Particularized Preferences for Civilian Protection? A Survey Experiment

Abstract

Even as the protection of civilians becomes a widely held norm, there is substantial variation in public support for humanitarian policy efforts. We use a survey experiment in Sweden to gain insights into this puzzle. Our survey confirms that citizens generally support military, but particularly non-military, means of civilian protection. Yet, we also find that support is partly particularized. Specifying that civilians may have ties to extremist groups (as victims or supporters) reduces support for proposals to provide humanitarian aid, contribute to UN observer missions and accept refugees. We trace this reduced support to lower moral obligation and higher threat perceptions. In contrast to expectations, respondents do not prioritize the protection of co-nationals, or women and children. Manipulation checks suggest the explanation that perceptions of who constitutes a civilian are subjective. Our findings provide insights into the domestic political determinants of atrocity prevention abroad.

Full text

Foreign Policy Analysis (2023), orad031
Particularized Pr efer ences for Civilian
Protection? A Sur vey Experiment
SOPHIA HATZ AND LISA HULTMAN
Uppsala University, Sweden
Even as the protection of civilians becomes a widely held norm, there is
substantial variation in public support for humanitarian policy efforts. We
use a survey experiment in Sweden to gain insights into this puzzle. Our
survey confirms that citizens generally support military, but particularly
non-military, means of civilian protection. Yet, we also find that support
is partly particularized. Specifying that civilians may have ties to extremist
groups (as victims or supporters) reduces support for proposals to pro-
vide humanitarian aid, contribute to UN observer missions and accept
refugees. We trace this reduced support to lower moral obligation and
higher threat perceptions. In contrast to expectations, respondents do not
prioritize the protection of co-nationals, or women and children. Manipu-
lation checks suggest the explanation that perceptions of who constitutes
a civilian are subjective. Our findings provide insights into the domestic
political determinants of atrocity prevention abroad.
Incluso cuando la protección de los civiles se convierte en una norma gen-
eralmente aceptada, existe una variación sustancial en el apoyo público
a los esfuerzos de política humanitaria. Utilizamos un experimento de
encuesta llevado a cabo en Suecia con el fin de comprender mejor
este enigma. Nuestra encuesta confirma que los ciudadanos, en general,
apoyan los medios militares de protección civil, pero que apoyan en mayor
medida aquellos medios de carácter no militar. Sin embargo, también en-
contramos que este apoyo está parcialmente particularizado. El hecho de
especificar que los civiles puedan tener vínculos con grupos extremistas
(ya sea como víctimas o como simpatizantes) reduce el apoyo a las prop-
uestas para brindar ayuda humanitaria, para contribuir a las misiones de
observación de la ONU y para aceptar refugiados. Atribuimos este menor
apoyo a la existencia de una menor obligación moral y de una mayor
percepción de amenaza. Contrariamente a lo esperado, los encuestados
no dan prioridad a la protección de sus compatriotas, ni de las mujeres
y los niños. Los controles de manipulación sugieren como explicación
para esto que las percepciones sobre quién es realmente un civil son sub-
jetivas. Nuestros hallazgos proporcionan información sobre los determi-
nantes políticos internos en materia de prevención de atrocidades en el
extranjero.
Sophia Hatz is an Associate Senior Lecturer at the Department of Peace and Conflict Research at Uppsala Univer-
sity, Sweden. Her research interests relate to state repression, human rights, and public opinion.
Lisa Hultman is a Professor at the Department of Peace and Conflict Research, Uppsala University, Sweden. Her
research focuses on international interventions, civil war dynamics, and protection of civilians.
Authors’ note : This research was funded by the Knut and Alice Wallenberg Foundation (Grant no.
2014.0162/2018.0455) and the Swedish Research Council (Grant no. 2018–00835). All data files necessary to repli-
cate the analysis presented in the article are available on Foreign Policy Analysis Dataverse. R 3.6.2 was the statistical
package used in this study. We are grateful to all colleagues who provided valuable feedback, in particular Charles Crab-
tree, Annekatrin Deglow, Kristine Eck, Linnea Gelot, Niklas Karlén, Sara Lindberg Bromley, and Gudlaug Olafsdottir.
We also want to extend a special thanks to Sophie Cassel and the whole team at LORE for a professional implementation
of the survey.
Hatz, Sophia, and Lisa Hultman. (2023) Particularized Preferences for Civilian Protection? A Survey Experiment. Foreign Policy
Analysis , https://doi.org/10.1093/fpa/orad031
©The Author(s) (2023). Published by Oxford University Press on behalf of the International Studies Association. This is an
Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs licence
( https://creativecommons.org/licenses/by- nc- nd/4.0/ ), which permits non-commercial reproduction and distribution of the
work, in any medium, provided the original work is not altered or transformed in any way, and that the work is properly cited.
For commercial re-use, please contact journals.permissions@oup.com
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

2 Particularized Pr efer ences for Civilian Pr otection?
Même si la protection des civils est une norme de plus en plus respectée,
les efforts politiques humanitaires reçoivent un soutien très inégal de la
part des populations. Nous utilisons une expérience de sondage en Suède
pour mieux comprendre cette énigme. Notre sondage confirme que les
citoyens sont généralement favorables aux modes de protection militaires
des civils, mais plus souvent aux modes non militaires. Pourtant, nous con-
statons aussi que le soutien est en partie discriminé. En précisant que les
civils peuvent avoir des liens avec des groupes extrémistes (comme vic-
times ou partisans), on observe une réduction du soutien aux propositions
d’aide humanitaire, de missions d’observation de l’ONU et d’accueil de
réfugiés. Nous relions cette diminution du soutien àun affaiblissement de
l’obligation morale et une accentuation des menaces perçues. Contraire-
ment aux attentes, les personnes interrogées n’accordent pas la priorité
aux autres ressortissants, ou aux femmes et enfants. Pour l’expliquer, la
perception de la définition d’un civil serait subjective, selon des tests de
manipulation. Nos conclusions renseignent les déterminants politiques
nationaux de la prévention des atrocités commises à l’étranger.
Introduction
The war in Syria has led to one of the worst humanitarian crises of our time.
Yet, international responses to protect civilians have been limited. This is diffi-
cult to explain given the development of strong norms such as “the Responsibil-
ity to Protect (R2P)” and the “Protection of Civilians (POC)” (e.g., Finnemore
2004 ; Bellamy 2009 ; Hunt 2019 ; Rhoads and Welsh 2019 ). On the international
level, the entrenchment of these norms coincides with an increase in humanitarian
inter ventions—inter ventions carried out with the aim to protect civilians. On the
individual level, decades of survey research shows that even though domestic sup-
port may vary with partisanship and ideology ( Hildebrandt et al. 2013 ), individuals
are broadly supportive of humanitarian interventions (e.g., Jentleson and Britton
1998 ; Eichenberg 2005 ). This support also rests on a sense of moral obligation con-
sistent with civilian protection norms (e.g., Kreps and Maxey 2018 ; Wallace 2019 ).
Given these general patterns, it is surprising that we have not seen stronger public
pressure to take action to protect civilians.
In this paper, we make two contributions to make sense of deviations from the
expectations that come from international norms and public support for the POC.
First, we examine public support for civilian protection broadly defined, includ-
ing military and nonmilitary forms of protection. We note that prior opinion re-
search has primarily measured support for state-led military humanitarian inter-
vention, which makes it difficult to separate support for the use of military force
from support for protecting civilians through other means. People may be support-
ive of doing something but perceive some types of action to be less effective toward
that end (e.g., Davies and Johns 2016 ). Second, we empirically assess whether indi-
viduals express particularized preferences for civilian protection. We draw on the
concept of “particularised protection” in UN peacekeeping ( Shesterinina and Job
2016 ) and its implications for the prioritization of “particularly vulnerable groups”
( Carpenter 2005 ) and group-based biases influencing public support for foreign
policy ( Boettcher 2004 ; Grillo and Pupcenoks 2017 ). We thus contribute to the lit-
erature that explores conditions for individual preferences for civilian protection.
For our empirical strategy, we use a population-based survey with a vignette ex-
periment. We conducted our survey in Sweden: a country with a history of support-
ing human rights and providing protection to civilians in foreign conflicts ( Doeser
2014 ). Respondents read a vignette describing the ongoing humanitarian crisis in
Syria and were then asked their support for several proposed actions Sweden could
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

