Abstract

Conventional wisdom on alliances proposes that leaders comply with alliances because the public opposes violating alliance commitments. However, this assumes that the public can easily judge whether or not a particular policy violates an alliance treaty. This article challenges this assumption and develops a theory that elites have the opportunity to shape public understanding as to whether an action violates an alliance treaty. It shows that while alliance commitments continue to have an important impact on public opinion, signals from unified elites can significantly reduce public pressure to support an ally by arguing that the alliance treaty does not create a legal obligation to intervene. In a pair of experiments on large samples of American adults, we found that a unified signal from the president and the Senate opposition leader can significantly reduce support for sending troops to the embattled ally. Consistent with elite cueing theory, the president’s ability to move public opinion in this manner is eliminated if the Senate opposition leader disagrees with his argument.

El conocimiento popular en materia de alianzas sugiere que los líderes cumplen con las alianzas porque el público se opone a violar los compromisos de estas alianzas. Sin embargo, esto asume que el público puede juzgar fácilmente si una política en particular viola o no un tratado de alianza. Este artículo desafía esta suposición y desarrolla una teoría según la cual las élites tienen la oportunidad de moldear la comprensión pública con referencia a si una acción viola o no un tratado de alianza. El artículo demuestra que, si bien los compromisos de las alianzas siguen teniendo un impacto importante sobre la opinión pública, las señales de las élites unificadas pueden reducir significativamente la presión pública para apoyar a un aliado argumentando que el tratado de la alianza no crea una obligación legal de intervenir. Concluimos, mediante un par de experimentos con grandes muestras de adultos estadounidenses, que una señal unificada por parte del presidente y del líder de la oposición en el Senado puede reducir significativamente el apoyo al envío de tropas a un aliado que está siendo asediado. De acuerdo con la teoría de la señalización de élite, la capacidad del presidente para modificar de esta manera la opinión pública desaparece si el líder de la oposición en el Senado no está de acuerdo con sus argumentos.

D'après la sagesse populaire, les dirigeants respectent les alliances, car la population s'oppose aux violations d'engagements pris dans le cadre de celles-ci. Cependant, ce n'est vrai qu’à la condition que la population puisse facilement juger si une certaine politique viole un traité d'alliance. Cet article remet en question cette hypothèse avant de développer une théorie selon laquelle les élites ont la possibilité de façonner la compréhension publique d'une violation d'un traité d'alliance par une action. Il montre que bien que les engagements pris dans le cadre d'alliances aient toujours un effet important sur l'opinion publique, les signaux envoyés par les élites unifiées peuvent réduire de façon importante la pression de la population en faveur du soutien d'un allié, en affirmant que le traité d'alliance ne les contraint pas légalement à intervenir. Dans deux expériences menées sur de larges échantillons d'adultes américains, l'article conclut qu'un signal unifié du président et du leader de l'opposition au Sénat peut considérablement diminuer le soutien à l'envoi de troupes pour aider un allié assiégé. Conformément à la théorie des indices des élites, la capacité d'un président à orienter l'opinion publique de cette façon est réduite à néant si le leader de l'opposition au Sénat exprime son désaccord avec cet argument.

Introduction

An important theoretical puzzle in international relations is “Why do international institutions matter under anarchy?” Concerns about a state’s international reputation can go some way toward explaining this, but another part of the answer lies in domestic politics. The theory of audience costs posits that the public wants their government to comply with international agreements and will punish leaders who violate these commitments (Leeds 1999). This gives leaders a strong incentive to comply, especially in democratic regimes.

However, existing work on this proposition assumes that public beliefs about whether a given action violates a treaty are uncontroversial and are guided by indisputable, common-knowledge facts. This article proposes that this assumption may not be valid, both because many international treaties are not completely clear about what actions constitute compliance and because most members of the public hold relatively unsophisticated beliefs about their country’s obligations under international treaties.

It further proposes that political elites, especially national leaders, can shape the public’s understanding of whether a particular action would violate an international treaty commitment. This theory is connected to the broader theory that public opinion can be guided by elite statements, known as “elite cueing.” However, past work on elite cueing has tended to focus on the desirability of certain policies, such as whether the public should support policies that violate international treaties. There is almost no scholarship on whether political elites can shape the public’s understanding of what obligations exist in the first place. This article aims to understand whether elite signals can shape whether the public believes that a particular action violates an international treaty and whether this can in turn shape the public’s opinions on policy choices.

This article applies this theory to the specific area of international alliances. Alliance scholarship has long proposed that states, especially democracies, abide by alliance commitments because leaders fear domestic audience costs if they abandon an ally. This domestic audience cost assumption has also been used to explain a variety of different alliance patterns, including that democracies are more likely to comply with alliance agreements (Leeds 2003), alliances of democracies last longer (Leeds, Mattes, and Vogel 2009), alliances of democracies are more powerful deterrents (Kroenig 2020), democratic alliances perform better during wartime (Choi 2012), and democratic great powers are better able to build durable, constitutional postwar orders (Ikenberry 2001).

This theoretical and empirical work presumes that alliance treaties, especially those with security guarantees, impose a fairly straightforward set of obligations on their member states: Intervention on behalf of an embattled ally constitutes treaty compliance, whereas failure to intervene violates the treaty. However, in reality, most international alliances contain language that is flexible and somewhat vague. This creates opportunities for leaders to argue that a treaty might not require intervention in favor of an embattled ally, such as not specifying exactly what actions must be taken if an ally is at war. Indeed, leaders deliberately include such language in alliance treaties in order to provide them with just this sort of flexibility in the event that an alliance signatory finds itself at war and demands assistance.

This article proposes that public support for intervening can be reduced if unified elites argue that doing so is not a legal requirement of the alliance treaty. We test this hypothesis with two survey experiments on American adults, comparing public support for intervening on behalf of an embattled state that is not an ally, an embattled state that is an ally, and an embattled state that is an ally but for which the president argues that the treaty does not require intervention. Consistent with elite cueing theory, we find that certain signals reduce public support for sending troops on behalf of an embattled ally. Specifically, in one experiment, we find that if subjects are told that both the president and Senate opposition leader believe that the treaty does not require the United States to send troops to the embattled country, the effect of an alliance agreement on support for sending troops to the embattled country is reduced by about half. There is a similar effect if the president argues that the United States can fulfill its treaty requirements by sending aid but not troops, even without opposition support. However, there is no statistically significant effect if the president does not argue that aid fulfills the treaty requirement, or if the Senate opposition leader disagrees with the president’s argument. In a second experiment, we consider the difference between signals that focus on legal/compliance arguments and those that focus on interest-based arguments. Interestingly, we find that the legal/compliance arguments have a significantly larger effect on reducing domestic audience costs than the interest-based arguments. That is, elites have greater leeway to reduce public support for intervening on behalf of an ally by interpreting the compliance requirements of a treaty than by trying to argue that it is not in the national interest to defend the embattled ally.

These findings make important contributions to international relations scholarship. First, they highlight the importance of exploring how the public comes to understand the compliance requirements of treaties. Past work on domestic audience costs of noncompliance has mostly taken the question of treaty violation as a given and has then examined whether the fact of treaty violation affects public opinion. This study stresses the importance of endogenizing the public’s understanding of treaty compliance requirements. Second, the results show that while alliance treaties continue to have a strong influence on public opinion, there are some limits to their ability to tie the hands of elected governments to intervene on behalf of allies. Audience costs scholarship proposes that alliances of democracies are especially effective at deterring aggressors because elected leaders of alliance members pay high domestic audience costs if they abandon their allies (Tomz and Weeks 2021). However, our results suggest that some signals from elites can substantially reduce public support for sending troops to support an ally, thereby giving governments more leeway to abandon or at least limit their support for allies. This may have the effect of undermining the deterrence power of alliances by mitigating the domestic audience costs governments would suffer by abandoning an ally.

Treaty Compliance and Flexibility

A long-standing argument has been that states that violate their treaty obligations risk facing international audience costs once other actors come to believe that the state is less likely to abide by its treaty commitments in the future. This will make it more difficult for the state to work cooperatively with other states across a range of different issues. As a result, states will feel compelled to comply with their treaty obligations even when their short-term interests suggest otherwise.

In recent years, scholars have also argued that a government’s failure to abide by its international treaty commitments risks domestic audience costs as well. Domestic and international audience costs are linked because the public imposes political audience costs on the leader in response to the possibility of damage to the nation’s international reputation. Some have also proposed that, international audience costs aside, the public may place a normative value on abiding by commitments to other actors (Tomz and Weeks 2021). We do not challenge the assumption that the public opposes violating international treaties, though some other studies have done so (Chaudoin 2014; Morse and Pratt 2022). Instead, we challenge the assumption that the public has firm ideas about what might or might not constitute a violation of their country’s treaty obligations.

