‘Partyism’ and Commitment to Democracy

What does extant work teach us about partisans’ willingness to forego basic democratic rules if it benefits their party? Party identity represents an increasingly salient demarcation in American politics today (Huddy et al. Reference Huddy, Mason and Aarøe2015; Mason Reference Mason2018), leading some scholars to conclude that ‘it is difficult to overstate the importance of party loyalty’ (Barber and Pope Reference Barber and Pope2019, 39). Perhaps the most concerning feature of widespread ‘partyism’ is the potential to undermine Americans’ commitment to the rules and principles of democracy, driving them to see political losses as ‘existential threats that must be averted – whatever the cost’ (Finkel et al. Reference Finkel, Bail, Cikara, Ditto, Iyengar, Klar, Mason, McGrath, Nyhan, Rand, Skitka, Tucker, Van Bavel, Wang and Druckman2020, 533) and to ‘prioritize “victory” over the other side above winning “fairly”’ (Webster Reference Webster2020, 109). Partisanship has even been linked to the dehumanization of opposing-party members and support for political violence (Cassese Reference Cassese2021; Kalmoe and Mason Reference Kalmoe and Mason2022; Martherus et al. Reference Martherus, Martinez, Piff and Theodoridis2021, but see Westwood et al. Reference Westwood, Grimmer, Tyler and Nall2022).

This view seems to fit with prior work on democratic violations, which has found that voters often support, or at least disapprove less of, anti-democratic behavior if it benefits their side (Claassen and Ensley Reference Claassen and Ensley2016; Gidengil et al. Reference Gidengil, Stolle and Bergeron-Boutin2022). In a seminal study, Graham and Svolik (Reference Graham and Svolik2020, 392) find that only a few Americans ‘punish’ undemocratic behavior (such as banning rallies or restricting voting access) if doing so implies voting for the opposition, leading to the well-known conclusion that voters are ‘partisans first and democrats only second’. To be sure, there are negative consequences of anti-democratic actions on support, but these are ‘much smaller than the effects of partisanship or policy’ (Nyhan and Titiunik Reference Nyhan and Titiunik2024, 269, see also Frederiksen Reference Frederiksen2024). There is also evidence that partisans support giving elected leaders discretion over democratic principles but only when their party is in power (Simonovits et al. Reference Simonovits, McCoy and Littvay2022).

Another explanation is that partisan identification leads to such pervasive biases that voters perceive undemocratic behavior through a partisan double standard. Tomz and Weeks (Reference Tomz and Weeks2020), for example, show that partisans disapprove more strongly of foreign electoral interference helping political opponents. More broadly, people may rationalize their understanding of democracy when confronted with undemocratic behavior to realign the in-party’s behavior with democratic principles (Krishnarajan Reference Krishnarajan2023). Partisans may even have motivations to reward in-party transgressions (Solaz et al. Reference Solaz, De Vries and De Geus2019) or to derogate the opposing party in response to them (Rothschild et al. Reference Rothschild, Keefer and Hauri2021).

Different cognitive mechanisms notwithstanding (see Druckman Reference Druckman2024 for an overview), the overarching consensus in existing research is that parties and party leaders can break democratic rules with relative impunity (Finkel et al. Reference Finkel, Bail, Cikara, Ditto, Iyengar, Klar, Mason, McGrath, Nyhan, Rand, Skitka, Tucker, Van Bavel, Wang and Druckman2020; Nyhan and Titiunik Reference Nyhan and Titiunik2024; Graham and Svolik Reference Graham and Svolik2020). Partisans may somehow legitimize, tolerate, or even endorse democratic transgressions that benefit their party. When given a choice between a co-partisan candidate who has violated democratic norms and an opposing-party candidate who has not, most partisans tend to stick with their party.

