A Pact with the Devil: The Madman Theory and Nonaggression Pacts

Abstract

This study explores the onset of nonaggression pacts at the individual level, focusing on how leaders’ perceived madness influences security agreements. While the Madman Theory suggests that perceived madness could be an asset in crisis bargaining, it also signals uncertainty and hostility to other states. Consequently, states seek security assurances through nonaggression pacts to manage the risks posed by such leaders. On the other hand, leaders adopting the Madman strategy may strategically pursue these pacts as signaling devices to alleviate diplomatic isolation and mitigate their overall reputation for madness for strategic purposes. By utilizing McManus’s perception-based measures, my findings indicate that leaders with a strong reputation for madness significantly influence the formation of nonaggression pacts, whereas moderate levels of madness do not have the same effect. The impact of a strong reputation for madness diminishes when states are distant from the mad leader or are protected by more military alliances. This study sheds light on the role of leaders’ reputations in security agreements and extends the scope of the Madman Theory beyond crisis bargaining to broader contexts of security cooperation.

Cette étude s'intéresse au déclenchement des pactes de non-agression au niveau individuel, en se concentrant sur l'influence de la folie perçue des gouvernants sur les accords de sécurité. Bien que la théorie du fou suggère que la folie perçue puisse constituer un atout dans la négociation de crises, elle indique également une certaine incertitude et de l'hostilité aux autres États. Par conséquent, les États souhaitent obtenir des garanties de sécurité par le biais de pactes de non-agression afin de gérer les risques posés par lesdits gouvernants. D'autre part, la signature de tels pactes peut faire partie intégrante de la stratégie du fou des gouvernants : ce sont des outils qui permettent de réduire l'isolation diplomatique et d'atténuer leur réputation globale de folie à des fins stratégiques. En m'appuyant sur les mesures fondées sur la perception de RW McManus, mes conclusions indiquent que les gouvernants fortement réputés pour leur folie ont une incidence importante sur la formation de pactes de non-agression, tandis que des niveaux modérés de folie n'ont pas le même effet. L'effet d'une forte réputation de folie perd de l'ampleur quand les États sont éloignés du gouvernant fou ou protégés par davantage d'alliances militaires. Cette étude met en lumière le rôle de la réputation des dirigeants dans les accords de sécurité et élargit la portée de la théorie du fou au-delà de la négociation de crise, à des contextes plus larges de coopération sécuritaire.

Este estudio analiza el inicio de los pactos de no agresión a nivel individual, centrándose en cómo el hecho se perciba a un líder como loco influye sobre los acuerdos de seguridad. Si bien la teoría del loco sugiere que la locura percibida en un líder podría representar una ventaja en la negociación de crisis, también manifiesta incertidumbre y hostilidad hacia otros Estados. En consecuencia, los Estados buscan garantías de seguridad a través de pactos de no agresión con el fin de gestionar los riesgos que plantean dichos líderes. Por otro lado, los líderes que adoptan la estrategia del loco pueden perseguir estos pactos de manera estratégica con el fin de utilizarlos como dispositivos de señalización para aliviar el aislamiento diplomático y mitigar su reputación general de locura con fines estratégicos. Utilizamos las medidas basadas en la percepción de McManus, lo que nos permite concluir que la existencia de líderes con una fuerte reputación de locura influye de forma significativa sobre la formación de pactos de no agresión, mientras que los niveles moderados de locura no tienen el mismo efecto. El impacto que tiene una fuerte reputación de locura disminuye cuando los Estados se encuentran distantes del líder loco o están protegidos por más alianzas militares. Este estudio arroja luz sobre el papel que ejerce la reputación de los líderes en los acuerdos de seguridad y amplía el alcance de la teoría del loco más allá de la negociación de crisis, hacia contextos más amplios a nivel de cooperación en materia de seguridad.

Introduction

Adolf Hitler was known for a reputation for madness based on his erratic behaviors and aggressive coercion. When Hitler came to power, Chamberlain soon labeled Hitler as a “mad dictator” (Feiling 1946, 279). In the lead-up to World War II, Hitler took risky gambles by remilitarizing the Rhineland and annexing Austria without facing opposition from Britain and France. Soon after, he asserted his demand for control over the Sudetenland, a region within Czechoslovakia predominantly inhabited by ethnic Germans. Britain and France eventually acquiesced to Hitler’s demand in 1938 due to the belief that the Sudetenland was Hitler’s last territorial ambition (Goddard 2015; Churchill 2023). The Sudetenland crisis illustrated that Hitler demonstrated the advantage of perceived madness through the perception of unpredictability and the impression of having situational extreme preferences and limited ambitions (Ellsberg 1959; McManus 2019).

Naturally, Hitler’s reputation for madness generated fear and uncertainty, motivating other states to seek security solutions. For instance, in an effort to counter Hitler’s threat, the USSR, Britain, and France engaged in negotiations for a triple alliance in 1939. However, the alliance negotiation was not productive due to mutual distrust (Roberts 1992) and conflicting perspectives on war planning and coordination regarding Eastern Europe (Cienciala 2003; Poast 2019). While having doubts about a favorable deal with the British and the French, Moscow consistently kept the door open to a peace agreement with Berlin at the same time since Stalin’s primary goal in 1939 was to avoid war with Hitler in Eastern Europe (Roberts 1992; Cienciala 2003). Meanwhile, Nazi Germany actively pursued numerous attempts to negotiate with the USSR in an effort to prevent the formation of a coalition with Britain and France (Roberts 1992). As a result, in August 1939, the infamous Molotov–Ribbentrop Pact was signed in Moscow, ensuring peace between the parties along with other secret protocols.

Not only did the USSR attempt to negotiate with Nazi Germany before World War II, but other countries, including Poland, France, Denmark, Estonia, and Latvia, also tried to reach a deal and signed nonaggression pacts with Hitler between 1934 and 1939. The motivation was that signing nonaggression pacts could normalize relations with Nazi Germany and pacify Hitler, possibly redirecting his attention elsewhere, because they saw pursuing nonaggression pacts with Hitler was seen as a form of peaceful measure to ensure security (Cienciala 2003). Seeking peace treaties with others was also a component of Hitler’s pre-war strategy, as signaling peaceful intentions was meant to break diplomatic isolation, lower the guard of the Allies, and prevent coalitions against Nazi Germany.

In addition, Saddam Hussein, often labeled as “the madman of the Middle East,” was notorious for his unpredictable behavior and expansionist regional ambitions due to initiating costly wars, pursuing weapons of mass destruction, and using chemical weapons against civilians (Post 1991, 283; Karsh and Rautsi 2002). He was often perceived as having dispositional extreme preferences and a willingness to take high-risk actions, and this perception conveyed important information and influenced diplomatic and strategic calculations by other states (Post 1991, 279; McManus 2019). Saudi Arabia and Iraq, both important players in the Middle East, have a complex history of rivalry. During the Iran-Iraq War, Saudi Arabia adopted a pro-Iraq neutrality position, providing financial support to Iraq due to the shared threat of Iran and concerns about revolutionary influence in the region (Herrmann 1991; Gause III 2001, 2002). Their relationship was relatively manageable during the Iran-Iraq War, but tensions between Saudi Arabia and Iraq resurfaced after the war due to the diminished Iranian threat. This shift was further exacerbated by Hussein’s increasingly paranoid worldview, believing that Iraq was being seriously threatened and infiltrated by external forces, including Iran, the United States, and Saudi Arabia (Gause 2001).

To alleviate uncertainty and signal peaceful intentions with the “madman,” Saudi Arabia and Iraq signed a nonaggression pact in 1989, promising not to use force against each other and not to interfere in domestic affairs. This agreement can be seen as an attempt by Saudi Arabia to constrain Hussein’s unpredictable behavior and manage border security through a formal peace treaty. Hussein was also incentivized to signal peaceful intentions by reaching out to Saudi Arabia due to economic concerns and diplomatic isolation. However, the normalized relations established by the nonaggression treaty were short-lived. Hussein’s invasion of Kuwait in 1990 forced Saudi Arabia to cut ties with Iraq and join the coalition against the “madman” during the Gulf War. The case of Hussein and Saudi Arabia also illustrates the complex dynamics of security cooperation with the “mad” leader.

The twin examples of nonaggression pacts formed with “mad” leaders raise a broader question about how leaders’ reputations influence security cooperation.¹ This pattern is demonstrated not only with Hitler and Hussein but also with other well-known mad leaders such as Mussolini, Gaddafi, Kim Jong-il, Milosevic, Botha, Yeltsin, and Putin, where many states were willing to sign nonaggression pacts with them. Why are states willing to make a deal with “devils” by signing nonaggression agreements despite their reputation for madness? Equally puzzling, why would these mad leaders seek to negotiate nonaggression pacts that could potentially downplay their reputation and even run counter to the strategy of Madman diplomacy?

