Identity–Policy Fusion and Polarization in United States House Campaign Language

Abstract

Campaign language in the United States has grown markedly polarized, with candidates increasingly speaking in party-distinctive ways. Yet the mechanisms driving this linguistic divergence remain insufficiently understood. This study proposes identity–policy fusion, a framing strategy in which candidates embed distinctive biographical vocabulary into policy statements, as one factor shaping this divergence. By framing policy as an authentic extension of lived experience, fusion may tie stances to in-group identity and biographical authority, raising the psychological and social costs of disagreement. Using computational text analysis of 41,842 policy statements from 3,343 U.S. House candidates in the 2018, 2020, and 2022 election cycles, the study operationalizes fusion as term frequency–inverse document frequency (TF–IDF)-weighted lexical overlap between biographies and policy texts, and polarization as a statement’s relative similarity to in-party versus out-party linguistic norms within policy domains. Ordinary least squares regression shows that fusion is significantly associated with higher polarization in campaign language, with the association approximately 26 percent stronger for Republicans than Democrats. This partisan asymmetry is consistent with fusion serving as an alternative source of legitimation under conditions of contested institutional authority, illuminating a potential mechanism through which elite messaging may harden partisan boundaries.

Keywords

political communication affective polarization computational text analysis candidate biography policy statements partisan language

Introduction

Partisan conflict in the United States has intensified across multiple dimensions over the past four decades, including a sharp rise in mass issue polarization from 1988 to 2024, with especially pronounced divergence between political opinion clusters from 2008 to 2020 (Young et al. 2026). Affective polarization, the intensification of partisan dislike and moralized out-group hostility, has emerged as a particularly consequential dimension of this trend, threatening core democratic functions by fueling gridlock, degrading institutional trust, and undermining collective problem-solving (Iyengar and Westwood 2015; Mason 2018).

Studies have further characterized this phenomenon as “political sectarianism,” wherein partisan identity establishes rigid moral boundaries that cast opponents not merely as wrong, but as alien and immoral (Finkel et al. 2020). Importantly, however, the behavioral consequences of this hostility are conditional rather than constant: partisan animosity most powerfully shapes action when politicians send clear signals (Druckman et al. 2024), placing elite communication at the center of whether partisan divisions degrade democratic functioning. This conditional logic underscores why elite language that activates partisan identities and sharpens partisan distinctions, even when ostensibly aimed at persuasion, warrants particular scrutiny. Indeed, empirical work shows that congressional floor speech has become markedly more party-distinctive over time, such that an observer can now infer a speaker’s party from a short speech with unprecedented accuracy (Gentzkow et al. 2017).

This growing party distinctiveness in elite speech—polarization in political language—raises questions about the message-framing strategies and mechanisms driving it, particularly in high-stakes contexts such as election campaigns. To address these questions, this paper investigates polarization in campaign language, operationalized as linguistic distinctiveness: the degree to which a candidate’s language aligns with in-party norms while diverging from those of the opposition. The paper proposes identity–policy fusion as one factor shaping this polarization, a framing strategy in which candidates anchor policy arguments to their own identity and lived experience. To contextualize why candidates adopt this strategy, the next section surveys how political actors adapt their communication under conditions of eroding institutional trust and contested authority, using this synthesis to motivate and elaborate the proposed framing strategy.

This study develops and tests this proposition through a computational text analysis of campaign communications from 3,343 U.S. House candidates across the 2018, 2020, and 2022 election cycles, drawing on 41,842 policy statements (the unit of analysis) from the CampaignView database (Porter et al. 2025). The analysis then examines whether identity–policy fusion is associated with more polarized campaign language, and whether this association differs by party.

Framing Strategies Under Declining Institutional Trust

Scholarship on political communication and related fields suggests three specific ways political actors, including candidates, navigate environments where traditional credibility signals have weakened.

First, a substantial body of work documents the rise of populist communication styles that explicitly reject institutional authority in favor of “authentic” ordinary experience. Moffitt (2016) identifies the stylistic performance of authenticity and anti-elitism as central to populist political style, while Mudde and Rovira Kaltwasser (2017) emphasize how populist messaging constructs a virtuous “people” in opposition to corrupt elites, positioning popular will as superior to institutional expertise. Importantly, these appeals do not simply critique institutions; they elevate ordinary citizens’ lived experience and popular judgment as alternative grounds for epistemic authority. Experimental work on populist communication supports this broader anti-institutional logic: Hameleers et al. (2017) show that exposure to populist messages using emotionalized blame attribution can strengthen populist attitudes and increase blame directed at political elites, particularly among citizens with weaker collective identity attachments. Here, claims to authenticity rest on presenting oneself as an ordinary person rather than an expert. Yet in these accounts, authenticity functions primarily as a source of speaker credibility, not as a mechanism through which personal experience is integrated into substantive policy claims.

Second, research on political personalization documents a shift in power away from collective entities, particularly parties, toward individual political actors (Rahat and Sheafer 2007). Personalization manifests through two simultaneous processes: centralizing personalization, which concentrates authority in top leaders, and decentralizing personalization, which grants individual politicians greater autonomy from party structures (Balmas et al. 2014). This shift toward the individual politician as the primary unit of political appeal has encouraged greater emphasis on personalized communication. Experimental studies examining the electoral consequences of this trend show that when online communication focuses on an individual politician rather than a party, it can increase citizens’ feelings of political involvement and closeness to politics (Kruikemeier et al. 2013). These effects, however, vary by candidate characteristics and partisan alignment (McGregor 2018). Within this literature, personal biography is treated primarily as a standalone appeal to character, rather than as a component of policy arguments.

Third, a parallel body of work in elite rhetoric and political psychology has identified the increasing moralization and emotionalization of policy. Clifford and Jerit (2013) demonstrate that elites’ use of moral language in policy debates is a strategic choice that can shape mass attitudes. This focus on moralization is critical because, as Ryan (2017) finds, when citizens themselves hold moral convictions on an issue, they become fundamentally opposed to political compromise and are more likely to penalize politicians who negotiate, creating electoral incentives for elites to adopt more rigid, moralized stances. In a complementary line of work, Feinberg and Willer (2015) identify a key reason moral arguments often fail: advocates spontaneously argue from their own moral framework, failing to bridge a “moral empathy gap.” However, their experiments demonstrate that when arguments are consciously reframed to appeal to an audience’s distinct moral foundations, they can successfully increase persuasion across partisan lines. Crucially, these accounts treat moral framing as issue-centered: the moral weight attaches to the policy position itself, not to the candidate’s biographical identity or lived experience.

