RESEARCH DOSSIER
Affective Polarization and Partisan Discrimination
Abstract. Affective polarization describes antagonistic evaluations across political-party boundaries; partisan discrimination concerns differential choices toward people bearing party cues. This critical narrative research monograph reconstructs the social-identity history of the field and integrates experimental, survey, cross-national, digital-platform, and intervention research. It distinguishes interpersonal aversion from ideological disagreement, elite evaluations, discriminatory behavior, democratic norm violations, and support for political violence. The evidence establishes party-cue effects under controlled conditions and multiple strategies for reducing expressed animosity, while showing that effects on institutional conduct and democratic attitudes must be tested separately. Measurement targets, political systems, contact duration, sampling, and causal identification determine the limits of inference. This work synthesizes published scholarship; it reports no new experiment, preregistered systematic review, or quantitative meta-analysis.
Keywords: affective polarization; partisan discrimination; partisan animosity; outparty hostility; intergroup bias; political social identity; feeling thermometers; social distance; partisan sorting; party cues; behavioral discrimination; democratic norms; depolarization interventions; cross-partisan contact; comparative political psychology
Introduction: Interparty Affect as a Psychological Research Object
Affective polarization concerns how membership in a political party becomes a basis for emotional preference, distrust, avoidance, and sometimes unequal treatment of other people. In its most studied form, respondents favor members of their own party and evaluate supporters of another party with suspicion or hostility. The object is relational: what matters is the evaluation of an identifiable political outgroup, the contrast with the ingroup, and the contexts in which those evaluations influence conduct. In an experiment, discrimination becomes observable when otherwise comparable partners, applicants, or proposals receive different treatment after a party cue is manipulated. The distinction makes attitudes, perceptions, and actual choices separately measurable outcomes. Iyengar et al. (2012) Iyengar & Westwood (2015)
The present monograph examines the history of this construct, partisan social identity, ideological sorting, feelings about parties and voters, methods of measuring bias, experimental party-cue effects, stereotypes, media settings, comparative party systems, and interventions. Its boundary runs through interpersonal political animosity and differential treatment. Group discussion that shifts consensus, political violence, generalized prejudice, and authoritarian beliefs are relevant neighboring phenomena whose causal connections must be tested rather than assumed. A partisan may dislike an opposing politician without discriminating against an opposing voter, or oppose that voter's policies while behaving fairly toward them. Those distinctions organize the analysis throughout. Druckman & Levendusky (2019) Broockman et al. (2023)
This is a critical narrative research synthesis of identifiable original studies, experiments, methodological analyses, and reviews. The evidence discussed ranges from survey time series and cross-sectional comparisons to randomized messages, encounters, party-label cues, and platform interventions. Studies answer different questions depending on their design, outcomes, sampling frames, and timing. Cross-sectional correlations can suggest mechanisms; randomized assignment of party information can identify an effect of that information on a measured choice; experimental treatment of animosity can evaluate whether changing feelings changes another outcome. None of these designs alone establishes every link in a proposed causal chain. Iyengar et al. (2019) Jost et al. (2022)
The central scientific conclusion developed here is that partisan identity is capable of organizing social perception and selective behavior beyond issue disagreement, while the downstream consequences are heterogeneous. It follows that research needs a family of related outcomes rather than one global polarization score. A precise model specifies the political categories available in a society, the people to whom attitudes refer, the decision context, the mechanism potentially changing behavior, and the period over which any effect persists. Such specification permits genuine comparison between United States findings, multiparty settings, postreferendum divisions, and digitally mediated encounters. Reiljan (2020) Voelkel et al. (2024)
The Formation of the Affective Polarization Research Program
Work on political partisanship long emphasized party identification, electoral choice, ideology, and spatial distance between policy preferences. The affective turn drew attention to a different observation: citizens can become socially estranged across party lines even when the distribution of particular issue positions does not show equivalent divergence. Iyengar, Sood, and Lelkes's 2012 article explicitly contrasted issue-based polarization with increasingly negative interparty evaluations and social distance. Its title, Affect, Not Ideology, functioned as a research-program challenge to the assumption that polarized opinion requires a corresponding expansion of policy disagreement. Iyengar et al. (2012)
The 2012 contribution also connected political divisions with the older social-psychological problem of categorical ingroups and outgroups. Political labels can summarize reputations, moral expectations, and perceived similarity; their behavioral influence may be amplified by campaigns that repeatedly represent the other party as dangerous or untrustworthy. Historically, this explanation did not replace work on actual ideology. It asked researchers to estimate the relative contributions of identity, evaluative hostility, and issue distance using measures designed for each. The published article subsequently received a correction concerning the description of certain party-identification questions, an example of why long-run historical interpretations must retain attention to instrument wording. Iyengar et al. (2012) Iyengar et al. (2019)
Iyengar and Westwood's 2015 study expanded the research program from survey sentiment toward explicit, implicit, and behavioral indicators. In controlled contexts, party cues influenced judgments and allocation choices that were not themselves elections. This was significant because an interpersonal choice involving an opposing partisan offers a stronger test of discrimination than a respondent's general statement that the opposing party is unlikable. Yet the experiments should be interpreted as evidence within their recruited populations, cues, and decision environments. An experimentally observed penalty is neither an estimate of every workplace nor an automatic prediction of every real-world hiring decision. Iyengar & Westwood (2015)
Subsequent work developed the difference between strong partisan identity, social sorting, political contempt, and democratic consequences. Huddy, Mason, and Aarøe in 2015 tested expressive party identity and emotional mobilization; Mason's 2015 analysis described how identities aligning across social and ideological lines can intensify social antagonism despite limited issue movement. The 2019 synthesis by Iyengar and colleagues consolidated a literature on sources and consequences of outparty dislike, while later measurement studies questioned whether a familiar thermometer question actually targets ordinary opposing voters or elites. This progression changed the field from one headline trend into a program of construct validation and causal investigation. Huddy et al. (2015) Mason (2015) Iyengar et al. (2019) Druckman & Levendusky (2019)
Conceptual Boundaries: Disagreement, Identity, Animosity, and Discrimination
Ideological polarization denotes distance in political issue preferences, policy packages, or ideological locations. Affective polarization denotes emotionally negative evaluation of a political outgroup, commonly considered together with ingroup warmth. Partisan sorting denotes increasing alignment between party identification and other ideological or social identities. Partisan discrimination denotes unequal evaluation or treatment conditional on another person's partisan identity, ideally demonstrated with equivalently situated targets. These are connected constructs with different empirical indicators. One cannot infer behavioral discrimination from a thermometer gap, and one cannot infer identical ideology from shared party membership. Mason (2015) Iyengar & Westwood (2015)
