Introduction

Over the past couple of years, newspapers, cable news, and social media platforms have consistently covered international conflicts, as well as the extent of US support for its allies involved in those conflicts. This media coverage has featured images of destroyed buildings and devastated infrastructures, reports on civilian and military casualties from both ally countries and their adversaries, and narratives highlighting civilians impacted by the conflict. Meanwhile, support for allied nations like Ukraine has remained a constant topic of discussion, with frequent media coverage of debate surrounding military, economic, and diplomatic assistance for these nations embroiled in conflict. But do reports of civilian casualties in conflicts where the US is not directly involved militarily influence Americans’ support for policies aimed at assisting other nations in their own war efforts? Are these reports more influential than reports of military casualties, and do the effects depend on whether the civilian casualties are from the US ally’s side in the conflict? What types of information have the most impact, and what mechanisms drive the US public to support foreign involvement in these contexts?

These questions are critical to a variety of disciplines that seek to understand how we respond to contemporary events in the world, including political communication, international relations, and political and social psychology. The literature has long acknowledged that war casualties represent the most visible and salient information that media outlets present when covering foreign conflicts. It is also clear that when the US is engaged in combat, media reports on American military casualties (i.e., dead and injured armed force personnel) turn the US population against the war effort (Gartner and Segura, 1998, 2021), although the onset of a conflict and major military developments can initially increase patriotic support (Brody, 1991). Similarly, when American forces are involved in the war, foreign civilian casualties (i.e., dead and injured civilians) elicit public outcry, increasing calls to withdraw from the conflict (Crawford et al. 2017; Friedman and Sutton, 2013; Johns and Davies, 2019; Walsh, 2015). Yet, our theoretical and empirical understanding of the relation between information about casualties and American public support for involvement in military conflicts is unlikely to be complete without evidence about the impact of information on civilian harm in conflicts without American forces’ involvement. This paper concerns this impact.

Current theories about responses to civilian casualties propose that citizens in affluent Western societies have adopted a norm that emphasizes the protection of civilians during wartime (Crawford, 2013; Dill et al. 2023; Pinker, 2012). Thus, we hypothesize that when the US is not responsible for civilian casualties, as is the case when it is not directly fighting in the war, the public may still seek to protect foreign civilians by helping them defend themselves against attacks. One mechanism that may promote support for other countries when the media report on harm to civilians is empathy. Empathy can facilitate a deeper understanding of others’ perspectives (Gadarian and Brader, 2023) and is an important consideration in political contexts. For example, cues that encourage individuals to “step into” the perspective of other nations can significantly enhance support for international cooperation (Adida et al. 2018; Casler and Groves, 2023). War correspondents, photographers, and other media professionals play a critical role in fostering this empathy by capturing civilian suffering in relatable ways and humanizing victims through compelling narratives (Höijer, 2003). These media professionals may be motivated by various factors, including moral imperatives to expose human rights violations, professional prestige associated with war coverage, or financial incentives from news organizations seeking compelling stories. Regardless of their motivation, their work serves to make distant suffering more immediate and tangible to the public.

During wartime, exposure to distressing information about destruction and loss of civilian life should be able to elicit empathetic responses, prompting attitudes and support of policies aimed at helping those in need (Gutsell and Inzlicht, 2010; Hudson et al. 2019; Morelli and Lieberman, 2013; Rameson et al. 2012). The protection of human life is often considered a “sacred” or core moral value (Tetlock, 2003), with “infinite or transcendental significance that precludes comparisons [and] trade-offs” (Tetlock et al. 2000, p. 853). Thus, even though casualties are often considered a signal of cost and the need to abandon a “losing proposition” (Gartner and Segura, 1998), information about casualties might also evoke compassion and care for those affected (Agerberg and Kreft, 2023; Kreft and Agerberg, 2024). Information about casualties is likely to elicit empathy especially when victims are innocent and not in competition with the perceiver (Batson, 2018; Eisenberg and Miller, 1987; Graziano et al. 2007; Oswald, 1996; Paciello et al. 2013; Preston, 2013). As a result, compared to depictions of military casualties, depictions of civilian harm may produce greater empathy and ultimately support for policies to assist foreign countries. Depictions of civilian casualties may also have more pronounced effects in conflicts that involve an ally country with strong political, social, and economic bonds with the US. In contrast, information about civilian casualties from an adversary nation may be overlooked, not eliciting the same emotions or support for US involvement.

Possible preconditions for the effect of media portrayal of civilian casualties should be considered as well. Outstanding questions include whether information about civilian casualties is influential when it involves verbal descriptions and images or only verbal descriptions; whether information about damage to civilian infrastructure has an effect regardless of references to casualties; whether concrete information about casualties is necessary; and whether the duration of the war matters. To begin, war images are frequently used to induce an emotional response, but verbal depictions and stories are powerful as well (Griffin, 2010; Joseph et al. 2020). Contemporary coverage of wars often includes detailed descriptions of infrastructural damage and casualties in specific contexts (Hoskins and O’loughlin, 2010). However, the effect of such information on empathy and public support for a foreign country remains unclear. Additionally, a war that lasts longer may lead to an erosion of support over time, as prolonged conflicts are perceived as endless (Sanaei, 2019), prompting considerations to end the war.