These three literatures point to a common strategic adaptation: in the face of eroding traditional authority, political actors increasingly rely on authenticity claims, candidate-centered self-presentation, and moral framing of policy. Although conceptualized separately in the literature, these strategies are arguably complementary in practice, plausibly converging to form a more comprehensive political appeal. However, existing studies have largely analyzed these appeals in isolation, leaving undertheorized the ways they may be combined. Such convergence is also enabled by the broader structural context in which these adaptations have unfolded: the rise of the hybrid media system and the decline of traditional journalistic gatekeeping (Chadwick 2017), a dynamic that political elites have strategically leveraged. As Heseltine (2025) demonstrates, Members of Congress (particularly Republicans since 2016) increasingly bypass mainstream outlets to share and amplify ideologically extreme alternative media, a strategy associated with higher public engagement. Insofar as candidates operate within and are rewarded by these largely unmediated, engagement-driven environments, they may face stronger incentives to combine these distinct appeals into a unified frame.

Beyond the structural context, the logic of consolidation is also functional: each strategy, deployed in isolation, carries vulnerabilities that the others offset. Authenticity claims grounded solely in lived experience risk inviting challenges to technocratic competence, personalization detached from issues risks appearing self-serving, and moral framing without biographical grounding can read as strategic posturing rather than genuine conviction. When consolidated, these vulnerabilities are mutually neutralized: biographical grounding lends moral claims the credibility of lived conviction, moral substance anchors personalization in policy rather than self-promotion, and policy engagement insulates authenticity appeals from challenges to competence. Together, these linkages make the policy itself more resonant.

Providing conceptual grounding for such integration, Wellings et al. (2024) introduce the concept of “narrative fusion” to describe how politicians deliberately interweave otherwise independent narratives into a unified frame to serve multiple objectives. Analyzing government communication during the COVID-19 crisis in the United Kingdom, they demonstrate that elite political narratives are not single-issue constructions; rather, elites actively localize and fuse them with broader political projects to resonate with an audience’s pre-existing cultural identities. This concept describes a broader logic of strategic communication, particularly in contexts where politicians must simultaneously establish personal credibility and build support for their policies.

To conceptualize how authenticity claims, candidate-centered self-presentation, and moral framing of policy may converge into a unified political appeal, the present paper extends the concept of narrative fusion to candidate-level communication and introduces identity–policy fusion: the embedding of a candidate’s distinctive biographical vocabulary (e.g., references to occupation, family background, religion, military service) into policy statements. This strategy presents policy as an authentic and inherently moral outgrowth of the candidate’s life experience, thereby potentially increasing the psychological and social costs of disagreement. Existing research has not yet formally operationalized or empirically tested identity–policy fusion as a measurable feature of political communication; the present study addresses this gap.

Identity–Policy Fusion and the Polarization of Campaign Language

Identity–policy fusion is conceptualized here as a strategic response to the limitations of classic persuasion models under conditions of entrenched affective polarization, declining trust, and increasingly unmediated campaign communication. Fenno’s (1978) seminal work on “home style” demonstrated that representatives build trust through curated presentations of self. Complementing this insight, the Elaboration Likelihood Model further posits two persuasion pathways: a central route, which engages audiences with substantive arguments, and a peripheral route, which relies on heuristic cues such as speaker credibility (Petty and Cacioppo 1986). These frameworks have been highly influential, but they typically treat personal biography as a peripheral cue and policy substance as a central argument, without theorizing their systematic integration.

These classic models face increasing challenges in contemporary political environments marked by heightened partisan-motivated reasoning and declining trust. The central route is often compromised when partisans are predisposed to reject opposing arguments (Kunda 1990), facilitating selective processing that blunts the force of factual appeals (Ecker et al. 2022). At the same time, the peripheral route may weaken as public trust in government erodes (Hetherington and Rudolph 2015; Pew Research Center 2025). In response to the reduced reliability of both routes, and enabled by the affordances of the hybrid media environment, this paper theorizes identity–policy fusion as a strategic adaptation. Rather than operating as a third “route” of persuasion, fusion embeds policy within identity claims in ways that may raise the costs of disagreement without fitting neatly into the central–peripheral distinction. This strategy is theorized to operate through distinct yet complementary psychological and social pathways.

The Psychological Pathway: Identity-Based Insulation

The psychological mechanism of identity–policy fusion is theorized to stem from its potential to activate identity-based defenses among supporters. Drawing on social identity theory (Tajfel & Turner, 1979), fusion can position policy agreement as a test of in-group loyalty. When a candidate presents a policy as a natural extension of their biographical identity, particularly when that identity is salient to supporters, acceptance can become a means of affirming group membership. Cognitive dissonance (Festinger 1957) suggests why rejecting a fused policy might generate psychological discomfort for in-group members. Supporters who disagree may need to reconcile their group identity with personal dissent, potentially creating heightened tension. Identity-protective cognition (Kahan 2017) may further reinforce this resistance by motivating individuals to dismiss counterarguments to safeguard both identity and political commitments.

Together, these mechanisms can create dual insulation: rejecting the policy threatens both the politician’s biographical credibility and the shared identity of supporters. For audiences who value the invoked identity, opposition may generate greater psychological discomfort than disagreement with a standalone policy position, making compromise less tenable. However, fusion’s polarizing potential is not limited to its effects on co-partisans; it may also operate through social pressures that shape how opponents can respond.