The referent of dislike is indispensable. A question about the Democratic or Republican Party can activate thoughts about officeholders, activist organizations, campaign messaging, or ordinary supporters. Druckman and Levendusky showed that different established measures do not behave interchangeably and that respondents can hold more negative opinions of opposing-party elites than opposing-party citizens. This means that a reported 'rise in polarization' may partly reflect changed elite evaluations, changed category salience, or changes in the respondent's interpretation of the item. A sound instrument therefore identifies whether the target is a party organization, politician, politically active partisan, average voter, or interpersonal partner. Druckman & Levendusky (2019)
Political identity can operate as an expressive commitment, a decision heuristic, a marker of social belonging, or a record of issue preferences. These functions need not coincide. A voter may use party as a source cue while keeping little emotional attachment to the organization; another may feel strong identity-based threat despite relatively flexible policy opinions. The expressive model tested by Huddy, Mason, and Aarøe suggests that threats and reassurances to party standing can mobilize anger and enthusiasm beyond responses explained by individual issue positions. Such experimental findings support motivational mechanisms while leaving room for instrumental party considerations under different conditions. Huddy et al. (2015)
Hostility itself is multidimensional. Coldness, distrust, contempt, perceived moral inferiority, reluctance to befriend, and willingness to allocate less money to another partisan are related observations with potentially different causes. Finkel and colleagues' account of political sectarianism highlights othering, aversion, and moralization; the framework is a useful interpretive map but cannot substitute for direct measurement of each element. Scholars should distinguish privately experienced aversion, publicly expressed condemnation, reported intentions, and consequential conduct. The strongest claim available from a study follows its observed outcome, not the most alarming implication that could be imagined from it. Finkel et al. (2020) Broockman et al. (2023)
From Individual Evaluation to Intergroup Relations
Affective polarization exists in social relations as well as in personal attitudes. People encounter opponents through family life, workplaces, social media, institutions, and electoral campaigns. Exposure may reveal common human characteristics, signal antagonistic norms, or increase perceptions of disagreement, depending on the encounter's design. A survey answer elicited during a campaign is therefore a snapshot of a participant's evaluation under specific informational conditions. Research must examine how repeated interactions, perceived audience expectations, and significant political events alter the accessibility and expression of partisan labels over time. Iyengar et al. (2019) Jost et al. (2022)
A causal account needs to specify at least four links: the cues that activate partisan categorization, the beliefs and feelings associated with those cues, the decision processes that translate them into choices, and the feedback that may reinforce future evaluations. Hostility can increase selective contact, which changes the information available about opponents, which then influences subsequent hostility. But such feedback is a hypothesis about dynamic relations unless measured longitudinally or manipulated. In many datasets, the same individuals select both their social contacts and their media environments, complicating causal separation. Ahler & Sood (2018) Bail et al. (2018)
This distinction clarifies why societal rhetoric is an imperfect measure of everyday discrimination. Public claims about opposing voters may be exaggerated, strategic, or directed at elites, while private interactions can be cooperative. Conversely, superficially civil relationships can include biased allocations when party labels are visible. Evidence should accordingly combine different units of analysis: individual responses, dyadic conversations, hiring or resource choices, campaign communications, and institutional rules. The construct becomes more precise when researchers locate the behavior in a known context rather than treating political tension as one homogeneous psychological state. Druckman & Levendusky (2019) Iyengar & Westwood (2015)
Within the Artificial Era, human interaction with conversational systems offers an additional test bed for the effects of partisan cues. Randomized assignments of apparent party endorsement in otherwise equivalent artificial messages could isolate user responses to displayed affiliation, while longitudinal designs could test whether perceived similarity changes trust and information selection. Such proposals concern human psychological responses to artificial interlocutors; stable psychological organization in an artificial system would require distinct operational evidence. No claim about demonstrated machine partisanship follows from studies of human political social identity. Jost et al. (2022) Guess et al. (2023)
Measurement: Feeling Thermometers, Social Distance, and Trait Judgments
The conventional feeling thermometer asks respondents to rate an object along a cold-to-warm continuum. For a partisan, the gap between evaluations of the ingroup and outgroup can approximate relative affective polarization. Its interpretation depends on whether researchers analyze outparty coldness alone, the ingroup–outgroup difference, the absolute gap, or combinations across parties. Two respondents can have identical gaps despite different warmth toward their own party and different hostility toward the opponent. A thermometer is a measure of expressed evaluation; it does not establish that the respondent will discriminate in a consequential encounter. Iyengar et al. (2012) Druckman & Levendusky (2019)
Social-distance measures ask about relationships such as friendship, residence, family ties, or marriage. They have a longer sociological ancestry and can reveal how strongly party membership is perceived as an obstacle to personal association. But discomfort with a hypothetical relationship differs from the social opportunities actually available to respondents. Marriage aversion may incorporate expectations about household conflict, religion, geography, age, or family approval, and an unanticipated party cue may alter how such factors are considered. The construct requires validated items suited to the research question, not the assumption that any adverse answer measures the same latent emotion. Iyengar et al. (2012) Mason (2015)
Trait batteries ask whether opposing partisans are selfish, reasonable, intelligent, honest, or trustworthy. These measures capture stereotypes and inferred dispositions, but they can blur differences between moral condemnation, expectations of strategic behavior, and interpersonal warmth. Druckman and Levendusky's instrument comparison underscores that party referents and item formats matter: respondents may picture elites rather than ordinary citizens and express different intensities of hostility toward them. A robust battery should include explicitly named target populations, balanced positive and negative items, and tests of item functioning across time and party groups. Druckman & Levendusky (2019) Iyengar et al. (2019)
Instrument validity is an empirical question. Convergent validity asks whether conceptually related affect measures correlate; discriminant validity asks whether they can be distinguished from ideology, ideology-related fear, candidate approval, political efficacy, and general distrust. Predictive validity asks whether expressed affect anticipates a separately measured choice. Repeated panels can evaluate change, while test–retest studies should consider whether a campaign event genuinely changed opinion between waves. Researchers should report scale reliabilities, subgroup distributions, alternative operationalizations, and the degree to which a conclusion survives changes in the affect index. Druckman & Levendusky (2019) Reiljan (2020)