SOPHIA HATZ AND LISA HULTMAN 3
take to protect civilians. The proposals span humanitarian aid, military and non-
military contributions to UN peacekeeping, and accepting refugees; allowing us to
measure support for various forms of civilian protection. By varying the identity
characteristics of the civilians in need of protection across treatment groups, we
empirically test whether support is particularized along identity lines. If such biases
exist, it could be one factor that accounts for the low public support for action in
Syria.
In line with civilian protection norms, we find that individuals are broadly sup-
portive of a variety of proposals to protect civilians in humanitarian crises. In a
close comparison of military and non-military forms of protection, we find that
respondents prefer the latter. We also find some support that civilian protection
is conditional. In line with our expectations, our results show that when civilians
are described as possibly having ties to extremist groups (as victims or supporters)
support for protection efforts is lower. Using causal mediation analysis, we trace
reduced support to a lower sense of moral obligation and increased threat per-
ceptions. However, we do not observe any effect from informing respondents that
there were women and children or Swedish citizens among those in need of protec-
tion. Examining the results of manipulation checks, we find some explanation for
these null results: respondents inferred that civilians described as Swedish citizens
were supporters of the Islamic State (IS), an extremist group. This is a plausible
inference, given that IS recruited foreign fighters, among them Swedish citizens.
Further, respondents did not seem to perceive women and children as particularly
vulnerable—as civilians or victims. This could either be due to the nature of the
conflict in Syria or a consequence of gender equality in Sweden making the gen-
dered protection norm less pronounced. Qualitatively, this demonstrates that iden-
tity characteristics do matter, but also that perceptions of who constitutes a civilian
or a victim are highly subjective.
In sum, we shed light on the complex manner in which civilian protection norms
manifest in the public and highlight important variations in public support for civil-
ian protection. Understanding the variation is critical: in democracies especially,
public support is fundamental for generating the political will to respond to hu-
manitarian crises and mass atrocities abroad.
Civilian Protection Norms
Many scholars have associated the increasing attention to humanitarian interven-
tions globally since the end of the Cold War with a changing normative context:
a growing consensus around the imperative to protect civilians suffering from
conflict-related violence (cf., Finnemore 1996 ; Thakur and Weiss 2009 ; Bellamy and
Williams 2011 ; Hunt 2019 ). This is most visible in the co-evolution of two specific
norms: R2P and POC ( Hunt 2019 ; Rhoads and Welsh 2019 ). We will refer to “civil-
ian protection norms” more broadly here, understood as the collection of norms
and normative principles related to the idea that the international community has
an obligation to offer protection to civilians suffering from harm.
Fundamentally, civilian protection norms provide a basis for third-party interven-
tion and protection efforts in the context of armed conflict and mass atrocities.
First, they invoke the responsibility of international actors to employ various in-
struments to protect civilian populations. R2P stipulates the responsibility of third
parties to intervene expediently, using military force if necessary. However, impor-
tantly, the idea of R2P has also broadened the scope of protection efforts beyond
military force to include preventative measures, diplomacy, judicial measures, eco-
nomic measures, peacekeeping operations, and capacity-building ( Bellamy 2009 ,
4). Similarly, POC mandates entail authorizing the use of force to protect civil-
ians from physical violence, but the concept of POC reflects a broader approach to
protecting civilians, for example, by enhancing compliance with international hu-
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

4 Particularized Pr efer ences for Civilian Pr otection?
manitarian law, ensuring humanitarian access, and tackling impunity ( Hunt 2019 ,
634–5).
Second, civilian protection norms establish a shared understanding that all civil-
ians warrant protection. The concept of POC within the UN relates specifically
to civilians , including specific types of non-combatants such as humanitarians,
health workers and journalists ( Hunt 2019 , 634). Further, the universality of hu-
man rights—the idea that all people have the right to freedom from violence and
human rights abuses, no matter their race, ethnicity, or other identity attributes—is
central to R2P ( Bellamy 2009 , 19; Murray 2012 , 68–9). In particular, R2P extends
to internally displaced persons (IDPs), who had previously been outside the reach
of the international community when sovereignty was viewed as a barrier to inter-
national involvement ( Bellamy 2009 , 22). As such, civilian protection norms can
be seen as indicative of a more general normative trend over time, toward the in-
clusion of civilians of all backgrounds under the rubric of protection ( Finnemore
1996 ; Grillo and Pupcenoks 2017 ).
Given civilian protection norms, we should thus expect to see an increased will-
ingness to intervene when civilians are being harmed. Studies confirm that atrocities
at least partly increase the likelihood of the UN taking action in the form of peace-
keeping or more coercive measures (e.g., Hultman 2013 ; Binder 2015 ). However,
agreement in the Security Council does not necessarily translate into public sup-
port for interventions to protect civilians, and R2P is still controversial ( Crossley
2018 ). Moreover, country-level support for R2P is influenced by domestic norms
( Negrón-Gonzales and Contarino 2014 ). Therefore, we move to the literature on
public opinion.
Public Opinion and Humanitarian Intervention
Much of what we know about public opinion on intervention in humanitarian crises
comes from studies of American public opinion. A common starting point in re-
views of previous research is Jentleson’s characterization of the American public as
“pretty prudent”: opinion on the use of military force abroad depends on the princi-
pal policy objective of the intervention ( Jentleson 1992 ; Jentleson and Britton 1998 ;
Eichenberg 2005 ). Studying public opinion of major US military interventions dur-
ing 1980–2005, Eichenberg (2005) found humanitarian intervention (providing
emergency relief in the context of a humanitarian crisis) to be one of the most
favored reasons for the use of military force abroad.
Why should the public be supportive of humanitarian intervention? Given that no
obvious national interest is at stake, support for humanitarian intervention requires
a normative motivation ( Finnemore 1996 ). Normative considerations and moral
mechanisms are often viewed in juxtaposition to instrumental motivations—such as
perceptions of costs, risks, security threats, and strategic consequences.
To date, the balance of evidence from public opinion surveys primarily points to
moral mechanisms. Eichenberg (2005) and Jentleson and Britton (1998) found that
the objective of protecting civilians outweighed other considerations such as risk, ca-
sualties, multilateral participation and security interests. Through a series of survey
experiments with US samples, Kreps and Maxey (2018) find that support for hu-
manitarian intervention rests on the perception that the United States has a “moral
obligation” to intervene to protect civilians; perceptions of costs and strategic con-
sequences were not important. Tomz and Weeks (2020) find altered perceptions of
morality to be the most important driver of support for military intervention among
US survey respondents, followed by threat perceptions. Drawing on surveys of the
British public, Davies and Johns (2016) find that individuals perceive humanitar-
ian interventions as ethical. Maxey (2020) provides suggestive evidence: mobilizing
public support for security-driven military interventions is enabled by humanitarian
narratives.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

SOPHIA HATZ AND LISA HULTMAN 5
The conclusion we draw from existing studies is that support for humanitarian
intervention is often grounded in a sense of moral obligation. While this does not
preclude the importance of other considerations, such as the prospect of success, it
does point to the relevance of normative principles.
However, previous research tends to equate humanitarian intervention with the
use of military force. Many definitions are restricted to state-led interventions using
military means (e.g., Boettcher 2004 , 332; Kreps and Maxey 2018 , 1814). Even stud-
ies that conceptualize humanitarian intervention as including non-military means
empirically measure support for the use of military force, with the goal of protecting
civilians (e.g., Jentleson and Britton 1998 ; Eichenberg 2005 , 399–400). The prob-
lem in our view is that individuals may be supportive of interventions for the pur-
pose of protecting civilians, but still disagree with the use of military force abroad.
Indeed, Davies and Johns (2016) find that although individuals think humanitarian
interventions are morally justifiable, they question their efficacy. People are likely
to have quite varying beliefs about efficacy, as a result of limited public knowledge
about the impact of foreign policy instruments. By differentiating between different
measures, we acknowledge that people may support the idea of intervention due
to moral obligation and at the same time have doubts about the efficacy of some
forms of intervention. Public support for military intervention specifically may also
depend on the foreign policy tradition of the state. For example, while unilateral
military intervention is a realistic alternative for the United States, it is not for a
smaller country like Sweden. Instead, people may support multilateral action and
non-military options. These varying preferences, however, have largely gone unmea-
sured in previous research.
1
Particularized Pr efer ences for Protection
Another possible explanation for variations in public support for civilian protection
is that while people may agree with civilian protection norms in principle, they hold
particularized preferences in terms of who they perceive as more or less deserving
of protection.
Prior research points out that humanitarian actions tend to be biased toward
certain populations and subgroups ( Finnemore 1996 ; Murray 2012 ; Shesterinina
and Job 2016 ). During the 19th century, humanitarian interventions were aimed
at protecting either interveners’ own nationals or other in-groups such as fellow
Christians ( Finnemore 1996 ). Some have argued that US interventions have been
racially biased; the United States was willing to intervene on behalf of white Kosovar
Albanians but not on behalf of black Hutu and Tutsi in Rwanda ( Boettcher 2004 ).
Within the context of peacekeeping, Shesterinina and Job (2016) conceptualize
a related phenomenon: “particularised protection.” Analyzing several UN peace-
keeping missions, the authors find that mandates are adjusted over time to pro-
vide direct protection to specific population subsets: nationals of intervening states,
UN personnel, state officials, humanitarian aid workers, and journalists. The con-
sequence, they argue, is that the majority of civilian populations effectively receive
little protection.
Other research suggests the existence of a gendered protection norm. Studying
the rhetoric used by transnational humanitarian networks in advocating for the pro-
tection of war-affected civilians, Carpenter (2005) suggests that women and chil-
dren are viewed as “particularly vulnerable groups” and are prioritized for protec-
tion. The UN Security Council is more likely to take action in response to sexual
1
One of the few studies that explores support for peacekeeping shows that American public support for peacekeep-
ing for humanitarian purposes is high ( Burk 1999 ) but does not compare that to other measures. Lyon and Malone
(2010) compare support for peacekeeping in the United States and ten European countries to support for military ac-
tion for other policy goals. However, they do not evaluate support for different responses for the purpose of protecting
civilians.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

