Introduction
Societies across the Globe face the threat from terrorist attacks. Although the 2001 9/11 attack on the United States may hold historical prominence for many, the more recent attacks in Bali (2002), Paris (2015), Manchester (2017), Sri Lanka (2019), and Magdeburg (2024) – to name just a few – all testify to the constant threat from terrorism since the 9/11 attacks and show that terrorist threat is not confined to North America. Consequently, political science research has explored how terrorism and terrorist threat affects attitudes about minority groups and immigration (e.g., Nussio, Bove and Steele Reference Nussio, Bove and Steele2019; Bulbulia, Usman Afzali, Yogeeswaran et al. Reference Bulbulia, Usman Afzali, Yogeeswaran and Sibley2023; Lueders, Wollast, Nugier et al. Reference Lueders, Wollast, Nugier and Guimon2024; but see Turkoglu and Chadefaux Reference Turkoglu and Chadefaux2023), cause ‘conservative shifts’ (e.g., Bonanno and Jost Reference Bonanno and Jost2006; Nail and McGregor Reference Nail and McGregor2009; Huddy and Feldman Reference Huddy and Feldman2011; Schüller Reference Schüller2015), change voting behavior (e.g., Getmansky and Zeitzoff Reference Getmansky and Zeitzoff2014; Vasilopoulos, Marcus, Valentino et al. Reference Vasilopoulos, Marcus, Valentino and Foucault2019; Geys and Hernæs, Reference Geys and Hernæs2021; but see Vlandas and Halikiopoulou Reference Vlandas and Halikiopoulou2025), and even affect public employees’ professional decisions (Emeriau Reference Emeriau2024). Importantly, recent meta-analytical and experimental work shows that terrorism still exerts significant effects on political behavior and attitudes despite the 9/11 attacks taking place more than two decades ago (Germann, Godefroidt and Mendez Reference Germann, Godefroidt and Mendez2026).
This paper tests whether terrorism possibly also affects political representation. Specifically, we test whether the terrorist threat may cause citizens to increase their relative demand for men over women candidates. We generate this prediction from three well-established interdisciplinary findings about gender stereotypes and citizens’ candidate evaluations. First, numerous social psychology studies on person perception and stereotypes show that men, compared to women, are seen and stereotyped as higher in agentic traits such as assertiveness and dominance (e.g., Ellemers Reference Ellemers2018; Eagly, Nater, Miller, et al. Reference Eagly, Nater, Miller, Kaufmann and Sczesny2020). Similar gender differences also obtain for perceptions of political candidates (e.g., Huddy and Terkildsen Reference Huddy and Terkildsen1993; Sanbonmatsu Reference Sanbonmatsu2002; Dolan Reference Dolan2010). Second, security crises and conflict situations such as wars (e.g., Gadarian Reference Gadarian2010; Laustsen, Sheng, Ahmad et al. Reference Laustsen, Sheng, Ahmad, Al-Shawaf, Banai, Banai, Barlev, Bastardoz, Bor, Cheng and Chmielińska2025; Hargrave, Smith, Valgarðsson et al. Reference Hargrave, Smith, Valgarðsson, Devine, Bunting and Leicht2026) and terrorist attacks (Merolla and Zechmeister Reference Merolla and Zechmeister2009) are shown to increase citizens’ preferences for masculine, assertive, dominant, and ‘Hawkish’ political leaders. That is, citizens upregulate their preferences for the traits that men are stereotypically advantaged on when threats from wars and terrorism rise. Finally, turning from trait stereotypes to policy competence stereotypes, previous work documents how voters typically perceive men as more capable than women candidates to handle national security, military, and defense issues (e.g., Huddy and Terkildsen Reference Huddy and Terkildsen1993; Lawless Reference Lawless2004; Sanbonmatsu and Dolan Reference Sanbonmatsu and Dolan2009; Holman, Merolla and Zechmeister Reference Holman, Merolla and Zechmeister2016; although see Gothreau and Laustsen (Reference Gothreau and Laustsen2025) for indications of changing stereotype content). In total, this leads us to formulate the Terrorist Threat Hypothesis that terrorist threats cause citizens to upregulate their relative preference for men compared to women political candidates. This hypothesis was pre-registered with AsPredicted: https://aspredicted.org/g7yq-fptv.pdf.Footnote 1