The Social Pathway: Constraining Counterarguments

Socially, identity–policy fusion is theorized to constrain opponents’ discursive options by leveraging norms that valorize personal testimony and lived experience. Critiquing a fused position may place opponents in a reputational bind: they face a choice to either explicitly dismiss the biographical experience as irrelevant, which risks violating these salient social norms and appearing morally callous, or to acknowledge the experience as valid while still rejecting the policy, a move that can appear internally inconsistent or be framed as a denial of the narrative’s claimed implications. This dynamic suggests a structural constraint on political communication in an era of affective polarization, where moral and identity-based attacks are common (Finkel et al. 2020). Thus, fusion potentially elevates the reputational risks of policy engagement, making substantive counterargument socially more costly.

Strategic Advantages Over Alternative Approaches

Building on these psychological and social mechanisms, and returning to the functional logic outlined earlier, identity–policy fusion offers a way to address vulnerabilities in existing framing strategies for elites seeking to mobilize supporters and deter opposition. A standalone personal biography may enhance credibility but risks dismissal as anecdotal, especially under conditions of low trust. A standalone policy argument is more directly exposed to motivated reasoning and partisan filtering, and nuanced arguments are easily misremembered in polarized settings (Novoa et al. 2023). Identity–policy fusion may mitigate both vulnerabilities: the policy is shielded by biography, and the biography is validated through its policy linkage.

Although the mechanics of fusion are ideologically neutral, its functional utility is theorized to differ across the political spectrum. While moderates can employ fusion to signal credibility, the strategy’s unique value lies in authenticating and insulating more partisan or controversial positions by reframing them as natural extensions of personal conviction. Framing positions as rooted in biography potentially converts divisive stances into expressions of personal authenticity. Moreover, because fusion ties policy to identity markers that resonate primarily with the candidate’s base, it should produce language that signals strong in-party authenticity while appearing alienating or irrelevant to out-party audiences. The result is not merely rigid positioning, but distinctively partisan language: vocabulary and framing that align closely with in-party norms because they are grounded in identities the out-party does not share, thereby increasing the linguistic distinctiveness of campaign language across parties. This logic motivates the first hypothesis:

Hypothesis 1 (H1): Greater identity–policy fusion in a candidate’s policy statement will be associated with more polarized campaign language in that statement (Identity–policy fusion is operationalized as the degree of distinctive lexical overlap between the statement and the candidate’s own biography; polarized campaign language is operationalized as the degree to which a statement’s language aligns with in-party norms while diverging from opposing-party norms within the same policy domain).

Partisan Asymmetry in Fusion’s Polarizing Effects

The differential function of fusion also points to potential asymmetry across the political spectrum. Its power is theorized to be magnified within a political ecosystem that cultivates skepticism toward traditional arbiters of evidence such as science, academia, and the news media. For political actors who define themselves in opposition to traditional elites and institutions, a pattern documented extensively in populist radical right movements (Mols and Jetten 2017), identity–policy fusion can provide an alternative basis of authority, substituting the external, verifiable authority of data for the internal, less easily falsifiable authority of personal experience. This dynamic has clear partisan implications in the U.S. context. Extensive research documents a pronounced asymmetry in trust toward scientific experts and institutions, with Republicans expressing significantly more skepticism than Democrats (Funk et al. 2019; Gauchat 2012). This divide has widened markedly in recent years (Kennedy et al. 2022).

When candidates use identity–policy fusion to authenticate claims that are at odds with institutional consensus, its legitimizing power should be most pronounced among audiences most distrustful of these institutions. This suggests that the strategy may more effectively enable polarized language for Republican candidates than for Democratic candidates, at least within the contemporary U.S. party system. Accordingly, the theory predicts a partisan asymmetry in the strength of the association between identity–policy fusion and polarized campaign language, leading to a second, conditional hypothesis:

Hypothesis 2 (H2): The positive association between identity–policy fusion and polarized campaign language will be stronger for Republican candidates than for Democratic candidates.

Materials and Methods

Research Design and Sample

The research design requires linking individual policy statements, the unit of analysis, to candidate biographies and situating them within party-level linguistic norms. Given the scale of the corpus, this study adopts a computational text analysis approach, permitting precise, replicable measurement of the study’s core constructs across tens of thousands of documents.

Data are drawn from the CampaignView database (Porter et al. 2025), a corpus of personal biographies and policy statements from 5,228 U.S. House of Representatives candidates across the 2018, 2020, and 2022 election cycles, representing 86.9% of all major-party, ballot-eligible contenders. The database’s compilers collected textual data directly from active candidate websites approximately one week prior to each state’s primary election, capturing messaging at a point of high strategic refinement and reflecting the language candidates actually presented to voters. This provides a more direct measure of strategic framing than retrospectively archived web data, which may have been modified or incompletely preserved after the campaign. The 2018–2022 period is also theoretically appropriate: as established in the preceding sections, both hypotheses treat heightened affective polarization and declining institutional trust as antecedent conditions for identity–policy fusion, and this window captures precisely that environment, including the documented elite turn toward unmediated communication.

The analytical sample draws on this database but is restricted to candidates for whom both a personal biography and at least one policy statement were available. Of the 3,545 candidates who published at least one policy position (43,465 statements total), 202 candidates (1,623 statements) were excluded because they did not publish a personal biography, yielding an analytical dataset of 41,842 policy statements from 3,343 unique candidates. Policy statements spanned fourteen domains (thirteen substantive issue domains and one residual “Unknown” domain), with the most prevalent topics being Economics and Commerce (n = 5,685), Crime (n = 4,163), and Civil Rights (n = 4,031; see Supplemental Material, Note S1, for complete descriptive statistics by issue domain).

A standardized pipeline implemented in R processed all textual data from this corpus. Specifically, all textual data underwent conversion to lowercase, tokenization, and the removal of standard English stop words. The pipeline deliberately avoided stemming and lemmatization to preserve the precise lexical choices made by candidates, as the theoretical mechanism relies on lexical matching between biographical self-descriptions and policy statements. The analysis focuses on unigrams (single-word tokens) and does not incorporate higher-order n-grams, because the theoretical interest is in the reuse of unique biographical vocabulary rather than in syntactic or phrasal structure.