Despite robust evidence that partisanship often trumps democratic principles, concerns about Americans’ willingness to tolerate undemocratic behavior may be overstated. Most existing research measures ‘punishment’ through vote choice (for example, Carey et al. Reference Carey, Clayton, Helmke, Nyhan, Sanders and Stokes2022; Frederiksen Reference Frederiksen2024; Gidengil et al. Reference Gidengil, Stolle and Bergeron-Boutin2022; Graham and Svolik Reference Graham and Svolik2020). The idea is that voters only deter candidates and parties from violating democracy’s rules if they are willing to trade off their policy preferences, ideological leanings, or partisanship for democratic compliance. While this is undoubtedly an important metric of an electorate’s ability and willingness to defend democracy against undemocratic candidates, it also sets a high bar for observing costs. Because many different considerations shape an individual’s electoral decision, anti-democratic behavior may not always outweigh these other motives in any one given choice-scenario (Aarslew Reference Aarslew2023). In this view, voter defection represents an arguably coarse measure of partisans’ tolerance of democratic norm violations. Voters may strongly disapprove of democratic transgressions yet nevertheless vote for the norm-violating co-partisan candidate.

In this study, I propose complementing the analysis of partisans’ tolerance for undemocratic behavior by using a higher-resolution measure of support: partisan affective attachments. Party affect is best described as a ‘summary evaluation’, which ‘collapses judgments about an object into a single dimension of favorability’ (Broockman et al. Reference Broockman, Kalla and Westwood2023, 809). Partisan affective attachments thus capture the individual’s sense of affective orientation towards a party, which has long been seen as the defining feature of party identification (Campbell et al. Reference Campbell, Converse, Miller and Stokes1960; Dias and Lelkes Reference Dias and Lelkes2022; Huddy et al. Reference Huddy, Mason and Aarøe2015; Iyengar et al. Reference Iyengar, Sood and Lelkes2012). Attachments and electoral support are thus two points on a support scale, moving from more to less granularity. To be clear, I am not proposing that vote choice is unimportant or that attachments represent a more accurate metric of the costs of rule-breaking. The purpose is instead to highlight that there could be real political costs associated with undermining democratic rules even though defection rates remain low. Specifically, any erosion of partisan attachments indicates that parties and candidates who violate democratic principles risk alienating their core supporters, effectively chipping away at their base.

I focus on electoral integrity violations for three reasons. First, elections are the cornerstone of democracy (Collier and Adcock Reference Collier and Adcock1999; Dahl Reference Dahl1971; Przeworski Reference Przeworski1991). Although many advocate that democracy involves more than holding free and fair elections (for example, Dahl Reference Dahl1971), democracy without elections cannot work. Second, voters consistently rate free and fair elections as essential for democracy (Carey et al. Reference Carey, Helmke, Nyhan, Sanders and Stokes2019; Chu et al. Reference Chu, Williamson and Yeung2024; Davis et al. Reference Davis, Gåddie and Goidel2022). This is important, as some of the instances of anti-democratic behavior used in previous research, such as restricting senators’ campaign activities near hospitals (Krishnarajan Reference Krishnarajan2023), may be too subtle to trigger substantive voter reactions, though they are undemocratic in principle (see Adserà et al. Reference Adserà, Arenas and Boix2023; Haggard and Kaufman Reference Haggard and Kaufman2021). But if citizens do not support democracy’s basic framework, the diffuse support that undergirds democratic legitimacy evaporates (Reference Anderson, Blais, Bowler, Donovan and ListhaugAnderson et al., 2005; Reference EastonEaston, 1975). Finally, as elections create winners and losers, they highlight partisan interests. Election losses can trigger partisan anger and sadness, and present a status-loss threat, which has been argued to increase tolerance of undemocratic actions (Finkel et al. Reference Finkel, Bail, Cikara, Ditto, Iyengar, Klar, Mason, McGrath, Nyhan, Rand, Skitka, Tucker, Van Bavel, Wang and Druckman2020; Kalmoe and Mason Reference Kalmoe and Mason2022; Webster Reference Webster2020). Electoral subversion thus pits strong partisan interests and democracy’s foundation directly against each other.

