Abstract

World leaders are increasingly turning to social media to engage in crisis signaling. This raises important questions about the effects of emerging communication technologies on international politics. In particular, are threats issued via social media seen as more or less credible than those issued through traditional channels such as government press releases? Using survey experiments fielded both on unique samples of foreign policy experts in the United States, India, and Singapore and on US public samples, we find that threat medium generally generates no significant difference in perceived credibility among national security experts and members of the public. Put differently, tweeted threats are not seen as “cheaper talk” than threats issued through more traditional channels. This project extends work on crisis signaling, elite decision-making, and the domestic politics of international relations by taking into account an increasingly common technology.

Los líderes mundiales recurren, cada vez más, a las redes sociales con el fin de participar en la señalización de crisis. Esto plantea preguntas importantes sobre los efectos que tienen las tecnologías de comunicación emergentes sobre la política internacional. En concreto, ¿se consideran más o menos creíbles las amenazas emitidas a través de las redes sociales que las emitidas a través de canales tradicionales, como los comunicados de prensa por parte del Gobierno? Concluimos, mediante el uso de experimentos de encuestas realizadas tanto en muestras únicas de expertos en política exterior en los Estados Unidos, India y Singapur como en muestras públicas de los Estados Unidos, que el medio por el que se produce la amenaza no suele generar, normalmente, una diferencia significativa en la credibilidad percibida por parte de los expertos en seguridad nacional y de los miembros del público. Dicho de otra manera, las amenazas tuiteadas no se consideran «palabras más vacías» que las amenazas emitidas a través de canales más tradicionales. Este proyecto amplía el trabajo existente en materia de señalización de crisis, de toma de decisiones de las élites y de política interna en las relaciones internacionales, teniendo en cuenta una tecnología cuyo uso es cada vez más común.

Les dirigeants du monde entier se tournent de plus en plus vers les réseaux sociaux pour signaler une crise. Cette tendance soulève des questions importantes quant aux répercussions des technologies de communication émergentes sur la politique internationale. Notamment : les menaces émises par le biais des réseaux sociaux sont-elles perçues comme plus ou moins crédibles que celles formulées par le biais des canaux traditionnels, comme les communiqués de presse du gouvernement ? À l'aide d'expériences de sondage menées à la fois à partir d’échantillons de la population américaine et d'un échantillon unique d'experts politiques étrangers aux États-Unis, en Inde et à Singapour, nous observons que le moyen employé pour menacer n'a généralement pas de forte incidence sur la crédibilité perçue chez les experts de la sécurité nationale ou le grand public. Autrement dit, les tweets de menace ne sont pas pris moins au sérieux que les menaces diffusées sur des canaux plus traditionnels. Ce projet prolonge les travaux sur le signalement de crises, la prise de décisions des élites et la politique nationale des relations internationales en prenant en compte une technologie qui gagne en popularité.

Introduction

As tensions between the United States and Iran surged in January 2020, President Donald Trump took to Twitter: “Let this serve as a WARNING that if Iran strikes any Americans, or American assets, we have targeted 52 Iranian sites … and those targets, and Iran itself, WILL BE HIT VERY FAST AND VERY HARD.”1 Threats like these, delivered in real time and in full public view, have become increasingly common as policymakers around the world turn to Twitter and other social media outlets, raising important questions about the effects of social media on international politics.2 In particular, how should scholars and policymakers think about the credibility of interstate signals in an era where social media use is ubiquitous? Do national security practitioners perceive threats issued via social media as equally credible as those issued through traditional channels such as formal government press releases? Do domestic audiences in the threat-sending state view their leader’s Twitter-based threats as credible? Does the channel through which a threat is issued affect perceived credibility among the threat-receiving state’s population?

The credibility of a leader’s statements—particularly during crises—is central to the study of international relations (IR). Threats or promises that are perceived as credible can shape another state’s behavior, while those that are viewed as “cheap talk” often have little sway on a rival’s behavior, degrade the reputation of the sending actor, and generate audience costs among the sending state’s domestic population. As such, leaders must make their signals credible to multiple audiences, including members of the domestic public and foreign policy decision-makers at home and abroad. When sending these public signals, leaders and governments have historically relied on public addresses or written public statements.

The proliferation of social media platforms as a low-cost means of rapidly communicating information to multiple audiences has increasingly led leaders and governments to use these channels as diplomatic messaging tools (Seib 2012; Collins, DeWitt, and LeFebvre 2019; Ven Bruusgaard and Kerr 2020). Indeed, at least 189 world leaders maintain Twitter accounts and routinely tweet about policy matters, including foreign policy issues (Twiplomacy Study 2020). Studying whether and how elite social media use affects crisis signaling therefore has important implications for IR scholarship. Specifically, it allows us to explore how well core IR concepts related to interstate signaling hold up to changing trends in technology use.3

To study whether and how a leader’s use of social media can affect perceptions of crisis signaling, we first synthesize disparate literatures from IR, communications studies, and political psychology to lay out three potential outcomes for how a leader’s use of social media to issue threats affects perceptions of threat credibility. The domestic public in threat-issuing and threat-receiving states and national security decision-makers at home and abroad might view a tweet-issued threat as equally as, less, or more credible than a threat issued using a more traditional platform.

To empirically assess the effect of threat medium, we leverage data from a series of complementary survey experiments fielded on US public samples and samples of foreign policy practitioners and experts in the United States, India, and Singapore. The experiments vary whether the US president uses a tweet or White House press release to issue a threat during an interstate crisis. This multi-pronged research design allows us to assess whether and how a leader’s use of social media to issue threats affects perceptions of threat credibility among two core audiences: the domestic public and foreign policy practitioners.

Our experiments reveal that, in general, varying threat medium between Twitter and more traditional channels generates little difference in perceived threat credibility. Put differently, threats issued via social media are not necessarily seen as “cheaper talk” than more traditionally issued threats. At most, a president’s threat issued using Twitter is seen as very slightly less credible among the US general public than an identical threat issued via White House press releases. Among national security experts, however, tweets are viewed as being just as credible as White House press releases. In a series of follow-on experiments, we explore the effects of varying the president’s identity and tweet wording. We also field an experiment where Iran threatens the United States.

We seek to contribute to two key areas of IR scholarship. First, this project extends work on crisis signaling by taking into account new technologies that leaders use to transmit signals. Existing explanations for signal credibility often invoke a leader’s past behavior, her disposition, the degree to which she has taken steps that sink costs or tie hands, and her ability to follow through on threats or promises (Fearon 1997; Kertzer 2016; McManus 2017b; Yarhi-Milo, Kertzer, and Renshon 2018; Lupton 2020). These studies, however, typically overlook whether the channel through which a leader issues a threat affects its credibility.4 While the academic literature has not examined whether communication medium affects threat credibility, policy pieces, and journalistic accounts often assume that tweets are less credible, potentially leading to a premature consensus on an under-researched topic (Shapiro 2018; Shepp 2018).

Second, the project has implications for scholarship on the politics of emerging technologies, particularly social media. Recent work on social media highlights how platforms like Twitter add to the chorus of voices the public must sort through when developing positions on foreign policy issues (Baum and Potter 2019; Kreps 2020); explores specific cases (Duncombe 2017; Williams and Drew 2020; Freedman and Williams 2023); studies the relationship between social media use and escalation (Narang and Williams 2022); assesses domestic political implications (Mir, Mitts, and Staniland 2023); and highlights the dangers of misinformation (Chesney and Citron 2018). This paper builds upon this growing body of research by more fully exploring social media’s effects on the core IR concept of crisis signaling.

