Introduction
Emerging wars around the globe and risks of democratic backsliding provide two dire challenges to the political system of the United States. A historical and contemporary adversary to the United States, Russia, increasingly constitutes a direct threat to the NATO alliance, most prominently exemplified by the war in Ukraine. A historical ally, Israel, is often engaged in conflicts in the Middle East. Finally, the great power of China provides an ever-changing, latent threat to the geopolitically crucial Taiwan. Meanwhile, American democracy is under pressure internally and even considered a backsliding democracy by several metrics (Levitsky and Ziblatt Reference Levitsky and Ziblatt2018; Bartels and Carnes Reference Bartels and Carnes2023).
In this letter, we examine how these two challenges to the political system of the United States interact – with a focus on how interstate conflict affects American voters’ propensity to support candidates behaving undemocratically. Our argument is that the interaction between war and democratic backsliding is gloomy: threats of interstate conflict increase the appeal of undemocratic candidates to citizens. This argument brings together recent work on citizens’ tolerance of undemocratic behavior (Graham and Svolik Reference Graham and Svolik2020; Simonovits et al Reference Simonovits, McCoy and Littvay2022; Krishnarajan Reference Krishnarajan2023) with social psychological insights on voters’ preferences for candidate personality and character in different contexts (for example, Laustsen and Petersen Reference Laustsen and Petersen2020; Kakkar and Sivanathan Reference Kakkar and Sivanathan2017). We specifically argue that undemocratic behavior enhances dominance impressions – being tough-minded, aggressive, and intimidating – which, in turn, are valued candidate qualities in the eyes of citizens during intergroup conflict.
We leverage a pre-registered survey experiment (N = 2,611) – randomly assigning undemocratic behavior to in-party presidential candidates and priming threats of real-world disputes with China and Russia in a factorial design – to test our novel argument that interstate conflict increases the appeal of undemocratic candidates to citizens. Besides this main study, we fielded two pre-registered conjoint experiments to examine the relationship between undemocratic candidate behavior and dominance impressions in a rich information environment (N = 18,395 candidate observations).
The conjoint experiments show that candidates who behave undemocratically are perceived as more dominant than democratically compliant candidates. In further support of our theoretical argument, the main study shows that voters dislike an undemocratic presidential candidate significantly less during threats of interstate conflict than during peace. We also find substantial partisan heterogeneity in the findings of our main study, revealing that Republican voters – as compared to Independents and Democrats – are most swayed by threats of interstate conflict.
To the best of our knowledge, we provide the first study theorizing and empirically testing how contemporary threats of interstate conflict affect support for would-be authoritarians in the United States. In doing so, we contribute to three streams of research. We contribute to the literature on how international relations affect American public opinion (Tomz and Weeks Reference Tomz and Weeks2020; Berinsky Reference Berinsky2009; Hetherington and Suhay Reference Hetherington and Suhay2011) by bringing in support for undemocratic politicians as an important outcome. Prior studies have examined different explanations such as terror threats and different outcomes such as the importance of leadership that does not necessarily violate democratic principles (for example, Merolla and Zechmeister Reference Merolla and Zechmeister2009).
Vice versa, we contribute to a recent stream of research on why citizens support undemocratic politicians – which hitherto has been more focused on domestic contexts (for example, Braley et al Reference Braley, Lenz, Adjodah, Rahnama and Pentland2023; Graham and Svolik Reference Graham and Svolik2020) – by bringing in interstate conflict as a possible explanation. Finally, we contribute to research on citizens’ preferences for candidate traits and personality (for example, Bittner Reference Bittner2011; Hayes Reference Hayes2005) by illustrating how far-reaching the consequences of citizens’ preferences for dominant leaders in times of conflict are: such preferences may lead to an increase in the appeal of undemocratic candidates.