6 Particularized Pr efer ences for Civilian Pr otection?
violence, which is often assumed to target women ( Hultman and Johansson 2017 ;
Benson and Gizelis 2020 ). Carpenter (2005) argues that this contradicts the prin-
ciple that civilians (non-combatants) are the population who warrant protection.
Given the nature of many conflicts today, the prioritization of women and children
can end up including some combatants (adult female combatants) while exclud-
ing some non-combatants (adult civilian men) from protection (see also Bond and
Sherret 2012 ).
In the literature on public opinion and humanitarian intervention, there is rela-
tively little prior research on whether individuals exhibit particularized preferences
for protection. Some evidence of racial, religious and gender bias has been found
in studies with US samples. Grillo and Pupcenoks (2017) find that Americans are
more willing to support a military humanitarian intervention in the Syrian civil war
when the affected civilians are described as Christian, compared to Muslims. The
authors also find that priming respondents to consider the suffering of women and
children increases support for military humanitarian intervention, but only among
female survey respondents. Similarly, Agerberg and Kreft (2022) find that Amer-
icans are more supportive of intervention when women, compared to civilians in
general, are victims of torture. In a survey experiment with a US student sample,
Boettcher (2004 , 346) finds indicative evidence that the “race/ethnicity/religion”
of an endangered population affects support for military intervention in humani-
tarian crises.
Outside the literature on humanitarian intervention, identity-based biases have
been found in studies of public tolerance of human rights violations (e.g., Mondak
and Hurwitz 2012 ; Piazza 2015 ; Conrad et al. 2018 ; Carriere et al. 2019 ; Hatz 2021 ).
A general insight is that individuals are more tolerant of human rights abuses di-
rected toward members of out-groups; specifically, groups that are perceived as
threatening, disruptive, or violent are those who are considered legitimate targets
of civil and human rights abuses. For example, several studies conducted with US
samples have found that individuals are more permissive of human rights abuses
directed at members of extremist groups, insurgent groups, Arabs, and Muslims
( Piazza 2015 ; Conrad et al. 2018 ; Wallace 2013 ). Similar results have also been found
with samples of Israeli Jews ( Shamir and Sagiv-Schifter 2006 ; Lupu and Wallace
2019 ), Indians and Argentinians ( Lupu and Wallace 2019 ), Germans ( Asbrock and
Fritsche 2013 ), and Swedes ( Lindén et al. 2018 ).
Naturally, there is a tension between the argument that civilian protection norms
motivate support for civilian protection and the argument that people may prior-
itize the protection of certain population subgroups. The universality of human
rights and the principle of civilian immunity are central to both R2P and POC, so
particularized protection and the prioritization of particularly vulnerable groups
seem to contradict the core value that all civilians warrant protection. Yet, the very
definition of “civilian” is complex and perhaps subjective considering the nature
of many conflicts and humanitarian crises today, in which insurgent or extrem-
ist groups are closely intertwined with civilian populations. Ultimately, particular-
ized preferences may be defined as a bias toward protecting subgroups perceived
as civilians—as innocent, victimized, and vulnerable—while excluding subgroups
perceived as culpable or threatening.
Drawing on our notion of particularized preferences for protection, we expect
that support for civilian protection varies according to the identity characteristics of
the civilians in need of protection. Based on the discussion above, we identify three
such relevant characteristics.
First, we expect that individuals are more supportive of protection proposals aim-
ing to protect citizens of their own country. This could be due to in-group favoritism
or due to an assumption that co-nationals in a humanitarian crisis abroad are civil-
ians.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

SOPHIA HATZ AND LISA HULTMAN 7
H1: Individuals are more supportive of civilian protection including the protec-
tion of citizens of their own country.
Second, we expect individuals to be more supportive of protection proposals aim-
ing to protect women and children, since these are likely to be perceived as “partic-
ularly vulnerable groups.”
H2: Individuals are more supportive of civilian protection including the protec-
tion of women and children.
Third, we expect individuals to become less supportive of protecting civilians who
have ties to armed extremist groups; these are likely to be perceived as threatening
or culpable.
H3: Individuals are less supportive of civilian protection including the protection
of civilians affiliated with extremist groups.
With regard to each of our hypotheses, our underlying expectation is that identity
characteristics serve as cues that allow individuals to update their perceptions of
the extent to which the civilians in question are innocent, victimized, vulnerable,
and deserving of protection. Yet, identity characteristics may be imperfect cues. For
example, we have noted that the assumption that women and children are victims of
conflict is not universal; meaning that this cue may not always increase perceptions
of vulnerability. In a similar vane, the affiliation with an extremist group could signal
guilt and threat—in the case of civilians being voluntary and active supporters—or
innocence and vulnerability—in the case of coerced recruitment of civilians. In our
empirical analysis, we will probe our implicit assumptions regarding how identity
cues map onto perceptions.
Research Design
To test our hypotheses, we use a population-based survey experiment: an exper-
iment within a survey of a representative population sample ( Mutz 2011 ). This
method offers the advantage of maximizing the potential for both internal and
external validity. As in a traditional experiment, random assignment in a survey ex-
periment allows for the estimation of treatment effects in isolation from pre-existing
differences across survey respondents. Additionally, by employing a representative
population sample, the results of the experiment can be generalized to the larger
population.
We conducted the survey in Sweden, a country where civilian protection norms
are internalized in foreign policy. Sweden has a strong identity of internationalism
and commitment to global justice ( Bergman Rosamond 2016 ). Yet, even in Sweden,
there is variation in public support for civilian protection, providing a plausible set-
ting for our investigation of “particularised preferences.” A few recent studies pro-
vide interesting comparisons of public attitudes in Sweden to other Western states.
Swedes have more positive attitudes toward refugees compared to people in Bel-
gium, France, and the Netherlands ( Coninck 2020 ). Support for humanitarian aid
is slightly higher in Sweden as compared to the United States but lower than in the
United Kingdom ( Dumitrescu and Bucy 2021 ). Support for military intervention
in response to widespread violence is similar in Sweden as compared to the United
Kingdom, and slightly higher than in the United States ( Agerberg and Kreft 2023 ).
All in all, this suggests that Sweden is similar to many Western countries, but with a
slightly higher expectation to see norms of protection expressed in public attitudes,
making it a hard test for our hypotheses. Regarding the gendered protection norm,
Sweden has a tradition of promoting gender equality and was the first country to
formally introduce the concept of feminist foreign policy. At the same time, other
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