This paper explores how exposure to information about civilian casualties in an ally country at war influences foreign policy attitudes. We hypothesize that exposure to civilian casualties can evoke empathy for the victims in the ally country, thereby increasing support for US involvement in the conflict. To test this hypothesis, we analyze both the real-world example of the Ukraine-Russia war and various hypothetical international conflict scenarios, achieving a comprehensive examination of our hypothesis. The Ukraine war is interesting because we do not yet know how the public is affected by the high human toll of the conflict.Footnote 1 Despite the passage of time since the war began on February 24, 2022, a majority of Americans continue to hold positive attitudes toward Ukraine and support military and economic aid to the country (Kafura and Smeltz, 2024; Poushter et al. 2023). Accordingly, the Biden administration has continued to provide military, economic, and diplomatic aid to UkraineFootnote 2, which the president has justified by repeatedly emphasizing the democratic values and aspirations of Ukrainians vis-a-vis Russia’s authoritarian ambitions (Wertheim, 2022). However, the question remains: to what extent does information about civilian casualties in Ukraine shape public support for US involvement, and through what mechanisms does this influence occur?

To address these issues, we longitudinally surveyed a probability sample of US residents and conducted four experiments, three of them preregistered. First, in the longitudinal survey (Study 1), we ask respondents to identify the aspect of the Ukraine war they hear most often about in the news (similar to “most important issue/problem” indicators, see: Moniz and Wlezien (2020)). Then, in the experiments, we directly manipulate exposure to information about civilian casualties. Support for US involvement is measured in several ways, including support for the democratic cause of the war (i.e., agreement that the war is a fight for democracy); backing of US policies for military, diplomatic, and economic assistance; approval of the actions taken by the countries in conflict; general attitudes toward US foreign involvement; and direct choices of what allies to support in particular conflicts. Our hypothesis is that exposure to information about civilian casualties should generate stronger support for the ally country compared to either no information about casualties or information about military casualties. The positive impact of civilian casualty information on support, however, may or may not persist when the ally country is perceived as socio-politically dissimilar from the US, an issue examined in Study 2. Studies 3-5 investigate other critical factors by measuring empathy, using media reports involving only text, comparing the impact of civilian and military casualties, analyzing the influence of descriptions of damaged civilian infrastructure, contrasting civilian casualties from ally and adversary countries, as well as manipulating the presence of concrete information about numbers of casualties in particular events and different durations of a war. Studies 1-4 focused on the Ukraine-Russia war, and Study 5 used conjoint analyses of hypothetical conflicts involving an ally country.

Being considered a US ally is important for countries that depend on American backing and support, whether for security, economic aid, or diplomatic influence. However, which countries are regarded as allies and the strength of these relationships are part of a dynamic process that evolves over time. Public opinion in the United States can also influence these alliances, particularly when the relationships involve risks or contentious decisions (Tomz et al. 2020). This study explores whether, in conflicts like the Ukraine-Russia war, the portrayal of civilian casualties affects support for the conflict and whether Americans’ empathy and support for suffering civilians differs across ally and adversary countries. Russia has repeatedly claimed that Ukraine is intentionally targeting Russian civilians in an effort to counter Ukraine’s narrative as a victim of Russian aggression and thus shift international attention. This raises questions about whether these claims should depend on American’s empathy for US allies and adversaries.

Study 1—Probability Panel of US Residents

In our first study (N = 2020; see Methods for details), we collected longitudinal data (six waves) from a probability, phone/Internet hybrid panel of adults residing in the US. To assess the prominence of various aspects of the war in media coverage, respondents were asked, ’What aspect of the Ukraine war do you hear most often about when following the news?’ This question allowed us to examine the salience of different topics within the media narrative. The average percentage of people who said the issue of civilian casualties was “the most salient issue” across the six waves is about 21 percent (Appendix, Table A3). This high percentage underscores that this issue has consistently been a focal point of public attention throughout the study period.Footnote 3

As a preliminary analysis, we examined the relation between US news on Ukrainian civilian casualties and the salience of civilian casualties as reported in our survey. To obtain the news about this issue, we used Nexis Uni with the search terms “Ukraine," “civilian," and “casualties" in English, covering the period from February 2022 to September 2023, and focusing on news publications in the US.Footnote 4 The results of the mixed-effects logistic regression (see Appendix, Table A1) suggest a higher likelihood of survey respondents indicating civilian casualties as the most salient issue with increased news coverage of civilian casualties. This finding thus indicates that our survey measure of issue salience is indeed related to actual news coverage.

To examine how media salience of civilian casualties shaped attitudes in support of US involvement in the war, we focus on two survey items. As mentioned, since the beginning of the Ukraine war, the Biden administration has framed its support for Ukraine as a global fight for democracy (Wertheim, 2022). On waves 1-6, we, therefore, asked how much respondents agreed that “the war between Russia and Ukraine is, on a larger scale, a fight for democracy”, on a 4-point scale, with 1 indicating “Strongly disagree” and 4 indicating “Strongly agree.”. On waves 5-6, we also asked respondents if they “approve or disapprove of US support to Ukraine?”, on a 7-point scale, with 1 indicating “Strongly disapprove” and 7 indicating “Strongly approve.”

Table 1 shows how the salience of civilian casualties affects justification for the war as a fight for democracy, using mixed-effect models with lagged independent variables. The first model includes only lagged salience as the main predictor. The results indicate a significant positive effect of the salience of civilian casualties on reported democratic justification for the war. The second model adds lagged values of reception of right-wing media, which was measured on a scale of 1-7, as the number of days during a week respondents receive information from “sources such as Fox News, Breitbart News, One America News, or The Drudge Report”. The third model adds lagged indicators for party identification, measured on a 3-point scale (1 Democrat, 2 Independent, and 3 Republican), and ideology measured on a five-point scale (from 1, very liberal, to 5, very conservative). The fourth model adds a counter variable for the survey wave (1-6) as control variables. Consistent with our hypothesis, even when these control variables are included in the model, the impact of the salience of civilian casualties remains positive and significant.