Theory: Threat Transmission Medium And Credibility

Crisis signaling is a prominent feature in IR scholarship. Whether a leader’s threats and promises are viewed as credible or as “cheap talk” has important implications for both international and domestic politics. On the international stage, signaling is critical to deterrence, compellence, and broader diplomatic efforts. A signal that other states perceive as truthful can shape a rival’s behavior without the need for further escalation. Credibility is therefore a key determinant of which signals influence an adversary’s actions and which are ignored. Drawing from Schelling’s lexicon, a credible signal that convincingly demonstrates the “power to hurt” can coerce without resorting to potentially more costly and dangerous “brute force” tactics (Schelling 1966). Domestically, leaders who make insincere threats may face political punishment from domestic audiences.5 This can diminish a leader’s ability to pursue their policy agenda (Gelpi and Grieco 2015), raise questions about their competence (Smith 1998), and generally decrease public support for the leader (Tomz 2007).

Simply put, credibility is the perceived likelihood that an actor will follow through on their threats or promises. Prominent explanations suggest that leaders can enhance signal credibility by “sinking costs” through military deployments that are ex ante costly or by “tying hands” and making pledges that are politically or reputationally costly to back down from ex post (Schelling 1966; Fearon 1997). A leader’s willingness to accept these potential costs is thought to help observers distinguish credible threats from non-credible threats.6

Scholars have long debated the underlying determinants of credibility (Jervis, Yarhi-Milo, and Casler 2021). Some work suggests signal credibility is shaped primarily by a leader’s reputation—whether they followed through on past threats and promises (Guisinger and Smith 2002; Lupton 2020). Other scholars suggest that reputation plays a less important role than an actor’s capability to follow through on a threat or promise, specifically whether they possess the military or political wherewithal to do so (Press 2007; McManus 2017b). Scholars have also explored the extent to which a leader’s individual-level dispositions affect perceptions of a rival’s threats (Yarhi-Milo, Kertzer, and Renshon 2018).

Less well studied is the extent to which the channel or medium through which a signal is transmitted affects perceptions of the signal’s credibility. Existing IR research that explores variation in signal transmission typically examines whether publicly visible signals are perceived differently than secret signals visible only to decision-makers (Kurizaki 2007; Carson and Yarhi-Milo 2017). The traditional logic suggests leaders tie their hands through publicly issued threats and promises, only making statements they intend to follow through on to avoid domestic political repercussions for backing down (Fearon 1994). More recent scholarship, however, suggests that leaders may perceive covert threats that are not publicly visible as highly credible (Carson and Yarhi-Milo 2017).

Although existing IR scholarship has studied the distinction between public and secret signals, it has largely overlooked whether variation in the public channels that leaders use to issue threats affects perceived signal credibility.7 This dearth of research is problematic given that leaders and governments now select from a growing menu of public platforms for use in crisis signaling, including a variety of social media outlets. Paying greater attention to threat medium is important as communications scholars have long suggested that information-transmission channels can shape how signals are perceived. Indeed, as Marshall McLuhan (1964) famously argued, “the medium is the message.”8 Does this adage apply to signaling in international politics?

Signaling in an Era of Social Media

New communications technologies are transforming how leaders engage in coercive diplomacy and crisis bargaining (Freedman and Williams 2023). Social media platforms, which now feature prominently in political communication at the domestic and international levels, allow users to rapidly create and disseminate information about important political events without the need to wait for gatekeepers or intermediaries—such as traditional media outlets—to broadcast reports (Zeitzoff 2017; Kreps 2020).9 Moreover, increased global use of platforms like Twitter, Facebook, and Instagram enables leaders to quickly convey information to massive international and domestic audiences with relatively little cost. Indeed, in 2021, Twitter had more than 206 million daily users, while Facebook had 2.8 billion users active on a monthly basis (Statista 2021).

World leaders have recognized social media as an inexpensive, instantaneous, and highly visible channel for communicating with constituents, rivals, and allies. Although the use of platforms like Twitter is commonly associated with Donald Trump’s contentious and saber-rattling messaging, Barberá and Zeitzoff (2018) find that more than three-quarters of world leaders are active on social media, with some using these platforms to issue threats. For instance, Austria’s then-chancellor responded to the 2020 incursion of a Turkish survey ship into Greek waters with a tweet: “If #Turkey continues to violate international law, it faces the threat of #EU sanctions.”10 Dmitry Medvedev, Deputy Chairman of Russia’s Security Council and former Russian President, also issued threats via social media. In a series of Telegram posts, Medvedev issued veiled threats of missile strikes on NATO countries and on the Hague in response to NATO’s support for Ukraine and the International Criminal Court’s issuance of an arrest warrant for Vladimir Putin (Balachuk 2023; Reuters 2023). One Telegram post explained that Russia’s hypersonic missiles “will bring to their senses anyone who will pose a direct threat to Russia and our allies (Reuters 2023).”

Beyond leaders themselves, government agencies maintain social media accounts and regularly use Twitter and other similar channels to communicate with domestic and international audiences, including during high-stakes international crises (Williams and Drew 2020). In March 2018, for instance, the Russian Embassy in London tweeted, “Any threat to take “punitive” measures against Russia will meet with a response. The British side should be aware of that.” This warning came on the heels of the British government accusing Russia of poisoning a former Russian spy residing in the United Kingdom.11

These social media-issued statements fall along a continuum of specificity. Some, like Trump’s January 2020 threat to strike Iranian targets, are explicit and include clear red lines. Others are less direct. For instance, in January 2018, Trump warned North Korean leader Kim Jong Un, tweeting, “I too have a Nuclear Button, but it is a much bigger & more powerful one than his, and my Button works!”12 These examples highlight the role social media platforms, particularly Twitter, play in contemporary crisis signaling.

To be sure, social media operates in parallel with other public and private channels for diplomatic communication. A leader might, for instance, issue a threat via Twitter and through a formal press release, while simultaneously sending private diplomatic communiques to a target state. While the interaction of signals in different mediums will affect threat credibility, exploring the effect of social media in isolation is still important. If public audiences are not privy to the private signals that leaders send to their rivals, the public may subsequently base assessments only on publicly visible signals.

Threat Credibility

How might a leader’s use of social media to issue coercive threats affect perceived signal credibility? Journalistic accounts yield mixed predictions. Some journalists and practitioners suggest that tweeted threats put a state’s international credibility at stake. These arguments, which focus largely on the Trump administration, suggest that tweeted threats are often idle. Social media saber-rattling is frequently intended to appeal to a leader’s domestic base and may not always align with the government's actual foreign policy (Erlanger 2018; Shepp 2018; Laine 2019). Other reports suggest that rivals treat tweeted threats seriously. In 2017, for example, a Kremlin spokesperson announced that “everything which is published from [President Trump's] Twitter account is perceived by Moscow as his official statement (Shelbourne 2017).”

Scholarly research—primarily from psychology and communication studies—also sheds little light on the relationship between threat communication medium and credibility in foreign policy contexts. Some research suggests that information conveyed on social media is seen as less credible than information presented on more traditional sources, with respondents perceiving social media as less accurate, less believable, and less in-depth than information from traditional media outlets (Johnson and Kaye 2014; Besalú and Pont-Sorribes 2021). Yet, leaders and governments increasingly rely on social media.

Another body of scholarship examines how characteristics of social media posts—such as formatting, originating source, or metrics such as the number of “likes”—affect perceived credibility (Morris et al. 2012; Westerman, Spence, and Van Der Heide 2012; Jahng and Littau 2016). These studies, however, are narrowly focused on what elements of a social media post bolster credibility, rather than exploring how a leader’s choice of transmission medium affects the credibility of a specific message.