Importantly, previous work already provides some support for the Terrorist Threat Hypothesis. Directly related to the 9/11 attacks, Lawless (Reference Lawless2004: 482) reports that 61% of American respondents find ‘men candidates and officeholders more likely to possess the traits or competency to deal with a military crisis’, while only 3% find women more likely to possess the same qualities. Likewise, Falk and Kenski (Reference Falk and Kenski2006) find that Americans who listed terrorism, homeland security, and US involvement in Iraq as the most important issues, also said that, a man would do a better job as President. Across a series of studies, Holman, Merolla and Zechmeister (2011, 2016, Reference Holman, Merolla and Zechmeister2017) add candidate partisanship and experience to the equation. Their results show that terrorist threat enhances the expression of gender-trait stereotypes (Holman, Merolla and Zechmeister Reference Holman, Merolla and Zechmeister2011), that in the absence of clear threats ‘masculine stereotypes’ may hurt Democratic candidates – particularly women (Holman, Merolla and Zechmeister Reference Holman, Merolla and Zechmeister2016), and that Republican partisanship or national security experience may counteract negative electoral effects of terrorist threat for women candidates (Holman, Merolla and Zechmeister 2011, Reference Holman, Merolla and Zechmeister2017). Finally, Kim and Kang (Reference Kim and Kang2022) broaden the analytical scope and leverage several waves of the World Values Survey (1995–2014) to show that ‘the more hostile a country’s security environment, the more individuals are likely to agree that men make better leaders than women do’ (Kim and Kang Reference Kim and Kang2022: 1). Altogether, these results suggest that conflicts and security threats move candidate preferences in the direction of men over women.
Despite the many important insights generated by existing studies, several theoretical, methodological, and empirical questions remain unaddressed. First, according to recent meta-analytical evidence and multi-country studies, citizens across most democracies generally hold preferences for rather than against women candidates (Schwarz and Coppock Reference Schwarz and Coppock2022; Gothreau and Laustsen Reference Gothreau and Laustsen2026). While the meta-analyzed and multi-country studies do not relate to conflicts and threats, they question whether contemporary citizens – in times characterized by preferences for women candidates – still find men candidates more appealing under conflict and threat as indicated by previous studies? Second, existing work suffers from different methodological shortcomings. Reliance on observational data (Lawless Reference Lawless2004; Falk and Kenski Reference Falk and Kenski2006; Kim and Kang Reference Kim and Kang2022) does not allow for a strong causal identification of the potential effect of a terrorist threat on candidate gender preferences. Results from smaller convenience samples (Holman, Merolla and Zechmeister Reference Holman, Merolla and Zechmeister2011) inevitably cause questionable generalizability, and one may consider the realism and ecological validity of using mock news stories to manipulate the terror context (Holman, Merolla and Zechmeister 2016, Reference Holman, Merolla and Zechmeister2017). The experiment reported in this manuscript seeks to simultaneously meet these challenges to internal, external, and ecological validity of previous research. Third, previous experimental work investigating how the threat from terrorism (and other related threats) affects citizens’ preferences for candidate gender relies exclusively on American respondents despite the threat from terrorism – as indicated by the terrorist attacks listed earlier in the introduction – not being confined to North America. This raises the question whether citizens from other countries and geographic regions also find men candidates relatively more appealing when facing a terrorist threat. To address these questions, we employ a conjoint candidate choice experiment among a sample of nationally representative Danes and randomize whether respondents read the Danish Security and Intelligence Service’s (Politiets Efterretningstjeneste, PET) assessment of the terrorist threat against Denmark before choosing their preferred candidates in the conjoint experiment.
