Imperfect Victims? Civilian Men, Vulnerability, and Policy Preferences

ho is deemed vulnerable and in need of protection has a bearing on important policy decisions, such as refugee acceptance or provision of aid. In war, dominant narratives construe women as paradigmatic victims, even while civilian men are disproportionately targeted in the most lethal forms of violence. How are such gender-essentialist notions reflected in public opinion? Do regular citizens have inaccurate perceptions of male victimization in war, and with what consequences for their policy preferences? We carried out survey experiments among U.S. and U.K. respondents on both real and hypothetical conflicts, where we emphasized or varied the gender of the victims. In support of our expectations, respondents consistently underestimate the victimization of men, perceive civilian male victims as less innocent, and hold anti-male biases when it comes to accepting refugees and providing aid. However, informing respondents of the vulnerability of male civilians to targeted assassinations and massacres mitigates these effects.


INTRODUCTION
A cross armed conflicts, civilian targeting is more than an isolated occurrence: besides less frequent mass atrocities such as genocide or mass killings, armed actors often "undertake a fairly constant level of low-intensity violence" against civilians (Eck and Hultman 2007, 234). We know from prior public opinion research that citizens in Western democracies care about such human rights violations and are more likely to support military intervention in armed conflicts with large-scale targeting of civilians (Agerberg and Kreft 2022;Kreps and Maxey 2018;Tomz, Weeks, and Yarhi-Milo 2020). Whose fate and vulnerability in war is recognized or discounted in these processes has not been sufficiently examined, however. In light of the highly gendered nature of conflict violence on the one hand, and of gender-essentialist narratives of victimization that predominate in public and policy discourses on the other, the gender of the victim is a particularly interesting factor to consider. This is the angle we pursue in this study.
Prior research has amply demonstrated that armed conflict affects men and women in distinct ways, in terms of the roles they play within armed groups or in war-affected societies, and in terms of the conflict violence they experience (Buvinic et al. 2013;Carpenter 2006;Henshaw 2016;Kreft 2019;Shekhawat 2015).
Whereas women and girls are more likely to suffer conflict-related sexual violence (Buvinic et al. 2013;Cohen and Nordås 2014), men and boys are disproportionately affected by lethal violence, including massacres like those in Srebrenica, but also by (legal) forced conscription or (illegal) forced recruitment into nonstate armed groups, as recently observed in the Russian war against Ukraine (Carpenter 2006;Ormhaug, Meier, and Hernes 2009).
This violence commonly targets men specifically as men-because in accordance with gender-essentialist notions, they are perceived as naturally belligerent and distinctly threatening (Carpenter 2006;Elshtain 1982;Glick and Fiske 1996). Such male-directed forms of violence, in short, are "gender-based, because they are rooted in assumptions about male wartime roles, assumptions that both reflect and reproduce gendered hierarchies prevalent in both peacetime and war" (Carpenter 2006, 88). The consequences for civilian men are detrimental. Estimates across conflicts classify men as between 1.3 and 8.9 times as likely to be killed in war as women (Ormhaug, Meier, and Hernes 2009). In Syria, for example, available data indicate that almost 90% of adult civilians killed have been men. 1 The result is a pattern where men constitute the majority of direct, battle-related casualties in wars, whereas women's life expectancy tends to be affected more indirectly, including via displacement, economic disparities, and unequal access to healthcare infrastructure (Ghobarah, Huth, and Russett 2003;Plümper and Neumayer 2006).
Studies show that despite their gendered vulnerability, male civilians 2 tend to be overlooked in humanitarian responses to armed conflict (Griffiths 2015;Krystalli, Hawkins, and Wilson 2017;Olivius 2016;Turner 2019). This is in part because discursive gender essentialisms facilitate the construction of female victim-male perpetrator or female victim-male protector binaries in policy circles and the media (Carpenter 2005;Johnson 2011;Kronsell 2016). We lack systematic insight, however, into the salience of these conceptions in society at large, their underlying micro-foundations, and their consequences. In this study, we therefore tackle the following questions: To what extent and how are people's assessments of vulnerability and innocence in war gendered? Specifically, do regular citizens have accurate perceptions of men's vulnerability in war? How do they perceive male victims of conflict violence? And, how do these perceptions link to policy preferences?
Theoretically, we leverage the discursive feminization of victimhood and civilians, the construction of civilian males as legitimate targets of political violence in war, and the relative invisibility of male victimization and vulnerability in policy and media responses to armed conflict. Based on these, we hypothesize that the general public: (1) underestimates men's victimization in war, (2) is less likely to perceive civilian men victims as innocent, and is less supportive of (3) providing humanitarian aid when most victims are (perceived to be) men and (4) accepting refugees when most victims are (perceived to be) men. In three different pre-registered survey experiments carried out among U.S. respondents, as well as two replications with U.K. respondents, we find support for our expectations. These results have significant implications. They expose anti-male biases in protection imperatives that can endanger the human rights of men affected by war. Moreover, prior research indicates that gendered notions of who is an acceptable refugee can be strategically instrumentalized to generate negative public opinion toward liberal refugee regimes (Griffiths 2015;Scheibelhofer 2017). The latter not only harms men, but also women, children, and people with nonbinary gender identities affected by war. A central question then becomes whether we can correct antimale biases and mitigate their consequences. In a second step, we examine whether receiving factual information about men's vulnerability in war can reduce anti-male biases in policy preferences. We find that, indeed, it does.