Drawing from these journalistic and scholarly insights, we identify three potential outcomes for the perceived credibility of a leader’s tweeted threats relative to the credibility of the same threat issued via a more traditional government channel: The outcome in which tweets are perceived as more credible; a null effect in which a tweeted threat is seen as equally credible; and the outcome in which tweet-issued threats are perceived as less credible.

The first possibility is that a leader’s tweeted threat might be perceived as more credible than the same threat disseminated through more traditional channels. This could be the case if members of the public and foreign decision-makers believe social media posts represent the tweeting leader’s unfiltered thoughts, and critically, also think the leader is able to implement their tweeted threat. Underlying this line of reasoning is the belief that off-the-cuff comments are often highly credible (Jervis 1970). These statements are thought to be imbued with emotion and offer valuable insight into a leader’s thinking.

The second plausible outcome is that domestic and international audiences perceive threats issued via social media to be just as credible as those issued using traditional channels. Making a threat on social media is an inherently public act that potentially ties a leader’s hands in the same way as a public threat conveyed using more traditional channels. A tweet, for instance, can reach millions of Twitter users worldwide, even without being magnified by coverage in traditional media outlets. Indeed, a leader’s tweet might disseminate far faster than if the message were issued through a press release or a formal statement to traditional media. Given the highly public nature of many social media platforms, a tweet provides a written record of a leader’s threats or promises that the public can use to hold a leader accountable should she subsequently fail to follow through.

According to the “tying hands” logic of costly signaling, tweets should be perceived as credible. Since issuing an “empty tweet” is highly visible and could lead to reputational and political consequences, a leader should only tweet a threat they intend to carry out. Moreover, members of the public and international decision-makers may not differentiate between a tweet-issued threat and a public threat issued through a more traditional channel. Put differently, they might assume tweets and more traditional messages go through the same drafting and coordination process (or that staffers issue a leader’s tweets) and are therefore one and the same.

The third possibility is that a leader’s social media posts are seen as less credible than the same message conveyed using more traditional platforms. Underpinning this potential outcome is the perception that a leader can issue a tweet without the extensive interagency coordination and vetting required to craft and publish other forms of public-facing government communications.13 Journalistic and insider accounts of the Trump administration highlight this lack of synchronization and describe how the absence of interagency coordination resulted in tweeted threats that the US government could not follow through on (Shepp 2018; Esper 2022). Former US Secretary of Defense Mark Esper, for instance, explained that President Trump often developed his Twitter messages “with just one or two other people … They were never coordinated outside of the Oval Office, it seemed, let alone with the departments” (Esper 2022, 158). Trump’s threat to strike fifty-two Iranian targets was tweeted, “without any consultation with military and civilian leaders responsible for effecting such a mission” (Esper 2022, 157), and the military was unprepared to execute such an operation that, according to Secretary Esper, “clearly had no operational logic behind it” (Esper 2022, 156).

Research on coercive signaling suggests that the degree of coordination and preparation associated with issuing coercive statements can affect perceptions of threat credibility. Specifically, messages that are perceived as off-the-cuff remarks issued without fully considering potential risks and costs may be seen as less accurately conveying resolve than carefully prepared statements that are thought to involve significant planning and coordination (McManus 2017a). Moreover, coordination and planning by groups is often thought to be more rational and capable than individual efforts (Laughlin 2011; Kugler, Kausel, and Kocher 2012). As such, a message thought to be issued without coordination with a leader’s advisors should be seen as a less credible representation of the government’s position during a crisis.

Which of these three outcomes is the most plausible? We believe it is unlikely that a tweeted threat will be viewed as more credible than a threat issued via a more traditional channel. Although a leader can tie their hands by tweeting to a large audience, recent empirical work suggests more scripted and formal communications typically convey greater resolve (McManus 2017a). As a result, a threat issued via a more traditional and formal communication channel, such as a government press release, should be seen as more credible. Interagency coordination involved in crafting press releases and other traditional messages should help ensure a cohesive and credible message that signals the government’s actual intentions. This leads to our testable hypothesis:

H1: National security experts and members of the general public are likely to perceive a threat issued via social media as less credible than a threat issued via a more traditional communication channel.

Our hypothesis assumes that those assessing credibility believe that statements issued through more traditional communication channels involve greater coordination and planning. If, however, those assessing credibility believe that tweets and more traditional communications involve similar interagency coordination, threats may be viewed as equally credible. Familiarity with interagency coordination and government communications processes could therefore shape perceptions of credibility. In particular, government insiders may have a better understanding of the coordination involved with issuing foreign policy communications than members of the public. As a result, they may be more likely to treat threats equally, regardless of communication medium.

Method

To assess whether and how social media use as a signaling medium affects perceived threat credibility, we turn to original survey experiments that manipulate whether the US president issues a threat to a rival using Twitter or a White House press release. To study the effects on both domestic public and international expert audiences, we adopt a two-prong recruitment strategy that includes complementary experiments fielded on a US public sample and samples of national security practitioners and experts in the United States, India, and Singapore.14 While past studies have examined specific cases in which leaders used Twitter for crisis signaling (Duncombe 2017; Williams and Drew 2020), our experimental approach allows us to precisely identify the effects of social media-issued threats on perceived threat credibility.

Our survey instrument presents respondents with a hypothetical, but plausible, crisis involving the United States and Iran. All respondents are told:

Over the past several months, the Iranian government has provided funding, training, and weapons to militia groups that have launched several attacks on U.S. forces and partners throughout the Middle East. Earlier this week, Iranian-backed militias attacked two oil tankers in the Red Sea that were transporting fuel to the United States and fired rockets at the U.S. Embassy in Yemen. The attacks caused significant damage to the oil tankers and the embassy and killed eight people, including one American.

We then randomly assign respondents to one of two experimental conditions that manipulate whether “President Biden made his first statement about the situation by issuing the [Tweet or official White House Press Release] below.”15 We describe this as the “first statement” to suggest that the president has not yet made similar announcements on other platforms.

Unlike many survey experiments that present respondents solely with textual information, we attempt to enhance the realism of the treatments by presenting realistic facsimiles that closely mirror the format of tweets and White House press releases (figure 1a and b). While the experiment varies the medium through which President Biden issues the threat, the treatments hold threat content constant. We then ask respondents a series of questions about their perceptions of the threat’s credibility. In the follow-on experiments described below, we vary the identity of the president (i.e., Trump or Biden) and the wording of the tweet, and present a scenario in which Iranian officials issue threats against the United States.

(a) Tweet treatment (Main experiment). (b) Press release treatment (Main experiment)
Figure 1.

(a) Tweet treatment (Main experiment). (b) Press release treatment (Main experiment)

We made several key design choices when developing the main experiment. First, we explicitly name President Biden to avoid introducing ambiguity about the president’s identity. This is critical because foreign policy elites and members of the public are thought to consider the leader’s reputation when judging threat credibility (Lupton 2020). Leaving the president unnamed could lead respondents to make assumptions about the president’s identity, potentially resulting in assessments based on assumptions rather than on the treatment of interest. Moreover, recent work finds that “real and highly salient cuegivers” often generate stronger effects than hypothetical ones (Brutger et al. 2023). Second, we name Iran as the target of President Biden’s threats. As with identifying the president, naming a specific rival mitigates the risk that respondents will make assumptions about the identity or other attributes of an unnamed rival (Dafoe, Zhang, and Caughey 2018). Finally, we use Twitter as our social media treatment. To be sure, world leaders maintain a presence on other platforms such as Facebook and Instagram, but several governments have used Twitter to issue coercive threats. We compare the tweet to a written White House press release, rather than a more dissimilar medium, like a presidential press conference. In sum, naming the president and rival country and using likenesses of tweets or press releases allow respondents to engage with a realistic vignette.