These questions motivate this study to explore the connection between the Madman Theory and nonaggression pacts. The Madman Theory suggests that being perceived as a madman is often considered an asset in crisis bargaining (Ellsberg 1959; Schelling 1980); however, it also generates uncertainty and fear that urges other states to seek security solutions. The anecdotal cases demonstrate that leaders with a strong reputation for madness might motivate other states to seek security assurances and sign nonaggression pacts with them. To manage risks and uncertainties associated with “crazy” leaders, states may attempt to seek security assurances through various means. While other security strategies are available, such as balancing or bandwagoning, nonaggression pacts serve as effective tools for conflict management (Mattes and Vonnahme 2010), alleviating the threat from the mad leaders.²

On the other hand, security agreements take two to tango. It might seem somewhat counterintuitive that “mad” leaders find nonaggression pacts favorable, but I argue that leaders practicing “Madman diplomacy” are also inclined to pursue nonaggression pacts for strategic purposes. Leaders known for their strong reputation for madness may often find themselves diplomatically isolated or facing potential security coalitions formed by wary states. To moderate their perceived madness and alleviate international isolation, such leaders may strategically pursue nonaggression pacts as costly signals of peaceful intentions (Long, Nordstrom, and Baek 2007; Lupu and Poast 2016). Occasionally and selectively signaling peaceful intentions through nonaggression treaties can also bolster the effectiveness of their reputation for madness, as it may be more beneficial when perceived as situational rather than dispositional madness (McManus 2019). Taken together, I argue that leaders with a strong reputation for madness increase the likelihood of forming nonaggression pacts, with this tendency moderated by geographic proximity and the number of alliances held by threatened states.

To test my theoretical argument, I utilize McManus’s perception-based measure of leaders’ perceived madness (McManus 2021) and the Alliance Treaty Obligations and Provisions (ATOP) dataset (Leeds et al. 2002). The empirical results indicate that leaders’ strong reputations for madness increase the likelihood of forming nonaggression pacts, with this effect moderated by geographic proximity and the number of alliances held by threatened states. This supports my theoretical expectation that leaders’ reputations for madness play a significant role in security cooperation. My findings contribute to the growing consensus that leaders’ characteristics and reputations matter in international politics, carrying important policy implications, especially given the rising presence of “mad” leaders on the current international stage.

The remainder of this paper is organized as follows: First, I review the literature on nonaggression pacts and the Madman Theory. Subsequently, I explore the link between leaders’ reputations for madness and the formation of nonaggression pacts. While I acknowledge that other factors also contribute to the formation of these pacts and do not claim that leaders’ reputations for madness are the paramount factors—since such well-known mad leaders are rare in diplomatic history—I carefully discuss the scope, condition, and limitation in my theoretical argument. In the subsequent sections of the paper, I detail the research design and then discuss the empirical findings, robustness, and implications.

What Do We Know about Nonaggression Pacts

Nonaggression pacts are formal security treaties signed by states wherein the parties commit to refraining from the use of military force against each other, explicitly incorporating a nonaggression provision. Specifically, signatory states pledge “to refrain from the use of force in relations with the alliance partner, to refrain from participating in any action against the alliance partner and/or to settle all disputes peacefully in relations with the alliance partner” (Leeds 2005, 22). While the alliance literature is extensive, nonaggression pacts receive comparatively less scholarly attention than traditional military alliances, despite being more frequent.³

It is crucial to distinguish nonaggression pacts from other alliance treaties. Unlike military alliances, nonaggression pacts do not entail an obligation for active military support during wartime, and scholars acknowledge that their purpose differs significantly (Leeds et al. 2002; Leeds and Savun 2007; Long, Nordstrom, and Baek 2007; Mattes and Vonnahme 2010; Owsiak and Frazier 2014; Lupu and Poast 2016; Edry, Johnson and Leeds 2021). Despite being formal security agreements, their validity and sincerity may be questioned, as they could be perceived as mere cheap talk. It is possible that nonaggression pacts offer states a convenient way to extend an olive branch at minimal cost, and it might be seen as a reflection of pre-existing peaceful relations rather than a proactive tool for maintaining peace. From this perspective, nonaggression pacts may serve as symbolic gestures that only parties already dedicated to peaceful interaction sign nonaggression pacts, yet these agreements may have limited independent causal impact.

The question that arises is how should we conceptualize nonaggression pacts, and what motivates states to sign such security agreements? Firstly, previous studies suggest that nonaggression pacts should be conceptualized as costly signaling devices (Long, Nordstrom, and Baek 2007; Lupu and Poast 2016), akin to other formal security treaties that convey signals regarding future behaviors (Smith 1995; Fearon 1997; Morrow 2000). The signal of nonaggression pacts entails costs, as security agreements involve audience costs (Fearon 1997) and trade-offs with other areas of interest during negotiations (Altfeld 1984; Morrow 1991). Leaders encounter domestic audience costs and potential punishment for not honoring such agreements (Fearon 1997; Long, Nordstrom, and Baek 2007), as well as international reputational costs resulting from violations of international agreements (Gibler 2008; Crescenzi et al. 2012; Mattes 2012). Hence, nonaggression pacts should not be dismissed as cheap talk. If the inclusion of a nonaggression provision were entirely costless, it would be a standard component in all alliance treaties, which is not the case (Long, Nordstrom, and Baek 2007). Moreover, according to Lupu and Poast (2016), nonaggression pacts serve as communicative tools conveying peaceful intentions in the future not only between the involved parties but also with third parties. States sign nonaggression pacts to signal peaceful intentions and the termination of rivalries, addressing information asymmetry and reducing uncertainty (Lupu and Poast 2016).

Secondly, nonaggression pacts serve as tools for conflict management. In situations where the expectation of conflict is minimal, signing nonaggression pacts may be unnecessary, as peace can be sustained without them. Nonaggression pacts are often formed by states with close ties but a perceived high risk of conflict, such as between rivals and neighbors. Consequently, dyads anticipating higher levels of hostilities are more likely to seek peace assurances through nonaggression pacts (Leeds and Mattes 2007; Long, Nordstrom, and Baek 2007; Mattes and Vonnahme 2010). Mattes and Vonnahme (2010) argue that nonaggression pacts play a crucial role in reducing the likelihood of conflict by alleviating uncertainty and fears among rivals and establishing a peaceful channel for dispute resolution.⁴ As tools for conflict management, nonaggression pacts become essential for parties engaged in prolonged, antagonistic interactions, particularly among rival states, in order to sustain peace. Moreover, nonaggression pacts serve as instruments for managing internal threats and preventing foreign intervention (Owsiak and Frazier 2014). In particular, states may form nonaggression pacts to thwart international support for a rebel group, breaking the connection between rebel groups and foreign governments (Edry, Johnson, and Leeds 2021). Nonaggression pacts can effectively mitigate the risk of both international and internal conflicts.

Furthermore, the role of reputation has been extensively explored in the alliance literature, and empirical evidence largely supports the notion that reputation matters in alliance politics. Negative reputation for unreliability hinders future alliance formation (Miller 2003; Gibler 2008; Crescenzi et al. 2012; LeVeck and Narang 2017), and it also influences the design of treaties when partners have a questionable reputation (Mattes 2012; Narang and LeVeck 2019). However, existing alliance literature has primarily focused on explaining the impact at the system-level and state-level dynamics while overlooking the individual-level factors. Reputation, as commonly discussed in alliance literature, is primarily treated as a state-level characteristic with an assumption of persistence beyond leaders’ tenure (Crescenzi et al. 2012; LeVeck and Narang 2017; Narang and LeVeck 2019). The role of leaders’ reputations has received limited attention in the examination of alliance behaviors and security cooperation. While past behaviors signal states’ reputation, I argue that states’ reputations are also closely intertwined with the reputations of their leaders, aligning with the arguments put forth by McGillivray and Smith (2000, 2008), Gibler (2008), and Mattes (2012).

In summary, nonaggression pacts are not mere cheap talk. Rather, the literature suggests that nonaggression pacts serve as costly signals, signaling peaceful intentions, alleviating uncertainty, and maintaining peace among rival states. In the subsequent section, I delve into the Madman theory literature and examine the connection between leaders’ reputations and nonaggression pacts, exploring the onsets of security agreements at the individual level.