PUBLIC OPINION ON CONFLICT VIOLENCE AND GENDER
Prior public opinion research has illustrated that citizens in Western democracies, such as the United States, the United Kingdom, and Sweden, are moved by moral concerns about democracy, human rights, and the physical integrity rights of citizens in other countries, and that these shape their policy preferences (Agerberg and Kreft 2022;Kreps and Maxey 2018;Leiby, Bos, and Krain 2021;Tomz and Weeks 2013;. Thus, different survey experimental studies show that large-scale human rights violations and the targeting of civilians are important drivers of citizens' support for humanitarian intervention in, or war against, other countries (Agerberg and Kreft 2022;Kreps and Maxey 2018;. Importantly, such moral concerns remain salient even when issues of state and international security are at stake, that is, when the country in question is developing nuclear weapon capabilities . A recent survey experimental study likewise indicates that even in a hypothetical scenario in which North Korea launches a nuclear attack, a clear majority of U.S., Japanese, and South Korean publics reject nuclear retaliation, a decision that is driven at least in part by concern about civilian casualties in North Korea (Allison, Herzog, and Ko 2022). These findings are a strong indicator that human rights violations and the targeting of civilians are important factors that shape public opinion toward states and societies affected by war, in both Western and non-Western democracies. Where we need greater nuance, however, is in understanding how concern for the human rights and physical integrity of citizens living in war-affected countries is shaped by entrenched notions of who is vulnerable and in need of protection.
Specifically, we extend prior survey experimental research by taking a close look at gender dynamics, and in particular at how gendered victimization patterns in war affect public opinion and perceptions. In doing so, we bring (experimental) public opinion research into dialogue with the rich qualitative literature that has identified, first, entrenched genderessentialisms that uphold a female victim-male perpetrator binary (Carpenter 2005;Johnson 2011;Kronsell 2016) and, second, the neglect of male victims in both discourse and humanitarian practice (Griffiths 2015;Krystalli, Hawkins, and Wilson 2017;Olivius 2016;Turner 2019). Prior survey experimental work has certainly capitalized on the gender-essentialist framings of civilian victims that we often encounter in news reporting and public discourse. In their study exploring under what conditions U.S. citizens support humanitarian intervention, Kreps and Maxey (2018) use a treatment of a "massacre of civilians, including innocent women and children" (1827) and a protection treatment that mentions civilians, "many of whom are women and children" (1825). These formulations ensure high ecological validity, as they accurately mirror the way in which civilians in war tend to be strategically framed to elicit empathy and support from states and international actors (Carpenter 2005). In this study, we theorize and empirically examine the gendered underpinnings of these narratives, and their consequences for policy preferences, further. Agerberg and Kreft (2022) have taken a first step toward integrating a gender dimension into survey experimental public opinion research relating to armed conscription or forced recruitment) also apply to adolescents or even children. Therefore, we use terminology such as "male victims" rather than "men victims" throughout, unless we refer specifically to adult men in our experimental design and results. conflict. They find that when sexual violence, as a distinctly gendered form of violence that is associated with female victimization, is prevalent in a conflict, support for military intervention increases. The authors touch upon actual gendered victimization patterns only tangentially, however, finding that respondents generally consider women to be more likely to be victims in war than men. They also do not explore gender-based violence beyond sexual violence. We complement this prior research by tackling the following questions: Where are the men in public conceptions of victimization, vulnerability, and innocence in war? How do gendered perceptions of victimization and vulnerability affect policy preferences, specifically refugee acceptance and support for aid provision?

MEN PERCEIVED AS IMPERFECT VICTIMS
We draw on three themes in the existing literature to formulate our theoretical expectation of anti-male biases in public opinion relating to civilians in armed conflict: (1) the discursive feminization of victimhood and civilians, (2) the construction of civilian males as legitimate targets of political violence, and (3) the relative invisibility of male victimization and vulnerability in policy and media responses to armed conflict. As they are to varying extents embedded in, exposed to, and participating in societal, public, and policy discourses, we expect that citizens hold many of the gender-essentialist notions and ideas underlying these themes.

Feminization of Victimhood and Civilians
Perceptions of victimization and vulnerability in warin policy and public discourses-are distinctly gendered, and in ways that are not congruent with the gendered victimization patterns that conflict researchers have identified. Instead, in the popular imaginary as in policy circles, women tend to be viewed as the paradigmatic victims (and victims as women), whereas men are more readily conceptualized as either heroic protectors or perpetrators (and protectors and perpetrators as men) (Åse 2015;Carpenter 2005;Nagel 2019;Peet and Sjoberg 2019;Young 2003). Forty years ago, Elshtain (1982) described the resulting discursive dichotomy as one between the woman as the "beautiful soul" and the man as the "just warrior." In this reading, the woman is ascribed attributes such as innocence, passivity, and worthiness of masculinist protection.
The equation of innocent civilians with women and the feminization of victimhood are not accidental. As Carpenter (2005) illustrates, such gender essentialisms make it easier to attract international attention and to motivate international involvement in humanitarian crises, because these essentialisms remain entrenched also in allegedly liberal and gender-equal democracies. Empirical evidence of this can be found, for example, in the United States, the United Kingdom, and, to a lesser extent, even in Sweden, where the (suspected) victimization of women yields stronger support for military intervention (Agerberg and Kreft 2022). The narrative of women's innocence, vulnerability, passivity, and need for protection is at the core of protective paternalism: the notion that men should protect and provide for "the weaker sex" (Glick et al. 2000;Glick and Fiske 1996, 492-3).
Human rights and protection advocates hoping to direct policy attention to the plight of civilians thus face strategic incentives to avail themselves of salient gender-essentialist messaging, that is, of "tropes associating men and women with mutually exclusive and oppositional attributes" (Carpenter 2005, 296), whereby women are paradigmatic victims and men are not. It is no coincidence that conflict-related sexual violence, conceptualized as targeting women and girls, has received such policy and academic attention that it has come to epitomize gender-based conflict violence (Crawford 2017;Meger 2016), and become an important factor in attracting international attention and involvement in conflict (Benson and Gizelis 2020;Hultman and Johansson 2017;Kreutz and Cardenas 2017).
In the refugee context-one of the immediate results of armed conflict-the feminization of vulnerability is likewise highly prevalent: "these are images that constitute imagined refugees: masses of humanity, nameless women and children fleeing violent conflict and living in destitution, the victims of tragedy searching for a place where they can rebuild shattered lives" (Johnson 2011(Johnson , 1015. Humanitarian actors, Johnson argues, play on the vulnerability, voicelessness, passivity, and innocence that women (and children) refugees stereotypically evoke, in order to attract sympathy and donations.
In sum, we observe "an essentialized 'protection' discourse that associates women but not men with civilian status … and with vulnerability to armed attack" (Carpenter 2005, 310). This constrains the agentive space for women and the conceptual space for male vulnerability, especially in contexts characterized by violent masculinity as the primary form of agency, such as armed conflict. In the next sections, we elaborate this masculinized armed conflict context and its implications for the invisibility of male victimization and male vulnerability further.