8 Particularized Pr efer ences for Civilian Pr otection?
surveys show that Swedes are less prone to favor humanitarian responses when chil-
dren or sexual violence is highlighted ( Dumitrescu and Bucy 2021 ; Agerberg and
Kreft 2023 ). For the hypothesis about extremist groups: due to the general support
for protection and equal treatment, Sweden could provide a hard test. However, it
is perhaps more likely that Sweden is similar to other countries in Western Europe,
given the attention to the risk of terrorism and the role of foreign fighters in politics
and media all over Western Europe when the war in Syria escalated.
The survey was fielded by the Laboratory of Opinion Research (LORE) at
Gothenburg University, during December 3, 2020–January 5, 2021. A total of 2,149
Swedish citizens, a subset of LORE’s web-based Citizen Panel, participated in our
study.
2 Prior to collecting the data, we obtained ethical approval from the Swedish
national ethical board (dnr. 2020–04,408). We registered our survey design, hy-
potheses, and pre-analysis plan (PAP) on OpenScience Forum (OSF).
3 Details on
sampling, randomization procedure and other aspects of the survey and experi-
ment design can be found in our PAP. At the start of our survey, respondents were
presented with information about the purpose of the study, their role as voluntary
participants, and the procedure in place for ensuring confidentiality, anonymity,
and protection of personal data.
4
In the vignette experiment, respondents were randomly assigned to receive one
of six possible vignettes describing the ongoing humanitarian crisis in Syria.
5 All
vignettes contained the following background information, formulated based on
information available to the public on the websites of the Swedish International De-
velopment Cooperation Agency (SIDA) and Sweden’s Ministry of Foreign Affairs
6
:
The conflict in Syria is in its tenth year and has led to the world’s largest refugee
crisis. Over 6 million civilians are internally displaced within Syria and 5,6 million
have fled to neighboring countries. The situation for civilians in Syrian refugee camps
is critical: they suffer from extreme poverty, starvation, disease, violence and serious
human rights abuses. The Syrian government has not provided basic public services
such as medical care, water, electricity, sanitation, and protection from harm.
With this background held constant, the vignettes differ only in the identity
characteristics of the Syrian civilians. Specifically, we varied additional information
which places emphasis on the civilians as Swedish citizens , women and children , or af-
filiated with IS . We combined these three identity factors to construct five different
treatment conditions.
7 The control condition received no additional information
about the civilians’ identity. Table 1 shows the text corresponding to the treatment
condition in each of the six vignettes.
We then measured our outcome variable: support for civilian protection. We used
a multi-item question in which respondents were presented with several proposals
for how Sweden could respond to the crisis. Respondents were asked to rate each of
these proposals on a numerical 5-point scale with labeled end-points, ranging from
“(1) very bad proposal” to “(5) very good proposal.” The question items, shown
2
Information about LORE, its Citizen Panel, and its sampling methodology can be found at https://lore.gu.se .
3
See https://osf.io/pra52 . This paper focuses on three of the four hypotheses listed in the pre-analysis plan.
4
Our survey was inserted into a longer survey: LORE’s December 2020 wave of the Citizen Panel. We only describe
the questions in our survey here.
5
Respondents were also randomly assigned to a priming experiment. We do not describe the design of this exper-
iment or its additional survey questions here, but we control for treatment assignment to the priming experiment in
our analyses.
6
SIDA: https://www.sida.se/sida- i- varlden/lander- och- regioner/syrien , accessed June 19, 2020, and Sweden
Ministry of Foreign Affairs: https://www.government.se/international- development- cooperation- strategies/2015/12/
swedens- regional- strategy- for- the- syria- crisis/ , accessed June 24, 2020 (both websites have been updated since we ac-
cessed them).
7
To clarify our language here: we use “factor” to refer to the three different identity characteristics we vary: Swedish
citizens, women and children, and IS affiliation. We use “treatment condition” to refer to the factor or the combination
of factors in a single vignette.For example, the treatment condition in vignette 5 is “Swedish citizens” and the treatment
condition in vignette 6 is “Swedish citizens + women and children” (see table 1 ).
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

SOPHIA HATZ AND LISA HULTMAN 9
Tab le 1. Survey experiment treatment conditions and text
Vignette Treatment condition Vignette text
1 Control [no additional information]
2 IS affiliation Among the civilians in need, some have ties to the Islamic
State. It is difficult to determine who has been a victim of
the Islamic State and who supports the Islamic state.
3 Women and children Among the civilians in need, many are women and young
children.
4 Women and
children + IS affiliation
Among the civilians in need, many are women and young
children. Some have ties to the Islamic State. It is difficult to
determine who has been a victim of the Islamic State and
who supports the Islamic state.
5 Swedish citizens Among the civilians in need, there are Swedish citizens.
6 Swedish
citizens + women and
children
Among the civilians in need, there are Swedish citizens.
Many are women and young children.
Tab le 2. Outcome variables and text
Outcome variable Question item text
Humanitarian aid Increase humanitarian aid to Syrian refugee camps.
UN observers Contribute unarmed observers to a UN peacekeeping operation.
UN police Contribute police to a UN peacekeeping operation.
UN military troops Contribute military troops to a UN peacekeeping operation.
Accept refugees Accept more Syrian refugees into Sweden.
Repatriation Return Swedish citizens in Syrian refugee camps to Sweden.
in table 2 , correspond to a range of different forms of civilian protection efforts,
including humanitarian aid, UN peacekeeping, and aiding refugees. We further
distinguished between different kinds of personnel contributions to a UN opera-
tion: unarmed observers, police, and military troops ( Kathman 2013 ). This allows
for a close comparison of military and non-military forms of civilian protection.
8
For respondents assigned to vignettes 5 and 6 (which include the Swedish citizen
identity factor), we additionally asked about a proposal corresponding to repatria-
tion. Because support for repatriation is only measured for a subset of respondents,
we refer to the first five items listed in table 2 as our primary outcome variables.
We chose to use a real-world crisis in our vignettes, rather than an abstract sce-
nario. This choice presents a trade-off between experimental control and general-
izability ( Brutger et al. 2023 ). On the one hand, the use of a real-world crisis allows
us to control the background attributes of the crisis by either fixing or random-
izing information. A risk with abstract scenarios is that respondents infer specific
information (such as the location or type of conflict)—these inferences are both
correlated with respondent characteristics and unmeasured. On the other hand, a
real-world crisis may come with lower generalizability, since the crisis in Syria differs
from past and ongoing crises. Recent methodological work, however, has shown
that increasing the level of abstraction in a survey vignette, e.g., by describing a
situation as hypothetical, using artificial actors or unnamed countries, offers no ad-
8
We specifically construct these items as proposals that involve Sweden “contributing” to UN efforts for two reasons.
First, this allows us to hold constant whether there is multilateral cooperation, since prior research shows that public
support for humanitarian intervention depends on whether the intervention is multilateral ( Wallace 2019 ). Second,
this form of response is much more realistic for Sweden compared to hypothetical unilateral intervention.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

10 Particularized Pr efer ences for Civilian Pr otection?
vantage over the use of real-world scenarios ( Brutger et al. 2023 ). We reflect further
on generalizability as we draw conclusions about the findings.
Our primary concern with our use of a real-world crisis is that Swedish respon-
dents may hold pre-existing beliefs about the case, and these pre-existing beliefs
are not randomly assigned. In particular, respondents may have pre-existing beliefs
about civilians in Syrian refugee camps—e.g., the extent to which they are victims of
atrocities, targets of coercive recruitment into IS, or voluntary IS supporters; these
labels feature prominently in media stories and political debates. Although we deal
with this problem by randomly assigning civilians’ identity factors, we do anticipate
that pre-existing beliefs could still cause problems.
First, pre-existing beliefs could make respondents less sensitive to information
treatments. Second, pre-existing beliefs could cause respondents to make subjective
interpretations about vulnerability and culpability; making the identity treatments
imperfect cues. Third, respondents could draw on pre-existing beliefs in order to
make inferences about other background factors, which are correlated with the
information treatments and the outcome ( Dafoe et al. 2018 ). We employ a manipu-
lation and placebo check question to assess the extent to which pre-existing beliefs
cloud or confound our treatments. Directly after the outcome variable, we asked
a question that measures the treatment respondents received, that is: respondents’
perceptions of the identity of the civilians in the crisis description. This question
allows us to validate whether the experiment manipulated perceptions of identity
as intended, and to assess if the treatments caused respondents to make inferences
about other identity characteristics which we did not manipulate.
We then measured two possible mediating variables identified in prior research:
moral obligation and threat perceptions. To encourage respondents to answer in
reference to the specific vignette they read, we repeated the vignette on the same
page as each of the mediating variable questions. We also collected data on re-
spondents’ demographic characteristics: gender, age, country of birth, education,
left-right political ideology, and political party preference. We use these as pre-
treatment covariates for balance tests and adjustments in regression models.
9 A
sample survey including the full text of the vignettes and all survey questions is
given in Online Appendix C.
Results
In line with our expectations about Sweden as a case, our survey of the Swedish pub-
lic confirms that individuals are generally supportive of the protection of civilians
in the context of humanitarian crises, and also support a variety of different forms
of civilian protection. Figure 1 shows the average levels of support for each of the
six protection proposals we asked about.
There are a few interesting patterns. First, respondents generally preferred
non-military forms of civilian protection. Humanitarian aid was the most favored
(mean = 4.03, sd = 0.87), followed closely by contributing unarmed observers
to a UN peacekeeping operation (mean = 3.98, sd = 0.85). Consistent with
prior opinion studies on military humanitarian intervention, respondents also sup-
ported the proposal to contribute military troops to a UN peacekeeping operation
(mean = 3.62, sd = 1.10). However, support for contributing military troops was sig-
nificantly lower compared to support for contributing unarmed UN observers; the
95 percent confidence intervals do not overlap. This suggests to us that the ques-
tion of military force is important to separate out. Also of interest is that accepting
9
Gender, age, and country of birth were drawn from LORE’s prior surveys of the Citizen panel. Education, political
party preference, and political ideology were measured in our survey, after a block of questions in LORE’s longer
Citizen panel survey. Although our covariate measures are post-treatment, we argue they are unlikely to be affected by
the treatment and can be treated as pre-treatment covariates.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

