Military Alliances and Public Support for War (2024)

Abstract

Introduction

Do military alliances matter and, if so, when and why? These questions have long been debated in academic and policy circles. Some argue that military alliances are merely “scraps of paper” that countries can disregard when it suits them. According to skeptics, alliances place minimal constraints on state behavior, including decisions to use military force. Others take for granted that alliances are consequential. For centuries, American leaders have warned that decisions to enter alliances should not be taken lightly. Thomas Jefferson railed against “entangling” alliances and George Washington warned against “permanent” ones. In this view, alliances shape behavior because breaking them would have consequences.

In this article, we examine one piece of the alliance puzzle: how alliances affect public support for war. Many scholars argue that public opinion is important to decisions about the use of military force. Leaders of democracies, in particular, rarely go to war without public backing (Reiter and Stam 2002). Leaders expect to be penalized domestically for fighting wars the public opposes and for failing to fight wars the public supports (Tomz, Weeks, and Yarhi-Milo 2020). For these reasons, domestic audiences can be “pivotal in the choice to intervene” on behalf of an ally (Gartzke and Gleditsch 2004, 782; Chiba, Johnson, and Leeds 2015).

To date, however, scholars have not investigated whether and how alliances shape public opinion about defending countries that have been attacked.¹ In this article, we develop three hypotheses. First, citizens should be more supportive of military intervention on behalf of a formal military ally than on behalf of an otherwise similar country with which the United States has not signed an alliance. Second, alliances should affect public opinion through two main mechanisms: increasing public fears about the reputational costs of nonintervention and heightening the perceived moral obligation to intervene. Finally, alliances can create domestic pressure for war even when honoring the commitment would be extremely inconvenient, underscoring one reason why leaders should take great care when deciding which alliances to form.

We tested these hypotheses by conducting survey experiments in the United States. Participants read a vignette in which one country attacked a country that shared US interests, in an attempt to seize some of that country's territory. We randomized whether the United States had signed a formal defense pact or had not made any pledge to defend the invaded country. To estimate the potential effects of alliances across a wide range of circ*mstances, including ones in which actual alliances might be more or less common and intervention might seem more or less attractive, we also varied four contextual features: the costs of intervention, the stakes for the United States, the political regime of the invaded country, and the location of the conflict. We then measured whether and why respondents would support or oppose using the US military to defend the invaded country.

Our experiments yielded several important findings. First, military alliances substantially changed public preferences about war. Participants were, on average, 33 percentage points more supportive of intervention to help an ally than to help an otherwise equivalent country to whom the United States had not made a pledge. Thus, in cases where the public has influence, alliances can raise the probability of intervention and potentially contribute to the credibility of deterrence (Leeds 2003a; Johnson and Leeds 2011).

Second, the effects of alliances on public opinion arose through two mechanisms: reputation and morality. Alliances increased public support for war by raising concerns that inaction would hurt America's reputation for reliability and violate ethical norms, including concerns about fairness and loyalty. These findings have broader implications for international relations, by revealing the criteria citizens use when thinking about compliance with treaties and other international commitments.

Finally, although alliances swayed public opinion across a wide range of circ*mstances, they made the biggest difference when the costs of intervention were high, the stakes of intervention were low, and the country needing aid was not a democracy. In precisely those situations, alliances turned public skepticism about intervention into majority support for war. Thus, our experiments suggest that alliances are not mere scraps of paper that countries can dismiss when it suits them. Instead, alliances are weighty commitments that can increase domestic support for war, even when intervention would otherwise seem inadvisable.² These findings, too, have broader implications for international relations, by not only confirming the power of treaties, but also highlighting when treaties would be most consequential.

Hypotheses about Alliances and Public Opinion

In recent years, scholars have examined how public opinion influences the foreign policies of democratic states.³ Studies have shown that voters tend to elect leaders who share their views about the use of military force, and that leaders, once in office, respond to public opinion when making decisions about war.⁴ Of course, public opinion is not the only factor influencing leaders’ decisions about war and elites may be able to shape public opinion to some extent.⁵ Nevertheless, understanding how the public responds to information about military crises is valuable for explaining and predicting when democracies use military force.

Accordingly, scholars have studied factors that affect public support for war. Research has shown, for example, that voters are sensitive to the human and economic costs of conflict (Mueller 1973) and are more willing to support conflicts involving low casualties and financial expenditures (Gartner and Segura 1998; Gelpi, Feaver, and Reifler 2009; Flores-Macias and Kreps 2017). Citizens also weigh the stakes of a conflict and the likelihood of success, becoming more likely to support military intervention when they believe US interests can be pursued successfully (Eichenberg 2005; Gelpi, Feaver, and Reifler 2009). Finally, democratic publics are more supportive of intervention to help democracies than to help dictatorships (Herrmann, Tetlock, and Visser 1999).

To date, however, scholars have overlooked a potentially important factor influencing public support for war: military alliances.⁶ Although researchers have begun to examine how international commitments affect public opinion in other domains, they have not investigated whether and why military alliances influence public willingness to defend countries that have been attacked. Moreover, scholars have not studied how alliances interact with other key drivers of public opinion, including costs, stakes, and regime type.

In this article, we hypothesize that formal alliances affect public support for war through three potentially complementary mechanisms: reputation for being a reliable military ally (“military reputation”), reputation for being a reliable partner in other domains (“nonmilitary reputation”), and moral considerations including fairness and loyalty.⁷ Much research has argued that violating alliance commitments can hurt a country's reputation in the eyes of foreign observers (e.g., Crescenzi et al. 2012; Crescenzi 2018). We connect this insight to domestic politics by arguing that voters anticipate how abandoning an ally would compromise their country's military reputation. We further argue that citizens expect collateral reputational damage: leaving an ally in the lurch could tarnish the country's reputation for reliability in nonmilitary domains such as trade or the environment. Finally, we argue that alliances trigger public concerns about fairness and loyalty. Beyond all of this, we hypothesize about the circ*mstances under which alliances might be most potent.

Reputation for Military Reliability

First, alliance commitments could affect public support for war by raising concerns about a country's reputation as a reliable military partner.⁸ A poor reputation could undermine the ability to attract military cooperation down the road, because states form alliances only when “they believe there is a reasonable probability of successful cooperation” (Crescenzi et al. 2012, 263), and even if willing to align, demand costly terms from countries with flawed reputations.⁹ Moreover, the perception that a country will disregard alliance commitments could embolden aggressors, weakening its national security.