Civilian Men as Legitimate Targets in War?
The discursive feminization of civilians and of victimization stands in stark contrast with the reality of gendered civilian targeting in war. As noted above, the scholarship on gender and conflict has identified clear -albeit cross-case variant-patterns of the disproportionate targeting of men (and boys) in specific kinds of conflict violence, especially massacres and other types of targeted lethal violence against civilians (Carpenter 2006;Jones 2004;Ormhaug, Meier, and Hernes 2009). It is worth noting that in these processes, armed actors target "military-age males" strategically based on their gender because they perceive them as particularly threatening to an extent that women and (female) Imperfect Victims? Civilian Men, Vulnerability, and Policy Preferences children never would be (Carpenter 2006;Davis 2016). As Carpenter (2006, 88) notes: …military-age men and adolescent boys are assumed to be "potential" combatants and are therefore treated by armed forceswhether engaged in formal battle, in lowintensity conflict, or in repression of domestic civilian populationsas though they are legitimate targets of political violence.
Such gendered threat perceptions are not entirely without empirical basis. There is now plenty of evidence that women, too, are actively involved in armed groups in support and combat roles across contexts (Henshaw 2016;Thomas 2017;Thomas and Bond 2015) and that they too are active participants in atrocities, such as conflict-related sexual violence (Cohen 2013). Nonetheless, men constitute the majority of perpetrators of conflict violence, and armed groups remain heavily male-dominated (Buvinic et al. 2013;Loken 2017;Shekhawat 2015). This empirical reality does not, however, justify targeting people based on their gender. Blurring the line between civilian and combatant by preemptively killing military-age civilian males just because they are statistically more likely than other groups to take up arms in future violates international law (Carpenter 2006;Ni Aolain 2021).
It is also worth noting that armed actors themselves are deeply implicated in upholding the gendered conflict patterns to which they then respond with gender-based lethal violence. Practices of male-specific conscription or (forced) recruitment, and the conscious nurturing of violent masculinities within military structures, are key drivers behind masculinized combat (Carpenter 2006;Eichler 2012;Goldstein 2001). Even though norms linking masculinity to warring may be contested even within traditionally militaristic societies (Eichler 2012), male-specific conscription is globally not recognized as a gender issue of sociopolitical importance (Heikkilä and Laukkanen 2022). This is although conscription-age males are during episodes of armed conflict-as recently in Syria or Ukraine-often prohibited and actively prevented from leaving the country without special permission, which pushes male refugees who do manage to flee into the sphere of illegality (Davis, Taylor, and Murphy 2014). As Davis (2016) notes, the failure to perceive as vulnerable civilians those military-age males seeking to avoid conscription or forced recruitment is not only normatively problematic from a protection standpoint, but it also forces many men who do not wish to take up arms to stay in a setting in which they are ultimately forced to kill and/or be killed. This is why Carpenter (2006) classifies male-targeted forced conscription as a form of genderbased violence.
In brief, armed groups actively reinforce an understanding of war as a distinctly masculinized affair, in terms of who is coerced into combat, who is perceived as a suspected enemy or potential future threat, and who is a "legitimate target" of lethal violence (Carpenter 2006, 88;Ni Aolain 2021). Of particular interest to our theoretical argument, there is evidence that male-directed lethal violence is about more than empirical realities of gendered perpetration patterns. The disproportionate targeting of males thus persists even in contexts in which women's (and girls') participation in armed groups is disproportionately high, such as in Colombia, where the FARC rebel group boasted 30%-40% female members (Carpenter 2006, 90). Likewise, recent data from the Syrian war show that while both targeted and indiscriminate state violence against female civilians increase in areas where armed rebel groups with higher levels of women's participation are present, the vast majority (ca. 85%-90%) of those targeted are still male (Gohdes 2019). These patterns suggest that gender essentialisms that help determine who is viewed as innocent or threatening, as a civilian or a potential combatant, largely overshadow the empirical reality of armed group membership. We extend this logic of pervasive gender essentialisms to the general population in our theoretical expectations articulated below. In our experiments, we focus primarily on massacres/targeted killings, for which we have the most reliable data from more recent conflicts, but consider also conscription/forced recruitment as forms of gendered conflict violence that disproportionately target men.

Invisibility of Male Victimization and Vulnerability
Stereotypically gendered conceptions of men's and women's roles in war make it more difficult for men to conceptually occupy the victim space in the eyes of humanitarian actors and observers (Carpenter 2006;Jones 2004;Schulz 2020;Touquet et al. 2020;Turner 2019). In both research and practice, the emphasis tends to be on violent or hegemonic masculinities and on male-perpetrated violence in war, which sidelines nonviolent masculinities and male vulnerabilities (Duriesmith 2020;Myrttinen, Khattab, and Naujoks 2017). In fact, vulnerable civilian men are likely to face suspicion or even hostility as suspected or potential perpetrators much more than they are considered victims (Krystalli, Hawkins, and Wilson 2017;Myrttinen, Khattab, and Naujoks 2017;Olivius 2016;Turner 2019).
Such perceptions build on a problematic and empirically inaccurate dichotomy between victimhood and agency, which is salient in the context of armed conflict and has distinctly gendered overtones (Kreft and Schulz 2022). Reporting on his research of the Jordanian refugee response to the Syrian war, Turner (2019, 608) thus notes: "[w]hile Syrian women were assumed to need 'empowering,' Syrian men were assumed to be independent and agential." Men being perceived as agentic thereby becomes the justification for them not requiring specific support. Similarly, Olivius (2016) finds that humanitarian actors in Thailand and Bangladesh adopt three representations of male Rohingya refugees: (1) as perpetrators of violence and thus responsible for women's vulnerability, (2) as gatekeepers that can promote or obstruct gender equality and improvements for women, and

Anne-Kathrin Kreft and Mattias Agerberg
(3) as emasculated troublemakers and security threats, unable to fulfill their roles as protectors and providers. Viewed as potential perpetrators but also through the prism of male agency generally, male refugees are thus assessed in relation to the enabling or obstructing role they can play for the empowerment of women. Their own vulnerability and need for protection or support are eclipsed in the process (Olivius 2016;Turner 2019).
The consequences of these narratives are detrimental for vulnerable civilian men, especially those displaced by war. With men being considered agentic, and refugees imagined as passive, traumatized, and feminized, male refugees are more frequently viewed as "bogus asylum seekers" from the outset, as Griffiths (2015) illustrates in the U.K. context. In Colombia, displaced families, and women in particular, are likewise prioritized over single men in policy areas such as protection, border crossings, or food distribution because they are seen as more vulnerable, innocent, and in need (Krystalli, Hawkins, and Wilson 2017). As a result of these gendered prioritizations, many men and male minors are left to fend for themselves and are often pushed into transactional sex in order to ensure their basic livelihood and survival (Krystalli, Hawkins, and Wilson 2017, 32). This exposes them to renewed vulnerability and, particularly in the case of minors, victimization. In Austria, in turn, the discursive reframing of refugees as young, male, and Muslim helped legitimize restrictive policies in response to an influx of refugees in 2015 (Scheibelhofer 2017). 3 In brief, a growing body of qualitative studies has elucidated how humanitarian and other policy actors overlook men's vulnerabilities and victimization, while perceiving and framing them as resourcefully agentic, sometimes with distinctly aggressive, dangerous, and with criminal connotations. However, we lack a systematic overview of how widespread such perceptions of men's victimization/vulnerability in war are among the general public in Western democracies, and whether and how they affect political attitudes. Is the neglect of men in humanitarian programming mirrored in a decreased support for humanitarian aid when men rather than women are victims? Likewise, is acceptance of men refugees lower than that of women refugees? Turning to micro-level analysis of public opinion allows us to tease out (some of) the components of anti-male biases individuals may hold (underestimation of male victimization and male civilians perceived as less innocent), how these affect policy preferences (reduced support for policies benefiting male victims), and whether updating people's priors on the vulnerability of men may reduce such anti-male biases.