The Madman Theory

The Madman Theory remained well known in the field, particularly during the Cold War era with its significant nuclear threats. The case of Hitler exemplifies the use of “Madman diplomacy,” a strategy embraced by numerous leaders throughout diplomatic history. Notable leaders such as Khrushchev, Nixon, Hussein, and Gaddafi have also employed this approach (McManus 2019). In recent years, the concept of the “Madman” has received increasing attention in the media and the literature, reflecting the growing presence of “mad” leaders on the current international stage. For instance, former US President Donald Trump is undoubtedly associated with a reputation for madness due to his emphasis on being “unpredictable” and “mad” in his foreign policy, as reflected in his day-to-day rhetoric on Twitter, characterized by impulsivity, emotion, and provocation (Turner and Kaarbo 2021). Trump’s reputation draws comparisons to Nixon (Boys 2021), and the Madman Theory provides a helpful framework for understanding Trump’s unpredictable diplomacy in various areas (Bentley and Lerner 2021; Boussios 2021; Boys 2021; Devanny 2022).

The philosophical roots of the concept of madness in politics can be traced back to Machiavelli and Hobbes, but the Madman Theory was first fully defined and outlined by Ellsberg. Ellsberg argues that a leader, as a blackmailer, can succeed more in making demands and threats of war, especially nuclear war, if they appear “convincingly mad” in others’ view (Ellsberg 1959, 2). Two subtypes of reputations for madness are highlighted in Ellsberg’s argument: Firstly, the subtype of unpredictability, where a leader deviates from predictable decision-making, indicates that leaders possess an unusual aggressiveness and could choose to do anything at any moment. Secondly, the subtype of abnormal payoffs, where a leader sees war as an uncostly option or even treats total victory as the only acceptable outcome. The second type of madness is predictable and consistent compared to the first one; however, it is based on extreme payoffs. The framework and the two subtypes of perceived madness provided by Ellsberg illustrate a greater propensity for leaders with perceived madness reputation to use force aggressively and initiate conflicts, making leaders who appear mad or insane more likely to make credible and successful threats even if the conflict is very costly.

Similarly, Schelling also reasons that leaders with a reputation for madness can be an effective bargaining tool in coercive bargaining (Schelling 1980), suggesting that perceived madness is often considered an asset in foreign policy. Contrary to the expectation of rational leadership, Schelling notes that it is not always beneficial to be rational and calm while making a threat due to the paradox of deterrence (Schelling 2020); therefore, successful coercion may require irrationality, raising threats’ credibility.⁵  Boys (2021) illustrates that Ellsberg, Schelling, and Kissinger developed the Madman Theory independently and inadvertently influenced Nixon’s approach to foreign policy.

While the Madman Theory is widely recognized and theoretically plausible, the growing research in the field finds mixed empirical support. On one hand, some scholars suggest that the Madman strategy could have certain benefits. Rationalist scholars have incorporated the concept of craziness and irrationality into bargaining models, demonstrating that acting irrationally or pretending to be mad can occasionally enhance bargaining outcomes (Acharya and Grillo 2015; Little and Zeitzoff 2017). Studies that took psychological approaches have demonstrated that emotional inconsistency and anger are also helpful bargaining strategies in negotiations (Sinaceur and Tiedens 2006; Van Kleef and Côté 2007; Sinaceur et al. 2013; McDermott 2014; Wong 2019). Experimental evidence also supports the notion that Madman diplomacy enhances threat effectiveness, as the public may be more willing to concede to a leader perceived as erratic in certain circumstances (Schwartz 2023). The effectiveness of the Madman Theory depends on the types of madness perceived by others, as such reputation is more effective when leaders have a perception of situational extreme preferences in coercive bargaining (McManus 2019). While it is rare, leaders like Hitler did demonstrate the advantage of a reputation for madness in coercive bargaining when it was perceived situationally (Ellsberg 1959; McManus 2019).

On the other hand, many studies have concluded that the Madman Theory might be more of a myth (Sechser and Fuhrmann 2017), with limited concrete historical evidence to support its effectiveness in foreign policy as an asset (Walt 2017; Seitz and Talmadge 2020). Although the Madman strategy may be advantageous in coercive bargaining, it carries significant domestic costs, requiring leaders to navigate a two-level game (Schwartz 2023), and perceived madness may even be detrimental due to potential commitment problems (McManus 2021). The potential commitment problem highlights that successful coercion relies on the ability to credibly promise not to attack, which is a major drawback of the Madman Theory (McManus 2021). The lack of commitment to peace due to leaders’ unusual aggressiveness and the commitment problem makes it more difficult to achieve lasting peace (Weisiger 2013; Kydd and McManus 2017). Additionally, being perceived as having extreme preferences and deviating from consequence-based decision-making is detrimental because opponents believe future conflicts are inevitable, making them less likely to back down (McManus 2019). Empirical evidence also indicates that leaders’ reputations for madness may not always be that helpful in crisis bargaining (McManus 2019, 2021). In particular, different types of perceived madness may have varying effects, and only reputation for madness perceived in a situational manner is more likely to be successful compared to dispositional perceived madness (McManus 2019). In short, scholars have found mixed support regarding the effectiveness of leaders’ reputations for madness in coercive bargaining, which varies depending on different circumstances and types of madness. This underscores the need for further empirical studies.

In addition, one significant challenge in empirically testing the Madman Theory is the lack of a clear and consistent definition of perceived madness, which varies across studies. The concept of perceived madness is either defined as too narrow or too broad, necessitating further refinement (McManus 2019, 2021). Also, studies that adopt a psychological approach often measure leaders’ actual madness instead of leaders’ reputations for madness (Boys 2021; McManus 2021), which fails to capture the essence of the Madman Theory, as it is about assessing leaders’ perceived reputation rather than their actual mental health. Therefore, it is more suitable to conceptualize perceived madness as a reputation of madness and unpredictability perceived by other actors with an abnormally high likelihood of using force. I follow McManus and define perceived madness as “deviation from normal payoffs or decision making in a way that makes a leader more likely to use force” (McManus 2021, 4). Such a definition aligns with the essential logic of the Madman Theory, which asserts that leaders with reputations for madness are more successful at coercive bargaining because their threats of force become credible, thereby increasing the likelihood of adversaries backing down when they perceive the threat as genuine.

However, a notable gap in the existing literature on the Madman Theory is its primary focus on the reputation for madness in coercive bargaining but not in other circumstances. The theory only highlights the interaction with adversaries involved in crisis bargaining while disregarding its broader consequences (Schwartz 2023). Schwartz (2023) has explored the interaction between leaders with perceived madness with domestic audiences, but the interaction with international audiences is understudied. To address this limitation, I explore the linkage between the Madman Theory and security cooperation and the impact of leaders’ perceived madness in the onset of nonaggression pacts.

Nonaggression Pacts with Mad Leaders

The aforementioned cases illustrate the strategic interactions between “mad” leaders and other states, often resulting in nonaggression agreements.⁶ The question at hand is two-fold: Firstly, why are states willing to engage in nonaggression agreements with “mad” leaders? Secondly and equally puzzling, why would these mad leaders themselves prefer nonaggression pacts that potentially downplay their reputation for madness or even run counter to the strategy of Madman diplomacy?

According to the Madman theory, cultivating a reputation for madness can be advantageous in crisis bargaining, but the benefit may be limited in other contexts. Leaders typically showcase and intensify their reputation for madness in crisis bargaining situations, which persists beyond the bargaining scenario. This reputation remains under careful scrutiny by other states, particularly those facing threats from mad leaders. A reputation for madness signifies heightened aggressiveness, uncertainty, and unpredictability, which provoke fear and security concerns among other states. This dynamic is akin to a scenario in a bar where encountering someone who seems drunk and exhibiting erratic behavior, yelling loudly and acting bizarrely, would prompt others to avoid eye contact and maintain a distance from the madman due to fear of unwanted troubles. Similarly, this logic may apply to international politics, in which states leverage the reputational information and actively seek security strategies to manage and mitigate threats and uncertainties posed by leaders perceived as extremely aggressive or erratic.

To mitigate the uncertainty of the “madman” and avoid becoming the victim of the madman’s wrath, it is a logical response for other states to pursue security assurances with leaders known for their reputation for madness. Seeking security assurances often involves signing formal security treaties aimed at addressing specific security concerns and establishing some “strategic understandings” among states, a practice that has been fundamental throughout diplomatic history (Langer 1950, 1960). Nonaggression pacts serve as valuable conflict management tools for resolving existing or potential disputes between states, thereby easing tensions and decreasing the likelihood of conflicts (Mattes and Vonnahme 2010). Nonaggression pacts also signal peaceful intentions between parties, facilitating information exchange and reducing uncertainty and potential miscalculations (Long, Nordstrom, and Baek 2007; Lupu and Poast 2016). Additionally, nonaggression pacts are less costly compared to seeking formal alliance treaties because nonaggression pacts do not require any active behaviors. Therefore, to address the security concerns posed by mad leaders, negotiating and signing nonaggression pacts represents one of the solutions for managing relations with mad leaders without fighting against or allying directly with the mad leaders.