How Dominance Perceptions Make Undemocratic Candidates Attractive During Conflict
We theorize that undemocratic behavior – which refers to violations of core democratic principles such as free and fair elections, civil liberties, checks and balances, and the rule of law (Krishnarajan Reference Krishnarajan2023) – affects dominance impressions; undemocratic leaders should, all else equal, be perceived as more dominant than democratically compliant leaders. This is the first step of our theoretical argument, which remains undocumented in existing studies.
Conceptually, leader dominance relates to an individual’s attempt to attain high social rank through the ‘induction of fear through intimidation and coercion’ (Cheng et al Reference Cheng, Tracy, Foulsham, Kingstone and Henrich2013). Studies have investigated the different ways through which social impressions of dominance are formed. These cover both seemingly shallow and superficial types of information such as first impressions from visual (Oosterhof and Todorov Reference Oosterhof and Todorov2008; Laustsen and Petersen Reference Laustsen and Petersen2016; Cogsdill et al Reference Cogsdill, Todorov, Spelke and Banaji2014) and vocal (Puts et al Reference Puts, Gaulin and Verdolini2006) appearances, but also richer types of information such as personality descriptions and behavioral records (Chen et al Reference Chen, Zhang, Laustsen and Cheng2021). Specifically, we argue that because undemocratic behavior often involves disrespecting authorities – such as courts – for own gain or harassing or obstructing political opponents and journalists (Graham and Svolik Reference Graham and Svolik2020), citizens have good reasons to infer dominance from such undemocratic acts.
The second part of our theoretical argument is that undemocratic candidates via this effect on dominance become more attractive to citizens in times of interstate conflict. Prior studies have, indeed, documented that dominant leaders are seen as more competent during conflict (Laustsen and Petersen Reference Laustsen and Petersen2020; Merolla and Zechmeister Reference Merolla and Zechmeister2009). This is because people up-regulate the value attached to dominance in contexts requiring coercion to ensure group-based co-ordination (Laustsen Reference Laustsen2021; Chen et al Reference Chen, Zhang, Laustsen and Cheng2021). Democratically compliant leaders may thus be seen as ‘too soft’ and incapable of ensuring the needed coordination among all citizens to meet the threat from the enemy out-group and, thus, safely steer the nation through the conflict. In times of peace, the psychology of followership reverses such that people fear being exploited by dominant leaders due to their selfish nature (Cheng et al Reference Cheng, Tracy, Foulsham, Kingstone and Henrich2013; Laustsen Reference Laustsen2021). In combination, this makes dominant leaders relatively more attractive to citizens in times of conflict, whereas non-dominant leaders are more attractive to citizens in times of peace.
However, our argument certainly sheds light on a high possible cost of achieving dominance in political leaders. Dominant leaders need not be undemocratic; they can be tough-minded without challenging the democratic system. Citizens may therefore achieve dominance in political leaders without having to vote for undemocratic ones. Additionally, undemocratic behavior might come with other consequences such as the risk of democratic backsliding. Specifically, a short-term gain in capability to deal with conflict resulting from dominance may come with a long-term loss in the ability to select leaders in the future due to politicians undermining democracy (Luo and Przeworski Reference Luo and Przeworski2023; Frederiksen Reference Frederiksen2022). Thus, when citizens have the option to vote for a candidate behaving undemocratically during conflict, they face trade-offs that they do not necessarily face when merely having the option to vote for dominant candidates. Nevertheless, we expect undemocratic behavior to increase impressions of dominance which, in turn, leads citizens to find undemocratic candidates more appealing during conflict.
Study 1
Research Design and Data
We fielded two pre-registered conjoint experiments (N = 18,395 candidate observations) in 2020–21 on Amazon Mechanical Turk to test whether undemocratic behaviors affect dominance impressions and thereby the first part of our theoretical argument.Footnote 1 The conjoint set-up enables us to examine the undemocratic behavior–dominance relationship in a rich information environment, where we can probe the robustness of the relationship across several other factors. Specifically, we randomly assigned undemocratic and democratic behaviors – along with background attributes, policy positions, and party – to potential future candidates in gubernatorial elections and measured how the respondents perceived these candidates in terms of dominance.