We extend these ideas to domestic politics by hypothesizing that voters will internalize the international reputational costs of violating alliance commitments.¹⁰ A large literature has demonstrated that voters punish leaders who tarnish their country's reputation. For example, voters impose “domestic audience costs” on leaders who make empty threats, in part because of the reputational damage empty threats would incur.¹¹ Building on this idea, we argue that voters anticipate that breaking an alliance would weaken their country's reputation for being a reliable military ally, and that voters will weigh these reputational concerns when deciding whether to support military intervention.¹² In sum, alliances will increase public support for military intervention by raising the specter of reputational damage.

Our hypothesis, while intuitive, is not foreordained. Citizens might think that skirting an alliance obligation in one specific case would not harm their country's reputation as a sound ally more generally.¹³Morrow (2000, 71–72) argues that “every decision to intervene is unique, and the interests and values that drive decisions to intervene vary from case to case.” If voters expect foreign countries to see a particular act of betrayal as sui generis, rather than evidence of a general proclivity to break promises and abandon allies, the mere existence of an alliance might not affect voter calculations about whether to support a particular war.

Alternatively, voters might anticipate reputational damage but let other considerations, such as financial costs or strategic interests, drive their opinions about war. In other words, voters might believe that breaking an alliance would harm their country's reputation as an ally, yet place little weight on that factor when judging a potential military intervention.

Reputation for Nonmilitary Reliability

Alliance behavior could also influence a country's reputation in other areas. Jervis (1989) suggested that countries can form a “signaling reputation,” a general reputation for keeping their word. Others have claimed that cheating in one area of international relations could tarnish a country's reputation in other domains, a phenomenon known as “reputation spillover” (Cole and Kehoe 1998; Guzman 2008, 103). We develop these ideas in the context of alliances. Voters, we argue, may worry that reneging on an alliance could make it harder to recruit partners not only in military affairs, but also in nonmilitary contexts like trade, finance, immigration, and the environment.

As with military reliability, however, this hypothesis is not self-evident. Citizens might be skeptical that reneging on a military agreement would hurt the country's reputation in other areas. Downs and Jones (2002) contend that the reputational consequences of breaking an international agreement are usually limited to similar agreements and voters might follow similar reasoning. Or, as Goldsmith and Posner (2005, 102) put it, “it is not clear how much the violation of one treaty says about a state's propensity to violate other treaties” (see also Posner and Sykes 2013, 33). Moreover, even if voters anticipate reputational spillover, they may focus on other considerations when judging potential military interventions. Our empirical tests allow us to adjudicate between these competing possibilities.

Moral Obligation

Finally, citizens could deem it morally wrong to break a promise, even in the absence of reputational consequences. Although morality is largely absent from the literature on alliances,¹⁴ a growing body of scholarship argues that moral considerations influence public thinking about foreign policy (Hurwitz and Peffley 1987; Hermann, Tetlock, and Visser 1999; Herrmann and Shannon 2001; Liberman 2006; Tomz and Weeks 2013; Wallace 2013; Kertzer et al. 2014; Reifler et al. 2014; Kreps and Wallace 2016; Kreps and Maxey 2017). Drawing on this literature, we propose that alliances could alter public preferences by triggering perceptions of a moral obligation to intervene.

How, specifically, could alliances activate moral concerns? Recent scholarship on moral foundations theory (MFT) argues that conceptions of morality are based on five or six “moral foundations” (Haidt and Joseph 2004; Graham et al. 2013). Two foundations, care/harm and fairness/cheating, invoke longstanding Western theories about “how individuals ought to relate to, protect, and respect other individuals” (Graham et al. 2013, 59). The care/harm foundation holds that people should help rather than harm others and the fairness/cheating foundation emphasizes the importance of equal treatment and reciprocity. Other MFT principles relate to membership in groups and deference to authorities: loyalty/betrayal (associated with virtues such as loyalty and patriotism), authority/subversion (associated with obedience and deference), and sanctity/degradation (related to qualities such as chastity, piety, and cleanliness). Finally, some have proposed a sixth moral foundation, liberty/oppression, which emphasizes freedom from domination and coercion (Haidt 2012).

We predict that alliances will activate several of these moral foundations. First, alliances should raise concerns about fairness toward citizens in the other country. Alliances are typically reciprocal, in that countries promise to defend each other in the event of attack. Countries and their citizens profit from having allies, for example, because alliances deter foreign aggression. Failing to defend an ally could be viewed as unfair: reaping the benefits of an alliance but refusing to pay the costs.

We also predict that alliances will prompt questions of loyalty. As Haidt and Graham (2007, 105) write, “Because people value their ingroups, they also value those who sacrifice for the ingroup, and they despise those who betray or fail to come to the aid of the ingroup, particularly in times of conflict.” If alliances create an “ingroup” comprising citizens of the allied nations, failing to help an ally could be viewed as an immoral betrayal of ingroup members.

Alliances could also raise issues related to the authority moral foundation, which involves respect for hierarchy and the rule of law. The public could reason that abrogating an alliance violates international law—contravenes the international legal principle that agreements must be kept—and is therefore morally wrong. Finally, rebuffing an ally could generate concerns about care. Voters might reason that spurning an ally generates greater harm than rejecting a non-ally, because the ally was operating under the assumption that the alliance would be upheld. As Crescenzi et al. (2012, 260) put it, “the failure of an alliance likely renders the abandoned partner more vulnerable than it was prior to its formation.”

It is difficult to imagine why alliances would elicit concerns about the remaining two foundations: liberty and sanctity. The liberty foundation involves freedom from domination. Voters might feel a moral duty to protect citizens of other countries from domination by an invading army, but it is unclear why alliances would heighten those concerns. It also seems improbable that alliances would trigger concerns about sanctity, which has to do with spiritual and bodily purity. Our experiments test how alliances affect the six moral foundations and how those foundations, in turn, shape views about the morality of intervention.

Context and the Effects of Alliances

Previous research has found that support for intervention varies with the stakes of the dispute, the anticipated cost of intervention, and the regime type of the country needing help. We hypothesize that these same contextual variables should moderate the effect of alliances. In general, alliances should make more of a difference when there are not already good reasons for war than when the case for war is already compelling because important US economic and security interests are at stake, the costs of intervention are expected to be low, and/or intervention would help a democracy. Put differently, when support for intervention is already high, there is a ceiling on the extent to which alliances can have an additional effect.¹⁵

As a corollary, we also predict that alliances will temper the explanatory power of contextual variables. If citizens insist that governments honor their alliance commitments even when doing so would be inconvenient, the presence of an alliance should make costs, stakes, and regime type less reliable predictors of support for war.