However, one might argue that there are alternative solutions available, and states may not necessarily need to seek security assurances directly from mad leaders due to many other factors. Indeed, states can employ various security strategies to address external threats, such as self-help, balancing, bandwagoning, hedging, nonaggression pacts, or all of the above. This paper does not intend to argue that seeking nonaggression pacts is superior to other security strategies, but rather that it is one among several available options for states while dealing with mad leaders. While states have alternative security strategies to counter external threats, they must carefully weigh the trade-offs involved (Morrow 1993). My argument acknowledges the trade-offs between different security strategies and explains why negotiating nonaggression pacts with mad leaders can be a prudent and common, albeit may not necessarily be the optimal, approach compared to other security strategies.

To begin with, as a classic form of external balancing, military alliances can be costly and effective signals for deterrence, often driven by significant external threats (Morrow 1993; Johnson 2017). Scholars found that military alliances can effectively deter external threats (Smith 1998; Leeds 2003; Johnson and Leeds 2011), depending on allies’ capabilities and reliability (Johnson, Leeds, and Wu 2015; Johnson and Joiner 2021). Many states strengthen their alliance ties to ensure security against mad leaders.⁷ However, alliances come at a cost and typically involve some trade-offs and policy concessions, especially for smaller states seeking protection from stronger powers (Morrow 1993; Johnson 2015). In addition, concerns about alliance reliability persist, and many alliance studies indicate that states choose their alliance partners carefully due to concerns about credibility (Gibler 2008; Crescenzi et al. 2012; Mattes 2012; LeVeck and Narang 2017). Even though formal alliances have significant strategic value, there remains an inherent risk of abandonment or entrapment. States may be betrayed by their allies or drawn into unwanted conflicts, highlighting the precarious nature of such partnerships.⁸ In this sense, while balancing is often preferred (Walt 1985), it can also be costly and risky at times.⁹

Moreover, while the literature demonstrates that alliances provide effective signals for deterrence, their effectiveness in deterring mad leaders may vary. When dealing with crazy leaders, deterrence may falter because mad leaders may be less responsive to traditional deterrence mechanisms and may miscalculate the consequences of initiating conflicts. This may be attributed to the tendency of mad leaders to pursue extreme foreign policies or simply their inability to accurately calculate the consequences when deviating from consequence-based decision-making (McManus 2019, 2021). For instance, Hitler disregarded the deterrence signals of military alliances held by Czechoslovakia and Poland, even though the Allies possessed greater military capability than Nazi Germany. Hitler was confident that the Allies would continue to pursue appeasement and abandon Czechoslovakia and Poland. Similarly, despite Kuwait’s membership in the Arab League, which promises defense by its members, Hussein gambled that intervention by other members in the region would not occur, as well as falsely believing the United States was uninterested. Although these may be rare examples, it appears traditional alliances may struggle to deter well-known mad leaders, prompting states to explore alternative security strategies.¹⁰

Pursuing alliances or a military buildup (i.e., external and internal balancing) can be costly and resource-intensive (Yarhi-Milo, Lanoszka, and Cooper 2016; Alley and Fuhrmann 2021). States may often struggle to find suitable allies capable and credible enough to counter mad leaders, and many lack the resources to fully rely on self-help to address their security concerns (Walt 1985; Morrow 1993; Crescenzi et al. 2012). If other options are unavailable, bandwagoning with the threat by allying directly with the mad leader becomes an attractive alternative (Walt 1985; Schweller 1994). However, bandwagoning is less driven by security concerns and more by the desire for gains and self-extension rather than self-preservation, achieved by directly allying with the stronger side (Schweller 1994). Allying with a mad leader can reduce the likelihood of becoming a target of aggression from mad leaders and allow states to share in the spoils of war during wartime. For example, not all European states joined the Allies against Nazi Germany; instead, Hungary, Romania, and Bulgaria joined the Axis powers by taking sides and allying directly with the mad leader for gains. However, bandwagoning with mad leaders might come with significant costs. Taking clear sides by forming alliances with mad leaders can incur reputational costs, as it positions one as an accomplice to their actions. Bandwagoning with mad leaders could lead to diplomatic isolation since allying with mad leaders is often not appreciated by the majority of states, thereby limiting opportunities to seek additional alliances. States also face the potential costs of war if dragged into unwanted conflicts due to alliances with mad leaders. States that bandwagoned with Nazi Germany were also punished and attacked by the Allies. Even adopting a pro-madman neutrality stance can lead to isolation and criticism from the international community (e.g., the Molotov–Ribbentrop Pact).¹¹

The literature often presents a dichotomous approach between balancing and bandwagoning, the two most extreme policies a state can choose by taking clear sides. I argue that a nonaggression pact represents a more nuanced and prudent option in between these extreme strategies. It offers the advantage of obtaining security assurances directly from the threat without the costs and risks associated with balancing against a mad leader or the reputational costs of directly allying with them. In some cases, the line between bandwagoning—allying directly with the strong for gains—and signing nonaggression pacts with mad leaders can be blurred. Some may argue that signing such a pact can be considered bandwagoning depending on the context and treaty details, which also bring reputational costs. However, I contend that nonaggression pacts should not be simply classified as bandwagoning.¹² Nonaggression pacts require no active support for the mad leader and are usually driven by security concerns for self-preservation, not self-extension, and not by the desire to benefit from the rise of the madman.¹³ Additionally, seeking peace through a nonaggression pact should incur less reputational costs, as it might be impractical to expect all states, especially weaker states with fewer allies, to stand up to mad leaders on their own, which can be very dangerous. Even for states with allies’ protection, seeking peace with a mad leader is still a reasonable strategy, and it is not necessarily to be opposed by their allies, as it reduces the likelihood of defending their allies against the mad leader if the target state maintains relatively peaceful relations with the threat.¹⁴ Moreover, nonaggression pacts are less likely to trigger mad leaders’ security concerns compared to balancing strategy, thereby reducing the likelihood of extreme behaviors from them due to their heightened paranoia and sensitivity. In this sense, a nonaggression pact may not be the ideal solution, but it provides a more prudent approach between balancing and bandwagoning, offering security for weaker states threatened by a mad leader, even if only temporarily buying time.

While negotiating a nonaggression pact with mad leaders can be tempting, it comes with significant risks. Mad leaders are more likely to pursue extreme policies, sometimes even deviating from consequence-based decision-making (McManus 2019). Mad leaders may view the honoring of nonaggression pacts as unnecessary and may perceive the benefits of violating such agreements as outweighing the reputational costs. A few historical examples illustrate the risks of trusting these leaders. Hitler, for instance, signed nonaggression pacts with Poland, France, Denmark, the USSR, Estonia, and Latvia, yet these agreements did not prevent war. Similarly, Mussolini and Stalin formed nonaggression pacts strategically but violated them during wartime. Although these were extreme pre-Cold War examples, similar dangers persist today. The current war in Ukraine suggests that Putin showed scant regard for nonaggression pacts signed with Ukraine by occupying Crimea in 2014 and invading Ukraine in 2022.¹⁵ Although nonaggression pacts generally reduce the likelihood of conflicts (Long, Nordstrom, and Baek 2007; Mattes and Vonnahme 2010) and are often durable (Leeds and Mattes 2007), nonaggression pacts with mad leaders can be exceptions. Therefore, relying solely on nonaggression pacts with such leaders is risky, akin to making a pact with the “devil.”

In this sense, it is possible that states opt for a hedging strategy rather than relying solely on a single security approach, especially for weaker states (Lim and Cooper 2015; Ciorciari and Haacke 2019). I argue that nonaggression pacts align more closely with the concept of a hedging strategy rather than balancing or bandwagoning. Such an approach requires states to adopt multiple strategies when available and carefully navigate their relations with potential threats without taking clear sides.¹⁶ When threatened by mad leaders, states might pursue multiple strategies simultaneously, including nonaggression pacts to buy time if necessary, as they prefer to keep their foreign policies flexible. For instance, facing threats from mad leaders on both Eastern and Western fronts, Poland first pursued nonaggression pacts with Stalin and Hitler in 1932 and 1934, respectively, shortly after these mad leaders came to power (Cienciala 1975). However, Poland recognized the limitations of nonaggression pacts and the risk of betrayal (Cienciala 2003). Nonaggression pacts bought time, and in 1939, Poland turned to the United Kingdom, signing formal defense agreements for protection before the formation of the Molotov–Ribbentrop Pact.¹⁷ In this sense, pursuing multiple strategies to counter mad leaders is reasonable and even preferable, as relying on a single strategy can be risky. While this does not imply that one strategy is inherently more likely or superior to others, my argument suggests that states threatened by mad leaders are more likely to pursue nonaggression pacts, which is a viable and prudent option at hand.