Research design, data, and methods
Data collection
Our experiment was included as part of a larger omnibus survey fielded by the Danish survey agency Epinion between 25 October and 8 November 2024. Epinion cleaned the dataset in accordance with standard procedures before sharing it with the authors, which yielded a final sample of 1,034 respondents. The sample closely resembles the Danish population on age, gender, broader geographic regions, and educational level (see Supplementary Online Materials A.1).
Experimental design
We test the Terrorist Threat Hypothesis using a combination of a vignette experiment – manipulating terrorist threat – and a paired candidate choice conjoint experiment – manipulating candidate gender as one among five different candidate attributes.
In the paired conjoint experiment, respondents performed eight choices between two fictitious and randomly generated candidate profiles varying on five attributes (levels in parentheses): Gender (woman, man), age (young (29–35 years), middle-aged (41–47 years), older (52–60 years)), country of origin (Denmark, or Middle Eastern country (specifically displayed as Syria, Turkey, or Pakistan)), education (short, medium length, or longer education (reflecting six more specific educational categories familiar to Danish respondents)), and political experience (low (maximum 2 years), medium (4–5 years), high (minimum 8 years)). Candidate gender constitutes our primary interest with women candidates coded as the reference category. In total, respondents were exposed to sixteen different candidates with equal probability of exposure to a given combination of attribute levels across rounds and respondents. Included attributes match related conjoint experiments (e.g., Teele, Kalla and Rosenbluth Reference Teele, Kalla and Rosenbluth2018) and reflect an attempt to balance risks of masking effects by having too few attributes against risks of survey satisficing by having too many (Bansak, Hainmueller, Hopkins et al. Reference Bansak, Hainmueller, Hopkins, Yamamoto, Druckman and Green2021). To enhance realism, ecological, and external validity, we applied weighted distributions for the attributes ‘country of origin’ and ‘education’ matching displayed candidate profiles to the demographic composition of Danish political candidates (Hainmueller, Hopkins and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014; De la Cuesta, Brandon and Imai Reference la Cuesta, Brandon and Imai2022) (see Supplementary Online Materials A.2 for attribute-level details). Following recent studies on voters’ candidate gender preferences, we hold candidates’ partisanship constant (e.g., Teele, Kalla and Rosenbluth Reference Teele, Kalla and Rosenbluth2018; Gothreau and Laustsen Reference Gothreau and Laustsen2025). Specifically, for each of the eight choices, respondents were asked: ‘Which of the two candidates would you prefer as the leading candidate for your party in the upcoming parliamentary election?’ The decision to hold partisanship constant reflects our specific interest in testing whether terrorist threat changes citizens’ relative preference for men and women candidates. However, this design decision obviously confines our results to intra-party candidate preferences in national elections. We return to this point in the discussion. Finally, because respondents’ choices express favorability, we employ them as a proxy for candidate preferences (Hainmueller, Hopkins and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014), coding non-chosen and chosen candidates ‘0’ and ‘1’, respectively.
Importantly, the described conjoint experiment was embedded in a vignette experiment manipulating terrorist threat against Denmark. This experiment consisted of a Terrorist Threat condition (coded ‘1’) and a Control condition (coded ‘0’). In the Control condition, respondents were sent directly to the conjoint experiment without further instructions than what was explained above. In contrast, in the Terrorist Threat condition, respondents were exposed to the following official assessment of the risk of terrorism against Denmark (the italicized part of the below text) from the Danish Security and Intelligence Service (Politiets Efterretningstjeneste, PET):
Over the last decade Europe and Denmark have experienced several terrorist attacks and attempts. Along with recent tensions in domestic terrorist groups this has led to the following severe assessment from PET: ‘PET assesses that the terrorist threat to Denmark is significant. This means that there is a known threat to Denmark and Danish interests abroad and that there is intent, planning and capability among individuals or groups to commit a terrorist attack against Denmark.’ (Source: PET, 2024)
We drafted and included the first two sentences to contextualize PET’s quoted assessment, while the quote from the official Danish authority enhances realism and ecological validity of the Terrorist Threat condition through reliance on factually correct and (at the time of data collection) up-to-date information. We included a hyperlink to PET’s official website to further strengthen authenticity and credibility. To ensure that PET’s terror assessment remained salient to respondents assigned to the Terrorist Threat condition throughout the experiment, the following sentence was added to instructions before each conjoint task: ‘Based on PET’s assessed terrorist threat to Denmark as significant, please choose between the following two candidates’ (see Supplementary Online Materials A.3 for original Danish language study materials.) Finally, at the end of the survey, respondents assigned to the Terrorist Threat condition were thoroughly debriefed and explained to not be overly concerned despite the content of PET’s assessment (directing respondents to PET’s website for further information).
