1. Introduction
A growing number of studies suggest that, across different contexts, partisan group identities may divide citizens to a greater extent than social identities such as race, ethnicity, or religion (e.g., Iyengar et al., Reference Iyengar, Sood and Lelkes2012; Iyengar and Westwood, Reference Iyengar and Westwood2015; Carlin and Love, Reference Carlin and Love2018; Westwood et al., Reference Westwood, Iyengar, Walgrave, Leonisio, Miller and Strijbis2018; Martini and Torcal, Reference Martini and Torcal2019; Aldrich et al., Reference Aldrich, Bae and Sanders2024; Sleiman et al., Reference Sleiman, Melios and Dolan2025; but see Helbling and Jungkunz, Reference Helbling and Jungkunz2020). This tendency has possibly increased in recent years. As the comparative literature on affective polarization highlights, antagonistic feelings toward other parties represent now a global trend which has intensified over recent decades (e.g., Gidron et al., Reference Gidron, Adams and Horne2020; Garzia and Ferreira da Silva, Reference Garzia and Ferreira da Silva2025). Two main explanations have been advanced for the predominance of partisan over social identities in generating social divisions (e.g., Mason, Reference Mason2018; Westwood et al., Reference Westwood, Iyengar, Walgrave, Leonisio, Miller and Strijbis2018). First, partisan ties may partially absorb existing social identities, leading to an observed outcome where social identities matter less. Second, social norms of respect and tolerance may not apply to the same extent to groups defined by political characteristics as they do to groups defined by more inherent social traits. This may lead individuals to feel more encouraged to express hostility toward political rather than social out-groups.
Despite valuable past contributions of this literature, we argue it faces three main limitations. First, although using a variety of empirical strategies (such as observational designs, implicit association tests, or trust games), much of the existing work typically contrasts partisanship with a single social trait, thereby overlooking that most societies are structured around multiple salient social cleavages (e.g., Iyengar and Westwood, Reference Iyengar and Westwood2015; Carlin and Love, Reference Carlin and Love2018; Westwood et al., Reference Westwood, Iyengar, Walgrave, Leonisio, Miller and Strijbis2018). This approach risks oversimplifying social reality. Second, not only partisanship but also ideological orientations can function as important group identifiers, particularly in multiparty systems (e.g., Vegetti and Širinić, Reference Vegetti and Širinić2019). Yet previous studies have not sufficiently emphasized this distinction. Finally, the geographical scope of prior research remains limited largely to the US and Western Europe, which constrains the external validity of its findings.
Our study aims to unpack the contribution of different salient social and political group identities to group homophily, defined as the tendency to prefer social interactions with members of one’s own group (e.g., McPherson et al., Reference McPherson, Smith-Lovin and Cook2001). Homophily implies reduced cross-group interactions and, consequently, greater social distance (e.g., Wark and Galliher, Reference Wark and Galliher2007). This preference for in-group members may have significant social consequences, as it can lead to segregated social networks, foster out-group antagonism and discrimination—especially when the out-group is perceived as a threat to the in-group—and erode social tolerance, thereby undermining peaceful cohabitation and social cohesion (e.g., Mummendey and Wenzel, Reference Mummendey and Wenzel1999; Velthuis et al., Reference Velthuis, Verkuyten and Smeekes2021). The political consequences can also be harmful, since deep partisan divides may fuel political intolerance toward out-party supporters, undermine democratic norms, and, in extreme cases, even trigger political violence (e.g., Kingzette et al., Reference Kingzette, Druckman, Klar, Krupnikov, Levendusky and Ryan2021; Kalmoe and Mason, Reference Kalmoe and Mason2022; but see Broockman et al., Reference Broockman, Kalla and Westwood2023; Berntzen et al., Reference Berntzen, Kelsall and Harteveld2024).
We address the aforementioned limitations by, first, emphasizing that individuals typically hold multiple group identities rooted in both social characteristics and political affinities, and second, considering both partisanship and ideology as relevant political group identifiers. In addition, we extend the analysis to less-explored multiparty systems in Southern Europe and Latin America.
Our first expectation, consistent with prior research (e.g., Carlin and Love, Reference Carlin and Love2018; Westwood et al., Reference Westwood, Iyengar, Walgrave, Leonisio, Miller and Strijbis2018; Martini and Torcal, Reference Martini and Torcal2019; Helbling and Jungkunz, Reference Helbling and Jungkunz2020; Sleiman et al., Reference Sleiman, Melios and Dolan2025), is that group identities rooted in salient social and political traits fuel homophily and social distance. Second, as we further detail below, we posit that political identities may be more consequential than social ones in triggering social distance. Finally, we propose that these effects may vary according to individuals’ left-right orientations. Specifically, while right-wing individuals could be more sensitive to differences in social characteristics, left-wing individuals may be more sensitive to differences in political ones.
To test these conjectures, we conducted a conjoint experiment in five countries—Argentina, Chile, Italy, Portugal, and Spain.Footnote 1 This research design is particularly well suited for our goal of simultaneously assessing the impact of multiple key political and social characteristics on homophily and social distance. Specifically, respondents were asked to choose between two hypothetical neighbors moving next door, each varying in political and social characteristics. We focus primarily on two political traits—party and ideological affinities—and three salient social attributes—migration background, religious denomination, and education. By observing the extent to which respondents systematically favor certain neighbor profiles over others, we can assess their tendency to prefer social interactions with people who share similar characteristics and, by extension, have a weaker preference for those who differ. Crucially, the randomization of attributes enables us to test which specific social and political characteristics contribute more strongly to social closeness to in-groups and distance from out-groups. We note, however, that due to our theoretical focus on political identities (along social ones), our empirical analyses center on individuals whose partisan affinities and ideological leanings could be identified, necessarily leaving out respondents with no preferred party or ideological affiliation.