SOPHIA HATZ AND LISA HULTMAN 11
Figure 1. Support for civilian protection across six proposals.
Support is measured on a five-point scale, where higher numbers correspond to greater
support. Means and 95 percent confidence intervals are plotted over the distribution of
responses.
refugees stands out as being relatively unpopular, with more respondents rating this
as a bad proposal (mean = 2.63, sd = 1.43).
Although there was variation in levels of support across the measures, an ex-
ploratory factor analysis suggests the questions measure the same underlying con-
cept.
10
We therefore constructed a composite index (which we call the POC index)
using the five main outcome measures, which account for the relative importance
of each measure. The POC index is the factor score of the variables, rescaled to a
five-point scale (mean = 3.6, sd = 0.35).
Protection for Whom?
Given respondents’ generally positive predisposition toward civilian protection, did
they also express particularized preferences—preferences that vary with the identity
characteristics of those in need of protection?
To answer this question, we turn to the vignette experiment. The independent
variables in the experiment are dummy variables indicating a comparison of two
groups of respondents assigned to vignettes that differ along one identity factor.
Each dummy independent variable is coded 1 for the identity factor of interest and
0 otherwise. For example, to test Hypothesis 1, we use a dummy variable coded 1
if the respondent was assigned to vignette 5 or 6 (which both include the Swedish
citizen’s identity factor) and coded 0 if the respondent was assigned to vignette 1, 2,
3, or 4 (which do not include this identity factor). Figure 2 shows the results of the
experiment used to evaluate our three hypotheses.
Looking at the top two rows of figure 2 , there is little support for Hypotheses 1
and 2. Mentioning that there are Swedish citizens among those in need of protec-
tion appears to increase support for the proposal that Sweden contribute unarmed
observers to a UN peacekeeping mission in Syria, but this effect is not statistically sig-
10
Cochran’s alpha, a measure of internal reliability, is greater than 0.65 for the five measures of civilian protection.
See Online Appendix A, Section 2.2.1 for details.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

12 Particularized Pr efer ences for Civilian Pr otection?
Figure 2. Effect of identity factors on support for civilian protection.
Support is measured on a five-point scale, where higher numbers correspond to greater
support. Estimates and 95 percent confidence intervals are calculated using robust lin-
ear regression. Average treatment effects (Difference in mean support) are estimated
including mean-centered pre-treatment covariates. P -values (except for the POC index)
are adjusted using the Holm’s correction.
nificant after applying a Holm’s adjustment to correct for multiple comparisons.
11
There are also no significant effects from informing respondents that many civilians
were women and children, to respondents assigned to other vignettes not including
this information.
Looking at the bottom row of plots, we do find support for Hypothesis 3. Pro-
viding the information that the civilians in need have ties to the IS significantly
reduced support for the proposals the Sweden increase humanitarian aid to Syria,
contribute with unarmed observers, and accept more Syrian refugees. The effect of
IS-affiliation is also negative and significant for the POC index. It is interesting to
note that IS affiliation affects only support for non-armed forms of civilian protec-
tion; it is possible that respondents’ preferences regarding the deployment of the
Swedish military or police are more rigid.
12
We were surprised to find no effect from informing respondents that Swedish cit-
izens or women and children were among those in need of protection. In Online
Appendix B, we explore whether the null results could be due to the pooling of
treatment conditions and reference categories in our comparisons. We compare
each treatment to the control group only and conduct additional difference of
means comparisons across pairs of treatment conditions. This allows us to assess
interaction effects—whether the effect of one identity factor depends on the pres-
ence of another. The results remain the same for citizens and women and children,
when compared individually to the control. While we do not identify any signifi-
cant interaction effects, the findings underscore the importance of IS affiliation:
11
We apply the Holm’s correction to the group of tests run for each hypothesis. The correction is applied to the
vector P -values from N tests where N is number of independent variables tested (1) times the number of outcome
variables (5). We do not include the p-value of the POC index in the adjustment, as the index produces a single test of
the five outcome variables.
12
Full results of regression models are reported in Online Appendix B. As we measure our outcome variables us-
ing numerical 5-point scales, we use linear regression. The results are consistent when we instead estimate logistic
regressions with dichotomous versions of the outcome variables and ordinal logistic regressions with ordered categor-
ical outcome variables. All models include mean-centered pre-treatment covariates, but we also show that the results
hold when using unadjusted models. Section 5 of Online Appendix B shows the results of sub-group analysis for het-
erogenous treatment effects. We generally do not find any meaningful patterns, although we note that the effect of
women and children treatment remains insignificant among female respondents, in contrast to the findings in Grillo
and Pupcenoks (2017) .
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

SOPHIA HATZ AND LISA HULTMAN 13
Figure 3. Factual manipulation checks.
The dependent variables are dummy variables coded 1 if the respondent received the
treatment factor assigned and 0 otherwise. Estimates and 95 percent confidence inter-
vals are calculated using robust linear regression.
even when respondents are informed that many of the civilians in need are women
and children, mentioning that some have ties to IS significantly reduces support for
protection efforts.
Perceived Identity
Our primary concern with using a real-world humanitarian crisis in the vignette ex-
periment was that respondents’ pre-existing beliefs about the identity characteristics
of Syrian civilians in refugee camps could cloud or confound the effects of the iden-
tity treatments. To assess the extent to which this is problematic, we incorporated
a question that measured the treatment respondents received, that is: respondents’
perceptions of the identity of the civilians in the crisis description.
We first used this question to conduct factual manipulation checks: to assess if re-
spondents correctly perceived civilian identity according to the identity treatment
factor assigned ( Gerber and Green 2012 ; Kane and Barabas 2019 ). In figure 3 , the
top row of plots indicates the presence of two-sided non-compliance in each of the
comparison groups. For example, among the respondents assigned to the women
and children identity factor, some correctly answered that there were women and
children among the civilians (they received the treatment) while other respondents
did not. However, the extent of non-compliance is negligible for each identity factor,
and in the bottom row of figure 3 , we consistently observe a significant difference
in the probability of correctly receiving the treatment, given the identity factor as-
signed. We can therefore rule out non-compliance as a cause of the null results, and
we gain confidence that the difference in support for civilian protection comparing
IS affiliation to otherwise is due to respondents’ perception of civilians as affiliated
with IS.
We gain a bit more insight into the null findings on Hypothesis 1 by assessing if
the information treatments caused respondents to perceive the Syrian civilians as
civilians, combatants, victims of IS, or supporters of IS. These can be thought of
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

14 Particularized Pr efer ences for Civilian Pr otection?
Figure 4. Subjective manipulation checks .
The dependent variables are dummy variables coded 1 if the respondent perceived the
identity characteristic of interest and 0 otherwise. Estimates and 95 percent confidence
intervals are calculated using robust linear regression.
as subjective manipulation checks (
Kane and Barabas 2019 ), since the treatment
received is not necessarily factually “correct” with respect to treatment assignment.
The results of these tests are shown in figure 4 .
Importantly, the Swedish citizen’s treatment significantly reduced the perception
that the persons in need of protection were civilians. It is likely that respondents
assumed that Swedish citizens were in Syrian refugee camps because they voluntar-
ily joined IS, although other possible assumptions would be that they were civilians
kidnapped or coerced to join IS. Indeed, the Swedish citizen’s treatment also sig-
nificantly increased the tendency to describe the Syrian refugees as supporters of IS
and significantly decreased the tendency to describe them as IS victims, even while
the vignette text specified both possibilities. These perceptions could have reduced
support for civilian protection in the same manner as IS affiliation, possibly cancel-
ing out any tendency to favor fellow citizens.
We also gain a bit of insight into the lack of a women-and-children effect (Hy-
pothesis 2). In contrast to our expectations, the assignment to the women and chil-
dren identity factor did not coincide significantly with the perception of persons in
the crisis description as civilians or victims. This suggests that women and children
may not have been judged as particularly innocent or vulnerable. Lastly, assignment
to the IS-affiliation identity factor significantly increased perceptions that the indi-
viduals were combatants, which helps to explain the overall negative effect of the
treatment condition (in line with Hypothesis 3).
We conducted placebo tests in order to assess another problem related to re-
spondents’ pre-existing beliefs. That is, the treatments could cause respondents to
change their beliefs about other background factors, creating a risk that these per-
ceptions (which are not randomly assigned) drive differences in outcome ( Dafoe et
al. 2018 ; Eggers et al. 2023 ). This can be thought of as hidden confounding ( Samii
2011 ) or a violation of the information equivalence assumption in survey experi-
ments ( Dafoe et al. 2018 ).
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