Table 1 The Effects of Media Salience of Civilian Casualties on Democratic Cause and Support Approval (Study 1 - Probability Panel of US Residents).
Full size table

To test the robustness of our findings, we also examined whether a different measure of support for Ukraine, the direct measure of approval of support included in waves 5 and 6, is also influenced by the salience of civilian casualties. The results are reported in the last column of Table 1. The model includes lagged salience as the main predictor, with reception of right-wing media, party identification, and ideology as control variables. The results again demonstrate that respondents who chose civilian casualties as the most important issue in the previous wave of the survey were more likely to approve of US support for Ukraine in the next wave of the survey. This finding thus supports our hypothesis that media reports of civilian casualties can increase support for US involvement in the Ukraine war.Footnote 5

Study 2—Preregistered Experiment Manipulating Exposure to Information on Civilian Casualties

Study 1 shows a positive association between categorizing civilian casualties as the most salient issue and our two measures of support for Ukraine, but cannot firmly establish the direction of causality. In particular, through a self-selection bias, respondents who are more concerned with the Ukraine war may seek media about civilian casualties in the conflict, thus complicating the interpretation of our findings. To address this possibility, Study 2 directly manipulates exposure to information about civilian casualties in the Ukraine war. We predict that individuals who are presented with information about civilian casualties will have more support for Ukraine than those not presented with this information. This study (N = 1200; see Methods for details) used a convenience sample from Prolific, with quotas for Republicans and Democrats.

Our experiment also examines the impact of information concerning a foreign country’s sociopolitical characteristics. While the Biden administration continues to justify the war in Ukraine as a fight for democracy, some political pundits portray Ukraine as a corrupt and undemocratic country undeserving of US support (Geissler et al. 2023). For example, Tucker Carlson stated on Fox News (February 2022): ’You can’t say it enough, Ukraine is not a democracy. … In American terms, you would call Ukraine a tyranny’ (Kessler, 2022). This aspect aligns with ongoing research highlighting the potential role of social, political, and economic characteristics in shaping empathy and support for other countries (Mutz, 2021). Specifically, we introduce information about Ukraine as either socio-politically similar or dissimilar to the US. Perceiving Ukraine as akin to the US may heighten support for the war and also increase the impact of information about civilian casualties.

In our preregistered experiment, participants were randomly assigned to one of four conditions in a 2 (casualty information: present vs. absent) by 2 (similarity: similar vs. dissimilar) factorial design. Initially, participants received a fact sheet that described Ukraine as socio-politically similar to or different from the US (Appendix, Table A4). Subsequently, they read an article, accompanied by images, that either detailed Russian attacks on Ukrainian civilians or discussed the war in general terms without mentioning civilian casualties (Table 2). Then we measured participants’ foreign policy attitudes. To this end, informed by seminal works in the field of public opinion on foreign policy (Gelpi et al. 2005; Herrmann and Keller, 2004; Milner and Tingley, 2013), we constructed three indices: (1) US Support for Ukraine Index, which measures approval of US support for Ukraine, using six items on a 7-point scale, with 1 indicating “strongly disapprove" and 7 indicating “strongly approve."; (2) Ukraine Approval Index, which measures approval of Ukraine’s actions in the conflict, using two items on a similar 7-point scale; and (3) Perspectives on Foreign Involvement Index, which measures foreign policy preferences related to the global role of the US, using four items on a 7-point scale with 1 indicating “strongly disagree" and 7 indicating “strongly agree" (see Methods for details). Afterward, participants answered a series of manipulation check questions, to assess their evaluation of civilian casualties in the war and perceived similarity to the US. Regressing these subjective estimates on the manipulations of exposure to civilian casualties and information on similarity to the US reveals that our manipulations were perceived as intended (see Methods for details).

Table 2 Manipulations in Studies 2–5.
Full size table

The first three columns in Table 3 show the results for each dependent measure in Study 2. All models include the two manipulations (exposure to civilian casualties and similarity between Ukraine and the US), and party identification, measured as a binary variable (1 Democrat and 3 Republican). The main finding from these models is, like in Study 1, that exposure to civilian casualties has a positive and significant effect on support for Ukraine. Respondents who are (vs. are not) exposed to information about Ukrainian civilian casualties are more likely to approve of US’s support to Ukraine, Ukraine’s actions, and US’s foreign involvement.

Table 3 Regression Analyses: Impact of Civilian Casualties (studies 2, 3 and 4).
Full size table

Overall, our results support the hypothesis that participants are more likely to support Ukraine when they receive information about civilian casualties than when they do not. Furthermore, contrary to the prediction that civilian casualties may have stronger impacts when Ukraine is perceived as similar to the US, this effect does not interact with information about the similarity to the US (Appendix, Table A6, Models 14, 16, and 18). Moreover, despite Democrats having stronger support for Ukraine relative to Republicans, our results generalize across Democrats and Republicans, suggesting that information about casualties heightens support in both groups (Appendix, Table A6, Models 15, 17 and 19).Footnote 6

Study 3—Experiment Exploring the Role of Empathy in a Nationally Representative Sample

Although the prior studies are consistent with the hypothesis that information about Ukrainian civilian casualties produces empathy in recipients, the studies do not directly investigate the underlying process. Therefore, Study 3 (N = 800; see Methods for details) involved a representative US sample and manipulated the presence or absence of information about civilian casualties in the Ukraine war, in a 2-cell design. Then, respondents’ support for US involvement in Ukraine was measured, using the same measurement of US Support for Ukraine as in Study 2. Subsequently, empathetic responses were measured using an Empathy Index, which includes six items on a 5-point scale, with 1 indicating “strongly disagree” and 5 indicating “strongly agree”. Unlike Study 2, Ukraine was described as moderately similar to the US in this experiment. The information about casualties, however, was identical to Study 2 (see Table 2).

