Let’s Intervene! But Only If They’re Like Us: The Effects of Group Dynamics and Emotion on the Willingness to Support Humanitarian Intervention

ABSTRACT International relations (IR) studies on humanitarian intervention have debated both the nature and strength of intervention norms. This article contributes to this debate by exploring under what conditions individuals are willing to support military humanitarian intervention (MHI) and the psychological factors that influence whether, and the degree to which individuals support MHI. Taking a psychological approach, we hypothesized that individuals’ decision to support MHI is influenced by in-group favoritism and emotional responses to in-group suffering. We tested our theory with two experiments, each of which recruited roughly 200 American participants. Both experiments centered on the ongoing Syrian civil war and assessed Americans’ willingness to support intervention to protect different civilian groups. The results suggested that support for intervention was widespread, but not a majority view in most cases. The findings also suggested that participants exhibited slightly higher rates of support for intervention when those suffering were Christian, as opposed to Muslim. Furthermore, we found that the dynamics of support for intervention changed when chemical weapons were introduced into the scenario, which reframed the the crisis as a national security issue. Overall, our results suggest that individuals’ decisions to act upon norms can be influenced by the context of a crisis and individual level psychological factors, which have been under explored in IR scholarship on norms.

American participants' attitudes about whether the United States should militarily intervene in Syria to stop human rights abuses. The first experiment (E1) focuses on the Syrian regime's targeting of civilians. The second experiment (E2) is identical to the first, with the exception of adding the use of chemical weapons against civilians to the list of human rights abuses, which serves to reframe the crisis as a security issue.
This study provides substantive contributions to research on norms and IR theory. First, we empirically test under what conditions people are more likely to support military humanitarian intervention (MHI) and how the context of a crisis shapes attitudes about the scope of intervention. Second, we seek to examine how psychological factors such as in-group favoritism and emotional responses to in-group suffering influence normative behavior and the degree to which norms are acted upon. This is particularly important because as critics have noted, recent IR scholarship has tended to downplay or completely reject the role that psychological processes play in influencing whether individuals comply with norms (Rosati 2000). An exploration of such mechanisms can be extremely useful in that they can help scholars understand the factors that influence the extent to which individuals and collectives adhere to various social norms (Shannon 2012). Third, we demonstrate how the experimental method, which is still underutilized in IR scholarship, can be a valuable tool for examining the determinants of agents' normative behavior, as it allows researchers to explore the psychological dimensions of decision making in a way that other methodologies cannot. Previously, IR scholars have tended to make assertions about the underlying causes of behavior from observing external actions alone, which can be misleading since there can be an array of motivations behind any given behavior. Indeed, IR scholarship could benefit from examining the oftenneglected individual level and the causal role that identity, preexisting beliefs, and emotions play in the decision-making process. Experimental methods are a powerful tool for examining such causal mechanisms.
The article is structured as follows. First, we discuss current trends in research on MHI. Second, we discuss how in-group favoritism and emotional responses to in-group suffering can influence the decision-making process and provide researchers with a richer understanding of the motivations behind normative behavior. Third, drawing from psychology research, we develop a theory to explain the public's willingness to support MHI. We posit that individuals will be much more supportive of MHI when the targets of violence are members of their own group. This occurs because individuals experience a strong emotional response when confronted with the suffering of fellow in-group members. In the case of our experiment, we contend that Americans will be more likely to support MHI to save Christians, given America's predominately Christian heritage and culture. Fourth, we test our theory with two experiments. For each experiment, we recruited roughly 200 participants through Amazon Mechanical Turk (MTurk). The treatments for both experiments consist of realistic vignettes about the targeting of civilians in the current Syrian civil war. Examining the data with analysis of variance (ANOVA) and mediation analysis, the results of the first experiment confirmed our theoretical model, while the second partially did so. Lastly, we conclude with a discussion of the findings and their implications for IR research. We argue that many studies have painted an incomplete picture of the process of normative action because of their dismissal of both underlying psychological processes and methodological individualism.
Overall, we find that participants were more willing to support intervention when those suffering are Christians (as opposed to Muslims), which challenges arguments found in some liberal and constructivist research that contends that the scope of MHI norms has evolved to the degree that all people, regardless of race and religion, are believed to be worthy of MHI and not just White Christians, as it was a century ago. However, the results of both experiments suggest that public support for MHI may be widespread but is in most cases not a majority view. Furthermore, the findings suggest that context matters, as the dynamics of support changed from E1 to E2, where (after the addition of chemical weapons use by the Syrian government) MHI was only supported by those respondents who most likely viewed the conflict as a potential security issue to the United States. Moreover, the findings also surprisingly suggest that even highly educated liberals-who generally tend to support humanitarian intervention-largely oppose it.