Why Do Mad Leaders Favor Nonaggression Pacts?

While it is understandable for states that are threatened to pursue nonaggression pacts with mad leaders, security cooperation takes two to tango, suggesting such pacts cannot be formed without the consent of mad leaders. Therefore, it is equally important—and perhaps more puzzling—to explain why these mad leaders would also prefer nonaggression pacts that might downplay their reputation for madness or even contradict their strategy of Madman diplomacy. I argue that mad leaders favor nonaggression pacts with other states because these agreements can alleviate their diplomatic isolation and temper their overall reputation for madness for strategic purposes.

Firstly, “mad” leaders, especially those well known for their strong reputation for madness, are more likely to be diplomatically isolated in the international community. These leaders are often isolated because they frequently face sanctions and are disliked by other states due to their erratic and aggressive behaviors. As discussed, it is relatively rare to see other states bandwagon and ally with mad leaders because of the potential reputational costs. Although a reputation for madness might be an asset in coercive bargaining, cultivating such a reputation often comes at the cost of isolation, with fewer friends in the international community.

Hence, mad leaders might need to reach out and alleviate diplomatic isolation due to economic challenges to address the downside of reputation for madness. For instance, North Korea faced sanctions and isolation for decades. To ease international pressures and internal economic challenges, including shortages of food and energy supplies, North Korea made concessions and eased tensions with South Korea in 1991 by signing a nonaggression pact and downplaying its unification policy, at least symbolically (Sang-Woo 1992). This move aimed to alleviate isolation and address economic challenges. It does not mean that North Korean leaders have abandoned their pursuit of nuclear programs or given up madman diplomacy. Instead, they employed a temporary strategy to establish less intense relations with South Korea, Japan, and the United States (Sang-Woo 1992; Merrill 1993). In this sense, nonaggression pacts serve as useful signaling tools for mad leaders to signal peaceful intentions and temporarily alleviate isolation with minimal costs.¹⁸

Secondly, nonaggression pacts can be useful tools for mad leaders to address, or at least alleviate, the drawbacks of madman diplomacy. Reputation is categorized as situational or dispositional in the eye of the beholder (Mercer 2010; McManus 2019). The effectiveness of madman diplomacy depends on how these leaders are perceived by others. McManus found that commitment problems arise when the perceived madness is viewed as dispositional rather than situational, which undermines the advantage of the Madman strategy in coercive bargaining (McManus 2019, 2021). Ideally, leaders employing madman diplomacy would prefer to cultivate a perception of situational madness if they understood how such a strategy works. This approach makes their coercion more credible and increases the likelihood of success in crisis bargaining. On the other hand, the risk of pursuing dispositional perceived madness exacerbates the commitment problem, as other states may find it futile to negotiate or make concessions with mad leaders (McManus 2019). This diminishes the effectiveness of bargaining and likely leads to military coalitions against the mad leader.¹⁹ Following this logic, “smart” mad leaders should not pursue extreme preferences on every issue all the time. Instead, they are likely to cultivate and demonstrate their madness occasionally and selectively and even show peaceful gestures sometimes, making it more likely to be perceived as situational madness, assuming they understand how the strategy works.

This logic aligns with McManus’s framework (McManus 2019), and it is possible that leaders with perceived madness tend to selectively pursue extreme preferences in a situational manner while adopting more moderate preferences or making concessions on other matters. Therefore, it is a reasonable and even preferable strategy for mad leaders to pursue nonaggression pacts with other states to signal peaceful intentions and reduce the likelihood of being balanced against. Nonaggression pacts function as peaceful signals (Long, Nordstrom, and Baek 2007; Lupu and Poast 2016), demonstrating the willingness to negotiate issues peacefully, and are more likely to cultivate perceived madness in a situational manner. This was evident in Hitler’s strategy: despite his aggression on certain issues, he reached peaceful agreements with many states, which diluted his reputation to a more situational manner as a distraction. This approach suggests that other states perceive the mad leader, such as Hitler, as pursuing limited ambitions on certain issues and being willing to compromise on others, viewing their madness as situational. Thus, leaders with perceived madness may find it beneficial to pursue nonaggression pacts to avoid being balanced by military coalitions and to dilute their reputation for madness by limiting it to situational perceived madness.

However, one limitation of my argument is that it relies on the assumption that leaders rationally adopt madman diplomacy, knowing how to exploit the leverage of reputation situationally and understanding the utility of nonaggression pacts. Additionally, McManus highlights another dimension of perceived madness: Whether such leaders are seen as merely pursuing extreme preferences or actually deviating from consequence-based decision-making (McManus 2019). Leaders likely to deviate from consequence-based decision-making are often extremely emotional or even delusional, making decisions randomly. They may be actually suffering from mental illness and unable to make rational decisions. In such cases, it is difficult to determine whether leaders are truly mentally ill or adopting madman diplomacy as a rational strategy. If they deviate from consequence-based decision-making, they may not be able to make such calculations and see the utility of nonaggression by downplaying their reputation situationally or using nonaggression pacts strategically. This might be the case for the perceived madness demonstrated by Khrushchev and Gaddafi, as they are more likely to be seen as deviating from consequence-based decision-making.²⁰ Therefore, such perceived madness may challenge my argument that other states might be unwilling to make a deal with the mad leader if they believe the mad leader is truly crazy and deviating from consequence-based decision-making. However, it is difficult to distinguish true mental illness from perceived madness empirically, even though they are rare but correlated.

Another limitation of my theory is the frequency of mad leaders in diplomatic history, especially those with a strong reputation for madness. Leaders adopting madman diplomacy are quite rare, which hinders my theory’s generalizability. While it is a common challenge for studies exploring the role of leaders, this issue needs to be addressed both theoretically and empirically. It is quite possible that this mechanism only works for well-known mad leaders. Almost all the leaders described in the anecdotal cases are well known for their strong reputations for madness, widely recognized by the international community. However, leaders with a somewhat moderate reputation for madness, or those perceived as slightly mad by a small audience, may be less likely to lead to the onset of nonaggression pacts. It is more challenging to determine whether they are truly perceived as mad or are actually adopting madman diplomacy, raising doubts and debates. Therefore, other states may feel less threatened by them, and these leaders are less likely to be isolated or need to address their reputation strategically. Hence, I acknowledge that well-known mad leaders are rare in international politics, and my theory may work best only for those with a strong reputation for madness. It is also necessary to distinguish strong and moderate madness reputations empirically, which I discuss further in the research section.

In sum, I argue that leaders’ strong reputation for madness plays a significant role in the onset of nonaggression pacts. On the one hand, states are more likely to seek security assurances and pursue nonaggression pacts with mad leaders to address security concerns, as these pacts serve as tools for conflict management and signals of peaceful intention. On the other hand, leaders employing the madman strategy are willing to form nonaggression pacts to temper their reputation for madness situationally and to alleviate diplomatic isolation strategically. This leads to the first hypothesis: 

Though states might be willing to pursue nonaggression pacts with mad leaders, it is important to consider that this impact may not be consistent and should be influenced by other conditions. Therefore, I theorize two additional conditions that could potentially diminish the impact of a leader’s strong reputation for madness on the onset of nonaggression pacts.

It is crucial to consider the geographic proximity between the threatened states and the mad leader. If a country is much closer to the mad leader, it will naturally feel more pressure as the likelihood of conflict increases when states are closer or contiguous (Vasquez 1995). Conversely, if states are far from the mad leader, although the reputation for madness is concerning, there might be less immediate need for forming a nonaggression pact as the likelihood of being directly threatened and attacked by mad leaders is much lower with increasing distance. Many discussed cases seem to support this logic. For instance, one major reason that Poland, France, Estonia, Latvia, Denmark, and the USSR sought nonaggression agreements with Hitler was their proximity to the mad leader. In contrast, states outside of Europe had fewer incentives to do so. Similarly, Saudi Arabia and South Korea, being contiguous with mad leaders, had more incentives to sign nonaggression pacts with them. For states far from the mad leader, the threat should be much lower, reducing the impact on the need for nonaggression pacts, as they have more time to consider other security strategies if needed. This leads to the second hypothesis: 

In addition, my theoretical argument highlights the trade-off between different security strategies that threatened states can adopt when encountering mad leaders. Although alliances come with costs and risks, it is common for states to seek balancing by forming military alliances to deter external threats (Walt 1985; Morrow 1993; Johnson 2017). When states lack military alliances, they should be more likely to seek nonaggression pacts with mad leaders because they are less likely to deter mad leaders and are more vulnerable during conflicts. For instance, Poland, without military alliances to deter Hitler and Stalin in the 1930s, was almost forced to seek nonaggression pacts with those mad leaders before negotiating a military alliance with the United Kingdom eventually. Similarly, Finland sought peace and bandwagoned with Stalin due to a lack of protection from external allies. However, the likelihood of seeking nonaggression pacts should be much lower when threatened states have sufficient military alliances for deterrence and protection, reducing the need to make deals with the “devil.” Therefore, I consider the impact of the threatened states’ alliance status, which should tip the balance in seeking nonaggression pacts with mad leaders. In particular, I expect that with more military alliances held by the threatened state, the impact of a reputation for madness is diminished as the threatened state is less vulnerable. This leads to the third hypothesis: 

Research Design

To test my theoretical argument, I employ a dyad-year design to investigate whether leaders’ perceived madness reputation tips the balance of signing nonaggression pacts at time t. To identify the formation of nonaggression pacts, I utilize the ATOP Project data set (Leeds et al. 2002) as it offers the most comprehensive information on alliance treaty obligations. The dependent variable, nonaggression pact formation, is binary: It takes a value of 1 if states i and j joined nonaggression pacts in year t, and 0 otherwise.