SOPHIA HATZ AND LISA HULTMAN 15
Figure 5. Placebo tests.
The dependent variables are dummy variables coded 1 if the respondent perceived the
identity characteristic of interest and 0 otherwise. Estimates and 95 percent confidence
intervals are calculated using robust linear regression.
Using the same question measuring perceived identity, we assessed whether the
information treatments caused respondents to interpret the civilians as “Christians”
or “Muslims.” We chose religion as a placebo attribute because religion is likely cor-
related with the identity characteristics we varied, and prior research suggests that
the religious traits of civilian populations affect support for humanitarian interven-
tion ( Boettcher 2004 ; Grillo and Pupcenoks 2017 ). The results, shown in figure 5 ,
show that there is no significant difference in the probability respondents perceived
the civilians as Christians or Muslims, across any of the comparison groups in our
hypothesis tests. This gives us confidence that the negative effect of IS affiliation on
support for civilian protection is not due to any confounding effect of Religion.
13
Causal Mechanisms
Our manipulation checks help us to understand the null results on hypotheses 1
and 2: identity cues can be imperfect and perceptions of who constitutes a “civilian”
may be subjective. We also gained some insight into the reason for a dampening
effect of IS affiliation; IS-affiliated civilians were likely to be perceived as combatants.
To further understand how identity characteristics of civilians in need influence
support for protection, we sought to measure the importance of threat perceptions
and moral obligation as causal mechanisms. These are two considerations found to
be important in prior research ( Larson 1996 ; Tomz and Weeks 2020 ). Drawing on
the wording in Tomz and Weeks (2020) survey experiment, we asked respondents
whether Sweden had a strong or weak moral obligation to intervene in the crisis
we described in the vignette. In order to capture whether the civilians described
in the vignette were perceived as threatening, we asked about the likelihood they
might pose a security risk in countries where they settle. Both questions are coded
on numerical five-point scales with labeled end-points.
We assessed the causal pathways connecting the independent variables in our
hypothesis tests, the mediating variables, and the outcome variables through medi-
ation analysis ( Imai et al. 2010 ; Tingley et al. 2014 ). As a first step, we examined
the effect of civilian identity factors on each of the mediating variables. Next, we
estimated the effect of the mediating variables on each of the outcome variables,
13
Additional tests of design, including balance tests, for question ordering effects and systematically missing data
are reported in Online Appendix A.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

16 Particularized Pr efer ences for Civilian Pr otection?
controlling for each of the independent variables in turn. Finally, we estimated av-
erage causal mediation effects: the indirect effect of the independent variables on
the outcome through the mediating variable ( Imai et al. 2010 ). In all three steps,
we controlled for pre-treatment covariates.
Figure 6 summarizes the main results of the mediation analysis. From the first
step of the analysis (estimates are denoted with squared-shaped points), we find
that associating the civilians with IS significantly lowered respondents’ views that
Sweden has a moral obligation to intervene, while it significantly increased respon-
dents’ perception that the civilians pose a security risk.
14 In the second step (de-
noted with round points), we find that both of these considerations, in turn, shape
respondents’ attitudes toward civilian protection. Looking down the left column,
higher levels of moral obligation significantly increase support for all five proposals
for civilian protection, and this is also reflected in the POC index. In the right-hand
column, we also see that higher levels of threat perceptions reduce support for each
proposal for civilian protection, with the exception of proposals involving UN po-
lice and UN military troops. In the final step of the mediation analysis, we link the
two causal chains to assess the extent to which the effect of the independent variable
travels through the mediating variables. This is primarily relevant for IS affiliation,
the independent variable that produced a statistically effect direct effect. The aver-
age causal mediation effect is statistically significant here for all outcome measures,
meaning that respondents’ lowered sense of moral obligation contributed to lower
support for civilian protection among respondents who were informed that some
civilians had ties to IS. Threat perceptions also moderate the impact of IS affiliation,
but to a lesser extent; playing no significant role in support for civilian protection
via UN police and UN military troops.
In the analyses above, we are unable to tease out the importance of the expected
efficacy of various measures. In our Online Appendix B, we therefore explore a
number of additional considerations suggested in prior research: expectations of
success, risk, cost, and benefit (e.g., Jentleson and Britton 1998 ; Eichenberg 2005 ;
Eckles and Schaffner 2011 ; Davies and Johns 2016 ; Kreps and Maxey 2018 ; Tomz
and Weeks 2020 ). Because of the question formulation, we can only assess these
mechanisms with regard to the UN military troops outcome variable. Results of that
extended mediation analysis indicate that expectations about success and benefits
are relevant for support for UN troops, but these effects are not conditioned by the
identity characteristics of the victims.
Discussion
Our survey sheds light on the complex manner in which civilian protection norms
manifest in the public. In line with internationally and nationally established norms,
Swedish survey respondents are broadly supportive of proposals to protect civilians
in humanitarian crises. There is, however, some variation across different forms of
civilian protection: support for contributing military troops is significantly lower
compared to similar non-military proposals.
We find some support for the idea of particularized preferences. In the context of
our vignette experiment, respondents become significantly less supportive of civil-
ian protection when the civilians in need are specified as affiliated with the extrem-
ist group IS. From our mediation analysis, we find that specifying civilians’ IS affil-
iation lowered respondents’ sense of moral obligation to intervene and increased
perceptions that the civilians in need pose a security threat. These considerations,
in turn, contributed to lower support for civilian protection.
14
In figure 6, the coefficient for the effect of an identity factor on a mediator (denoted with a square-shaped point)
is the same across the outcome variables, we include it in each plot for reference.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

SOPHIA HATZ AND LISA HULTMAN 17
Figure 6. Mediation analysis: moral obligation and threat perceptions.
Outcome variables and mediating variables are coded on five-point scales. Estimates
and 95 percent confidence intervals for the effect of identity factors, and the effect
of mediators are calculated using robust linear regression. Estimates and 95 percent
confidence intervals for average causal mediation effects are estimated via the Quasi-
Bayesian Monte Carlo method (1000 simulations) using robust standard errors. All mod-
els include mean-centered pre-treatment covariates. Estimates with significant P -values
( P < 0 . 05) are colored red. For ACMEs (except for the POC index) P -values have been
adjusted with the Holm’s correction.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

18 Particularized Pr efer ences for Civilian Pr otection?
In contrast to our expectations, we did not observe any effect from informing re-
spondents that there were particularly vulnerable groups—women and children—
or co-nationals—Swedish citizens—among those in need of protection. The results
of subjective manipulation checks shed some light on possible reasons for these
null results. Respondents assigned to the Swedish citizen identity factor were less
likely to describe the Syrian refugees in the crisis vignette as civilians or IS victims,
and more likely to describe them as supporters of IS. Hence, the non-finding is less
about Swedes adhering to the R2P norm and more about threat perceptions that
in this particular case are linked to the many Swedish citizens found among foreign
fighters in Syria. The subjective manipulation checks also showed that respondents
assigned the women and children identity factor did not necessarily perceive the
Syrian refugees in the crisis vignette as particularly vulnerable—as civilians or vic-
tims, as we would expect based on the gendered protection norm.
15 It is possible
that the gendered protection norm is less pronounced in Sweden due to its rela-
tively high gender equality, something suggested also by Agerberg and Kreft (2022) .
Outside the hypothesis tests, there are several other interesting results. First, while
the forms of civilian protection we measured were diverse—spanning humanitar-
ian aid, UN peacekeeping and accepting refugees—exploratory reliability analysis
suggests they belong to the same underlying concept. Preferences for civilian pro-
tection can be measured as a whole, and we chose to do this by creating a POC
index. At the same time, some forms of civilian protection are more sensitive to
variations in civilian identity than others: IS affiliation only affected support for
non-military/armed means of protection: humanitarian aid, contributing unarmed
observers, and accepting refugees. We infer that attitudes toward the use of military
or police forms are less sensitive to changes in context.
Second, it appears that the different considerations we measured as mediating
variables shape preferences toward different forms of civilian protection in differ-
ent ways. Moral obligation appears to be the most important consideration; it influ-
ences preferences for all forms of civilian protection. Threat perceptions are also
influential, but not when it comes to preferences on the contribution of police
or military troops to UN missions. The clear role of moral obligation in influenc-
ing preferences is consistent with our underlying expectation that individuals are
guided by normative principles as articulated in civilian protection norms.
Conclusions
Intervention for the purpose of protecting civilians is at once one of the most ac-
ceptable forms of third-party intervention and one of the most debated topics in
international politics ( Fixdal and Smith 1998 ). We hope that the findings of our
survey with Swedish citizens will illuminate how individuals in the public form pref-
erences about different forms of civilian protection, and in this way contribute to
our understanding of domestic political debates. We make three key contributions
to prior research.
First, our study measures public opinion on a variety of different forms of civil-
ian protection, including humanitarian aid, peacekeeping operations, accepting
refugees, and the repatriation of citizens. While much prior research focuses on
the use of military force, our survey separates the question of intervention from the
question of military force, as people may be in favor of a humanitarian response
but also think that military intervention is unlikely to enhance civilian protection
or is too costly. Contrasting non-military and military forms of civilian protection,
15
This can be compared to the findings by Getmansky et al. (2018) who even found a negative effect of informing
people that Turkey’s refugee policy has saved innocent women and children on their support for refugees. Their inter-
pretation is that people’s views toward women and children were influenced by the perceived negative impact they had
on Turkish society through, for example, increased begging on the streets and increased divorce rates as Turkish men
married Syrian women.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