In sum, the employed research design allows for a realistic and ecologically valid manipulation of terrorist threat and for estimating how such a threat may affect Danes’ candidate gender preferences. Moreover, the rich information environment of the paired conjoint experiment arguably minimizes social desirability and demand bias risks (Hainmueller, Hopkins and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014). Consequently, the experimental design is well-equipped to validly test the Terrorist Threat Hypothesis.
Respondent characteristics
The omnibus survey in which our experiment was embedded also permits measuring various respondent characteristics of which two are particularly relevant. First, respondents’ binary gender was measured coding women ‘0’ and men ‘1’. Second, respondents’ broader political orientations were captured based on intended voting behavior if an election was held tomorrow. Following traditional categorizations of Danish parties, respondents reporting preferences for left-wing and right-wing parties were coded ‘0’ and ‘1’, respectively (respondents failing to report a preferred party are excluded from analyses involving political orientations).Footnote 2
The case of Denmark
Denmark constitutes an interesting country for testing the Terrorist Threat Hypothesis using the described research design for at least three reasons. First, over the last 25 years, Denmark has only experienced one terrorist attack, which took place in Copenhagen on 14 and 15 February 2015, causing two civilian deaths and six wounded police officers (dr.dk). Thus, terrorist acts are (luckily) very rare in the Danish society although Danish citizens may occasionally consider such risks. Given this point of departure, we see ample potential for PET’s quite severe terror risk assessment having an effect on Danes’ candidate preferences.
Second, comparatively speaking, women are well represented in Danish politics. At the time of our data collection, women made up 44% of the members of the Danish parliament, Folketinget, with this proportion rising to 48% with the 2026 election (Folketinget 2026). Interestingly, in 2026 women only constituted 39% of the nominated candidates suggesting that women candidates are quite popular among the Danish electorate (Holck Reference Holck2026). Because of this comparatively high baseline preference for women candidates, we see strong potential for terrorist threat increasing citizens’ preferences for men candidates insofar as the theoretical rationale for the Terrorist Threat Hypothesis is correct.
Third, an important feature of the Danish electoral system is that multiple candidates running for multiple parties (twelve in the 2026 election) compete for several seats in each of the ten different electoral districts (‘storkredse’). Thus, the allocation of seats in each district – and, thus, electoral competition between candidates – can crudely and somewhat simplifiedly be described as following two steps: First, seats are allocated between parties based on received votes. Second, a given party assigns seats among its candidates based on candidates’ numbers of personal votes. When voting, citizens can choose to either cast a party vote or a personal vote. Party votes will only count toward the allocation of seats between parties, whereas a personal vote both counts when allocating seats between parties (based on the chosen candidate’s party) and when allocating seats between candidates nominated and running for the same party.Footnote 3 Approximately 50% of Danish voters cast personal votes in national elections with somewhat higher percentages in elections for municipal and regional councils and the European Parliament (Jørgensen Reference Jørgensen2025). Interestingly, some theories about Danish voters’ decision-making strategies also suggest a two-step structure, such that voters first find their favored party and subsequently (if casting a personal vote) then choose their favored candidate among this party’s nominees (Hansen and Stubager Reference Hansen, Stubager, Hansen and Stubager2017). Importantly, this process is closely resembled by the setting of our conjoint candidate choice experiment in which respondents choose between candidates from their favored party, which ensures high ecological validity and realism of our test.
