Abstract

American politics is rife with messages designed to anger one’s political enemies. In this paper, we propose and test a model suggesting that such inflammatory messages are effective because they signal that the messenger is unwilling to compromise with the groups they have offended. Taking the example of inflammatory pro-gun messaging, we show, in three preregistered experiments with American conservatives (total n = 1742) that conservatives view such messages as offending liberals, and, to the extent that they feel negatively toward Democratic politicians, they view the producers of these inflammatory messages positively, seeing them as the sort of politician who should be trusted by conservatives and as the sort of politician that they would be willing to vote for. However, this is not true across the entire conservative electorate, as conservatives who feel more warmly toward Democratic politicians view politicians posting inflammatory messages (and therefore offending liberals) as less trustworthy and less vote-worthy. As affective polarization is on the rise in the United States, these inflammatory, bridge-burning messages may become more prevalent with politicians trying to appeal to voters who are increasingly suspicious of anyone who might be tempted to compromise.

INTRODUCTION

In April 2022, Mehmet Oz—a former TV doctor who was then running in the Republican primary in Pennsylvania for the Senate—aired a controversial advertisement. The ad included scenes where Oz was talking with a gun in his hand, scenes of men shooting firearms with Oz narrating, and a scene of Oz reloading an AR-15-style rifle. His narration included the statement that the “Second Amendment is not just about hunting. It’s about our constitutional right to protect ourselves from intruders, or an overly intrusive government.” Such ads were relatively commonplace in the 2021–2022 cycle: the Center for American Progress, a liberal think tank, noted that there were more than 100 ads displaying weapons, and these ads were often accompanied by threats against political opponents (2022). For example, Mike Collins, then a Republican candidate for a House seat in Georgia, produced an August 2021 ad where he shot a box of papers with “Nancy Pelosi’s plan for America,” and then declared “if you think I’m just going to fight the Democrats, think again. Because in the swamp, it’s RINO season too,” proceeding to shoot a cardboard cutout of a rhino. Why have these ads proliferated in recent years, and why would Oz, a seasoned media professional with notably liberal prior views, air something so controversial?

In this paper, we study this puzzle through the lens of inflammatory pro-firearm rhetoric. Such communications do not merely affirm support of the right to own guns; instead, inflammatory pro-firearm communications are messages that are likely to offend their political opponents, such as liberals and gun-control advocates. Inflammatory rhetoric is not merely about the style of language. A speech that makes frequent use of swearing is no doubt offensive; however, it is possible that this speech is found to be equally offensive by most. By contrast, the narration in Oz’s pro-gun advertisement did not include offensive phrases; however, that does not dilute the underlying inflammatory message, suggesting that it might be commendable to mount an armed rebellion against the government.

One potential reason to post inflammatory rhetoric is to seem “authentic.” Several recent studies have suggested that unfettered or politically incorrect speech can be seen by voters as a sign of authenticity, at least in part because it suggests that the speaker is unfiltered or not speaking strategically (Pillow et al., 2018; Rosenblum et al., 2020). By breaking taboos, speakers can show that they are simply unable to hold their true selves back. When it comes to inflammatory pro-firearms rhetoric, however, authenticity seems like an unlikely full explanation. In the aforementioned ad by Oz, he was holding a rifle with two optics attached to it—effectively rendering them both useless. If his intention was to communicate how much he loves guns, it appears that his attempt was halfhearted at best.

It may be that this sort of unfettered speech suggests not just that speakers are speaking honestly, but instead that they are resistant to outside influence—that they are unaffected by political correctness or other norms imposed by outsiders (Hahl et al., 2018; Pillow et al., 2018; Rosenblum et al., 2020). If someone is resistant to outside meddling, then maybe they can be truly themselves. In one study, for example, unfettered speech was found to be positively associated with perceived resistance to external influence (Pillow et al., 2018). At the same time, it is reasonable to suggest that not all displays of resistance to external influence are perceived positively by potential voters. For example, in the 2024 cycle of Republican presidential primaries, Chris Christie campaigned aggressively against Donald Trump. By campaigning against the dominant figure of the Republican Party, Christie certainly showed his resistance to external influence; however, his authenticity was not rewarded by the Republican primary voters—he dropped out well before the first primary due to lack of support. Intuitively, it is probable that one’s resistance to external influence is perceived positively only when they are resisting external influences that their audience does not like. A Republican politician who changes their stance after being criticized by Donald Trump may appear inauthentic, but it is unlikely that this failure to resist external influence (or their lack of general authenticity) would be punished by the voters who strongly support Donald Trump. We suggest, then, that communicating unwillingness to compromise with political opponents specifically may be seen as the key component of resistance to external influence.