Humanitarian intervention and public support
The purpose of the current study is to develop a better understanding of the factors that prompt mass support for MHI. In this section, we review relevant IR studies on humanitarian intervention and key studies about the impact of public opinion on MHI and the relationship between public opinion and elite decision making. We discuss how these studies have influenced our experiments, as well as how our experiments contribute to these scholarly discussions.
The first set of questions our experiments test is whether people believe that a grave humanitarian crisis warrants intervention and what factors influence that decision. On one hand, a longitudinal analysis of public opinion polls suggest that the American public has shown notable support for US military involvement abroad for a number of decades (Kull 2002). On the other hand, Jentleson and Britton (1998) analyze American public opinion on the use of military force in the post-Cold War period and conclude that the American public was less interested in foreign policy in the decade following the end of the Cold War and was less in favor of using military force with regards to humanitarian actions abroad. Likewise, Eichenberg (2005:141) argues that both the articulated objective of a US military mission abroad and the resulting outcome (success or failure) are important determinants of public support for military actions; and public support increases when military interventions are perceived to be successful (regardless of the level of casualties). Our experiments contribute to this discussion by assessing how the circumstances of the crisis (whether conventional or chemical weapons are used) and emotional responses to atrocities influence individuals support for MHI.
A second set of questions we test is whether people believe that humanitarian interventions should be conducted on a multilateral basis and whether the United Nations should play a role in that process. Since the end of the Cold War, it has been increasingly agreed upon that a legitimate humanitarian intervention has to be done on a multilateral basis, usually with UN Security Council (UNSC) authorization (Finnemore 1996:180-185;Pattison 2010). For example, Nicholas Wheeler's (2000:8) seminal study found that a new norm of UN-authorized humanitarian intervention emerged in the 1990s. Similarly, Schultz (2003:108-109) finds that during the post-Cold War period American presidents used these norms and the institutions that support them to overcome Congressional resistance and increase public support for humanitarian intervention abroad (due to the added legitimacy and promises of burden sharing that working multilaterally through intentional organizations offer). However, Welsh (2006:3) notes that there is still a contentious debate about whether UNSC authorization is necessary for intervention. We aim to both engage with this quantitative literature on multilateralism and public opinion (for example, see Grieco, Gelpi, Reifler, and Feaver 2011) and-more importantly-to contribute to IR research on norms, which tends to be less quantitative in nature.
Third, our experiments test whether people believe that the motivation for intervention by their country should be driven by the perceived selfinterests of their state, which in the current study is the United States. Critics have questioned whether humanitarian intervention has actually been driven by egalitarian norms, as opposed to state interests and other pragmatic motivations. For example, Butler (2005) uses a series of regression analyses to show that American and Canadian military interventions during the Cold War were only selectively employed in instances when such military actions supported those countries' broader foreign policy goals. Aydin (2012) uses both quantitative and qualitative evidence to argue that economic ties between the potential intervening power and the belligerent state increase the likelihood of intervening. Looking at state-level factors, Koch and Sullivan (2010) suggest that the domestic political climate within democratic states affects their ability and willingness to become involved in military interventions abroad. Based on the aforementioned evidence, critics assert that norms of humanitarian intervention have been selectively applied and that decisions to engage in intervention have been based on self-interest and other pragmatic considerations. Our study contributes to this discussion by again moving beyond elites and testing scholars' assertions at the mass level.
Lastly and most importantly, we test the assertion that beliefs regarding the scope of intervention have expanded to include people of all racial and religious backgrounds. Numerous IR studies address the question of who should be protected when conducting humanitarian intervention. Finnemore (1996Finnemore ( , 2003 argues that while the notion of humanitarian intervention is not a recent development, the scope of and justifications for such actions have changed over time. The most notable change has been in regards to which human beings merit intervention (Finnemore 1996:159). Finnemore contends that during the nineteenth century, the key reasons for humanitarian intervention dealt with the protection of Christians from Ottoman Turks (1996:161). Later, due to the abolitions of slavery, decolonization, and the spread of the idea of self-determination, egalitarian movements swept across Europe, and the understanding of when a humanitarian intervention is warranted greatly expanded. As a result, humanitarian intervention since 1945 has been almost exclusively on behalf of non-Christians and/or non-Europeans. While Finnemore states that these norms are merely permissive, and thus do not necessarily prescribe intervention, our study contributes to this debate by testing Finnemore's assertion via experimental manipulation of the group membership of the victims. Given that in-group/outgroup dynamics tend to be influenced by psychological biases, we contend that Finnemore's assertions of inclusiveness in regarding who should be protected may be overstated.
We contend that an examination of mass-level attitudes toward humanitarian intervention is warranted for the following reasons. First, in addition to engaging with IR literatures on MHI, we also aim to contribute to the substantive quantitative literature on US public opinion on intervention. As noted earlier, much of the public opinion research on intervention tends to approach the decision to intervene from an outcomes-oriented perspective that focuses on public support for certain mission objectives and the role that public perceptions about the success of specific operations plays in mass desire to support humanitarian missions. We seek to contribute to this discussion by examining the impact of nonrational psychological factors such as in-group favoritism and emotional responses to ingroup suffering.
Second, while it has been demonstrated that elites can shape public opinion on foreign policy issues (Brody 1991:166-167;Recchia 2015)-with some going so far as to argue that the masses have little influence on the elite opinions (Zaller and Chiu 2000)-several studies have shown that American foreign policy decision making can indeed be influenced by popular opinions on given foreign policy issues (Aldrich, Gelpi, Feaver, Reifler, and Sharp 2006;Berinsky 2007;Burstein 2003). For example, one study found that increased public pressure prompted Congress to support a number of humanitarian interventions in the 1990s (Hildebrandt, Hillebrecht, Holm, and Pevehouse 2013). Furthermore, as Van Der Meulen and Soeters (2005) show, a lack of public support for particular foreign policy initiatives can enhance reluctance within the government to embrace them. Baum and Groeling (2010:12) suggest that presidents who lose public support for their foreign policy initiatives face notable difficulty sustaining them.
Third, given that by definition a norm is a standard of behavior that is adhered to by the vast majority of a society, one can make the case that norms can be measured through public opinion. Such an approach can give scholars a more detailed understanding of the degree to which norms are adhered to and the factors that contribute to differing degrees of adherence. As Kowert (2012) observes, norms are based on societal beliefs that establish a sense of what is appropriate. These beliefs, which constitute the foundation of norms, can be supported and adhered to within a society in varying degrees. While some segments may strongly abide by a particular norm, others may not (Kowert 2012). In the following section, we explain our theoretical framework regarding the factors that shape attitudes about whether to intervene, as well as the scope of intervention. We theorize that our respondents will be more likely to support intervention in instances where they perceive that the individuals suffering are relatively similar to them.