These hypotheses have methodological implications for research about international commitments. If commitments and context are partial substitutes for one another, one cannot fully understand the effects of commitments by regressing outcomes on commitments while treating contextual features as additive control variables. Likewise, one cannot fully understand the effects of context by regressing outcomes on contextual variables while treating commitments as additive controls. It would be more informative to interact commitments and context. By modeling the interplay between commitments and context, one can see how each moderates the other, thereby revealing when changes in commitments—or changes in context—would be more or less consequential.

Our hypotheses about the interaction between alliances and context also put previous research about alliance reliability—the likelihood that a state will uphold an existing alliance rather than violate it—in a new light. Previous research has found that countries honor their formal alliance agreements most of the time (Leeds, Long, and Mitchell 2000).¹⁶ But showing that countries honor their alliances does not prove that alliances change behavior. Suppose that, when the stakes are high, the costs are low, and the country needing defense is democratic, voters support intervention at exactly the same rate, whether or not the victim is an ally. In such circ*mstances, alliances would be reliable but inconsequential. That is, the ally would send aid reliably, but contextual factors would have caused the same behavior even if no alliance existed, rendering the apparent correlation between alliances and intervention spurious. To find out if and when alliances are consequential, one must test whether, for each potential configuration of contextual variables, support for war is greater when alliances exist than when they do not.¹⁷ An experimental approach is ideally suited to assess these counterfactuals.

Research Strategy

To study how alliances affect support for war, we conducted survey experiments in the United States. Respondents read a hypothetical situation in which one country invaded another in an effort to seize territory. We randomized whether the United States had a formal alliance with the invaded country and contextual factors that could make alliances more or less consequential. After describing the scenario, we measured support for US military intervention to stop the invasion. We also measured perceptions of three mechanisms: how the US response would affect America's reputation for upholding alliance agreements; how it would affect America's reputation in the nonmilitary realm; and whether the United States had a moral obligation to intervene.¹⁸

Our main experiment was administered to a nationally representative sample of 1,200US adults by YouGov in April 2017.¹⁹ In the scenario we presented, the leader of a country wanted more power and resources, so he sent his military to attack another country and take part of that country's territory. All participants read that the attacking country was nondemocratic, did not have a military alliance with the United States, and did not share many interests with the United States. In all scenarios, subjects learned that the United States shared many interests with the country that had been attacked.

To isolate the effect of alliances, we randomized whether the United States had a formal alliance with the attacked country. Half of the respondents read that the United States “does not have a military alliance” with the country that was attacked. The other half read that “the country that was attacked has a written military alliance with the United States. The agreement, which was signed and ratified three years ago, says: ‘If one member of the alliance is attacked, the other member will take all necessary actions, including the use of armed force, to defend its ally’.”²⁰ Our experiment therefore allowed us to compare how Americans would react upon learning about an attack on a formal US ally versus an attack on a country with which the US shared interests but not an alliance agreement.

In addition to randomizing alliances, we independently randomized four contextual variables—stakes, costs, regime type, and region—resulting in a 2×2×2×2×4 design. To vary the stakes of the dispute for the United States, we told half the sample, “If the attacker succeeds in taking part of the other country, this would weaken US military security and hurt the US economy.” The other half read that a victory by the attacking country would “neither weaken US military security nor hurt the US economy.” We also randomized whether the military operation would or would not be “very costly for the United States,” and whether the country under attack was or was not a democracy. Finally, we randomized whether the dispute took place in Africa, Asia, Eastern Europe, or South America, both to make the vignette more concrete and to make sure our findings were not unique to a particular region.

Our approach allows us to estimate the effect of alliances for each possible combination of contextual variables, including situations in which actual alliances might be more or less common and intervention might be more or less appealing. For example, we can estimate the effect of alliances when the stakes are high and the target is a democracy—a situation in which US alliances might be especially likely. We can also calculate the influence of alliances when contextual factors mitigate against intervention. This is informative both because alliance agreements can outlive the security conditions that originally gave rise to them and because it is important to know what would happen if leaders signed alliances that later obligated the United States to intervene in unattractive disputes.

After describing the scenario, we measured support for sending the US military to stop the invasion. There were five response options, ranging from “Favor strongly” to “Oppose strongly.”²¹ For our main analyses, we dichotomized the dependent variable: 100 if respondents favored military intervention strongly or somewhat and 0 otherwise. Coded this way, our dependent variable measures the percentage of Americans who favored military intervention, and our treatment effects are percentage-point changes in public support for intervention. Focusing on percentages simplifies the presentation and allows a natural interpretation that matches how news organizations and political analysts present public opinion data. Nevertheless, our conclusions did not change when we analyzed public opinion on a five-point scale (see the online appendix).

To estimate the effects of the randomized treatments, we regressed support for war on all interactions of alliances, costs, stakes, regime, and region, while controlling for demographic and attitudinal variables (gender, race, age, education, party identification, hawkishness, internationalism, and nationalism) that might affect support for war. This approach not only corrected for minor imbalances in treatments and demographic/attitudinal variables, but also facilitated the analysis of both main effects, and effects for each possible combination of contextual variables.²² Similarly, we estimated the conditional effects of alliances by averaging the effects in selected strata of interest.

The Effect of Alliances on Public Support for War

Figure 1 shows the effect of alliances on public support for war, averaging over the other randomized treatments. Approximately 79 percent of respondents supported war when the victim was a US ally, whereas only 46 percent supported intervention to help an otherwise identical non-ally. Thus, other factors equal, having an alliance increased public support for intervention by 33 percentage points, swinging opinion from majority opposition to majority support for war.²³ As the online appendix shows, the same conclusions held for different subsets of the population. Alliances mattered for respondents with high as well as low levels of political interest and were consequential regardless of political affiliation.²⁴

Figure 2 presents the effects of the other randomized treatments—averaging over alliance status and all of the remaining treatments—and compares them to alliances. Ceteris paribus, intervention was 15 points more popular when the stakes were high than when the stakes were low. The public was also 12 points more willing to intervene on behalf of a democracy than an autocracy and 10 points more willing to intervene when the expected costs of action were low rather than high. While these effects may initially appear modest, they average across situations with and without alliances. As we show below, the effects of stakes, regime, and cost were larger in scenarios without alliances. Finally, regional differences (with South America as reference category) were small and statistically indistinguishable from zero.

In summary, alliances shaped public support for war. Respondents attached high importance to honoring military alliances and were willing to send American forces into battle to uphold prior commitments.