While nonaggression pacts may include other obligations, my theoretical focus is on how a leader’s reputation for madness influences the formation of treaties with nonaggression obligations, which promise to refrain from using force as a tool for conflict management. Consequently, treaties signed in a given year that include nonaggression obligations are included in the sample as nonaggression pacts, consistent with prior research (Long, Nordstrom, and Baek 2007; Edry, Johnson, and Leeds 2021).²¹ Due to the binary nature of my dependent variable, I employ the probit regression model. I limit my analysis to the period from 1986 to 2010 due to the limited data availability on my primary independent variables.

My theoretical argument requires measuring leaders’ reputations for madness as independent variables; however, this is always challenging due to the lack of a clear and consistent measurement for perceived madness, which varies across studies. While the psychological approach may provide a better understanding of leaders’ psychological patterns or actual madness (McDermott 2007; Yarhi-Milo 2018; Ramey, Klingler, and Hollibaugh 2019), it fails to capture the essence of the Madman Theory, as it is about assessing leaders’ perceived reputation rather than their actual mental health. An alternative approach would be to use a behaviorally based method by measuring various behaviors that might indicate a leader's madness. However, coding such behaviors systematically poses challenges, as they may appear insane in one situation but entirely rational and strategic in another, depending on the circumstances and context. These approaches make it difficult to capture leaders’ reputations for madness fully. Consequently, McManus (2021) argues that the Madman Theory hypotheses should be tested through the perception of madness, leaders’ reputations perceived by others, rather than leaders’ actual mental health or specific behaviors, aligning more closely with the conceptualization put forth in the Madman Theory.

Hence, the goal is to capture the perception of leaders’ madness, which should be based on public perception, as reflected in the media (McManus 2021). McManus (2021) proposes an alternative approach: measuring the frequency with which leaders’ perceived madness is reflected and called in the press, using adjectives such as crazy, insane, and irrational. These adjectives are commonly used to describe leaders’ perception of madness. The perceptions of the informed public, as reflected in the media, are likely to closely correlate with policymakers’ perceptions, and this method is more suitable for coding reputations for leaders worldwide with less personal biases (McManus 2021). While the reputation of madness in the global media may be subject to potential bias, and the image portrayed in the news or press may slightly differ from policymakers’ perceptions,²² this approach captures leaders’ widespread reputations and the strength of their reputation for madness. Most significantly, when a leader is consistently branded as “mad” or “crazy” in global media, the reputation for madness will likely be more pervasive and shared among policymakers, as reputation is in the eye of the beholders, which is suitable for testing my argument as it emphasizes the degree of perceived madness.

McManus (2021) presents a continuous measurement for the reputation for madness, ranging from 0 to 6.46103.²³ A higher score of perceived madness indicates a stronger madness reputation for a leader, as reflected by the media’s frequent depiction of them as mad. The higher madness score indicates greater collective confidence that the leader’s reputation for madness is widespread and acknowledged by the public. However, a continuous measurement of the reputation for madness may not be most suitable for this research. As discussed, my theoretical arguments may be best applied to well-known mad leaders, as they are more likely to pose security concerns and suffer international isolation. Leaders with only slight or moderate reputations for madness may not exhibit the mechanisms I describe. Although leaders with a strong and well-known reputation for madness are rare in diplomatic history, which might limit the generalizability of my argument, it is still necessary to distinguish between leaders with strong madness reputations and those with lesser reputations, as this aligns more closely with my theory. Empirically, this distinction is also necessary because the distribution of continuous madness reputation scores is highly skewed. Most leaders have no signs of a madness reputation, scoring zero. Among leader-years with scores above zero, most leaders demonstrate only a slight reputation for madness with relatively low scores, while a few leader-years show much higher reputations, indicating a well-known madness reputation.²⁴

Given the theoretical and empirical concerns, I utilize the alternative dichotomous measurement provided by McManus, Strong Madness Reputation and Slight Madness Reputation, as my main independent variables.²⁵ The Strong Madness Reputation is a dummy variable that indicates that a leader’s continuous madness score falls within the top 15 percent among non-zero values, while the Slight Madness Reputation is a dummy variable that indicates a score within the lower 85 percent among non-zero values. The Strong Madness Reputation and the Slight Madness Reputation are mutually exclusive. This approach addresses the concern of the skewed distribution and captures the nuances between strong and moderate reputations for madness, closely aligned with my theory. In addition, this approach is more suitable as it captures the context of reputation by comparing leaders’ reputations with their peers. It aligns with prior studies suggesting that reputation should be measured in context through comparisons with counterparts (LeVeck and Narang 2017; Narang and LeVeck 2019). Moreover, the top 15 percent cutoff was not chosen randomly but based on the observation that the continuous reputation for madness score increases dramatically around this point (McManus 2021). This indicates that the top 15 percent of samples have a much stronger reputation for madness and are significantly different from the rest, making them more likely to be well-known mad leaders.

I list and rank the top thirty leaders with a reputation for madness based on their total madness score during their term in Table 1. I also list the frequency of being labeled as slight or strong madness and their average madness reputation score. The distribution largely aligns with what we might expect for the image of well-known mad leaders, such as Saddam Hussein, Kim Jong Il, Robert Mugabe, Mahmoud Ahmadinejad, Slobodan Milosevic, and Muammar Gaddafi. Some leaders frequently demonstrate a reputation for madness but never reach the threshold to be coded as strong mad leaders, such as John Howard, Tony Blair, Hugo Chavez, Ariel Sharon, Boris Yeltsin, and George W. Bush. This indicates that they do demonstrate a moderate reputation for madness, but the level of madness is no match compared to those well-known mad leaders. If my theoretical argument is correct, a strong madness reputation should have a more significant impact, while the explanatory power diminishes for a slight madness reputation.

Lastly, there is a concern that the reputation for madness is influenced and correlated with other factors that might affect its explanatory power, raising broader questions about what contributes to perceived madness and how it influences my studies. Following McManus’s approach (McManus 2021), I briefly test some factors that might influence leaders’ perceived madness. McManus found that perceived madness is influenced by leaders’ recent MID initiations and years in office, while democracy affects only the continuous measure and slight madness, not the strong madness (McManus 2021). I replicate McManus’s model based on the author’s Online Appendix and add several other potentially influential factors.²⁶ Firstly, I considered leaders’ capabilities by including their Composite Index of National Capability (CINC) score and major power status (Singer et al. 1972), as powerful leaders are more likely to initiate conflicts or be more active on the international stage. Similarly, I controlled for the number of alliances they share and their alliance formation behaviors as international activities (Leeds et al. 2002). I also controlled whether a leader is labeled as a “rogue state” by the United States (Caprioli and Trumbore 2005), as such leaders are more likely to be called crazy. Additionally, I considered the number of rivals a leader has in a given year (Rasler and Thompson 2006), as more enemies could result in more criticism. Lastly, similar to recent Militarized Interstate Disputes (MIDs) initiation, I controlled for war initiations and whether the country was in wartime in the model (Sarkees and Wayman 2010).

The results are displayed in Online Appendix Table S3 and are largely consistent with McManus’s findings. I do not find evidence that such reputation is strongly correlated with a state’s capabilities or major power status. Leaders of major powers are more likely to generate slight madness reputation, but not in predicting strong madness as virtually none of the major power leaders are identified as having strong madness. Additionally, I found no evidence to suggest that this reputation is strongly affected by alliance behaviors, the number of alliances, or the number of rivals. Although leaders are more likely to be called crazy during wartime, it is not statistically significant in predicting strong madness. I do find that rogue state status consistently predicts leaders being called crazy, which is not surprising. In sum, most variables are not statistically significant, indicating that the determinants of perceived madness require further exploration. Due to space constraints, I leave further exploration for future studies.