Our first proposition is that public beliefs about whether a treaty has been violated can be malleable. This proposition rests on two assumptions. First, the public is unlikely to have strong opinions on technical questions concerning the interpretation of international treaties. Scholars have built on this assumption to argue that political elites can shape public opinion through “elite cueing” (Berinsky 2009). We propose that elite cueing is more likely to succeed in shaping public beliefs regarding technical issues and/or issues that are vague, such as compliance requirements of some treaties. An example of a less vague issue is the number of friendly casualties a country has suffered, given the usually accurate fatality lists issued by elected governments during wartime (Berinsky 2009, e.g., 75).

Several studies have supported, built on, and/or challenged elite cueing theory as applied to foreign policy decisions. In the context of the 2000s Iraq War, Baum and Groeling (2010; see also Gartner and Segura 2021) found support for elite cueing theory but also showed that the ability of elite cues to shape public opinion shrank over the course of the war as the public received growing amounts of objective information. Gelpi (2010) was skeptical, finding that public opinion was guided more by objective information than by elite cues. Guisinger and Saunders (2017) demonstrated that the ability of elite cues to move public opinion varied across different issue areas (they did not examine opinion on alliances). A number of studies suggest that elite cueing guides public opinion on climate policy (for a review, see Van Boven and Sherman 2021). Bisbee and Lee (2022) showed the limited effects of elite cueing on beliefs about COVID-19.

Our second proposition is that the language on compliance requirements in treaties is sometimes vague or flexible. Treaty language is vague if it is imprecise as to exactly what the compliance requirements of the treaty are, such as banning torture without precisely defining what interrogation tactics constitute torture. Economists sometimes describe this as a problem of “incomplete contracting” (Koremenos 2013, 666). A treaty is flexible if it directly gives the signatories the ability to determine what constitutes compliant behavior (Benson 2012). State signatories sometimes deliberately include flexible language to give themselves greater leeway to argue that a wider array of actions does not violate the treaty (Downs et al. 1996;,Benson 2012; Johns 2014; Koremenos 2016; Fjelstul and Reiter 2019).1

Almost all alliance treaties include elements of flexible language. Many alliance treaties are defense pacts, requiring allies to intervene only if a signatory is attacked, yet the treaties provide no provisions for determining who attacked first. Because it is often difficult to know exactly who started a war, either due to the fog of war or because war sometimes erupts after an escalation of mutual hostility, allies have more leeway to argue that intervention is not required. In 1914, Italy refused to intervene on behalf of its German and Austria-Hungarian allies, despite their pleas to do so, because their common alliance treaty required intervention when a signatory was attacked, and Italy declared the war to be one of Austro-German aggression.2 The United States made this argument in 1965 after the Rann of Kutch War broke out between India and Pakistan, ignoring Pakistan’s demands for US intervention on the basis of the Manila Pact and the bilateral US–Pakistan defense pact, arguing that Pakistan was the aggressor and that US intervention was not required, though scholars are still divided about who the actual initiator was.3

Several alliance treaties also give allies the formal ability to decide their obligations. Some alliance treaties, such as the 1954 Manila Pact, indicate that military action is required only if the allies unanimously agree, giving each signatory the power to veto intervention without violating the treaty. American defense pacts since 1950 covering Taiwan, Japan, South Korea, the Philippines, Australia, New Zealand, and Pakistan require states to act only as long as they are consistent “with constitutional processes.”4 This means that the United States would not be in violation of the treaty if Congress elected not to formally declare war on behalf of the ally.

Other treaties directly give signatories the ability to decide what actions to take in the event of war. The 1949 North Atlantic Treaty and the 1955 Warsaw Pact both do this, with the former treaty declaring that in the event of war, each signatory will take “such action as it deems necessary.” NATO members made use of this language following the September 11, 2001, terrorism attacks—the only time Article 5 of the North Atlantic Treaty has been invoked—when signatories agreed to do so provided that they were not required to send troops to fight in Afghanistan (Buckley 2006). The 1954 US–Taiwan alliance allows that it will be decided by “mutual agreement” whether the treaty covers Taiwan-controlled islands beyond the island of Taiwan and the Pescadores, such as the islands of Quemoy and Matsu. This means the United States could elect not to send troops in the event of an attack on these other islands—an issue that was of great salience in the 1950s when China attacked these islands (Benson 2012).

Other treaties include flexible language of different kinds. In 1940, Turkey refused to enter World War II on behalf of its Anglo-French allies, declaring that their alliance treaty excused intervention if doing so risked attack by the Soviet Union, an interpretation that Britain and France vehemently disagreed with.5 Most treaties do not specify precisely what level of violence is necessary to trigger the treaty requirements, meaning that allies can argue that minor acts of aggression do not trigger the treaty. The North Atlantic Treaty is flexible in this manner, making reference only to an “armed attack.” Accordingly, NATO has declined to intervene on Turkey’s behalf following relatively minor clashes emerging from the recent civil war in Syria, including strikes on Turkish soil that killed dozens of Turkish soldiers (Sari 2020).

Public Opinion and Compliance

So far, most experimental studies that have examined public support for treaty compliance have not explored the possibility of elite rhetoric shaping public expectations of compliance requirements. Rather, the experiments are typically structured in a way that involves the survey instrument telling the respondent whether or not the treaty in question has been violated (Wallace 2013; Chaudoin 2014; Kreps and Wallace 2016;,Strezhnev, Simmons, and Kim 2020; Tomz and Weeks 2021;,Morse and Pratt 2022; Powers 2022). An exception is Greenhill and Reiter (2022), which demonstrated that American subjects were more likely to support the solitary confinement of prisoners when presented with an argument from the president claiming that this practice does not violate international law.

There is a small but growing body of survey experimental work on whether alliance ties affect the willingness of the public to support intervening on behalf of another embattled state. Scholars have proposed that democratic citizens oppose the violation of international alliance commitments for a variety of reasons that include concerns about the state’s international reputation as an ally as well as a more general desire to see their state abide by its international obligations (Tomz and Weeks 2021).

In a recent study, Tomz and Weeks (2021; see also Tomz, Weeks, and Bansak 2023; Xu, Quek, and Souv 2023) asked subjects to read a vignette describing a hypothetical situation where an unnamed state was attacked by an aggressor. They were then asked whether they would support sending troops to help defend the embattled state. The experiment manipulated contextual factors such as the regime type of the state, its geographical region, the presence of an alliance tie, the stakes involved, and the projected casualties that would result from the intervention. Tomz and Weeks found a very strong effect of alliances: Even when comparing across different combinations of conditions, the presence of an alliance tie made respondents significantly more likely to support military intervention on behalf of the embattled state.

Berejikian and Justwan (2022) found more nuanced results. One of their experiments found that merely reminding subjects that the United States and South Korea were allies did not increase the subjects’ willingness to intervene on South Korea’s behalf, but this may have been because of widespread prior knowledge of the US–South Korea alliance. However, in another experiment, a treatment stressing that the US–South Korea alliance required intervention did significantly increase subjects’ willingness to support intervention.

Other recent experimental studies support this view. Levin and Kobayashi (2022) showed the potential audience costs of an American leader exiting an international alliance. Musgrave and Ward (2023, 12, 14) found that an alliance made subjects significantly more willing to support using force, though their observed substantive effect was smaller than that observed by Tomz and Weeks (2021).

Justwan and Berejikian (2023) examined the ability of presidents to manipulate public willingness to support an ally. They considered the American public’s willingness to intervene on South Korea’s behalf in the event of a conflict between South and North Korea, where South Korea is the initiator. They found that a presidential statement reminding the subjects of the presence of the US–South Korea alliance did not have a statistically significant effect on the respondents’ willingness to provide military support, although it did make them more willing to support cruise missile strikes. However, respondents were less willing to support South Korea in a scenario where North Korea tries to deter US involvement by threatening a nuclear attack against the United States, but even in that case, they found that the decrease in willingness disappears if there is a presidential statement reminding subjects of the US alliance with South Korea.

As discussed in the previous section, in practice, leaders deliberately include flexible language in alliance treaties to provide greater latitude to argue that the treaties do not require intervention, and then sometimes use this language to justify nonintervention. There is not yet an experimental study that directly assesses whether this tactic works—in other words, whether a leader can reduce the domestic audience costs of nonintervention by attempting to shape public expectations of alliance treaty compliance requirements. Some work has, however, examined whether vague language within nonalliance threats mitigates audience costs. In a survey experiment of American adults, Smetana, Vranka, and Rosendorf (2024) demonstrated that leaders pay no domestic audience costs when failing to follow through on ambiguous nuclear threats but do pay such costs when failing to follow through on more specific threats. However, in a survey experiment of Chinese adults, Quek and Johnston (2017/2018) found that subjects were not less likely to impose costs when the government did not follow through on an ambiguous threat compared to a more specific threat. But these studies do not speak to the specific question of whether leaders can shape public expectations about compliance requirements after the commitment has been made. Our experiment begins to address this gap.

Research Design

We examine whether leaders can alleviate the audience costs of not intervening on behalf of an ally by sending signals about the compliance requirements of the alliance agreement. By “intervene,” we mean sending military forces. We focus on sending military forces because this is especially costly in the sense of casualties, military expenditures, and the risk of further escalation. Sending troops is also the most consequential action in terms of the potential deterrence and warfighting effects of alliances, as the dispatch of troops directly shifts the balance of power. For all these reasons, sending troops is the dependent variable used in a large body of theoretical and empirical work on alliance compliance.