Before examining how the effects of alliances varied by context, we connect our findings to the growing literature about domestic audience costs: the domestic penalties leaders would suffer for making commitments and failing to follow through (Fearon 1994). Audience cost experiments have traditionally investigated the domestic costs of reneging on military threats (e.g., Tomz 2007), but leaders could also incur audience costs for violating legal commitments (Wallace 2013), including alliance treaties.

Our data suggest that leaders would suffer substantial alliance audience costs; they would pay a much greater public opinion penalty for declining to help an ally compared to a non-ally (Chiba, Leeds, and Johnson 2015). Although our research design differed from classic audience cost experiments, in which respondents judge leaders who refrain from acting after committing or not committing at an earlier stage, one can—with some assumptions—translate our estimates into the metric of audience costs.

Suppose that participants who did not favor intervention would have approved of a leader who stayed out of the conflict. By that assumption, 100 − 46=54 percent would have approved of staying out absent a prior alliance commitment, compared with only 100 − 79=21 percent given a prior alliance commitment. Thus, a leader who refrained from helping an ally would be 54 − 21=33 points less popular than a leader who refrained from helping a non-ally in an otherwise equivalent crisis.²⁵ Our experiments could, therefore, be interpreted as extending research on audience costs to include formal alliance commitments.

Does Context Moderate the Effect of Alliances?

In this section, we investigate whether alliances could be more consequential in some contexts than in others. We hypothesized that alliances would be more important when the case for war is weak than when war is more appealing. As a corollary, we predicted that alliances would reduce the explanatory power of contextual variables.

Figure 3 depicts support for war given different combinations of alliances and our three contextual variables: the stakes for the United States, the expected costs of military intervention, and the regime type of the victim. We averaged over the fourth contextual variable, region, which proved relatively unimportant.²⁶ The hollow markers measure support for war in scenarios without an alliance, while the solid markers represent support for war in scenarios with an alliance.

In the absence of alliances, context powerfully influenced public opinion. The hollow marker in the top left shows that when stakes for the United States were low, the victim was an autocracy, and the costs of intervention were high, only 18 percent of respondents supported military intervention to defend a non-ally. As background conditions changed, support for war surged. For instance, enthusiasm for war was much greater when the stakes were high (bottom half of the figure) than when the stakes were low (top half), a pattern that held for every combination of regime type and costs. Regime type and cost operated similarly, driving support for war not only on average, but also for each combination of the other contextual variables.²⁷

As predicted, however, contextual variables became far less important when we introduced alliances. The public was not substantially more willing to defend an ally when the stakes for the United States were high than when the stakes for the United States were low. Likewise, opinions were not systematically different when the ally was a democracy rather than an autocracy and costs mattered only when the stakes were low. In general, alliances deprived contextual variables of their explanatory power.

These same data show that alliances exerted stronger effects in some contexts than in others.Figure 4 displays the data fromfigure 3 as treatment effects; each dot indicates how alliances increased support for intervention under the stated conditions, relative to an otherwise identical situation without an alliance commitment. The first row shows that when the stakes for the United States were low, the costs of intervention were high, and the victim was an autocracy, alliances boosted support for war by 50 percentage points. The bottom row shows the opposite situation: when the stakes were high, the costs low, and the victim a democracy, alliances still moved opinion, but by a smaller margin (15 points).

Using the estimates from figures 3 and 4, one can speculate about how public support for intervention would vary in crises involving real countries. Imagine that China attacked Cambodia, a non-democracy with whom the United States does not share an alliance. Assuming the costs of fighting would be high and stakes for the United States would be relatively low, only 18 percent of Americans would support intervention in this kind of situation. What if China instead attacked Thailand, a nondemocratic ally? Assuming similar costs of war and similar stakes for the United States, 68 percent of Americans would support intervention, a swing of fifty points.²⁸

One can also speculate about how alliances would affect public willingness to defend real-world democracies. Imagine a Russian attack on Sweden, a democratic non-ally. Assuming high stakes and high costs, 60 percent of voters would support intervention even though the United States was not obligated by treaty to defend Sweden. If instead Russia targeted the UK, a democratic ally, support would be twenty-four points higher. Thus, majorities would support intervention in both cases, but alliances would cause the majority to swell.

Our findings have several implications. First, alliances are powerful commitments that can sway public opinion even—and especially—when war would otherwise seem unattractive. In our study, alliances made the biggest difference when the United States did not already have strong military, economic, and political reasons to intervene. This domestic dynamic could help explain why democratic leaders are particularly selective about the alliances they sign and validate concerns about alliances that survive beyond the conditions in which they were signed. At the same time, alliances were consequential even when war seemed relatively attractive—potentially the circ*mstances under which alliances are the most common in practice. Thus, unlike mere scraps of paper, alliances can be potent in a wide range of circ*mstances.

Second, our findings underscore the difference between alliance reliability and alliance effectiveness. When the target was a democracy, the stakes were high, and the costs were low, 89 percent of voters supported intervention to help an ally, implying that the alliance was highly reliable. But 73 percent of voters would have supported intervention even absent an alliance, implying that the alliance, though reliable, was not as consequential in this context as in others. As this example illustrates, showing that an alliance is reliable says little, on its own, about whether the alliance made a difference. One must test, for each configuration of contextual variables, whether support for war is substantially higher in the presence of alliances than in their absence.

Finally, our findings about context contribute to the literature about domestic audience costs in international relations. Experiments have shown that the costs of reneging on military threats vary by context. Tomz (2007, 831), for example, found that audience costs were larger in scenarios involving low stakes. Likewise, Davies and Johns (2013) found that audience costs were larger in their hostage and invasion scenarios, where most citizens approved of staying out absent a military threat, than in nuclear scenarios, where most citizens disapproved of staying out even in the absence of an antecedent threat.

Our research further demonstrates the contextual nature of audience costs. Once again, suppose that participants who did not favor intervention would have approved of a leader who abstained from conflict. If the stakes were low, the costs of war were high, and the target was an autocracy (top row offigure 3), a leader who stayed out would enjoy 100 − 18=82 percent approval without an alliance, versus 100 − 68=32 percent approval with an alliance, implying an audience cost of 50 percentage points. If instead the stakes were high, the costs of fighting were low, and the target was a democracy (bottom row offigure 3), the audience cost would have been less than a third as large. Thus, our findings suggest the intriguing hypothesis that the more citizens would have supported war in the absence of a commitment, the smaller the audience costs of reneging on the commitment.

Why Do Alliances Affect Support for War?