Of course, these choices limit the generalizations that can be drawn from experimental findings. Critics might suggest several limitations to our research design. First, one might argue that leaders use social media to issue threats only in circumstances involving lower stakes where credibility might be less important than a high-stakes crisis. The empirical record, however, suggests that governments routinely use Twitter even during major crises involving highly contentious disputes with peer competitors. Second, critics might argue that the experimental manipulations are more salient in the experimental context than they would be in the real world. During an actual crisis, members of the public and national security experts would receive information not only via Twitter or White House press release, but from a variety of other traditional and social media sources. Moreover, a public threat might be coupled with other signals, such as the posturing of military forces and a diplomatic communique. While we concede this is true, we believe our research design maximizes internal validity while still offering externally valid insights. In an actual crisis, members of the public and national security experts may not be aware of the full range of actions a threat-issuing state has taken. As a result, they may formulate their judgments of threat credibility based solely on publicly available information—like tweets and press releases.

More broadly, scholars continue to debate whether findings from survey experiments are valid beyond the controlled confines of survey instruments. To be sure, experimental subjects face different stakes than those faced during actual crises and also receive different degrees of information than their counterparts confronting real-world crises. We believe, however, that survey experiments remain useful tools for studying expert and public preferences because experiments are grounded in the assumption that respondents rely on the same cognitive processes they would use when making judgments in the real world (Schelling 1961, 55).

We fielded the main experiment on a 977 respondent US public sample recruited using Lucid, an online sampling service, in July 2021. Lucid relies on quota sampling to recruit samples that align more closely with US Census demographics than many online convenience samples (Coppock and McClellan 2019). Lucid samples, however, are not nationally representative across all dimensions. Our sample, for instance, skews older than a national sample; underrepresents Hispanic Americans and those with household incomes greater than $100,000; and overrepresents college-educated Americans and White Americans. Still, Lucid samples are generally more representative than other online convenience samples (e.g., Amazon Mechanical Turk).

Our Lucid sample allows us to study domestic public perceptions, which a large body of research on signaling suggests are important for assessing threat credibility (Fearon 1994; Tomz 2007). The American public’s ability to levy political consequences on leaders who make threats perceived as idle or non-credible should lead the president to only issue threats that will be viewed as credible. At the same time, studying domestic and international policymaker perceptions is also important as officials in the United States, rival, and allied states make assessments about the credibility of US presidents and Washington’s foreign policy. Whether a president’s threats are seen as credible can ultimately shape other states’ policy choices.

To recruit foreign policy experts, we follow Clark (2021) and use LinkedIn messages targeted at government and think tank employees who work for institutions that make or advise on national security policy (e.g., the Ministry of Defense, the Ministry of Foreign Affairs).16 These respondents’ domain-specific expertise and experience offer unique analytic leverage for understanding the perceptions of foreign policy decision-makers (Dietrich, Hardt, and Swedlund 2021; Kertzer and Renshon 2022). We field the expert survey in the United States, India, and Singapore because these states feature professional, English-speaking, foreign policy bureaucracies and vibrant think tank communities that engage in policy debates.17 Expert participants completed the survey between August and October 2021.18 Given our online recruitment strategy, we assume both our public and expert respondents regularly use the internet.

Main Experiment Results

The main experiment suggests that signaling medium has at most, a small effect on perceived threat credibility. In a simple model with no demographic covariates, variation in communication medium has no significant effect on perceived credibility among both the public and elites.19 When standard demographic controls (e.g., gender, political ideology, race, etc.) are included, a Twitter-issued threat is viewed as slightly less credible among the public sample than a threat issued via more traditional channels (marginally significant at the p < 0.10 threshold). However, variation in communication medium continues to yield no significant results in the elite sample.

Assessing Credibility

To assess whether the channel the leader uses to issue a threat affects perceptions of credibility, we ask both the public and expert respondents, “In your opinion, how likely or unlikely is it that the president will follow through on his threat?” We use the perceived likelihood of following through as a proxy for credibility. If a threat is perceived as credible, respondents should be more likely to believe the president will carry out the threat. Respondents rate the likelihood of following through using a 5-point Likert scale ranging from “Extremely unlikely” (1) to “Extremely likely” (5).

Among respondents in the expert sample, we find that threat medium has no statistically significant effect on the perceived credibility of President Biden’s threats. Put differently, national security experts from all three countries viewed the tweet-issued threat to be just as credible as the same threat issued via a White House statement.20 On average, respondents across the full sample of experts viewed the credibility of tweeted threats to be 3.11 on the 5-point scale compared to 3.10 for threats announced through a press release (table 1). Figure 2 plots the average treatment effect of threat medium on credibility with the baseline condition as a threat issued via press release. The plot on the upper half of the figure accounts for several covariates, including respondent age, education, gender, veteran status, and past government work experience, while the plot on the lower half does not. Although we are hesitant to draw too many conclusions due to the small sample sizes, the findings reveal no significant cross-national variation in perceived credibility. Among the demographic covariates, three results were significant at the p < 0.10 threshold in the full expert sample: Veteran status and female identity were associated with decreased perceptions of credibility, while age varied directly with perceptions of credibility, regardless of threat medium.21

Average treatment effect of tweet medium on credibility, expert sample
Figure 2.

Average treatment effect of tweet medium on credibility, expert sample

Table 1.

Mean perceived credibility, US public sample

US public sample
Tweet
Mean credibility
3.22
(0.06)
n = 491
Press release
Mean credibility
3.30
(0.06)
n = 486
US public sample
Tweet
Mean credibility
3.22
(0.06)
n = 491
Press release
Mean credibility
3.30
(0.06)
n = 486

Note: Standard errors in parentheses.

Table 1.

Mean perceived credibility, US public sample

US public sample
Tweet
Mean credibility
3.22
(0.06)
n = 491
Press release
Mean credibility
3.30
(0.06)
n = 486
US public sample
Tweet
Mean credibility
3.22
(0.06)
n = 491
Press release
Mean credibility
3.30
(0.06)
n = 486

Note: Standard errors in parentheses.

We find similar results among members of the US public when not including demographic controls (figure 3). Respondents, on average, rated tweeted threats to be 3.22 on the 5-point credibility scale compared to 3.30 for threats announced through a press release (table 1). Although the results are not statistically significant, they are in the hypothesized direction with tweet threats, on average, perceived as slightly less credible than one issued via a White House press release. When demographic controls are included, the coefficient for the tweet medium remains negative and is statistically significant at the p < 0.10 threshold.

Average treatment effect of tweet medium on credibility, public sample
Figure 3.

Average treatment effect of tweet medium on credibility, public sample

Unsurprisingly, liberal respondents are, on average, likely to view President Biden’s threat as more credible than conservatives do, regardless of threat medium. Further, older Americans are, on average, slightly less likely to view the president’s threat as credible, while veterans are likely to view the threat more credibly, irrespective of whether the threat is tweeted or issued via a press release. We find no interaction effects between any of our significant demographic covariates and the Tweet medium.