Measuring Identity–Policy Fusion

The independent variable, identity–policy fusion, is operationalized as the degree of distinctive lexical overlap between a candidate’s personal biography and their policy statements. This approach provides a scalable, replicable measure of the linguistic tether between personal biography and policy, which is the core of the proposed mechanism. The measure captures the extent to which a candidate embeds personally distinctive language from their biography into policy statements, rather than assessing narrative structure or rhetorical quality. Biography is treated as an “identity text”: a concentrated source of self-descriptive language whose reuse in policy statements is interpreted as evidence of identity–policy fusion rather than generic political phrasing. In this sense, the measure operationalizes fusion as repeated, distinctive self-referential language in policy contexts, which is a necessary but not sufficient condition for fully articulated narrative integration. A high score indicates that the candidate’s personal lexicon is a salient feature of the policy statement.

The score for each policy statement was calculated using a three-step process relying on term frequency–inverse document frequency (TF–IDF) and cosine similarity. TF–IDF is a standard text-analysis technique designed to identify words that are distinctively important to a specific document relative to a larger corpus.

Global IDF calculation: A global inverse document frequency (IDF) table was constructed from a comprehensive corpus comprising every personal biography and policy statement from all candidates in the dataset (Democratic and Republican, 2018–2022). This critical step down-weights high-frequency political terms (e.g., “government,” “America,” “economy”), thereby prioritizing more distinctive lexical choices.

Document vectorization: For each candidate, the personal biography and each policy statement were converted into TF-IDF vectors. Term frequency (TF) was calculated within each document, while IDF values were drawn from the global table constructed in step one.

Similarity scoring: For each policy statement, the identity–policy fusion score was computed as the cosine similarity between the TF-IDF vector of that statement and the TF-IDF vector of the candidate’s own personal biography. Cosine similarity measures the similarity in weighted term distributions of two documents in a high-dimensional space, irrespective of length. The metric ranges from 0 to 1, where 0 indicates no weighted vocabulary overlap and 1 indicates identical term distributions after TF–IDF weighting. A score of zero thus represents the theoretical minimum of identity–policy fusion.

Formally, for candidate $i$ and policy statement $s$ :

F u s i o n_{i s} = \cos (b_{i}, p_{i s}) = \frac{b_{i} \cdot p_{i s}}{∥ b_{i} ∥ ∥ p_{i s} ∥}

(1)

where $b_{i}$ is the TF–IDF vector for candidate $i$ ’s biography and $p_{i s}$ is the TF–IDF vector for policy statement $s$ . A 30-case construct validation with verbatim text excerpts across low-, middle-, and high-fusion segments is provided in Supplemental Material (Note S2).

Measuring Polarization in Campaign Language

This study’s dependent variable, polarization in campaign language, measures the degree to which a policy statement’s language is more similar to the linguistic norms of the candidate’s own party than to those of the opposing party, within a specific policy domain.

This measure is calculated in two steps using an issue-specific, party-centroid comparison method:

Centroid Creation: First, for each policy domain (e.g., Economy, Healthcare), two “party centroids” were created. A centroid is the average TF-IDF vector of all policy statements within that domain, one for all Democratic candidates and one for all Republican candidates. These policy domains are derived from the dataset’s pre-existing policy codes (see Supplemental Material, Note S1, for the domains).

Polarization Calculation: Second, the polarization score for any given policy statement is calculated as the cosine similarity between the statement and its in-party issue centroid, minus the cosine similarity between the statement and its out-party issue centroid. This relative difference score serves as a direct measure of partisan differentiation, capturing both in-group alignment and simultaneous linguistic distancing from the out-group. A high positive value indicates the statement’s language is distinctly partisan, aligning closely with the candidate’s party and diverging from the opposition. A score near zero represents linguistic bipartisanship or centrism, indicating equal similarity to both party centroids. Negative values indicate greater similarity to the opposing party’s issue centroid than to the candidate’s own party centroid.

Formally, for statement $s$ by candidate $i$ on issue $k$ :

P o l_{i s k} = \cos (p_{i s k}, C_{k}^{i n}) - \cos (p_{i s k}, C_{k}^{o u t})

(2)

where $p_{i s k}$ is the TF–IDF vector for statement $s$ , $C_{k}^{i n}$ is the in-party centroid for issue $k$ , and $C_{k}^{o u t}$ is the out-party centroid for that issue.

Analytical Strategy

The hypotheses were tested using ordinary least squares (OLS) regression. The final analytical sample consists of 41,149 policy statements obtained after listwise deletion for observations with missing data on control variables. This sample includes 22,747 statements from Democratic candidates and 18,402 from Republican candidates.

To test Hypothesis 1 (H1), the polarization in campaign language score, the dependent variable, was regressed on the identity–policy fusion score, the independent variable. The models include a standard set of controls to account for potential confounders: incumbency status, election year, candidate party, district-level partisanship, candidate quality, and primary election performance. These covariates capture well-established sources of variation in both campaign rhetoric and partisan polarization, such as systematic differences between incumbents and challengers, temporal shifts across election cycles, and the partisan and competitive structure of districts.

Furthermore, to distinguish candidate-specific fusion from broader partisan identity signaling, the models include a key control for generic partisan language. This variable captures the extent to which a policy statement uses language that resembles the typical biographical language of the candidate’s party.

For each party $g \in {D e m o c r a t, R e p u b l i c a n}$ , a party biography centroid $B_{g}$ was constructed as the average TF–IDF vector of all personal biographies from candidates in that party (non-issue-specific). The generic partisan language score for each policy statement is then defined as the cosine similarity between the statement and the centroid of biographies from the candidate’s party. Higher values indicate that the policy statement’s language more closely matches the generic biographical language associated with the candidate’s party, independent of the candidate’s own biography. Because party-archetype language may partially overlap with individual candidates’ biographies, this control helps make the test of identity–policy fusion more conservative by attributing any shared, non-idiosyncratic identity tokens to generic partisan identity rather than to candidate-specific fusion.