We next investigated three mechanisms through which alliances could drive public support for war: reputation for military reliability, reputation for nonmilitary reliability, and moral obligation. We measured these three potential mediators by asking how much respondents agreed or disagreed with each of the following statements: (1) “If the United States does not send its military, other countries will doubt America's willingness to honor military alliance agreements in the future”; (2) “If the United States does not send its military, other countries will doubt America's willingness to honor nonmilitary agreements in areas such as trade or the environment”; and (3) “The United States has a moral obligation to send its military to defend the country that was attacked.”²⁹ In each case, we used the answers to construct a scale with five levels: disagree strongly (0), disagree somewhat (25), neither agree nor disagree (50), agree somewhat (75), or agree strongly (100).³⁰

We regressed each mediator on all interactions of alliances, costs, stakes, regime, and region, while controlling for demographic and attitudinal variables (gender, race, age, education, party, hawkishness, internationalism, nationalism) that might affect how respondents thought about reputation or morality. From the regressions, we estimated the main effect of alliances on each mediator by averaging over the effects in the 2×2×2×4 strata defined by the other randomized treatments.³¹

Figure 5 shows that respondents were thirty-eight points more likely to agree that staying out would hurt America's reputation for military reliability when the victim was an ally, than when it was not. Respondents also felt that abandoning an ally would taint America's reputation for nonmilitary reliability; concerns about reputational spillovers were seventeen points higher when the United States had previously made an alliance commitment. Citizens expected more damage to America's military reputation than to its nonmilitary reputation, suggesting that reneging causes more reputational damage within the immediate issue area than across issue areas.

Finally, alliances generated a strong sense of moral obligation. Respondents were twenty-seven points more likely to perceive a moral obligation to intervene when the victim was an ally than when the victim was not. Overall, figure 6 confirms that alliances affected perceptions of all three mediators: reputation for military reliability, reputation for nonmilitary reliability, and moral obligation.

Moving down the causal chain, how did these mediators affect support for war?³² The online appendix shows that, other factors equal, willingness to intervene was 49 percentage points higher among people who strongly agreed that inaction would undermine America's reputation for honoring military alliances than among people who strongly disagreed that staying out would cause this kind of reputational damage. In contrast, support for war was not significantly higher among respondents who anticipated that inaction would cause reputational spillovers than among people who doubted reputational spillovers. Finally, independent of reputational concerns, citizens who perceived a moral obligation to intervene gave thirty-eight points more support for war than citizens who denied any moral obligation.

Finally, we used these estimates to infer the importance of each causal mechanism.³³ Recall that alliances increased public support for intervention by thirty-three points. Approximately 57 percent of this effect flowed from concerns about a reputation for being a reliable ally. To see why, note that the expected damage to America's reputation for military reliability was thirty-eight points higher in scenarios with an alliance than in scenarios without an alliance. Moreover, a one-unit change in expected damage to the country's reputation for military reliability was associated with a 0.49-point surge in support for intervention. Combining these estimates, we get 38×0.49=19, representing 19 ÷ 33≈57 percent of the total effect of alliances.

Using a similar procedure, we calculated that fears of reputational spillovers mediated only 4 percent of the total effect of alliances. This was because, although alliances raised expectations about reputational spillover, fears of spillover had little relationship with support for war. Finally, alliances generated a sense of moral duty, which was strongly related to attitudes about intervention. By our calculations, this moral pathway accounted for 31 percent of the total treatment effect. We present these estimates infigure 6, which suggests that alliances drove public opinion primarily by changing perceptions of moral obligation and America's reputation as a reliable ally.

Thus, our findings show strong support for two of our hypotheses about mechanisms: that alliances influence support for war by raising fears about America's reputation as a reliable ally and by triggering a feeling of moral duty. In contrast, we found relatively little support for our hypothesis that alliances mattered because of concerns about reputational spillover. Voters anticipated negative spillovers but placed little emphasis on them when weighing military intervention.

What Drives Perceptions of Moral Obligation?

We found evidence that alliances affect support for war partly by triggering a sense of moral obligation. Why do citizens regard alliance commitments as morally binding? To find out, we fielded a follow-up study to 2,703US adults, recruited via Lucid in December 2017 and June 2018.³⁴ We began with text identical to our main YouGov study and found that our previous estimates about the effects of alliances, the role of context, and the underlying mechanisms replicated well (see the online appendix).

The follow-up study concluded with a battery of questions designed to gauge which specific moral foundations influenced perceptions of moral obligation. We asked whether failing to intervene would violate principles of fairness, loyalty, authority, care, liberty, and sanctity, the six main constructs in MFT. For example, we captured beliefs about fairness by asking how much respondents agreed or disagreed that “If the United States does not send its military, this would be unfair to the country that was invaded.” Similarly, we studied perceptions of loyalty by measuring reactions to the claim, “If the United States does not send its military, this would be disloyal to the country that was invaded.”³⁵ All six moral foundation items were coded on a five-point scale from 0 (disagree strongly) to 100 (agree strongly).

We regressed each moral foundation on all interactions of alliances, costs, stakes, regime, and region, controlling for demographic and attitudinal variables (listed earlier).Figure 7 gives the average effect of alliances on each moral foundation. As hypothesized, respondents were far more likely to view nonintervention as unfair and disloyal when the victim was an ally than when the victim was not. Alliances also raised concerns about authority; subjects were substantially more likely to think staying out of the conflict would “show disobedience to higher authorities” if the United States had previously cemented an alliance agreement. This effect makes sense to the extent that citizens view breaches of treaties as violations of international law.

Alliances had a more modest effect on perceptions of care. In vignettes with alliances, respondents were eight points more likely to agree that “If the United States does not send its military, the country that was invaded would suffer serious harm.” This effect, though relatively small, fits the hypothesis that allies might suffer disproportionately if abandoned, because they had made military plans on the assumption that the ally would uphold its side of the agreement.

Surprisingly, alliances also raised concerns about liberty. Subjects were twelve points more likely to believe failing to intervene “would be a threat to the liberty of people in the invaded country” when that country was an ally, though it is not clear why they would have this perception. Finally, as expected, we found no evidence that alliances increased concerns about sanctity/purity, operationalized as the belief that nonintervention would “increase the spread of germs and diseases.”

We next studied the relationship between the six moral foundations and overall beliefs about a moral obligation to intervene (see the online appendix).³⁶ Other factors equal, perceptions of fairness and loyalty were strongly associated with the belief that the “United States has a moral obligation to defend the country that was attacked.” Three other moral foundations—authority, care, and liberty—had smaller but still positive relationships with perceptions of moral obligation, while sanctity did not appear to drive thinking about the morality of military intervention.