Leaders can offer a variety of interpretations of the compliance requirements of an alliance. We focus here on a common argument that providing military aid, rather than troops, satisfies the compliance requirements of an alliance.6 Especially for a great power, the provision of military aid promises limited economic costs, no immediate troop commitments, and limited risks of escalation. Military aid can also be thought to fulfill the obligations implied in frequently occurring treaty language that calls upon states to “meet the common danger.” Indeed, the United States has often supplied military aid without troops to friendly states in order to limit economic costs and escalation risks, as it did for France during its early 1950s Indochina War, Israel in the 1973 Yom Kippur War, Britain during the 1982 Falklands War, Iraq during the Iran–Iraq War, Ukraine in its recent war with Russia, and others. After the January 1973 Paris Peace Accords and the withdrawal of American troops, President Gerald Ford argued that America’s still-active alliance commitment to South Vietnam under the Manila Pact would be satisfied with military aid rather than US combat forces.7 If anything, the legal argument for the United States to send troops to South Vietnam was even stronger in 1975 compared with 1964 (when President Johnson argued that the Manila Pact compelled US intervention), because in 1964 the threat was primarily intrastate, and in 1975 it was primarily interstate, and the Manila Pact states that in the event of internal subversion, consultation but not intervention is required (Article IV, Part 2). However, there was no call within American domestic politics in 1975 for the United States to redeploy troops to South Vietnam to fulfill its Manila Pact obligations.

Rather than trying to provide legal justifications for not sending troops, leaders might try to deflect potential audience costs by simply saying that doing so is not in the national interest. Studies generally find that alliance commitment factors do not overwhelm other factors pertaining to sending troops, and when presented with seemingly objective information on elements such as stakes and prospective costs, respondents are less willing to support sending troops (Tomz and Weeks 2021). Other studies indicate that leaders’ rhetoric about the costs and benefits can be important in shaping public opinion. Levendusky and Horowitz (2012) found that the audience costs for not implementing a threat were significantly reduced when the president provided a justification for backing down.

We also consider two other factors. One is the extent to which the governing party and the opposition are united in their interpretation of treaty obligations. Elite cueing theory proposes that leader signals are especially likely to move public opinion when the opposition signals its agreement with the leader (Trager and Vavreck 2011). Conversely, the effect of a leader’s signal can be reduced when the opposition signals its disagreement with the leader (Berinsky 2009).

The second is the party identification of the citizens. Citizens who identify with the same political party as the leader will presumably be more likely to be affected by the leader’s signal, although Levendusky and Horowitz (2012) found only very limited support for such partisanship effects. Alley (2023) examined whether partisan elite cues affected public opinion on forming and maintaining alliances. In survey experiments, he found that respondents followed elite cues from their own parties, except for specific wings within each party that maintained rigid alliance attitudes.

There is also the possibility of a backlash or backfire effect, suggesting that if the leader supports a policy, individuals from the opposite party will then be less likely to support that policy. However, recent work provides mixed results for the existence of backlash effects (Zhou 2016; Swire-Thompson, DeGutis, and Lazar 2020; Greenhill and Reiter 2022; Tingley and Tomz 2022).

Our study therefore considers a range of ways in which elite signals can undermine the public’s preference for complying with alliance obligations. In doing so, we built upon the framework developed by Tomz and Weeks (2021) and preregistered the following hypotheses:8

  • H1 (baseline alliance effect): Popular support for sending troops to defend a country will be higher when the public is told that their government has a formal alliance commitment to the country under attack.

  • H2 (presidential signal, national interest): Popular support for sending troops to defend an ally will be lower if the president signals that doing so is not in the national interest.

  • H3 (presidential signal, alternative assistance): Popular support for sending troops to defend an ally will be lower if the president recommends alternative ways of assisting the ally.

  • H4 (presidential signal, opposition statement): The effect of a presidential statement on popular support for sending troops to defend an ally will increase (or decrease) if the leader of the opposing party adopts a position that is in alignment with (or is opposed to) the position taken by the president.

  • H5 (presidential signal, respondent partisan alignment): When the respondent identifies with the same political party as the president, the effect of a presidential statement on support for sending troops will be larger in comparison with respondents who do not identify with the president’s party.

  • H6 (presidential signal, respondent backlash): When the respondent identifies with the opposite political party as the president, a presidential statement will cause support for sending troops to move in the opposite direction as compared with respondents who identify with the same party as the president.

Experiment #1

The first experiment was carried out with a sample of 2,343 US adults recruited by the survey firm Dynata (formerly Survey Sampling International). Respondents from various demographic groups were selected with a view to achieving a sample that approximates the national distribution of respondents with respect to age, gender, ethnic and racial identities, as well as geographic region.9 The sample collection took place between December 5 and December 15, 2022.

The experiment involved asking our respondents to read a vignette describing a hypothetical attack on a US ally in Asia. This was intended to closely follow the design used by Tomz and Weeks (2021).10 The following vignette served as the control group in which the United States does not have an alliance with the attacked country.11

The leader of a country in Asia wanted more power and resources, so he sent his military to attack another country in Asia and take part of that country’s territory. Here are some facts about the two countries.

The country that attackedThe country that was attacked
Type of governmentNot a democracyNot a democracy
Shared interests?Does not share many interests with the U.S.Shares many interests with the U.S.
Alliance with the U.S.?Does not have a military alliance with the U.S.Does not have a written military alliance agreement with the U.S.
The country that attackedThe country that was attacked
Type of governmentNot a democracyNot a democracy
Shared interests?Does not share many interests with the U.S.Shares many interests with the U.S.
Alliance with the U.S.?Does not have a military alliance with the U.S.Does not have a written military alliance agreement with the U.S.
The country that attackedThe country that was attacked
Type of governmentNot a democracyNot a democracy
Shared interests?Does not share many interests with the U.S.Shares many interests with the U.S.
Alliance with the U.S.?Does not have a military alliance with the U.S.Does not have a written military alliance agreement with the U.S.
The country that attackedThe country that was attacked
Type of governmentNot a democracyNot a democracy
Shared interests?Does not share many interests with the U.S.Shares many interests with the U.S.
Alliance with the U.S.?Does not have a military alliance with the U.S.Does not have a written military alliance agreement with the U.S.

In the other versions of the vignette, the respondents were told that the attacked country is indeed an ally of the United States. In these cases, the respondents were presented with more information about the alliance and, in some cases, the reactions of political leaders. Our simplest alliance scenario involved the following:12

As noted, the country that was attacked has a written military alliance with the U.S. The agreement, which was signed and ratified three years ago, says: “If one member of the alliance is attacked, the other member will act to meet the common danger in accordance with its constitutional provisions and processes.”

This language is essentially identical to the language in American alliances with Asian and Pacific countries, including Japan (“Each Party recognizes that an armed attack against either Party in the territories under the administration of Japan would be dangerous to its own peace and safety and declares that it would act to meet the common danger in accordance with its constitutional provisions and processes”), Taiwan (1954 alliance: “Each Party recognizes that an armed attack in the West Pacific Area directed against the territories of either of the Parties would be dangerous to its own peace and safety and declares that it would act to meet the common danger in accordance with its constitutional processes”), South Korea, Pakistan, the Philippines, Australia, and New Zealand, as well as in the multilateral Manila Pact.

In the first manipulation of the alliance scenario, the respondents were told that the president does not believe that the alliance requires a military response and, furthermore, does not believe that military intervention would be in the national interest.13 Specifically, the respondents were told that

The President, a Democrat, stated, “The Attorney General has advised me that the terms of the alliance do not require the U.S. to send its troops. I do not believe that, on balance, it is in our national interest to deploy troops to aid our ally.”

We chose to include the Attorney General (AG) because the president would be very likely to refer to the AG in making an argument about treaty requirements. The AG is a presidentially appointed cabinet member and is formally tasked with being the president’s legal advisor. Other US presidents have publicly used their AGs to make or support legal arguments regarding matters of foreign policy and national security, including President Ronald Reagan’s use of AG Edwin Meese regarding the Iran–Contra Affair, President George W. Bush Jr.’s use of AG John Ashcroft regarding the constitutionality of the USA Patriot Act, and President Barack Obama’s use of AG Eric Holder to argue that drone strikes did not violate international law.

In the second manipulation, we repeated this information while having the president offer to help our ally through the provision of military aid:

The President went on to say, “Instead of sending in U.S. troops, I recommend that we send weapons and other needed supplies. This will fulfill the requirements of the alliance agreement and help our ally defend itself.”

The third maintained the language from the first part of the president’s comments while adding an endorsement from a leading member of the political opposition14:

The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. Our alliance treaty does not require us to send troops.”

The fourth used the same structure but indicated that the Republican Senate leader disagreed with the president’s interpretation of the alliance obligations:

However, the leader of the opposition in the Senate, a Republican, stated, “I disagree with the President’s assessment. Our alliance treaty requires us to send troops.”