A 2026 measurement synthesis by Druckman, Klar, Krupnikov, Levendusky, and Ryan makes the concept-to-indicator match a primary methodological requirement. Researchers studying evaluations of party elites, feelings about ordinary partisans, partisan social avoidance, or differential treatment should construct measures for that specific outcome, describe the target population, and test sensitivity to alternative instruments. The availability of shared replication data makes it possible to test which established scales generalize across samples and which give diverging answers. This contemporary work reinforces the dossier's central distinction between measured affective evaluations and behavioral discrimination. Druckman et al. (2026) Druckman & Levendusky (2019)
Party-Cue Experiments and Behavioral Discrimination
Randomized party cues provide one of the clearest approaches to estimating partisan discrimination. A researcher presents otherwise comparable candidates, partners, or recipients with different partisan labels and records evaluations or allocations. Randomization breaks systematic association between the assigned label and observed attributes under the experimental conditions. When the resulting choices differ, the evidence supports a causal effect of the cue on that outcome. The underlying psychological pathway can include dislike, perceived moral character, anticipated reciprocity, cultural similarity, or inference about unobserved preferences; distinguishing these pathways requires further experimental contrasts. Iyengar & Westwood (2015)
Iyengar and Westwood used explicit and implicit indicators along with behavioral decision tasks to demonstrate that party affiliation can influence nonpolitical judgments. Their finding that partisan discrimination was substantial in studied tasks was historically influential because it extended the phenomenon beyond abstract political evaluations. Comparisons with discrimination involving race in those tasks drew attention to differences in the social norms governing openly expressed bias. Such results are task-specific estimates, however: the meaning of a political cue, the historical status of protected groups, and the costs of a decision vary considerably across settings. A comparison of numerical effects in one experiment is not a universal hierarchy of social harms. Iyengar & Westwood (2015)
Behavioral discrimination can be operationalized as a difference in ratings, opportunities, hiring recommendations, resource allocations, or actual treatment. These outcomes vary in incentive compatibility. A survey vignette with no consequences may be vulnerable to expressive responding, whereas a consequential choice can involve competing material motives. Strong designs randomize labels within a factorial structure that also varies competence, stated policy preference, and nonpolitical similarity. Outcomes should be recorded before respondents are asked to rationalize their choice, and manipulation checks should distinguish failed recognition of party information from deliberate refusal to use it. Iyengar & Westwood (2015) Broockman et al. (2023)
The transport of party-cue effects into real institutions requires separate evidence. Employment decisions may involve legal protections, professional norms, organizational procedures, repeated interaction, and high stakes. Online choice tasks may reveal dispositions but differ from legally regulated hiring or long-term collaboration. Field experiments, administrative data, audit designs, and incentivized lab tasks can supply complementary information where ethically appropriate. Bias estimates should be accompanied by the population, decision environment, randomization unit, precision, and plausible channels of selection; a general declaration that partisans 'discriminate' does not identify how widely a particular mechanism operates. Iyengar & Westwood (2015) Jost et al. (2022)
Implicit Measures, Expressive Responding, and Behavioral Convergence
Implicit association procedures can detect relative associations between political labels and evaluative categories under timed-response conditions. Their attraction is that respondents cannot simply provide the same form of deliberated verbal rating as in a survey. Yet a reaction-time contrast has multiple determinants, including familiarity, task demands, relative category frequencies, and learned semantic associations. It is useful as an additional observation about automatic responding, while its interpretation as individual prejudice requires a validated mapping from task performance to the latent construct. Implicit results should be interpreted alongside explicit judgments and observed choices. Iyengar & Westwood (2015)
Expressive responding arises when respondents use a survey to signal allegiance, moral condemnation, or group loyalty instead of reporting an exact private belief. In political research, exaggerated ratings can be genuine communicative acts, even when their magnitude differs from a person's likely behavior. Researchers can compare sensitive items with incentivized accuracy tasks, indirect question formats, and repeated measures to evaluate how much expressive incentives contribute. The party-composition experiments by Ahler and Sood indicate that misperceptions of party membership are substantial and not adequately explained by simple survey-performance artifacts. Accordingly, neither 'all hostility is sincere' nor 'all extreme reporting is performative' is a defensible default. Ahler & Sood (2018)
Convergence across methods strengthens inference only when methods truly supply independent constraints. If a vignette, thermometer, and adjective rating are all administered in one session with highly salient partisan primes, shared response demands can induce convergence. If an independently randomized party label predicts actual resource allocation in a different context, the evidence covers a wider behavioral domain. Measurement designs should document order effects, survey wording, social desirability pressures, and the age and organizational meanings of party identifiers. Cross-validation on withheld samples and across elections can help reveal instruments whose apparent stability is produced by a single moment's context. Druckman & Levendusky (2019) Iyengar et al. (2019)
Researchers should also separate between-person correlations from within-person change. People with more negative affect may make more discriminatory choices because of stable ideology, exposure history, personality, or group networks. An intervention that makes the same individual temporarily warmer toward opponents tests a different causal relationship. When improved warmth has no reliable effect on a consequential decision, this is evidence against a simple mediation account under the specified conditions, even if the population-level association remains substantial. This distinction becomes particularly consequential in studies of democratic norms. Broockman et al. (2023) Voelkel et al. (2023)
Cross-National Measurement and Multiparty Systems
Most early research used the two-party context of the United States, where a respondent can often be classified by feelings toward Democrats and Republicans. In a multiparty system, the outgroup may include several parties whose ideological distance, coalition history, and salience differ. Evaluating only the most disliked party can exaggerate polarization relative to a measure weighting every relevant party. Conversely, collapsing all competitors into one outgroup can hide consequential animosity toward the major coalition alternative. Cross-national work therefore requires explicit definitions of which outgroups count and how they are weighted. Reiljan (2020)
Reiljan introduced an Affective Polarisation Index for multiparty contexts and compared 22 European democracies with the United States using data from 2005 through 2016. The contribution is methodological as well as substantive: comparative inference depends on estimating weighted evaluations across a party system rather than importing an unmodified binary ingroup–outgroup contrast. Comparisons also face differences in party systems, coalition politics, respondent identification, survey languages, and electoral institutions. A valid cross-national index should report which parties were included, how party size or preferences entered the weighting, and whether thermometers were functionally comparable. Reiljan (2020)
Political divisions also arise around referendums and identities not equivalent to formal party membership. The British Leaver–Remainer divide provided a context for contact experiments in which the opposing group is defined by referendum position. Such research usefully tests whether familiar intergroup mechanisms apply outside conventional party boundaries, but changes the object being measured. A respondent's referendum identity may be embedded within geographic, generation, or constitutional conflict differently from an enduring party affiliation. Comparative studies should model these identity bundles rather than assuming labels have identical psychological meaning. Tausch et al. (2024)