Analytical strategy
To estimate the causal effect of candidate gender on respondents’ candidate preferences, we follow standard practices for conjoint experiments relying on the average marginal component effect (AMCE). Because women candidates are coded as the reference category, the AMCE of candidate gender represents the marginal effect of being a man (rather than a woman) candidate on being chosen over the joint distribution of all other assigned candidate attributes (Hainmueller, Hopkins and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014 and should not be seen as reflecting respondents’ majority preferences (Abramson, Kocak and Magazinnik Reference Abramson, Kocak and Magazinnik2022)). However, our primary interest is whether terrorist threat changes citizens’ relative preferences for women and men candidates. Hence, our key test of the Terrorist Threat Hypothesis is captured by the average marginal component interaction effect (AMCIE) between the assigned vignette (Terrorist Threat vs. Control conditions) and candidate gender as it tests how the assigned vignette possibly moderates the effect of candidate gender (Hainmueller, Hopkins and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014). Following Hainmueller, Hopkins and Yamamoto (Reference Hainmueller, Hopkins and Yamamoto2014), standard errors are clustered at the respondent-level accounting for the repeated nature of the data from the conjoint experiment. Finally, an a priori power analysis (using the R-package ‘cjpowR’) based on an expected sample size of 1,000 respondents (or 16,000 candidate profiles (eight rounds of two candidates for each respondent)) shows that with 80% power we can detect an AMCIE of d = 0.044, which according to Schuessler and Freitag (Reference Schuessler and Freitag2020) corresponds to a small interaction effect for conjoint experiments.
Results
Below, we report key AMCEs and AMCIEs directly in the text and illustrate findings using marginal effect plots of candidate gender. Supplementary Online Materials A.4 and A.5 display the underlying full regression models and a supplementary marginal means plot for key findings, respectively.
We start our analysis focusing on the Control condition. Verifying that our study produces meaningful results, we find positive AMCEs of increasing levels of political experience (ps < 0.001) and a negative AMCE of immigrant background (p < 0.001; paralleling for instance Portmann and Stojanović Reference Portmann and Stojanović2019). Directly related to candidate gender, we also replicate recent studies finding that respondents prefer women over men candidates (e.g., Schwartz and Coppock Reference Schwarz and Coppock2022; Gothreau and Laustsen Reference Gothreau and Laustsen2026). Specifically, we estimate an AMCE of gender of −0.086 (95% CI [−0.110, −0.062], p < 0.001), reflecting that marginal means of being chosen for men and women candidates are 45.7% and 54.3%, respectively. These results are illustrated in Figure 1’s left-hand panel A.
Average marginal component effects (AMCEs) of candidate attributes across Control (left-hand panel) and Terrorist Threat (middle panel) conditions and the interactions between candidate attributes and experimental condition (right-hand panel). Point estimates and 95% confidence intervals. Supplementary Online Materials A.4 display the full regression models.
Moving on to the Terrorist Threat condition, we estimate an AMCE of candidate gender of −0.053 (95% CI [−0.075, −0.030], p < 0.001). Thus, when respondents read PET’s terror assessment before performing the conjoint tasks, marginal means for choosing men and women candidates are 47.4% and 52.6%, respectively. Along with estimated AMCEs of the other candidate attributes, this result is illustrated in the middle panel B of Figure 1.
Our results reveal that men candidates are less likely to be chosen than their women counterparts, but this tendency is strongest in the Control condition. Consequently, we turn to the key test of the Terrorist Threat Hypothesis – if terrorist threat upregulates respondents’ relative preference for men compared to women candidates – estimating the interaction between assigned vignette (Terrorist Threat vs. Control) and candidate gender. In support of this hypothesis, we find a positive and significant AMCIE of 0.034 (95% CI [0.001, 0.066], p = 0.042),Footnote 4 which is displayed in the top rows of Figure 1’s right-hand panel C. This result reflects that although women candidates are preferred in both experimental conditions, men candidates’ marginal mean increases by 1.7 percentage points (from 45.7% to 47.4%) in the Terrorist Threat relative to the Control condition, with a symmetric 1.7 percentage point decrease for women candidates. In other words, terrorist threat makes men candidates less unappealing to respondents.
