The Coercive Limits of Nuclear Weapons*

Transcription

1 The Coercive Limits of Nuclear Weapons* Todd S. Sechser University of Virginia Matthew Fuhrmann University of South Carolina *Prepared for the Annual Meeting of the International Studies Association, Montreal, Canada, March 16-19, Please do not quote or cite without permission from the authors.

2 Abstract What are the coercive effects of nuclear weapons? Sixty-five years into the nuclear age, we still lack a complete answer to this question. This article employs a new dataset of nearly 200 militarized compellent threats from 1918 to 2001 to evaluate whether states with nuclear weapons are better able to make successful compellent threats than non-nuclear states. Using a bivariate probit model designed to jointly model the initiation and outcomes of threats, we find that states possessing nuclear weapons are not more likely to make successful compellent threats. If anything, possessing the bomb reduces the likelihood that targets will accept challengers demands without a fight. While nuclear weapons may carry coercive weight as instruments of deterrence, it appears that these effects do not extend to compellent threats.

3 Project Overview What are the coercive effects of nuclear weapons? This is a central question in international relations, but sixty-five years into the nuclear age, we still lack a complete answer. The Cold War prompted a colossal body of research about the utility of nuclear weapons for purposes of deterrence that is, to prevent challenges to the status quo. In contrast, our understanding of the usefulness of nuclear threats to compel targets to relinquish possessions or change their behavior is comparatively limited. 1 Can nuclear weapons be used to blackmail other states into making concessions, or are they effective mainly as instruments of deterrence? 2 There is surprisingly little agreement about the answer. At the outset of the Cold War, American policy makers were optimistic that nuclear threats both explicit and implicit could be used to bully the Soviet Union into conforming with U.S. demands. In 1946, Bernard Baruch, a close adviser to President Truman on nuclear issues, boasted that America can get what it wants if she insists on it. After all, we ve got [the atomic bomb] and they haven t (Lilienthal 1964, 123). Indeed, Presidents Truman and Eisenhower believed that tacit nuclear threats had played a central role in compelling the Soviet Union and China to comply with U.S. demands during crises in Azerbaijan and the Taiwan Strait. The belief that nuclear weapons can compel as well as deter continues to be shared by some policy makers today: for example, the 2002 U.S. National Strategy to Combat Weapons of Mass Destruction argued that the United States must oppose proliferation because nuclear weapons 1 See Schelling (1960, 1966) for discussions about the distinction between deterrence and compellence. 2 Many scholars characterize compellence as the actual use of military force to inflict punishment until the adversary complies (e.g., Art 1980). However, we use the term compellent threat to refer to any demand that a target modify the status quo in some way, even if force was not being used at the time the threat was made. This conceptualization is useful because it creates a direct parallel to deterrent threats, which demand that a target refrain from modifying the status quo. In both cases, the most successful threat is one in which the use of force is not necessary to secure the target s compliance. 1

4 are useful as tools of coercion and intimidation. Diplomatic historians have reached similar conclusions based on studies of several high-profile crises (e.g., Trachtenberg 1991). And recent quantitative research likewise appears to support the notion that nuclear weapons are useful for compellence and not merely deterrence (Beardsley and Asal 2009b; see also Asal and Beardsley 2007; Beardsley and Asal 2009a). Yet several scholars and practitioners have disputed this optimistic assessment, concluding that threats to use nuclear weapons for purposes other than self-defense are simply not credible. For example, McGeorge Bundy, national security adviser to Presidents Kennedy and Johnson, argued that nuclear weapons have not been of great use to any government for such wider purposes beyond deterring another nuclear attack (Bundy 1984, 43). Several other recent studies have argued along the same lines, asserting that a normative opprobrium against using nuclear weapons for offensive purposes makes them unusable as instruments of compellence (Tannenwald 2007; Paul 2009). Efforts to resolve this debate empirically have suffered from two important limitations. First, many studies of the coercive effects of nuclear weapons evaluate nuclear threats in isolation (e.g., Bundy 1984; Halperin 1987; Betts 1987). Because these studies do not compare nuclear and non-nuclear threats, they cannot tell us whether nuclear weapons lend additional coercive leverage to compellent threats. Second, quantitative studies of nuclear compellence have suffered from a lack of appropriate data. Rather than evaluating the effectiveness of threats, these studies tend to correlate the possession of nuclear weapons with crisis outcomes. The problem with this approach is that it equates crisis victories that are achieved by force with those achieved through threats. At best, we can conclude from these studies that nuclear states tend to win the wars they fight but not that their threats are more effective. This article corrects for these two inferential problems by employing a new dataset of nearly 200 militarized compellent threats to evaluate whether states with nuclear weapons are more likely to make successful threats than non-nuclear states. We find that, after accounting for confounding factors, states possessing nuclear weapons are not more likely to make successful compellent threats. If anything, possession of the bomb reduces the like- 2