In contemporary American politics, it seems as if it is increasingly desirable for politicians to signal their unwillingness to compromise with the members of the opposing party. Partisan Americans distrust those on the other side, and this antipathy is motivated not only by their policy positions, but by their social identity (Mason, 2018). Right-leaning American citizens especially tend to disfavor political leaders who compromise, and they are more likely to equate compromise with “selling out” (Davis, 2019). This observation is consistent with the general trend of increasing affective polarization in the United States—political polarization based not just on policy positions but rather motivated by a general sense of antipathy and distrust toward the members of the opposing political party (Iyengar et al., 2019; Yair, 2020). There is evidence from both Europe (Bettarelli & Van Haute, 2022) and the United States (Wolf et al., 2012) that more affectively polarized and ideological voters oppose politicians who are open to compromise, and that politicians who are receptive to the opposing party’s viewpoint are perceived negatively (Hussein & Wheeler, 2024).

It is possible that the uses of highly inflammatory rhetoric confer certain electoral benefits to politicians, even when such communication tends to be perceived negatively by the broader audience. In recent years, there has been an increase in incivility and inflammatory communications in political discourse, especially in political opinion media (Berry & Sobieraj, 2013). Nevertheless, it is reasonable to suggest that politicians have different incentives from personalities on partisan media. Existing studies indicate that politicians who make inflammatory remarks or display incivility tend to be perceived negatively (Feinberg & Frimer, 2023; Frimer & Skitka, 2018). Although angry tweets made by politicians tend to receive more engagement (Frimer et al., 2023; Webster, 2021), it is unclear if increased social media engagement outweighs the cost of being perceived negatively by their voters. In attempting to unpack the puzzle, some researchers suggest that, since voters in general tend to dislike negative partisan rhetoric, politicians who use such rhetoric must be motivated by other incentives (Costa, 2025).

How can inflammatory pro-firearms rhetoric communicate a candidate’s unwillingness to compromise with political (and ideological) opponents? We propose that signaling theory can provide a simple, effective explanation. Signaling theory (Grafen, 1990; Spence, 1973) is a game-theoretic framework that focuses on situations with asymmetrical information. In signaling theory, there are two parties with conflicting interests. The sender of a signal possesses a quality that is invisible to the receiver of the signal. The receiver would like to accurately determine the quality of the sender because the outcome of their interaction with the sender depends on the sender’s quality. The sender has the incentive to appear as if they possess high quality because the sender can benefit from interacting with the receiver. In such a situation, signaling theory suggests that a certain behavior can function as a reliable signal of quality if the cost of producing the signal is lower for those of high quality than those of low quality.

We argue that a similar signaling dynamic exists between political candidates and their audience. It is plausible that a Republican candidate running in a primary election would want to signal their loyalty toward conservative causes—for example, by communicating that they support issue positions popular among Republican voters. However, such communications would not make effective signals themselves because the cost of producing the communications would not be related to how loyal they are to the conservative cause. Any Republican candidates running in a primary would be expected to express support for gun owners. Indeed, even Kamala Harris, running as a Democratic presidential candidate, proclaimed that she is a gun owner (Levien, 2024).

Using inflammatory pro-firearms rhetoric, on the other hand, could be seen as costly because doing so is likely to evoke a very negative reaction from the Democrats. We suggest that a person who does not wish to maintain a relationship with someone else would lose nothing by ruining their relationship with this person (after Ohtsubo & Watanabe, 2009). Using inflammatory rhetoric, which can potentially burn bridges with the members of the other party, would be very costly for a politician who wishes to maintain a working relationship with them. However, using the same rhetoric would be much less costly for a politician who has no intention to work with the other party to begin with. Therefore, the use of such language would be able to function as a reliable signal of one’s unwillingness to work with one’s political opponents. For example, talking about an armed rebellion against the government would be perceived as deeply offensive to the Democrats and traditional mainstream Republicans. Offending these groups should be less costly for a politician who does not wish to work with these groups and, therefore, does not care about how they are perceived by the members of these groups. For voters who abhor any potential compromises with the Democrats (or with traditional Republicans), a communication that threatens armed rebellion against the government—as Oz did in his ad—may be seen as a signal of desirable quality even if voters themselves do not endorse the idea of armed rebellion; the point of the signal is whom the signaler antagonizes, rather than the actual issue position being expressed by the signaler.

In signaling theory, the reliability of a signal can be ensured by the association between higher quality being signaled and lower cost of producing the signal. Therefore, inflammatory rhetoric that offends everyone would not be an effective signal of one’s unwillingness to compromise with political opponents; doing so would be similarly costly regardless of one’s willingness to compromise with their opponents. Instead, to function as an effective signal of one’s unwillingness to compromise with a certain group, the inflammatory rhetoric needs to specifically antagonize the members of this very group, so that they can signal their unwillingness to cooperate with this group. Opinions related to gun control may be especially useful, as firearms are increasingly becoming important parts of partisan identities (Lacombe, 2019), where the pro-gun control view is associated with the Democrats and the pro-Second Amendment (or anti-gun control) view is associated with the Republicans. Contemporary pro-gun rhetoric frequently adapts the narrative that the Democrats are trying to infringe on the constitutional rights of gun owners (Lacombe, 2021), therefore, serving to antagonize the Democrats.