Finally, we included a control version in which the country under attack was described as not being an ally of the United States. In that case, there was no mention of the president (or anyone else) endorsing or opposing the idea of sending troops, but the other information presented in the vignette remained the same.

The various treatment groups are summarized in table 1 below.

Table 1.

Summary of the treatment arms for Experiment #1.

Treatment arm (with abbreviation)Is the invaded country an ally of the United States?What position did the president take?What position did the leader of the opposition in the Senate take?
Control (“C”)NoNoneNone
Baseline (“Ally”)YesNoneNone
President (“Pres.”)YesOpposes sending troops.None
AidYesOpposes sending troops. Offers military aid as an alternative.None
Senate supports (“S+”)YesOpposes sending troops.Opposes sending troops.
Senate opposes (“S−”)YesOpposes sending troops.Supports sending troops.
Treatment arm (with abbreviation)Is the invaded country an ally of the United States?What position did the president take?What position did the leader of the opposition in the Senate take?
Control (“C”)NoNoneNone
Baseline (“Ally”)YesNoneNone
President (“Pres.”)YesOpposes sending troops.None
AidYesOpposes sending troops. Offers military aid as an alternative.None
Senate supports (“S+”)YesOpposes sending troops.Opposes sending troops.
Senate opposes (“S−”)YesOpposes sending troops.Supports sending troops.
Table 1.

Summary of the treatment arms for Experiment #1.

Treatment arm (with abbreviation)Is the invaded country an ally of the United States?What position did the president take?What position did the leader of the opposition in the Senate take?
Control (“C”)NoNoneNone
Baseline (“Ally”)YesNoneNone
President (“Pres.”)YesOpposes sending troops.None
AidYesOpposes sending troops. Offers military aid as an alternative.None
Senate supports (“S+”)YesOpposes sending troops.Opposes sending troops.
Senate opposes (“S−”)YesOpposes sending troops.Supports sending troops.
Treatment arm (with abbreviation)Is the invaded country an ally of the United States?What position did the president take?What position did the leader of the opposition in the Senate take?
Control (“C”)NoNoneNone
Baseline (“Ally”)YesNoneNone
President (“Pres.”)YesOpposes sending troops.None
AidYesOpposes sending troops. Offers military aid as an alternative.None
Senate supports (“S+”)YesOpposes sending troops.Opposes sending troops.
Senate opposes (“S−”)YesOpposes sending troops.Supports sending troops.

In keeping with Tomz and Weeks (2021), we presented all respondents with the following information about the costs and benefits of the proposed military intervention. (We held the descriptions of the costs and benefits constant at the levels that Tomz and Weeks (2021) had found to maximize the contrast between the ally and nonally conditions.)

If the attacker succeeds in taking part of the other country, this would neither weaken U.S. military security nor hurt the U.S. economy.

The U.S. military could stop the invasion, but the military operation would be very costly to the United States.

After asking the respondents to review a summary of the situation one more time, they were asked the question, “Do you favor or oppose sending U.S. troops to stop the invasion?” They were then asked to indicate their support on a scale that ranged from 1 (“oppose strongly”) to 5 (“favor strongly”).

Results

The left-hand panel of figure 1 shows the proportion of respondents who support (either “strongly” or “somewhat”) sending troops for all six arms of the experiment.15 Moving from left to right, the first bar shows the results for the control group (“C”)—that is, the respondents who were told that the country under attack is not an ally of the United States. In that case, only 32 percent of respondents supported sending troops to defend the attacked country.

The panel on the left shows the impact of each of the manipulations on the proportion of respondents who favor sending troops to defend the embattled state. The vertical lines represent 95 percent confidence intervals. The names of the treatment groups are indicated as follows: “C”—control; “Ally”—alliance baseline; “Pres.”—president opposes sending troops; “Aid”—president offers military aid instead; “S+”—Senate agrees with the president; “S−”—Senate disagrees with the president. The panel on the right shows the differences between each of the experimental groups when compared to the “Ally” condition.
Figure 1.

The panel on the left shows the impact of each of the manipulations on the proportion of respondents who favor sending troops to defend the embattled state. The vertical lines represent 95 percent confidence intervals. The names of the treatment groups are indicated as follows: “C”—control; “Ally”—alliance baseline; “Pres.”—president opposes sending troops; “Aid”—president offers military aid instead; “S+”—Senate agrees with the president; “S−”—Senate disagrees with the president. The panel on the right shows the differences between each of the experimental groups when compared to the “Ally” condition.

The second bar from the left (“Ally”) shows the dramatic rise in support (relative to the control) that results from telling our respondents that the attacked country is an ally of the United States. In that case, support for sending in troops rises to 52 percent—a highly statistically significant increase (p < 0.001). This result thereby replicates Tomz and Weeks’ (2021) finding that the presence of a formal alliance significantly increases public support for war.

The third bar (“Pres.”) shows the impact of adding the President’s statement about how the terms of the alliance agreement necessitate sending troops. This leads to no discernible change relative to the “Ally” condition. This result provides evidence against Hypothesis 2.

The fourth bar (“Aid”) shows the effect of the president suggesting providing military aid, rather than sending troops, to the attacked ally. We find that support for the use of force decreases to 41 percent once this option is put on the table. This represents a highly statistically significant decline relative to the previous condition (“Pres.”; p = 0.01) and the baseline condition (“Ally”; p < 0.001). This result provides support for Hypothesis 3.

The fifth bar (“S+”) shows support for sending troops when the respondents are told that both the president and the leader of the opposition in the Senate agree that the alliance commitment does not require military intervention. Here we see that the signal from unified elites results in a significant decline in support for sending troops. This difference is statistically significant when compared to the presidential condition (“Pres.”; p = 0.02) and the baseline condition (“Ally”; p = 0.03). This supports Hypothesis 4.

Finally, the sixth bar (“S−”) shows the impact of conflicting messages from elected leaders. In this case, the president’s attempt to release the United States from its obligations to its ally is neutralized by the opposing message from the Senate. The average level of support for sending troops is almost identical to that of the baseline (“Ally”) condition. This supports Hypothesis 4.

To provide a clearer picture of the differences between the experimental groups, the right-hand panel of figure 1 plots the differences between the proportion of people who support sending in troops in the baseline (“Ally”) group and each of the others. Here we see that the differences are statistically significant at the 0.05 level for the “Aid” and “S+” groups, but not for “Pres.” or “S−.”

Variation by Party Identification

An interesting question to consider is whether Democrats, Republicans, and Independents respond in different ways to the signals from the president. As explained earlier, our vignettes had explicitly identified the president as a Democrat and the leader of the opposition in the Senate as a Republican. Given the timing of the survey, we would expect Democratic respondents to be more strongly influenced by the signals from the president, while Republicans would be more receptive to the messages from the leader of the opposition in the Senate. One might also expect to find effects of negative partisanship (Abramowitz and Webster 2016), whereby respondents react in the opposite direction to the recommendations made by leaders of the opposing political party.

The three panels in figure 2 show the results of the experimental manipulations among the subsets of respondents who identify as Democrats, Republicans, or Independents. Consistent with Tomz and Weeks (2021, 5), we find evidence of a strong contrast between the control (“C”) and the baseline (“Ally”) condition in all three groups. However, the presidential statements have a discernible impact on lowering support for sending troops only for the Democratic respondents, consistent with the view that presidential signals matter more for members of the same party (Hypothesis 5). It is also interesting to note that Republicans are, on average, still not moved away from their pro-intervention position when told that the leader of their party in the Senate agrees with the president’s position.

Impact of each of the manipulations on the change in the proportion of respondents supporting sending US troops relative to the “Ally” baseline condition, disaggregated by party identification of the respondents.
Figure 2.

Impact of each of the manipulations on the change in the proportion of respondents supporting sending US troops relative to the “Ally” baseline condition, disaggregated by party identification of the respondents.

Hypothesis 6 had suggested that presidential statements could result in a backfire effect for members of the opposing political party. If true, this would suggest that Republican respondents would become more, rather than less, willing to support the sending troops when the Democratic president had argued against doing so. However, we do not find evidence consistent with this hypothesis.

Attention and Manipulation Checks

The survey included a series of questions that tested the respondents’ ability to recall details from the vignette—for example, whether the attacked country was an ally of the United States, and whether the president was mentioned in the text. These were generally well answered; the median respondent correctly answered six out of the seven questions presented to them. However, to test whether our results were affected by some respondents not properly paying attention to important parts of the vignette, we reran the analysis using only the subset of respondents who had correctly answered at least six out of the seven questions. The results, shown in figure 3, confirm—and, indeed, sharpen—the findings reported based on the full sample.

Results for the subset of respondents who achieved a high score on the attention and manipulation checks. The panel on the left shows the impact of each of the manipulations on the proportion of respondents who favor sending troops to defend the embattled state. The panel on the right shows the differences between each of the experimental groups when compared to the “Ally” condition.
Figure 3.

Results for the subset of respondents who achieved a high score on the attention and manipulation checks. The panel on the left shows the impact of each of the manipulations on the proportion of respondents who favor sending troops to defend the embattled state. The panel on the right shows the differences between each of the experimental groups when compared to the “Ally” condition.