Our results show that both political and ideological affinities significantly impact the choices of individuals with partisan sympathies regarding their preferred neighbors, while social attributes have a weaker impact, with notable exceptions. These findings suggest that, for this group, political identities contribute more strongly to group homophily than other forms of social identity. Interestingly, we also find that partisan affinities have a stronger impact than ideology in almost all the countries we examined. Finally, our analyses suggest that left-wing citizens express greater social distance from neighbors supporting opposing parties than their right-wing counterparts, whereas right-wing individuals are more likely to favor neighbors who share their social characteristics.
Our paper is structured as follows. We first outline our theoretical argument and the empirical implications. Next, we present our research design. After detailing our main results, we conclude by discussing the implications of our findings.
2. Theoretical framework
Our theoretical argument proceeds stepwise. We first discuss how group identification based on either social or political traits can lead to homophily and social distance, building on scholarship drawing on the social identity theory (Tajfel and Tuner, Reference Tajfel, Tuner, Austin and Worchel1979) and the affective polarization research. Social and political identities do not exist in a vacuum, however. People may be bound by multiple relevant identities, be they social or political, a nuance that has been largely overlooked in prior research (for a recent exception, see Sleiman et al., Reference Sleiman, Melios and Dolan2025). Importantly, these identities can inform each other, or even trump one another, when it comes to shaping homophilic preferences. Accordingly, we discuss in a second step to what extent political identities may outweigh social ones in fostering social distance and whether these dynamics are more prevalent among specific voter constituencies defined in terms of ideological leanings.
2.1. Group identification and social distance
Homophily is a basic principle according to which “contact between similar people occurs at a higher rate than among dissimilar people” (McPherson et al., Reference McPherson, Smith-Lovin and Cook2001: 416). It implies that inter-group differences in social characteristics translate into social distance, typically conceptualized as “the degree of intimacy and understanding that exists between individuals or social groups” (Wark and Galliher, Reference Wark and Galliher2007: 389–390). According to social identity theory (Tajfel and Tuner, Reference Tajfel, Tuner, Austin and Worchel1979), people have a natural inclination to categorize the world into in-groups and out-groups and to positively distinguish their group from others. This inclination arises even under the most basic definition of a group (Tajfel et al., Reference Tajfel, Billig, Bundy and Flament1971) and intensifies with the strength of in-group identification (e.g., Leszczensky and Pink, Reference Leszczensky and Pink2019). Relatedly, in-group favoritism is driven not only by cognitive factors but also by motivational ones, such as the need for positive distinctiveness and the maintenance of personal self-esteem (e.g., Huddy, Reference Huddy2001).
While identities can be grounded in multiple social factors, the most significant are those social and political variables that have become central in contemporary societies. This makes identities rooted in these factors salient and predispose individuals to evaluate others based on differences and similarities along these traits, potentially generating in-group favoritism and social distance and, under conditions of perceived threat and inter-group competition, out-group derogation and discrimination (e.g., Mummendey and Wenzel, Reference Mummendey and Wenzel1999; Brewer, Reference Brewer and Crips2010).
To begin with, cleavage theory has long emphasized a bottom-up approach to political conflict. In this tradition, historical processes of socio-economic transformation—such as industrialization, modernization, or globalization—created new social divisions which fostered the crystallization of self-aware, socially distinct groups (e.g., workers and owners, or winners and losers of globalization) through patterns of everyday interactions and association (e.g., Lipset and Rokkan, Reference Lipset, Rokkan, Lipset and Rokkan1967; Hooghe and Marks, Reference Hooghe and Marks2018). Rooted in demographic and social characteristics such as ethnicity, religion, class, education, or gender (i.e., socio-structural conditions), these group identities have become socially articulated and given rise to enduring social cleavages, which, in turn, eventually informed political conflict.
Three social markers exert particular social and political relevance in today’s European politics and beyond. National identification and opposition to non-natives, such as immigrants and asylum seekers, are among the most enduring factors explaining the remarkable transformation of the European political space at the hands of radical right actors (e.g., Kriesi et al., Reference Kriesi, Grande, Lachat, Dolezal, Bornschier and Frey2006; Hooghe and Marks, Reference Hooghe and Marks2018). Immigrants are often perceived as an out-group by natives and can trigger nativist and anti-immigrant attitudes (e.g., Hangartner et al., Reference Hangartner, Dinas, Marbach, Matakos and Xefteris2019). While immigration has long been politicized in Europe (e.g., Hutter and Kriesi, Reference Hutter and Kriesi2022), it has also gained prominence in other regions, such as Latin America (e.g., Díaz et al., Reference Díaz, Rovira Kaltwasser and Zanotti2023).
At the same time, an education cleavage has been identified in several Western European countries, strongly structuring political conflict on the libertarian-authoritarian dimension (e.g., Stubager, Reference Stubager2009; Acemoglu, Reference Acemoglu2026). Finally, although secularization has long characterized many European and other societies (e.g., Goldberg, Reference Goldberg2020), religion remains significant for political behavior (e.g., Montero et al., Reference Montero, Segatti and Calvo2023), particularly when it comes to opposition to Muslim minorities (e.g., Helbling and Traunmüller, Reference Helbling and Traunmüller2018).
Group identity may be grounded not only in social but also partisan and other political traits. Political parties and elites do not merely mobilize pre-defined social groups. They also actively contribute to shaping group boundaries and identities (e.g., Carty, Reference Carty2020; Westheuser and Zollinger, Reference Westheuser and Zollinger2025). Partisanship is therefore seen by some scholars not only as an expression of policy preferences or instrumental considerations (e.g., evaluations of party performance), but also as an enduring social identity that affects individuals’ emotions, attitudes, and behavior (e.g., Huddy et al., Reference Huddy, Mason and Aarøe2015; Bankert, Reference Bankert2023).