In summary, then, we believe that this rise in pro-firearms inflammatory rhetoric among Republican politicians is explicitly designed to indicate to voters that the politician is unwilling—and after the expression of the rhetoric, unable—to work with or compromise with members of the Democratic Party, regardless of whether the voters believe that the politician is serious in their speech. “Triggering the libs” is a bridge-burning exercise, and we expect that Republicans will see bridge-burners as more trustworthy to Republican causes to the extent that they understand their inflammatory rhetoric as antagonizing their enemies in the Democratic Party.

To test this framework, we designed three preregistered experiments. We recruited self-identified conservatives from a large online sampling pool, and we showed them a series of social media posts that we had constructed, based on preliminary analyses of social media posts by the Republican congressional candidates in the 2022 cycle, to be either inflammatory or not. By then measuring these participants’ opinions about the hypothetical posters of these messages, we can test whether inflammatory rhetoric holds our proposed signaling function. See Figure 1 for a proposed model.

Proposed conceptual model.

STUDY 1

In Study 1, we test our proposed model in a preregistered fully within-subjects design by presenting participants with hypothetical social media posts made in the aftermath of a mass shooting by imaginary candidates running in a Republican primary for the House of Representatives. A mass shooting event would make the issue of gun control highly salient, and it is plausible that congressional candidates would make social media posts in response to a mass shooting event. We expected that candidates who made an inflammatory pro-firearms tweet would be perceived as less likely to compromise with the Democrats than candidates who made a tweet expressing thoughts and prayers, and that this relationship would specifically be mediated by the perceived (negative) reaction to the tweet by the liberal audience. We further expected that these perceptions would have consequences for the candidate: that participants would be more willing to vote for the candidates who made an inflammatory pro-firearms tweet than for the candidates who made a thoughts and prayers tweet, and that this relationship would again be mediated by the candidate’s perceived willingness to compromise with the Democrats.

Finally, following from the predictions of signaling theory, we suspected that the benefits of inflammatory language would be especially strong for candidates who would otherwise read as somewhat suspicious to our participants. There may be no need to worry about potential partisan betrayal by stereotypical conservatives, but voters may need to be convinced that non-stereotypical conservatives will not sell them out. We therefore created two sets of candidates, one with stereotypical backgrounds for a Republican primary and one with less traditional backgrounds. The less traditional backgrounds are designed to imply that the candidates’ loyalty toward the conservatives might be suspect; in creating these profiles, we were inspired by recent high-profile political candidates with nontraditional backgrounds who are known for their use of inflammatory rhetoric, including Oz. We suspected (albeit without making it a confirmatory hypothesis) that the benefits of inflammatory rhetoric for partisan trustworthiness would be especially strong for candidates with otherwise suspicious backgrounds. We did predict, in a confirmatory way, that candidates with more stereotypical backgrounds would be treated as more trustworthy.

In these and other studies in this manuscript, we focus exclusively on the attitudes of conservatives. While we do not think that this phenomenon of signaling by inflammatory rhetoric is exclusive to the right wing of American politics, we chose to focus on conservatives alone in order to maximize the power of our statistical tests. Between-subjects interactions require notoriously large samples to properly power for (see Sommet et al., 2023 for a discussion), and so we chose to drop liberals from our participant sample for the sake of efficiency. Given that it is plausible to assume that liberal voters are not the intended audience of inflammatory pro-firearms tweets, we focus our analysis on conservative-leaning participants.

Method

Participants

We recruited 392 self-reported conservatives from Prolific in September 2023. 224 (57.1%) of the participants reported being male. The median age range selected by the participants was between 35 and 45 years old, and 192 (49%) of the participants had a bachelor’s degree or more. After removing participants who failed attention checks and participants who did not agree to be included in the analysis, we were left with 301 participants. All participants had identified themselves as conservative before participating in the study; therefore, it is unsurprising that the majority of the participants rated themselves as very conservative. The median self-rated ideology was conservative (six on a seven-point scale), and 184 out of 301 participants rated themselves as very conservative (7) or conservative (6). Nevertheless, 81 participants rated themselves as somewhat conservative (5), and 29 participants rated themselves as moderate (4). Because our respondents were already screened based on their self-identification on Prolific, as per our preregistration, we did not remove the seven participants who identified themselves as liberal during the experiment. Given that they identified themselves as conservatives on one occasion (on Prolific), we do not believe that their inconsistency warrants their exclusion. However, when, in exploratory analyses, we exclude these seven participants, our conclusions do not change.