Finally, we estimated the importance of each causal pathway, using methods described earlier. Asfigure 8 shows, alliances appeared to affect overall beliefs about morality primarily by raising concerns about fairness and loyalty. Together, these two factors mediated nearly 60 percent of the effect of alliances on moral obligation. Three other moral foundations—authority, care, and liberty—played small roles, each accounting for 3–5 percent of the total effect, and concerns about sanctity did not appear to mediate the effect of alliances on moral obligation.

In summary, alliances raise concerns about fairness and loyalty, contributing to the sense of moral obligation that helps make alliances bind. These findings underscore the importance of studying moral inclinations when explaining attitudes about foreign policy (Kertzer et al. 2014; Kreps and Maxey 2017). Although fairness and loyalty have received relatively little attention in the literatures on alliances and military intervention, our findings reveal that these are important considerations in the public mind.

Our results suggest both challenges and opportunities for leaders seeking to sway public opinion about supporting an ally. When voters see policies in terms of moral values, they report being much less open to changing their minds than when issues do not involve questions of morality (Mooney and Schuldt 2008). Indeed, people express “moral outrage” at the suggestion that nonmoral considerations such as economic costs could eclipse moral concerns (Tetlock et al. 2000). Leaders wishing to mute public support for intervention would therefore need to explain why refusing to help an ally would be neither unfair nor disloyal. On the other hand, when voters do not already view an issue in moral terms, moral rhetoric can shift public opinion (Feinberg and Willer 2013). In some contexts, therefore, moral rhetoric might shape the public debate.

Our findings about the role of moral reasoning also shed light on contemporary debates about US alliances such as NATO. On the one hand, our findings suggest that if a US president failed to defend an ally, the backlash in terms of public opinion could be significant, with voters decrying such behavior as disloyal and unfair. On the other hand, former President Donald Trump frequently said it was unfair to expect the United States to defend NATO allies who are spending less than 2 percent of annual GDP on defense. Such rhetoric has the potential to undercut Americans’ perception of a moral obligation to help those countries militarily, which, according to our findings, could erode public support for defending NATO down the road.

The Political Importance of Public Opinion about Alliances

Our experiments showed that alliances increase public support for war, demonstrated how the effects of alliances vary by context, and revealed why alliances shape public opinion. In this section, we discuss how public opinion about alliances could influence state behavior and why citizens would be likely to know about alliances during military crises.

Public opinion could influence state behavior by affecting the incentives of elected officials. Research has found that democratic officeholders pay close attention to public opinion and expect serious domestic political costs for contradicting the public on matters of war (Reiter and Stam 2002). To measure the anticipated costs, Tomz, Weeks, and Yarhi-Milo (2020) surveyed current and former members of the Israeli Knesset about what would happen if the government waged a war the public opposed or abstained from a war the public wanted. In both scenarios, more than 90 percent of the Knesset sample thought it was at least somewhat likely—and more than half said it was very or extremely likely—that the government would lose public support, lose seats in the legislature, find it difficult to pass other policies, and/or fall from power. Given these concerns, democratic leaders would have powerful incentives to respect public opinion about alliances.

Of course, public opinion operates alongside other incentives leaders might have to wage war or stay out. We found that alliances shape public opinion by raising concerns about reputation and morality. If leaders independently felt similar concerns, they might respond to alliances in similar ways even in the absence of public pressure.³⁷

Nevertheless, we offer four observations. First, situations could arise in which elected officials disagree with voters about military force. Polls have shown that elites and masses do not always see eye-to-eye about war (e.g., Holsti 2004; Page and Bouton 2006). Reflecting on these differences, elites have long criticized the American public for being excessively moralistic and legalistic, and insufficiently attuned to the national interest (e.g., Lippmann 1955; Morgenthau 1985). We found a powerful public urge to honor alliances, not only under favorable conditions, but also when elites might shy from conflict because the stakes were low and the costs were high. Thus, public pressure could incentivize leaders to uphold alliances when they might otherwise be tempted to avoid conflict for realist or idiosyncratic reasons.³⁸

Second, even when elites and masses agree, public opinion could reinforce existing incentives. The fear of public backlash—and the allure of public approval—provide additional reasons to honor alliances, supplementing reasons leaders might already have. If instead our experiments had shown little public enthusiasm for upholding alliances, public opinion would have pulled in the opposite direction: undermining, rather than augmenting, any existing incentives leaders might have to wage war on behalf of formal allies.

Third, to the extent that elected officials think like ordinary citizens, the correspondence could reflect a selection mechanism, in which citizens choose leaders who share their preferences. Studies have shown that citizens elect politicians based on expectations about the kinds of foreign policies they would pursue (e.g., Tomz, Weeks, and Yarhi-Milo 2020). Thus, our research could help explain why alliance-respecting leaders tend to win democratic elections.

Finally, even if public opinion had no effect on the incentives of leaders or the outcomes of elections, understanding public opinion would be essential for judging the representativeness of democracies. By estimating public support for war across a wide range of conditions, with and without alliances, our experiments provide a normative benchmark for determining which decisions might reflect or contradict the will of the people.

Our experiments showed how the public would respond to information about alliances. In practice, though, one might wonder whether the public would actually know whether the United States had an alliance treaty with a country under attack. Logically, there are three pathways by which voters could learn about US alliance obligations. First, voters might gain this knowledge prior to any crisis, through formal education or by following the news. In the online appendix, we report the results of a survey we fielded in September 2020 to measure American knowledge of alliances. On average, voters correctly classified allies and non-allies, even in the absence of a crisis that would make alliances salient.

Second, political elites would have strong incentives to educate the public once a crisis emerged. At the outset of a crisis, US leaders might publicize the alliance for its deterrent value, reminding potential aggressors that the United States is treaty-bound to defend the victim in the event of an attack. If an attack nonetheless occurred, proponents of US intervention would find it politically beneficial to emphasize any relevant alliance commitments. For example, President Eisenhower explained in a major speech at the outset of the 1958 Taiwan Straits crisis—one of the few cases in which a US alliance was tested—that the 1955 alliance treaty obligated the United States to defend Taiwan. Conversely, opponents of intervention would have reason to emphasize the absence of alliances. In the Taiwan case, elites noted that the American commitment to Taiwan did not cover the offshore islands under direct attack by Communist China.

Finally, during a crisis, we expect that the media would cover alliance commitments. Indeed, during the 1958 Taiwan crisis, news articles discussed the specifics of the US–Taiwan treaty from the very first day of the hostilities. In sum, during actual crises many voters would probably know—based on preexisting knowledge, elite rhetoric, and/or the media—whether the United States was bound by an alliance.