Lastly, to account for the impact of other factors, I include a number of control variables related to the onset of nonaggression pacts. Firstly, I take into account the impact of regime differences and democracy by controlling the joint democracy and controlling the regime type differences (Marshall, Jaggers, and Gurr 2010).²⁷ Secondly, I control for the impact of major power status and the relative military capabilities according to the COW project (Singer et al. 1972).²⁸ Next, it is necessary to control for distance, as nonaggression pacts are less needed when states are far apart and not directly threatened. I control for the capital-to-capital distance, which more comprehensively captures the level of threat compared to contiguity (Bennett and Stam 2000). I follow McManus (2021) and incorporate control for the recent MID initiated by the mad leaders over the past 5 years in the dyad (Palmer et al. 2022).²⁹ Additionally, rivals are more likely to engage in hostile relations and have a greater need for nonaggression pacts (Mattes and Vonnahme 2010). I control for the rivalry relations as well as the duration of the rivalry (Rasler and Thompson 2006).³⁰ To address the concern that states only need to sign nonaggression pacts when they are vulnerable and threatened by mad leaders, I identify the less powerful state in the dyad and control for their number of active alliances (Leeds et al. 2002), number of neighbors (Douglas et al. 2002), and number of other active rivals (Rasler and Thompson 2006), as these factors influence a threatened state’s likelihood of signing a nonaggression pact with a mad leader. Lastly, following McManus (2021), I control for the temporal domain by including a cubic polynomial for peace years.

Main Results

The empirical results are displayed in Table 2. The three models presented each test with a different hypothesis. The first column reports the test of Hypothesis 1, while Models 2 and 3 include interaction terms for testing Hypotheses 2 and 3, respectively. Each model includes the main independent variables and the discussed control variables.³¹

Model 1 explores the role of a reputation for madness in forming nonaggression pacts. Specifically, my theoretical argument suggests that having a strong reputation for madness is more likely to form nonaggression pacts compared to having a slight reputation for madness. I find both the strong and slight madness variables to be positive and statistically significant; however, the difference between them is less clear in the main model since both strong and slight madness are statistically significant. Though my theory emphasizes the impact of strong madness, it is necessary to further evaluate the substantive effects between these two variables.

I calculate and demonstrate the predicted probabilities of strong and slight madness, and the comparison of their effects is reported in Figure 1. The graph on the left demonstrates the predicted probabilities of slight madness in predicting nonaggression pacts. The predicted probability increases from 0.0023 to 0.0029 when leaders demonstrate a slight madness reputation, representing an increase of approximately 25 percent. The impact of slight madness is meaningful but is dwarfed when compared to the substantive impact of a strong madness reputation. The graph on the right reports the predicted probabilities of strong madness. The predicted probability increases from 0.0023 to 0.0102, representing approximately a 337 percent increase. The comparison of substantive effects indicates that leaders with a strong reputation for madness are much more impactful in predicting nonaggression pact onset than those with a slight madness reputation. The comparison supports my hypothesis that while leaders identified with a strong madness reputation are quite rare, they tend to be more impactful compared to those with a slight madness reputation.³²

To test Hypothesis 2, Model 2 evaluates the interaction between a leader’s reputation for madness and distance. The coefficient for the interaction between strong madness and distance is negative and statistically significant. This suggests that the effect of the strong madness reputation decreases as distance increases because the threat posed by a mad leader is less urgent when they are farther away, supporting Hypothesis 2. As previously demonstrated, a significant coefficient alone may not provide conclusive evidence. I computed the marginal effects of the interaction, shown on the left in Figure 2. The graph indicates that the marginal effect of a strong madness is about 0.085 and significant when states are close. However, as the distance between states increases, the marginal effect decreases significantly to near zero. This indicates that the threat posed by leaders with strong perceived madness is immediate and pressing when they are closer. However, the urgency to form a nonaggression pact with such leaders decreases as the distance increases, indicating that the impact of a strong madness reputation is significantly influenced by geographic proximity. This further supports Hypothesis 2 that the impact of strong madness is more pronounced when distances are shorter but diminishes with greater distances. Similarly, the interaction between strong madness and contiguity, demonstrated in the robustness check in Online Appendix Table S16, aligns with the notion that states are more likely to sign nonaggression pacts with nearby strong mad leaders but feel less compelled to sign such pacts with distant mad leaders.³³

Model 3 evaluates Hypothesis 3 and includes the interaction between leaders’ reputations for madness and the number of alliances held by threatened states. Threatened states are expected to be more likely to seek peace with mad leaders, but they may have fewer incentives to do so when protected by military alliances. The impact of strong madness is anticipated to be moderated by the presence of more active alliances. The coefficient of the interaction between strong madness and the number of alliances is negative, indicating that the influence of a strong madness reputation is significantly diminished when more alliances are held. Similarly, I illustrate the marginal effect of this interaction on the right side of Figure 2. The marginal effect of a strong madness reputation decreases from approximately 0.018 to 0.011 as the number of alliances held by threatened states increases. This suggests that when threatened states have more military alliances, they are less likely to seek nonaggression pacts with mad leaders as they are less vulnerable, thereby supporting Hypothesis 3.³⁴

Overall, I find clear evidence that a leader’s strong reputation for madness significantly impacts the onset of nonaggression pacts, but not for leaders’ slight reputation for madness. However, this effect is substantially moderated by greater distances and a greater number of available alliances, supporting my hypotheses.

Robustness

While I find strong evidence supporting my argument, there are some potential concerns regarding potential measurement bias. As highlighted by McManus (2021), a potential issue is the possibility of reporting bias in how the use of madness adjectives is reported in the press. I address these concerns and conduct several robustness checks to increase confidence in these findings. Robustness estimates for all three models are displayed in the Online Appendix.

I first investigate the potential bias arising from variations in reporting frequency across a variety of factors. To mitigate this issue, I controlled for time-fixed effects, region-fixed effects, country-fixed effects, and dyad-fixed effects in the regression model. The results, as displayed in Online Appendix Tables S4–S7, are consistent with the main models. It is important to note that the data largely rely on Western sources; there might be a significant concern about a pro-Western or pro-US bias against leaders who challenge Western or US hegemony (McManus 2021). These leaders are more likely to be labeled negatively with terms like “mad” or “crazy” instead of positive adjectives (McManus 2021).

To address potential concerns that arise regarding measurement bias related to the use of the madness adjectives reported in the press. I follow McManus (2021) and implement additional measures to mitigate the pro-Western or pro-US bias. The affinity with the United States is likely to influence how leaders are described in the media, with states having lower affinity with the United States being more frequently criticized and labeled as mad. To account for the influence of the affinity with the United States, I first control for UN voting affinity with the United States in the regression model (Gartzke and Jo 2006; Voeten and Merdzanovic 2009). The results are consistent and reported in Online Appendix Table S8. Next, following McManus (2021), I further address the potential pro-US bias by excluding US sources from the madness reputation measurement. Using a measure based solely on non-US sources reduces US media’s potential influence and bias. The consistent results are reported in Online Appendix Table S9. Additionally, I address broader pro-Western bias by excluding all G7 countries from the sample, as they are the most influential Western countries, and their sources are more likely to present disadvantages to non-Western leaders. Finally, I consider potential bias against “rogue states” as labeled by the United States, which are more frequently characterized as having mad leaders (Caprioli and Trumbore 2005). My results remain consistent even after excluding major Western countries or “rogue states,” as reported in Online Appendix Tables S10 and S11.

Next, I have included a variety of controls in the regression model to address the threatened states’ strategic vulnerability. Firstly, I control for the level of external threat faced by the threatened states. If states feel more threatened or under more external pressure, they are more likely to seek nonaggression pacts to mitigate the threat from mad leaders. Following Leeds and Savun (2007), I calculate the external threat based on the formula that considers both the similarity and capability of the states’ threat environment. The results are displayed in Online Appendix Table S12, and the overall results remain unchanged. Next, although I control for the states’ affinity with the United States in previous models, it is also important to control for the affinity between dyadic members, as nonaggression pacts are likely influenced by ideological similarities and affinity. The consistent results are demonstrated in Online Appendix Table S13. Additionally, I include the interaction between the number of alliances and the CINC ratio as a more comprehensive way to control the capabilities and vulnerabilities that the threatened states may face. The results are displayed in Online Appendix Table S14, and my results still hold. Moreover, as previously demonstrated, the years a leader has been in office are positively related to their reputation for madness. I control for the mad leaders’ years in office, and the results, shown in Online Appendix Table S15, remain the same.