Why Study Public Opinion?
Investigating the drivers of policy preferences in such high-stakes issue areas as refugee acceptance and aid provision is more than an academic exercise. Humanitarian and human rights advocates "need the material and moral support of the public if they are to act freely and effectively … [and] the reactions of governments and the UN to major crises are inextricably bound up with public opinion" (Brauman 1993, 149;cited in Carpenter 2005, 315). The influx of refugees to Europe since the beginning of the Syrian war, in particular, has cast the spotlight on the importance of public support for refugee and immigration policy, which (unlike support for humanitarian aid) has received considerable scholarly attention.
Accordingly, recent studies have investigated the determinants of citizens' attitudes toward immigrants (Hainmueller and Hopkins 2015;Ward 2019) and asylum seekers (Hainmueller and Hopkins 2015). A conjoint experiment spanning 15 European countries finds, inter alia, that respondents are more accepting of asylum seekers with vulnerabilities and experiences of prior victimization, such as those who have escaped torture or who have disabilities (Bansak, Hainmueller, and Hangartner 2016). Likewise, a conjoint experiment among U.S. citizens shows more favorable attitudes toward immigrants who have escaped prosecution in their country of origin compared to those who seek to reunite with family (Hainmueller and Hopkins 2015). That is, Western respondents are sympathetic with refugees who are vulnerable or who have experienced various harms. Both studies also indicate that acceptance of male refugees and immigrants is below that of their female counterparts, although this finding receives little or no discussion in the respective studies. Examining the effect of immigrant gender explicitly, Ward (2019) finds that among German respondents, support for receiving a group of immigrants decreases as the share of young men in that group increases. Further analyses reveal that threat perceptions appear to help explain this pattern: the greater the share of young men in a refugee group, the greater the perceived security threat that group poses (Ward 2019).
In this study, we build on, extend and further nuance this existing public opinion research. We employ survey experiments to examine in greater depth to what extent people hold empirically (in)accurate perceptions of male victimization in war; how the gender of the victims affects perceptions of innocence; and how such gendered perceptions translate into refugee acceptance and support for humanitarian aid in conflict-affected settings. In doing so, we integrate, synthesize, and extend prior experimental research on attitudes toward immigration (Bansak, Hainmueller, and Hangartner 2016;Hainmueller and Hopkins 2015;Ward 2019) and on public support for military intervention in war (Agerberg and Kreft 2022;Kreps and Maxey 2018;Tomz, Weeks, and Yarhi-Milo 2020). Our main expectations are, based on the rich qualitative literature on gender essentialisms and neglect of men in humanitarian responses, that respondents: (1) underestimate the victimization of men in war; (2) perceive male civilian victims as less innocent; and are less supportive of (3) providing humanitarian aid and (4) accepting refugees if conflict victims are predominantly men. We develop specific hypotheses to match the respective experimental designs in each of the sections below.

RESEARCH DESIGN
We designed three experimental studies with different parts to explore how people perceive the victimization and vulnerability of men in war, and with what consequences. We focus specifically on forms of violence that conflict researchers have identified as primarily affecting men as men. This choice is theoretically motivated: lethal violence, including targeted killings and massacres, as well as forced recruitment/abduction are customarily overlooked as forms of gender-based violence in policy and discourse, 4 which further sidelines men's victimization in war. Here, we examine to what extent we find evidence that the public is indeed less aware or more dismissive of men victims in war.
All research questions and the corresponding hypotheses, which are detailed below, as well as the experimental designs and planned analyses were preregistered on OSF 5 prior to data collection. The text of the entire surveys for all three experiments is included in the Supplementary Material (Kreft and Agerberg 2023). The survey experiments were administered through Qualtrics and carried out with U.S. respondents recruited through Prolific, an online platform shown to yield high data quality (Peer et al. 2022). Studies 1 and 2 used Prolific's representative sample option, whereas the larger sample size required precluded this option for Study 3. This means that the samples for the first two studies are representative of the U.S. population on age, gender, and race. Yet Democrats and people with a college degree are overrepresented in the samples. The sample for the third study is younger and includes a slightly higher share of white/Caucasian respondents, but is balanced when it comes to the share of Democrats and Republicans. Descriptive statistics for all three samples can be found in Appendix A of the Supplementary Material. To explore the potential consequences of nonrepresentativeness, we explore treatment effect heterogeneity in Appendix B of the Supplementary Material. In short, we find little evidence of effect heterogeneity. This makes us cautiously optimistic with regard to the generalizability of the overall results to the U.S. population (Coppock, Leeper, and Mullinix 2018;Hartman 2021). The lack of heterogeneity together with the fact that we successfully replicated the main results with U.K. respondents, and in a very different geopolitical context (see below and Appendix C of the Supplementary Material), also suggests that the results might be generalizable to other Western democracies. The latter point is, however, still uncertain and something for future research to explore.

Study 1: Syria Experiment
We designed Study 1 to test the following hypotheses: H1: Individuals hold empirically inaccurate perceptions of victimization patterns in war, in that they underestimate the victimization of men.
H2a: When correct information on actual gender patterns of victimization is provided, civilian causalities are less likely to be viewed as innocent bystanders.
H3a: When correct information on actual gender patterns of victimization is provided, support for increasing humanitarian aid decreases.
H4a: When correct information on actual gender patterns of victimization is provided, support for accepting refugees decreases.
Study 1 focuses on the civil war in Syria. Referring to an actual, ongoing conflict has the obvious advantage of a high degree of realism and relevance. Different organizations have also been tracking the number of civilians casualties in the conflict, including the gender distribution of the victims, which gives us a benchmark to which we can compare respondents' perceptions. All estimates of civilian casualties in the experiment were taken from the U.K.-based organization The Syrian Observatory for Human Rights (SOHR). 6 The numbers reported were the most recent at the launch of the experiment in November 2021. In the experiment, we first provided all respondents with the estimated number of civilian adults killed in the war (135,000) and then asked them to estimate the share of men and women among the victims using a slider running from 0 (0% male victims) to 100 (100% male victims). 7 In Bayesian terms, our assumption is that respondents have an overly negative prior about male victimization (Hill 2017). Our main expectation, as indicated by H1, is thus that respondents' mean prior is significantly below available estimates of actual men's victimization. We should also expect the variance of the distribution of priors to be relatively high, given that respondents, as per our argument above, are likely to have little actual information about male victimization. 4 By contrast, men's victimization in sexual violence-a form of gender-based violence that primarily affects women and girls-has received increased and overdue academic attention in recent years (see, e.g. Edström and Dolan 2019; Schulz 2020), including, more tangentially, in survey experimental work (Agerberg and Kreft 2022). 5 https://osf.io/bgstz/. 6 https://www.syriahr.com/en/. 7 Respondents were randomly assigned (with p = 0:5 ) to either estimate the share of "male victims" or the share of "female victims." The estimates were combined and coded so that higher values always refer to a higher share of male victims. We do not wish to reinforce a binary understanding of gender, but for the sake of keeping survey design and response options simple, we ask about men and women only. This also maps onto the gender-essentialist narratives that are theoretically relevant for this study.
After providing their prior beliefs about men's victimization, respondents were either assigned to a control group (with p = 0:5 ) that received no additional information or to a treatment group that received the following information on the next screen: "According to the Syrian Observatory for Human Rights, about 88.6% of the civilian victims in the war in Syria are men and 11.4% are women." Next, respondents in both groups were asked three different outcome questions: "How likely would you say it is that the civilian casualties of the war in Syria generally have been innocent bystanders in the conflict?"; "Do you support or oppose the United States increasing its humanitarian aid to Syria by 10%?"; and "Do you support or oppose accepting 500 additional Syrian refugees into the United States?" 8 These outcomes correspond to hypothesis H2a-H4a. All answers were provided on five-point Likert scales that were normalized to range from 0 to 1.