To assess what underpins these perceptions, we collected qualitative data by asking respondents to “write a sentence or two explaining [their] response.” Just 2.2-percent of respondents explicitly mentioned the medium, potentially suggesting that threat medium is not particularly salient to respondents. The responses that do mention medium, however, yield interesting insights into respondents’ thinking. Some responses indicate that tweets are viewed like other forms of public statements. One respondent in the public sample notes that “President Biden didn't write that tweet, he has people that do that for him. He would only do what a tweet says if his staff told him to.” Another respondent alluded to the tying hands logic, by explaining that “if he tweeted that he is likely to follow up on it or else people would take him as a joke.” Other respondents, however, explicitly raised questions about the credibility of tweets. One respondent explained that “A threat on social media is a weak response and indicates weak follow through.” Several others suggested Twitter was not the venue for engaging in crisis signaling. “Very inappropriate if a president would do that,” noted one respondent. “A tweet is not the place! Address the Nation on prime-time saying that and it would be extremely likely [that the president would follow through, ]” said another. Although these qualitative responses demonstrate how the threat communication channel affected some respondents’ perceptions of credibility, the very limited reference to medium suggests the public pays little attention to the channel through which leaders issue foreign policy threats.

To further assess our results, we turn to data from a series of manipulation check questions built into the experiment fielded on the public sample. One question asked respondents to identify the treatment they received: Whether they read a tweet or a White House press release. Three other questions asked respondents to recall details of the scenario. Respondents who were presented a tweet were much more likely to fail the treatment manipulation check than respondents presented with a White House statement.22 In other words, respondents in the tweet treatment were more likely to incorrectly report that they had read a White House press release (51.94 percent) than respondents in the press release treatment were to incorrectly state that they had read a tweet (30.45 percent).23 Despite having a higher likelihood of inaccurately identifying the treatment they received, respondents in the tweet treatment were no more likely than respondents in the White House statement treatment to incorrectly recall specific vignette details, including the target of the attack, the country supporting the attack, or the sea in which the oil tankers were located. These findings could suggest many members of the public view tweets as a formal White House communication, which potentially explains the relatively small and statistically insignificant results.24

Given the unexpected limited findings and the possibility that respondents viewed tweets and more traditional communication mediums as one and the same, we assess whether the public believes tweeted threats involve the same degree of interagency coordination as White House press releases. In a follow-on experiment that mirrors the main experiment, we ask respondents, “Do you think the president coordinated [the Tweet or White House Press Release] with his advisors and other key members of the government?”25 We find that 73.78 percent of respondents in the White House press release treatment and 74.18 percent of respondents in the Tweet treatment believed the threat was coordinated. This difference is not statistically significant, as shown in figure 4.26 These results potentially help explain the limited effect of threat medium in the main experiment.

Average treatment effect of tweet medium on perceived coordination, public sample
Figure 4.

Average treatment effect of tweet medium on perceived coordination, public sample

In sum, our complementary public and expert experiments do not yield strong support for our hypothesis that tweeted threats should be perceived as less credible than those issued using more traditional mediums. Although social media-issued threats very slightly reduce perceived threat credibility for some members of the public, many respondents see little difference between a president’s tweets and more traditional press releases. Put differently, for many members of the public, a president’s tweets are themselves viewed as formal statements. Moreover, national security experts, on average, perceive a tweeted threat as being just as credible as one issued via a White House press release. This is perhaps because individuals with national security experience understand that a leader’s social media posts often go through a similar vetting and coordination process as statements issued using more traditional channels.

These findings offer valuable insights for the study of crisis signaling and the politics of emerging technologies. Critically, our findings suggest threats issued via social media should not be discounted as being less credible than those issued using more conventional channels. Moreover, given that perceived credibility hovers around a mean of three on a 5-point scale regardless of medium, the results suggest that the public generally views publicly issued threats as only moderately credible.

Exploring Variation in Leader and Tweet Wording

Because different leaders may use social media in different ways, we field a follow-on experiment that varies the president’s identity and the level of formality in the tweet language.27 The follow-on experiment is similar in design to the main experiment, featuring the same crisis scenario and the president threatening retaliation to future attacks either via tweet or White House press release. The follow-on, however, introduces new manipulations that allow for additional analysis. To assess whether the tweet-issuing leader’s identity has an effect on perceived credibility, we vary whether Donald Trump or Joe Biden is president. To make this plausible, we situate the scenario in September 2026, when either individual could be president. We also vary whether the text in the tweet is formal or informal to assess how the specific language a leader uses might affect outcomes. The formal language is identical to the original experiment, while the informal language is depicted in figure 5.

Informal Trump tweet
Figure 5.

Informal Trump tweet

Based on the main experiment findings, we expect variation between a formal tweet and a White House statement to yield no significant differences in credibility. Informally worded tweets, however, might be viewed as less credible than formally worded ones, given that more carefully scripted statements are thought to convey more resolve (McManus 2017a). When varying the president’s identity, we expect heterogeneous treatment effects based on respondents’ party affiliation. Conditional on threat medium, Democrats are likely to view President Biden’s threat as more credible than President Trump’s threat.

The follow-on experiment yields results that align with those from our original experiment: neither the communication medium nor text formality exerts significant influence over perceived threat credibility (figure 6). Table 2 reports the mean perceived credibility levels for each treatment group (on a 5-point scale).28

Average treatment effect of tweet medium, formality, and president on credibility
Figure 6:

Average treatment effect of tweet medium, formality, and president on credibility

Table 2.

Mean perceived credibility, presidential variation follow-on

BidenTrump
White House press release3.00
(0.09)
n = 230
3.39
(0.09)
n = 236
Formal tweet3.01
(0.09)
n = 227
3.31
(0.09)
n = 232
Informal tweet3.04
(0.09)
n = 225
3.10
(0.09)
n = 233
BidenTrump
White House press release3.00
(0.09)
n = 230
3.39
(0.09)
n = 236
Formal tweet3.01
(0.09)
n = 227
3.31
(0.09)
n = 232
Informal tweet3.04
(0.09)
n = 225
3.10
(0.09)
n = 233

Note: Standard errors in parentheses.

Table 2.

Mean perceived credibility, presidential variation follow-on

BidenTrump
White House press release3.00
(0.09)
n = 230
3.39
(0.09)
n = 236
Formal tweet3.01
(0.09)
n = 227
3.31
(0.09)
n = 232
Informal tweet3.04
(0.09)
n = 225
3.10
(0.09)
n = 233
BidenTrump
White House press release3.00
(0.09)
n = 230
3.39
(0.09)
n = 236
Formal tweet3.01
(0.09)
n = 227
3.31
(0.09)
n = 232
Informal tweet3.04
(0.09)
n = 225
3.10
(0.09)
n = 233

Note: Standard errors in parentheses.

In a simple model without demographic controls or interaction terms, the only significant result was the president’s identity: respondents rated threats issued by President Trump as more credible, regardless of threat medium.29 For our demographic covariates, Black and Asian racial identity as well as age are correlated with an increase in perceived credibility. When adding interaction terms, we find no significant effects for the interactions between the president’s identity and language formality or president’s identity and the tweet medium. Unsurprisingly, we again find that liberal respondents view a threat issued by President Biden to be more credible, while more conservative respondents find a threat issued by President Trump to be more credible. In sum, threat medium continues to have little effect on perceived threat credibility.

When examining the manipulation check questions, we once again find that respondents are more likely to incorrectly identify their treatment if they read a tweet (42.09 percent) rather than a press release (30.04 percent).30 Oddly, respondents in the tweet treatment are also slightly more likely to correctly identify the president (92.80 percent) compared to those in the press release treatment (89.06 percent).31

Exploring Variation in Threat-Issuing State

While the experiments featuring US presidents shed light on perceived credibility among the US public and experts in the United States, India, and Singapore, crisis signaling is ultimately intended to shape a targeted state’s behavior. To explore whether communication medium shapes how a rival’s threats are perceived among a target state audience, which in turn can inform government actions (Chu and Recchia 2022), we field a second follow-on experiment in which Iran’s Supreme Leader threatens the United States.32

All respondents are told:

After an Iranian-backed rocket attack on the U.S. Embassy in Yemen killed eight people, including one American, President Biden publicly warned the Iranian government that the United States would conduct military strikes on Iranian military facilities in Yemen if Iran supported further attacks on Americans. The Iranian Supreme Leader responded to President Biden's warning by issuing the [official statement or Tweet] below.

