What explains the American public’s support for military conflict with North Korea? North Korea’s nuclear ambitions, as well as how the regime frames rhetoric around its nuclear program for foreign observers, is well documented (Rich Reference Rich2014). American responses appear to only reinforce North Korean bellicosity. For example, in August and September of 2017, the United States heard Kim Jong Un threatening to strike American bases in Guam, Donald Trump suggesting military solutions against the North Korean issue and calling Kim the “Rocket Man,” and North Korea conducting its sixth nuclear test. The resulting debates on North Korean military capabilities (Albert Reference Albert2018) and subsequent rhetoric from both Pyongyang and the Trump administration indicate that the potential for large-scale military conflict is non-negligible. Yet, despite increased rhetoric and even Pentagon assertions that any military action would require significant commitments in US ground troops (Talmazan and Associated Press 2017) beyond the roughly 28,000 currently stationed in South Korea, little attention appears to focus on the human costs of such conflict or that the specter of nuclear weapons and the potential for mass casualties hampers any consideration of conventional strikes. In addition, few public-opinion surveys ask about perceptions of North Korea (Poushter Reference Poushter2017), and much less about the probability of conflict with the country. Chicago Council on Global Affairs surveys found that Americans increasingly support the use of American troops if North Korea invaded South Korea, with 62% in favor in 2017 (Friedhoff and Smeltz Reference Friedhoff and Smeltz2017). Meanwhile, a 2017 Gallup poll found that a clear majority (58%) supported military action if economic and diplomatic efforts failed, up from 47% in 2003, with clear variation by partisanship (Reuters 2017).
Although the American public’s view of military force remains largely shaped by the objective behind such actions (Eichenberg Reference Eichenberg2005; Jentleson and Britton Reference Jentleson and Britton1998), a sizable literature reveals a negative correlation between casualties in conflict and public support for the military action itself, whether limited to the Korean and Vietnam wars (Gartner and Segura Reference Gartner and Segura1998; Reference Gartner and Segura2000; Lorell, Kelley, and Hensler Reference Lorell, Kelley and Hensler1985), the Iraq War (Berinksy and Druckman 2007), or a combination (Eichenberg Reference Eichenberg2005).
Experimental work, whether within the context of the actual conflicts or a hypothetical intervention (Gartner Reference Gartner2008; Gelpi Reference Gelpi2010), largely follows the same pattern. Crawford, Lawrence, and Lebovic (Reference Crawford, Lawrence and Lebovic2017) identified variance in public support based on the value of targets and the potential for civilian casualties. Similarly, Gelpi, Feaver, and Reifler (Reference Gelpi, Feaver and Reifler2006) found the American public more tolerant of casualties in conflicts perceived as just and where beliefs in success are high. However, past conventional conflicts may shed little light on a conflict with North Korea in which a nuclear exchange would far surpass previous casualties. Experimental work suggests that Americans are supportive of the use of nuclear weapons, despite noncombatant casualties, if such action is framed as an alternative to American military casualties (Sagan and Valentino Reference Sagan and Valentino2017).
More broadly, the priming and framing literature demonstrates that how issues are presented affects perceptions (Druckman Reference Druckman2001; Iyengar, Peters, and Kinder Reference Iyengar, Peters and Kinder1982; Scheufele Reference Scheufele2000). Iyengar and Simon (Reference Iyengar and Simon1993) found that survey respondents weighed evaluations of George H. W. Bush more heavily on foreign policy after the Gulf War, whereas exposure to television news corresponded with greater support for a military rather than a diplomatic response. Likewise, additional information about the risks and/or likelihood of success in conflict should shift the risk–reward calculus of the public. The heavy media coverage of North Korea and the potential for conflict, if not shaping perceptions of conflict themselves, potentially prime the public to consider the possibilities of such conflict.
Likewise, additional information about the risks and/or likelihood of success in conflict should shift the risk–reward calculus of the public. The heavy media coverage of North Korea and the potential for conflict, if not shaping perceptions of conflict themselves, potentially prime the public to consider the possibilities of such conflict.
The assumption, based on existing work, is that the presentation of military conflict with North Korea in terms of lives lost would lower public support for military response. Furthermore, with more than one million North Korean troops, even without the use of nuclear weapons, the potential for tens of thousands of military casualties in the first days of conflict—on either side—is not unrealistic. However, estimating American casualties remains difficult in large part due to uncertainty about the initial additional ground-force commitment the United States would make during this conflict and the evacuation of American civilians in South Korea (i.e., more than 120,000), as well as the defensive capabilities of both Koreas (McInnis et al. Reference McInnis, Feickert, Manyin, Hildreth, Nikitin and Chanlett-Avery2017). Predictions before conflicts often miss their mark, as evident by estimates of 7,000 to 30,000 American dead and wounded expected in the Gulf War (1990–1991)Footnote 1 compared to actual results of fewer than 1,200. However, the literature is less clear in terms of whether losses—not to one’s own side but rather in general—would depress public support. In other words, would the American public’s support for military conflict decline when emphasizing enemy or ally casualties?