Design

The study used a within-subject design. Participants were asked to evaluate four vignettes, each containing a brief profile of a Republican House primary candidate and their hypothetical social media post in the aftermath of a mass shooting. There were four possible candidate profiles (two with loyal backgrounds and two with suspect backgrounds) and four possible social media posts (two inflammatory and two thoughts and prayers). Each participant evaluated all four combinations of treatment conditions (loyal/inflammatory, loyal/thoughts and prayers, suspect/inflammatory, suspect/thoughts and prayers) in random order and encountered all candidate profiles and social media posts. Actual combinations of profiles and social media posts were randomized, and all potential combinations of profiles and social media posts had an equal likelihood of being shown to the participants. For each profile/social media post combination, a participant was asked to answer six questions. At the end of the survey, the participants answered questions on their demographics and attention check questions.

In the loyal background condition, a focal candidate had a background that would be seen as typical for a Republican political candidate: one candidate was a successful local businessman who had been a prolific Republican donor, and the other was a former Republican political operative. In the suspect-background condition, a focal candidate had a background that would be seen as unusual for a Republican political background, and their profile implies that they may have switched political affiliation relatively recently: one candidate was a writer who used to be liberal-leaning, and the other was a former environmental activist. We conducted an initial pretest with 99 self-identified conservatives recruited from an undergraduate research pool and found no evidence that the two different loyal candidates, without any tweets, were viewed as differently unwilling to compromise with Democrats (Ms = 3.5 and 3.6, p = .59) or as people that conservatives could trust (Ms = 5.6 and 5.6, p = .95); while we similarly found no evidence that the two different suspect candidates were viewed as differently willing to compromise (Ms = 4.9 and 4.8, p = .61). We did find a suggestion that the two different suspect candidates were viewed as somewhat different in terms of the degree to which they were viewed as someone that conservatives should be wary of (Ms = 3.0 and 3.5, p = .024). Regardless of this difference, we prioritized having similar perceived willingness to compromise within each condition. In our pretest data, those with loyal backgrounds were, on the whole, viewed as less likely to compromise with Democrats (p <.001) and as people that conservatives could trust (p <.001).

We modeled inflammatory pro-gun social media posts from actual messages sent by Republican politicians during the 2022 primaries. They contain strong pro-gun sentiment and are highly antagonistic toward their political opponents. We then constructed messages conveying “thoughts and prayers” expressing condolences for the victims of the hypothetical mass shooting event, but otherwise abstaining from discussion of issues related to gun ownership. It is important to note that the thoughts and prayers tweets do not express pro-gun-control views; rather, they tacitly support the status quo related to firearms in the aftermath of a mass shooting event by refusing to engage in gun-related discussions.

Each social media post was designed to simulate a tweet, and each tweet is accompanied by a computer-generated picture of a middle-aged white male. Specifically, the two inflammatory tweets read: “Yet another mass killing by a crazed criminal, and the Delusional Dems are still obsessing on taking guns away from law-abiding Americans. I’ll fight tooth and nail to protect our Second Amendment rights and ensure American families can defend themselves” and “The bodies are still warm after the horrific shooting and the Radical Left are already talking about taking away our guns. They are scared that we can fight back against their tyrannical agenda with our guns. I will always fight to stop the Left’s anti-gun agenda!” By contrast, the two thoughts and prayers tweets read: “My heart goes out to the victims of this tragic event and their families. This senseless act of violence is devastating, and we must come together as a nation to support each other during this difficult time. We should not widen the political divide by politicizing the tragedy.” and “As we mourn the lives lost in yet another tragic mass shooting, we must examine our society’s cultural challenges. Promoting empathy, respect, and understanding is essential. We can work together to overcome these challenges.” The images used for these tweets are in the Supplemental Material.

A study of tweets made by congressional members in response to the Parkland shooting shows that Republican members of Congress were more likely to express support and were less likely to criticize others or comment about policies (Copenhaver et al., 2023). Therefore, we argue that our thoughts and prayers tweets represent a common type of Republican response to a mass shooting event. We avoided using more direct pro-firearms statements because we suspected that making such statements in response to a mass shooting event was likely to be seen as confrontational. Finally, we note that neither the inflammatory pro-gun social media posts nor the thoughts and prayers social media posts include concrete policy suggestions—instead, they convey a more general attitude.