The Impact of Electoral Malpractice on Partisan Attachments

How should we expect violations of electoral integrity to shape party attachments? I argue that there are good reasons to expect that electoral transgressions change how partisans feel about their party, alienating even ardent partisans. Political psychological work on procedural fairness shows that fairness perceptions shape evaluations of authorities and institutions and are associated with considerations of institutional legitimacy (Becher and Brouard Reference Becher and Brouard2020; Hibbing and Alford Reference Hibbing and Alford2004; Smith et al. Reference Smith, Larimer, Littvay and Hibbing2007; Tyler Reference Tyler2006; Wilking Reference Wilking2011). For example, there is evidence that fairness principles constrain partisan self-interest in evaluations of election reform proposals such as same-day registrations (Mccarthy Reference Mccarthy2019; Plescia et al. Reference Plescia, Blais and Högström2020; Virgin Reference Virgin2021). And comparative research has shown that voters tend to equate democracy with free and fair elections (Adserà et al. Reference Adserà, Arenas and Boix2023; Carey et al. Reference Carey, Helmke, Nyhan, Sanders and Stokes2019; Chu et al. Reference Chu, Williamson and Yeung2024; Davis et al. Reference Davis, Gåddie and Goidel2022). Building on this body of work, I argue that electoral fairness constitutes an important democratic norm, and that election subversion, therefore, represents a direct norm violation.

What happens when the individual’s party subverts elections? Drawing on social psychological work on social identity and intergroup emotions, I argue that electoral violations by the in-party push individuals to disassociate themselves from the party. What is more, I argue that violations of electoral integrity have the effect of causing voters to report greater warmth towards the opposing party. The paper’s main argument is thus that electoral subversion by the in-party leads to a heightened sense of partisan ambivalence.

From a social identity perspective, partisans want to positively differentiate the in-party from the out-party (Greene Reference Greene2004). Evidence that the in-party has violated basic rules or norms undermines these goals, causing negative emotional reactions (Druckman et al. Reference Druckman, Gubitz, Levendusky and Lloyd2019; Van Kleef et al. Reference Van Kleef, Wanders, Stamkou and Homan2015). These emotions have direct behavioral and attitudinal consequences by making people more uncertain about their preconvictions and more open to new information (Gervais Reference Gervais2019; Valentino et al. Reference Valentino, Hutchings, Banks and Davis2008,Reference Valentino, Brader, Groenendyk, Gregorowicz and Hutchings2011). As Gervais (Reference Gervais2021, 2) notes, ‘(w)hen in-elites initiate hostile, norm-violating aggression, it undermines the perception of in-group moral superiority; it is viewed as degrading behavior, which can trigger disgust and lead co-partisans to distance themselves from the elite’. Moreover, the so-called ‘black sheep effect’ suggests that group members tend to derogate in-group deviance more than otherwise identical out-group deviance to maintain or restore a positive in-group differentiation (Eidelman et al. Reference Eidelman, Silvia and Biernat2006; Fousiani et al. Reference Fousiani, Yzerbyt, Kteily and Demoulin2019; Marques et al. Reference Marques, Yzerbyt and Leyens1988; Marques and Paez Reference Marques and Paez1994).

Recent empirical work corroborates such a line of reasoning, documenting the negative effects of major norm violations on in-group attachments. Eady et al. (Reference Eady, Hjorth and Dinesen2023), for example, demonstrate that the January 6 2021 insurrection caused a strong and lasting reduction of self-expressed partisanship among Republicans on Twitter (now X). Similarly, research on political incivility has shown that excessive insults reduce partisan attachments among supporters of the perpetrating party (Costa Reference Costa2021; Druckman et al. Reference Druckman, Gubitz, Levendusky and Lloyd2019; Frimer and Skitka Reference Frimer and Skitka2018; Skytte Reference Skytte2022). And while partisans may desire preferential treatment of the in-party, they oppose intentionally harming the opposition (Lelkes and Westwood Reference Lelkes and Westwood2017). Hence, there is some evidence that parties risk alienating their base when they go too far, breaking established norms.