Study 1: Results
A total of 1,294 respondents on the Prolific platform completed Study 1. The data were collected between November 6 and 8, 2021.
Do respondents underestimate the victimization of men? The first outcome question (prior to the information treatment) allows us to explore this question by comparing respondent estimates to the estimates reported by SOHR for the Syria civil war. These results are presented graphically in Figure 1.
As shown in Figure 1, most respondents' estimates of the share of male casualties are substantially lower than the estimate reported by SOHR of 88.6%. The mean respondent estimate is about 57%. 9 Put differently, almost 99% of respondents provide an estimate that is lower than that of SOHR. In this sense, respondents clearly have an overly negative prior. This is initial evidence that people underestimate the extent of male victimization in conflicts. At the same time, respondents do correctly estimate that a majority of civilian casualties are men; we thus do not find evidence of a sweeping conceptual feminization of victimhood. As shown in Figure 1, the variance of the distribution of priors is also high, with an estimated standard deviation of 17.3. The spread-out distribution indicates that many respondents likely have weak priors based on little information.
What happens, then, when we provide respondents with more specific information about male victimization

Civil war in Syria
Note: The dashed vertical line indicates the estimate reported by the Syria Observatory for Human Rights. 8 We specified the exact increases to make responses easier to interpret and to avoid a situation where respondents have very different conceptions of what an "increase" would entail. We opted for numbers that would not seem too drastic, but that still would seem substantial for regular citizens without in-depth knowledge of U.S. refugee policy or aid budget. For reference, the United States admitted about five hundred Syrian refugees each year between 2019 and 2021 (https://www.statista.com/statistics/742553/syrian-refugeearrivals-us/). 9 We pre-registered a very conservative test of H1: whether the mean respondent guess was below 50%. A conservative hypothesis is certainly warranted, as considerable uncertainties exist about quantitative data in war contexts. Disaggregation by gender further magnifies these uncertainties in the data. Given the extremely high share of male casualties reported by SOHR, however, the 50% threshold was, in hindsight, unnecessarily conservative. Since the mean respondent guess was 57%, the hypothesis was not supported by the test that we registered. However, it is clear that respondents Imperfect Victims? Civilian Men, Vulnerability, and Policy Preferences patterns? Given prevailing gendered understandings of vulnerability and victimization and gendered protection norms, we predicted that respondents would be less benevolent toward the civilian victims of the war upon learning that a very high share are men. We test this conjecture by regressing our three different outcome questions on a treatment indicator that equals 1 for respondents who received the information with the SOHR estimate of the share of male victims and 0 for respondents in the control group who received no additional information. 10 The results are presented in Figure 2 (for full results, see Table A4 in Appendix D of the Supplementary Material).
All estimates go in the expected direction: respondents in the treatment group learning that a very high share of the victims are men are less likely to view the civilian victims of the war in Syria as innocent bystanders, are less supportive of U.S. humanitarian aid to the country, and are somewhat less supportive of the U.S. accepting Syrian refugees. However, while the two former contrasts are statistically significant, the latter is not. The corresponding effect sizes are 0.4, 0.13, and 0.07 (Cohen's d). This, hence, lends strong support for H2a: when the share of civilian victims that are men is perceived to be higher, respondents are substantially less likely to view the civilians as innocent. As shown with regard to the aid outcome, and in line with H3a, this can also affect policy preferences. Simply put, Syrian civilians are viewed as less in need of U.S. humanitarian aid when male victimization is emphasized. This, hence, constitutes evidence that men's vulnerability and victimization might be partly overlooked in the conflict in Syria. However, are these patterns limited to the Syrian case, which had been frequently covered in the news in the years preceding the survey? Study 2 approaches the same overarching research questions from a more abstract perspective to gage respondents' general intuition about gendered patterns of victimization.

Study 2: Vignette Experiment
In Study 2, we test the following hypotheses: H2b: Civilian men are less likely to be perceived as innocent bystanders when they become victims of conflict violence than civilian women.
H3b: When men are explicitly mentioned as victims, support for increasing humanitarian aid decreases, compared to when women are explicitly mentioned as victims.
H4b: When men are explicitly mentioned as victims, support for accepting refugees decreases, compared to when women are explicitly mentioned as victims.
We constructed a vignette based on a hypothetical conflict scenario. The hypothetical scenario gave us more freedom in experimental design than the real case of Syria, as it allowed us to randomly vary several different factors (while avoiding deception) to further tease out any general, gendered patterns in perceptions of vulnerability and victimization. Respondents were given a vignette resembling a news article format (randomized components are shown in brackets):

Syria study
Note: The figure shows predicted values along with 95% confidence intervals. All predictions based on a linear regression model where the dependent variable is regressed on a treatment indicator. The outcome variables were normalized to range from 0 to 1. Dependent variable means: 0.74 (Innocent bystanders), 0.68 (Aid support), and 0.72 (Support for refugees). The y-axes are scaled by the variable mean AE0:5 SD. The full results table is included in Table A4 in Appendix D of the Supplementary Material.
underestimate the share of male victims in relation to SOHR's estimate of 88.6%. This is qualitatively in line with H1. In Study 3, we pre-registered a different, arguably more reasonable, test of the same hypothesis that is supported by existing data. 10 The models were estimated using OLS and robust standard errors. years. Conflict intensity has generally been low, with skirmishes concentrated primarily in the areas surrounding the capital city. But in the past 6 months, the conflict violence has flared up. Last week, a medium-sized town 375 miles from the capital city was attacked by armed rebels.
The brutality of the attack shocked international observers. Official sources reported a massacre of civilians in the central village square, usually the source of buzzing market activity. An estimated 370 civilians were shot or stabbed with machetes. Most of the causalities were [men/women]. The town is situated in a part of the country that had not previously been affected by the conflict. International observers express concern about an escalation of the violence and about an emerging pattern of massacres of primarily [men/women] by the rebel group.
Half of the respondents were assigned to a near identical version of the vignette, but where the victims instead were abducted by the rebel group (both versions can be found in the Supplementary Material). The main treatment in the experiment is thus simply whether the victims of the attack are described as mainly men or women. By mentioning that the town had not previously been affected by the armed conflict, we sought to convey a very low probability that the civilians targeted may have in any way been affiliated with any conflict actor-patterns that, as discussed in the theory section, tend to be gendered due to both forced and voluntary recruitment. We randomized the region to introduce some specificity and relation to the real world in the scenario respondents received, while reducing the likelihood of people's associations with a specific world region affecting the results. After reading the vignette, all respondents were asked three outcome questions similar to those in the Syria experiment: "How likely would you say it is that the casualties in the described scenario are innocent bystanders in the conflict?"; "How likely would you be to support the U.S. increasing humanitarian aid to help the civilian population in the described conflict scenario?"; and "How likely would you be to support the U.S. accepting 500 refugees from the conflict hotspot?" Answers were again provided on five-point Likert scales that were normalized to range from 0 to 1.