Cross-cultural comparisons are strongest when they use invariant target descriptions, transparently document translation and item design, and test measurement equivalence. A lower average outparty rating in one country may reflect scale response tendencies, differences in partisan attachment, or a genuinely hostile environment. Longitudinal comparisons must consider changing party alignments and new political organizations, which can alter whom participants imagine when they read a familiar label. A theory of partisan discrimination should predict effects for identifiable individuals under locally meaningful cues; it need not predict identical gap scores in all national survey instruments. Reiljan (2020) Druckman & Levendusky (2019)
Evidence Quality, Replication, and Causal Interpretation
The evidence base contains descriptive time series, cross-sectional surveys, panel studies, vignettes, randomized experiments, platform field interventions, and reviews. Each design identifies a different estimand. Survey trends can document the evolution of a measure in a sampled population but require stable question wording and sampling procedures. Panel data can establish within-person temporal patterns yet remain exposed to time-varying confounding. Cue randomization establishes an experimentally identified treatment contrast, while a measure of implicit association indexes task performance. Research reports should describe precisely what an estimated difference represents and whom it represents. Iyengar et al. (2019) Druckman & Levendusky (2019)
Statistical significance alone provides little information about social consequence. Confidence intervals, effect magnitudes, base rates, outcome prevalence, and the durability of treatment effects matter. A small shift in warmth may be meaningful in a large population, whereas an identical numerical change on a low-prevalence violence endorsement scale can have a different interpretation. Preregistered analysis protects against some researcher degrees of freedom, but still depends on construct-valid instruments, sample coverage, and a suitable control condition. The 2024 megastudy's distinct outcomes and numerous treatments illustrate the need to compare effects within a transparent shared protocol. Voelkel et al. (2024)
Replication includes repeating a manipulation with another sample, changing institutions, testing delayed outcomes, and transporting interventions between political contexts. A failure to replicate an immediate conversational effect one month later may indicate temporal decay rather than invalidate the short-term mechanism. Likewise, an unchanged political-attitude measure after a platform feed alteration can be a meaningful null result even if on-platform exposure was altered. Null findings must be interpreted in relation to achieved statistical power, measurement sensitivity, and the breadth of plausible effects excluded by the confidence interval. Tausch et al. (2024) Guess et al. (2023)
This monograph reports a structured critical narrative assessment, not a systematic review with exhaustive registered search, dual-screening protocol, or pooled meta-analysis. The cited research is selected because it represents foundational theory, core methods, influential experimental contrasts, comparative contexts, and pivotal contrary findings. Conclusions therefore distinguish convergent patterns in the reviewed studies from universal prevalence claims. Future systematic synthesis should preregister databases and coverage, document study selection, code outcome definitions and risks of bias, and estimate heterogeneity separately for attitudes, interpersonal behavior, and political-institutional choices. Iyengar et al. (2019) Broockman et al. (2023)
Partisan Sorting, Stereotypes, and Perceived Social Distance
Partisan sorting aligns political identity with other social categories and reduces occasions when a person experiences potentially conflicting loyalties. The mechanism matters because a party label can come to signify many things at once: expected moral values, educational experience, cultural preferences, geographic origin, and social networks. A cue that appears to convey political alignment can therefore also change beliefs about an unfamiliar person's likely interests and traits. Sorting does not prove that every partisan is stereotypical; it provides a social environment in which a party label becomes an especially dense predictor in observers' minds. Mason (2015) Iyengar et al. (2019)
Ahler and Sood's research on perceptions of party composition demonstrates another layer: people substantially exaggerate the presence of party-stereotypical demographic groups. Their original observational and experimental evidence indicates that such distorted perceptions are consequential and cannot be fully reduced to unwillingness to give accurate survey answers. In theoretical terms, a partisan can dislike a caricatured opponent whose characteristics do not match the actual distribution of party members. Correcting that caricature can alter affect, but the effect depends on whether hostility was generated by inaccurate beliefs, moral condemnation, personal experiences, or other sources. Ahler & Sood (2018)
Social distance is both an attitude and a potential generator of further misunderstanding. People who avoid members of an opposing group encounter fewer opportunities to learn about within-group diversity; the resulting social environment can make extreme portrayals seem plausible. Yet avoidance may also be the outcome of preexisting geographic and cultural segmentation. Observed friendship networks therefore provide ambiguous evidence about direction of causation. Designs that randomize accurate information, assign interaction opportunities, or track network change over time can separate at least some of these possibilities. Such research needs baseline contact, identity strength, and exposure opportunity recorded before a treatment is administered. Iyengar et al. (2012) Levendusky (2018)
An important comparative hypothesis is that outparty animosity rises when elite rhetoric and social sorting jointly make a rival party seem homogeneous and threatening. This does not entail symmetric mechanisms, equal magnitudes, or invariant effects in every party system. Comparative researchers should model institutional incentives and the social meaning of coalition possibilities, while experimentalists identify whether a given stereotype affects personal judgment after actual issue preference is controlled. The outcome of interest is the effect on evaluations or treatment of ordinary people carrying party cues, rather than the descriptive truth of any political program. Jost et al. (2022) Reiljan (2020)
Information Environments, Social Media, and the Limits of Exposure
Communication research commonly considers whether partisan news, elite statements, social networks, and platform ranking reinforce evaluations of opponents. It is tempting to infer that mere exposure to divergent opinions creates tolerance or that homogeneous feeds inevitably create animosity. Both claims require experimental evidence. Social platforms simultaneously select what appears in feeds, encourage users to choose whom to follow, and reward particular expressive behavior. A study of posts encountered is therefore partly a study of the platform's ranking rules and partly of the user's previous preferences. Observational correlations between platform use and polarization cannot uniquely identify an algorithmic effect. Bail et al. (2018) Guess et al. (2023)
Bail and colleagues compensated United States partisans to follow automated accounts exposing them to political messages from the opposing side. Their 2018 randomized field study found that exposure could increase rather than uniformly decrease measured political polarization, with treatment effects dependent on studied groups and outcomes. This result challenged the simple contact-as-information model: opposing messages delivered through an adversarial, one-directional political feed differ from respectful interpersonal contact. Their findings establish the response to that intervention in its study setting rather than a universal rule that all cross-party exposure increases hostility. Bail et al. (2018)