Interestingly, the effect of terrorist threat is not confined to preferences for candidate gender. Preferences for older aged, and longer relative to shorter educated candidates increase in the Terrorist Threat compared to the Control condition. Moreover, a non-significant increase in preferences for political experience and a non-significant decrease in preferences for Middle Eastern candidates obtain. AMCIEs for all candidate attributes are displayed in Figure 1’s right-hand panel C. The effect of terrorist threat on increased preferences for men (relative to women) candidates is smaller than the simultaneous effects on preferences for older (relative to young) and longer (relative to shorter) educated candidates but matches the simultaneous – but non-significant – changes in preferences for middle-aged (relative to young) and highly (relative to low) experienced candidates. Importantly, our pre-registration only contained hypotheses for candidate gender. We therefore stress that the reported moderations by terrorist threat on effects of other candidate attributes should only be interpreted as exploratory. Yet, we return to these interesting patterns in the conclusion.
Finally, we explore two potential moderators of respondents’ gendered candidate preferences and how they potentially interact with the effect of terrorist threat.Footnote 5 First, research supports the existence of a gender affinity effect with voters exhibiting same-gender candidate preferences (Sanbonmatsu Reference Sanbonmatsu2002; Badas and Stauffer Reference Badas and Stauffer2019; Schwarz and Coppock Reference Schwarz and Coppock2022). Because women (also politically) constitute the disadvantaged gender group, the gender affinity effect is expected to be most pronounced among women voters (Sanbonmatsu Reference Sanbonmatsu2002; Badas and Stauffer Reference Badas and Stauffer2019). Second, existing studies also show that left-wing and liberal voters are generally more supportive of and positively oriented toward women candidates (Dahl and Nyrup Reference Dahl and Nyrup2021; Schwarz and Coppock Reference Schwarz and Coppock2022; Gothreau and Laustsen Reference Gothreau and Laustsen2025), which possibly reflects these voters’ preferences for leveling (gender) hierarchies and inequalities and/or changing rather than preserving existing (gender) norms and conventions (Federico Reference Federico and Adam2020; Osborne, Costello, Duckitt et al. Reference Osborne, Costello, Duckitt and Sibley2023). Regarding respondent gender and the gender affinity effect, our results show that terrorist threat exerts substantially identical and statistically indistinguishable effects on candidate preferences regardless of respondent gender. That is, men and women respondents to the same extent come to find men candidates relatively more appealing in the Terrorist Threat compared to the Control condition (see Supplementary Online Materials A.6.6). Next, testing if respondents’ political orientation moderates the reported results, we observe two interesting patterns. First, among right-wing respondents, non-significant AMCEs of candidate gender obtain across experimental conditions (ps > 0.05). In other words, right-wing respondents do not display gendered candidate preferences in neither the Control nor in the Terrorist Threat conditions (see right-hand panel in Figure 2). Second, left-wing respondents strongly prefer women candidates in the Control condition (b = −0.153, 95% CI [−0.192, −0.113], p < 0.001), and while this preference also obtains in the Terrorist Threat condition (b = −0.087, 95% CI [−0.124, −0.050], p < 0.001), it is substantially and significantly reduced compared to the Control condition (testing the interaction between experimental condition and candidate gender among left-wing respondents: b = 0.066, 95% CI [0.012, 0.120], p = 0.016). That is, left-wing respondents seem to drive the overall result that men candidates appear less unappealing when terrorist threat is made salient (see Supplementary Online Materials A.6.7). Figure 2 displays AMCEs of candidate gender by experimental condition among left-wing and right-wing respondents, respectively.
AMCE of candidate gender (men relative to women candidates) for left-wing and right-wing respondents and across assigned experimental condition (Control vs. Terrorist Threat condition). Point estimates and 95% confidence intervals.