RESEARCH DESIGN AND RESULTS
To address the role of casualties, I used an experimental survey design through mTurk Amazon to survey 1,035 Americans on October 20, 2017, regarding North Korea.Footnote 2 After a series of demographic, ideological, and personality questions, the first North Korea–related question asked for the first three words or phrases that come to mind when thinking about North Korea. This open-ended question allowed for insight on views before conflict-specific questions, views that potentially were influenced by previous exposure to media framing. The most common words, appearing in at least 20% of responses, were largely predictable (figure 1). For example, “nuclear” (and variations such as “nukes”) appeared most often, mentioned 295 times by 27.9% of respondents, followed by references to “danger” or “threat” (245 times), the regime as “authoritarian” (228 times), and “war” (219 times).
Figure 1 Word Cloud of Survey Responses about North Korea
Respondents then were randomly assigned one of four versions of a question regarding military conflict with North Korea (table 1). The baseline version stated: “Based on recent events, some have proposed an American military response to North Korea. On the following five-point scale ranging from strongly opposing such a response to strongly approving, where would you place yourself?” This wording, without specifically mentioning North Korea’s nuclear tests, acknowledged that many people likely read any prompt in the context of well-known contemporary events and the remote possibility of nuclear war rather than a baseline of ignorance on potential American military engagement in North Korea. Additional versions added one sentence in the middle of this prompt, referencing 20,000 casualties in North Korea, South Korea, or both. This estimate was chosen because it is consistent with other estimates (Daniels Reference Daniels2017; Friedman Reference Friedman2017; Gady Reference Gady2017; McInnis et al. Reference McInnis, Feickert, Manyin, Hildreth, Nikitin and Chanlett-Avery2017). The intentional omission of American casualties provided a means to simplify the experimental design and not arbitrarily choose an estimate of casualties. Any casualty assumption would rely on additional assumptions about American commitments to ground troops and broader military support.
Table 1 Experimental Survey Question Versions
Note: Underlining for emphasis not in the original survey.
For clarity, figure 2 presents the treatment of effects on the experimental design on rates of opposition (combining “strongly oppose” and “oppose”) because opposition comprised the plurality, if not majority, response in each version. For the baseline, 49.08% opposed a military response, compared to 58.46%, 59.68%, and 61.36% for versions 2, 3, and 4, respectively. The results clearly show that simply mentioning potential casualties on either side of the Korean peninsula increased opposition, with chi-square tests finding vrsions 2 through 4 statistically significant from the baseline (i.e., p=0.053, 0.038, and 0.017, respectively).Footnote 3 Additional tests focused on common terms in the open-ended question to identify whether the respondents’ word choice primed subsequent responses to the experimental question (omitted due to space limitations). Of particular note, among those who mentioned danger or threat of war, treatment effects of 10% or greater were seen across versions 2 through 4, with smaller effects among those who mentioned authoritarianism and nuclear issues. However, treatment effects were rarely statistically significant.
Figure 2 Treatment Effects on Opposition to a Military Response to North Korea (by percentage)
As an additional test, a series of regressions used the five-point Likert scale on support of a military response as a dependent variable. The first model included only whether a respondent was assigned one of the versions mentioning casualties, leaving the generic prompt (version 1) as the baseline. The second model controlled for gender (female), age, education (seven-point scale), household income (seven-point scale), a five-point measure of interest in politics, a seven-point measure of political ideology (from very liberal to very conservative), and a dummy variable for whether the respondent was currently serving or had served in a branch of the US military.
The results strongly suggest that the mere mentioning of casualties—even if not American ones—dampens public support for conflict and thus may indirectly encourage greater attention to alternatives. However, the underlying motivation behind variance in perceptions is less clear.