A large platform experiment during the 2020 United States election assigned consenting Facebook and Instagram participants to reverse-chronological feeds instead of the default algorithms. Guess and colleagues reported substantial changes in platform use and content exposure but no statistically significant changes in issue or affective polarization over the three-month observation period. This null finding is crucial to causal interpretation. A major platform-level intervention can alter information consumption without producing an observed shift on the specified political-attitude measures. A different platform design, historical moment, population, or longer duration could yield other results; those possibilities do not erase the experiment's evidence about its actual treatment. Guess et al. (2023)
The contrast between these studies demonstrates the importance of specifying the intervention. Following opposition-message bots is not the same treatment as replacing algorithmic ranking with chronological ordering. Both differ from dialogue with a personally encountered opposing voter. Measures also differ: attitude extremity, outgroup warmth, exposure diversity, and discriminatory selection may respond along distinct trajectories. Future experiments should cross exposure content with presentation mode, opportunities for reciprocal interaction, perceived credibility, and individual attachment strength, then record immediate and sustained outcomes. Designs should avoid equating a user's engagement metric with a change in hostility. Bail et al. (2018) Guess et al. (2023) Santoro & Broockman (2022)
The original 2023 platform-feed article has associated formal correction notices published in 2024 and 2026. The later notice addresses supplementary exposure and engagement classifications involving Instagram carousel posts; those revisions matter when interpreting precisely what types of content participants encountered. The existence of corrections does not itself establish a reversal of the preregistered attitude-outcome findings. A cumulative assessment should consult the corrected article and supplements, preserve the versions of measures being compared, and separate changes to descriptive exposure metrics from changes to causal estimates of political attitudes. Guess et al. (2023) Science erratum (2024) Science erratum (2026)
Motivated Reasoning, Group Threat, and Moralization
People evaluate political information through commitments involving identity, moral values, beliefs about institutions, and concerns about the legitimacy of social arrangements. Jost, Baldassarri, and Druckman's cognitive–motivational framework integrates self-, group-, and system-justifying processes with social communication. It helps researchers ask whether a particular hostility response emerges from perceived threats to group standing, defensive evaluation of inconvenient evidence, or broader preferences for institutional order. These mechanisms must be discriminated with tailored manipulations and separate indicators; observing a negative outparty rating does not tell us which motive produced it. Jost et al. (2022)
Emotion may be triggered by the anticipated victory or loss of one's party, especially among strongly identified partisans. Huddy, Mason, and Aarøe's experiments demonstrate the capacity of group-status communications to elicit anger and enthusiasm beyond the influence of holding strongly consistent policy positions. Those emotions can orient attention, motivate campaign involvement, and shape evaluations, but the evidence concerns the specified political-emotion and engagement outcomes. Whether a moment of anger produces unfair resource allocation requires additional observation. Scholars should therefore treat mobilization, hostility, and discrimination as potentially connected steps rather than equivalent measures. Huddy et al. (2015) Iyengar & Westwood (2015)
Political moralization can convert disagreement over specific policies into judgments about opponents' entire character. Finkel and colleagues describe political sectarianism using the elements of othering, aversion, and moralization. The formulation identifies an important possible mechanism: an opposing view can be treated as evidence of a defective or threatening person rather than a contestable public position. Yet the component parts still require measurement, and the framework should not be reified into one universal cause of democratic decline. For example, a morally intense attitude can motivate civic participation while simultaneously increasing reluctance to cooperate with opponents. Finkel et al. (2020)
Institutions affect whether rivalry is expressed through regulated disagreement, constructive compromise, avoidance, or hostile exclusion. A political system can generate meaningful collective conflict without reproducing every pattern observed in the American two-party environment. Researchers should examine candidate incentives, group threat messages, coalition structures, and media cues as contextual moderators of psychological processes. Such work can identify mechanisms whose expression depends on a system's arrangements without claiming an invariant national character or treating all disagreement as pathological. Jost et al. (2022) Reiljan (2020)
Political Accountability, Democratic Norms, and Causal Separation
A major public concern is that dislike of political opponents may prompt citizens to excuse wrongdoing by their own leaders, support undemocratic candidates, reject legitimate election results, or tolerate violence. These outcomes are substantively distinct from social-distance ratings. A person can report intense hostility while insisting on institutional rules that constrain their own party. Another can express moderate warmth but endorse an undemocratic proposal for strategic reasons. It follows that the relevant causal question is whether changing interparty affect changes institutional judgments while other influences are controlled, rather than whether two attitudes correlate at one time. Broockman et al. (2023) Voelkel et al. (2023)
Broockman, Kalla, and Westwood reported five experiments involving a total of 12,341 participants in their published study of affective polarization and democratic outcomes. Interventions capable of shifting interpersonal evaluations did not yield corresponding evidence of change across many measured political judgments, including aspects of democratic accountability. Their conclusion places an important limit on narratives that assign broad political outcomes to a single hostility construct. The experiments do not establish that affect never matters or that democratic institutions are immune to hostile group dynamics; they demonstrate that large shifts in measured affect can occur without detectable shifts in the particular downstream outcomes studied. Broockman et al. (2023)
Voelkel and colleagues similarly investigated interventions reducing affective polarization and found that changes in negative partisan feelings did not necessarily translate into improvements in antidemocratic attitudes. This distinction is compatible with the idea that interparty warmth and democratic norm support have partly different determinants. Political institutions, strategic calculations, false beliefs about opponents, perceived stakes of elections, and commitment to procedures may shape democratic judgments apart from interpersonal sentiment. Analyses should measure each pathway directly and avoid treating an effective depolarization intervention as a proven democratic reform in the absence of downstream behavioral evidence. Voelkel et al. (2023)
At the same time, the later 25-treatment megastudy showed that interventions sometimes reduced both partisan animosity and certain antidemocratic outcomes. Such findings do not contradict experiments that found narrower effects, because treatments target different mechanisms and outcomes. Correcting beliefs about rivals' commitments to democratic principles may affect political judgments through a route different from making an opponent appear personally sympathetic. The key research task is to identify which treatments work for which outcomes, populations, and time horizons. Success on one indicator should remain an explicitly delimited finding even when related indicators change in a favorable direction. Voelkel et al. (2024)
Comparative and Social Consequences of Partisan Discrimination
Discriminatory selection based on party labels can undermine cooperation across local, professional, and community settings if party information is routinely used to infer trustworthiness or shared purposes. The magnitude of this risk depends on opportunities for encounters, the visibility of party labels, stakes, social norms, and the behavior of organizations. Controlled evidence demonstrates that party cues can matter, but generalization requires direct measurement of interaction outside electoral research. A comprehensive research program should therefore compare hypothetical vignettes, incentivized exchange, real-world audit designs, and longitudinal interpersonal outcomes using separately defined estimands. Iyengar & Westwood (2015)