A growing body of literature analyzes how partisan group identities fuel affective polarization among ordinary supporters of different parties, both in the US (e.g., Iyengar et al., Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019) and globally (e.g., Gidron et al., Reference Gidron, Adams and Horne2020; Reiljan, Reference Reiljan2020; Garzia and Ferreira da Silva, Reference Garzia and Ferreira da Silva2025). These partisan identities can trigger homophily, social distance, and discriminatory behavior in a variety of settings—for instance, when searching for potential dating partners (Huber and Malhotra, Reference Huber and Malhotra2017; Sleiman et al., Reference Sleiman, Melios and Dolan2025), interacting with neighbors (Chopik and Motyl, Reference Chopik and Motyl2016; but see Mummolo and Nall, Reference Mummolo and Nall2017), or in everyday economic decisions (McConnell et al., Reference McConnell, Margalit, Malhotra and Levendusky2018).
Left-right ideological labels may also function as group identifiers in contemporary democracies in ways that are not fully captured by partisanship. Several authors have argued that ideology encompasses not only a belief system but also affective attachments to ideological labels and to people and symbols associated with those labels (e.g., Vegetti and Širinić, Reference Vegetti and Širinić2019; Comellas and Torcal, Reference Comellas and Torcal2023). Ideological identity seems particularly relevant in multiparty systems, where a single ideological bloc typically encompasses several parties, and where some citizens may identify less with a particular party than with an ideological label (e.g., Garry, Reference Garry2007). Similar to partisan identity, ideological identity has been found to fuel political engagement and affective polarization, both in the US and in various multiparty systems (e.g., Mason, Reference Mason2018; Oshri et al., Reference Oshri, Yair and Huddy2022).
Building on the above, we first expect that salient social and political identities contribute to social distance from out-groups (general expectation 1). More concretely, individuals should have stronger preferences for interacting with co-partisans, co-ideologues, or, for example, co-nationals, thereby increasing social distance from those who support another party, hold a different ideology, or were born in another country.
At the same time, in multiparty systems—where parties are usually positioned along the left-right axis—not all out-parties are equally distant from partisans. Previous literature indicates that dislike toward out-parties and their supporters increases with ideological distance (e.g., Westwood et al., Reference Westwood, Iyengar, Walgrave, Leonisio, Miller and Strijbis2018; Van Erkel and Turkenburg, Reference Van Erkel and Turkenburg2022). Voters tend to like parties within their ideological bloc while disliking those from the opposing bloc (e.g., Bantel, Reference Bantel2023). We can therefore expect that social distance from supporters of an out-party will increase as the ideological distance from that out-party grows (general expectation 2).
2.2. Political affinities versus social traits
While group identities rooted in either political or social characteristics may lead to homophily and social distance from out-groups, the former may be more consequential. This is for two reasons. First, partisan (and ideological) identities may partially “absorb” existing salient social identities (e.g., Westwood et al., Reference Westwood, Iyengar, Walgrave, Leonisio, Miller and Strijbis2018: 334). This idea mirrors core insights from the cleavage theory discussed earlier, whereby partisan ties, nurtured by party elites, channel and partially integrate social cleavages within a given society (e.g., Lipset and Rokkan, Reference Lipset, Rokkan, Lipset and Rokkan1967; Westheuser and Zollinger, Reference Westheuser and Zollinger2025).
Recent research on social sorting highlights the increasing alignment of partisan identities with social ones (e.g., race, religiosity, class) across different national contexts (e.g., Harteveld, Reference Harteveld2021). This cumulative alignment can create a psychologically durable partisan social identity that intensifies perceived threats and competition among partisan groups, heightening both in-group sympathy and animosity toward out-parties and their supporters (Mason, Reference Mason2018). A similar dynamic can occur with ideological identities, given their capacity to integrate new issues and social divisions that emerge in contemporary politics (e.g., Steiner, Reference Steiner2024).
The second explanation for why political identities may be more consequential than social ones in triggering social distance stems from our current understanding of the role of prevailing social norms. Described as the “unwritten grammar of society” (Bicchieri, Reference Bicchieri2005: ix), social norms are informal rules dictating appropriate social behavior. While relationships between groups defined by physical or social traits are usually regulated by norms of respect and tolerance (e.g., Sigall and Page, Reference Sigall and Page1971; Iyengar and Westwood, Reference Iyengar and Westwood2015; Alvarez-Benjumea and Valentim, Reference Alvarez-Benjumea and Valentim2024), these norms may do substantially less to preclude intolerance toward groups defined by partisan or ideological identities. This could be because individuals tend to blame and hold others personally accountable for their partisan or ideological affiliations, but less so for their more inherent demographic or social characteristics. Accordingly, prejudiced attitudes and behaviors may be less stigmatized when directed toward the former than the latter. Moreover, political parties competing in the electoral arena often portray out-party views and associated out-groups in negative terms (e.g., Nai, Reference Nai2021), thereby signaling to voters that hostility toward political opponents is acceptable and fuelling affective polarization (e.g., Hernández et al., Reference Hernández, Anduiza and Rico2021; Martin and Nai, Reference Martin and Nai2024).
Accordingly, our third expectation is that political identities may outweigh other social identities in explaining social distance from out-groups (general expectation 3).
2.3. The conditioning role of individuals’ ideology
In the previous section, we proposed that political identities may be more consequential in triggering social distance than social identities. There are reasons, however, to believe that this may vary by individuals’ ideological orientations. In particular, the values underpinning right-wing and left-wing worldviews may predispose people toward different mechanisms of social categorization. On the one hand, conservative ideology is generally characterized by a preference for order, stability, and the maintenance of established social hierarchies, which entails an acceptance of social inequalities (e.g., Jost et al., Reference Jost, Glaser, Kruglanski and Sulloway2003; Jost and Van der Toorn, Reference Jost, Van der Toorn, Van Lange, Kruglanski and Higgins2012). For right-wing individuals, characteristics that signal conformity to established social hierarchies may then be particularly salient, whereas deviations from the existing social order tend to be interpreted as threats (e.g., Nilsson, Reference Nilsson2024).