What is more, work in intergroup emotions theory shows that hostility towards the out-group perceived as unjustified can cause reparative action and feelings of empathy towards the out-group (Ferguson and Branscombe Reference Ferguson, Branscombe, von Scheve and Salmela2014; Hakim et al. Reference Hakim, Branscombe and Schoemann2021; Maitner et al. Reference Maitner, Mackie and Smith2006). In a political domain, Druckman et al. (Reference Druckman, Gubitz, Levendusky and Lloyd2019) find that incivility in co-partisan media causes voters to express lower affect for the in-party and higher affect for the out-party (see also Frimer and Skitka Reference Frimer and Skitka2018). Based on this, we should expect that in-party election cheating makes voters less reliant on partisan preconvictions and less sure about the positive group differentiation. I argue that electoral violations cause co-partisans to disassociate themselves while also causing them to report less hostile attitudes towards the opposing party.Footnote ²

What happens when the opposing party violates electoral integrity? Partisans already hold the opposing party in low esteem (Abramowitz and Webster Reference Abramowitz and Webster2018); out-party cheating only serves to justify their views. For example, Gervais (Reference Gervais2019) finds that exposure to incivility by political opponents triggers feelings of anger, leading people to cling to their preexisting convictions (see also Webster Reference Webster2020). I therefore also expect that out-party transgressions exert a positive comparison effect, leading partisans to double down on support for their party. Election cheating by the out-party may cause people not only to be less favorable towards the opposing party but to simultaneously feel more attached to their party.

Research Design and Data

I conducted three survey experiments in the United States to examine partisans’ reactions to electoral subversion. All studies recruited descriptively representative samples of the US adult population using quotas based on age, sex, and region from Lucid’s online panels. Data for Study 1 (N = 5,000) were collected between 26 December 2021 and 6 January 2022, data for Study 2 (N = 1,800) were collected in September 2022, and data for Study 3 (N = 2,000) were collected between 30 November and 6 December 2022. All experiments use real-world examples of election cheating to inform the treatment material. The experiments were preregistered with OSF.Footnote ³ Table 1 presents an overview.

Table 1.

Overview of experimental studies

Note: all studies were collected through Lucid using quotas based on age, sex, and region.

I focus on the United States, as it has become central to discussions of democratic backsliding in recent years (Gidengil et al. Reference Gidengil, Stolle and Bergeron-Boutin2022; Graham and Svolik Reference Graham and Svolik2020; Krishnarajan Reference Krishnarajan2023; Simonovits et al. Reference Simonovits, McCoy and Littvay2022; Tomz and Weeks Reference Tomz and Weeks2020). On the one hand, this does pose limitations for generalizability. For example, the findings may not extend to multiparty systems, where voters can more easily shift their votes to proximate alternatives to penalize undemocratic behavior, and partisan identification may function differently (Wagner Reference Wagner2020).

At the same time, the United States also represents a ‘hard test’ for detecting changes in partisan attachments. As noted above, partisanship is particularly salient in American politics (Barber and Pope Reference Barber and Pope2019; Huddy et al. Reference Huddy, Mason and Aarøe2015; Mason Reference Mason2016). Given the high levels of polarization and partisan hostility in the United States (Abramowitz and Webster Reference Abramowitz and Webster2018; Webster Reference Webster2020), we might expect voters to be especially unwilling to update partisan sentiments, even in the face of democratic rule-violations. More importantly, the two-party system also allows me to clearly pit partisan self-interest against democratic principles, as any losses for one party necessarily advantage the other.

Treatment Material

The three studies were designed to build on each other as stepping-stones, with each experiment adding nuance to the previous ones. Specifically, because experiments aim to isolate the effects of one causal variable, ‘which means removing the “noise” of real life . . . it becomes inevitable that they look different from the multidimensional nature of “real life”’ (Druckman Reference Druckman2022b, 53). Accordingly, the experiments progressively reintroduce such ‘noise’ by moving from unambiguous violations in a clean setting to subtler offenses and, finally, to information delivered by different media sources.

First, Study 1 sets aside uncertainty about illegitimacy, aiming to examine how unequivocal electoral transgressions change how partisans feel about their party. Study 1, therefore, presents participants with unambiguous information about electoral subversion. Although this is necessarily an artificial set-up, it provides a crisp and unconfounded benchmark of partisans’ reactions. Specifically, Study 1 randomly assigned participants to read about either the Republican or the Democratic Party in a hypothetical election in their state, in which the party either won fairly (control condition) or engaged in misinformation or mail-in fraud (treatment conditions). Table 2 presents the exact wording.

Table 2.

Experimental wording (Study 1)

Note: N = 5,062. Control group = 1,684 N (33.2 per cent), treatment group 1 = 1,690 N (33.4 per cent), treatment group 2 = 1,693 N (33.4 per cent).