As a first step, we assess the direct effect of information about civilian casualties on support for Ukraine while controlling for party identification, which is measured on a 3-point scale (coded as 1 for Democrat, 2 for Independent, and 3 for Republican). The results, presented in the fourth column of Table 3, reproduce the findings from Study 2. Specifically, exposure to information about civilian casualties leads to stronger approval of US support for Ukraine. These results are important because they not only support our earlier findings with a nationally representative sample but also demonstrate that they hold when Ukraine is portrayed as moderately similar to the US. Following this foundational analysis, we proceed to examine the role of empathy in shaping public support for Ukraine.

Figure 1a presents the results of the mediation analysis, conducted using structural equation modeling (SEM), to assess the potential intervening role of empathy in the relation between exposure to civilian casualties and approval of US support for Ukraine. We estimate the model using Maximum Likelihood estimation and obtain key fit statistics, including the Root Mean Square Error of Approximation (RMSEA) and Comparative Fit Index (CFI). Overall, these statistics indicate that the proposed mediational chain (Model 1a) fits the data.

Fig. 1: Path Analyses Study 3 - Experiment: Exploring the Role of Empathy.
figure 1

a Proposed chain model; b first alternative chain model; c second alternative chain model.

Full size image

Even though mediation analyses provide evidence of the mechanism underlying the experimental effect, the relation between empathy and support for the war remains correlational. To address potential reverse causation, we compared our proposed chain model with two alternative specifications. The first alternative (casualties ⟶ Ukraine support ⟶ empathy) showed notably poorer fit (AIC = 7491.184, BIC = 7551.434, RMSEA = 0.397, CFI = 0.695) compared to our hypothesized model (AIC = 7373.795, BIC = 7434.045, RMSEA = 0.055, CFI = 0.994). A second alternative model testing whether empathy precedes casualties (empathy ⟶ casualties ⟶ Ukraine support) also showed inferior fit (AIC = 7462.571, BIC = 7522.821, RMSEA = 0.346, CFI = 0.770). These comparisons provide robust support for our interpretation that exposure to civilian casualties triggers empathy, which subsequently increases support for Ukraine.

Study 4—Preregistered Experiment Manipulating Exposure to Information on Civilian vs. Military Casualties

Building on the experimental design of the previous studies, Study 4 (N = 800; see Methods for details) investigates the differential impacts of exposure to information about civilian versus military personnel casualties. This preregistered study, which involved a quota sample balanced for party identification and sex, also examines if the effect of casualties is present not only with text and images, as in Studies 2 and 3, but also with only text (see Table 2). Participants were randomly assigned to one of two groups. The first group received information about over 100,000 civilian casualties in Ukraine, resulting from targeted attacks on residential areas and hospitals. The second group was presented with information about an equivalent number of Ukrainian military personnel casualties, due to attacks on barracks and training facilities (Table 2). The dependent variables measured in this study were the same as in Study 3 and included two indices: US Support for Ukraine and Empathy. Also included is a 3-category measurement for respondents’ party identification, similar to Study 3.

The results, detailed in the last column of Table 3, indicate a distinct difference in public support for Ukraine depending on the type of casualties reported. Participants exposed to information about civilian casualties showed significantly higher support for Ukraine than those who received reports of military personnel casualties. This finding suggests that, as predicted, civilian casualties evoke greater support for US involvement, signaling a special significance of civilian harm in an ally country. Similar to Study 3, we further find that empathy mediates this relation (Appendix Fig. B2). Study 5 also addresses this issue and again tests the role of empathy, yet in a broader range of hypothetical conflicts.

Study 5—Preregistered Experiment Evaluating the Impact of Civilian Casualties in Hypothetical Wars in a Nationally Representative Sample

Study 5 (N = 300; see Methods for details) seeks to extend the findings from previous studies by exploring support for US involvement in hypothetical international conflicts involving allies. While earlier studies focused on the effects of information about civilian casualties in the Ukraine conflict, this study adopts a broader framework to explore whether the observed patterns hold in a more generalized model of international relations. Specifically, Study 5 tests the impact of several factors, including the number of casualties, the severity of infrastructure damage, concrete details about the circumstances of casualties, and the duration of the conflict, for both ally and non-ally countries. By addressing these variables, this study provides a more comprehensive understanding of the mechanisms and preconditions influencing public opinion in diverse international crises.

Our sample took part in a conjoint experiment designed to simulate decision-making processes related to international conflicts. Each participant was presented with ten hypothetical scenarios, each detailing two separate conflicts in different global regions, involving various countries. Each conflict was structured to include detailed attributes of about two countries: one a US ally and the other an adversary. The manipulated attributes included the number of civilian casualties in the ally and adversary countries (a randomly selected number between 0 and 100,000), the extent of damage to civilian infrastructure in each country (a randomly selected number between 0 and 40%), concrete information about numbers and circumstances of recent casualties (a randomly selected number between 0 and 200 casualties in a specific event; e.g., #civilians were injured and #were killed “in shelling of a residential area”), and the duration of the conflict (a randomly selected number between 0 and 60 months). Participants then chose whether or not they approve of more US support (military, diplomatic, and economic) to either ally country in the described choice, followed by a report of their level of empathy for each ally country.