Formally, for candidate $i$ in party $g (i)$ and policy statement $s$ :

G e n e r i c P a r t i s a n_{i s} = \cos (p_{i s}, B_{g (i)}) = \frac{p_{i s} \cdot B_{g (i)}}{∥ p_{i s} ∥ ∥ B_{g (i)} ∥}

(3)

where $p_{i s}$ is the TF–IDF vector for policy statement $s$ , and $B_{g (i)}$ is the TF–IDF centroid of all biographies from candidates in $i$ ’s party.

Policy code fixed effects were included in all OLS models to account for baseline differences in polarization across topics. To account for the non-independence of multiple statements from the same individual, standard errors were clustered at the candidate level (see Supplemental Material, Note S3, for detailed variable descriptions, control variable measurements, and sources). Given the large number of clusters (3,343 candidates), inference relies on the asymptotic normal distribution; test statistics reported in the text are therefore z values, computed from unrounded coefficients and clustered standard errors. Additional robustness checks that (a) remove potential measurement artifacts, (b) add controls for text length and lexical diversity, and (c) substitute alternative fusion operationalizations were conducted; details are reported in the Supplemental Material, Note S2 (Construct Validation of the Identity–Policy Fusion Measure).

To test Hypothesis 2 (H2), the primary model was extended to include an interaction term between the identity–policy fusion score and a binary indicator for the candidate’s party.

Results

Descriptive Statistics

Table 1 presents descriptive statistics for all variables in the main analytical sample (N = 41,842 policy statements from 3,343 candidates). The dependent variable, polarization in campaign language, had a mean of 0.01 (SD = 0.017), with values ranging from −0.086 to 0.434, indicating that the majority of statements occupy a relatively centrist linguistic position, with a positively skewed tail of highly partisan language. The independent variable, identity–policy fusion, was similarly right-skewed (M = 0.091, SD = 0.125, range = 0–1.00), reflecting that strong biographical–policy lexical overlap is present but concentrated among a subset of statements. Figure 1 plots mean identity-policy fusion against mean polarization in campaign language across the thirteen substantive policy domains (excluding “Unknown” domain).

Table 1.

Descriptive Statistics.

Variable	Mean	SD	Min	Max
Polarization in Campaign Language	0.010	0.017	−0.086	0.434
Identity–policy Fusion	0.091	0.125	0.000	1.000
Generic Partisan Language	0.088	0.039	0.000	0.350
District-Level Partisanship	50.263	15.397	16.900	93.800
Primary Election Vote Share	42.646	34.430	0.005	100.000
Incumbency Status	0.247	0.431	0.000	1.000
Primary Election Win	0.480	0.500	0.000	1.000
Candidate Quality	0.622	0.854	0.000	2.000
Candidate Party	—	—	—	—
Election Year	—	—	—	—
Policy Code (Fixed Effect)	—	—	—	—

Note. N = 41,842 policy statements from 3,343 candidates across the 2018, 2020, and 2022 U.S. House election cycles, except District-Level Partisanship (n = 41,149). For binary variables, the mean equals the proportion. Candidate Quality is coded 0 = no prior office-holding experience, 1 = local or state office, 2 = federal or statewide office. District-Level Partisanship is the Democratic presidential vote share in the candidate’s district. The sample included 23,202 Democratic and 18,640 Republican statements across the full fourteen policy domains.

Figure 1.

Mean identity–policy fusion and polarization in campaign language across policy domains.

Three examples from the construct validation analysis illustrate what the identity–policy fusion score captures. At the low end (fusion = 0.000), a candidate whose biography centered on law enforcement and military service scored zero on an energy policy statement, sharing no substantive vocabulary with it. At the median (fusion ≈ 0.042), a family physician’s biography overlapped modestly with a firearm policy statement reframing gun violence as a public health issue, sharing terms such as “health,” “care,” and “system.” At the high end (fusion = 0.796), a candidate retold the same schools, family circumstances, and “return home” narrative inside an education policy statement to argue that similar opportunities should be available to all children in the district. Together, these cases suggest the measure discriminates meaningfully across the construct’s full range (see Supplemental Material, Note S2, for the full 30-case validation).

Hypothesis Tests

To evaluate the hypotheses, a series of regression models with policy-code fixed effects and standard errors clustered by candidate were estimated. H1 predicted that greater identity–policy fusion would be associated with more polarized campaign language, whereas H2 predicted that this association would be stronger for Republican candidates than for Democratic candidates. Table 2 summarizes the key results. Diagnostic tests supporting model assumptions and full regression results are provided in the Supplemental Material (Notes S4 and S5).

Table 2.

The Effect of Identity–Policy Fusion on Polarization in Campaign Language.

Variable	Model 1 (Baseline)	Model 2 (+Key Control)	Model 3 (Interaction + Key Control)	Model 4 (Interaction Only)
Identity–policy fusion	.020*** (.001)	.026*** (.001)	.023*** (.002)	.017*** (.002)
Identity–policy fusion × Republican	—	—	.006** (.002)	.007* (.003)
Generic partisan language	—	−.163*** (.004)	−.163*** (.004)	—
Fixed effects & statistics
Policy-code FE	Yes	Yes	Yes	Yes
Observations	41,149	41,149	41,149	41,149
R² (within)	.245	.360	.360	.246

Note. n = 41,149 (Final analytical sample, after listwise deletion for missing district-level partisanship values). OLS estimates with standard errors clustered by candidate in parentheses. The dependent variable is polarization in campaign language. All models include standard controls for incumbency, election year, candidate party, district-level partisanship, candidate quality, primary performance, and policy-code fixed effects. Standard errors were clustered at the candidate level. OLS = ordinary least squares.

p < .05. **p < .01. ***p < .001.

Model 1: Baseline Association

As predicted by H1, the baseline model revealed a significant and positive association between identity–policy fusion and polarization in campaign language (b = 0.020, SE = 0.001, z = 13.54, p < .001; see Table 2, Model 1). To interpret the effect size, the standardized coefficient was calculated. A one standard deviation increase in the identity–policy fusion score was associated with a 0.14 standard deviation increase in polarization in campaign language (β = 0.14). This indicates that as the linguistic overlap between a candidate’s personal biography and a policy statement increases, the language of that statement becomes significantly more aligned with the candidate’s own party and less aligned with the opposing party. These results support H1.