Experiment #2

The previous experiment demonstrated that unified signals from political leaders about compliance requirements could reduce public support for sending troops to support an ally. The key treatment arm in that experiment (“S+”) combined a signal about the legal obligation to intervene with a signal about why it is in the national interest to do so. This makes it difficult to discern whether one or both signals were moving public opinion. In the second experiment, we divided the compliance and national interest signals into separate treatment arms. Specifically, in one treatment, the message from the political leaders focused only on the question of whether coming to the defense of an ally is in the US national interest. In the other treatment, the message from the political leaders focused only on the question of whether the United States has a legal obligation to defend its ally.

To operationalize this, we reproduced the “Senate supports” message from Experiment #1 and introduced two subvariants—one focusing on only the national interest component of that message, and the other focusing on only the legal component. Table 2 below shows the content of these three messages.

Table 2.

Summary of treatment arms for Experiment #2.

Treatment armText
Senate Supports (same as “S+” in Experiment #1)The President, a Democrat, stated, “The Attorney General has advised me that the terms of the alliance do not require the U.S. to send its troops. I do not believe that, on balance, it is in our national interest to deploy troops to aid our ally.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. Our alliance treaty does not require us to send troops.”
Interest-only justification (“Int.”)The President, a Democrat, stated, “I do not believe that, on balance, it is in our national interest to deploy troops to aid our ally.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. It’s not in our national interest to send troops.”
Legal-only justification (“Leg.”)The President, a Democrat, stated, “The Attorney General has advised me that the terms of the alliance do not require the U.S. to send its troops.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. The terms of the alliance do not require us to send troops.”
Treatment armText
Senate Supports (same as “S+” in Experiment #1)The President, a Democrat, stated, “The Attorney General has advised me that the terms of the alliance do not require the U.S. to send its troops. I do not believe that, on balance, it is in our national interest to deploy troops to aid our ally.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. Our alliance treaty does not require us to send troops.”
Interest-only justification (“Int.”)The President, a Democrat, stated, “I do not believe that, on balance, it is in our national interest to deploy troops to aid our ally.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. It’s not in our national interest to send troops.”
Legal-only justification (“Leg.”)The President, a Democrat, stated, “The Attorney General has advised me that the terms of the alliance do not require the U.S. to send its troops.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. The terms of the alliance do not require us to send troops.”
Table 2.

Summary of treatment arms for Experiment #2.

Treatment armText
Senate Supports (same as “S+” in Experiment #1)The President, a Democrat, stated, “The Attorney General has advised me that the terms of the alliance do not require the U.S. to send its troops. I do not believe that, on balance, it is in our national interest to deploy troops to aid our ally.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. Our alliance treaty does not require us to send troops.”
Interest-only justification (“Int.”)The President, a Democrat, stated, “I do not believe that, on balance, it is in our national interest to deploy troops to aid our ally.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. It’s not in our national interest to send troops.”
Legal-only justification (“Leg.”)The President, a Democrat, stated, “The Attorney General has advised me that the terms of the alliance do not require the U.S. to send its troops.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. The terms of the alliance do not require us to send troops.”
Treatment armText
Senate Supports (same as “S+” in Experiment #1)The President, a Democrat, stated, “The Attorney General has advised me that the terms of the alliance do not require the U.S. to send its troops. I do not believe that, on balance, it is in our national interest to deploy troops to aid our ally.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. Our alliance treaty does not require us to send troops.”
Interest-only justification (“Int.”)The President, a Democrat, stated, “I do not believe that, on balance, it is in our national interest to deploy troops to aid our ally.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. It’s not in our national interest to send troops.”
Legal-only justification (“Leg.”)The President, a Democrat, stated, “The Attorney General has advised me that the terms of the alliance do not require the U.S. to send its troops.”
The leader of the opposition in the Senate, a Republican, stated, “I agree with the President’s assessment. The terms of the alliance do not require us to send troops.”

All other aspects of the original experimental design were kept the same. For ease of comparison, we also included two treatment arms from the original study to serve as benchmarks: the baseline “Ally” condition (with no signal from political leaders) and the no alliance control group (“C”). This led to the follow-up study having five treatment arms, three of which were identical to those used in the first experiment. This second experiment and its hypotheses were preregistered.16

We conducted the experiment on a nationally representative sample of US adults between April 26 and 28, 2023, using the survey firm Lucid Theorem.17 We collected a total of 2,426 usable responses. The results are shown in figure 4.

Results from the follow-up study. The panel on the left shows the impact of each of the manipulations on the proportion of respondents who favor sending troops to defend the embattled state. The vertical lines represent 95 percent confidence intervals. The names of the treatment groups are indicated as follows: “C”—control; “Ally”—alliance baseline; “S+”—Senate agrees with the president; “Int.”—interest-based justification only; “Leg.”—legal justification only. The panel on the right shows the differences between each of the experimental groups when compared to the “Ally” condition.
Figure 4.

Results from the follow-up study. The panel on the left shows the impact of each of the manipulations on the proportion of respondents who favor sending troops to defend the embattled state. The vertical lines represent 95 percent confidence intervals. The names of the treatment groups are indicated as follows: “C”—control; “Ally”—alliance baseline; “S+”—Senate agrees with the president; “Int.”—interest-based justification only; “Leg.”—legal justification only. The panel on the right shows the differences between each of the experimental groups when compared to the “Ally” condition.

These results offer support for our theory. First, they confirm the findings from the first study: The alliance effect remains strong in comparison to the control of no alliance, yet this can be reduced substantially when the respondents are provided with signals from the president and opposition suggesting that it is not necessary to send troops to defend the ally (see the bar labeled “S+” in figure 4). This leads to a reduction of 14 percentage points in the proportion of people supporting sending troops relative to the baseline (“Ally”) group.

Second, the results decompose the possible effects of the political leaders’ signal into their interest-based and compliance components. When we consider only the interest-based element of the political leaders’ message (“Int.”), we find a more modest—albeit still statistically significant—decline of 9 percentage points in support for sending troops relative to the baseline (“Ally”) condition.

However, when we focus the respondents’ attention on the compliance element of the political leaders’ message, we see a much stronger effect. The level of support for sending troops in this case drops by 16 percent relative to the baseline (“Ally”) condition. At this level, it is no longer statistically distinguishable from the no alliance control condition (“C”; p = 0.21). This result suggests the possibility that a politically savvy leader could, under the right circumstances, craft a message that allows them to virtually eliminate the audience costs associated with failing to defend an ally. However, this result needs to be viewed with caution, as in Experiment #1, a roughly comparable treatment reduced but did not eliminate the effect of an alliance in comparison to the control group.

Figure 5 shows how these results compare when we disaggregate our sample by political party identification.18 We again find that the effects of presidential signals are driven by the Democrats in the sample, although we see some evidence to suggest that the legal justifications are more persuasive than the interest-based justifications for the other groups.

Impact that each of the manipulations has in the follow-up study, disaggregated by party identification of the respondents. The y-axis represents the change in the proportion of respondents supporting sending US troops relative to the “Ally” baseline condition.
Figure 5.

Impact that each of the manipulations has in the follow-up study, disaggregated by party identification of the respondents. The y-axis represents the change in the proportion of respondents supporting sending US troops relative to the “Ally” baseline condition.

Figure 6 presents the results obtained when restricting the sample to the respondents who did especially well on the attention and manipulation check questions. Although this leads to a significant decline (53 percent) in sample size, the results are very similar to those shown for the full sample.

Results for the subset of respondents from the follow-up study who achieved a high score on the attention and manipulation checks. The panel on the left shows the impact of each of the manipulations on the proportion of respondents who favor sending troops to defend the embattled state. The panel on the right shows the differences between each of the experimental groups when compared to the “Ally” condition.
Figure 6.

Results for the subset of respondents from the follow-up study who achieved a high score on the attention and manipulation checks. The panel on the left shows the impact of each of the manipulations on the proportion of respondents who favor sending troops to defend the embattled state. The panel on the right shows the differences between each of the experimental groups when compared to the “Ally” condition.

Discussion

The results presented here have important implications for our understanding of the impact of alliance commitments on public opinion. First, they show that certain types of signals from elites can significantly reduce the audience costs that publics impose on leaders who break alliance commitments. We found evidence of this when the political opposition agreed with the government’s interpretation of the treaty, or when military aid was offered as an alternative to sending troops. This implies that presidents have some leeway when it comes to wriggling out of alliance commitments, although their success depends upon the broader political context. Importantly, the presidents who choose to do so would not necessarily need to lie or dissemble to make such an argument; as we discussed earlier, most US alliance treaties (as well as other contemporary alliances) do provide significant flexibility. As a result, the domestic political obligations implied by alliance commitments would appear to be less binding than previous work has suggested.

Two aspects of the results deserve closer discussion. The first concerns the finding that the impact of presidential signals depends on the signals sent by the political opposition. One could argue that this limits the significance of the finding because, in a real war, the opposition might disagree with a president’s unwillingness to send troops to an embattled ally. As a result, the president would, in practice, be unlikely to significantly shift public opinion in a way that justifies abandoning the ally.