Conclusion

When and why do military alliances affect public opinion? Our experiments revealed three main findings. First, having a military alliance profoundly influenced support for war. On average, Americans were 33 percentage points more supportive of defending a US ally than an otherwise identical non-ally. Second, alliances mattered in all contexts but held the most sway when circ*mstances least favored war. Third, alliances shaped public opinion about war by raising concerns about America's reputation for military reliability and by generating a sense of moral obligation. When we investigated the roots of these moral beliefs, we found that alliances activated two important moral foundations: fairness and loyalty.

Our findings have implications for both research and practice. First, our results provide microfoundations for the empirical finding that democracies make particularly reliable allies (Leeds 2003b). We found that a strong majority of the public supported intervention on behalf of allies, suggesting that democratic leaders would face backlash for leaving an ally hanging. To the extent that democratic leaders are more sensitive to public pressure, our findings could help explain why democracies are less likely to break their alliances than nondemocracies.³⁹

Our results also furnish domestic political foundations for the deterrent value of alliances (Leeds 2003a; Johnson and Leeds 2011). If potential aggressors understand that alliance agreements increase domestic political support for helping the victim and raise the domestic costs of staying out, alliances could discourage attacks and help maintain peace in the international system.⁴⁰ Our findings therefore provide microfoundations for the idea that alliances, like public threats, are potent tools of international politics.

At the same time, one might wonder why democratic leaders occasionally violate their alliance commitments and why aggressors do not always see alliance commitments as credible. Our findings help answer this puzzle as well. By testing the effects of alliances in many contexts, we identified scenarios in which the domestic penalty for reneging on an alliance would be relatively small. Respondents were most amenable to violating an alliance when the costs of intervention were high, the stakes of intervention were low, and the ally was nondemocratic. In these types of situations, a democratic leader might calculate that the benefits of violating an alliance outweigh the domestic political costs, tempting foreign aggressors.

Our analysis of mechanisms provides further clues about alliance violations, by suggesting that citizens would tolerate violations that do not threaten reputation, fairness, or loyalty. For example, voters might have fewer reputational and moral concerns about violating an alliance if conditions had changed since the agreement was signed (Leeds 2003b; Leeds and Savun 2007; Leeds, Mattes, and Vogel 2009); if the agreement contained vague language and escape clauses (Leeds 2003a; Benson 2012; Beckley 2015; Chiba, Johnson, and Leeds 2015; Fjelstul and Reiter 2019); or if the commitment was merely an informal promise or a crisis-induced coalition (Wolford 2015). Future studies could randomize these features and test their effects on public opinion.

Our findings not only illuminate the domestic incentives to honor alliances, but also help explain why democratic leaders tend to choose and design alliances carefully (Chiba, Johnson, and Leeds 2015; Fjelstul and Reiter 2019). Traditional explanations for alliances focus on international strategic factors,⁴¹ but leaders also have domestic political reasons to be cautious. If the public thought that alliances were merely scraps of paper, democratic leaders would have the same flexibility as nondemocratic leaders to forge alliances with countries they would rather not defend, knowing there would be few domestic repercussions for reneging. On the contrary, we found that alliances moved public opinion even—and especially—when intervention would otherwise seem imprudent. Given the risk of electoral backlash, democratic leaders may need to choose allies and design agreements more carefully than nondemocratic leaders. By studying alliances across a wide range of circ*mstances, including ones in which intervention might be inconvenient and actual alliances might be rare, we learned not only about the potential effects of alliances, but also about the incentives to sign or avoid alliances in the first place.⁴²

Our work contributes not only to the study of alliances, but also to the literature on audience costs, i.e., the domestic penalties leaders would face for reneging on international commitments. Previous experiments estimated the audience costs of reneging on military threats. We find audience costs for reneging on alliance treaties as well. At the same time, our experiments underscore how the magnitude of audience costs depends on context. Our findings suggest that the more citizens support action—military or otherwise—in the absence of a commitment, the smaller the domestic audience cost leaders would pay for reneging on a commitment.

Our experiments also suggest how audience costs could vary with the nature of the commitment. Earlier studies found that reputational concerns motivate voters to impose audience costs on leaders who back down from threats (e.g., Tomz 2007; Brutger and Kertzer 2018). We find that audience costs can also arise from a perceived moral obligation to honor existing commitments. We expect, therefore, that the magnitude of audience costs will vary across commitments, depending on how seriously the commitments raise concern not only about reputation, but also about morality. Future research could, for example, compare the audience costs associated with threats versus promises or formal commitments versus informal ones.

Finally, our experiments contribute to the broader literature about international law.⁴³ As Martin (2012) has noted, the most interesting questions about international law are about when and how legal commitments shape state behavior. By varying the existence of alliances, our experiments provide micro-evidence of the mechanisms through which, and conditions under which, international law changes outcomes. While our focus was on military alliances, our insights about the role of reputation, morality, and context could apply to other legal commitments as well.

Future research could explore additional questions about alliances and public opinion, including the effects of elite rhetoric. Studies have shown that leaders who backtrack on international commitments can mitigate public disapproval by providing justifications (Levendusky and Horowitz 2012). Knowing this, leaders who violate alliances could attempt to minimize the damage by misrepresenting the terms of the alliance, claiming that circ*mstances have changed, or painting the ally's behavior in an unflattering light. Political opponents, on the other hand, could stoke public outrage by highlighting the reputational consequences and moral ramifications of breaching an alliance. Future research could evaluate whether, and when, elite cues moderate the effects of alliances on public opinion.

Future research could also explore how the origins and nature of the conflict influence public reactions to alliances. In the experiments we conducted, the ally was attacked. Future experiments could examine how alliances affect the likelihood of intervention when the ally instigates the dispute, to find out whether alliances not only entangle but also entrap (Kim 2011). Finally, scholars could investigate whether alliances have similar effects in other countries. We fielded our study in the United States, a superpower and one of the most militarily active democracies in the world. Researchers could replicate our experiments on other samples—including citizens of other countries and foreign policy elites—to explore the effects of alliances in countries that are militarily weaker or that differ from the United States in other ways.

Observers often decry world politics as a realm devoid of enforcement. Without a central authority to punish countries that renege on their promises, why should international commitments such as alliances carry any weight when it comes to life-or-death decisions such as sending military forces into battle? Our findings suggest that by engaging concerns about reputation and morality, military alliances have potent effects on public support for military intervention. The power of alliances to sway public opinion should be taken lightly neither by leaders contemplating new agreements nor by foreign powers tempted to test existing ones.