Discussion
Previous American studies – based widely on observational data – show that voters find men (compared to women) candidates relatively more appealing when the social context is characterized by threat and conflict such as war or terrorism (eg, Lawless Reference Lawless2004; Holman, Merolla and Zechmeister Reference Holman, Merolla and Zechmeister2016). Combining vignette and conjoint experimental elements, we show that this pattern replicates more than two decades after most of the existing studies were conducted (in the aftermath of the 9/11 attacks). Importantly, despite the fact that contemporary citizens across democracies are found to express a preference for women candidates (eg, Schwartz and Coppock Reference Schwarz and Coppock2022), we find that the threat from terrorism moves Danish citizens in the direction of men candidates: Absent a terrorist threat, women candidates are advantaged by 8.6 percentage points, but this drops to 5.3 percentage points – almost halving women candidates’ favorability – when respondents read the official Danish security and intelligence authority’s (PET) assessment of risks of terrorism against Denmark. Below, we first discuss the effect size of our key findings and the potential for terrorist threat to exert real electoral consequences in Denmark and other countries. Second, we point out important limitations of our study and suggest directions for future research.
Effect size and potential real-world implications
Focusing on the effect of candidate gender in the control condition, the estimated AMCE of 8.6 percentage points matches the effect of candidate age but is only about half the size of the effect of medium vs. low political experience (see Figure 1, Panel A). Assessing the effect against results obtained in other recent Scandinavian conjoint experiments gives rise to mixed conclusions: Dahl and Nyrup’s (Reference Dahl and Nyrup2021) effect of 1.8 percentage points is clearly smaller, but other recent conjoint experiments provide effects close to our results (see Arnesen, Duell and Johannesson Reference Arnesen, Duell and Johannesson2019; Gothreau and Laustsen Reference Gothreau and Laustsen2026). In total, we take this as an indication that Scandinavian respondents (at least when candidates’ policy positions remain unrevealed) rather strongly favor women over men candidates.Footnote 6
Turning to the role of terrorist threat, skeptics may question the substantial importance of the results arguing that terrorist threat does not push citizens to outright prefer men candidates – it only reduces favoritism of women candidates by 3.4 percentage points. While there are reasons to see this reduction as modest – for instance when compared against the non-preregistered changes in preferences for candidates’ age and education – we also think that the obtained result can potentially have real consequences for descriptive representation of women and men. First, one may consider that if only a few lines of text can almost halve respondents’ preferences for women candidates, then exposure to more vivid, online, or TV-based coverage of terrorist threats may plausibly cause a full reversal on candidate gender preferences. Second, one also needs to take the two-step nature of the Danish electoral system into account when considering real-world implications of our results. As explained above, party votes are first used for allocating seats between parties, and then personal votes for candidates nominated by a given party are used for assigning that party’s seats. Because a few hundred votes can be decisive for winning a seat or not (especially among candidates competing for a party’s last seat in a given district), we see it as quite plausible that the estimated reduction in the effect of candidate gender caused by terrorist threat can affect whether a given seat is won by a man or woman candidate. Consequently, although the estimated effect of terrorist threat appears modest, we believe it may exert real electoral impact.
Limitations and future research
Due to space limitations, our survey regrettably did not contain a manipulation check of respondents’ threat perception from PET’s official assessment of the risk of terrorism against Denmark. However, media coverage of PET’s assessment included headlines like ‘PET sees intensified terrorist threat in Denmark’ (our translation; Politiken.dk 2024), which signals that journalists perceive the threat as substantial and real. Still, we urge future research to directly measure threat perceptions to verify that employed treatment materials work and illuminate the psychological mechanisms underlying a threat’s effect on candidate preferences.
A second limitation relates to the follow-up analyses suggesting that left-wing respondents drive the aggregate results. Sample size was primarily determined for testing the Terrorist Threat Hypothesis and, thus, one should be particularly cautious when interpreting the reported moderations by political orientation. Future studies should employ larger samples to permit deeper exploration of heterogeneity in effects of terrorist threat on candidate preferences.