Patterns observed in the United States require comparison with different institutional structures. Reiljan's multiparty analysis demonstrates the importance of measuring affinity and hostility across several competing parties rather than assuming one opponent. Coalition ties may permit affinity with parties outside a respondent's preferred organization, while intense hostility may concentrate on one ideologically distant competitor. The consequences of discrimination are accordingly shaped by whether people expect political rivals to become collaborators in governing institutions and whether social norms discourage public animosity. Comparable psychological processes can generate different aggregate indices as the party system changes. Reiljan (2020)
Discrimination is especially important when political cues contaminate decisions nominally governed by other criteria. If a participant penalizes a competent applicant solely because of perceived party affiliation, the observed outcome reveals a problem of equal evaluation. Yet in some tasks party affiliation may have informational relevance to the stated role, making it necessary to identify which attributes were experimentally held constant. A careful audit of party discrimination records the decision rule and whether ideological knowledge, prospective behavior, and role requirements were controlled. The burden of inference is on a study's design rather than on a label attached after the result. Iyengar & Westwood (2015) Jost et al. (2022)
Research on the social consequences of affective polarization should also record counterexamples. Many political opponents remain colleagues, relatives, neighbors, and cooperative interlocutors. A positive interaction may weaken stereotypes, though its persistence and transfer to new outgroup members are empirical questions. Conversely, intervention success among relatively moderate participants may conceal resistance among the most committed. Recognizing heterogeneity sharpens the substantive object: the science studies conditions under which political classifications alter psychological evaluation and behavior, including conditions that preserve fair treatment despite disagreement. Santoro & Broockman (2022) Tausch et al. (2024)
Common Identities, Corrective Information, and Contact Interventions
Interventions can target partisan animosity by altering the category through which opponents are perceived, correcting false beliefs, or creating constructive interpersonal encounters. These approaches address different mechanisms. Common-identity framing increases the salience of a larger group to which both sides belong; stereotype correction modifies beliefs about the composition or motivations of an outgroup; contact permits participants to observe another person's conduct directly. A treatment's apparent success depends on its comparator, the timing of outcome assessment, the specific affect measure, and whether effects transfer beyond the people or messages encountered. Levendusky (2018) Ahler & Sood (2018) Wojcieszak & Warner (2020)
Levendusky's study of American identity used experiments and natural-event comparisons involving national celebrations to examine whether a superordinate identity could soften judgments of opposing partisans. The results support the proposition that shared membership can temporarily alter the relative salience of rival partisan categories. The mechanism is conditional: a national identity cue is credible only to the extent that participants accept the common category and do not interpret it as another exclusionary signal. No general guarantee follows that shared-identity primes can resolve entrenched institutional conflict or change discriminatory behavior across every domain. Levendusky (2018)
Wojcieszak and Warner examined forms of interparty contact, including friendship correlations and experimental vicarious or imagined contact. Such studies are valuable because contact can be indirect, symbolic, or personally experienced and the mechanisms may differ. A friendly portrayal of an opposing voter may weaken an extreme stereotype, while reciprocal discussion may alter perceived trust. Causal conclusions require attention to which forms were randomized and which relationships were merely observed. Friendship patterns in survey data cannot by themselves identify the effect of friendship because people select contacts partly according to preexisting dispositions. Wojcieszak & Warner (2020)
Across these approaches, outcome specificity matters. A participant might become more willing to have a conversation yet remain unwilling to support an opposing-party candidate. Another might revise perceptions of opposing voters' demographic composition without changing their own ideological commitments. Mechanism tests should therefore pair a manipulation with direct mediator measurement and multiple outcomes, including warmth, perceived similarity, trust, avoidance, and behavior. Longer follow-up is essential when policy discussions, family relationships, or workplace contact are expected to produce durable change. Ahler & Sood (2018) Santoro & Broockman (2022)
Randomized Cross-Partisan Conversations and Durability
Santoro and Broockman's randomized experiments paired political opponents for video conversations on assigned topics. Their findings illustrate the promise of interpersonal contact, especially when conversation avoids the most divisive areas of disagreement, while identifying limits in persistence and downstream democratic outcomes. The design isolates an intervention that participants actually experienced, rather than asking whether they would hypothetically enjoy meeting an opponent. At the same time, participants willing to enter such a study and speak to a stranger may differ from those who refuse. Effects estimated for volunteers in a video-call protocol require further transport testing. Santoro & Broockman (2022)
An encounter can change immediate affect through personalization, perceived politeness, and improved understanding without permanently altering the social environment that made hostility plausible. After the call, participants return to political networks, news sources, and group norms that may reinforce established categories. A short-term gain is meaningful evidence of malleability, but persistence must be measured with delayed outcomes and account for attrition. Repeated contact, structured cooperation, and opportunities to correct stereotypes may plausibly increase durability; such proposals require experiments that compare doses and control conditions rather than extrapolation from a single session. Santoro & Broockman (2022) Wojcieszak & Warner (2020)
Tausch and colleagues evaluated a facilitated contact intervention between self-identified Leavers and Remainers after Brexit. Sixty dyads comprising 120 participants were randomly assigned to an interaction condition or comparison interaction. The intervention improved some immediate affective and compromise-related measures, but longer-term effects were not statistically significant at follow-up; selective attrition further complicated generalization, especially for more extreme participants. The study provides a useful boundary condition: political contact can change specific interpersonal outcomes in the near term while durability and the inclusion of highly committed opponents remain unresolved. Tausch et al. (2024)
Intervention research should distinguish an effect of conversation content from an effect of being listened to, meeting a respectful stranger, or participating in a research task. Stronger designs can compare neutral cross-party talk, political disagreement, same-party conversation, and a noncontact control, keeping time and social attention comparable. Measurements collected both before and after interaction, coupled with delayed follow-up, clarify whether a change persists and generalizes to unfamiliar outgroup members. Such experiments should report participant refusal, dropout by baseline ideology, and adverse or reactance responses, because successful contact among self-selected participants may conceal different population-level effects. Santoro & Broockman (2022) Tausch et al. (2024)
Large-Scale Intervention Evidence and Distinct Policy Outcomes