Model 2: Main Effect with Generic Partisan Language Control

When the generic partisan language control was introduced, the coefficient for identity–policy fusion increased and remained highly significant (b = 0.026, SE = 0.001, z = 21.05, p < .001; Model 2). Notably, the generic partisan language variable showed a strong negative association with polarization in campaign language (b = −0.163, SE = 0.004, z = −40.75, p < .001). This pattern indicates that, conditional on identity–policy fusion and the other covariates, generic party-identity language is associated with less polarized campaign language, whereas candidate-specific identity–policy fusion is associated with more polarized campaign language. Because identity–policy fusion remained positive and significant after accounting for generic partisan language, Model 2 provides a more conservative test of H1 and shows that the effect is not reducible to generic partisan language.

Model 3: Interaction Effect With Generic Partisan Language Control

Turning to H2, the results showed a significant interaction between identity–policy fusion and candidate party, indicating that the polarizing effect of identity–policy fusion is conditional on party (b = 0.006, SE = 0.002, z = 3.00, p < .01; Model 3). In this model, the main coefficient for identity–policy fusion represents the simple slope for the baseline group, Democratic candidates. A simple slopes analysis established that while the relationship was significant and positive for both Democrats (b = 0.023, SE = 0.002, p < .001) and Republicans (b = 0.029, SE = 0.002, p < .001), the effect was approximately 26 percent larger for Republican candidates than for Democratic candidates (see Figure 2). These results support H2.

Figure 2.

The conditional effect of identity–policy fusion on polarization in campaign language.

Model 4: Sensitivity to Generic Partisan Language Control

To assess sensitivity to the inclusion of the generic partisan language control, Model 4 re-estimated the interaction model without this control. The interaction effect for Republican candidates remained significant when the generic partisan language control was excluded (b = 0.007, SE = 0.003, z = 2.39, p < .05; Model 4).

Discussion

This study introduces identity–policy fusion as a novel framing strategy and provides large-scale, quantitative evidence that it is strongly associated with polarization in campaign language. By extending Wellings et al.’s (2024) concept of “narrative fusion” to candidate-level communication, this research shows that policy statements that more strongly fuse candidates’ personal biographies with their policy claims tend to exhibit higher measured polarization (H1), defined here as the degree to which language aligns with in-party norms while diverging from those of the opposition. Moreover, this association is significantly stronger for Republican candidates than for Democratic candidates (H2). Because the study draws these texts from campaign websites, which are platforms, candidates deliberately craft to anchor core messaging (Porter et al. 2025), the observed pattern plausibly reflects strategic choices in message design and framing rather than incidental or ephemeral language use.

Theoretically, the findings clarify how contemporary campaigns reorganize the relationship between “who the candidate is” and “what the candidate stands for.” Classic persuasion frameworks such as the Elaboration Likelihood Model (Petty and Cacioppo 1986) and Fenno’s (1978) account of “home style” conceptually separate personal presentation from policy argument: biography functions as a credibility cue, while candidates offer issue positions as the substantive object of evaluation. The present results indicate that, in contemporary campaigns, this separation is often not maintained in campaign language. Candidates embed distinctive biographical vocabulary on issue pages, so that biographical identity is not merely adjacent to policy but textually integrated into policy claims. In doing so, candidates appear to integrate what prior research has treated as distinct strategic adaptations: populist appeals to biographical authenticity (Moffitt 2016; Mudde and Rovira Kaltwasser 2017), candidate-centered self-presentation (Balmas et al. 2014; Rahat and Sheafer 2007), and moralized issue framing (Clifford and Jerit 2013; Ryan 2017). Rather than deploying these as separate strategies of appeal, candidates combine them into a unified frame. This consolidation is particularly adaptive within a contemporary hybrid media system (Chadwick 2017), where candidates leverage owned platforms—such as the campaign websites analyzed here—to bypass traditional journalistic gatekeepers and deliver unfiltered, identity-fused policy messaging directly to voters.

Identity–policy fusion, in this sense, functions as a structural property of campaign messaging and framing rather than as a new “route” of persuasion. It describes a configuration in which candidates saturate policy statements with candidate-specific identity markers, making it harder—at the level of language—to disentangle who is speaking from what is being said. This structure is consistent with the theoretical account developed in this study, in which fusion can position policy agreement as a signal of in-group loyalty and conformity to group norms (Tajfel & Turner, 1979), and may make dissent from a fused policy more psychologically and socially costly for co-partisans by triggering cognitive dissonance and identity-protective reasoning (Festinger 1957; Kahan 2017). The observed link between fusion and higher polarization in campaign language suggests that this configuration systematically aligns issue language with identity-coded vocabulary that is unevenly distributed across parties. This alignment is consistent with a discursive environment that may reinforce the identity-protective and dissonance-related dynamics theorized here, though the present study measures language rather than individual-level processing.

Rather than simply adding authenticity to pre-existing positions, fusion appears to shape how candidates talk about issues in ways that are more distinctive across parties. This pattern is analytically distinct from the broader democratic ideal that representatives draw on their experiences and backgrounds when justifying policy positions. The contribution of the present analysis is to isolate a specific textual configuration (repeated use of distinctive, candidate-specific biographical vocabulary within policy statements) and to show that greater use of this form of fusion is systematically associated with more linguistically polarized campaign language.

The partisan asymmetry in the fusion–polarization relationship helps illuminate when and why identity–policy fusion is likely to be deployed. The theory motivating H2 proposed that fusion should be especially useful where target audiences place less trust in institutional sources of knowledge and authority, including scientific expertise, mainstream news media, and government agencies, and therefore treat them as less persuasive bases for policy claims. Consistent with this expectation, the association between identity–policy fusion and polarization in campaign language is substantially stronger among Republican candidates, who operate in an ecosystem characterized by higher levels of institutional distrust among Republican identifiers (Funk et al. 2019; Gauchat 2012). In such a context, appeals to biographical authority and lived experience can function as alternative legitimating resources for positions that diverge from expert consensus, echoing populist narratives that valorize ordinary citizens over expert elites (Mols and Jetten 2017) and broader ideological efforts to politicize and challenge scientific authority (Oreskes and Conway 2011). The stronger fusion–polarization link among Republican candidates is thus compatible with an interpretation in which identity–policy fusion is an adaptive framing strategy in campaign messages under conditions of institutional skepticism.