After respondents are presented with either the tweet (figure 7a) or an official statement from the Office of the Supreme Leader (figure 7b), the survey instrument asks about perceived credibility. As with the other experiments, we measure perceived credibility by asking, “In your opinion, how likely or unlikely is it that the Iranian Supreme Leader will follow through on his threat if the United States strikes Iranian facilities?” In line with the experiments described above, threat medium has no significant effect on perceived credibility of the Supreme Leader’s threat. Although a tweeted threat yields a very slightly lower mean credibility level than one issued via a press release, the difference is small and not statistically significant (table 3; figure 8). We find that female respondents are more likely to view the threat as credible, while Asian respondents and those with a high school education are less likely to view the threat as credible. Because our sample includes very few respondents without a high school diploma, we refrain from making inferences from these findings.33

(a) Tweet treatment (Iran follow-on). (b) Press release treatment (Iran follow-on)
Figure 7.

(a) Tweet treatment (Iran follow-on). (b) Press release treatment (Iran follow-on)

Average treatment effect of tweet medium on credibility, Iran follow-on
Figure 8.

Average treatment effect of tweet medium on credibility, Iran follow-on

Table 3.

Mean perceived credibility, US public sample, Iran follow-on

US public sample
Tweet
Mean credibility
3.56
(0.04)
n = 751
Press release
Mean credibility
3.58
(0.04)
n = 739
US public sample
Tweet
Mean credibility
3.56
(0.04)
n = 751
Press release
Mean credibility
3.58
(0.04)
n = 739

Note: Standard errors in parentheses.

Table 3.

Mean perceived credibility, US public sample, Iran follow-on

US public sample
Tweet
Mean credibility
3.56
(0.04)
n = 751
Press release
Mean credibility
3.58
(0.04)
n = 739
US public sample
Tweet
Mean credibility
3.56
(0.04)
n = 751
Press release
Mean credibility
3.58
(0.04)
n = 739

Note: Standard errors in parentheses.

As before, we find that respondents in the tweet treatment are more likely to miss the treatment manipulation check question (26.76 percent) than those in the official statement treatment (14.21 percent).34

Conclusion

Much existing work on crisis signaling and credibility centers on a leader’s reputation, a state’s military capabilities, or whether a leader has taken actions that tie their hands or sink costs. These studies are generally agnostic to the medium through which a public threat is issued. Yet, leaders today have an increasing menu of platforms through which to publicly threaten rivals. Our analysis explores whether variation in the platform used to communicate coercive signals affects perceived threat credibility among both domestic and international audiences. Drawing data from a series of survey experiments fielded on samples of foreign policy experts and on members of the US public, we find that variation in threat channels has at most a limited effect on perceptions of threat credibility. The public sometimes views tweet-issued threats as slightly less credible than those issued using press releases, but in general, tweeted threats are seen as no different than those transmitted using more traditional channels.

Our analysis moves beyond existing scholarship on interstate signaling by taking into account world leaders’ increasing use of social media as a medium of coercive diplomacy. In doing so, we contribute to the study of signaling and perceptions, the domestic politics of foreign affairs, and the implications of emerging technologies on international security.

Our findings point to several additional avenues for future research. Since signaling does not occur in a vacuum, scholars might study whether variation in the framing of tweeted threats affects perceived credibility. For instance, future work might explore what happens when a senior official’s tweets diverge from more traditional government statements. Or, additional experiments might explore whether the amplification of tweets through retweeting or rebroadcasting in traditional outlets shapes how people perceive the threats.

Future studies might also study other communication mediums. For instance, the effect of communication medium might have been more pronounced had the traditional channel in the experiment been a more significant event—like a West Wing address—rather than a written press release. Moreover, some studies suggest the use of different social media platforms can yield divergent effects during international crises (Narang and Williams 2022). Additional experiments might explore whether threat issuance using different social media platforms affects perceived threat credibility among both decision-makers and the public.

Finally, scholars might explore the generalizability of our findings. Our research design asks respondents in the United States, India, and Singapore to assess the credibility of a threat. Future work might assess whether our results hold among respondents in other states or if different states are named as the issuing and target actors. Pursuing these questions will further enrich our understanding of the interactions between emerging technology and international politics.

Author Biography

Benjamin Norwood Harris is a PhD Candidate in the Department of Political Science at the Massachusetts Institute of Technology.

Erik Lin-Greenberg is the Leo Marx Career Development Assistant Professor in the Department of Political Science at the Massachusetts Institute of Technology.

Notes

Author’s note: We thank Kjell Engelbrekt, Mauro Gilli, Jonas Heering, Rich Nielsen, and workshop participants at Georgetown University, the University of Oslo, the University of Pennsylvania, the Weatherhead Center for International Affairs, and ISA 2022 for helpful feedback on earlier drafts. Ella von Baeyer provided research assistance. Any remaining errors are ours alone. Author names in alphabetical order. This project was preregistered with OSF: https://osf.io/rcb5a. The data underlying this article are available on the ISQ Dataverse at https://dataverse-harvard-edu.libproxy.ucl.ac.uk/dataverse/isq.

Footnotes

1

The tweet was widely covered in media. See, for example, Diamond, Kelly, and Clary (2020).

2

Twitter was rebranded as X in 2023. We use “Twitter” throughout the article as many scholars and policymakers continue to refer to the platform by its original name. Additionally, most of our experiments were fielded prior to the platform’s rebranding.

3

On the importance of studying “digital diplomacy,” see Hedling and Bremberg (2021).

4

Past work has compared private signaling to public signaling (for instance, Carson and Yarhi-Milo 2017). We focus here on assessing whether variation in public platforms (i.e., social media versus press releases) affects perceived credibility.

5

The concept of audience costs is the subject of an ongoing scholarly debate. For instance, see Snyder and Borghard (2011) and Trachtenberg (2012).

6

To be clear, this logic is up for debate. For instance, Fuhrmann and Sechser (2014) find no evidence of the sinking costs logic.

7

One important exception is McManus (2017a), which explores whether the context in which presidents make public statements affects perceived resolve and dispute outcomes. McManus finds that highly publicized national speeches and scripted remarks have the most significant effects on crisis outcomes.

8

While debates over the reputational, dispositional, and capabilities-based determinants of credibility remain unresolved, we bracket these factors and focus on assessing whether variation in public signaling channel affects perceived credibility.

9

To be sure, traditional media outlets often rebroadcast social media posts. Rebroadcasts, however, still typically identify the original source.

11

For a detailed summary of the tweets, see Williams and Drew (2020).

13

Of course, tweets can also go through a similar coordination and vetting process, but the ability of a leader to issue a tweet on his or her own means they can circumvent this process.

14

For more on the utility of “paired experiments” and “complementary designs” that feature experiments fielded on both public and elite samples, see Kertzer and Renshon (2022).

15

Because describing the press releases as “official” may imply that other communication forms (e.g., social media) are unofficial and inherently less credible, we field a follow-on experiment that does not include the term “official.” The analysis yields similar effects as the main experiment. See Online Appendix F.

16

See Online Appendix A for additional information on the recruitment strategy. The Online Appendix includes the recruitment text and the organizations from which expert respondents were recruited.

17

In the ideal case, we would have also fielded an expert survey in the target country of Iran, but this was not logistically feasible.