Group dynamics, emotion, and support for intervention
With regards to the scope of intervention, building on findings from social and cognitive psychology, we contend that individuals' decisions about whether humanitarian intervention should be pursued by their state rely heavily upon individual-and group-level psychological variables. These variables include individuals' group identity (which can be comprised of race, religion, ethnicity, etc.) and the tendency of people to be more sensitive to the suffering of those who are similar to them. Drawing on this research, we put forth and test two hypotheses regarding how in-group/out-group dynamics and emotion influence mass-level perceptions about whether to intervene, as well as the scope of intervention.
First, while Americans in the past have supported MHI-and certainly humanitarian assistance-to non-Christians (Lyon and Malone 2010), we hypothesize that individuals will be significantly less willing to support intervention when the populations suffering are perceived to be members of an out-group (for example, Muslims). We base this assertion on social and cognitive psychology research on intergroup bias, which has empirically demonstrated that people have a tendency to prefer members of their own group, dislike members of perceived out-groups, and not empathize with out-group members' suffering (Avenanti, Sirigu, and Aglioti 2010;Cikara, Bruneau, and Saxe 2011). Thus, we contend that American participants (most of whom are Christian or come from Christian backgrounds) will be less likely to support intervention for non-Americans and non-Christians (for example, Muslims).
Our second hypothesis is that individuals will experience a negative emotional response when reading about the suffering of members of their own identity group (for example, Christians), which will in turn prompt a willingness to support action to assist said group. We derive this hypothesis from research in psychology, which suggests that emotions play a determining role in the decision-making process. For example, studies have found that strong negative emotions prompt people to behave impulsively, experience optimism, and be more willing to take risks (Cyders and Smith 2008;Lerner and Keltner 2001). Building on these findings, we argue that people's decisions about whether to intervene for the purposes of humanitarian intervention are based on strong emotional reactions based on their affinity with the victims of human rights abuses. Additionally, we contend that the willingness to support military intervention in another country after encountering stimuli about in-group member suffering is an example of an emotional and impulsive response-one that does not take into consideration the logistics that such an action would require or the costs and benefits to the state.
Utilizing insights from these different strands of research, we contend that affinity with the group facing human rights abuses will influence individuals' emotional responses to a humanitarian crisis. Specifically, we argue that our American participants will experience a negative emotional response when reading about the suffering of Christians, which will prompt them to support a hypothetical military action. Conversely, they will not experience an emotional response and exhibit less of a willingness to intervene when Muslims are the victims of violence. Our theoretical model is detailed in Figure 1.

Methodology
We tested our theoretical model with two experiments that were conducted prior to American involvement in Syria and President Obama's September 13, 2013, speech stating that the United States will take action in Syria if Assad's regime were to use chemical weapons (E1 on April 30, 2013 and E2 on September 1, 2013). The experimental method is appropriate here because it allows us to assess people's immediate responses to stimuli, and the causal effects of said stimuli, while controlling for extraneous variables. Statistically, we analyzed our data with mediation analysis, which is a form of causal path modeling widely used in psychology because it allows researchers to determine the degree to which a mediating variable (that is, negative affect) facilitates the relationship between the independent (experimental treatments) and dependent (support for intervention) variables (Hayes 2013;MacKinnon 2008). Thus, for present purposes, experimentation coupled with mediation analysis is a powerful methodological tool because it allows us to test our posited causal chain.
To test the degree to which, and under what conditions, individuals support an inclusive and intercommunal enforcement of MHI norms, we exposed participants to a real news report about the current Syrian conflict. With the exception of slight modifications for each experimental condition to focus on a different victim group (Christians, women and children, or Muslims), we used the article verbatim. We believe that our use of a real-world scenario is appropriate and does not bias results for a number of reasons. First, a common criticism of experiments is that they have low external validity because the treatments and tasks utilized in most psychological studies are not reflective of the kinds of situations and tasks that people would be exposed to, or asked to do, in the real world. The best way to overcome this problem is to utilize treatments or tasks that are reflective of what one would encounter in the real world (Abrahms 2013:666-667). Second, while it is possible that some participants may follow world news or have an existing knowledge of the Syrian conflict, random assignment greatly reduces the effects of these and other extraneous variables because it makes the distribution of such variables equal across all experimental groups. Furthermore, we believe that previous knowledge of world affairs or the Syrian conflict would not be an issue because Americans tend not to be very informed about foreign affairs (Aldrich, Sullivan, and Borgida 1989;Bennett, Flickinger, Baker, Rhine, and Bennett 1996). Lastly, we are confident about the external validity of our results because, as we will later discuss, our overall findings are consistent with trends found in previous studies and recent public opinion polls.

Participants
Approximately 200 American participants were recruited for each experiment through MTurk, which is an online marketplace that scholars in many fields are now using to conduct human subjects research. 1 Each participant was compensated $0.50 through MTurk for completing the study. MTurk's reliability as a participant pool has been demonstrated through the successful replication of many classic social science experiments, which suggests that the samples obtained through the marketplace produce results comparable to those obtained through traditional recruitment methods (Berinsky, Huber, and Lenz 2012).
Since MTurk workers volunteer to participate in research studies as opposed to being randomly selected, the samples obtained from the marketplace are not representative and are therefore not appropriate for conducting public opinion surveys that seek to make inferences about the distribution of certain traits or opinions in the population as a whole. Furthermore, our sample cannot offer insights into elite opinions and behaviors. However, we contend that these limitations are not problematic for the current study for a number of reasons. First, the main goal of this study is not to make inferences about where the American public stands on humanitarian intervention but rather to establish the causal impact that psychological variables (for example, in-group favoritism and emotional responses to in-group suffering) have on the decision to support intervention within our sample of Americans. Second, representative samples are not necessary to conduct experiments because that methodology employs random assignment, which as noted earlier can control for the effects of a variety of extraneous variables. Furthermore, the fact that our samples, like many MTurk samples, skew liberal (Berinsky et al. 2012) makes it more difficult to obtain significant results for our hypothesized model because that group tends to be more supportive of human rights and MHI (Western 2002). Therefore, significant counterintuitive findings from a liberal sample can only bolster our case. Lastly, recent research in foreign policy and leadership strongly suggests that the decision-making process of elites, like those of the ordinary people in our study, are heavily influenced by bias and emotion (Mintz and DeRouen 2010;Renshon and Renshon 2008).