Third, in our experiment, respondents always share partisanship with the candidates they are exposed to. This design choice rests on pragmatism and the wish to avoid partisanship washing out effects of all other candidate attributes (Hjortskov and Andersen Reference Hjortskov and Calmar Andersen2024). Moreover, this choice scenario also corresponds to theories about Danish voters’ decision-making strategies when casting personal votes (first finding their favored party and subsequently their favored candidate; see section ‘The case of Denmark’). Still, to explore the scope conditions of the effect of terrorist threat on preferences for candidate gender, future studies may vary whether respondents and candidate profiles share partisanship or not. Relatedly, one may speculate whether the reported results would replicate in other countries. Based on the low occurrence of terrorist attacks and women’s high political representation in Denmark, we saw ample potential for finding support for the Terrorist Threat Hypothesis in our Danish study. Consequently, we hope future studies will explore the generalizability of our findings in countries characterized by a less fortunate experience with terrorist attacks and/or by lower political gender equality.
Finally, we want to return to the exploratory results regarding the effect of terrorist threat on preferences for other candidate attributes than gender (see right-hand panel in Figure 1). Besides making men candidates relatively more appealing, the Terrorist Threat condition also increases respondents’ preferences for older candidates and (although non-significantly) decreases preferences for Middle Eastern candidates. That is, increased salience of terrorist threat seemingly exacerbates many well-known challenges to political representation with respondents choosing fewer women, younger, and immigrant candidates. On this background, future hopes for alleviating these existing challenges to political representation quickly appear gloomy, especially given expert assessments of current times being the most threatening since World War II (eg, Toft Reference Toft2025). Obviously, given their exploratory and non-preregistered nature, these results are only suggestive of broader effects of contextual threats on voters’ preferences for candidate attributes besides gender. Nevertheless, we see our results as underlining the need for deeper scientific scrutiny of the representational consequences of terrorism, war and related threats, not least if democracies are to counter these consequences effectively.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1475676526101649
Data availability statement
Data and replication files are available at Harvard Dataverse: https://doi.org/10.7910/DVN/F7WN3A.
Acknowledgements
The authors thank Jennifer Piscopo, Leonie Huddy, and three anonymous reviewers for valuable feedback.
Author contributions
Brian Littmann: Conceptualization-Lead, Data curation-Lead, Investigation-Lead, Methodology-Lead, Project administration-Lead, Visualization-Lead, Writing – original draft-Equal, Writing – review & editing-Equal; Nadia Natasha James Nielsen: Conceptualization-Lead, Data curation-Lead, Investigation-Lead, Methodology-Lead, Project administration-Lead, Visualization-Lead, Writing – original draft-Equal, Writing – review & editing-Equal; Lasse Laustsen: Conceptualization-Supporting, Data curation-Supporting, Investigation-Supporting, Methodology-Supporting, Project administration-Supporting, Supervision-Lead, Visualization-Supporting, Writing – original draft-Equal, Writing – review & editing-Equal.
Competing interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Ethical standards
The analyzed dataset was collected for the first and second authors’ Master Thesis project. Due to severe time limitations for MA thesis projects, it is common practice that Danish students do not obtain formal ethical approval for data collections. Instead, they must follow a set of written departmental guidelines (see: https://ps.medarbejdere.au.dk/en/about-the-department/policies-and-agreements/guidelines-for-students-collection-of-data). Moreover, because data were collected via the survey agency Epinion, all of Epinion’s standard procedures for ethically sound survey data collection were followed. Finally, according to Danish legislation, survey research is only required to obtain approval from the regional ethics committee system if it ‘involves human biological material’ (section 14(2) underlying the Danish National Research Ethics Committee). Consequently, the reported study lives up to the formal Danish legislation and follows standard ethical practices for conducting survey research.
AI use
Generative artificial intelligence (ChatGPT Pro, version 4/4.5) was used to inform and streamline the coding procedure of the dataset to ensure among other things transparency and replicability of the reported analyses.
Open access