Two considerations were central to the design of Study 1. First, the treatments build on real-world violations of electoral integrity in recent US elections. Second, the experimental prompt needed to be sufficiently unambiguous to create a situation where partisans faced tension between election fairness norms and partisan instincts. It is worth pointing out that while there is no evidence of systematic election fraud in US elections, there have been minor, non-systematic attempts at cheating. Misinformation (such as robocalling, push polling, or sending fake letters) has featured non-trivially in American elections in recent years (Common Cause Foundation 2008; Washington Post 2020). Similarly, the absentee ballot fraud treatment draws on the 2019 election for the Ninth Congressional District in North Carolina, in which Republican candidate Mark Harris admitted to harvesting mail-in votes (New York Times 2019; Politico 2021).

The second experiment extends Study 1’s focus to more ambiguous electoral integrity violations. A concern with powerful survey experimental prompts is that the observed effects are not generalizable (Barabas and Jerit Reference Barabas and Jerit2010). As unequivocal electoral malpractice is rare, one may worry that the findings may not say much about the effects of cheating on mass behavior more broadly. Partisans may, in other words, respond quite differently to transgressions that are legally more obfuscated. Study 2 therefore adds a layer of ‘real-world noise’ by introducing ambiguity to the treatments. Specifically, it replaces the mail-in fraud condition with a gerrymandering condition (see the wording in Table 3).

Table 3.

Experimental wording (Study 2)

Note: N = 1,779. Control group = 594 N (33.4 per cent), treatment group 1 = 592 N (33.3 per cent), treatment group 2 = 593 N (33.3 per cent).

Gerrymandering is ideal for this purpose as it represents a subtle but common violation of electoral integrity in the United States (Keena et al. Reference Keena, Latner, McGann and Smith2021; Kenny et al. Reference Kenny, McCartan, Simko, Kuriwaki and Imai2023). State Assemblies redrew election districts ahead of the 2022 midterm elections, and redrawing districts is not in itself illegitimate. However, for parties and lawmakers, gerrymandering represents an opportunity to significantly change the lean and composition of Congress. Because gerrymandering gives one party an edge in a given election but is not necessarily illegal, it is an ‘example of forbearance since it involves the spirit rather than the letter of the law’ (Druckman et al. Reference Druckman, Klar and Krupnikov2024, 144). This, in turn, makes gerrymandering well suited to examine how subtle malpractice shapes political support. From a voter perspective, it may often be unclear how gerrymandering influences the outcome of an election or undermines electoral integrity. Nevertheless, scholars typically consider partisan gerrymandering a type of institutional electoral subversion (Birch Reference Birch2011).

The first two experiments provide unmediated information about election cheating, as is common in the literature. However, as voters rarely observe democratic rule-breaking directly, providing such information may not be an accurate approximation of how voters learn about violations. To test Hypothesis 3, the final experiment introduces variation in which media source provides information about in-party gerrymandering. The vignettes are presented to respondents as excerpts from mock news articles (from either MSNBC or Fox News). MSNBC is the congenial source for Democratic partisans and the uncongenial source for Republicans, and vice versa.Footnote ⁴ As in Study 2, the experiment prompts respondents to imagine that there had just been a congressional election in a typical ‘toss-up’ state, after which they read an article about the election. Table 4 shows the four possible conditions for a Republican respondent.

Table 4.

Experimental wording (Study 3)

Note: N = 2,181. In-party control = 550 N (25.2 per cent), out-party control = 549 N (25.2 per cent), in-party treatment = 541 N (24.8 per cent), out-party treatment = 541 N (24.8 per cent).

A common thread in the three experiments is the prompt to ‘imagine a scenario’ about a congressional election. To alleviate ethical concerns about deception, I opted to use hypothetical scenarios. Importantly, recent evidence shows that situational hypotheticality does not affect responses (Brutger et al. Reference Brutger, Kertzer, Renshon, Tingley and Weiss2023), and that it therefore does not lead to inflated treatment effect estimates. What is more, all treatments were based on real-world examples of electoral violations, thus avoiding constraining the findings’ inferential potential (Barnfield Reference Barnfield2022).