Procedure
The experiments were conducted through SurveyMonkey, the link for which was embedded in MTurk. The design of the first experiment was as follows: Participants first completed a pretest survey eliciting demographic information, religious affiliation, religiosity, and political views. Next, SurveyMonkey randomly assigned participants to one of four conditions. In three of the conditions, participants received a treatment that 1 When conducting experiments, the general practice is to have a minimum of 30 participants per condition, which we exceed. Furthermore, our N size is more than adequate for the statistical procedures we use (for example, ANOVA requires a minimum of 15 cases per group and regression 30 cases total). consisted of a slightly modified UN News Centre report detailing gross violations of human rights in the Syrian civil war, specifically the killing of civilians (UN News Centre 2012). The articles used in the three experimental conditions were identical, except for the group membership of the targeted civilians (condition 1 = Christians, condition 2 = women and children, condition 3 = Muslims). Condition 4 served as a control, where participants did not read an article and continued to the posttest survey. The following is an excerpt from the treatment (group frames are italicized): Syria has been wracked by violence, with an estimated 19,000 people, mostly Christians/women and children/Muslims, killed since the uprising against President Bashar al-Assad began some 18 months ago. Violations conducted by Government death squads include murder, summary executions, torture, arbitrary arrest and detention, rape and pillaging and destruction of civilian objectsincluding hospitals and schools (UN News Centre 2012).
After reading one of the treatment articles, participants completed a posttest survey that assessed their emotional response to the article, the extent to which they supported US intervention in Syria, and under what conditions that intervention should occur (for example, if other states also commit, if it benefits America's security interests, etc.). We believe that the treatment used in our study was framed in a way that our subjects could infer that an intervention in Syria would be short-term and military in nature.
E2's design was identical to E1, with the exception of adding the use of chemical weapons to the list of atrocities to assess if support for intervention would change. The rationale for conducting E2 was that it allowed us to investigate whether increasing the level of human suffering and reframing the crisis as a national security issue involving the violation of an international "taboo" of nonuse of chemical weapons would impact support for MHI. Therefore, the inclusion of chemical weapons into the experiment is important because the results can have important implications for IR theory and practice. First, Saddam Hussein's suspected proliferation of weapons of mass destruction (WMDs) was among the main reasons why the United States invaded Iraq in 2003. Interestingly, after we conducted our experiments, the Obama administration noted that the Syrian government's use of chemical weapons would signify traversing a "red line" that the United States would not accept, which could prompt US action or even military intervention (Baker and Gordon 2012). Second, numerous and well-institutionalized international treaties against the use of chemical weapons exist, which arguably make the use of such weapons a "taboo" (see Price 1997). Therefore, in our second experiment, we test whether our subjects would be more likely to endorse MHI if a well-established norm against the use of chemical weapons has been violated.

Experiment 1
Demographics A total of 204 subjects participated in E1. The demographics of the participants are as follows. The majority of participants were male (63.2%). Participants' ages ranged from 18 to 66 years of age (M = 32). In regards to race, 77.8% identified as "White alone." Religious affiliation was divided between Christians (42.7%) and agnostics/atheists (43.1%). Ideologically, participants identified as "slightly liberal" on the ANES scale (M = 3.19). In regards to education, 56.9% of participants had earned an undergraduate degree or higher.

Variables
The dependent variable was support for US intervention in Syria, which we gauged with six items, each of which were measured on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). We employed these different measures of intervention to establish under which conditions our participants would be willing support intervening on behalf of different victim groups. Participants were asked the extent to which they agree or disagree with the following statements, "America should intervene in Syria to … The independent variable was the experimental treatment, the values for which included 1 (Christian frame), 2 (women and children frame), 3 (Muslim frame), and 4 (control). The mediating variable was negative affect, which was measured with a composite scale comprised of four items taken from PANAS-X, which is an emotion instrument widely used in psychology (Watson and Clark 2008). Immediately after reading the treatment article, participants were given a list of positive and negative emotions and were asked the extent to which they were feeling each emotion (for example, sad, distressed, angry, and disgusted) "at the present moment." Responses ranged from 1 (very slightly or not at all) to 5 (extremely). The composite measure, which was unidimensional and reliable (varimax rotated factor loadings ≥ .84, α = .909), was created by summing scores for the four items (M = 9.07).
We also examined a number of covariates. We gauged religiosity with the RCI-10, which measures religious commitment in 10 areas (for example, religious reading, influence of religious beliefs in daily life, volunteering, financial contributions, etc.) (Worthington, Wade, Hight, Ripley, McCullough, Berry, Schmitt, Berry, Bursley, and O'Connor 2003). Scores in each area range from 1 (not at all true of me) to 5 (totally true of me). An index was created by summing scores for all RCI-10 items (M = 24.98). Factor and reliability analyses suggest that the 10 items form a unidimensional construct with strong reliability (varimax rotated factor loadings ≥ .73, α = .956). We also examined ideology, age, and dummy variables for race (White only = 1), education (bachelor's degree or higher = 1), gender (female = 1), and Christians (all Christian groups = 1).