The initial analysis examines the direct impact of information about civilians on public support for an ally country. As predicted, participants supported a US ally more when presented with scenarios featuring higher numbers of civilian casualties, more substantial damage to civilian infrastructure, and concrete information about civilian harm in the ally country. The marginal effects from the binary logistic regression, presented in Fig. 2a, illustrate the dynamics of these influences. Specifically, we examine the difference in the probability of supporting an ally country for each attribute’s full range effect (from minimum to maximum values). The difference in probability of support is 0.4 higher for civilian casualties, 0.1 higher for civilian infrastructure damage, and 0.2 higher for concrete information about casualties in the ally country. However, the impacts of attacks on civilians or infrastructure in the adversary country have no significant effects on the probability of support. Nor does the duration of the war have a significant effect.

Fig. 2: Results from Conjoint Experiment, Study 5 - Preregistered Experiment Evaluating the Impact of Civilian Casualties in Hypothetical International Conflicts.
figure 2

a Predicted values of war components; b mediation analysis: empathy.

Full size image

In the next analysis, we focus on the mediating effect of empathy. The models, as depicted in Fig. 2, illustrate the total, direct, and indirect effects of each significant manipulation from the previous analysis. As predicted, the effects of civilian casualties, civilian infrastructures, and higher numbers of concretely described casualties on support for an ally country are mediated through empathy. Moreover, for all three attributes, we observe partial mediation, with civilian casualties showing the strongest indirect effect. Overall, these findings indicate that empathy channels the impact of information about civilian casualties on support for US involvement.

Discussion

Five studies examined how information about civilian war casualties from an ally country influences support for US involvement in a conflict. Across correlational studies and experiments, our findings consistently demonstrate that the type of information disseminated has the potential to shape public support for US foreign involvement. We started with a longitudinal analysis of a large national study (Study 1), which showed that the salience of civilian casualties in the Ukraine war was positively associated with a democratic justification for the war. This finding was robust across different model specifications and alternative attitudinal measures of support for Ukraine.

Next, we used experiments to directly manipulate information exposure in our case study – the war in Ukraine – and examine its impact on public opinion. Study 2 expanded our analysis by establishing the causal effect of exposure to text and images of civilian casualties on support for US involvement in Ukraine. Study 3 provided a replication of these experimental findings and added critical mediational evidence about the role of empathy, showing that the emotional response to civilian suffering is a key factor in shaping US public opinion. Study 4 extended this analysis by comparing the effects of information presented solely through text about civilian casualties versus military personnel casualties. Highlighting the unique impact of civilian-focused narratives, the findings revealed that information about civilian casualties elicited significantly more support for US involvement compared to information about military casualties.

Study 5 expanded the scope of our analysis beyond the context of the Ukraine conflict by exploring responses to hypothetical international conflict scenarios involving US allies. The findings identified four key drivers of empathy and ultimate support for a country at war: the number of civilian casualties from the ally side, the extent of damage to ally’s civilian infrastructure, and concrete accounts of recent attacks on civilians. The mediation analysis demonstrated that empathy played a role in translating these attributes into increased support for US involvement. Yet, of these three components, we found that the number of civilian casualties had the strongest direct and indirect impact on support for the ally country. Equally important, as anticipated, information about civilian casualties and damaged civilian infrastructure did not boost support when they affected a nation that was not a US ally.

Our research contributes to the existing literature on the influence of portrayals of militarized conflicts on public opinion regarding foreign policy (Kraitzman and Genauer, 2023; Kraitzman and Ostrom, 2021). We demonstrate that exposure to war casualties does not necessarily lower support for involvement in the conflict. Instead, our findings suggest a more nuanced public response, wherein the depiction of civilian casualties generates selective public support for government action that could prevent further atrocities against civilians abroad. While previous studies have found that pride and enthusiasm are key emotions mediating public support during wartime (Aday, 2010), our study reveals that empathy can also be a powerful motivator for public opinion, above and beyond the desire to minimize the costs of a war. Our study further reinforces Gartner and Gelpi’s (2016) finding that the content of war information matters more than its format of delivery – whether the information is conveyed through images or text.

However, our findings must be interpreted in light of certain limitations. One concern is the potential bias in self-reported measures of empathy and policy support, especially given the highly polarized state of US politics on foreign policy issues (Smeltz, 2022). Similar to challenges in estimating military casualties in modern conflicts, contextual factors and measurement tools introduce uncertainty (Khorram-Manesh et al. 2022) that also affects media coverage and public opinion. In a polarized environment, the public may rely on often conflicting estimates of war casualties, which can shape their levels of empathy and influence support for foreign policy decisions. Future research could address this issue by exploring the sources of casualty estimates and how they are perceived by the public, as well as by examining the role of media framing in shaping these perceptions. Such an effort could help uncover how discrepancies in reported estimates affect empathy and foreign policy support across different segments of the population.

Furthermore, to avoid distress and mimic the most common media coverage of war, our study did not expose respondents to highly vivid images of casualties. While such images may evoke stronger emotional responses, potentially amplifying empathy and increasing support for an ally country, they could also provoke discomfort, avoidance, or even desensitization, which might reduce empathy and public support. The impact of highly vivid and distressing imagery on public opinion, therefore, remains an open question.

Finally, the potential influence of ongoing conflicts on participants’ responses represents an additional limitation. While Experiment 5 attempted to mitigate this concern by utilizing a hypothetical conflict scenario without referencing any specific country, the ongoing salience of the Ukraine conflict at the time of data collection may have implicitly shaped participants’ responses. This real-world backdrop may complicate efforts to isolate the effects of our experimental manipulations. Future research could explicitly examine perceptions of civilian casualties in conflicts involving non-European US allies, such as those in Asia, Africa, or the Middle East. Such studies would help disentangle the influence of geographic and cultural proximity from broader humanitarian norms, offering a more comprehensive understanding of how empathy and policy support are shaped in different conflicts and geographic contexts.