Findings

Before presenting the findings, I checked that randomization produced balanced groups. Next, I examined whether respondents could correctly recall the topic of the experimental vignettes (Study 1: recall = 95 per cent; Study 2: recall = 95.5 per cent; Study 3: recall = 69.1 per cent).Footnote ⁹ I return to the noticeably lower recall in Study 3 below. Comparing perceived electoral legitimacy across conditions indicates that the treatments in all three experiments were successful: respondents in the treatment groups considered the elections significantly less fair (Appendix D). Details on these tests and survey procedures, screening questions, and attention checks, as well as a discussion of ethics, are in the supplementary material. In the interest of simplicity, the main findings are presented in Figures 1–3. The supplementary material contains all accompanying tables. As preregistered, all estimates are unstandardized OLS regression coefficients with robust standard errors.

Figure 1.

Effect of unequivocal electoral integrity violations on vote choice and attachments (S1).

Note: difference between a fair election win and election cheating, estimated for both the in-party (a) and out-party (b). Dots represent point estimates and bars are 95 per cent confidence intervals using robust standard errors. N = 4,117. Associated regression tables in Appendix E.

For all three experiments, I begin by examining whether partisans are willing to defend democracy in the face of electoral rule-violations if doing so implies voting for an opposing-party candidate, which has been the predominant focus in prior studies (for example, Carey et al. Reference Carey, Clayton, Helmke, Nyhan, Sanders and Stokes2022; Frederiksen Reference Frederiksen2024; Gidengil et al. Reference Gidengil, Stolle and Bergeron-Boutin2022; Graham and Svolik Reference Graham and Svolik2020). Specifically, I compare the estimated likelihood of voting for the opposing-party candidate when the in-party has violated electoral integrity and when it has not.

Study 2: Even Subtle Election Cheating Alienates Partisans

Study 2 requires the transgression to be on the part of the respondent’s in-party and randomization into experimental conditions, therefore, occurred after blocking on participants’ self-reported partisanship. As treatment assignment was randomized conditional on an individual’s partisanship, the following regressions control for party identification.

Figure 2 presents the effects of misinformation and gerrymandering. First, the findings on misinformation closely resemble the findings in Study 1. Compared to a fair election win, misinformation causes partisans to feel markedly less attached to their party and less fervently opposed to the opposing party, but only slightly increases intentions to vote for an out-party candidate.

Figure 2.

Effects of subtle and unequivocal violations (S2).

Note: difference between a fair election win and election cheating, estimated for the in-party. Dots (black = misinformation, white = gerrymandering) represent point estimates and bars are 95 per cent confidence intervals using robust standard errors. N = 1,777. Associated regression tables in Appendix E.

Second, gerrymandering produces strikingly similar effects, albeit noticeably smaller in magnitude. When the in-party redraws electoral districts to their own advantage, only 2 percentage points are willing to cross party lines (p = 0.017). Compared to the 6 percentage point (p< 0.001) ‘punishment’ for misinformation, it is clear that sensitivity to electoral malpractice differs depending on type: voters perceive misinformation to be a greater infringement on electoral integrity than gerrymandering.Footnote ¹²

Regarding partisan attachments, co-partisan gerrymandering caused respondents to express lower affect towards their party (4.4 percentage points, p = 0.002). Recall that the treatment material did not explicitly state that the redistricting was illegitimate. In that light, a decrease of almost 5 percentage points is substantial. On the other hand, gerrymandering does not consistently improve out-party attitudes (1 percentage point, p = 0.57).

Overall, the findings from Study 2 thus demonstrate that although voters differentiate between degrees of malpractice, they disapprove of both subtle and severe transgressions. Yet only severe electoral norm violations spark reparative action on out-party attitudes.

Study 3: The Moderating Effects of Partisan Media

Study 3 moves away from the unmediated scenarios in Studies 1 and 2 by introducing media sources with known partisan leanings (Fox News and MSNBC). I consider respondents to be in one of four conditions: (1) in-party control (for example, Republicans reading about a fair Republican win from Fox News), (2) out-party control (for example, Republicans reading about a fair Republican win from MSNBC), (3) in-party treatment (for example, Republicans reading about a gerrymandered Republican win from Fox News), or (4) out-party treatment (for example, Republicans reading about a gerrymandered Republican win from MSNBC).