Study 2: Results
The experiment was administered to 1,280 respondents on the Prolific platform between November 7 and 8, 2021. There are no overlapping respondents with Study 1. Similar to Study 1, we regress the three outcome questions on an indicator that equals 1 if the victims in the vignette were described as "mostly men" and 0 if the victims were described as "mostly women." 11 The results are displayed graphically in Figure 3 (for full results, see Table A5 in Appendix D of the Supplementary Material).
The general patterns are very similar to Study 1. Again, respondents are substantially less likely to view the victims in the scenario as innocent bystanders when the civilian casualties are described as "primarily men." Respondents are also less likely to support aid provision to help the civilian population and to accept refugees from the conflict hotspot when the victims of the attack are described as men. All three contrasts are statistically significant with the corresponding effect sizes of 0.52, 0.17, and 0.20, respectively (Cohen's d).
We thus find strong support for H2b, H3b, and H4b.
This further strengthens our argument that people have a gendered perception of civilian victims in conflicts. Men are viewed as less innocent, less vulnerable, and less in need, or less deserving, of help. Crucially,

Vignette study
Note: The figure shows predicted values along with 95% confidence intervals. All predictions based on a linear regression model where the dependent variable is regressed on a treatment indicator. The outcome variables were normalized to range from 0 to 1. Dependent variable means: 0.82 (Innocent bystanders), 0.75 (Aid support), and 0.71 (Support for refugees). The y-axes are scaled by the variable mean AE0:5 SD. The full results table is included in Table A5 in Appendix D of the Supplementary Material. 11 The models were estimated using OLS and robust standard errors. Imperfect Victims? Civilian Men, Vulnerability, and Policy Preferences these patterns are present both in the specific case of the war in Syria and in a more general hypothetical case. Overall, the results from Studies 1 and 2 may seem depressing: by providing accurate information about gendered patterns of victimization, people become less benevolent toward civilian victims of conflict violence.
Some might conclude in light of these results that, from a normative standpoint, it may be best to refrain from correcting misperceptions about gendered victimization patterns in armed conflict. Clearly, drawing attention to the killing (and abduction) of men depresses support for providing humanitarian aid and accepting refugees, while men victims are perceived as less innocent than women victims. Implicitly perpetuating gender essentialisms, however, gives rise to different harms. The most prominent are the neglect and suspicion of male refugees that prior qualitative studies have documented across contexts, as laid out in the theory section. With this in mind, we explore next whether updating respondents' priors about male victimization can shift negative attitudes toward refugee men. In Study 3, we first establish the existence of antimale biases in refugee acceptance and examine then whether providing accurate information on the vulnerability and specific targeting of male civilians can prompt people to reconsider these biases.

Study 3: Refugee Acceptance
We designed Study 3 to further illuminate the patterns uncovered in Studies 1 and 2, and to better understand how policy preferences are affected by gendered perceptions of conflict violence. The study is inspired by the literature showing that people tend to have a general preference against male refugees (Bansak, Hainmueller, and Hangartner 2016;Hainmueller and Hopkins 2015), in particular against younger men (Ward 2019). We are especially interested in whether information about male vulnerability, in particular the fact that civilian men are often the victims of targeted killings, shifts people's anti-male policy preferences.
We test the following specific hypotheses: H1: Individuals underestimate the victimization of men in war.
H5a: Individuals are less likely to support admitting a group of refugees into the United States the larger the share of men among the refugees.
H5b: Information on the vulnerability of men to lethal violence (massacres and targeted killings) mitigates the anti-male bias in refugee acceptance.
H5c: Information on the vulnerability of men to lethal violence (massacres and targeted killings) increases support for humanitarian aid that specifically addresses the needs of men.
We again examine H1, suggesting that respondents have an overly negative prior, but the framing of the question is more global in focus than in Study 1. H5a is a conceptual replication of Ward (2019) and tests to what extent people have anti-male biases when it comes to refugee acceptance. H5b and H5c focus on how information about men's vulnerability in war influences people's anti-male policy preferences.
We again administered Study 3 through Qualtrics, with respondents from the United States recruited through Prolific (due to the sample size of 2,590 respondents, the representative sample option was not available for this study).
To tap into respondents' priors about the rate of male victimization, we first asked them about their perception of male and female civilian casualties in conflicts around the world. We posed the question: According to your best guess, on average, what percent of civilian casualties in conflicts are men and women respectively? Respondents were asked to provide their answer using a slider running from 0% to 100%. We randomized whether respondents received a "% male casualties" or "% female casualties" label to accompany the slider. Respondents assigned to the treatment group (with p = 0:5) were then given the following information text and question (the control group skipped the block): Armed conflict affects men and women differently. Research shows that civilian men, that is, men who are not in any way involved in the fighting, are disproportionately likely to be killed in war. For example, in the war in Syria, the U.K.-based organization Syrian Observatory for Human Rights reports that 88.6% of civilian adults killed are men.
Researchers have found that such a disproportionate killing of men is common in conflicts all over the world. Often, armed groups even specifically target male civilians between the ages of 15 and 45 in massacres and assassinations.
Would you say that this information is new to you?
We included the question after the information text to further explore to what extent the provided information was novel to respondents. Again, we expect most people to have weak priors, and the provided information to be new to most respondents.
Subsequently, all respondents were administered three rounds of a conjoint experiment (Hainmueller, Hopkins, and Yamamoto 2014). The experimental design was inspired by Ward (2019). In each round, respondents were presented with information about two groups of about 50 refugees (the exact number varied between rounds) between the ages of 18 and 45, and then asked which group they would prefer to be admitted to the United States. For each group of refugees, we randomly varied the refugees' country of origin (Iraq, Yemen, Afghanistan, Nigeria, Eritrea, and Myanmar), share of men (0%, 25%, 50%, 75%, and 100%), share of women (100-the share of men), mean age (22, 23, 24, 25, 26, 27, 28, and 29), and share with a Anne-Kathrin Kreft and Mattias Agerberg university degree (5%, 10%, 15%, 20%, and 25%). The choice task was repeated two times (three choices in total), each time with new randomized profiles.
Following the conjoint experiment was a second outcome question: Do you agree or disagree that the United States should implement more humanitarian projects that specifically address the needs of men as victims in conflict hotspots around the world? Response options fall on a five-point Likert scale ranging from "strongly disagree" to "strongly agree." The design thus allows us to test whether the information given to the treatment group earlier in the study influences respondents' answers to the conjoint experiment and the aid question. As predicted by H5b and H5c, we expect respondents to revise their anti-male preference when learning about the vulnerability of male civilians in conflicts around the world.