The respondents faced ten different pairs of candidates. The following attributes were randomly assigned: age, gender, profession, party, policy positions, and democratic/undemocratic behavior. The undemocratic (and democratic) behaviors were assigned in relation to the democratic cornerstones of free and fair elections, civil liberties, and the rule of law. To enhance the external validity of the candidate profiles, distributions of age, gender, and profession were assigned to reflect real-world characteristics of current American governors (Cuesta et al Reference Cuesta, Egami and Imai2022). All candidate attributes are illustrated in Appendix A.
Study 1 contains two separate conjoint experiments (rather than just one) because we decided to examine whether the relationship between undemocratic behavior and dominance is robust to deliberately separating out any sign of aggression from the behaviors. The second conjoint experiment thus distinguishes between aggressive and non-aggressive undemocratic behaviors.Footnote 2 These aggressive and non-aggressive undemocratic behaviors were selected based on a pilot study – fielded in between the first and second experiment – testing perceived aggression and undemocraticness of twenty-seven different examples of undemocratic behaviors. An example of one of our non-aggressive undemocratic behaviors is emphasizing lack of trust in election officials, whereas legitimizing journalist harassment constitutes an aggressive undemocratic behavior. Appendix C shows that the results are robust to excluding aggression from the behaviors and thereby also that the undemocratic behavior–dominance link is not exclusively present for specific behaviors.
Regarding the outcome of the conjoint experiments, dominance impressions, we asked respondents to report on a seven-point scale ranging from ‘Strongly disagree’ (1) to ‘Strongly agree’ (7) how much they agreed that ‘Candidate 1 (2) is aggressive, threatening, and intimidating’. These adjectives capture the latent concept of dominance – which may not be captured by a single word – as referring ‘to the induction of fear through intimidation and coercion’ (Cheng et al Reference Cheng, Tracy, Foulsham, Kingstone and Henrich2013). The outcome also included a ‘Don’t know’ option, which we coded as missing, thereby not counting towards the total of 18,395 candidate observations. However, undemocratic behavior is not correlated with such attrition (p = 0.85). To facilitate reporting effects in percentage points, the outcome was rescaled to 0–1. We specifically estimate dominance impressions for undemocratic and democratic candidates (that is, marginal means) using linear regression with respondent-clustered standard errors, which is standard practice for conjoint analysis (Hainmueller et al Reference Hainmueller, Hopkins and Yamamoto2014; Adserà et al Reference Adserà, Arenas and Boix2023).
Undemocratic Behavior Increases Perceived Dominance
In support of our argument, Figure 1 shows that undemocratic candidates are perceived as 20 percentage points more dominant than democratically compliant candidates. To address possible concerns with the conjoint experiments (see also Clayton et al Reference Clayton, Horiuchi, Kaufman, King and Komisarchik2025; Bansak et al Reference Bansak, Hainmueller, Hopkins, Yamamoto, Druckman and Green2021; Jenke et al Reference Jenke, Bansak, Hainmueller and Hangartner2021), we show in Appendix B that this result is robust across the ten conjoint tasks, the partisan affiliations of the respondents, and the policy positions of the candidates.
Effect of undemocratic candidate behavior on dominance impressions illustrated by marginal means. N = 18,395 candidate observations.
To rule out that undemocratic behavior increases perceptions of any trait, some participants were with equal probability randomly assigned to state their impressions on warmth or competence instead of dominance (Fiske et al Reference Fiske, Cuddy and Glick2007). Here, we substituted the adjectives mentioned above with ‘warm, reliable, and friendly’ for warmth and ‘competent, knowledgeable, and skillful’ for competence. A final group of participants were assigned to a supplementary vote intentions outcome asking ‘How likely is it that you would vote for candidate 1 (2)?’.
