective polarization is a multidimensional phenomenon, shaped not only by attitudes toward political opponents but also by perceived emotional alignment with like-minded others, reinforced by structural embedding and social interactions, and characterized by contrasting emotional drives of approach and avoidance. Understanding this interplay helps illuminate the depth and persistence of political division in contemporary democracies and may inform the development of more effective responses.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

This work was supported by the Berlin University Alliance under contract #111_MC-SocCoh_4 and the Einstein Stiftung Berlin under contract #ERU-2023-782.

Author Contributions

Diego Dametto (Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review & editing), Stefanie Hechler (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review & editing), and Christian von Scheve (Conceptualization, Data curation, Formal analysis, Supervision, Funding acquisition, Investigation, Methodology, Writing—original draft, Writing—review & editing)

Data Availability

All data and scripts are available at https://osf.io/taks9/.