Left-wing ideology, in turn, is characterized by a commitment to equality, pluralism, and social change (e.g., Caprara and Vecchione, Reference Caprara and Vecchione2018). In principle, the values of pluralism and equality should entail an acceptance of both individuals who differ in terms of social characteristics and of those who differ in terms of political views. However, from a social identity perspective, perceived threats to the social predominance of values shared by the in-group can give rise to exclusionary reactions. This implies that left-wing individuals may not only prefer interactions with in-group members, who are perceived to share similar values, but should express particularly high social distance from those out-groups perceived as strongly threatening the in-group’s values (e.g., Nilsson, Reference Nilsson2024). The latter is most visibly represented by the radical right parties and their partisans (e.g., Gidron et al., Reference Gidron, Adams and Horne2023).Footnote 2
Based on the above, we expect social distance from out-groups defined on the basis of social traits, such as ethnicity, nationality, religious identification, and so on, to be more pronounced among individuals with right-wing orientations. By contrast, social distance from political opponents, both partisan and ideological, is expected to be stronger among left-wing individuals, especially in relation to threatening radical right opponents (general expectation 4).
2.4. Hypotheses
As we further detail in the “Research design” section, we employ a conjoint experiment to assess our general expectations. Respondents were asked to choose between two hypothetical next-door neighbors differing on the three social markers discussed earlier (education, religion, and migration background), as well as on two political traits (partisanship and ideological leanings).
The three initial general expectations discussed above are translated into four specific hypotheses, as follows:
Hypothesis 1Observing a difference in partisan and ideological characteristics between oneself and the hypothetical neighbor decreases the likelihood of preferring that neighbor’s profile.
Hypothesis 2The negative effect of observing a difference in partisan traits between oneself and the hypothetical neighbor is stronger when the difference occurs across ideological blocs than when it occurs within the same bloc.
Hypothesis 3Observing a difference in social traits between oneself and the hypothetical neighbor decreases the likelihood of preferring that neighbor’s profile.
Hypothesis 4The effect of observing a difference in partisan and ideological characteristics between oneself and the hypothetical neighbor is greater than that of observing a difference in social traits.
Our fourth and final general expectation—namely, that left- and right-wing individuals differ in their sensitivity to differences in social and political characteristics—translates into the following hypothesis (divided into two sub-hypotheses):
Hypothesis 5.1The effect of observing a difference in social traits between oneself and the hypothetical neighbor is greater among right-wing individuals than among left-wing individuals.
Hypothesis 5.2The effect of observing a difference in partisan and ideological characteristics between oneself and the hypothetical neighbor is greater among left-wing individuals, particularly when the neighbor supports a radical party from the opposing bloc.Footnote 3
3. Research design
To test these hypotheses, we conducted conjoint experiments in five countries—Argentina, Chile, Italy, Portugal, and Spain. These experiments were embedded in the third wave of an online panel survey, which applied non-probability quota sampling to ensure that the national samples reflect the characteristics of the general population in terms of gender, age, and region. Section A of the Appendix provides details on the socio-demographic characteristics of the survey participants, as well as information on panel attrition.Footnote 4
The conjoint experiment asked each respondent 12 times to choose among 2 hypothetical profiles of neighbors that randomly varied on a set of attributes, including those discussed above. In addition to (1) party support, (2) ideology, (3) birth in versus outside country, (4) religion, and (5) education, which are at the center of our empirical analyses, other attributes refer to sexual orientation, region/national identity, environmental concern, level of politicization, and pet ownership, as well as opinion on vaccines (Italy only) and language (Spain only). In Section B.I of the Appendix, we provide the specific question wording of the conjoint experiment, along with the complete list of possible neighbors’ attributes and levels in each country. The results for these additional attributes, although not discussed at length in the text, are presented in the Appendix in Sections E.I and E.II.Footnote 5
Regarding the political attributes, the original conjoint presented respondents with specific party and ideological orientation labels (left, right, and center). In the case of Italy, these attributes included an additional level: “does not identify with any party” and “does not identify with any ideological label,” respectively.Footnote 6 We re-coded these attributes to capture in-group/out-group dynamics and test the hypotheses more carefully. Specifically, we re-coded the party support attribute as 0 when the hypothetical neighbor supports the same party as the respondent (see below on our specific operationalization of the respondent’s in-party), 1 when the neighbor supports another party within the respondent’s ideological bloc, 2 when the neighbor supports a moderate party from the respondent’s opposing bloc, and 3 when the neighbor supports a radical party from the opposing bloc. For Italy, the re-coded attribute includes an additional level 4 corresponding to neighbors with no party affiliation. Similarly, the ideology attribute was re-coded as 0 when the neighbor’s ideology matches that of the respondent, 1 when it corresponds to a different but non-opposing ideology, and 2 when it corresponds to the respondent’s opposing ideology. Specifically, the value of 1 refers to the center for respondents located on the left or right; by contrast, for centrist respondents, level 1 refers to the ideology of respondent’s preferred party.Footnote 7 Again, in the Italian case, the re-coded ideological attribute includes an additional level 3 referring to neighbors with no ideological identification.
We identified respondents’ partisan in-group based on their most liked group of voters (for further details, see Section B.II in the Appendix). This allows us to go beyond the classical measure of partisanship relying on party identification per se and include all respondents who display some preference for a particular group of partisans (see Wagner, Reference Wagner2021). As our empirical analysis focuses solely on respondents for whom a particular preferred party can be identified based on this specific measure, our analytical samples are smaller than the full country panel samples. Specifically, respondents without a preferred party (that is, who do not display some preference for a particular group of partisans) excluded from the analysis represent 20.83% of the total national sample in Argentina, 34.78% in Chile, 28.49% in Italy, 23.98% in Portugal, and 28.74% in Spain.