We find that undemocratic behavior reduces perceived warmth, competence, and vote intentions by 11–12 percentage points each. Additionally, undemocratic candidates are less likely to attract votes when conflict is not manipulated (but see Study 2). Meanwhile, effects on dominance (20 percentage points) are not only distinctively positive but also stronger than those on warmth, competence, and vote intentions, highlighting the strength of the undemocratic behavior–dominance link.
Study 2
Research Design and Data
We fielded a factorial vignette experiment (N = 2,611 respondents meeting census-based quotas on gender, age, region, and education) with YouGov in December 2023 to test whether the appeal of undemocratic candidates is amplified during threats of interstate conflict. Aligning with our two-step theoretical argument, the experimental set-up allows respondents to first store candidate impressions about dominance and then evaluate candidate capabilities to deal with subsequently presented conflict. We deliberately avoided a design presenting candidate behaviors simultaneously with – or after – the introduction of conflict to make sure that the findings indeed reflect inferred leader character in advance of conflict.
Specifically, respondents first read about a potential presidential candidate (Thomas Johnson) who represents the respondent’s party and behaves either democratically or undemocratically. We decided to not randomize party, because the degree to which voters are willing to trade off democracy for partisan loyalty is well-established by prior research (Graham and Svolik Reference Graham and Svolik2020; Carey et al Reference Carey, Clayton, Helmke, Nyhan, Sanders and Stokes2022; Frederiksen Reference Frederiksen2024), enabling us to draw cumulative conclusions without potentially compromising statistical power by further increasing the number of experimental factors. The undemocratic behaviors include issuing executive orders to circumvent Congress and restricting media sources that are critical of government (see also Iyengar et al Reference Iyengar, Lelkes and Westwood2023). Thomas Johnson is either assigned both undemocratic behaviors or two democratic counterparts, which comprise an active, stated willingness to work with Congress and a statement that critical media should be respected.
Following the presentation of Thomas Johnson, we measured dominance impressions, assigned various survey items unrelated to the present project to minimize risks of demand effects, and also collected dominance impressions as a recall exercise. We do this to ensure that undemocratic behavior also affects dominance impressions in our vignette set-up. In the first measure of dominance, respondents are asked to state their agreement with the following statement: ‘Thomas Johnson is aggressive, threatening, and intimidating’. Responses are collected on a seven-point scale from ‘1’ reflecting ‘Strongly disagree’ to ‘7’ reflecting ‘Strongly agree’. The recall measure uses the same scale but asks about agreement with the statements (i) ‘Thomas Johnson is assertive’, (ii) ‘Thomas Johnson is dominant’, and (iii) ‘Thomas Johnson is tough-minded’, yielding a composite measure of dominance impressions. Indeed, undemocratic behavior increases dominance impressions by 28 percentage points (95 per cent CI: 0.26; 0.30) in the immediate task and 12 percentage points (95 per cent CI: 0.10; 0.13) in the recall task. These findings corroborate our finding from Study 1 that undemocratic behavior causes dominance impressions and show that the relationship is robust to using a vignette experiment instead of a conjoint experiment as well as different adjectives to capture dominance.
After the recall measure of dominance, respondents were primed with an escalation of conflicts with Russia (in relation to the war in Ukraine) or China (in relation to the conflict with Taiwan) – or peace. Emphasizing peace, and particularly emphasizing peace without accidentally sparking considerations about real-world conflicts, is difficult and easily becomes artificial. We therefore kept the peace condition as short as possible. Following the context information, respondents were asked (i) how capable Thomas Johnson as potential president is to handle the described situation and (ii) how likely they would be to vote for Thomas Johnson as president. Both outcomes were originally measured on a scale from 0 to 10, then rescaled to 0–1 in the main analyses. Appendix D shows all treatment conditions for Study 2.