We pretested both the backgrounds of the candidates and the content of the tweets with 99 self-reported conservatives. On the whole, the inflammatory tweets were rated as significantly more confrontational (p < .001) and significantly less conciliatory (p < .001) than the thoughts and prayers tweets. We found no evidence for difference between the two inflammatory tweets when rated on how confrontational (Ms = 5.5 and 5.6, p = .59) or conciliatory (Ms = 2.9 and 2.7, p = .38) they seemed; nor did we find evidence for difference between the two thoughts and prayers tweets when rated on how confrontational (Ms = 2.3 and 2.2, p = .58) or conciliatory (Ms = 5.3 and 5.4, p = .83) they seemed.

Measures

For each vignette, the participants were asked to answer the following questions. Questions about perceptions toward candidates were asked with 7-point strongly disagree to strongly agree questions. Perceived partisan trustworthiness was measured by: “[Candidate] is the type of politician [conservatives/liberals] can trust to be on their side.” Perceived willingness to compromise with Democrats was measured by: “[Candidate] is willing to make the compromises with the politicians of the Democratic Party.” Likely reaction by liberals/conservatives was measured by: “[Liberals/Conservatives] will react positively to [candidate]’s tweet.” Voting intention was asked by a 4-point (very likely, somewhat likely, somewhat unlikely, very unlikely) question: “If you were a voter in [candidate]’s district, how likely would you be to vote for him?”

We note that we used the terms “liberals” and “conservatives” to describe the partisan audience; these terms are more commonly used than more specific terms such as “supporters of the Democratic Party” or “Republican voters,” and these terms can encompass both the regular supporters and political elites. On the other hand, we used “the politicians of the Democratic Party” to describe the Democratic politicians because politicians in the United States are more often described as being Republicans or Democrats, rather than being liberal or conservative (see e.g., Kalmoe, 2020; Neiheisel, 2016 for discussions of partisanship vs. ideology).

Analytic plan

We report how we determined our sample size, all data exclusions (if any), all inclusion/exclusion criteria, whether inclusion/exclusion criteria were established prior to data analysis, all manipulations, and all measures in the study. The preregistration for this study, along with the preregistered analysis script, can be found at https://osf.io/gpz5f/. A sensitivity analysis showed that a sample size of 400 was able to achieve 80% power to detect a main effect of Cohen’s f = .073 for a four-condition within-subjects design. This provided a good tradeoff between the cost of sampling and statistical power. We planned to remove participants who completed the survey in less than 100 s, participants who indicated they did not wish their answers to be included in the analysis, and participants who failed the attention check questions.1

We used mixed-effect, random intercept models (with random intercepts for participants) to test the main effects of the treatment conditions in order to account for the issue of non-independent outcome variables from repeated measures (Brauer & Curtin, 2018). We used a random-intercept model rather than a random-slope model due to the limited number of treatments per treatment condition. Mediation models were tested using the mediation package in R (Tingley et al., 2014).

Results

Perceived willingness to compromise

We expected that candidates who made an inflammatory pro-firearms tweet would be perceived as less likely to compromise with the Democrats than the candidates who made a thoughts and prayers tweet, and that this relationship would specifically be mediated by the perceived (negative) reaction to the tweet by the liberal voters.

As predicted, we found that the candidates who made a thoughts and prayers tweet were perceived as more likely to compromise with the Democrats (Minflammatory = 2.41, SEinflammatory = .0598, Mt&p = 4.89, SEt&p = .053, b = 2.47 [2.32, 2.61], SE = .11, t = 24.20, p <.001, d = 1.93 [1.79, 2.06]).

As predicted, causal mediation analysis found that the relationship between the rhetoric of tweets and perceived willingness to compromise was mediated by the perceived liberal reaction, Average Causal Mediation Effect (ACME) = 1.13 [.98, 1.26], p <.001. 45.6% of the direct effect was explained by our mediating variable. See Figure 2 for the result of causal mediation analysis.

Causal mediation analysis, Study 1. Rhetoric is coded as 0 for inflammatory and 1 for thoughts and prayers. ***p < .001.

In an exploratory model, we tested whether our results were better explained by perceived responses to the tweets by liberals (as we had predicted) or perceived responses to the tweets by conservatives. Our signaling-based explanation of inflammatory pro-firearms rhetoric suggests that a signal’s effectiveness (in signaling unwillingness to compromise) is not driven by the issue position expressed in the signal, but by the perception that the signal would antagonize their political opponents—therefore, perceived unwillingness to compromise should be associated with the perceived reaction by liberals. On the other hand, if an inflammatory pro-firearms tweet signals unwillingness to compromise because of its support for the pro-firearms position, perceived unwillingness to compromise should be associated with perceived positive conservative reactions; it is the conservative audience that should appreciate the display of the strong pro-firearms issue position.