With respect to the social attributes, we re-coded them following the same in-group/out-group logic. First, migration background is re-coded so that 0 means the neighbor was born in the country and 1 means the neighbor was born outside the country.Footnote 8 Second, religion is re-coded as 0 when the hypothetical neighbor’s religious affiliation matches that of the respondent, 1 when it corresponds to a different affiliation (non-Muslim), and 2 when it corresponds to a Muslim affiliation. Since the Muslim level is not included for Argentina and Chile, in these two countries, the religion attribute only has two levels (respondent’s religion and another religion).Footnote 9 Finally, the education attribute (university versus non-university) is re-coded as 0 when the neighbor’s education level matches that of the respondent, and as 1 when it differs. For further details, see Sections B.II, B.III, and C in the Appendix.
This conjoint design aligns closely with our research goals and expectations, and represents an improvement over previous studies that, for example, relied on trust games (but see Sleiman et al., Reference Sleiman, Melios and Dolan2025, for a similar conjoint design). First, the proposed conjoint analysis allows us to identify and estimate the causal effects of a broader range of political and social group-based characteristics. Second, the conjoint design is particularly well suited to capture in-group favoritism and social distance through respondents’ systematic choices between pairs of hypothetical neighbors. To the extent that certain profiles are consistently preferred over others, we can identify the specific in-groups to which respondents express greater social closeness and, by extension, the out-groups to which they express greater distance. Third, this method also helps mitigate potential social desirability bias (i.e., the possibility that some respondents hide their social preferences for certain groups due to social pressures), since the experiment is conducted online and respondents are presented with multiple attributes, providing them with several justifications for any given choice (e.g., Hainmueller et al., Reference Hainmueller, Hopkins and Yamamoto2014; see also Horiuchi et al., Reference Horiuchi, Markovich and Yamamoto2022). This also diminishes the potential learning effects often associated with repeated trust games involving different players.
3.1. Case selection
We test our hypotheses in five countries from Southern Europe and Latin America, namely Argentina, Chile, Italy, Portugal, and Spain.
These five countries were selected because they meet several conditions particularly relevant to our study. First, they have been less explored in previous research on this topic and encompass two different world regions, allowing us to assess whether the identified patterns hold across a range of different contexts, thereby strengthening the external validity of our findings. Second, over the past decade, all selected countries have undergone significant transformations in their party systems following major social and political disruptions, including the emergence of new political forces and increased political polarization (e.g., Torcal and Comellas, Reference Torcal and Comellas2022; Moncagatta and Silva, Reference Moncagatta and Silva2024). These developments should make political identities particularly salient as bases for social categorization. Third, their party systems span the entire ideological spectrum, including significant radical left, center-left, center-right, and radical right parties. However, they differ in the electoral relevance of their most radical parties, the timing of their emergence, and the role these parties have played in government formation. For a more detailed explanation of their party systems, see Appendix, Section D.
4. Results
Figure 1 reports the average marginal component effects (AMCEs) of the five key attributes of interest, which indicate how much the probability of choosing a hypothetical neighbor would change on average if one of the neighbor’s attributes were switched from the reference category to the particular attribute level (Hainmueller et al., Reference Hainmueller, Hopkins and Yamamoto2014). Importantly, the reference category for each of these attributes corresponds to the scenario of congruence between the respondent’s characteristic and that of the hypothetical neighbor.Footnote 10 The full set of results is displayed in Section E.I. in the Appendix, whereas marginal means are shown in Section E.II.Footnote 11
Average marginal component effects with 95% confidence intervals.
Figure 1 Long description
The plots display average marginal component effects with 95 percent confidence intervals for five countries: Argentina, Chile, Portugal, Spain and Italy. The horizontal axis is labeled Estimated AMCE, ranging from negative 0.4 to 0.2. The vertical axis lists attributes: PARTY, IDEOLOGY, IMMIGRATION, RELIGION and EDUCATION. Each plot shows dots with whiskers representing confidence intervals. Argentina: Notable positive effects are seen for Radical party opposing bloc and Respondent′s ideology, while negative effects are observed for Born outside country and Other religion. Chile: Positive effects are seen for Moderate party opposing bloc and Respondent′s ideology, with negative effects for Born outside country and Other religion. Portugal: Radical party opposing bloc shows positive effects, while Born outside country and Other religion have negative effects. Spain: Positive effects are noted for Moderate party opposing bloc, with negative effects for Born outside country and Other religion. Italy: Radical party opposing bloc and Respondent′s ideology show positive effects, while Born outside country and Other religion have negative effects. Across all countries, attributes like Born outside country and Other religion consistently show negative effects, while Respondent′s ideology often shows positive effects. Colors differentiate attribute groups, aiding interpretation.
Consistent with H1, the results show that observing a difference in partisan and ideological characteristics between the respondent and the hypothetical neighbor decreases the likelihood of preferring that neighbor’s profile in all five countries, as indicated by the negative AMCEs. In other words, partisan and ideological identities appear to drive social distance. Moreover, such incongruence on the partisan trait exerts stronger negative effects on choice probability as ideological distance increases. For example, in Spain, a neighbor who supports a radical party from the respondent’s opposing bloc is, on average, 31.3 percentage points less likely to be chosen than a neighbor who supports the respondent’s preferred party. This negative effect amounts to 19.4 percentage points when the neighbor supports a moderate party from the opposing bloc, and 17.4 percentage points when the neighbor supports another party within the respondent’s bloc. Therefore, the results are also consistent with H2.
Interestingly, the negative impact of incongruence in party support is stronger than that of differences in ideological orientation in all countries except Chile. The results for Chile align with the dramatic decline in party identification in this country since the 1990s, as reported by De la Cerda (Reference De la Cerda2022), a trend that may have been accentuated by the weakening of traditional parties following the 2019 social uprising and the emergence of new parties and coalitions (Somma and Donoso, Reference Somma and Donoso2022).