SOPHIA HATZ AND LISA HULTMAN 19
we found overall weaker support for the latter. Future research should explore these
issues further; in particular, how people view efficacy versus moral obligation when
comparing different types of responses.
Second, we studied whether biases in civilian protection exist among individuals
in the public. We found that, to a certain extent, individual-level preferences for
protection are particularized. When civilians are perceived as having ties with ex-
tremist groups, support for protection drops. We have also highlighted how percep-
tions of who is a civilian are highly subjective, shedding light on apparent tensions
with core principles of civilian protection norms such as the universality of human
rights.
Third, the study brings evidence from a new empirical context. Prior research
on public opinion and humanitarian intervention predominantly uses US samples,
lending little insight into whether existing findings are generalizable to different
cultures and national contexts. We believe that Swedish public opinion is particu-
larly relevant, as Sweden has and continues to play an important role in UN- and
EU-led missions.
In addition to these contributions to research, we aim to generate knowledge
that is relevant for international advocacy groups and policymakers. For example,
our survey reveals that just mentioning that civilians may have ties to IS—even
while specifying that they could be victims or supporters—reduces support for many
forms of civilian protection. This should be an important insight for humanitarian
groups and foreign policymakers, as it suggests that public is sensitive to framing
and messaging concerning humanitarian crises. This is relevant also for the media
or any influential actor issuing statements on crises and potential interventions.
While we tested our hypotheses about particularized preferences in the specific
context of Swedish opinion on the refugee crisis in Syria, we believe the patterns we
observe generalize to other populations and similar crises. Outside of Sweden, many
countries are reluctant to aid civilians in Syrian refugee camps.
16 The crux of the
debate often revolves around civilians’ ties to IS: for example, obligation versus the
potential risks of repatriating citizens. The question of what to do becomes more
complex considering women and children, whose combat roles remain unclear.
17
Beyond Syria, recent analyses of the ongoing humanitarian crisis in Ukraine (2022)
suggest that the perception of Ukrainian refugees as ’civilized’ and “European” has
been a salient factor influencing the positive response across European countries.
18
We encourage future research to probe the extent to which particularized prefer-
ences guide international responses to humanitarian crises.
Supplementary data
Supplementary information is available at the Foreig n P ol icy Ana lysis data archive.
16
Peltier, Erin and Constant Méheut. 2021. “Europe’s Dilemma: Take in ISIS Families, or Leave them in Syria?” The
New York Times . May 21. Accessed 31 May, 2022. https://www.nytimes.com/2021/05/28/world/europe/ isis-women-
children-repatriation.html .
17
Margolin, Devorah and Austin C. Doctor. 2022. “Thousands of Men, Women and Children Remain in
Detention because of their Former Ties to ISIS.” The Washington Post . February 2. Accessed 31 May, 2022.
https://www.washingtonpost.com/politics/2022/02/02/thousands-men-women-children-remain-detention-because-
their -former -ties-isis/.
18
Abdelaaty, Lamis. 2020. “European countries are welcoming Ukrainian refugees. It
was a different story in 2015.” The Washington Post . March 23. Accessed 31 May, 2022.
https://www.washingtonpost.com/politics/2022/03/23/ukraine-refugees-welcome-europe/.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

20 Particularized Pr efer ences for Civilian Pr otection?
References
AGERBERG , MATTIAS , AND ANNE KATHRIN KREFT . 2023. “Sexual Violence, Gendered Protection and Support
for Intervention.” Journal of Peace Research 60 (5): 853–67
ASBROCK , FRANK , AND IMMO FRITSCHE . 2013. “Authoritarian Reactions to Terrorist Threat: Who is Being
Threatened, the Me or the We?” International Journal of Psychology 48 (1): 35–49.
BELLAMY , ALEX J. 2009. Responsibility to Protect: The Global Effort to End Mass Atrocities. Cambridge: Polity
Press.
BELLAMY , ALEX J. , AND PAUL D. WILLIAMS . 2011. “The New Politics of Protection? Côte d’Ivoire, Libya and
the Responsibility to Protect.” International Affairs 87 (4): 825–50.
BENSON , MICHELLE , AND THEODORA-ISMENE GIZELIS . 2020. “A Gendered Imperative: Does Sexual Violence
Attract UN Attention in Civil Wars?” Journal of Conflict Resolution 64 (1): 167–98.
BERGMAN ROSAMOND , ANNIKA. 2016. “Swedish Internationalism and Development Aid.” In The Oxford Hand-
book of Swedish Politics, edited by Jon Pierre, 462–78. Oxford: Oxford University Press.
BINDER , MARTIN. 2015. “Paths to Intervention: What Explains the UN’s Selective Response to Humanitar-
ian Crises?” Journal of Peace Research 52 (6): 712–26.
BOETTCHER , WILLIAM A. 2004. “Military Intervention Decisions Regarding Humanitarian Crises: Framing
Induced Risk Behavior.” Journal of Conflict Resolution 48 (3): 331–55.
BOND , JENNIFER , AND LAUREL SHERRET . 2012. “Securing Consistency for Consisent Security: Gender and the
Responsibility to Protect.” Chap. 13. In The Routledge Handbook of the Responsibility to Protect, edited by
W. Andy Knight and Frazer Egerton, 166–80. New York: Routledge.
BRUTGER , RYAN , JOSHUA D. KERTZER, JONATHAN RENSHON, DUSTIN TINGLEY, AND CHAGAI M. WEISS . 2023. “Ab-
straction and Detail in Experimental Design.” American Journal of Political Science 67: 979–995
BURK , JAMES. 1999. “Public Support for Peacekeeping in Lebanon and Somalia: Assessing the Casualties
Hypothesis.” Political Science Quarterly 114 (1): 53–78.
CARPENTER , R. CHARLI . 2005. “‘Women, Children and Other Vulnerable Groups’: Gender, Strategic Frames
and the Protection of Civilians as a Transnational Issue.” International Studies Quarterly 49 (2): 295–334.
CARRIERE , KEVIN R. , MARGARET J. HENDRICKS, AND FATHALI M. MOGHADDAM . 2019. “Sophisticated but Scared:
The Effects of Political Sophistication, Right-Wing Authoritarianism, and Threat on Civil Liberty
Restrictions.” Analyses of Social Issues and Public Policy 19 (1): 256–81.
CONINCK , DAVI D DE . 2020. “Migrant Categorizations and European Public Opinion: Diverging Attitudes
towards Immigrants and Refugees.” Journal of Ethnic and Migration Studies 46 (9): 1667–86.
CONRAD , COURTENAY R. , SARAH E. CROCO, BRAD T. GOMEZ, AND WILL H. MOORE . 2018. “Threat Perception and
American Support for Tortu r e .” Political Behavior 40 (4): 989–1009.
CROSSLEY , NOELE. 2018. “Is R2P Still Controversial? Continuity and Change in the Debate on ‘Humanitar-
ian Intervention’.” Cambridge Review of International Affairs 31 (5): 415–36.
DAFOE , ALLAN , BAOBAO ZHANG, AND DEVIN CAUGHEY . 2018. “Information Equivalence in Survey Experi-
ments.” Political Analysis 26 (4): 399–416.
DAVI ES , GRAEME A. M. , AND ROBERT JOHNS . 2016. “R2P from Below: Does the British Public View Humani-
tarian Interventions as Ethical and Effective?” International Politics 53 (1): 118–37.
DOESER , FREDRIK. 2014. “Sweden’s Libya Decision: A Case of Humanitarian Intervention.” International
Politics 51 (2): 196–213.
DUMITRESCU , DELIA , AND ERIK P. BUCY . 2021. “Visual Portrayals of People in Need: The Impact of Refugee
Depictions, Compassion, and Support for Humanitarian Aid.” International Journal of Public Opinion
Research 33 (4): 818–39.
ECKLES , DAVID L. , AND BRIAN F. SCHAFFNER . 2011. “Risk Tolerance and Support for Potential Military Inter-
ventions.” Public Opinion Quarterly 75 (3): 533–44.
EGGERS , ANDREW C. , GUADALUPE TUÑÓN, AND ALLAN DAFOE . 2023. “Placebo Tests for Causal Inference.” Amer-
ican Journal of Political Science . 1–16.
EICHENBERG , RICHARD C. 2005. “Victory Has Many Friends: US Public Opinion and the Use of Military
Force, 1981–2005.” International Security 30 (1): 140–77.
FINNEMORE , MARTHA. 1996. “Constructing Norms of Humanitarian Intervention.” In The Culture of National
Security: Norms and Identity in World Politics, edited by Peter Katzenstein, vol. 153. New York: Columbia
University Press.
———. 2004. The Purpose of Intervention: Changing Beliefs about the Use of Force. Ithaca, NY: Cornell Univer-
sity Press.
FIXDAL , MONA , AND DAN SMITH . 1998. “Humanitarian Intervention and Just War.” Mershon International
Studies Review 42 (Supplement 2): 283–312.
GERBER , ALAN S. , AND DONALD P. GREEN . 2012. Field Experiments: Design, Analysis and Interpretation, E dited by
Ann Shin. New York: W .W . Norton /Company, Inc.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