Data analysis
All data analysis was conducted with SPSS version 23. In regards to our six intervention variables, the percentage of those who agree to strongly agree that the United States should intervene in each context is well below 50%, with the exception of intervening if the UNSC grants its approval (57.3%) (see Table 1A in supplemental document).
ANOVA results suggest that with the exception of "intervene to protect civilians," F(3,197) = 2.167, p = .093, and "intervene if the UNSC gives approval to do so," F(3,195) = 2.087, p = .041, there were no significant differences in the level of agreement for the other intervention variables across the different treatments (see Table 2A in supplemental document). For both variables, ANOVA results suggest that support for intervention was highest among participants who received the Christian frame (M = 3.26 and 3.71 out of 5 respectively). Additionally, the least significant difference (LSD) post hoc test suggests that for both variables there was a statistically significant difference (at the .05 level) in responses between those receiving the Christian and Muslim frames. 2 Also, for the intervention to protect civilians variable there was a significant difference between the Christian and control conditions.
Descriptive statistics better illustrate these trends. For the first variable, intervening to protect civilians, 74.1% of participants in the Christian frame agree to strongly agree that the United States should intervene (at least 19 points higher than the other conditions). Likewise, 68% percent of participants randomly assigned to the Christian condition agree to strongly agree 2 Because ANOVA results suggested that the UNSC authorization variable did not meet the homogeneity of variance assumption, nonparametric tests were conducted to confirm that there was a difference. A Kruskal-Wallis ANOVA and Mann-Whitney U post hoc test produced results similar to the original ANOVA. that the United States should intervene if the UNSC grants its approval (at least 7 points higher than the other conditions) (see Tables 3A and 3B in supplemental document).
Given that intervening to protect civilians and doing so if the UNSC approves were the only variables that exhibited statistically significant differences in scores across the experimental conditions, we summed scores for these two items to create a single composite measure that would serve as the dependent variable in the mediation analysis (M = 6.37). Conceptually, the significance of these variables may be taken to mean that participants supported intervention to help Christians with UNSC approval in a very broad sense (that is, the sentiment that something should be done), without committing to specific policies such as regime change. Factor and reliability analysis suggest that the items created a unidimensional and reliable measure (varimax rotated factor loadings = .897, α = .758).
The independent variable in the mediation analysis was the experimental treatment. Since support for intervention was highest among participants receiving the Christian frame, condition was dummy coded (1 = Christian treatment, 0 = all others) for the mediation analysis, which we use to test the assertion that being confronted with in-group suffering prompts a negative emotional response that then leads to support for intervention.
We tested our hypothesized model with mediation analysis, which is a simple form of path analysis that examines how the effect of an independent variable (X) on a dependent variable (Y) is made possible by a mediating variable (M). We conducted the analysis using PROCESS, a widely used SPSS custom dialog for conducting conditional process analysis (Hayes 2013). Using this framework, we hypothesize that receiving the Christian treatment will prompt a negative emotional response that will lead to an increased likelihood of supporting intervention.
The results of the mediation analysis confirm our hypothesized model. As Table 1 shows, the path from the Christian treatment (X) to negative affect (M) was statistically significant and shows a positive relationship, where receiving the Christian treatment was associated with an increase in negative affect. Likewise, the path from negative affect (M) to the intervention index (Y) was also statistically significant and exhibited a positive relationship. This suggests that an increase in negative affect was associated with higher levels of agreement that the United States should intervene in Syria. The direct effect of the Christian treatment on support for intervention was not statistically significant at the .05 level. Bootstrap confidence intervals for the indirect effect (.001 and .316) did not include zero, suggesting that the indirect effect of negative emotion on the relationship between the Christian treatment and support for intervention was statistically significant. 3 The estimate for the indirect effect is .118, which suggests a medium effect size.
The estimates of the direct and indirect effects suggest that mediation occurred. Thus, the relationship between the Christian treatment and support for intervention is reliant upon the negative emotional response elicited by the treatment. In regard to the covariates, ideology had a statistically significant negative association with intervention, where an increase in conservatism was associated with a decrease in support for intervention. The Christian dummy variable was significant at the .10 level and exhibited a positive relationship, where being Christian was associated with an increase in support for intervention. Conversely, religiosity, gender, education, race, and age were not associated with support for intervention. The R-square for the overall model is .132, which suggests that the independent and mediating variables explain about 13% of the variation in the dependent variable in our mediation model for intervention.
To offer additional confirmation of these results, we reran the mediation model for each experimental condition, which produced interesting findings. As Table 2 shows, results for the women and children frame were strikingly similar to those of the Christian frame, where that condition was associated with a negative emotional response that led to support for intervention. However, what is interesting about this model is that while the women and children condition and the female dummy variable did not generate individual direct effects on support for intervention, an interaction between the two does have a significant positive effect. This suggests that for females, being primed about the suffering of women and children was associated with increased support for intervention. 4 Conversely, the Muslim condition did not have a significant association with negative affect. While the Muslim condition did have a statistically significant association with intervention, the relationship was negative where being primed about the suffering of Muslims prompted a decrease in support. Lastly, inconsistent mediation occurred when the control group was inserted as the independent variable. In this instance, the control group was negatively associated with negative affect, while negative affect was positively associated with intervention, and the indirect effect was negative.

Experiment 2
Demographics A total of 201 subjects participated in E2. The majority of participants were male (65.7%), with ages ranging from 18 to 70 years of age (M = 32). In regards to race, 82.1% of participants identified as "White alone." Unlike E1, where religious affiliation was divided between Christians and agnostics/ atheists, agnostics and atheists (48.7%) outnumbered Christians (33%). Ideologically, participants' tended to be slightly liberal (M = 3.14). Once again, the majority of participants (53.2%) had earned undergraduate and graduate degrees. We also tested for an interaction effect between females and the women and children frame on negative affect, which was not significant. Likewise, for the Christian frame we tested for interaction effects for being Christian and the Christian treatment on both negative affect and support for intervention, both of which produced nonsignificant results.