When we put both perceived reactions into a model as predictors, we find that the effect of perceived reaction by liberals is statistically significant (b = .69 [.65, .73], SE = .021, t = 32.98, p < .001), but the effect of perceived reaction by conservatives is not (b = −.005 [−.058, .049], SE = .027, t = −.17, p = .87). The absolute values of two 95% CIs do not overlap, indicating that the effect of perceived liberal reaction is significantly larger than the effect of perceived conservative reaction. This result indicates that the effect of rhetoric on perceived willingness to compromise is driven by perceived response by liberals, as we had predicted, with no evidence for the effect of the perceived response by conservatives. This result supports our arguments that inflammatory pro-firearms posts signal unwillingness to compromise by antagonizing the opposing side, rather than showcasing the pro-firearms issue position.

In another exploratory model, we tested how the candidate’s background affected perceived willingness to compromise. The candidates with suspect backgrounds were perceived as more likely to compromise with the Democrats (Msuspect = 3.82, SEsuspect = .077, Mloyal = 3.49, SEloyal = .073, b = .33 [.12, .54], SE = .11, t = 3.12, p < .001, d = .26 [.15, .37]). In a model which includes both treatment conditions (rhetoric and background), the interaction effect was not statistically significant (b = −.126 [−.416, .16], eta-squared = .0008 [.00, .01], SE = .15, t = −.85, p = .39). In the same model, we found that candidate background had a main-effect partial eta-squared of .02 [.01, .04]. On the other hand, we found the tweet rhetoric had a main-effect partial eta-squared of .55 [.45, .59], indicating that tweet rhetoric had a substantially larger effect on perceived willingness to compromise than did candidate background.

Voting intentions and perceived partisan trustworthiness

We further expected that participants would be more willing to vote for the candidates who made an inflammatory pro-firearms tweet than for the candidates who made a thoughts and prayers tweet, and that this relationship would again be mediated by the candidate’s perceived willingness to compromise with the Democrats. Additionally, we predicted that the candidates with the suspect backgrounds—the backgrounds which are unusual for Republican candidates—would be perceived as less trustworthy to conservatives and participants would be less willing to vote for such candidates.

As predicted, the participants were less willing to vote for more suspect-background candidates (Msuspect = 2.58, SEsuspect = .039, Mloyal = si, SEloyal = .036, b = −.37 [−.47, −.27], SE = .50, t = 7.39, p < .001, d = −.43 [−.54, −.31]), but unexpectedly, we found that the participants were not significantly more willing to vote for the candidates with inflammatory rhetoric (Minflammatory = 2.75, SEinflammatory = .042, Mt&p = 2.77, SEt&p = .035, b = .023 [−.078, .12], SE = .051, t = .45, p = .65, d = .027 [−.086, .14]). We found no evidence for a significant interaction effect of two treatment conditions (b = −.17 [−.37, .022], SE = .10, t = −1.73, p = .084, eta-squared = .00333 [.00, .01]).

As predicted, participants found the candidates with thoughts and prayers tweets to be less trustworthy toward conservatives (Minflammatory = 5.34, SEinflammatory = .063, Mt&p = 4.64, SEt&p = .063, b = −.701 [−.865, −.537], SE = .0837, t = −8.371, p < .001, d = −.510 [−.625, −.395]). As predicted, the candidates with suspect backgrounds were perceived as less trustworthy toward conservatives (Msuspect = 4.58, SEsuspect = .069, Mloyal = 5.39, SEloyal = .055, b = −.81 [−.97, −.65], SE = .083, t = −9.81, p < .001, d = −.59 [−.71, −.47]). Background and rhetoric significantly interacted when predicting perceived trustworthiness toward conservatives (b = −.31 [−.62, −.002], SE = .16, t = −1.97, p = .049, eta-squared = .0043 [.00, .02]), indicating that candidates with suspect backgrounds and thoughts and prayers tweets were perceived as especially untrustworthy toward conservatives.

Interactions by ideology

In exploratory models, we examined whether certain sorts of voters are more or less willing to vote for the more inflammatory candidates. We find that our hypotheses hold for voters who identify themselves as more conservative, but not among more moderate voters. The more conservative participants were, the more willing they were to vote for the candidates with inflammatory tweets (b = .31 [.24, .39], SE = .039, p < .001); but the less willing they were to vote for the candidates with thoughts and prayers tweets (b = −.13 [−.20, −.051], SE = .039, p = .0011); overall interaction: b = −.44 [−.53, −.35], SE = .047, t = −9.39, p < .001, eta-squared = .09 [.06, .13]; see Figure 3.

Voting intention by ideology, Study 1. Shaded areas indicate 95% CIs.

Similarly, we find evidence that more conservative participants were more likely to rate the candidates with inflammatory tweets as trustworthy toward conservatives (b = .18 [.053, .31], SE = .063, p = .0053), but not the candidates with thoughts and prayers tweets (b = −.083 [−.21, .041], SE = .063, p = .19, overall interaction b = −.26, 95% CI [−.41, −.11], SE = .077, t = −3.37, p < .001, eta-squared = .01 [.00, .03]); see Figure 4. All data and analysis scripts can be found at https://osf.io/gpz5f/.