The 2024 megastudy led by Voelkel evaluated 25 interventions in a common online survey-experimental framework involving 32,059 participants. In preregistered analyses, 23 of 25 treatments significantly reduced partisan animosity; six reduced support for undemocratic practices, and five reduced support for partisan violence. These are numbers of treatments achieving the specified statistical criteria, not proportions of all possible political interventions that work. The results underscore the value of directly contrasting multiple outcomes: strategies effective for warmth often differed from those effective for democratic-rule attitudes. Voelkel et al. (2024)
The strongest animosity-reduction strategies included highlighting sympathetic political opponents and emphasizing common identities. Some treatments aimed at democratic attitudes corrected perceptions of rivals' support for antidemocratic acts or highlighted the risks of democratic collapse. These results support a plural intervention science in which hostility, social trust, democratic procedure, and violence endorsement are measured separately. Researchers should not assume a common mechanism simply because the outcomes are politically related; the megastudy's covariance analyses suggested that support for antidemocratic practices and for partisan violence possess distinguishable empirical structures. Voelkel et al. (2024)
The megastudy also provides a model for relative-treatment comparisons: a shared outcome battery, randomized arms, clearly defined controls, and preregistered analyses make contrasts more interpretable than comparing effect sizes from unrelated studies. Yet the estimand remains the response to short exposures in an online sample under the study's conditions. Real-world campaigns, organizations, and long-term institutions may generate different intensities and feedback loops. Persistence, repeated dosing, behavior outside a questionnaire, and cross-national transport remain independent questions even when statistical precision is high. Voelkel et al. (2024) Reiljan (2020)
Earlier experiments reporting reduced animosity without broad democratic effects should be read alongside these later findings. Broockman, Kalla, and Westwood showed that experimentally changing affect need not alter many political outcomes; Voelkel and colleagues demonstrated that treatments specifically aimed at institutional beliefs can sometimes alter those outcomes. The combined evidence favors outcome-specific interventions and cautions against presenting a drop in a feeling-thermometer gap as a proven remedy for the political decisions that actually endanger democratic procedures. Broockman et al. (2023) Voelkel et al. (2023) Voelkel et al. (2024)
An Integrated Research Agenda: Identification, Equity, and Intervention
The next stage of research requires coordinated evidence across four observational layers: evaluative attitudes, explicit social expectations, consequential interpersonal decisions, and institutional political choices. At each layer, investigators should define the relevant partisan object, the population and political system, the method of assignment or sampling, and the expected causal mechanism. An index that averages warmth and trust can describe a broad pattern but should preserve item-level information if one component changes and another does not. A preregistered battery could report each outcome separately, alongside a theoretically justified composite and sensitivity tests. Druckman & Levendusky (2019) Voelkel et al. (2024)
To identify partisan discrimination, future field and laboratory experiments should randomize otherwise comparable party cues across settings with different norms and stakes. For instance, researchers could compare judgments about potential collaborators in a nonpolitical project, ethically designed hypothetical hiring evaluations, and incentivized allocations. Factorial manipulations of competence and issue similarity would test whether party labels affect decisions net of information that is legitimately relevant to the task. Transparent exclusions, power calculations, variation across participant identities, and accessible replication materials would permit cumulative inference while protecting participant privacy and legitimate institutional processes. Iyengar & Westwood (2015) Jost et al. (2022)
Longitudinal and network studies should track whether affective changes endure, spread through relationships, or vary by political event. Within-person designs can reveal temporal sequencing, though causal claims remain vulnerable to unmeasured time-dependent influences. Repeated randomized invitations to contact or information interventions offer stronger tests but can themselves generate differential attrition. Measurement should distinguish changes in overall warmth from changes in fairness toward an unfamiliar individual; effects on democratic-rule judgments and willingness to cooperate should be examined independently. Santoro & Broockman (2022) Tausch et al. (2024) Broockman et al. (2023)
Cross-national and human–artificial research can extend the scope under carefully specified operational definitions. Multiparty experiments should randomize labels that carry locally realistic coalition meanings and measure unequal treatment toward supporters of more than one competing organization. In digital interaction, researchers could present equivalent human or artificial messages with controlled partisan cues and test human trust, interpretation, and selection behavior. The hypothesis would concern the psychology of users and interactions, with any claim about an artificial system's own psychological characteristics evaluated through separate standards. Such work locates the dossier within the wider Artificial Era without collapsing political psychology into technology studies. Reiljan (2020) Guess et al. (2023)
Conclusion
Affective polarization is best understood as a family of intergroup evaluations organized around politically meaningful categories. Its most established explanatory resources include party identification, identity alignment, stereotypes, perceived social distance, group-status emotions, and communication contexts. Evidence from randomized party-cue studies supports the proposition that partisan categories can alter some nonpolitical judgments and choices. Evidence from surveys and cross-national comparisons documents variations in expressed hostility, while instrument research warns that evaluation of party leaders and ordinary voters can diverge substantially. A valid conclusion therefore states which target, design, and outcome a finding supports. Iyengar & Westwood (2015) Druckman & Levendusky (2019) Reiljan (2020)
The reviewed experiments also establish the importance of causal separation. Making a person warmer toward the opposing party is an observed psychological outcome; its effect on fair allocation, cooperation, candidate accountability, democratic norms, or political violence requires further measurement. Studies that shifted affect without detecting broad political change and studies that identified distinct successful intervention types together challenge one-dimensional explanations. They show that partisan discrimination and democratic conduct are related research problems, not interchangeable indicators of a single latent disposition. Broockman et al. (2023) Voelkel et al. (2023) Voelkel et al. (2024)
Methodologically, the field advances when it combines construct precision with behavioral observation. Thermometers, social-distance questions, trait ratings, implicit associations, randomized cue experiments, field interventions, and cross-national indices each reveal different facets of political relations. Samples, historical events, differential response propensities, local party systems, and intervention durability constrain generalization. Strong research makes uncertainty traceable through design, sampling, timing, and statistical precision instead of flattening heterogeneous evidence into an undifferentiated statement about the public. Iyengar et al. (2019) Jost et al. (2022)
The enduring research problem is to explain when political categorization becomes interpersonal exclusion, when that exclusion influences consequential decisions, and which conditions restore cooperative and equitable treatment among people who continue to disagree. The scientific goal is the identification of mechanisms that are observable, replicable, and transportable across relationships and institutions. The evidence synthesized here supports a plural account of partisan affect and discrimination: stable political commitments can coexist with fair conduct, and effective reduction of one form of hostility need not solve every other political conflict. This is the appropriate foundation for future experiments, comparative work, and research on human–artificial political encounters. Iyengar & Westwood (2015) Santoro & Broockman (2022) Voelkel et al. (2024)