Variables
As noted, the second experiment had the same design as the first, utilized the same survey instruments, and examined the same variables. 5 The only difference from E1 was the added frame about chemical weapons in the treatments.

Data analysis
Similar to E1, with the exception of intervening if authorization is granted by the UNSC (50.8%), the percentage of participants who agree to strongly agree that the United States should intervene in each context was below 50% (see Table 1B in supplemental document). Unlike E1, ANOVA results suggest that there was only a significant difference across experimental conditions for intervening when it is in the security interests of the United States, F(3,196) = 3.447, p = .018 (see Table 2B in supplemental document). However, similar to E1, ANOVA results suggest that on average support was highest in the Christian condition (3.18 out of 5). Also note that in E2 there was, on average, a higher level of support for intervention among those in the control group than the Muslim condition (M = 3.00 versus 2.88). The LSD post hoc test suggests that the Christian frame was statistically different than the women and children frame and that the women and children frame was different than the control. Descriptive statistics suggest that 70% of participants in the Christian frame agree to strongly agree that the United States should intervene if it is in its security interests (at least 7 points higher than the other conditions) (see Table 3C in supplemental document). Since support for intervention was highest among participants who read the article with the Christian frame, condition was again dummy coded (1 = Christian treatment, 0 = all others) for the mediation analysis. Since intervention when it is in the security interests of the United States was the only intervention variable to have group differences, it was examined as the dependent variable. Otherwise, the analysis utilizes the same independent, dependent, and mediating variables and covariates as E1.
The results of the mediation analysis suggest that mediation did not occur. While receiving the Christian treatment was associated with an increase in negative affect, there was no statistically significant relationship between negative affect and support for intervention. In regards to the covariates, only ideology and age were statistically significant (at the .10 level). Age had a positive association with intervention, where an increase in age was associated with higher support for intervention for security interests. Contrary to 5 E2 factor loadings and reliability estimates for negative affect and religiosity items were similar to those of experiment 1: four PANAS-X items (varimax rotated loadings ≥ .85, α = .910, M = 9.18) and RC-10 (varimax loadings ≥ .77, α = .951, M = 16.52).
the results for E1, the results suggest that the more conservative one is, the more likely s/he was to support intervention. The R-square for the second model is .104, which is in the same range as E1 (see Table 4 in supplemental document).
Similar to E1, we ran models for each experimental condition, which produced similar results in that there was no mediation. Interestingly, the women and children frame had no association with negative affect, while the Muslim frame had a positive association and the control a negative association, both of which were significant. In all four models, negative affect was not associated with support for intervention. The Christian and Muslim conditions were associated with increased support for intervention, while the women and children frame was associated with decreased support. Age was only associated with intervention in the Christian and Muslim conditions. The only consistent statistically significant finding across all four conditions was conservatism being associated with support for intervention.