However, history shows that the political opposition does not reflexively disagree with a president who chooses to abandon an ally. In 1938, all mainstream French political parties approved the abandonment of France’s Czech ally in the Munich agreement ceding the Sudetenland to Germany (Taylor 1979, 903–4). In the 1960s, the leaders of both Britain and France argued, with opposition assent, that the Manila Pact did not require their military intervention on behalf of South Vietnam (both countries were treaty signatories). Both Democrats and Republicans supported abandoning US ally Pakistan in its 1965 and 1971 wars with India. The Ford administration’s decision in 1974 and 1975 not to redeploy US troops to try to save South Vietnam, still at that point a US ally, was popular in both parties. Moreover, both Democrats and Republicans supported keeping the US out of the 1974 Cyprus War between Greece and Turkey and the 1982 Falklands War.

More recently, this pattern emerged in response to Turkey’s demands to invoke NATO’s Article 5 in the context of the Syrian Civil War, and, in the contemporary American political environment, it is not the case that one party is consistently internationalist and the other is isolationist. In short, though our results indicate that the political opposition can moot the signaling effects of the leader, there are still likely to be instances where the president can shift public opinion away from supporting an embattled ally.

A second issue is why national interest arguments were so much less effective than compliance arguments at moving public opinion. One possible explanation is that public opinion is more malleable over more specific policy questions. The public recognizes its ignorance over specific foreign policy questions, such as whether Iran is close to developing a nuclear weapon or the exact details of an alliance treaty, and is more open to listening to elite signals. But perhaps citizens have more settled opinions over more general issues, such as whether they are internationalist or isolationist, or whether they favor going to war to protect democracy abroad. The inability of elite national interest arguments to reduce popular support for intervening on behalf of an ally is consistent with the claim that populations value upholding alliance commitments. Their support for complying with alliance agreements is sufficiently strong that this belief cannot be dislodged by elite national interest arguments (Tomz and Weeks 2021).

The results might beg the question: If flexible alliances allow governments to dodge the domestic audience costs of abandoning allies, then why do states bother signing flexible alliance treaties at all? When states design alliance treaties, they are balancing the goals of deterrence and flexibility, in that the more flexible the alliance, the less it deters via tying hands dynamics. A state that values flexibility very highly over deterrence will likely not sign an alliance at all, or will sign a weaker alliance treaty, such as a consultation pact. A state signing a flexible alliance treaty with security guarantees is attempting to acquire some deterrence benefits while maintaining some flexibility, understanding that a flexible alliance treaty does not completely remove the possibility of suffering domestic audience costs from abandoning an ally. Our results support this perspective, as under the best conditions, the tying hands effects of an alliance are likely not eliminated, and disagreement from the political opposition can moot the president’s ability to eliminate audience costs. That is, flexible alliances provide a balance of some deterrence coupled with some flexibility, a combination that proved widely attractive in the post-1945 period to many states, especially the United States.

Additionally, alliances can provide deterrence benefits completely aside from tying hands effects. Some alliances, even flexible alliances such as NATO, the US–South Korea alliance, and the Warsaw Pact, are (or were) highly institutionalized and involve substantial peacetime military planning and coordination. Institutionalized alliances bolster deterrence by directly increasing the fighting power of the alliance, tying hands effects aside (Poast 2019). The opportunity to bolster fighting power through institutionalization and planning is another motivation for states to sign even flexible alliance treaties.

Conclusion

Scholarship on public support for compliance with treaties and the ability of elite signaling to shape public opinion continues to grow. This article is one of the first to bridge these two agendas, advancing a new direction. It demonstrates that some kinds of elite signals can substantially reduce public support for sending troops in support of an embattled ally. It also develops and provides empirical support for a novel argument that elites can manage audience costs by shaping public opinion about the compliance requirements of an agreement. In doing so, it suggests that compliance-based arguments are more powerful at moving public opinion than arguments based on the national interest. As predicted by elite cueing theory, opposition disagreements with the president eliminates these effects.

The findings have broader implications for international treaties, alliances, and larger debates about relationships between regime type and foreign policy behavior. They show that treaties in general and alliances in particular do not tie elected leaders’ hands as tightly as some work presumes. This gives leaders greater ability to abandon embattled allies without suffering prohibitively high costs, although their ability to do so depends upon the leaders being able to demonstrate cross-party support. The findings also provide some support for the proposition that an elected leader has at least some ability to shape public opinion. However, if elected leaders can shape public opinion, autocratic leaders have an even greater ability to do so, given the lack of political opposition and free speech in their societies (Reiter 2012).

Yet at the same time, these findings demonstrate that alliances have a remarkably resilient hold on the public imagination. The results from both experiments show that even when the president and leader of the opposition argue that the alliance treaty does not require sending troops in any particular instance, telling the public that the embattled state is an ally still leads to a significant increase in the public’s willingness to send troops relative to the nonally control condition. In the second experiment, the lesser ability of the “national interest” signal to move public opinion relative to the Ally condition is consistent with the view that the public supports upholding the nation’s alliance commitments. That is, though elites under some conditions can shape the public’s belief as to what constitutes fulfillment of an alliance commitment, they have less ability to persuade the public to accept that a violation of an alliance commitment is in the national interest. Even though elite signals can attenuate the impact of an alliance, the mere existence of an alliance has important effects on public opinion.

These findings imply several avenues for future research. First, future work can explore whether elite signals can shape public opinion in the opposite direction, toward supporting forceful intervention on behalf of an embattled ally. This was the dynamic during the 1964 Gulf of Tonkin Incident, when President Johnson argued that the Manila Pact required US intervention on South Vietnam’s behalf, despite the argument of some that the Manila Pact did not require US intervention (Brodie 1973). As noted, Berejikian and Justwan (2022; see also Justwan and Berejikian 2023) provide initial findings in this area.

Second, future studies can develop the nuances of how elite signals can shape public opinion on alliances. For example, referring to the advice of military advisers may make a leader’s argument even more effective (Kenwick and Maxey 2022). Future research could examine the conditions under which elite signals are more or less likely to work, especially in the context of shaping public beliefs about compliance requirements. Elite signals might be more effective in cases where aggression is more limited—for example, if an aggressor seizes only a small amount of an ally’s territory. Effects might also vary over the duration of a conflict, perhaps making it easier for leaders to justify abandoning an ally if the country has already made a diligent military effort, as President Ford argued regarding South Vietnam in 1975.

Third, future studies could examine the possibility that this kind of elite signaling might affect international as well as domestic audience costs. The international audience costs proposition assumes that other countries have a clear sense of whether a state has violated a treaty commitment. But it will be important to examine the extent to which leaders’ statements can shape international as well as domestic narratives around treaty compliance.

Fourth, follow-on work could explore the generalizability of our results, examining whether similar patterns are observed in other countries or across party divisions. This is an issue that is attracting increasing attention among experimentalists in IR (see Bassan-Nygate et al. 2023). While we have described a number of different non-US historical episodes in which leaders used flexible alliance language to justify nonintervention, including among some NATO members following 9/11, these efforts beg the question as to whether such tactics accomplish their intended domestic political goals. The theory here is certainly not limited to the US context, and it would be helpful to test it in other national contexts. Further, the research design here presumes a Democratic president, and though the theory forecasts that the results obtain across parties, future work could test this supposition by employing research designs that presume a Republican president.

Footnotes

1

Relatedly, signatories sometimes strategically include conditions limiting the scope of compliance requirements (Mattes 2012; Chiba, Johnson, and Leeds 2015).

2

The treaty reads: “If one, or two, of the High Contracting Parties, without direct provocation on their part, should chance to be attacked and to be engaged in a war with two or more Great Powers nonsignatory to the present Treaty, the casus foederis will arise simultaneously for all the High Contracting Parties.” https://gspi.unipr.it/sites/gspi/files/allegatiparagrafo/25-01-2016/triple_alliance.pdf. Accessed March 21, 2024.

3

The Manila Pact reads: “Each Party recognizes that aggression by means of armed attack in the treaty area against any of the Parties or against any State or territory which the Parties by unanimous agreement may hereafter designate, would endanger its own peace and safety, and agrees that it will in that event act to meet the common danger in accordance with its constitutional processes.” The 1959 US–Pakistan bilateral alliance refers to “In case of aggression against Pakistan.” Treaty texts available at https://avalon-law-yale-edu.libproxy.ucl.ac.uk/subject_menus/20th.asp (accessed March 21, 2024), and https://treaties-un-org.libproxy.ucl.ac.uk/doc/Publication/UNTS/Volume%20327/v327.pdf (accessed March 21, 2024). Regarding the United States using this defense pact loophole in 1965, see Memorandum for the Record, September 2, 1965. FRUS, 1964–1968, vol. XXV, South Asia, doc. 178. On scholarly division as to who initiated the war, see http://www.icb.umd.edu/dataviewer/?crisno=214 (accessed March 21, 2024).