As the experiment in Study 3 adds another layer of experimental variation, I first examine whether that treatment layer was successful by use of a treatment recall question, asking respondents to name the media source in their vignette. A total of 647 participants (30.9 per cent) could not recall the information source. Regressing recall on a treatment indicator, media source, treatment × media source, and partisanship shows that recall rates were significantly lower in both treatment conditions, which could suggest that the additional text in these conditions distracted respondents’ attention from the media source (see also Brutger et al. Reference Brutger, Kertzer, Renshon, Tingley and Weiss2023). As a consequence, the effect estimates obtained in Study 3 are best thought of as intention-to-treat (ITT) estimates, which does not assume full compliance (Gerber and Green Reference Gerber and Green2012). I estimate the differences between these conditions by OLS regression functions, controlling for partisanship. Figure 3 shows the effect of gerrymandering information from in-party (congenial) (difference between conditions 1 and 3) and out-party (uncongenial) sources (difference between conditions 2 and 4).

Figure 3.

Effects of source congeniality (S3).

Note: difference between a fair election win and gerrymandering, estimated for congenial (black) and uncongenial (white) sources. Dots are point estimates and bars are 95 per cent confidence intervals using robust standard errors. N = 1,779. Associated regression tables in Appendix E.

Figure 3 shows that treatment effects tended to be larger when the gerrymandering news excerpt provided a congenial source cue. Gerrymandering information increases out-party voting by 3 percentage points (p = 0.002), but only when provided by an in-party source. Otherwise identical information provided by an out-party source did not affect out-party voting (p = 0.90). Likewise, the effect on in-party affect is -4 percentage points if the news come from an in-party source (p = 0.005), twice as large as when the same information comes from an out-party source (p = 0.1). As in Study 2, I observe no effect of gerrymandering on attitudes towards the opposing party.

The results from Study 3 thus suggest that gerrymandering is mainly costly when the information is provided by a co-partisan media source. However, it is worth noting that most of the estimated differences in source effects fail to reach statistical significance (see Appendix L). Indeed, only the two-way interaction estimate on out-party voting is statistically significant (p = 0.029). Overall, Study 3 thus provides only suggestive evidence that partisans are more inclined to discount information about their party’s wrongdoings if such information is provided by out-party sources.

Robustness Checks and Auxiliary Analyses

The supplementary material presents a series of robustness checks and additional analyses. The Appendix shows similar findings on two alternative measures of partisans’ feelings towards the in- and out-party: evaluations of the parties’ policies as a threat to the country’s well-being and the parties’ perceived willingness to cheat to win elections (Appendix B). The results from all studies are similar when analyzing only strong and moderate partisans and when including pure independents. Using a mock-vignette approach (Kane et al. Reference Kane, Velez and Barabas2022), the Appendix shows that the findings are unchanged by conditioning on pre-treatment attention. Moreover, using a measure of self-monitoring, a trait that captures an individual’s tendency to misrepresent their views to appease others (Berinsky Reference Berinsky2004; Webster et al. Reference Webster, Connors and Sinclair2022), Study 2 shows that it is unlikely that socially desirable misreporting drives the findings. The Appendix also demonstrates that the findings from Study 2 are similar when using a measure of social identification with the parties (cf. Huddy et al. Reference Huddy, Mason and Aarøe2015). Moreover, the supplementary material presents the estimated treatment effects conditional on beliefs about the legitimacy of the 2020 election, demonstrating that so-called ‘election deniers’ are equally sensitive to blatant transgressions but less sensitive to gerrymandering. The Appendix also decomposes the effects on voting when allowing respondents to vote for an independent candidate (Appendix M). The analysis shows that, across all three experiments, most partisans who withdrew support from their co-partisan opted for the face-saving option of an independent candidate rather than cross the party lines. Appendix N shows that the findings from the three studies are similar for Republicans and Democrats. Finally, the supplementary material presents findings for all preregistered hypotheses and tests, and shows the estimated average treatment effects on all measured outcomes.