Discussion of findings
The results from our two experiments reveal a number of interesting insights that have important implications for both research on humanitarian intervention and IR scholarship on norms. First, in both experiments we found that while support for intervention was widespread in many contexts, it had the highest level of support in the context of intervening if the UNSC gives its approval to do so, which constituted a majority view (57% in E1 and 51% in E2). In E1, support for intervention was statistically significant for a general intervention and if the UNSC authorizes such action.
What is even more interesting is that in both experiments the context with the second-highest level of support was intervening if other countries do so (45% and 38% respectively). Taken together, these results suggest that there may be support among the public for military intervention in grave humanitarian crises abroad, with higher levels of support for multilateral solutions to the situation in Syria. This supports the common IR argument that, especially in the post-Cold War world, legitimate military action needs to be conducted in a multilateral manner and that the norm requiring UNSC authorization for legitimate humanitarian interventions may be on the path to becoming internalized.
Second, while the findings show some public support for multilateralism, our results suggest that on average, support for intervention across the six contexts in both experiments ranged from 30% to 35%, which is rather low. While our samples are not representative of the American public, the levels of support for intervention that we found among our participants are consistent with trends found in US public opinion polls taken around the time we conducted our experiments. For example, a December 2012 poll found that 64% of Americans believed that the United States has no responsibility to act in Syria (PEW Research Center 2013a). Five months later, a Gallup poll indicated that 64% of Americans believed that the United States should not use military means to end the conflict in a hypothetical situation when all other economic and diplomatic efforts failed (Gallup 2013a). Additionally, a major research study conducted in 2014 also shows trends similar to those in our study, where only 17% of Americans supported deploying US forces to Syria, which the authors contend may be a product of war weariness (Smeltz, Daalder, and Kafura 2014). While war weariness is not a variable that we explore in our study, it may be another factor that has an effect on support for intervention. Its potential effect does not refute our findings, as the crux of our argument is that there are many individual-level factors that may influence adherence to norms, though we focus our attention on the causal significance of in-group favoritism, emotion, and framing.
In regards to the Syrian government's use of chemical weapons, an early September 2013 poll suggested that only 36% of respondents supported US military actions against the Syrian government as a way to punish the regime for the use of chemical weapons (Gallup 2013b). Similarly, a PEW/USA Today poll taken that same month found that 28% of respondents favored airstrikes against Syria when it was made known that Syrian government used chemical weapons (PEW Research Center 2013b).
Furthermore, our results suggest that support for MHI was low even among a population who should theoretically support intervention. While the majority of participants in both experiments were college-educated and were on average liberal, support for intervention was still rather low. This finding is very important because oversampling this population should have theoretically made it more difficult for us to obtain results that supported our hypotheses.
These findings are important because if we accept the commonly held definition that a norm is a "standard of appropriate behavior for actors with a given identity" whose principles must be agreed upon by a large number of actors in a social context (Finnemore and Sikkink 1998:891), the results from our experiments, coupled with public opinion polls taken around the same time, suggest that the willingness to intervene to protect civilians is arguably widespread but not overwhelming. This distinction is important because it speaks to the relative strength of a norm. For example, if we were to poll Americans as to whether they believe slavery should be relegalized, an overwhelming majority of respondents would say that it should not because antislavery norms are very strong at the mass level. Based on the level of support found in our experiments and public opinion polls, it appears that humanitarian intervention is a contested norm at best. While it has been argued that humanitarian intervention is a permissive norm that allows states the option to intervene, as opposed to making it obligatory, the evidence suggests that the acquiescent nature of the norm has yet to foster a critical mass of support for actual intervention among the public.
Third, our findings suggest that perceptions about the interests of the state may play an important role in interventions in instances when the conflict is perceived to have security implications for said state. The treatments used in E2 were identical to those used in E1, the focus of which was civilians being targeted by the Syrian regime. The only difference was the addition of chemical weapons to the list of atrocities perpetrated by the regime. The inclusion of this additional frame changed the dynamics of the decisionmaking process to support intervention. As previously noted, mediation did not occur in the second experiment. Both the Christian and Muslim treatments were associated with an increase in negative affect, while the women and children treatment had no association with negative affect. Negative affect was not associated with support for intervention in any of the treatments. In regards to support for intervention, the Christian and Muslim conditions were associated with increased support, while the women and children frame was associated with decreased support. Furthermore, unlike E1, there was no interaction effect between gender and the women and children treatment. The only consistent finding in E2 was that ideology was statistically significant in all four conditions. However, unlike E1, the relationship was positive, where conservatism was associated with increased support for intervention.
We contend that the reason why the E1 and E2 findings differed so much was because the introduction of chemical weapons in the E2 likely changed the perception of the situation from a humanitarian or human rights issue to a national security issue. Additionally, since both experiments were conducted prior to President Obama's speech stating that use of chemical weapons by the Syrian government could warrant US intervention, we contend that concerns about US credibility are not driving the results. Furthermore, in E2 the only variable that had notable group differences was intervention when it is in the security interests of the United States. While levels of support for intervention were comparable to those of E1, the fact that security interest was significant suggests that the chemical weapons frame still primed individuals to think more about security and the possibility of intervening when such interests are at stake. Moreover, though ideology remained statistically significant, the directionality changed, as conservatives were more willing to intervene when chemical weapons were introduced into the scenario. This suggests that the chemical weapons frame strongly resonated with conservatives. One reason for this may be that conservatives believed that chemical weapons in Syria could eventually pose a security threat for the United States-which is consistent with findings in social psychology suggesting that conservatives have a higher need to manage threat (Jost, Napier, Thorisdottir, Gosling, Palfani, and Ostafin 2007).
It is also important to note that the findings suggest that while participants in the Christian and Muslim conditions experienced a negative emotional response when primed about the suffering of those groups, that emotional response did not warrant a support for intervention. Contrary to E1-where respondents expressed their support for intervention to protect Christians, and women and children-in E2 the Christian and Muslim treatment was associated with support for intervention only if it is in security interests of the United States. A possible explanation for the change in support in the Muslim condition may be that in E2 participants were less worried about wishing to help individuals from a certain group and more with wanting to intervene because of a concern that the use of chemical weapons by a Muslim government perceived to be hostile to the United States may pose a potential security threat. These findings are interesting, as they suggest that participants in the Muslim condition may have based their decisions on pragmatic national security considerations consistent with basic principles of realism. Conversely, participants in the Christian condition may still have based their support for intervention on latent in-group favoritism. Though negative affect did not lead to support in this condition, the frame prompted a negative emotional response nonetheless.
Fourth, the most important contribution of this study is that the willingness to support intervention was contingent upon a number of factors such as ingroup favoritism, emotion predispositions such as ideology and religious affiliation, and how the humanitarian crisis is framed. Regarding in-group favoritism, our results suggest that beliefs about who is worthy enough to be assisted via a military intervention may not be as inclusive and universal as some scholars have claimed. In E1, in-group favoritism was manifested in a few different ways. The highest level of support for intervention was in the condition that received the Christian frame. The results of the first mediation analysis suggest that when primed with the suffering of Christian civilians, participants experienced a negative emotional response that then led to increased support for intervention. This finding is notable, as our sample was not overly religious. However, as noted earlier, because the United States is a culturally Christian country, it is likely that even less-religious Americans are more likely to have an affinity with Christians.