A second empirical contrast clarifies what identity–policy fusion is not. Generic partisan language (a statement’s similarity to the party’s generic biographical lexicon) is negatively associated with polarization in campaign language, conditional on identity–policy fusion and the other covariates in the regression models. This pattern is consistent with a “lexical dilution” effect. Generic, non-issue-specific identity terms (e.g., community, country) inject broadly shared language into issue pages, reducing reliance on issue-distinctive partisan vocabulary and narrowing measured party distance. Because these words appear across many same-party candidates’ biographies regardless of specific policy positions, they function as party-typical identity cues that are decoupled from issue content. When imported into policy statements, they add identity-coded vocabulary without the sharp partisan framing that characterizes issue-specific debate. By contrast, identity–policy fusion imports candidate-specific, highly distinctive biographical terms (i.e., high-IDF words) that co-occur with sharper partisan framing, increasing the linguistic distance captured by the polarization metric. The fusion mechanism thus operates through the integration of distinctive personal biography into policy content, not through the mere presence of identity language in general.

Limitations and Future Directions

This study has several limitations that open avenues for future research. First, as an observational analysis of campaign websites, it establishes a robust association between identity–policy fusion and polarization in campaign language, but it cannot determine the causal direction of this relationship. Candidates who pursue more polarized rhetorical strategies might also be more inclined to foreground biography on issue pages, or strategic consultants might co-produce both elements as part of a unified campaign plan, or unobserved factors such as candidate ideology or district partisanship might jointly shape both.

Second, the analysis does not measure audience reception or effects. Future research should assess whether exposure to fused messages actually changes how citizens evaluate opponents, think about compromise, or recall issue content. Survey experiments and field studies that compare fused and unfused versions of the same issue text could test whether the patterns observed in language use translate into differences in affective polarization, trust, or willingness to engage opposing views.

Third, the computational operationalization of identity–policy fusion relies on biography–policy lexical overlap, weighted by TF–IDF. This approach scales to tens of thousands of pages and captures a central dimension of fusion. Validation analyses reported in the Supplemental Material show that the TF–IDF–based measure behaves as expected across policy domains, distinguishes qualitatively low-, medium-, and high-fusion cases, and converges with alternative overlap metrics, while remaining only weakly related to document length. These checks support the measure’s validity within its scope. However, lexical overlap cannot capture all forms of fusion. Candidates can also fuse identity and policy through narrative structure, metaphor, or emotional framing that does not involve repeating specific biographical tokens. Mixed-method designs that combine computational measures with qualitative content analysis, close readings of high- and low-fusion cases, or interviews with campaign practitioners could surface additional modes of fusion and help refine this study’s lexical measure.

Finally, the study focuses on U.S. House campaigns from 2018 to 2022. Whether similar patterns appear in presidential, gubernatorial, or local races, and in other party systems and media environments, remains an open question. Comparative work could test whether identity–policy fusion is a general feature of campaigning in high-polarization, low-trust democracies or a more distinctive product of the contemporary U.S. context. Future research should also examine how candidates’ use of identity–policy fusion varies with subnational region, sex, and gender, dimensions that district partisanship controls only partially capture.

Conclusion

This paper theorizes, operationalizes, and presents large-scale evidence of identity–policy fusion as a message framing strategy that is strongly associated with polarization in campaign language. By fusing personal biography with policy claims, candidates produce issue texts that are more sharply differentiated across party lines and more tightly bound to partisan identity markers. In a political environment characterized by affective polarization and declining trust in institutions, this configuration may offer clear strategic advantages to elites but can entail significant democratic costs. Mapping how campaign language intertwines identity and policy is a necessary step toward understanding the broader communication dynamics that may threaten the capacity for evidence-based, cross-partisan problem solving. In doing so, the study offers a foundation for future work aimed at designing institutions, platforms, and campaign norms that may help sustain disagreement over policy without collapsing it into conflict over identity.

Supplemental Material

sj-docx-1-hij-10.1177_19401612261452050 – Supplemental material for Identity–Policy Fusion and Polarization in United States House Campaign Language

Supplemental material, sj-docx-1-hij-10.1177_19401612261452050 for Identity–Policy Fusion and Polarization in United States House Campaign Language by Tenzin Tamang in The International Journal of Press/Politics

Footnotes

Acknowledgements

I would like to acknowledge Prof. Jonathan J. H. Zhu for his guidance and mentorship.

ORCID iD

Tenzin Tamang

Ethical Considerations

This research relies exclusively on publicly available, secondary data from campaign websites and associated public records, including the CampaignView database. In accordance with the author’s institutional policies on the use of publicly available data, formal review by an Ethics Committee or Institutional Review Board was not required.

Consent to Participate

Not applicable. The study uses publicly available text data and secondary data on political candidates, and does not involve human subjects or primary data collection.

Consent for Publication

Not applicable. The manuscript does not contain identifiable personal data, images, or other materials requiring individual consent for publication.

Author Contributions

The author confirms being the sole contributor to the conception, design, analysis, and writing of this manuscript.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The data supporting the findings of this research are secondary data sourced from the CampaignView database of policy platforms and biographical narratives for congressional candidates (Porter et al. 2025), published in Scientific Data, 12, 1237. https://doi.org/10.1038/s41597-025-05491-x. CampaignView is openly accessible at https://campaignview.org and via Harvard Dataverse ().

Supplemental Material

Supplemental material for this article is available online.

Author Biography

Tenzin Tamang is a postdoctoral fellow in the Web Mining Lab at the Center for Communication Research, Department of Media and Communication, City University of Hong Kong. He is a computational social scientist interested in public opinion, science communication, and political communication.

References

Balmas

Rahat

Sheafer

Shenhav

S. R.

2014. “Two Routes To Personalized Politics: Centralized and Decentralized Personalization.” Party Politics 20, no. 1: 37–51.