2

The domestic public is not the only relevant audience. Future research could explore how foreign publics/elites react to alliance violations.

5

We elaborate on the relationship between the public and elites later in the article.

6

We study formal alliance treaties. Future research could investigate the effects of informal alliance commitments.

7

Future research could explore other mechanisms. For example, alliances could increase public confidence that US intervention would succeed, due to peacetime military coordination between the United States and its allies (Johnson, Leeds, and Wu 2015).

12

Of course, domestic audiences care about reputation even when leaders have not made commitments (Snyder and Borghard 2011). We argue, however, that alliances heighten concerns about reputation by explicitly putting the nation's reputation on the line.

15

Moreover, this logic suggests that alliances and contextual variables can act as substitutes for each other. We thank an anonymous reviewer for making these points.

17

For a similar argument about “compliance” versus the effects of international law and institutions, see Martin (2012).

18

To avoid priming subjects, we measured the mediators after measuring the dependent variable. Future research could evaluate whether this design choice influenced our estimates of causal mechanisms (Chaudoin, Gaines, and Livny forthcoming).

19

The field period for the YouGov survey was April 7–29, 2017. YouGov interviewed 1,388 respondents and matched them by gender, age, race, education, party identification, ideology, and political interest down to a final sample dataset of 1,200 to produce the final dataset. The respondents were matched to a sampling frame on gender, age, race, education, party identification, ideology, and political interest. For full details about the questionnaire and sampling, see the online appendix.

20

We modeled this language on the North Atlantic Treaty.

21

At the end of the experiment, we administered five attention checks, shown in the online supplementary materials. YouGov delivered a nationally representative sample of 1,200 respondents who answered at least four attention checks correctly.

22

On using linear regression to estimate average marginal and conditional effects of all randomized treatments, see Hainmueller, Hopkins, and Yamamoto (2014). Based on the regression, we estimated the main effect of alliances as the average of the effects in the 2×2×2×4=32 strata defined by the other randomized treatments. Our subclassification estimator gave equal weight to each of the thirty-two strata.

23

If we had performed these calculations without controlling for demographic/attitudinal variables or other randomized treatments, the effect of alliances would have been the same. Our conclusions were also robust to using sampling weights or a five-point scale for the dependent variable (see the online appendix).

24

The effects of alliances among Democrats, Independents, and Republicans were all above 30 percentage points. See the online appendix for these and other subgroup analyses.

25

Thus, contra Snyder and Borghard (2011), domestic audiences cared not only about policy substance, but also about consistency between word and deed.

26

The effects of alliances were similar regardless of geographic region. See the online appendix.

27

The effects were always in the expected direction, though not always significant at p<.05.

28

The comparison also assumes the United States shares similar interests with Thailand as with Cambodia.

29

We asked respondents to react to each mediator individually, rather than choosing among them.

30

We used five-point scales to increase the precision of the mediation analysis.

31

If we had not run regressions, but instead calculated differences in the mean values of the mediators with and without alliances, the estimates would have been nearly identical.

32

We treat our analysis of this part of the causal chain with caution, as we measured the values of the mediators rather than randomly assigning them. The effects of the mediators on support for war are, therefore, causally identified only given the untestable assumption that the observed values of the mediators were independent of the treatment condition and all pretreatment confounders (Imai, Keele, Tingley, and Yamamoto 2011). Including demographic controls increases our confidence that this assumption holds, but future research employing alternative experimental designs (e.g., Acharya, Blackwell and Sen 2018) could further explore how the mediators affect support for war.

33

We calculated the strength of each pathway using the product-of-coefficients method (Baron and Kenny 1986). The online appendix shows that our conclusions remained the same when we instead applied the Imai et al. (2011) potential outcomes framework to nonlinear probit models.

34

Lucid used quota sampling to produce a sample that was diverse with respect to gender, age, ethnicity, and geographic region.

35

The answer options for each question were “Agree strongly,” “Agree somewhat,” “Neither agree nor disagree,” “Disagree somewhat,” and “Disagree strongly.” See below, and the online appendix, for the wording of other moral foundations.

36

As before, these estimates and the full causal mediation analysis are causally identified only under the untestable assumption that the observed values of the mediators were independent of treatment status and all pretreatment confounders (Imai et al. 2011).

38

If, as realists claim, the public is more moralistic than elites, then the effect of public opinion on elite incentives may be greater in situations where nonmoral reasons for compliance (such a concerns about reputation) are weak. Thus, public opinion could prove more influential in countries that are less reliant on alliance relationships. We thank an anonymous reviewer for making this point.

40

Snyder and Borghard (2011) argue, however, that authoritarian countries may not perceive the audience costs democratic leaders might incur for violating commitments. Though beyond the scope of this paper, alliances could also embolden allies to act aggressively (e.g., Benson 2012); encourage them to solve disputes peacefully (Fang, Johnson, and Leeds 2014); or make states seem more threatening (e.g., Vasquez 1993).

41

For example, leaders consider whether alliances would bolster their country's international position, which could depend on the partner's expected reliability (e.g., Walt 1985; Crescenzi et al. 2012).

Acknowledgments

This material is based in part upon work supported by the National Science Foundation under award numbers SES-1226855 and SES-1226824. For extremely helpful comments on earlier versions of this paper, we thank participants at the American Political Science Association conference (2016, 2018), the International Studies Association conference (2015, 2018), the Midwest Political Science Association conference (2015), the Peace Science Society conference (2014), and seminar participants at Academia Sinica, Harvard University, National Taiwan University, Northwestern University, Stanford University, the University of Miami, and the University of Wisconsin–Madison. We are also grateful for helpful comments and suggestions from Matthew Baum, Jonathan Chu, David Edelstein, Joshua Freedman, Kiera Hudson, Bruce Jentleson, Josh Kertzer, Ashley Leeds, Aila Matanock, Michaela Mattes, Harry Oppenheimer, Joe Parent, Jon Pevehouse, Phil Potter, Ryan Powers, Shawn Ramirez, Dan Reiter, Todd Sechser, and Rob Trager.

Michael Tomz is the William Bennett Munro Professor in Political Science at Stanford University and a Senior Fellow at the Stanford Institute for Economic Policy Research.

Jessica L.P. Weeks is a Professor of Political Science and H. Douglas Weaver Chair in Diplomacy and International Relations at the University of Wisconsin—Madison.

Notes

Author's note: Replication files are available on the ISQ Dataverse at https://dataverse.harvard.edu/dataverse/isq.

References

© The Author(s) (2021). Published by Oxford University Press on behalf of the International Studies Association.