Additionally, the same pattern of mediation was found for participants who received the women and children frame, where reading about the suffering of women and children in Syria led to a negative emotional response that then prompted support for intervention. While the women and children treatment and the female dummy variable did not have a significant effect on support for intervention, an interaction between those two variables did. This suggests that when primed with the suffering of women and children, women were more likely to increase their support for intervention. While the response among females was not driven by emotion in the same way that it was for other participants, the results suggest yet another manifestation of in-group favoritism.
Overall, while belief that the United States should intervene in Syria was low on average, the results suggest that support was more likely when the casualties were Christians and women and children, groups with whom Americans can empathize, which is not the case for Muslims. The mediation analysis did not produce the hypothesized effects for participants receiving the Muslim treatment. That condition had a negative statistically significant association with intervention. Furthermore, the Muslim condition did not have a significant association with negative affect. Thus, when confronted with Muslim suffering, participants did not have an emotional reaction and also exhibited a decreased willingness to intervene.
In terms of covariates, E1 results suggested that age, race, gender, education, and religiosity were not associated with support for intervention, while being ideologically conservative and being a Christian had statistically significant effects on support for intervention, though in different ways. Ideology was negatively associated with support for intervention, where the more conservative a participant was, the less likely they were to support intervention. Conversely, being Christian was associated with increased support for intervention. Given that Christianity is often tied to political conservatism in American politics, there may be a few explanations for these contrasting findings. One explanation may be that Christians are more likely to support intervention because of the belief that one should help and serve others, while some ideological conservatives are more secular and oppose intervention because of isolationistic tendencies. Yet another explanation may be that given the liberal skew of our sample, the Christian identifiers in our sample may simply be of a liberal variety.
Additionally, the results suggest that emotion can play an important role in the decision-making process, which is a psychological and physiological variable that most constructivists have not addressed. Ethnic conflict research has demonstrated in considerable detail how negative emotions such as anger, fear, resentment, and the like can prompt individuals and groups to take action (for example, assisting ethnic kin in other countries, taking action against rival groups, etc.) (Kaufman 2015;Saideman 1997). If such mechanisms work within the confines of ethnic politics, then it is also reasonable to conclude that the same emotions could drive mass populations in one country to be willing to intervene in another for the purposes of stopping grave atrocities being committed. Indeed, the internationalization of ethnic conflicts are often a result of group members in other states feeling a sense of empathy and outrage about the crimes committed against their kin in other states (Henderson 1997).
The results from E1 suggest that when individuals in a country of Christian heritage are primed about how Christians and women and children are being slaughtered in another country, it prompts an emotional response that leads to a willingness to support humanitarian intervention. The importance of emotion in this instance is essential because as the mediating variable, it facilitates, or makes possible, the relationship between the treatment and support for intervention. Without negative emotion, the effect of the treatment on support for intervention is greatly diminished or nonexistent. In E2, the context of the crisis was different enough to make it so that the emotion experienced when confronted with human suffering was not enough to prompt a willingness to support intervention in the context of a potential security threat.
These findings also have important implications for the larger body of IR research on norms. The general trend in that research has been to note the existence of a particular norm (usually set into motion by an individual activist or group), provide a few examples of when states' behavior was influenced by said norm, and then conclude with sweeping assertions about how the norm has changed notions of what is appropriate across the international community. Our results suggest that the process regarding the decision to intervene is complex and is shaped by a variety of factors, such as in-group bias, emotion, predispositions, and the framing of the crisis itself. Specifically, we demonstrate the causal significance that in-group favoritism, emotion, and the framing of a crisis have on the decision to support humanitarian intervention. These nuances are commonly not captured in IR scholarship, which privileges examining structural factors and frequently rejects both methodological individualism and the utility of psychology. As a result, IR scholars do not usually account for important psychological determinants of behavior related to norms, which as we have observed can influence the willingness to comply in a number of ways. Our results suggest that individual-level motivations are important and warrant further investigation.
As noted earlier, a limitation of our participant samples is that they do not include elites, which would prompt many to conclude that the applicability of our findings is limited because it is elites, and not laypeople, making the major foreign policy decisions related to intervention. However, as discussed earlier, a body of foreign policy research suggests that elite decision making is influenced by the same psychological variables as that of ordinary people. Hence, while elites have access to more information than ordinary citizens, they are still people, and as such their decisions are driven by personality, physiological, and cognitive factors. Based on this body of research, we contend that elite decision making is not immune to the effects of in-group favoritism, emotional responses to in-group suffering, and the like, which are very primitive psychological drives and responses.
In summary, while our experiments had varied results, they provide a number of important findings for the study of MHI. First, our study finds some support among our respondents for intervention in grave humanitarian interventions, especially with a UNSC authorization. Nevertheless, the results provide empirical evidence suggesting that norms of humanitarian intervention have been adopted partially and unevenly. While our experiments only examine the United States, the findings are very important because as the world's leading power, the United States wields a considerable amount of ideational strength in terms of demonstrating to other countries which norms matter. This raises the very interesting question of whether US reluctance to act in the Syrian conflict has influenced other countries to not do so.
Second, we find that perceptions about state interests may matter in instances where grave humanitarian catastrophes abroad may be perceived as representing a security threat to intervening states. In E2, when chemical weapon use by the Syrian government was introduced, our respondents were inclined to be more willing to support intervention if it is in security interests of the United States. Finally, regarding the scope of intervention, our results suggest that group dynamics, emotion, and predispositions such as ideology are extremely important in shaping the degree to which individuals adhere to international norms. Our results support the main conclusion of Johns and Davies (2012), who suggest that states' decisions to militarily intervene in other countries are determined by individual-and group-level psychological factors, as opposed to governments' self-interests or structural-level social forces such as norms and discourses. However, our study departs from Johns and Davies (2012) in that we give more attention to the role that negative emotions play in the decision-making process, which leads to our third point. Our results suggest that emotion plays an important role in the decision-making process regarding how individuals react to normative demands. Furthermore, research has shown that emotions can prompt both thoughtful and deliberate action as well as impulsive behavior (Lerner and Keltner 2001).
While critics may argue that the psychological and emotional variables that we highlight are obvious, discussions of these factors have been almost entirely absent from contemporary IR debates about norms, which often neglect individual level factors, and to a large degree discourage the exploration of such mechanisms. The results of our experiments suggest that they do matter and thus warrant exploration by IR scholars. Furthermore, our study demonstrates that norms, and the degree to which agents adhere to them, can be measured in a more direct manner and manipulated in an experimental context, which can provide a more solid empirical foundation for assertions about norms.

Funding
Schreiner University and Marist College provided generous financial support for this project.