Perceived partisan trustworthiness by ideology, Study 1. Shaded areas indicate 95% CIs.

Discussion

In a preregistered within-participants study of the effects of inflammatory pro-firearms rhetoric, we find evidence that the outrage this rhetoric creates among enemies is, in some sense, the point. Consistent with predictions derived from signaling theory, the perceived negative reaction by liberals acts as a signal that a candidate is less likely to compromise with the Democrats. Among more conservative participants, this outrage is regarded positively, as these participants tended to favor the candidates who made an inflammatory post, whereas the more moderate participants tended to favor the candidates who instead adopted a more conciliatory rhetoric.

Our finding suggests that not all conservatives reject candidates who may be interested in working with Democrats. In retrospect, this is not very surprising. For example, in a recent survey of Republican voters for the 2023 Republican presidential primary election, supporters of Donald Trump tended to prefer candidates who do not compromise with Democrats, whereas the supporters of Nikki Haley tended to prefer candidates who were more open to compromise with the other side (Pew Research Center, 2023). It is possible that the participants in this experiment held similarly heterogeneous attitudes toward compromises, and the participants’ attitudes toward potential compromise with the Democrats are associated with their self-described ideology. In our theoretical model, we suggested that affective polarization would lead the highly partisan to shun politicians who would be willing to work across the aisle. Therefore, it would be desirable for politicians targeting such an audience to signal that they are unwilling to compromise with the opposing side. We suggest that it is likely that the participants who described themselves as more conservative tended to feel more negatively toward the Democrats. Therefore, the more conservative participants tended to perceive the possibility of compromises with the Democrats more negatively and preferred the candidates whom they perceived as unwilling to compromise with the Democrats.

Contrary to our initial expectations, it seemed that our participants treated inflammatory candidates as inflammatory and conciliatory candidates as conciliatory, without reference to their backgrounds. Partially, we speculate that this may be because our manipulation of candidate background was too subtle: 87 (22.2%) of participants completely failed to pick it up in a manipulation check on candidate backgrounds, but it may also be the case that observers simply do not care about background in this case. In models with both candidate background and tweet content, the effect of tweet content (eta-squared = .55) in predicting perceived willingness to compromise was far larger than the effect of candidate background (eta-squared = .02). What may matter is what a candidate does, not who they are, and that bridge-burning is effective no matter who holds the match.

In our initial theorizing, we had focused more on the message, but it is clear that different audiences will receive the same message in different ways. Inflammatory pro-firearms rhetoric should only appeal to those who feel very negatively toward their political opponents and therefore regard compromise as a dirty word. We explicitly test this prediction in a follow-up study.

STUDY 2

In Study 2, we investigate whether affective polarization is a key variable in understanding who will be attracted to candidates who espouse inflammatory pro-firearm rhetoric. Specifically, we look to replicate and extend the findings of Study 1, examining, in a preregistered study, whether their dislike of Democrats, a key variable in operationalizing affective polarization, helps to predict approval of candidates that use inflammatory rhetoric.

In Study 1, we found exploratory evidence that those who described themselves as more strongly conservative were more likely to vote for candidates who used inflammatory language. Existing studies on affective polarization make a distinction between positive attitudes toward one’s own side and negative attitudes toward the opposing side (Iyengar et al., 2019). Given that our argument is based on the idea that the highly conservative participants, due to their negative out-partisan attitudes, would abhor candidates who would consider compromise with the Democrats, we infer that it is their negative attitudes toward the Democrats—rather than their positive attitudes toward the Republicans—that drove the observed interaction effect between the participants’ self-described ideology and rhetoric. Participants who have a more negative attitude toward the Democrats are more likely to perceive candidates’ willingness to compromise as a sign of partisan disloyalty, and they would be less inclined to vote for such candidates.

Because we found no evidence in Study 1 for a significant interaction effect between the contents of social media posts and candidate background for voting intention, and because the effects of background were relatively minimal, we dropped candidate backgrounds from this study.

As in Study 1, we predicted that candidates who made an inflammatory pro-firearms tweet are perceived as less likely to compromise with the Democrats than the candidates who made a thoughts and prayers tweet, and that this relationship would be mediated by the perceived (negative) reaction to the tweet by the liberal voters. We also expected that candidates who used more inflammatory language would be perceived as someone whom conservatives could trust and someone whom our participants would be more likely to vote for (again, as mediated through perceived negative reaction by liberals). Critically, we expected that partisan trustworthiness and voting intention would be especially strong for more conservative participants (mirroring Study 1) and for affective partisans. The more a person dislikes Democrats, the more appealing they should find the more inflammatory candidate.