For social attributes, the effects generally go in the expected direction, confirming H3. Specifically, incongruence in terms of religious denomination between the respondent and the hypothetical neighbor decreases choice probability in all five countries. In the Southern European countries that we focus on (Italy, Spain, and Portugal), we find the greatest social distance toward Muslims. Neighbors with an immigration background are also less favored in Italy, Portugal, and Chile—although levels of social distance from these profiles appear lower than those from Muslims in the first two countries, and lower than those from neighbors of different religious affiliations in the latter. However, a word of caution is warranted regarding this attribute. Specifically, since the profiles presented to respondents indicate that the hypothetical neighbor born abroad supports a specific party, respondents might have inadvertently assumed that the profile corresponds to a naturalized citizen—or even to a national born abroad. Such assumptions might have muted underlying concerns about immigration. Further research should delve into this aspect more carefully.
Finally, incongruence in education level decreases preferences for the hypothetical neighbor in Argentina, Italy, and Portugal, although the effects are relatively weak. In Chile and Spain, education has no significant impact on profile favorability. Overall, religious denomination emerges as the most consequential social trait fuelling social distance, with Muslim affiliation exerting the strongest effect in the three European countries. The stronger effects of this attribute might also explain why we find smaller effects for the attribute tapping into immigrant background.
With regard to H4, differences in political or ideological affinities generally have a stronger effect on reducing choice probability than differences in education, religious denomination, or immigrant background. This is particularly the case when the hypothetical neighbor allegedly supports a radical party from the respondent’s opposing bloc or, to a lesser extent, holds opposing ideological views. In other words, with few exceptions, political identities appear to fuel social distance to a greater extent than the social identities considered in this study, a result that supports H4. Since our focus is on individuals who manifested partisan and ideological leanings—as opposed to non-partisans or non-ideologues—this finding suggests that, within this constituency, social identities may be subordinated to political ones.
Still, the estimated negative AMCEs for Muslim affiliation on profile preferences are somewhat stronger than those for “other ideology” in Spain (although still weaker than the effects of all party support attributes) and equal to or stronger than those for “opposing ideology” and “other party (same bloc)” in Italy and Portugal. Similarly, in Argentina the AMCEs for “other religion” are somewhat stronger than those for “other ideology,” and in Chile, they are equal to or stronger than those for “other party (same bloc)” and “moderate party (opposing bloc).” These findings slightly qualify our H4.
Next, we assess the extent to which left- and right-wing respondents differ in their favorability toward groups demarcated by political and social identities (H5.1 and H5.2). To this end, we conducted sub-group analyses by computing marginal means separately for left- and right-wing respondents, and then taking the difference between the two groups (see Leeper et al., Reference Leeper, Hobolt and Tilley2020). These analyses allow us to capture potential treatment heterogeneity effects—i.e., whether some groups of respondents (here, left- versus right-wing individuals) respond differently to the treatments.
The results are shown in Figure 2, with statistically significant differences at p < 0.05 marked in red (for full information about the group-specific marginal means, see Section E.III in the Appendix). We classify left-wing respondents as those scoring below 5 on the 0 (left)—10 (right) ideological self-placement scale, and right-wing respondents as those scoring above 5.
Differences in marginal means between right-wing and left-wing respondents.
Figure 2 Long description
Forest plot titled, AR: Diff R vs L, illustrating estimated differences between right-wing and left-wing groups across various subgroups. The x-axis is labeled, Estimated Difference, with units not shown, ranging from minus 0.06 to 0.06. The y-axis lists categories: Radical party (opposing bloc), Moderate party (opposing bloc), Other party (same bloc), Respondent′s preferred party, Opposing ideology, Other ideology, Respondent′s ideology, Born outside country, Born in country, Other religion, Respondent′s religion, Other education level, Respondent′s education level. Points with horizontal intervals represent estimated differences, likely with confidence intervals. Several points lie to the right of zero, indicating positive differences, particularly in ideology categories. Image B shows a forest plot titled, CH: Diff R vs L, with the x-axis ranging from minus 0.10 to 0.10. Similar categories are listed on the y-axis. Points show positive differences in party and ideology categories, with intervals often crossing zero. Image C shows a forest plot titled, PT: Diff R vs L, with the x-axis ranging from minus 0.10 to 0.10. Points indicate positive differences in party and ideology categories, with some intervals crossing zero. Image D shows a forest plot titled, ES: Diff R vs L, with the x-axis ranging from minus 0.10 to 0.10. Positive differences are noted in party and ideology categories, while negative differences appear in immigration and religion categories. Image E shows a forest plot titled, IT: Diff R vs L, with the x-axis ranging from minus 0.10 to 0.10. Positive differences are observed in party and ideology categories, while negative differences are noted in immigration and education categories. Across all panels, intervals often cross zero, indicating variability in subgroup effects. Notable extremes include positive differences around plus 0.05 and negative differences around minus 0.08.
Regarding social attributes (H5.1), right-wing respondents tend to be less favorable to having foreign-born neighbors than left-wing respondents, but only in Spain and Italy. Concerning the religion attribute, we do not observe significant differences (at p < 0.05) in the marginal means between left- and right-wing respondents. At p < 0.1, however, right-wing respondents in Spain are less favorable than left-wing respondents to having a Muslim neighbor, while in both Italy and Spain they are more favorable to neighbors sharing their own religion. Finally, education appears to matter in Italy and Chile (although in the latter case at p < 0.1), but in the opposite direction: right-wing respondents are less favorable than left-wing respondents to having persons with the same education level as neighbors, and more favorable to neighbors with a different education level, although the magnitude of this difference is small. Overall, these results provide partial support for H5.1. The expected pattern, whereby right-wing individuals prioritize social characteristics as a basis for in-group favoritism more than left-wing individuals, is evident in some contexts for migration background and religion, consistent with our theoretical argument that these traits are particularly salient for those who value social order and established hierarchies.