To test whether we successfully made respondents reflect upon the context they were assigned (Kane and Barabas Reference Kane and Barabas2019), we employed a manipulation check asking them what they thought about while answering the outcome questions, with the four options being (i) the threat from China, (ii) the threat from Russia, (iii) peace, or (iv) none of the other options. Using linear probability models on three dichotomized outcomes (one for each condition), we find that the context treatments were successful: they each significantly increased respondents’ propensity to think about the ‘correct’ context by 27–39 percentage points.
As pre-registered, we pool the China and Russia conditionsFootnote 3 and rely on linear regression including an interaction term between threats of interstate conflict and undemocratic behavior. If this interaction term is significantly positive – that is, if an undemocratic presidential candidate becomes more appealing relative to a democratic candidate during conflict compared with peace – our argument is supported.
Interstate Conflict Increases the Appeal of Undemocratic Candidates
Figure 2(A) illustrates the support for an undemocratic versus democratic presidential candidate (‘Effect of undemocratic behavior’) during threats of interstate conflict and peace for each of the two outcomes, respectively, and also reports the interaction terms (‘Difference’). Supporting our argument, we find that an undemocratic presidential candidate is strongly disfavored during peace, but that the relative appeal of democraticness is significantly reduced when threats of interstate conflict appear. Respondents still prefer a democratic candidate during conflict, but recall that this is when party is held constant. Should voters, unlike in our study, face trade-offs between democracy and partisan loyalty during conflict, we interpret our findings as suggestive evidence – especially in combination with prior studies (for example, Graham and Svolik Reference Graham and Svolik2020) – that they would be significantly more likely to toe the party line and, thus, prefer the undemocratic option.
Panel A: interaction between undemocratic behavior and interstate conflict illustrated by marginal effects of undemocratic behavior on rated capability and vote intentions within each context condition. Panel B: interaction between dominance recall and interstate conflict illustrated by correlations between dominance and rated capability and vote intentions within each context condition. N = 2,611 respondents.
Figure 2(B) shows a correlation between the recall measure of perceived dominance and the two outcomes during peace and conflict, respectively. This test is non-causal, exploratory, and only complementary to our main test, but we include it to shed further light on the mechanism yielding increased support for undemocratic candidates during conflict. As Figure 2(B) shows, perceived dominance is not correlated with capability and is negatively correlated with vote intentions during peace. But during interstate conflict, perceived dominance is positively associated with both outcomes. This suggests that American voters become more tolerant of undemocratic behavior but outright value dominance during interstate conflict. Undemocratic behavior is something citizens might tolerate in exchange for a (perceived) increase in capability – resulting from dominance – to deal with conflict. In combination with the robust dominance–undemocratic behavior link demonstrated in Study 1, this exploratory finding provides suggestive evidence that assigning any type of undemocratic behavior in Study 2 would have yielded similar results (because dominance boosts support in times of conflict). In Appendix F, we complement this exploratory mechanism test with Structural Equation Modeling (SEM) analyses, which support our interpretation.
Partisan heterogeneity in interaction between undemocratic behavior and interstate conflict.
We now turn to heterogeneity across self-reported party affiliation. We pre-registered that we would conduct this additional test but did not register any specific predictions related to it. Nevertheless, we find evidence – illustrated in Figure 3 – that the increase in appeal of an undemocratic presidential candidate during interstate conflict is strongest among Republicans. Republicans’ preferences for a democratically compliant candidate are completely eroded during threats of interstate conflict. Independents are also significantly swayed by such threats and find democratically compliant and undemocratic candidates equally capable of dealing with conflict, but do retain voting preferences for a democratic candidate even in times of conflict. Democrats are less – yet still somewhat – affected by conflict.
Regarding statistical significance, the effects of undemocratic behavior are indistinguishable between partisan groups on both outcomes in the peace condition, whereas the partisan differences in the interstate conflict condition are all statistically significant on the 0.05 level except for the comparison between Independents and Republicans on the capability outcome (p = 0.14). This supports our interpretations of partisan heterogeneity in response to interstate conflict, and further suggests that Republican voters are not more open to undemocratic behavior per se – rather, their preferences are more sensitive to the shift in context imposed by conflict.