4

Treaty texts available at https://avalon-law-yale-edu.libproxy.ucl.ac.uk/subject_menus/20th.asp (accessed March 21, 2024).

5

The relevant clause indicates that Turkey need not intervene for Britain and France if “having as its effect, or involving as its consequence, entry into armed conflict with the Soviet Union.” https://history.state.gov/historicaldocuments/frus1951v05/d643#:∼:text=The%20Treaty%20of%20Mutual%20Assistance,the%20latter%20by%20a%20European (accessed March 21, 2024).

6

A comparable argument is that actions short of actually executing the stated military threat can nonetheless attract support. Specifically, Lin-Greenberg (2019) found that by making threats, leaders could reduce audience costs by not taking military action but instead imposing economic sanctions (see also Quek and Johnston 2017/2018).

8

These are the preregistered hypotheses for the first experiment. See https://osf.io/qvxgj. We had slightly different, preregistered hypotheses for the second experiment (see below).

9

The demographic composition of the final sample was in many respects representative of the US adult population. The median age was 47; 50 percent of our respondents identified as female, 49 percent as male, and less than 1 percent as another gender category; 73 percent identified as White; 18 percent as Hispanic; 15 percent as Black; 5 percent as Asian; and 2.5 percent as Native American. In terms of US Census regions, our respondents were split as follows: 18 percent from the Northeast; 40 percent from the South; 21 percent from the Midwest; and 20 percent from the West. The party identities of our respondents were 34 percent Democrat; 27 percent Republican; and 25 percent Independent.

10

Brutger et al. (2023) found that replacing hypothetical countries in experimental vignettes with real ones tends to make little difference to the results.

11

Tomz and Weeks (2021) explored the impact of varying several different contextual factors in the vignette—e.g., the region in which the attack took place, whether the attacked country was a democracy, etc. For our purposes, we fixed these variables at the levels that Tomz and Weeks (2021) had found to generate the sharpest contrast between allies and nonallies.

12

Our description of the alliance text differs slightly from that of Tomz and Weeks. Their description read, “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.” A key difference is that the language in our treatment provides more leeway in each signatory electing what action to take under the rubric “act to meet the common danger,” in contrast to the less flexible language of the Tomz and Weeks treatment of “take all necessary actions.” Our language is closer to the language used in all post-1945 US alliances in terms of the flexibility of these agreements, an important consideration given that the survey is conducted on American adults. As noted above, the 1949 North Atlantic Treaty allows each member state to determine what actions are necessary; the 1947 Rio Pact provides similar flexibility. The 1951 US–Japan alliance, 1951 US–Philippines alliance, 1951 Australia–New Zealand–US (ANZUS) alliance, 1953 US–South Korea alliance, 1954 US–Taiwan alliance, and the 1954 Southeast Asian Collective Defense Treaty (Manila Pact) all use the “common danger” language. Many other alliances provide similar degrees of flexibility.

13

To keep the experiment consistent with the political conditions that prevailed at the time it was conducted, we specified that the president was a Democrat rather than a Republican.

14

To again keep the scenario consistent with the current political environment, we asked the respondents to imagine that the leader of the opposition party in the Senate was a Republican.

15

Running the analysis using the mean levels of support on the 5-point scale produces similar results.

16

This second study was preregistered at https://osf.io/t2hyu. We registered the following hypotheses: H1: Popular support for sending troops to defend an ally will be lower when leaders argue that doing so is not a legal requirement of the alliance treaty, compared with when leaders make no argument. H2: Popular support for sending troops to defend an ally will be lower when leaders argue that doing so is not in the national interest, compared with when leaders make no argument. H3: Popular support for sending troops to defend an ally will be lower in the presence of both the legal and interest-based arguments, compared with when leaders make only the legal argument. H4: Popular support for sending troops to defend an ally will be lower in the presence of both the legal and interest-based arguments, compared with when leaders make only the interest-based argument.

17

The demographic composition of the respondents to the follow-up survey was very similar to that of the original survey (shown in parentheses). The median age was 44 (47); 50 percent (50 percent) identified as female, 47 percent (49 percent) as male, and 2 percent (< 1 percent) as another gender category; 75 percent (73 percent) identified as White; 12 percent (18 percent) as Hispanic; 13 percent (15 percent) as Black; 5 percent (5 percent) as Asian; and 2 percent (2.5 percent) as Native American. The geographic regions were split as follows: 21 percent (18 percent) Northeast; 36 percent (40 percent) South; 18 percent (21 percent) Midwest; and 23 percent (20 percent) West. The party political identities of our respondents were 28 percent (34 percent) Democrat; 25 percent (27 percent) Republican; and 20 percent (25 percent) Independent.

18

Though it was not one of our preregistered hypotheses, it is possible that, partisanship aside, respondents who are more ideologically predisposed to intervene might be less affected by the treatment. Examining samples of subjects who were more inclined to use force and who were more inclined toward internationalism, we found suggestive evidence in support of this possibility (see replication files), though these sample sizes are markedly smaller.

Author Biography

Dan Reiter is Samuel Candler Dobbs Professor of Political Science at Emory University. He is the award-winning author of dozens of scholarly articles as well as several books, including How Wars End (Princeton University Press, 2009), Democracies at War (Princeton University Press, 2002, with Allan C. Stam), and Crucible of Beliefs: Learning, Alliances, and World Wars (Cornell University Press, 1996).

Brian Greenhill is an Associate Professor in the Department of Political Science at the University at Albany, SUNY. His research and teaching focus on the politics of international institutions, human rights, and climate change. He is the author of Transmitting Rights: International Organizations and the Diffusion of Human Rights Practices (Oxford University Press, 2015).

Notes

Authors’ Note: Funding for this project was provided by Emory University, the University at Albany, SUNY, and the Charles W. Koch Foundation. For helpful comments and suggestions, the authors would like to thank Natalia Bueno, Chris Clary, Matt Ingram, Hyunki Kim, Niloufer Siddiqui, and Jessica Weeks. The interpretations of the results presented here are our own. Earlier versions of this paper were presented at the Emory University International Relations Journal Club, Rockefeller College’s Experiments in Politics and Policy Group, the Military Alliances and the Future of NATO conference at Texas A&M University, and at a department seminar at the University of Georgia. Participants in these seminars provided very helpful suggestions for improvement. The data underlying this article are available on the ISQ Dataverse at https://dataverse-harvard-edu.libproxy.ucl.ac.uk/dataverse/isq.

References

Abramowitz
 
Alan I.
,
Webster
 
Stephen
.
2016
. “
The Rise of Negative Partisanship and the Nationalization of US Elections in the 21st Century
.”
Electoral Studies
.
41
:
12
22
.

Alley
 
Joshua
.
2023
. “
Elite Cues and Public Attitudes towards Military Alliances
.”
Journal of Conflict Resolution
.
67
(
7–8
):
1537
63
.

Bassan-Nygate
 
Loten
,
Renshon
 
Jonathan
,
Weeks
 
Jessica L.P.
,
Weiss
 
Chagai M.
.
2023
. “
The Generalizability of IR Experiments beyond the U.S
APSA Preprints, June 12, 2023, Version 2. doi
:.

Baum
,
Matthew A.
and
Groeling
,
Tim
,
2010
. “
Reality asserts itself: Public opinion on Iraq and the elasticity of reality
.”
International Organization
.
64 (3)
:
443
479
.

Benson
 
Brett V.
 
2012
.
Constructing International Security: Alliances, Deterrence, and Moral Hazard
.
Cambridge
:
Cambridge University Press
.

Berejikian
 
Jeffrey D.
,
Justwan
 
Florian
.
2022
. “
Defense Treaties Increase Domestic Support for Military Action and Casualty Tolerance: Evidence from Survey Experiments in the United States
.”
Contemporary Security Policy
.
43
(
2
):
308
49
.

Berinsky
 
Adam J.
 
2009
.
In Time of War: Understanding American Public Opinion from World War II to Iraq
.
Chicago
:
University of Chicago Press
.

Bisbee
 
James
,
Lee
 
Diane Da In
.
2022
. “
Objective Facts and Elite Cues: Partisan Responses to COVID-19
.”
Journal of Politics
.
84
(
3
):
1278
91
.

Brodie
 
Bernard.
 
1973
.
War and Politics
.
New York, NY
:
Macmillan
.

Brutger
 
Ryan
,
Kertzer
 
Joshua D.
,
Renshon
 
Jonathan
,
Tingley
 
Dustin
,
Weiss
 
Chagai M.
.
2023
. “
Abstraction and Detail in Experimental Design
.”
American Journal of Political Science
.
67
(
4
):
979
95
.

Buckley
 
Edgar
.
2006
. “
Invoking Article 5
.” .

Chaudoin
 
Stephen
.
2014
. “
Promises or Policies? An Experimental Analysis of International Agreements and Audience Reactions
.”
International Organization
.
68
(
1
):
235
56
.

Chiba
 
Daina
,
Johnson
 
Jesse C.
,
Leeds
 
Brett Ashley
.
2015
. “
Careful Commitments: Democratic States and Alliance Design
.”
Journal of Politics
.
77
(
4
):
968
82
.