I also explore several interaction effects in Online Appendix Table S16, as the impact of a strong madness reputation might be influenced by various factors. First, I include the interaction term between perceived madness and rivalry, as rivals are more likely to engage in hostile relations and seek nonaggression pacts. However, the interaction effect with rivalry is not statistically significant. As discussed, similar to the interaction with distance, the interaction term between the reputation for madness and contiguity is positive and significant, indicating that this impact is amplified for neighboring mad leaders. Additionally, I include the interaction term between recent MID initiations and perceived madness, as such a reputation may be amplified through recent conflicts. Interestingly, the interaction term is negative, indicating that mad leaders involved in recent MIDs are less likely to form nonaggression pacts. This may suggest that states believe seeking nonaggression pacts, given recent conflicts initiated by mad leaders, might be futile and are more likely to pursue other strategies instead. Furthermore, I include the interaction terms for whether the threatened state is a member of NATO or the Warsaw Pact, as membership in powerful alliances might reduce the need for nonaggression pacts. Similarly, I include the interaction with nuclear capability for a similar logic. Lastly, since democracies are less likely to be associated with mad leaders, I include the interaction of democracy with the less mad state. However, I do not find evidence that these factors affect the impact of perceived madness.

Lastly, I examine the impact of perceived madness on other forms of treaties to see whether the impact extends to different types of treaty obligations as placebo tests. I use defense obligations, neutrality obligations, and consultation obligations as the dependent variables, and the results are displayed in Online Appendix Table S17.³⁵ Interestingly, leaders with a strong reputation for madness do not deter military alliances; instead, they are more likely to form defensive and consultation pacts. This supports the possibility of an alternative bandwagoning strategy with mad leaders that engaging in active military cooperation with them might be advantageous for potential gains (Walt 1985; Schweller 1994). Although uncommon, Putin’s recent visit to North Korea, where he signed a mutual defense agreement, indicates that mad leaders who are diplomatically isolated may actively seek military alliances. However, this raises the question of who they are approaching since they are isolated by the majority in the international system. Their choice of defense or consultation pacts might be limited, indicating an avenue for future studies.³⁶ To ensure that my main argument and results hold, given that mad leaders actively pursue security treaties in general, I test the formation of “pure nonaggression pacts,” which are nonaggression pacts that do not require any active military obligations (Mattes and Vonnahme 2010; Lupu and Poast 2016). The results in Online Appendix Table S17 remain consistent, showing that the role of perceived madness holds even in pure nonaggression pacts.

Overall, the robustness checks demonstrate the consistency of my main results, addressing potential measurement concerns, biases, and other related factors. The main independent variable, strong madness reputation, remains consistent across all models, indicating a significant impact on nonaggression pact formation even though such leaders are rare. In contrast, a slight reputation for madness is not significant or consistent across all models, suggesting that having a moderate reputation for madness is not as impactful. This supports my argument that the mechanisms I propose only work for extremely mad leaders.

Conclusion

While the Madman Theory suggests that a reputation for madness can be a strategic asset in crisis bargaining (Ellsberg 1959; Schelling 1980), existing research has primarily focused on its role in crisis situations, necessitating greater exploration of its broader implications (Schwartz 2023). This research investigates the impact of leaders’ reputations for madness on the formation of nonaggression pacts, extending the scope of the Madman Theory beyond crisis bargaining to broader security cooperation contexts.

My theoretical argument considers the trade-offs between various security strategies, suggesting that while nonaggression pacts may not be ideal, they offer a subtle and practical approach to managing and mitigating the risks posed by mad leaders. This aligns with existing research suggesting that nonaggression pacts are effective conflict management tools (Mattes and Vonnahme 2010) and valuable signals of peaceful intentions (Long, Nordstrom, and Baek 2007; Lupu and Poast 2016). Meanwhile, I argue that nonaggression pacts are also favored by leaders who adopt madman diplomacy as tools to alleviate diplomatic isolation and mitigate their overall reputation for madness by portraying their behavior as situational rather than dispositional (McManus 2019).

While acknowledging the challenge of empirically evaluating leaders’ actual madness and the rarity of well-known mad leaders in diplomatic history (McManus 2019, 2021), I utilize McManus’ perception-based data to test my argument. The results overall support the notion that leaders’ perceived madness plays a significant role in the onset of nonaggression pacts. In particular, only leaders with a strong reputation for madness consistently influence nonaggression pacts, whereas moderate levels of madness do not have a similar impact. I also explore the conditions under which states have less need to make a deal with the “devil.” The impact of a strong reputation for madness diminishes when states are distant from the mad leader or are protected by more military alliances, underscoring the intricate dynamics between a leader’s reputation and security agreements.

My findings support the growing consensus that leaders play an important role in international politics. While the role of reputation is well explored in security agreements (Gibler 2008; Crescenzi et al. 2012; Mattes 2012; LeVeck and Narang 2017), this research highlights that a leader’s reputation is also influential in security agreements. While leaders with a strong reputation for madness are rare, their impact remains significant even after controlling for country-level and dyadic-level factors. This indicates that not only a leader’s specific behaviors or experiences matter but also how a leader’s reputation is perceived by others on the international stage, conveying important reputational information. This study also contributes to and extends the Madman Theory, highlighting the broader impact of perceived madness in security cooperation. Building on McManus’s framework (McManus 2019), it is reasonable to expect that leaders who cultivate a reputation of madness are likely to favor a situational reputation rather than a dispositional reputation by occasionally demonstrating peaceful intentions as a strategy. My findings indicate the possibility that well-known mad leaders not only understand how Madman diplomacy works but also recognize the importance of managing their reputation for greater utility.

My findings illustrate notable policy implications as we face more “mad” leaders in the current international arena. While I discuss the trade-offs of different strategies, negotiating a nonaggression pact with such leaders can be tempting but comes with significant caveats. Although nonaggression pacts effectively reduce conflicts (Long, Nordstrom, and Baek 2007; Mattes and Vonnahme 2010), those involving mad leaders may be risky due to the commitment problem. Several historical examples from WWII support this notion, and this pattern may extend to the current international stage. For instance, while Trump did not violate nonaggression pacts, his reputation for disregarding international agreements—such as bullying NATO allies and withdrawing from the Paris Agreement—raises concerns about whether “mad” leaders can keep their word. Additionally, the current war in Ukraine highlights the risks of relying solely on nonaggression pacts with mad leaders, which cannot always guarantee peace. This indicates that while nonaggression pacts can be less costly security solutions, it is advisable to have other strategies in place due to the high cost of failure. Although well-known “mad” leaders are rare, the commitment problem they pose suggests that trusting them solely with nonaggression pacts can be risky, necessitating additional strategies when available. The question of whether “mad” leaders are less likely to violate these agreements and how treaty violations contribute to perceived madness remains an intriguing avenue for future study.³⁷

Following this logic, exploring what strategies are more effective in countering mad leaders would be another interesting avenue for future study. Conventional wisdom suggests that classical balancing behaviors, seeking alliances for deterrence, are most common and effective (Walt 1985; Johnson and Leeds 2011). However, as discussed, alliance deterrence can sometimes fail with mad leaders, possibly because they always pursue extreme preferences or cannot fully calculate the costs of conflicts by deviating from consequence-based decision-making (McManus 2019). Deterrence fails if the challenger cannot calculate these costs accurately. This insight could enhance the study of alliance deterrence, as the effectiveness of deterrence signals might vary when encountering different types of leaders.

Furthermore, there is another caveat: Although it is understandable for states to form nonaggression pacts with mad leaders driven by their own security concerns, this approach might inadvertently validate and assist these leaders in tempering their reputation for strategic purposes. It is possible that it could lead to their perceived madness being viewed as situational, thereby increasing its effectiveness in future crisis bargaining. While nonaggression pacts may bring temporary peace and time, it remains uncertain if they are the “best” long-term strategy for dealing with or tolerating well-known mad leaders, as these leaders are more likely to pose a significant threat to the international system. Further investigation is needed to determine whether such strategies truly moderate the reputation for madness or enhance these leaders’ effectiveness in future crisis bargaining.

Author Biography

Yinlong Li is a PhD candidate in the Department of Political Science at the University of North Texas. His research focuses on security cooperation, conflict management, and military alliances.

Notes

Author’s note: The author would like to thank the anonymous reviewers and the editors of Foreign Policy Analysis for their constructive comments and suggestions. The author also wishes to thank Marijke Breuning, Jacqueline DeMeritt, Andrew Enterline, J. Michael Greig, Paul Hensel, Roman Hlatky, John Ishiyama, Daniel S. Morey, Makito Takei, and Tsukasa Watanabe, as well as the participants at the 2022 International Studies Association-Midwest conference for their helpful comments.

Footnotes

References