References
- Ahler, D. J., & Sood, G. (2018). The parties in our heads: Misperceptions about party composition and their consequences. The Journal of Politics, 80(3), 964–981. https://doi.org/10.1086/697253
- Bail, C. A., Argyle, L. P., Brown, T. W., Bumpus, J. P., Chen, H., Hunzaker, M. B. F., Lee, J., Mann, M., Merhout, F., & Volfovsky, A. (2018). Exposure to opposing views on social media can increase political polarization. Proceedings of the National Academy of Sciences, 115(37), 9216–9221. https://doi.org/10.1073/pnas.1804840115
- Broockman, D. E., Kalla, J. L., & Westwood, S. J. (2023). Does affective polarization undermine democratic norms or accountability? Maybe not. American Journal of Political Science, 67(3), 808–828. https://doi.org/10.1111/ajps.12719
- Druckman, J. N., & Levendusky, M. S. (2019). What do we measure when we measure affective polarization? Public Opinion Quarterly, 83(1), 114–122. https://doi.org/10.1093/poq/nfz003
- Druckman, J. N., Klar, S., Krupnikov, Y., Levendusky, M., & Ryan, J. B. (2026). Measuring affective polarization: Matching concepts to measures. Advances in Political Psychology, 47(Suppl. 1), e70152. https://doi.org/10.1111/pops.70152
- Erratum for the research article “How do social media feed algorithms affect attitudes and behavior in an election campaign?” (2026). Science, 391(6791), eaeh2575. https://doi.org/10.1126/science.aeh2575
- Erratum for the research article “How do social media feed algorithms affect attitudes and behavior in an election campaign?” by A. M. Guess et al. (2024). Science, 386(6726), eadu8261. https://doi.org/10.1126/science.adu8261
- Finkel, E. J., Bail, C. A., Cikara, M., Ditto, P. H., Iyengar, S., Klar, S., Mason, L., McGrath, M. C., Nyhan, B., Rand, D. G., Skitka, L. J., Tucker, J. A., Van Bavel, J. J., Wang, C. S., & Druckman, J. N. (2020). Political sectarianism in America. Science, 370(6516), 533–536. https://doi.org/10.1126/science.abe1715
- Guess, A. M., Malhotra, N., Pan, J., Barberá, P., Allcott, H., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M., et al. (2023). How do social media feed algorithms affect attitudes and behavior in an election campaign? Science, 381(6656), 398–404. https://doi.org/10.1126/science.abp9364
- Huddy, L., Mason, L., & Aarøe, L. (2015). Expressive partisanship: Campaign involvement, political emotion, and partisan identity. American Political Science Review, 109(1), 1–17. https://doi.org/10.1017/S0003055414000604
- Iyengar, S., Sood, G., & Lelkes, Y. (2012). Affect, not ideology: A social identity perspective on polarization. Public Opinion Quarterly, 76(3), 405–431. https://doi.org/10.1093/poq/nfs038
- Iyengar, S., & Westwood, S. J. (2015). Fear and loathing across party lines: New evidence on group polarization. American Journal of Political Science, 59(3), 690–707. https://doi.org/10.1111/ajps.12152
- Iyengar, S., Lelkes, Y., Levendusky, M., Malhotra, N., & Westwood, S. J. (2019). The origins and consequences of affective polarization in the United States. Annual Review of Political Science, 22, 129–146. https://doi.org/10.1146/annurev-polisci-051117-073034
- Jost, J. T., Baldassarri, D. S., & Druckman, J. N. (2022). Cognitive–motivational mechanisms of political polarization in social-communicative contexts. Nature Reviews Psychology, 1, 560–576. https://doi.org/10.1038/s44159-022-00093-5
- Levendusky, M. S. (2018). Americans, not partisans: Can priming American national identity reduce affective polarization? The Journal of Politics, 80(1), 59–70. https://doi.org/10.1086/693987
- Mason, L. (2015). “I disrespectfully agree”: The differential effects of partisan sorting on social and issue polarization. American Journal of Political Science, 59(1), 128–145. https://doi.org/10.1111/ajps.12089
- Reiljan, A. (2020). ‘Fear and loathing across party lines’ (also) in Europe: Affective polarisation in European party systems. European Journal of Political Research, 59(2), 376–396. https://doi.org/10.1111/1475-6765.12351
- Santoro, E., & Broockman, D. E. (2022). The promise and pitfalls of cross-partisan conversations for reducing affective polarization: Evidence from randomized experiments. Science Advances, 8(25), eabn5515. https://doi.org/10.1126/sciadv.abn5515
- Tausch, N., Birtel, M. D., Górska, P., Bode, S., & Rocha, C. (2024). A post-Brexit intergroup contact intervention reduces affective polarization between Leavers and Remainers short-term. Communications Psychology, 2, Article 95. https://doi.org/10.1038/s44271-024-00146-w
- Voelkel, J. G., Chu, J., Stagnaro, M. N., Mernyk, J. S., Redekopp, C., Pink, S. L., Druckman, J. N., Rand, D. G., & Willer, R. (2023). Interventions reducing affective polarization do not necessarily improve anti-democratic attitudes. Nature Human Behaviour, 7, 55–64. https://doi.org/10.1038/s41562-022-01466-9
- Voelkel, J. G., Stagnaro, M. N., Chu, J. Y., Pink, S. L., Mernyk, J. S., Redekopp, C., Ghezae, I., Cashman, M., Adjodah, D., et al. (2024). Megastudy testing 25 treatments to reduce antidemocratic attitudes and partisan animosity. Science, 386(6719), eadh4764. https://doi.org/10.1126/science.adh4764
- Wojcieszak, M., & Warner, B. R. (2020). Can interparty contact reduce affective polarization? A systematic test of different forms of intergroup contact. Political Communication, 37(6), 789–811. https://doi.org/10.1080/10584609.2020.1760406
Social Identity and Expressive Partisanship
A political party can become a socially meaningful self-category through repeated electoral participation, social networks, media language, family transmission, and perceived collective success or defeat. Once party membership is salient, people may use it to distinguish 'us' from 'them,' regulate expectations of loyalty, and interpret ambiguous political information. Social identity reasoning predicts that the meaning of a category depends on comparison, the perceived status of relevant groups, and local norms, rather than on an invariant disposition toward every opponent. Political categories therefore acquire psychological force through history and interaction, while their intensity differs between individuals and situations. Iyengar et al. (2012) Jost et al. (2022)
Expressive partisanship treats group membership as a motivational resource. Across survey and experimental studies, Huddy, Mason, and Aarøe found that a multi-item partisan identity scale better accounted for campaign activity than some measures of ideological intensity and policy commitment. Threats to party status generated anger and reassurance generated positive emotion among strongly attached partisans. These findings make a narrower point than claiming ideology is irrelevant: experimentally presented political-group threats can elicit responses not reducible to respondents' reported issue positions. The strength of identification is consequently an important moderator to measure directly. Huddy et al. (2015)
Social sorting can strengthen partisan attachments because several memberships begin pointing in the same evaluative direction. Mason's analysis helps explain why changes in social distance can outpace movement in particular policy positions. Yet social categories also contain cross-cutting memberships, and their boundaries can change with context. A person who views another as an opposing partisan in a campaign can also view the same individual as a coworker, parent, neighbor, or fellow citizen. Identity-based theories consequently yield manipulable hypotheses about the salience of common membership, the credibility of counterstereotypical information, and the durability of changed judgments. Mason (2015) Levendusky (2018)
Identity processes coexist with accuracy-seeking, value-driven, and institutionally structured motives. Jost, Baldassarri, and Druckman's integrative review distinguishes motives related to protecting the self, defending a group, and endorsing a social system, alongside communication sources and channels. These processes supply different explanations for similar survey responses. An experiment designed to isolate party loyalty should therefore control the information available about competence, ideology, reciprocity, and fairness, and record when party identity was made salient. Without such controls, a party-cue effect may combine social preference with inference about policy, values, or anticipated cooperation. Jost et al. (2022) Iyengar & Westwood (2015)