Chadwick

2017. The Hybrid Media System: Politics and Power. Oxford University Press.

Clifford

Jerit

2013. “How Words Do the Work of Politics: Moral Foundations Theory and the Debate Over Stem Cell Research.” The Journal of Politics 75, no. 3: 659–71.

Druckman

J. N.

Klar

Krupnikov

Levendusky

Ryan

J. B.

2024. Partisan Hostility and American Democracy: Explaining Political Divisions and When They Matter. University of Chicago Press.

Ecker

U. K.

Lewandowsky

Cook

Schmid

Fazio

L. K.

Brashier

, et al. 2022. “The Psychological Drivers of Misinformation Belief and Its Resistance to Correction.” Nature Reviews Psychology 1, no. 1: 13–29.

Feinberg

Willer

2015. “From Gulf To Bridge: When Do Moral Arguments Facilitate Political Influence?” Personality and Social Psychology Bulletin 41, no. 12: 1665–81.

Fenno

R. F.

1978. Home Style: House Members in Their Districts. Little, Brown and Company.

Festinger

1957. A Theory of Cognitive Dissonance. Stanford University Press.

Finkel

E. J.

Bail

C. A.

Cikara

Ditto

P. H.

Iyengar

Klar

, et al. 2020. “Political Sectarianism In America.” Science 370, no. 6516: 533–6.

10.

Funk

Hefferon

Kennedy

Johnson

2019. Trust and Mistrust in Americans’ Views of Scientific Experts (Vol. 24). Pew Research Center.

11.

Gauchat

2012. “Politicization of Science in the Public Sphere: A Study of Public Trust in the United States, 1974 to 2010.” American Sociological Review 77, no. 2: 167–87.

12.

Gentzkow

Shapiro

J. M.

Taddy

N. M.

2017. Measuring Polarization in High-dimensional Data: Method and Application to Congressional Speech. Stanford Graduate School of Business.

13.

Hameleers

Bos

De Vreese

C. H.

2017. “‘They Did It’: The Effects of Emotionalized Blame Attribution in Populist Communication.” Communication Research 44, no. 6: 870–900.

14.

Heseltine

2025. “Asymmetric Polarization in Online Media Engagement in the United States Congress.” The International Journal of Press/Politics 30, no. 1: 299–325.

15.

Hetherington

M. J.

Rudolph

T. J.

2015. Why Washington Won't Work: Polarization, Political Trust, and the Governing Crisis. University of Chicago Press.

16.

Iyengar

Westwood

S. J.

2015. “Fear and Loathing Across Party Lines: New Evidence on Group Polarization.” American Journal of Political Science 59, no. 3: 690–707.

17.

Kahan

D. M.

2017. “Misinformation and Identity-Protective Cognition.” Yale Law & Economics Research Paper 587. https://doi.org/10.2139/ssrn.2973067

18.

Kennedy

Tyson

Funk

. 2022. Americans' Trust in Scientists, Other Groups Declines. Retrieved February 15, 2022, from https://www.pewresearch.org/science/2022/02/15/americans-trust-in-scientists-other-groups-declines/

19.

Kruikemeier

Van Noort

Vliegenthart

De Vreese

C. H.

2013. “Getting Closer: The Effects of Personalized and Interactive Online Political Communication.” European Journal of Communication 28, no. 1: 53–66.

20.

Kunda

1990. “The Case For Motivated Reasoning.” Psychological Bulletin 108, no. 3: 480.

21.

Mason

2018. “Ideologues Without Issues: The Polarizing Consequences of Ideological Identities.” Public Opinion Quarterly 82, no. S1: 866–87.

22.

McGregor

S. C.

2018. “Personalization, Social Media, and Voting: Effects of Candidate Self-Personalization On Vote Intention.” New Media & Society 20, no. 3: 1139–60.

23.

Moffitt

2016. The Global Rise of Populism: Performance, Political Style, and Representation. Stanford University Press.

24.

Mols

Jetten

2017. The Wealth Paradox: Economic Prosperity and the Hardening of Attitudes. Cambridge University Press.

25.

Mudde

Rovira Kaltwasser

2017. Populism: A Very Short Introduction. Oxford University Press.

26.

Novoa

Echelbarger

Gelman

S. A.

2023. “Generically Partisan: Polarization In Political Communication.” Proceedings of the National Academy of Sciences 120, no. 47: e2309361120.

27.

Oreskes

Conway

E. M.

2011. Merchants of Doubt: How a Handful of Scientists Obscured the Truth on Issues From Tobacco Smoke to Global Warming. Bloomsbury Publishing.

28.

Petty

R. E.

Cacioppo

J. T.

1986. The Elaboration Likelihood Model of Persuasion. Springer.

29.

Pew Research Center. 2025. Public Trust in Government: 1958-2025. Pew Research Center.

30.

Porter

Case

C. R.

Treul

S. A.

2025. “Campaignview, A Database of Policy Platforms and Biographical Narratives For Congressional Candidates.” Scientific Data 12, no. 1: 1237.

31.

Rahat

Sheafer

2007. “The Personalization (S) of Politics: Israel, 1949–2003.” Political Communication 24, no. 1: 65–80.

32.

Ryan

T. J.

2017. “No Compromise: Political Consequences of Moralized Attitudes.” American Journal of Political Science 61, no. 2: 409–23.

33.

Tajfel

Turner

J. C.

1979. “An Integrative Theory of Intergroup Conflict.” In The Social Psychology of Intergroup Relations eds. W. G. Austin and S. Worchel (pp. 33–47). Brooks/Cole.

34.

Wellings

Bonotti

Rublee

M. R.

Zech

S. T.

2024. “Narrative Fusion and Layering: Statecraft and the Shaping of Boris Johnson’s Pandemic Narrative, 2020–2021.” British Politics 19, no. 2: 234–52.

35.

Young

D. J.

Ackland

Kapounek

Madsen

J. K.

Greening

L. J.

de-Wit

2026. “A New Measure of Issue Polarization Using k-means Clustering: US Trends 1988–2024 and Predictors of Polarization Across the World.” Royal Society Open Science 13, no. 2: 251428.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.05 MB