Study 3: Results
The experiment was administered to 2,590 respondents on the Prolific platform between December 13 and 22, 2021.
First, do respondents underestimate the victimization of men in war when asked the general question about civilian casualties in conflicts around the world? Figure 4 shows respondents' estimates graphically.
Respondent estimates vary substantially. While the share of male victims is not underestimated by all respondents, the mean estimated share is clearly in the lower range of available estimates in existing studies (Ormhaug, Meier, and Hernes 2009), at 57.4 (SD = 17.6). 12 Once again, we thus find tentative support for H1 in terms of respondents underestimating the extent of male victimization, although they are once again aware that a majority of civilian casualties are men. 13 Next, we evaluate whether the information about high rates of male victimization and male vulnerability influences respondents' anti-male biases on the subsequent questions in the survey. As described above, respondents in the treatment group were asked whether the information displayed was new to them. Seventy-eight percentage of respondents answered that the information was either "definitely new" or "somewhat new." This is further evidence that respondents' priors are overly negative vis-à-vis available evidence and suggests that the information potentially also could have other downstream effects on items later in the survey.
We estimate the average marginal component effect for each treatment level of the different dimensions using OLS with standard errors clustered at the respondent level (Hainmueller, Hopkins, and Yamamoto 2014). The overall results from the conjoint experiment are displayed in Figure 5 (for the corresponding results table, see Table A6 in Appendix D of the Supplementary Material).
Similarly to Ward (2019), we find that respondents prefer groups of refugees with a higher degree of education. However, the most distinct pattern is that respondents, on average, show a very strong bias against men refugees: A group with 75% men refugees is 11% less likely to be preferred than a group with 0% (the reference group), whereas a group with 100% men  Conflicts around the world 12 The lower bound estimate in Ormhaug, Meier, and Hernes (2009) is at around 57% (or a male-female ratio of 1.3). 13 Formally, we tested whether respondents' mean guess was lower than 60 using a one-sided t-test (p < 0.001). The test was preregistered. refugees is over 33% less likely to be preferred. This is, hence, strong support for H5a. 14 To test H5b, we recoded the variable indicating the share of men refugees in a profile into a binary variable that equals 1 if a profile contains a majority of men refugees (75% or 100%) and 0 otherwise (0%, 25%, or 50%). This was in accordance with our pre-analysis plan and was done to maximize statistical power. We then interacted the binary majority men variable with the information treatment that was given randomly earlier in the survey to half of the respondents and re-estimated the main model for the conjoint experiment with the interaction included. We predicted that learning about male vulnerability in conflicts would tamper respondents' anti-male preference. As shown in Figure 6, this is indeed what we find (see the full results table in Table A7 in Appendix D of the Supplementary Material).
The treatment effect for the majority men variable is about −0.32 for respondents who did not receive information on male civilian vulnerability in war (control group). That means, refugee groups with a majority of men were 32% less likely to be preferred compared with profiles with 0%-50% men. For respondents in the treatment group, the corresponding coefficient is −0.22. Respondents receiving information about men's vulnerability in conflicts thus showed substantially less anti-male bias than those not receiving such information (the difference is statistically significant [p < 0:001]). In other words, the information treatment reduces the effect of the majority men variable by about 30%, which indicates strong support for H5b.
Finally, Figure 7 shows the treatment effect for the aid outcome, asked last in the survey (for the full results table, see Table A8 in Appendix D of the Supplementary Material).
As shown in the graph, respondents in the treatment group were on average slightly more likely to agree that the United States should implement more humanitarian projects that specifically address the needs of men as victims in conflict hotspots. The effect size is rather small (d ≈ 0:15) but statistically significant at the 0.001 level. This, hence, constitutes support for H5c.
In conclusion, across all three experiments, we find support for our hypotheses. U.S. respondents tend to underestimate the extent of victimization of men in types of violence that research has shown primarily affects military-age males, they are less likely to perceive male civilian victims of armed conflicts as innocent bystanders than their female counterparts, and they are less supportive of the provision of humanitarian aid and of accepting refugees if mostly men are affected by conflict violence. However, providing respondents with an information treatment about the vulnerability of male civilians to selective lethal violence-which a clear majority of them identified as new information-mitigates these patterns.
To probe the robustness of these findings, we conducted pre-registered direct replications of Studies 2 and 3 among U.K. respondents about 11 months after the original U.S. studies. This marks a very different geopolitical context due to the ongoing war in Ukraine. The U.K. experiments took place shortly after Putin announced Russia's partial mobilization and amid a mass exodus of Russian men seeking to avoid conscription. The U.K. replications are described in more detail in Appendix C of the Supplementary Material. In brief, we find support for all tested hypotheses also in the U.K. context: like in the United States, U.K. respondents underestimate the victimization of men in war and perceive male victims as less innocent but are also responsive to information about male vulnerability. This strongly suggests that the study's results are not specific to the United States or to a specific moment in time.

DISCUSSION
The results of the three studies and two replications indicate that, with respect to settings of armed conflict, U.S. and U.K. respondents hold anti-male biases. First, they underestimate the scope of victimization of men in lethal violence-a form of violence that conflict researchers have established targets men Effect on Pr(Group preferred)