5 lihood that targets will accept challengers demands. While nuclear weapons may carry coercive weight in a deterrent context, it appears that these effects do not extend to compellence. Our findings are important in part because they bear on two enduring policy questions. First, some prominent former officials in the United States have called for deep reductions to the American nuclear arsenal, 3 but arms control critics contend that such measures would undermine American influence around the world. 4 If nuclear weapons have helped the United States compel its adversaries to act in line with U.S. foreign policy goals in the past, then policy makers should be cautious about major cuts to the American nuclear arsenal. However, if nuclear weapons turn out not to be useful as tools of compellence, the case against American participation in arms reduction agreements would be weakened. Second, arguments in favor of aggressive measures to prevent nuclear proliferation hinge in part on whether one believes that nuclear weapons would enable nations to engage in nuclear blackmail against their adversaries. For example, some analysts have recently argued that a nuclear-armed Iran would be able to seize control of major oilfields in the Middle East. Iran would not even have to use its nuclear weapons to accomplish this, according to one commentator; intimidation and blackmail by themselves would do the trick (Podhoretz 2007, 17). Others counter that a nuclear Iran would quickly discover that nuclear bombs are simply not good for diplomatic leverage or strategic aggrandizement (Lindsay and Takeyh 2010, 37). Critical foreign policy choices depend on which argument is more persuasive: if the former perspective is correct, then the United States might be well-served to take military action to prevent Iran from acquiring a nuclear capability. The latter view, however, implies that attacking Iranian nuclear facilities may not be worthwhile. 3 For instance, see the opinion piece by George P. Shultz, William J. Perry, Henry A. Kissinger, and Sam Nunn titled A World Free of Nuclear Weapons in the Wall Street Journal, 4 January See, for example, Why the Rush to Cut Nukes? by John R. Bolton and John Yoo in the New York Times, 9 November

6 We proceed below in six parts. First, we review the competing schools of thought on the compellent effects of nuclear weapons. Second, we discuss the two problems with existing research that were highlighted above. The third section describes our research design, data, and variables, taking care to explain how our data resolve the two major empirical weaknesses in existing research on nuclear compellence. Fourth, we present empirical tests, using a variety of measures of nuclear capability. Fifth, we examine several cases from the dataset to verify our statistical results. We conclude with a discussion of the implications of our findings. Compellent Threats and the Bomb There is a voluminous literature on the relationship between nuclear weapons and interstate conflict. For decades, scholars have debated whether nuclear proliferation would make military crises and conflicts on the whole less likely by enhancing deterrence (e.g., Waltz 1990; Mearsheimer 1993; Karl 1996), more likely by contributing to strategic instability and accidents (e.g., Geller 1990; Sagan 1993, 1994; Feaver 1993; Kapur 2007), or neither by making large-scale conflicts less likely but raising the odds of low-intensity conflicts (e.g., Snyder 1965; Jervis 1984; Gartzke and Jo 2009; Rauchhaus 2009). This debate has focused primarily on the viability of nuclear deterrence, but often overlooked is the question of whether nuclear proliferation might enable states to use nuclear weapons for offensive diplomatic purposes in other words, whether nuclear weapons help make compellent threats more effective. In this section we detail competing views about the utility of nuclear blackmail. One common view is that nuclear weapons cast a shadow over international affairs even when their use is not explicitly threatened. Proponents of this perspective argue that compellent threats are more likely to be effective if the threatener possesses nuclear weapons. Skeptics, however, argue that nuclear weapons exert little, if any, impact on the effectiveness of compellent threats, in part because it is implausible that a state would use nuclear weapons for anything other than self-defense. Below we review both viewpoints and derive explicit hypotheses to 4