18

The United States withdrawal from Afghanistan and the Taliban’s subsequent takeover occurred shortly after we began fielding our expert experiment in the United States (the India and Singapore surveys were fielded entirely after Kabul fell). These events triggered significant policy debates about President Biden’s reputation and credibility. Because these real-world events might have influenced expert respondent perceptions of threat credibility, we include analysis in Online Appendix C that assesses whether respondents who completed the experiment before Kabul fell hold different perceptions than those who completed the survey after Kabul’s fall. We find no difference in perceptions.

19

For the public sample, the coefficient of the tweet medium on credibility is negative. In the expert sample, it is positive, but neither meets standard thresholds of statistical significance.

20

The experiments are fielded on three distinct samples (the United States, Singapore, and India). Because of the relatively small sample size, we present pooled results in addition to country-specific findings.

21

Online Appendix C includes regression models featuring the full set of covariates and a table displaying the mean and standard errors of the credibility estimate for all treatment groups.

22

This finding is significant at the p < 0.01 threshold.

23

See Online Appendix B for analysis of manipulation checks. As an additional check of our findings, we rerun our analysis of the effect of the tweet medium on perceived credibility using a subsample of only respondents who passed the manipulation check. The analysis yields similar null results to the full sample. We repeat this sensitivity check for all follow-on experiments and add an additional check that tests whether respondents who passed the attention check questions answer differently. We find null results in all cases. See Online Appendices B, D, E, and F.

24

Alternately, it is possible that a White House statement was the most logical guess for respondents who did not recall the treatment. This would also suggest that communication medium has relatively little effect on respondent perceptions.

25

The follow-on was fielded on a public sample of 1,099 adults in the United States recruited using Lucid in August 2023.

27

We field the experiment in February 2022 on a 1,383-respondent Lucid sample.

28

See Online Appendix D for Ordinary Least Squares (OLS) analysis.

29

It is possible that respondents, on average, thought President Trump would be more likely than President Biden to use force against Iran. During the Trump administration, Iran and the United States faced heightened tensions, whereas tensions subsided slightly over the course of the Biden administration. Future studies could explore this interesting finding.

30

This result is significant at the p < 0.01 threshold. See Online Appendix D.

31

This result is significant at the p < 0.05 threshold. See Online Appendix D.

32

The follow-on was fielded on a public sample of 1,490 adults in the United States recruited using Lucid in January 2023.

33

See Online Appendix E for the full OLS results.

34

This result is significant at the p < 0.01 threshold. See Online Appendix E.

References

Balachuk
 
Iryna
.
2023
. “
Medvedev Threatens to Strike Court in The Hague with Oniks Hypersonic Missile
.”
Ukrainska Pravda
.
March 20, 2023. Accessed August 10, 2023. https://www.pravda.com.ua/eng/news/2023/03/20/7394171/
.

Barberá
 
Pablo
,
Zeitzoff
 
Thomas
.
2018
. “
The New Public Address System: Why Do World Leaders Adopt Social Media?
International Studies Quarterly
.
62
(
1
):
121
30
.

Baum
 
Matthew A.
,
Potter
 
Philip B.K.
.
2019
. “
Media, Public Opinion, and Foreign Policy in the Age of Social Media
.”
The Journal of Politics
.
81
(
2
):
747
56
.

Besalú
 
Reinald
,
Pont-Sorribes
 
Carles
.
2021
. “
Credibility of Digital Political News in Spain: Comparison between Traditional Media and Social Media
.”
Social Sciences
.
10
(
5
):
170
.

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

Carson
 
Austin
,
Yarhi-Milo
 
Keren
.
2017
. “
Covert Communication: The Intelligibility and Credibility of Signaling in Secret
.”
Security Studies
.
26
(
1
):
124
56
.

Chesney
 
Robert
,
Citron
 
Danielle Keats
.
2018
. “
Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security
.”
SSRN Scholarly Paper ID 3213954
.
Rochester, NY
:
Social Science Research Network
.

Chu
 
Jonathan
,
Recchia
 
Stefano
.
2022
. “
Does Public Opinion Affect the Preferences of Foreign Policy Leaders? Experimental Evidence from the UK Parliament
.”
The Journal of Politics
.
84
(
3
):
1874
7
.

Clark
 
Richard
.
2021
. “
Pool or Duel? Cooperation and Competition Among International Organizations
.”
International Organization
.
75
(
4
):
1133
53
.

Collins
 
Stephen D.
,
DeWitt
 
Jeff R.
,
LeFebvre
 
Rebecca K.
.
2019
. “
Hashtag Diplomacy: Twitter as a Tool for Engaging in Public Diplomacy and Promoting US Foreign Policy
.”
Place Branding and Public Diplomacy
.
15
(
2
):
78
96
.

Coppock
 
Alexander
,
McClellan
 
Oliver A.
.
2019
. “
Validating the Demographic, Political, Psychological, and Experimental Results Obtained from a New Source of Online Survey Respondents
.”
Research & Politics
.
6
(
1
). https://journals-sagepub-com.libproxy.ucl.ac.uk/doi/full/10.1177/2053168018822174.

Dafoe
 
Allan
,
Zhang
 
Baobao
,
Caughey
 
Devin
.
2018
. “
Information Equivalence in Survey Experiments
.”
Political Analysis
.
26
(
4
):
399
416
.

Diamond
 
Jeremy
,
Kelly
 
Caroline
,
Clary
 
Greg
.
2020
. “
Donald Trump Warns Iran That ‘We Have Targeted 52 Iranian Sites’ If Any Americans Are Attacked
.”
CNN
. .

Dietrich
 
Simone
,
Hardt
 
Heidi
,
Swedlund
 
Haley J.
.
2021
. “
How to Make Elite Experiments Work in International Relations
.”
European Journal of International Relations
.
27
(
2
):
596
621
.

Duncombe
 
Constance
.
2017
. “
Twitter and Transformative Diplomacy: Social Media and Iran–US Relations
.”
International Affairs
.
93
(
3
):
545
62
.

Erlanger
 
Steven
.
2018
. “
Trump’s Twitter Threats Put American Credibility on the Line
.”
The New York Times
. .

Esper
 
Mark T.
 
2022
.
A Sacred Oath: Memoirs of a Secretary of Defense During Extraordinary Times
.
New York, NY
:
William Morrow
.

Fearon
 
James
.
1994
. “
Domestic Political Audiences and the Escalation of International Disputes
.”
American Political Science Review
.
88
(
3
):
577
92
.

Fearon
 
James
.
1997
. “
Signaling Foreign Policy Interests: Tying Hands versus Sinking Costs
.”
Journal of Conflict Resolution
.
41
(
1
):
68
90
.

Freedman
 
Lawrence
,
Williams
 
Heather
.
2023
.
Changing the Narrative: Information Campaigns, Strategy and Crisis Escalation in the Digital Age
.
London
:
International Institute for Strategic Studies
.

Fuhrmann
 
Matthew
,
Sechser
 
Todd S.
.
2014
. “
Signaling Alliance Commitments: Hand-Tying and Sunk Costs in Extended Nuclear Deterrence
.”
American Journal of Political Science
.
58
(
4
):
919
35
.

Gelpi
 
Christopher
,
Grieco
 
Joseph
.
2015
. “
Competency Costs in Foreign Affairs: Presidential Performance in International Conflicts and Domestic Legislative Success, 1953–2001
.”
American Journal of Political Science
.
59
(
2
):
440
56
.

Guisinger
 
Alexandra
,
Smith
 
Alastair
.
2002
. “
Honest Threats: The Interaction of Reputation and Political Institutions in International Crises
.”
Journal of Conflict Resolution
.
46
(
2
):
175
200
.