Method

Participants

We recruited 501 self-described conservatives in the United States on Prolific in January 2024. 45.7% of our participants reported being male. The median age category was 35 to 45 years old, and 258 participants (51.5%) had a bachelor’s degree or more. After removing participants who failed attention checks, participants who did not agree to be included in the analysis, and participants who completed the survey in an unusually short amount of time (below 100 s), we were left with 487 participants. The distribution of self-described ideology among the 487 participants was relatively similar to that of the participants from the first study. The median participant’s ideology was conservative. Two hundred eighty-two (57.9%) participants described themselves as conservative or very conservative, whereas 205 (42.1%) participants described themselves as somewhat conservative or less.2 Participants tended to have negative attitudes toward the Democrats, with a mean feeling thermometer rating of 27.1 (out of 100, with 0 being extremely cool and 100 being extremely warm) and a median rating of 23. These values are in line with previous research on partisan attitudes toward outpartisan members (Iyengar et al., 2019).

Design

Our study design was quite similar to Study 1—again, participants completed a fully within-subjects design in which they were given a tweet by a hypothetical candidate for a Republican primary posted in the aftermath of a mass shooting and asked to evaluate the hypothetical candidates. The simulated social media posts were identical to the ones from Study 1, but unlike in Study 1, participants received no additional information about the candidate beyond their name, photograph, and that the candidate was running for a Republican House primary.

Measures

For each vignette, participants were asked to answer the same perceived partisan trustworthiness toward conservatives, perceived willingness to compromise with the Democrats, likely reaction by liberals/conservatives, and voting intention questions as in Study 1. At the end of the survey, participants answered questions on their demographics and attention check questions. In addition to the question about participants’ self-described ideology, participants were also asked two feeling thermometer questions, one toward politicians from the Republican Party and one toward politicians from the Democratic Party. The feeling thermometer questions asked the participants to move sliders between the values of 0 (coldest) and 100 (warmest) when evaluating their feelings toward the focal group.

Analytic plan

We report how we determined our sample size, all data exclusions (if any), all inclusion/exclusion criteria, whether inclusion/exclusion criteria were established prior to data analysis, all manipulations, and all measures in the study. The preregistration, along with the preregistered analysis script, can be found at https://osf.io/uc2rs/. Sensitivity analyses suggested that a target sample size of 500 would provide 80% power to detect an effect of Cohen’s f of .052 in our design—more than enough power to detect effects in line with the f ~ .10 found in Study 1. We used the same exclusion criteria as in Study 1.

As in Study 1, we used mixed-effect models with random intercepts for participants to test our hypotheses.

Results

Main effects

Willingness to compromise

We predicted that candidates who made an inflammatory pro-firearms tweet would be perceived as less likely to compromise with the Democrats than the candidates who made a thoughts and prayers tweet, and that this relationship would be mediated by the perceived (negative) reaction to the tweet by liberal voters.

As predicted and consistent with the first study, we found that the candidates who made a “thoughts and prayers” tweet were rated as more likely to compromise with the Democrats than those with more inflammatory posts (Minflammatory = 2.20, SEinflammatory = .044, Mt&p = 4.94, SEt&p = .039, b = 2.74, SE = .052, t = 52.90, p < .001, d = 2.40 [2.28, 2.51]). As predicted, causal mediation analysis found that the relationship between the rhetoric of tweets and perceived willingness to compromise was mediated by the perceived liberal reaction, ACME = .82, [.73, .90], p < .001. 29.7% of the direct effect was explained by our mediator. Figure 5 shows the result of causal mediation analysis.

Causal mediation analysis, Study 2. Rhetoric is coded as 0 for inflammatory and 1 for thoughts and prayers. ***p < .001.

As in Study 1, we tested whether our results were better explained by perceived responses to the tweets by liberals (as we had predicted) or perceived responses to the tweets by conservatives. When we put both perceptions into a model as the sole predictors, we find that both the effect of perceived reaction by liberals (b = .69 [.66, .73], SE = .017, t = 40.91, p < .001) and the effect of perceived reaction by conservatives (b = −.05 [−.094, −.006], SE = .022, t = −2.24, p = .025) are both statistically significant. However, a direct comparison of the 95% CI of the absolute values of the two effects shows that the two do not overlap, and therefore that perceived responses by liberals have a significantly stronger effect on partisan trustworthiness than perceived responses by conservatives. Again, the result indicates that inflammatory pro-firearms posts signal unwillingness to compromise through antagonizing the opposing side, rather than through their implied pro-firearms issue position.

Trustworthiness and voting intentions

We predicted that candidates who used more inflammatory language would be perceived as someone whom conservatives could trust, and someone whom our participants would be more likely to vote for, with both effects mediated th