Our insights are important for policymakers and media outlets as they navigate the complex interplay between the harsh realities of war and the American public opinion on US involvement in global conflicts. Our argument may compel policymakers to stress the humanitarian dimension of an ally’s struggle to leverage public support for the government’s actions abroad. Furthermore, our research highlights the crucial role of media in shaping continued support for foreign intervention, indicating how empathy-inducing reporting on foreign conflicts can increase and sustain support for US involvement.

Future research should explore the role of trust in the government’s handling of foreign conflicts, particularly knowing that public trust in crisis management and emergency communication can shape individuals’ responses to media reports (Khorram-Manesh et al. 2021). In contexts where trust in media is low or misinformation is widespread, emotional responses to news may change or even disappear (Albarracín, 2020). For example, misinformation can lead to skepticism about the authenticity of casualty reports or downplaying the severity of civilian suffering. Therefore, misinformation about war casualties carries the risk of manipulating foreign policy attitudes, potentially driving support for or opposition to US intervention. This is an area ripe for additional research. Most critically, how to elicit US involvement on behalf of civilian casualties in an adversary country is important as well, particularly as the US historically has been striving to manage its international image (Snow, 2008). Recognizing that US media coverage of casualties may be biased for political reasons, it is vital to investigate these complexities and their effects on foreign policy support.

Methods

All data collection for this research was approved by the Institutional Review Board of the University of Pennsylvania (Protocol IDs: 852150 and 853762) and complies with all relevant ethical regulations. The first two studies were conducted while the first author was at the Australian National University, and the other studies were conducted while all authors were at the University of Pennsylvania. All experiments were designed in Qualtrics and performed on the survey platform Prolific.

Transparency and Openness

We report how we determined our sample size, all data exclusions, manipulations, and measures in the study. All data, analysis code, and research materials are available at OSF. Data were analyzed using Stata, version 18.0 (MP edition). The second, fourth, and fifth studies’ designs and analyses were pre-registered.

Study 1—Probability Panel of US Residents

Participants

In Study 1, longitudinal data were collected across six waves of a survey of a probability panel of US adults aged 18 and over, with the number of respondents starting at 2020 (see Table 4). Table 5 shows that the opinion panel, which is managed by SSRS, reflects the demographic composition of the US population in terms of age, sex, race/ethnicity, and education. It also has a similar percentage of self-reported Republicans, Independents, and Democrats. The data collection was conducted via web and telephone, in both English and Spanish.

Table 4 Longtidunal Study: Six Waves.
Full size table
Table 5 Demographic Data - Study 1.
Full size table

Independent Variables

To examine the media salience of civilian casualties, respondents were presented with the following question: “What aspect of the Ukraine war do you hear most often about when following the news?” Respondents were provided with ten distinct categories: Civilian casualties, Military aid, Military advancement, Threats of weapons of mass destruction, Refugees, NATO membership, Humanitarian efforts, Economic sanctions, Peace negotiations, or other. These topics were chosen based on a comprehensive review of New York Times front-page articles about the Ukraine conflict published from February to May 2022. A preliminary survey executed through Amazon Mechanical Turk on June 4th, 2022, indicated that only 2 percent of participants selected the “other” category. This finding implies that the selected topics were representative of the principal issues portrayed prominently in media coverage of the Ukraine war. Highly infrequent selection of the “other” category, consistently around 3 percent, reinforced the relevance and breadth of the included topics over several iterations of our panel survey. Our primary independent variable in this study is media salience of civilian casualties. We operationalize this variable as a binary indicator, taking the value of one when a respondent reported hearing about civilian casualties most often and zero otherwise.

To account for political attitudes, the statistical analysis also includes measures of ideology (measured on a five-point scale from 1, very liberal to 5, very conservative), party identification on a 3-point scale (coded as 1 for Democrat, 2 for Independent, and 3 for Republican), and reception of right-wing media, measured as the number of days per week respondents watched right-wing media from sources such as Fox News, Breitbart News, One America News, or The Drudge Report (ranging from 0 to 7 days). We expect that Republicans, those with a more conservative ideology, and individuals with higher levels of exposure to right-wing media, will be less likely to agree with the democratic cause of the war.

Dependent Variables

The study utilizes two alternatives to measure support for US involvement in the war. The first measure taps perceptions of the war as having a democratic cause, and the second is an assessment of approval for supporting Ukraine. In all six waves, respondents were asked to indicate their level of agreement with the following statement on a four-point scale, ranging from strongly disagree to strongly agree: “The war between Russia and Ukraine is, on a larger scale, a fight for democracy”. In addition, the last two waves of the survey ask about approval of support for Ukraine: ’Do you approve or disapprove of US support to Ukraine?’, with a 7-point scale from strongly disapprove to strongly approve.

Study 2 - Preregistered Experiment Manipulating Exposure to Information on Civilian Casualties

Participants

We conducted a pilot experiment that revealed a d = 0.22. Based on that, we planned a sample size of 1200, which is more than sufficient to detect this effect with α = 0.05 and power of a power of 95%. Thus, Study 2 involved 1200 participants, equally divided between Republicans and Democrats (Table 6). In accordance with our preregistration, participants who completed the survey in less than 2 minutes were excluded from the analysis. The experiment was designed in Qualtrics and performed on the survey platform Prolific.