7 be tested in the empirical sections later. A Looming Shadow Nuclear weapons are the most destructive weapons known to man. The prospect of facing an attack with nuclear weapons therefore ought to be sobering for any leader engaged in a diplomatic confrontation. This basic logic underlies one common perspective on nuclear weapons and compellence: that a leader faced with a threat is more likely to capitulate if the adversary has nuclear weapons at its disposal. In this view, few issues short of a nation s own survival would be worth enduring a nuclear attack. When a nuclear state issues a compellent demand and threatens possible nuclear punishment, the argument holds, the odds of success multiply. Pape (1996, 38) puts the argument succinctly: when nuclear use is threatened, the prospect of damage far worse than the most intense conventional assault will likely coerce all but the most resolute defenders. Of course, the use of nuclear weapons has rarely, if ever, been threatened explicitly in conjunction with a compellent threat. Yet many scholars argue that nuclear weapons influence crises even when their use is not explicitly threatened. As Henry Kissinger warned in 1956, overt threats have become unnecessary; every calculation of risks will have to include the Soviet stockpile of atomic weapons and ballistic missiles (351). The use of nuclear weapons need not be threatened in order to influence crises, according to this perspective. So long as nuclear weapons exist, their terrible destructive power will exert implicit coercive leverage against states whose adversaries possess them. The coercive effects of nuclear weapons are thought to be particularly powerful when nuclear capabilities are one-sided that is, when a challenger possesses nuclear weapons but the target does not. Since the target cannot threaten nuclear retaliation in response to the threat, the logic goes, it cannot match the challenger s bargaining leverage. Certainly, the view that nuclear weapons can compel as well as deter was pervasive among U.S. policy makers in the early nuclear era. Both Truman and Eisenhower, for example, believed that nuclear weapons had helped the United States compel the Soviet Union and 5

8 China to make concessions during various crises in the 1940s and 1950s (see Bundy 1984). Prior to negotiations with the Soviet Union at the end of World War II, James Byrnes, Truman s Secretary of State expressed confidence that the Soviets would have little choice but to make concessions to the United States due to its atomic monopoly (Alperovitz 1965). And Eisenhower s Secretary of State, John Foster Dulles, argued that the United States was able to secure a more favorable armistice at the end of the Korean War due to implicit nuclear compellent threats against China (Soman 2000). Indeed, there is some empirical evidence to support this notion. In perhaps the only quantitative study dedicated to evaluating the compellent effects of nuclear weapons, Beardsley and Asal (2009b) show that states possessing nuclear weapons tend to win crises more often, particularly against non-nuclear opponents. Betts conducted a detailed investigation of thirteen crises in which nuclear weapons appeared to play a role (Betts 1987), and concluded that nuclear superiority correlated with although did not clearly cause success during Cold War crises when their use was threatened, although such threats were rare then and rarer still today. Trachtenberg s 1991 review of the 1958 Berlin crisis and the Cuban missile crisis found that nuclear blackmail had indeed played a critical, even decisive, role. Snyder and Diesing (1977, ) likewise evaluated these crises and concluded that there are some empirical grounds for believing that nuclear weapons convey a coercive bargaining advantage when attempting to change an adversary s behavior. What are the testable implications of these arguments? One hypothesis is straightforward: compellent threats from nuclear states will succeed more often due to the looming shadow of nuclear punishment. Hypothesis 1. Compellent threats from nuclear states are more likely to succeed, on average, than compellent threats from non-nuclear states. A second hypothesis is more qualified, arguing that the compellent effects of nuclear weapons depend on the target s capabilities. If a target has nuclear weapons, this view holds, the compellent power of a nuclear arsenal will be neutralized. Hypothesis 2. Compellent threats from nuclear states are more likely to succeed, on average, 6

9 than compellent threats from non-nuclear states only if they are issued against non-nuclear states. Taboos and Traditions of Non-Use A contrary perspective suggests that beliefs about the usability of nuclear weapons particularly for aggressive purposes are anachronistic. Indeed, many scholars have recently argued that nuclear weapons have largely vanished from the decision-maker s menu of options during most interstate crises. Even if the use of nuclear weapons might be realistic for defending against threats to national survival, using them for offensive purposes is all but unthinkable today. This view holds that nuclear threats in support of compellence are simply not credible (Jervis 1989; Mueller 2010). Even if such threats are credible, the conditions under which nuclear weapons are useful for offensive diplomatic purposes are exceedingly rare (Pegahi 2009). Robert McNamara put it bluntly, arguing in an influential article in Foreign Affairs that nuclear weapons are totally useless except only to deter one s opponent from using them (McNamara 1983, 79). 5 Why might this be the case? A variety of scholars have suggested that the use (and perhaps merely the threat) of nuclear weapons for compellent purposes would carry prohibitive costs. Paul, for example, has suggested that using nuclear weapons against an adversary in support of compellent demands would be seen as a disproportionate, inviting both domestic and international condemnation (Paul 1998, 28-30). Huth and Russett concur, arguing that this dynamic is particularly acute for nuclear powers that threaten non-nuclear states: whereas some nuclear threats were made against non-nuclear powers in the first decade or so of the nuclear era, nuclear weapons now are seen as overkill (1988, 35). While the costs associated with nuclear use might be worthwhile for a state whose survival was being threatened, such is not the case for a state seeking to make territorial or other gains. 5 Dean Rusk made a similar claim, reflecting on his experience as Secretary of State: the simple fact is that nuclear power does not translate into usable political influence (quoted in Trachtenberg 1991, 235). 7