Regarding political attributes (H5.2), a consistent pattern, in line with our expectations, emerges across all countries except Argentina:Footnote 12 favorability toward neighbors who support radical parties from the opposing bloc is higher among right-wing respondents than among left-wing respondents. In other words, left-wing individuals display greater social distance from supporters of radical parties from the opposing bloc. This is consistent with our theoretical expectation that, for left-wing individuals, political identities function as the primary signal of normative alignment, making radical political opponents the main trigger of out-group distance. Neighbors supporting moderate parties from the opposing bloc are also less favored by left-wing respondents compared to right-wing respondents, though this is only observed in Chile and Spain, suggesting that the asymmetry is most pronounced when ideological distance is largest.
At the same time, favorability toward neighbors who support other parties from the respondent’s own ideological bloc is lower among right-wing respondents than among left-wing respondents. In other words, right-wing respondents appear to exhibit greater social distance from supporters of other right-wing parties. This is particularly the case in Spain, Portugal, and Chile. One plausible interpretation is that moderate right-wing individuals in these countries—all of which share a recent right-wing authoritarian past—may have internalized a social stigma around radical right-wing positions, since these are perceived as counter-normative and potentially harmful to democratic values (e.g., Harteveld et al., Reference Harteveld, Dahlberg, Kokkonen and van der Brug2019). This pattern, while not directly predicted by our theoretical framework, is consistent with the broader argument that social norms shape the expression of in-group favoritism.
When we examine ideology as a basis of in-group versus out-group dynamics, we find again that favorability toward neighbors with opposing ideology is higher among right-wing respondents than among left-wing respondents (with the exception of Argentina). This finding parallels the results for the party support attribute. Our data also reveal that favorability toward neighbors with the same ideology (hence, in-group) is in fact lower among right-wing respondents than among left-wing respondents, again with Argentina as the only exception. Taken together, these results provide support for H5.2, confirming that left-wing individuals rely more heavily on political dimensions, both partisan and ideological, as the primary basis for social distance.
Regarding the labels “no party” and “no ideology” specific to the Italian case, respondents tend to prefer a neighbor who supports their preferred party (or shares their ideology) over a non-partisan (or someone without an ideological identity). This difference in favorability is similar to that between a neighbor who supports the respondent’s party (or ideology) and a neighbor who supports another party within the same bloc (or holds a different, but not opposing, ideology).
Finally, the full AMCE results displayed in Figures E1–E5 in the Appendix portray a more nuanced picture. In Portugal and Italy, environmental orientations produce effects comparable to those of ideological incongruence. In Italy, vaccine attitudes generate some of the strongest social distance effects observed in the study, possibly reflecting the salience of COVID-related social norms (Graupensperger et al., Reference Graupensperger, Abdallah and Lee2021). These patterns highlight the potential role of issue-based group identities, which we leave for future research. Additionally, sexual orientation also appears to be consequential in Chile.
5. Robustness checks and extensions
To test the robustness of our results, we conducted additional analyses presented in Section E.V of the Appendix. First, repeating the analysis using only the first five tasks yields similar results (Figure E17). Second, excluding individuals with high in-party affinity (ratings above 6) supports our initial findings, with some reduction in partisan effects, particularly in Chile (Figure E18). Third, models weighted to correct for panel attrition show no relevant differences compared to unweighted models (Figures E19 and E20, which also provide additional details on weight construction and results). Fourth, excluding centrist respondents leaves the ideology attribute results largely unchanged (Figure E21).
Apart from these robustness tests, we also conducted three extensions. First, we explored whether the accumulation of differences in political and social traits may further contribute to social distance.Footnote 13 To assess this, we conducted average component interaction effects (ACIEs) within each country, specifically interacting party support and ideology incongruence with migration background, religion, and education, respectively. This resulted in six interactions between pairs of political-social attributes. In total, we performed 30 separate analyses shown in Figure E22. Most of the interaction terms do not reach statistical significance at a p < 0.05, and among those that do, the sign of the interaction term generally contradicts our expectations. Overall, we do not find evidence supporting a cumulative effect of incongruence in political and social characteristics on social distance.
Second, we tested whether the negative effect of differences in education level on respondents’ neighbor choices is stronger among respondents with a university degree than among those without one. Previous literature suggests that identification with the educational in-group is asymmetric and stronger among the highly educated (Stubager, Reference Stubager2009). The results of the sub-group analyses presented in Figures E23–E27 confirm that, in all five countries, respondents with a university education tend to prefer neighbors with the same level of formation over those with only basic schooling. Among respondents without a university education, there is also a preference for neighbors with a university degree in Argentina and Chile, while in the selected Southern European countries, they are equally likely to choose neighbors with either a university education or basic education.
Third, we examined the social preferences of respondents without a preferred partisan group, our measure of party preference (recall that these respondents are excluded from the analyses reported above). Figure E28 shows, country by country, these respondents’ preferences concerning hypothetical neighbors supporting different parties. In Portugal and Spain, we find that social distance is greatest toward supporters of the radical right (and some nationalist parties in Spain), slightly exceeding that toward hypothetical Muslim neighbors. In Italy and Chile, social distance from some out-partisans (on the right in Italy and at both ideological extremes in Chile) is comparable to that toward out-groups defined by social traits. By contrast, in Argentina, preferences do not differ across partisan groups; the only relevant differences are observed across religious denominations. Overall, non-partisans display notable social distance from supporters of extreme parties (particularly on the right). We replicated the analyses for respondents without an ideological identity (Figure E29), finding significant social distance from ideological groups only in Chile and Portugal.
6. Discussion and conclusions
This study aimed to expand previous research on how group partisan and social identities may trigger social divides by: (1) considering multiple social and political identities, including how they might inform or trump each other; (2) assessing the extent to which preexisting ideological predispositions condition these dynamics; and (3) extending research to less explored multiparty systems. A conjoint experiment was fielded in Argentina, Chile, Italy, Portugal, and Spain to test the causal effect of observing differences and similarities between respondents and hypothetical neighbors on a range of political and social attributes (party affinity, ideological orientation, migration background, religious affiliation, and education level) on homophily and social distance.