Table 6 Demographic Data - Study 2.
Full size table

Design

The experiment was executed on June 22, 2022, following its preregistration at AsPredicted (136418). Participants were randomly assigned to one of four conditions of a 2 (Casualty information: present vs. absent) X 2 (Similarity: similar vs. dissimilar) between-subjects experiment. They were first exposed to information that highlighted Ukraine as a nation that is either similar to the US or not (Appendix, Table A4). They then read an article, which contained images, that discussed either Russian attacks on Ukrainian civilians or the war in general but without mentioning civilian casualties (Table 2). The images were selected based on five criteria: (1) distribution by reliable sources (AP or AFP), (2) widespread use across US media outlets, (3) timing immediately prior to the study’s launch, (4) alignment with our experimental conditions, and (5) avoidance of gory or excessively graphic content to prevent severe emotional distress to research participants.

Manipulation Check Results

We included manipulation check questions to assess participants’ evaluation of civilian casualties in the war and perceived similarity to the US. Participants were asked, “To the best of your knowledge, how many civilians have died or been injured in Ukraine in the last seven days?” with response options: 0, 1-100, 101-500, 501-1000, 1001-2000, and more than 2000. They were also asked to rate the similarity between Ukraine and the US on various aspects, such as democratic institutions, personal liberty, social values, legal system, media and freedom of speech, economy, ethnicity, and religion, with response options: 1 (Very dissimilar), 2 (Moderately dissimilar), 3 (Slightly dissimilar), 4 (Equally similar/dissimilar), 5 (Slightly similar), 6 (Moderately similar), and 7 (Very similar).

We tested whether our manipulations affected the perceived number of casualties and perceived similarity using t-tests. Results indicated that participants who read the information about casualties estimated casualties to be significantly higher (M = 3.373, SD = 0.054) than participants who read the control information (M = 3.138, SD = 0.052). Similarly, participants who received information about Ukraine being similar to the US rated the similarity significantly higher (M = 4.774, SD = 0.048) than those who received information about Ukraine being dissimilar (M = 3.096, SD = 0.051). These results indicate that our manipulations were perceived as intended.

Dependent Variables

Three distinct indices are used to measure foreign policy attitudes: approval of US support for Ukraine, approval of Ukraine’s actions in the conflict, and broader perspectives on US’s foreign involvement.

  • US Support for Ukraine. This index assesses participants’ level of support for the United States’ foreign policy regarding Ukraine. It is calculated by averaging the responses to the following six questions: “Do you approve or disapprove of the US support to Ukraine?" and “Do you approve or disapprove the following types of US support to Ukraine: Sanctions against Russia; Humanitarian assistance to Ukraine; Sharing intelligence with Ukraine; Sending military weapons to Ukraine; Sending troops to Ukraine". Responses to each item are provided on scales from 1 (strongly disapprove) to 7 (strongly approve), with higher values indicating higher levels of support to Ukraine in various forms.

  • Ukraine Approval. This variable measures participants’ attitudes toward Ukraine’s actions in the conflict with Russia. It is calculated by averaging the responses to the following two questions: “To what extent do you support or oppose the actions taken by the Ukrainian people in response to the military conflict with Russia?" and “Do you approve or disapprove Ukrainian President Zelensky’s use of military force against Russia?".Footnote 7 This index ranges from 1 (strongly oppose/disapprove) to 7 (strongly support/approve), with higher values indicating higher approval of Ukraine’s actions.

  • Perspectives on Foreign Involvement. This index captures the average of participants’ general foreign policy preferences, which encompass their views on issues related to international cooperation, the United States’ global role, and attitudes toward foreign nations: “The United States needs to play an active role in solving conflicts around the world”; “It is essential for the United States to work with other nations to solve problems such as overpopulation, hunger, and pollution”; “America needs to cooperate more with the United Nations in settling international disputes”; and “The US government should just try to take care of the well-being of Americans and not get involved with other nations”. After reverse scoring the last item, the index ranges from 1 (strongly disagree) to 7 (strongly agree).

Study 3 - Experiment and The Role of Empathy in a Nationally Representative Sample

Participants

In this survey, we planned a sample of 800 participants, which exceeds the required number to detect a small effect size (r = 0.10) with α = 0.05 and a power of 0.95. Thus, Study 3 included 800 participants from a nationally representative sample collected from the survey platform Prolific. Table 7 describes the sample and shows not only that it is similar to the US population in age, sex, and race/ethnicity but also that it had similar percentages of self-reported Republican, Independent, and Democrat. The experiment was designed in Qualtrics.

Table 7 Demographic Data - Study 3.
Full size table

Design

This study is a 2-cell (Casualties: present vs. absent) between-subjects experiment, which was executed on October 30, 2023. Participants were randomly assigned to information about the Ukraine war that either did or did not mention casualties (Fig. 2). It differed from Study 2 in that similarity was set to be moderate (Appendix, Table A5), instead of being manipulated to be either high or low, and the addition of an empathy measure. The articles describing the war were identical to the ones used in Study 2.

Dependent Variables

The dependent measures used in Study 2 were also included in Study 3. In addition, this study included a measure of empathy, which was obtained by averaging agreement with the following five statements: (1) “The article made me feel compassion for those affected by this war.”; (2) “Thinking about the war was upsetting.”; (3) “As I was looking at the picture and reading the text, I thought about people living in the area.”; (4) “As I was looking at the picture and reading the text, I felt for people living in the area.”; and (5) “As I was looking at the picture and reading the text, the image and description disturbed me.” Responses to each question range from 1 (strongly disagree) to 5 (strongly agree), and higher values on this index indicate greater levels of empathy. Similar to Study 2, we also measured respondents’ party identification.