Choi
 
Ajin
.
2012
. “
Fighting to the Finish: Democracy and Commitment in Coalition War
.”
Security Studies
.
21
(
4
):
624
53
.

Downs
 
George W.
,
Rocke
 
David M.
,
Barsoom
 
Peter N.
.
1996
. “
Is the Good News about Compliance Good News about Cooperation?
International Organization
.
50
(
3
):
379
406
.

Fjelstul
 
Joshua C. Fjelstul
,
Reiter
 
Dan
.
2019
. “
Explaining Incompleteness and Conditionality in Alliance Agreements
.”
International Interactions
.
45
(
6
):
976
1002
.

Gartner
 
Scott Sigmund
,
Segura
 
Gary
.
2021
.
Costly Calculations: A Theory of War, Casualties, and Politics
.
Cambridge
:
Cambridge University Press
.

Gelpi
 
Christopher
.
2010
. “
Performing on Cue? The Formation of Public Opinion toward War
.”
Journal of Conflict Resolution
.
54
(
1
):
88
116
.

Greenhill
 
Brian
,
Reiter
 
Dan
.
2022
. “
Naming and Shaming, Government Messaging, and Backlash Effects: Experimental Evidence from the Convention Against Torture
.”
Journal of Human Rights
.
21
(
4
):
399
418
.

Guisinger
 
Alexandra
,
Saunders
 
Elizabeth N.
.
2017
. “
Mapping the Boundaries of Elite Cues: How Elites Shape Mass Opinion across International Issues
.”
International Studies Quarterly
.
61
(
2
):
425
41
.

Ikenberry
 
G. John
.
2001
.
After Victory: Institutions, Strategic Restraint, and the Rebuilding of Order After Major Wars
.
Princeton, NJ
:
Princeton University Press
.

Johns
 
Leslie
.
2014
. “
Depth versus Rigidity in the Design of International Treaty Agreements
.”
Journal of Theoretical Politics
.
26
(
3
):
488
95
.

Justwan
 
Florian
,
Berejikian
 
Jeffrey D.
.
2023
. “
Conditional Assistance: Entrapment Concerns and Individual-Level Support for US Alliance Partners
.”
Journal of Global Security Studies
.
8
(
3
):
ogad017
.

Kenwick
 
Michael R.
,
Maxey
 
Sarah
.
2022
. “
You and Whose Army? How Civilian Leaders Leverage the Military’s Prestige to Shape Public Opinion
.”
Journal of Politics
.
84
(
4
):
1963
78
.

Koremenos
 
Barbara
.
2013
. “
The Continent of International Law
.”
Journal of Conflict Resolution
.
57
(
4
):
653
81
.

Koremenos
 
Barbara
.
2016
.
The Continent of International Law: Explaining Agreement Design
.
Cambridge, MA
:
Cambridge University Press
.

Kreps
 
Sarah
,
Wallace
 
Geoffrey P.R.
.
2016
. “
International Law, Military Effectiveness, and Public Support for Drone Strikes
.”
Journal of Peace Research
.
53
(
6
):
830
44
.

Kroenig
 
Matthew
.
2020
.
The Return of Great Power Rivalry: Democracy versus Autocracy from the Ancient World to the U.S. and China
.
Oxford
:
Oxford University Press
.

Leeds
 
Brett Ashley
,
Mattes
 
Michaela
,
Vogel
 
Jeremy S.
.
2009
. “
Interests, Institutions, and the Reliability of International Commitments
.”
American Journal of Political Science
.
53
(
2
):
461
76
.

Leeds
 
Brett Ashley
.
1999
. “
Domestic Political Institutions, Credible Commitments, and International Cooperation
.”
American Journal of Political Science
.
43
:
979
1002
.

Leeds
 
Brett Ashley
.
2003
. “
Alliance Reliability in Times of War: Explaining State Decisions to Violate Treaties
.”
International Organization
.
57
(
4
):
801
27
.

Levendusky
 
Matthew A.
,
Horowitz
 
Michael C.
.
2012
. “
When Backing Down Is the Right Decision: Partisanship, New Information, and Audience Costs
.”
Journal of Politics
.
74
(
2
):
323
38
.

Levin
 
Dov H.
,
Kobayashi
 
Tetsuro
.
2022
. “
The Art of Uncommitment: The Costs of Peacetime Withdrawals from Alliance Commitments
.”
European Journal of International Relations
.
28
(
3
):
589
615
.

Lin-Greenberg
 
Erik
.
2019
. “
Backing Up, Not Backing Down: Mitigating Audience Costs through Policy Substitution
.”
Journal of Peace Research
.
56
(
4
):
559
74
.

Mattes
 
Michaela
.
2012
. “
Reputation, Symmetry, and Alliance Design
.”
International Organization
.
66
(
4
):
679
707
.

Morse
 
Julia C.
,
Pratt
 
Tyler
.
2022
. “
Strategies of Contestation: International Law, Domestic Audiences, and Image Management
.”
Journal of Politics
.
84
(
4
):
2080
93
.

Musgrave
 
Paul
,
Ward
 
Steven
.
2023
. “
The Tripwire Effect: Experimental Evidence Regarding U.S. Public Opinion
.”
Foreign Policy Analysis
.
19
(
4
):
orad017
.

Poast
 
Paul
,
2019
.
Arguing About Alliances: The Art of Agreement in Military-Pact Negotiations
.
Ithaca, NY
:
Cornell University Press

Powers
 
Ryan
,
2022
.
Is Context Pretext? Institutionalized Commitments and the Situational Politics of Foreign Economic Policy
 
Working Paper.
 https://ryanpowers.net/files/ContextPretextPowers.pdf.

Quek
 
Kai
,
Johnston
 
Iain
.
2017/2018
. “
Can China Back Down? Crisis De-Escalation in the Shadow of Popular Opposition
.”
International Security
.
42
(
3
):
7
36
.

Reiter
 
Dan
.
2012
. “
Democracy, Deception, and Entry into War
.”
Security Studies
.
21
(
4
):
594
623
.

Sari
 
Aurel
.
2020
. “
Turkey’s Troubles in Idlib: Does Article 5 of the North Atlantic Treaty Hold the Answer?
.

Smetana
 
Michal
,
Vranka
 
Marek
,
Rosendorf
 
Ondrej
.
2024
. “
The ‘Commitment Trap’ Revisited: Experimental Evidence on Ambiguous Nuclear Threats
.”
Journal of Experimental Political Science
.
11
(
1
):
64
77
.

Strezhnev
 
Anton
,
Simmons
 
Beth A.
,
Kim
 
Matthew A.
.
2020
. “
Rulers or Rules? International Law, Elite Cues and Public Opinion
.”
European Journal of International Law
.
30
(
4
):
1281
302
.

Swire-Thompson
 
Briony
,
DeGutis
 
Joseph
,
Lazar
 
David
.
2020
. “
Searching for the Backfire Effect: Measurement and Design Considerations
.”
Journal of Applied Research in Memory and Cognition
.
9
(
3
):
286
99
.

Taylor
 
Telford
.
1979
.
Munich: The Price of Peace
.
New York, NY
:
Doubleday
.

Tingley
 
Dustin
,
Tomz
 
Michael
.
2022
. “
The Effects of Naming and Shaming on Public Support for Compliance with International Agreements: An Experimental Analysis of the Paris Agreement
.”
International Organization
.
76
(
2
):
445
68
.

Tomz
 
Michael
,
Weeks
 
Jessica L.P.
,
Bansak
 
Kirk
.
2023
. “
How NATO Membership Transforms Public Support for War
.”
PNAS Nexus
.
2
(
7
):
pgad206
.

Tomz
 
Michael
,
Weeks
 
Jessica L.P.
.
2021
. “
Military Alliances and Public Support for War
.”
International Studies Quarterly
.
65
(
3
):
811
24
.

Trager
 
Robert F.
,
Vavreck
 
Lynn
.
2011
. “
The Political Costs of Crisis Bargaining: Presidential Rhetoric and the Role of Party
.”
American Journal of Political Science
.
55
(
3
):
526
45
.

Van Boven
 
Leaf
,
Sherman
 
David K.
.
2021
. “
Elite Influence on Public Attitudes about Climate Policy
.”
Current Opinion in Behavioral Sciences
.
42
:
83
8
.

Wallace
 
Geoffrey P.R.
 
2013
. “
International Law and Public Attitudes towards Torture: An Experimental Study
.”
International Organization
.
67
(
1
):
105
40
.

Xu
 
Weifang
,
Quek
 
Kai
,
Souva
 
Mark
.
2023
. “
Reneging on Alliances: Experimental Evidence
.”
Research & Politics
.
10
(
3
).

Zhou
 
Jack
.
2016
. “
Boomerangs versus Javelins: How Polarization Constraints Communication on Climate Change
.”
Environmental Politics
.
25
(
5
):
788
811
.

This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic-oup-com.libproxy.ucl.ac.uk/journals/pages/open_access/funder_policies/chorus/standard_publication_model)