Conclusion and Discussion
Threats of interstate conflict may push Americans to vote for politicians undermining democracy. This key finding illustrates that threats of war and democratic backsliding can interact with dire consequences. Specifically, we have argued and shown that undemocratic candidate behavior affects perceived dominance, which citizens find appealing during interstate conflict. A fruitful venue for future research would, therefore, be to examine whether dominant but democratically compliant candidates may constitute viable alternatives to undemocratic candidates in conflict situations.
Contributing to research on international relations and public opinion (for example, Tomz and Weeks Reference Tomz and Weeks2013), we highlight how external threats to the United States may exacerbate internal threats to the democratic system. Existing related studies have particularly focused on the consequences of terror threats and the war on terror (for example, Merolla and Zechmeister Reference Merolla and Zechmeister2009; Berinsky Reference Berinsky2009), whereas we have shifted focus to conflicts with other great powers. Future research may therefore also benefit from distinguishing the effects of these rather different types of conflict.
Contributing to research on citizens’ preferences for candidate traits (for example, Bittner Reference Bittner2011; Laustsen Reference Laustsen2021; Hayes Reference Hayes2005), our study sheds light on the boundaries of how far citizens are willing to go to elect leaders who embody dominance in times of conflict. Are they willing to sacrifice long-term prospects of democracy for an allegedly short-term gain in capability to deal with conflict? Our findings suggest that Republicans are willing to go further than Democrats, as they are pushed to a point by threats of interstate conflict where they no longer prefer democratically compliant Republican presidential candidates over undemocratic ones.
Building upon insights from prior studies (for example, Graham and Svolik Reference Graham and Svolik2020), this implies that Republican voters would be even more likely to prioritize partisan loyalty over democracy during times of conflict than under normal conditions if they – unlike in our main study – face a choice between an undemocratic Republican candidate and a democratically compliant Democratic candidate. However, we do acknowledge that this is an implication of our study in combination with prior studies. It is indeed a limitation of our study that the findings do not reflect two-party contests. At a minimum, our findings provide insights on how undemocratic candidates would fare in important one-party contests such as presidential primaries during threats of interstate conflict versus peace.
It is also important to highlight that conflicts that are more or less intense than those presented in our treatment material may affect public opinion differently. On the one hand, conflicts in regions such as the Middle East may seem less threatening to citizens and thus produce smaller effects. On the other hand, more intense interstate conflicts – for instance, directly on American soil – may produce stronger effects. We thus conceive of our study as the first rather than the last test of how interstate conflict affects support for undemocratic candidates.
Finally, interstate conflict may affect support for anti-democratic practices through additional mechanisms than the one suggested here. Our argument – and experimental set-up – is about leader character inferred in advance of conflict. Another possible mechanism is that conflict provides a venue for undemocratic solutions (see also Merolla and Zechmeister Reference Merolla and Zechmeister2009), or even affects the degree to which particular candidates are perceived as dominant. Future research may benefit from expanding on such possible mechanisms linking threats of interstate conflict to tolerance of undemocratic behavior.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0007123425101014.
Data availability statement
Replication data for this paper can be found in Harvard Dataverse at https://doi.org/10.7910/DVN/MQBORA
Acknowledgements
We thank the Political Behavior and Institutions Section at Aarhus University, participants at our presentations at APSA, EPSA, and HBES 2024, the Politics and Social Change Lab at Stanford University, Robb Willer, Joey T. Cheng, James N. Druckman, and three anonymous reviewers for helpful comments.
Funding
This work was supported by the Danish National Research Foundation (DNRF144).
Competing interests
None.
Ethical standards
All participants provided written consent and were compensated for their participation, and an ethical approval from the Research Ethics Committee at Aarhus BSS was obtained.
Open access