Study 4 - Preregistered Experiment Manipulating Exposure to Information on Civilian vs. Military Casualties

Participants

We conducted a pilot experiment that revealed a d = 0.3. Based on that, we planned a sample size of 800, which is more than sufficient to detect this effect with α = 0.05 and power of a power of 95%. Study 4 was conducted with a quota sample of 800 US adults, balanced by political identification and sex (Table 8). The experiment was designed in Qualtrics and performed on the survey platform Prolific. Following our preregistered plans, participants who failed an attention check or completed the survey in less than one minute were excluded from the analysis.

Table 8 Demographic Data - Study 4.
Full size table

Design

This experiment, conducted on February 21, 2024, after the study was preregistered at AsPredicted (163101), aimed to explore the impact of exposure to information on civilian versus military casualties on support for US involvement. Employing a 2-cell (civilian vs. military casualties) design, this between-subjects experiment allowed for the direct comparison of responses to Russian attacks on Ukrainian civilians with responses to attacks on Ukrainian military personnel (Table 2). Thus, this study extends the previous experiments by specifically testing whether information about civilian casualties elicits stronger support for intervention than information about military personnel casualties. For this study, we used text-only stimuli because we could not identify US media photos of military casualties that satisfied our criteria from the previous experiments. Most available photos depicted military units actively engaged in combat rather than suffering harm, while the few photos showing casualties were either excessively graphic or ambiguous – either about whether the individuals were military personnel or civilians, or about their nationality and side of the conflict – making them unsuitable for inclusion. Additionally, the text-only stimuli allowed us to examine whether our findings generalized to media without images (Table 2).

The dependent variables measured in this study were identical to those in Study 3 and included two indices: US Support for Ukraine and Empathy.

Study 5 - Preregistered Experiment Evaluating the Impact of Civilian Casualties in Hypothetical Wars with a Nationally Representative Sample

Participants

We conducted a pilot experiment that revealed a d = 0.5. Based on that, we planned a sample size of 300, which is more than sufficient to detect this effect with α = 0.05 and power of 0.95. Therefore, Study 5 included 300 participants from a nationally representative sample collected from the survey platform Prolific. Table 9 describes the sample and shows not only that it is similar to the US population in age, sex and race/ethnicity but also that it had similar percentages of self-reported Republican, Independent, and Democrat. The experiment was designed in Qualtrics, while using a conjoint JavaScript code. In accordance with our preregistration, participants failing to pass the attention test question or the Qualtrics bot check or whose response time was below 4 minutes (indicating insufficient engagement with the content) were excluded from the analysis.

Table 9 Demographic Data - Study 5.
Full size table

Design

The final experiment was executed on June 22, 2022, following its preregistration at AsPredicted (170103). Before encountering the hypothetical scenarios, participants navigated through several introductory pages that outlined the structure of the information presented and explained what each attribute indicated. Participants were then presented with ten hypothetical scenarios, each describing two separate conflicts in distinct regions of the world involving different countries. In every conflict, one of the countries was always a US ally and the other was always a US adversary. The conflict included information on five, randomly assigned attributes for each conflict (see Table 2 for an example): (a) the duration (0–60 months); (b) the number of civilian casualties in the ally country (0–100,000); (c) the percentage of civilian infrastructures damaged in the ally country (0-40%); (d) the number of civilian casualties in the adversary country (0–100,000); and (e) the percentage of civilian infrastructures damaged in the adversary country (0–40%).

To ensure realistic variability in the number of civilian casualties presented in the scenarios, the JavaScript code used in the Qualtrics survey platform randomly generated the number of civilian casualties as follows: a random number between 0 and 2 was generated to determine the range. If the number was 0, civilian casualties were set to 0. If the number was 1, civilian casualties were set to a value between 2000 and 5000. If the number was 2, civilian casualties were set to a value between 40,000 and 100,000. This process was repeated for each of the twenty hypothetical scenarios, ensuring diverse casualty numbers. This method provided a robust basis for analyzing participants’ decision-making processes related to international conflicts.

A sixth manipulated attribute was numbers of casualties in concrete recent events. Some conflicts did not include any concrete information and were coded 0. Other conflicts included such information with values for both the number of injured civilians (ranging from 1 to 100) and the number of killed civilians (ranging from 1 to 100). These conflicts were coded as the some of casualties. In addition, concrete information included one of the following descriptions: # civilians were injured and # were killed “in shelling of a residential area”, “during an airstrike on a marketplace”, “in crossfire near a school”, “after a bomb exploded in a train station”, “when a bridge was targeted”, “due to landmines in a village”, “in an attack on a hospital”, “in sustained artillery fire on a town”, “after an attack on a religious building”, “in a chemical attack on a suburb”, “in a sniper attack at a playground”, “during a siege on a city district”, “during a raid on a cultural center”, “in a series of coordinated attacks on public squares”, “during a mortar strike on a refugee camp”, “after a drone strike on a convoy”, “in a surprise assault on a peacekeeping mission”, “following a car bomb incident in a commercial district”, “amidst clashes at a border checkpoint”, or “when a missile hit a residential complex”.

Dependent Variables

After participants saw each pair of conflicts, they made a choice for the US to support either of the ally countries for which a conflict was depicted. The question was asked as follows: “Which country do you believe should receive increased support (e.g., military, diplomatic, and economic) from the United States?" This was measured as a discrete choice between the two ally countries presented in each scenario. The country that was selected received a 1 and the other a 0.

Additionally, after making their initial choices for all ten scenarios, participants were prompted to revisit each scenario to evaluate their empathetic responses towards the countries involved. They were asked: “Considering the conflicts described, for which country do you feel more empathy: Country X or Country A?". Then they rated their empathy on a scale from 1 (feeling strong empathy for Country X) to 7 (feeling strong empathy for Country A). These responses are used in mediation analyses.