SOPHIA HATZ AND LISA HULTMAN 21
GETMANSKY , ANNA , TOLGA SINMAZDEMIR, AND THOMAS ZEITZOFF . 2018. “Refugees, Xenophobia, and Domestic
Conflict: Evidence from a Survey Experiment in Turkey.” Journal of Peace Research 55 (4): 491–507.
GRILLO , MICHAEL C , AND JURIS PUPCENOKS . 2017. “Let’s Intervene! But Only If They’re Like Us: The Ef-
fects of Group Dynamics and Emotion on the Willingness to Support Humanitarian Intervention.”
International Interactions 43 (2): 349–74.
HATZ , SOPHIA. 2021. “What Shapes Public Support for Tort u r e, and Among Whom?” Human Rights Quar-
terly 43 (4): 683–98.
HILDEBRANDT , TIMOTHY , COURTNEY HILLEBRECHT, PETER M. HOLM, AND JON PEVEHOUSE . 2013. “The Domestic
Politics of Humanitarian Intervention: Public Opinion, Partisanship, and Ideology.” Foreign Policy
Analysis 9 (3): 243–66.
HULTMAN , LISA. 2013. “UN Peace Operations and Protection of Civilians: Cheap Talk or Norm Implemen-
tation?” Journal of Peace Research 50 (1): 59–73.
HULTMAN , LISA , AND KARIN JOHANSSON . 2017. “Responding to Wartime Sexual Violence: UN Peacekeeping
and the Protection Agenda.” Global Responsibility to Protect 9 (2): 129–46.
HUNT , CHARLES T. 2019. “Analyzing the Co-Evolution of the Responsibility to Protect and the Protection
of Civilians in UN Peace Operations.” International Peacekeeping 26 (5): 630–59.
IMAI , KOSUKE , LUKE KEELE, AND DUSTIN TINGLEY . 2010. “A General Approach to Causal Mediation Analysis.”
Psychological Methods 15 (4): 309–34.
JENTLESON , BRUCE W. 1992. “The Pretty Prudent Public: Post post-Vietnam American Opinion on the Use
of Military Force.” International Studies Quarterly 36 (1): 49–74.
JENTLESON , BRUCE W. , AND REBECCA L. BRITTON . 1998. “Still Pretty Prudent: Post-Cold War American Public
Opinion on the Use of Military Force.” Journal of Conflict Resolution 42 (4): 395–417.
KANE , JOHN V , AND JASON BARABAS . 2019. “No Harm in Checking: Using Factual Manipulation Checks to
Assess Attentiveness in Experiments.” American Journal of Political Science 63 (1): 234–49.
KATHM AN , JACOB D. 2013. “United Nations Peacekeeping Personnel Commitments, 1990–2011.” Conflict
Management and Peace Science 30 (5): 532–49.
KREPS , SARAH , AND SARAH MAXEY . 2018. “Mechanisms of Morality: Sources of Support for Humanitarian
Intervention.” Journal of Conflict Resolution 62 (8): 1814–42.
LARSON , ERIC V. 1996. Casualties and Consensus: The Historical Role of Casualties in Domestic Support for U.S.
Military Operations. Santa Monica, CA: Rand Corporation.
LINDÉN , MAGNUS , FREDRIK BJÖRKLUND, AND MARTIN BÄCKSTRÖM . 2018. “How a Terror Attack Affects Right-
wing Authoritarianism, Social Dominance Orientation, and their Relationship to To r ture Attitudes.”
Scandinavian Journal of Psychology 59 (5): 547–52.
LUPU , YONA T AN , AND GEOFFREY P. R. WALLACE . 2019. “Violence, Nonviolence, and the Effects of International
Human Rights Law.” American Journal of Political Science 63 (2): 411–26.
LYON , ALYNNA , AND MARY FRAN MALONE . 2010. “Where are the ’Good Samaritans’? Assessing Cross-National
Support for Peacekeeping Operations.” Journal of International Peacekeeping 13 (3-4): 239–66.
MAXEY , SARAH. 2020. “The Power of Humanitarian Narratives: A Domestic Coalition Theory of Justifica-
tions for Military Action.” Political Research Quarterly 73 (3): 680–95.
MONDAK , JEFFERY J. , AND JON HURWITZ . 2012. “Examining the Terror Exception: Terrorism and Commit-
ments to Civil Liberties.” Public Opinion Quarterly 76 (2): 193–213.
MURRAY , ROBERT W. 2012. “The Challenges Facing R2P Implementation.” Chap. 5. In The Routledge Hand-
book of the Responsibility to Protect, edited by W. Andy Knight and Frazer Egerton, 64–76. New York:
Routledge.
MUTZ , DIANA C. 2011. Population-based Sur vey Experiments. Princeton, NJ: Princeton University Press.
NEGRÓN-GONZALES , MELINDA , AND MICHAEL CONTARINO . 2014. “Local Norms Matter: Understanding National
Responses to the Responsibility to Protect.” Global Governance: A Review of Multilateralism and Interna-
tional Organizations 20 (2): 255–76.
PIAZZA , JAMES A. 2015. “Terrorist Suspect Religious Identity and Public Support for Harsh Interrogation
and Detention Practices.” Political Psychology 36 (6): 667–90.
RHOADS , EMILY PADDON , AND JENNIFER WELSH . 2019. “Close Cousins in Protection: The Evolution of Two
Norms.” International Affairs 95 (3): 597–617.
SAMII , CYRUS. 2011. “Hidden Confounding in Survey Experiments,” A ccessed December 7, 2021, http:
//cyrussamii.com/?p=613 .
SHAMIR , MICHAL , AND TAMMY SAGIV-SCHIFTER . 2006. “Conflict, Identity, and Tolerance: Israel in the Al-Aqsa
Intifada.” Political Psychology 27 (4): 569–95.
SHESTERININA , ANASTASIA , AND BRIAN L. JOB . 2016. “Particularized Protection: UNSC Mandates and the Pro-
tection of Civilians in Armed Conflict.” International Peacekeeping 23 (2): 240–73.
THAKUR , RAMESH , AND THOMAS WEISS . 2009. “R2P: from Idea to Norm—and Action?” Global Responsibility to
Protect 1 (1): 22–53.
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023

22 Particularized Pr efer ences for Civilian Pr otection?
TINGLEY , DUSTIN , TEPPEI YAMAMOTO, KENTARO HIROSE, LUKE KEELE, AND KOSUKE IMAI . 2014. “Mediation: R
Package for Causal Mediation Analysis.” Journal of Statistical Software 59 (5): 1–38.
TOMZ , MICHAEL R. , AND JESSICA L. P. WEEKS . 2020. “Human Rights and Public Support for War.” The Journal
of Politics 82 (1): 182–94.
WALLACE , GEOFFREY P. R. 2013. “International Law and Public Attitudes Towa r d Tort u re: An Experimental
Study.” International Organization 67 (1): 105–40.
———. 2019. “Supplying Protection: The United Nations and Public Support for Humanitarian Inter-
vention.” Conflict Management and Peace Science 36 (3): 248–69.
Hatz, Sophia, and Lisa Hultman. (2023) Particularized Preferences for Civilian Protection? A Survey Experiment. Foreign Policy Analysis ,
https://doi.org/10.1093/fpa/orad031
©The Author(s) (2023). Published by Oxford University Press on behalf of the International Studies Association. This is an Open Access
article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs licence
( https://creativecommons.org/licenses/by- nc- nd/4.0/ ), which permits non-commercial reproduction and distribution of the work, in
any medium, provided the original work is not altered or transformed in any way, and that the work is properly cited. For commercial
re-use, please contact journals.permissions@oup.com
Downloaded from https://academic.oup.com/fpa/article/20/1/orad031/7453166 by guest on 30 November 2023