Table 2 presents the results. In the first model, each version of the experimental design that mentions casualties corresponded with a statistically significant decline in support, with nearly identical coefficients on both that mention South Korean casualties. The same pattern continued after controlling for demographic variables. Meanwhile, age and income negatively corresponded with support, whereas ideology emerged as the strongest positive correlation. Of particular note, military service did not reach statistical significance. Additional models provided results consistent with those in table 2 (omitted due to space limitations). Replacing the seven-point political ideology variable with dummy variables for those identifying with the Democratic and Republican parties found both to be significant at p=0.001, whereas the former corresponded with a -0.26 decrease and the latter with a 0.78 increase in support. When including both the ideology measures and partisan dummies, both the ideological-spectrum variable and the Republican dummy variable positively corresponded with support, significant at p=0.001. Adding a dummy for respondents that had been to South Korea failed to reach significance or change the core models, as did a similar test for those who had been to East Asia. Finally, to assess whether initial perceptions of North Korea mentioned in the open-ended questions primed responses to the experimental design, dummy variables for references to themes such as nuclear weapons, war, danger, and authoritarianism were tested. However, none reached statistical significance or changed the core findings of the models in table 2.
Table 2 OLS Regressions on Support for a Military Response to North Korea
Note: ****p<0.001, ***p<0.01, **p<0.05, *p<0.10.
Additional questions after the experimental-design question asked about perceptions of North Korean conflict. These included evaluating on separate five-point scales whether either the United States or South Korea was vulnerable to a North Korean attack; a 10-point scale on the likelihood of North Korea launching a nuclear weapon at either country; a 10-point scale on confidence in President Trump’s leadership in regard to North Korea; a five-point scale on whether a US war in North Korea would affect the respondent’s daily life; and a 10-point scale on confidence of US success in a military conflict with North Korea. However, the version of the experimental question received did not influence values on any of these measures when tested in OLS regressions such as those in table 1.
In contrast, the initial experimental version influenced perceptions on one other question: Does North Korea have a right to nuclear weapons? Coded as binary (no/yes), the majority of respondents (79.39%) answered no. Figure 3 shows that those who received a version of the experimental question referencing casualties were less likely to claim that North Korea had a right to nuclear weapons, with chi-square tests finding the latter two statistically significant from the baseline (i.e., p=0.076 and 0.018, respectively). However, additional tests based on the most common terms in the open-ended question did not find similar consistency across subsets. Of particular note, treatment effects of 10% to 14% emerge across versions 2 through 4 among those who mentioned nuclear issues. Logistic models (table 3) controlling for the same factors as the models in table 2 found that only the fourth version (i.e., 20,000 casualties a day in both Koreas) was statistically significant in both models, whereas in the extended model, only the ideology reached significance.
Figure 3 Treatment Effects on Beliefs That North Korea Has a Right to Nuclear Weapons (by percentage)
Table 3 Logistic Regression Models on Perceptions of North Korea’s Right to Nuclear Weapons
Note: ****p<0.001, ***p<0.01, **p <0.05, *p<0.10.
CONCLUSION
The results strongly suggest that the mere mentioning of casualties—even if not American ones—dampens public support for conflict and thus may indirectly encourage greater attention to alternatives. However, the underlying motivation behind variance in perceptions is less clear. Declining support could be a result of general loss aversion, a reminder of the human cost of war. Alternatively, despite not mentioning it directly, respondents may have inferred American casualties based on Korean casualties in the prompts or external knowledge of potential casualties. The findings suggest the difficulties in harnessing public support for a conflict once the rhetoric of casualties emerges and that policy makers simply may want to avoid addressing any human cost of conflict in the lead-up to military action, further promoting wishful thinking about the costs of such conflict.
The results also suggest a paradox. The mentioning of casualties, especially those on both sides of the peninsula, decreased support for American military action but also decreased the belief of North Korea’s right to have nuclear weapons (and, conversely, increased beliefs that they should not have those weapons). However, how the United States or the international community more broadly convinces North Korea to relinquish those weapons is unclear. After all, North Korea claims that it needs these weapons to prevent an American invasion, while the regime cannot trust American assurances to forego regime-change efforts if the rogue state relinquishes those weapons (Fearon Reference Fearon2017).
Furthermore, while this experimental design shows clear patterns of tolerance for casualties, it admittedly cannot control for other potentially important factors that may influence perceptions. First, the influence of available heuristics to gauge potential casualties is difficult to measure. The most convenient heuristic—that is, previous casualties in conflicts—potentially varies generationally (e.g., Vietnam versus Iraq/Afghanistan) and it is unclear whether the specter of nuclear war and mass casualties precludes most available comparisons. Similarly, to what extent does knowledge of the Korean War and the more than 36,000 American military casualties and roughly 2.5 million military and civilian Korean casualties influence perceptions of the magnitude of a contemporary conflict? Finally, this experimental design does not parse out opposition to a military response in this context versus a more general opposition to military intervention, one that ideological measures are unlikely to capture adequately. Nevertheless, the results suggested sensitivity to casualties otherwise underexplored.