The main results indicate that, among respondents for whom we were able to identify a preferred party and ideological label (as well as information on religion and educational level), political and ideological characteristics strongly influence social distance across all five countries, while the impact of social characteristics is generally weaker but goes in the expected direction. Additionally, we find that the effect of ideological identities is generally weaker than that of partisan affinities.
These findings align with previous research conducted in other national contexts using different designs and methods (e.g., Iyengar and Westwood, Reference Iyengar and Westwood2015; Westwood et al., Reference Westwood, Iyengar, Walgrave, Leonisio, Miller and Strijbis2018), confirming the growing prevalence of a partisan/ideological cleavage. The stronger effects of partisan and ideological identities may stem from their capacity to channel the influence of social differences. We therefore encourage future research designs that explicitly capture mediation effects, enriching both our theoretical and empirical understanding of this topic.
The results reported in this study have important implications for social cohesion. Previous research has shown that political conflicts and divisions between parties may spill over into everyday relationships among partisan supporters and ideologues, even leading to the formation of partisan neighborhoods (Chopik and Motyl, Reference Chopik and Motyl2016), which might foster greater segregation and isolation and also affect social trust (Torcal and Thomson, Reference Torcal and Thomson2023). Our results suggest a similar dynamic, although caution is warranted regarding the extent to which reported social preferences translate into real-world behavior (see Mummolo and Nall, Reference Mummolo and Nall2017).
Our findings also highlight a conundrum. Expressing antipathy toward political out-groups may be less stigmatizing than expressing antipathy to social out-groups, especially in contexts with strong anti-discrimination norms (Iyengar and Westwood, Reference Iyengar and Westwood2015). At the same time, social traits tend to be more visible or inferable from everyday interactions (e.g., migration background and religion), whereas individuals can keep their partisan and ideological leanings private. Ultimately, the dynamics of social distance and segregation may depend on which group identities are made visible and politicized. Where people choose to disclose their political preferences in public settings, such as social media, social segregation along political lines may be reinforced.
In addition, our analyses reveal a significant pattern of in-group favoritism and social distance among left-wing citizens toward right-wing supporters, particularly those aligned with radical right parties. This aligns with research showing that left-wing individuals exhibit stronger animosity and a greater reluctance to engage socially with radical right supporters (e.g., Harteveld et al., Reference Harteveld, Dahlberg, Kokkonen and van der Brug2019). Since our findings come from five countries that all share right-wing authoritarian legacies, we encourage further research in other contexts (e.g., post-communist countries) to assess potential scope conditions.
The implications of these findings point to important trade-offs. On the one hand, strong resistance from the left to engaging with radical right supporters could serve as a protective measure, preventing the normalization of discriminatory discourses and behaviors and limiting the spread of harmful ideologies (e.g., Alvarez-Benjumea and Valentim, Reference Alvarez-Benjumea and Valentim2024). On the other hand, this stance risks further deepening the ideological divide, potentially making constructive dialogue across the political spectrum increasingly difficult. Such polarization may contribute to an environment where dialogue is limited, political conflicts are more likely to escalate, and the stability and inclusiveness of democratic processes are challenged.
Finally, a notable finding is the significant impact of specific “issue traits” (e.g., environmentalism in Portugal or vaccination in Italy) on respondents’ social preferences. This aligns with prior research showing that issue preferences can also trigger affective polarization (e.g., Mason, Reference Mason2018; Hobolt et al., Reference Hobolt, Leeper and Tilley2021). Unfortunately, our data do not capture social identity dynamics in these instances, as we lack information about respondents’ views on these issues, but further research is encouraged.
This study has several limitations. First, further research is required to better understand the reasons behind some of the cross-country differences in our results, such as the prominence of ideological divisions in Chile and the fact that Argentina stands out as the only case where favorability toward neighbors who support radical parties from the opposing bloc does not differ between left- and right-wing respondents. Moreover, to the extent that social divides are channeled by political ones, it is also important to further understand why we still observe effects for certain group identities (such as those based on migration background or education level) in some countries but not others.
Second, while we have strived to improve external validity, important questions remain regarding the differences in the variety and salience of social identities across countries and over time. Future research should also examine whether the patterns observed in our experimental setting translate into actual residential sorting behavior, going beyond the hypothetical scenario we examine here. Finally, this study focuses primarily on partisans (i.e., those expressing some affinity for a party). An additional exploratory analysis of our conjoint experiment suggests that non-partisans also tend to exhibit significant social distance toward supporters of radical (right-wing) parties. Future research on social and political polarization should also include a comprehensive analysis of non-partisans, using appropriate research designs, as they represent increasingly important segments in many contemporary democracies.
Funding statement
M.T. acknowledges support and funding from the ICREA ACADEMIA Excellence Award and “Excelencia María de Maeztu” CEX2025-001632-M, MICIU/AEI/10.13039/501100011033. This research has been made possible with the funding of the Ministerio de Economía y Competitividad, under grant number PID2019-106867RB-I00 (2020–2024). J.M.C. (josepmaria.comellas@uab.cat) acknowledges support and funding for her research through a Margarita Salas Fellowship, funded by European Union-NextGenerationEU, Ministry of Universities and Recovery, Transformation and Resilience Plan, through a call from Pompeu Fabra University (Barcelona). A.V. (alina.vranceanu@uab.cat) acknowledges financial support for her research through a Ramón y Cajal Fellowship, grant RYC2023-044435-I financed by MCIU/AEI/10.13039/501100011033 and FSE+.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2026.10144.
Declaration of interests
The authors declare no competing interests. None of the authors has any financial, personal, or professional relationship that could be perceived as influencing the research reported in this article.
Open access