Hedling
 
Elsa
,
Bremberg
 
Niklas
.
2021
. “
Practice Approaches to the Digital Transformations of Diplomacy: Toward a New Research Agenda
.”
International Studies Review
.
23
(
4
):
1595
618
.

Jahng
 
Mi Rosie
,
Littau
 
Jeremy
.
2016
. “
Interacting is Believing: Interactivity, Social Cue, and Perceptions of Journalistic Credibility on Twitter
.”
Journalism & Mass Communication Quarterly
.
93
(
1
):
38
58
.

Jervis
 
Robert
.
1970
.
The Logic of Images in International Relations
.
Princeton, NJ
:
Princeton University Press
.

Jervis
 
Robert
,
Yarhi-Milo
 
Keren
,
Casler
 
Don
.
2021
. “
Redefining the Debate Over Reputation and Credibility in International Security: Promises and Limits of New Scholarship
.”
World Politics
.
73
(
1
):
167
203
.

Johnson
 
Thomas J.
,
Kaye
 
Barbara K.
.
2014
. “
Credibility of Social Network Sites for Political Information Among Politically Interested Internet Users
.”
Journal of Computer-Mediated Communication
.
19
(
4
):
957
74
.

Kertzer
 
Joshua D.
 
2016
.
Resolve in International Politics
.
Princeton, NJ
:
Princeton University Press
.

Kertzer
 
Joshua D.
,
Renshon
 
Jonathan
.
2022
. “
Experiments and Surveys on Political Elites
.”
Annual Review of Political Science
.
25
:
529
50
.

Kreps
 
Sarah
.
2020
.
Social Media and International Relations
.
Cambridge
:
Cambridge University Press
.

Kugler
 
Tamar
,
Kausel
 
Edgar E.
,
Kocher
 
Martin G.
.
2012
. “
Are Groups More Rational than Individuals? A Review of Interactive Decision Making in Groups
.”
WIREs Cognitive Science
.
3
(
4
):
471
82
.

Kurizaki
 
Shuhei
.
2007
. “
Efficient Secrecy: Public versus Private Threats in Crisis Diplomacy
.”
American Political Science Review
.
101
(
3
):
543
58
.

Laine
 
Tamara
.
2019
. “
Thumb-Boat Diplomacy Could Undo U.S. Foreign Policy
.”
Foreign Policy (Blog)
. .

Laughlin
 
Patrick R.
 
2011
.
Group Problem Solving
.
Princeton, NJ
:
Princeton University Press
.

Lupton
 
Danielle L.
 
2020
.
Reputation for Resolve: How Leaders Signal Determination in International Politics
.
Ithaca, NY
:
Cornell University Press
.

McLuhan
 
Marshall
.
1964
.
Understanding Media: The Extensions of Man
.
New York
:
Signet Books
.

McManus
 
Roseanne W.
 
2017a
. “
The Impact of Context on the Ability of Leaders to Signal Resolve
.”
International Interactions
.
43
(
3
):
453
79
.

McManus
 
Roseanne W.
.
2017b
.
Statements of Resolve: Achieving Coercive Credibility in International Conflict
.
New York
:
Cambridge University Press
.

Mir
 
Asfandyar
,
Mitts
 
Tamar
,
Staniland
 
Paul
.
2023
. “
Political Coalitions and Social Media: Evidence from Pakistan
.”
Perspectives on Politics
.
21
(
4
):
1337
56
.

Morris
 
Meredith Ringel
,
Counts
 
Scott
,
Roseway
 
Asta
,
Hoff
 
Aaron
,
Schwarz
 
Julia
.
2012
. “
Tweeting is Believing? Understanding Microblog Credibility Perceptions
.” In
CSCW ’12: Proceedings of the ACM 2012 Conference on Computer Supported Cooperative Work
,
441
50
..
New York, NY
:
Association for Computing Machinery
.

Narang
 
Vipin
,
Williams
 
Heather
.
2022
. “
Thermonuclear Twitter
.” In
The Fragile Balance of Terror: Deterrence in the New Nuclear Age
,
edited by
 
Narang
 
Vipin
,
Sagan
 
Scott D.
,
63
89
..
Ithaca, NY
:
Cornell University Press
.

Press
 
Daryl G.
 
2007
.
Calculating Credibility: How Leaders Assess Military Threats. Cornell Studies in Security Affairs
.
Ithaca, NY
:
Cornell University Press
.

Reuters
.
2023
. “
Russia’s Medvedev Snaps Back after U.S. Appeal over Ukraine War
.”
Reuters
. .

Schelling
 
Thomas
.
1961
. “
Experimental Games and Bargaining Theory
.”
World Politics
.
14
(
1
):
47
68
.

Schelling
 
Thomas
.
1966
.
Arms and Influence
.
New Haven, CT
:
Yale University Press
.

Seib
 
Philip M.
 
2012
.
Real-Time Diplomacy: Politics and Power in the Social Media Era
. 1st ed.  
New York
:
Palgrave Macmillan
.

Shapiro
 
Ari
,
dir
.
2018
. “
Foreign Policy Expert Considers Repercussions Of Trump’s Twitter Diplomacy
.”
All Things Considered
.
National Public Radio
.

Shelbourne
 
Mallory
.
2017
. “
Kremlin: Every Trump Tweet Perceived as ‘Official Statement’
.”
Text
.
The Hill (Blog)
. .

Shepp
 
Jonah
.
2018
. “
How U.S. Foreign Policy Is Being Shaped by Trump’s Tweets
.”
Intelligencer
. .

Smith
 
Alastair
.
1998
. “
International Crises and Domestic Politics
.”
American Political Science Review
.
92
(
3
):
623
38
.

Snyder
 
Jack
,
Borghard
 
Erica
.
2011
. “
The Cost of Empty Threats: A Penny, Not a Pound
.”
American Political Science Review
.
105
(
3
):
437
56
.

Statista
.
2021
. “
Twitter Global mDAU 2021
.”
Statista
. .

Tomz
 
Michael
.
2007
. “
Domestic Audience Costs in International Relations: An Experimental Approach
.”
International Organization
.
61
(
4
):
821
40
.

Trachtenberg
 
Marc
.
2012
. “
Audience Costs: An Historical Analysis
.”
Security Studies
.
21
(
1
):
3
42
.

Twiplomacy Study 2020
.
2020
.
Burson Cohn & Wolfe
.

Ven Bruusgaard
 
Kristin
,
Kerr
 
Jaclyn A.
.
2020
. “
Crisis Stability and the Impact of the Information Ecosystem
.” In
Three Tweets to Midnight
,
edited by
 
Trinkunas
 
Harold
,
Lin
 
Herbert
,
Loehrke
 
Benjamin
,
137
58
..
Stanford, CA
:
Hoover Institution Press
.

Westerman
 
David
,
Spence
 
Patric R.
,
Van Der Heide
 
Brandon
.
2012
. “
A Social Network as Information: The Effect of System Generated Reports of Connectedness on Credibility on Twitter
.”
Computers in Human Behavior
.
28
(
1
):
199
206
.

Williams
 
Heather
,
Drew
 
Alexi
.
2020
.
Escalation by Tweet: Managing the New Nuclear Diplomacy
.
London
:
Kings College London
.

Yarhi-Milo
 
Keren
,
Kertzer
 
Joshua D.
,
Renshon
 
Jonathan
.
2018
. “
Tying Hands, Sinking Costs, and Leader Attributes
.”
Journal of Conflict Resolution
.
62
(
10
):
2150
79
.

Zeitzoff
 
Thomas
.
2017
. “
How Social Media is Changing Conflict
.”
Journal of Conflict Resolution
.
61
(
9
):
1970
91
..

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)

Supplementary data