Preference for group of refugees
Note: The figure shows estimates of the effects of the randomly assigned group attributes on the probability of being preferred for admission to the United States. Estimates based on a linear probability model (estimated with OLS) with standard errors clustered at the respondent level. The bars represent 95% confidence intervals. The full results table is included in Table A6 in Appendix D of the Supplementary Material.
disproportionately. While there is considerable variation in the respondents' estimates, on average, 57% of adult civilian victims in the Syrian war (Study 1) and across conflicts globally (Study 3) were estimated to be men. The results across the two studies are remarkably similar, that is, we are able to replicate the findings from Study 1 in Study 3. Whether we pose the question generally or with respect to a particular conflict that has received considerable media coverage, in other words, has no bearing on respondents' estimates. We can thus be confident that we are capturing respondents' general perceptions of gendered victimization patterns in war. In Syria, available data indicate that men constitute 88.6% of adult civilian victims-the respondents thus grossly underestimate the victimization of men. In a global perspective, the 57% estimate falls pretty much exactly on the lower bound identified in existing studies, which have established a range of men being 1.3-8.9 times as likely to be killed across wars. That respondents generally underestimate the victimization of men and overestimate the victimization of women in armed conflict also substantiates similar evidence found in prior experimental research (Agerberg and Kreft 2022). Contrary to the salience of gender-essentialist discourses on victimhood and vulnerability in war that have been discussed in prior studies (Carpenter 2005;Johnson 2011), respondents are-on average-clearly aware that men constitute the majority of casualties in war. We thus do not find evidence of an overwhelming conceptual feminization of victimhood. When we inform respondents of the scale on which those targeted are male-whether in the real-world scenario of Syria or in the hypothetical vignette study -we observe anti-male biases in perceptions and attitudes, however. The victims are then perceived as less innocent, and support for providing humanitarian aid and accepting refugees decreases, compared with when no information on gendered victimization patterns is provided (Syria) or, in particular, when women are mentioned as the majority of victims (vignette study). These results reflect our theoretical expectations, developed from a vibrant body of qualitative research spanning different conflict settings across the globe, which has laid out the different ways in which male victims of armed conflict are overlooked in humanitarian responses, dismissed as legitimate victims, and treated as inherently agentic, frequently with aggressive  Note: The figure shows the treatment effect for the majority men variable for respondents in the control group and in the treatment group, respectively, as well as the difference between groups. Estimates based on a linear probability model (estimated with OLS) with standard errors clustered at the respondent level. All other coefficients are omitted from the graph. The bars represent 95% confidence intervals. The full results table is included in Table A7 in Appendix D of the Supplementary Material. and threatening connotations (Griffiths 2015;Krystalli, Hawkins, and Wilson 2017;Olivius 2016;Turner 2019).
To further probe into anti-male biases in policy preferences regarding refugee acceptance-which had been identified in different studies (Bansak, Hainmueller, and Hangartner 2016;Hainmueller and Hopkins 2015; Ward 2019)-we first replicate the results of prior research. Respondents are 11% and 33% less likely to be accepting of refugee groups with a majority of men (75% and 100% men, respectively) than they are of groups consisting only of women. Informing respondents of the vulnerability of men to lethal violencethat is, providing not just statistics on male victimization patterns, but also on the fact that civilian men are often explicitly targeted by armed actors in massacres and assassinations-thereby mitigates this anti-male bias by almost a third. This indicates also that many people lack awareness of the ways in which male civilians, many of whom have no interest in fighting in an armed conflict at all, are often specifically targeted by armed groups. Indeed, roughly four out of five respondents who received the male vulnerability treatment in the conjoint experiment indicated that this information was definitely or somewhat new to them. While we know that gendered narratives of victimhood, vulnerability and worthiness of protection are firmly entrenched in Western societies (Glick et al. 2000) and dismantling these requires considerable effort, the results in Study 3 clearly illustrate that such gendered perceptions are not fixed but malleable.
Facts-based communication, in short, has the potential to counteract both disinformation about gendered victimization patterns in war and anti-male biases that are based in gendered conceptions of victimhood, vulnerability and worthiness of protection, and support. It is worth noting, in this context, that our male vulnerability treatment used relatively sanitized language. We referenced targeting patterns only and refrained from evoking international law or moral arguments, making emotional appeals, or humanizing the victims by providing details about individual victims' lives. We have reason to believe that any of these framings might have further reduced anti-male biases. A survey experimental study among U.S. respondents thus shows that humanizing male victims of human rights violations, that is, referring to men's parenting roles and their nonviolence, results in them being perceived as more innocent (Leiby, Bos, and Krain 2021).

CONCLUSION
In line with our theoretical expectations, the results of the three experimental studies and two direct replications point to the salience of gender-essentialist notions of civilians and victims in war, which are both empirically inaccurate and normatively problematic. They produce an under-appreciation of male victimization and vulnerability and promote anti-male biases in perceptions and policy preferences. Providing respondents with more information about the vulnerability of male civilians to selective lethal violence, that is, moving beyond merely providing information on the share of male victims, reduces anti-male biases. At the same time, our results show that even providing this kind of information does not fully undo the anti-male bias that exists in public opinion. Even those respondents who received the information treatment in Study 3 still exhibit anti-male biases, which in turn depress overall support for aid and refugee acceptance, as we show in Studies 1 and 2. This raises the question of whether providing the public with more information on the vulnerability of male civilians in war might do more harm than good. As social scientists, we believe that providing empirically accurate information is generally preferable to reinforcing narratives based on stereotypes and misperceptions. Moreover, our results corroborate the rich qualitative research reviewed in the theory section that identifies the different ways in which male victims-adults as well as adolescentsface suspicion, discrimination, and neglect in humanitarian responses and within asylum and refugee regimes. Also, from an ethical standpoint, it thus becomes imperative to correct gendered misperceptions about vulnerability and victimization.
Our findings make an important contribution to the literature on public opinion toward armed conflict, but also on public opinion generally, by highlighting the salience of traditional gender conceptions and their influence on individuals' political decisionmaking, at least in the United States and the United Kingdom. While our empirical analysis focuses on armed conflict, it is conceivable that the theoretical framework we develop here plays out similarly in other issue and policy areas where gender essentialisms and gendered protection norms may be salient. These include different types of violent crime occurring domestically, or natural or humanitarian disasters at home and abroad. For instance, it would be worth exploring whether gendered perceptions of vulnerability and worthiness of protection are more pronounced in settings of armed conflict, where male-dominated combat may affect people's assessments, than in natural disasters. This could provide greater clarity on the salience of innocence and need relative to other gendered perceptions, such as those relating to agency and resilience, in driving people's policy preferences. In short, this study opens up for further (survey experimental) studies in different domestic and foreign policy issue areas concerning violence, armed conflict, or other crises.
In addition, future research could examine the nexus of type of conflict violence and the gender of those targeted in greater depth. We focus in this paper on forms of violence that conflict researchers have established predominantly target men in war, specifically lethal violence and abductions/forced recruitment. Our three experiments focus primarily on the former, as this is a type of conflict violence on which we have reasonably reliable data on gender patterns within individual conflicts and across conflicts that allow us to determine whether respondents indeed underestimate the victimization of men. However, future research on gendered perceptions of civilian vulnerability and innocence in war could extend the scope to the fuller repertoire of violence and conflictaffected harms, including sexual violence, torture, disappearances, or displacement, many of which are gendered in their manifestations and impacts. For example, this could shed light on how perceptions of male victims vary depending on whether they are victimized in a type of violence that primarily targets women (e.g., different forms of rape, sexual slavery, or other forms of sexual violence) or not. One common (mis)conception, for example, is that sexual violence is even worse for men than it is for women because most people associate this violence with women (Touquet et al. 2020). Does this translate into greater sympathy with male victims of sexual violence, as compared with other forms of violence?
Nor is gender the only factor that we can expect to have a bearing on people's perceptions of civilians and victims in war. Ethnicity, religion, class, and age could be other factors that shape how the public perceives victims of armed conflict, as prior studies on refugee acceptance indicate (Bansak, Hainmueller, and Hangartner 2016;Hainmueller and Hopkins 2015;Ward 2019). Future survey experimental research may thus extend and adapt our theoretical argument to other structurally entrenched social hierarchies. It would be interesting, for instance, to probe how racialized perceptions of victimization patterns affect public opinion regarding victimhood and vulnerability, perceptions of innocence, and policy preferences, and whether providing factual information about the vulnerability of specific groups can likewise shift any biases respondents may hold. Of particular interest in this context is also how gendered misperceptions intersect with racialized identities (Krystalli, Hawkins, and Wilson 2017;Olivius 2016). Such extensions would shed light on the significance of gender relative to, and in interaction with, other individual characteristics and structurally entrenched hierarchies in shaping perceptions and attitudes.

SUPPLEMENTARY MATERIALS
To view supplementary material for this article, please visit http://doi.org/10.1017/S0003055423000345.

DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available in the American Political Science Review Dataverse at https://doi. org/10.7910/DVN/NX0QGA. The survey experiments were pre-registered on OSF prior to data collection. The pre-analysis plans can be found at https://osf.io/ bgstz/.