10 This dynamic, Paul argues, has helped engender a historical tradition of non-use of nuclear weapons among nuclear states (2009). Other scholars go even further, arguing that the opprobrium against the use of nuclear weapons amounts to more than a tradition in their view, it constitutes a powerful taboo that renders nuclear threats incredible except in extreme cases of self-defense. Nuclear weapons today, according to this argument, are increasingly seen as illegitimate, uncivilized, and unusable as weapons of warfare. The consequence, according to Price and Tannenwald (1996, 140), is that the strength of the nuclear taboo and the odium attached to nuclear weapons as weapons of mass destruction render unusable all nuclear weapons. Attempting to use nuclear weapons to compel an adversary would be seen as abhorrent and have little effect. Taboo theorists point to several sources of empirical support. Their research suggests that even in crises involving nuclear-armed states, the use of nuclear weapons is so unthinkable that it is rarely even mentioned and plays little role in leaders crisis calculations (Tannenwald 2007). When the United States threatened Serbia over ethnic cleansing in Kosovo in 1999, for instance, there was no suggestion that the use of nuclear weapons might be considered if Serbia did not comply. Nor was the nuclear option seen to be available in the U.S. invasion of Iraq in 2003, the Israeli war against Lebanon in 2006, or the Russian war against Georgia in When targets of compellent threats by nuclear states capitulate, according to this perspective, it is due to other factors not nuclear weapons. Several empirical studies appear to confirm these skeptical views about the coercive utility of nuclear weapons. In 1984, McGeorge Bundy reviewed the U.S. record of atomic diplomacy and concluded that it was surprisingly poor. His conclusions were all the more notable given that he concluded that nuclear weapons had not contributed to the success of U.S. compellent threats in the Cuban missile crisis, during which Bundy had served as national security adviser. Indeed, Bundy echoed the views of several other participants in that crisis, who argued that the Cuban missile crisis illustrates not the significance but the 8

11 insignificance of nuclear superiority. 6 Morton Halperin s 1987 review of U.S. atomic diplomacy reached a similar conclusion. Halperin considered nineteen nuclear crises between 1946 and 1981 and found that nuclear weapons have never been central to the outcome of a crisis (46). Moreover, even if nuclear weapons played a role in some crises, he argued, their effectiveness was limited to a deterrent capacity in other words, to prevent a government from acting rather than compelling it to act (47). Other scholars have questioned the importance of nuclear weapons in specific crises: Foot (1988), for example, argues that nuclear weapons did not help the United States achieve better settlement terms in the Korean War; Soman (2000) likewise concludes that nuclear weapons did not compel China to alter its behavior in the 1950s, despite officials assertions to the contrary. And Holloway (1994, 271) argues that there is little evidence to suggest that the United States was able to use the bomb to compel the Soviet Union to do things it did not want to do. Beyond these studies, there is also some quantitative evidence (e.g., Gelpi and Griesdorf 2001) suggesting that nuclear states are no more likely to accomplish their objectives in interstate crises once one controls for such factors as conventional military power, signaling behavior, and the balance of interests. While we do not distinguish between taboo-related arguments and other skeptical views of nuclear coercion in this article, we do test their common prediction about the impact of nuclear weapons on compellent threats: Hypothesis 3. Compellent threats from nuclear states are no more effective, on average, than threats from non-nuclear states. 6 See the joint statement in Time magazine by Bundy, Dean Rusk, Robert McNamara, George Ball, Roswell Gilpatric, and Theodore Sorensen (1982). Many other participants in the crisis expressed doubt about the importance of nuclear weapons; see the quotations in Trachtenberg (1985). 9

12 Inferential Problems in Existing Studies The studies reviewed in the previous section have contributed tremendously to our understanding of the potential coercive dynamics of nuclear weapons. However, they share two important empirical limitations. First, many of these studies adopt indeterminate research designs (King et al. 1994, ), attempting to draw conclusions about the effects of nuclear weapons without considering cases in which they were not present. Second, quantitative studies of nuclear coercion employ datasets that were not designed and are not appropriate for evaluating the effectiveness of coercive threats. These datasets contain few actual threats, and they conflate military victories with successful coercive diplomacy. Together, these flaws imply that the empirical conclusions of existing research about nuclear coercion cannot be considered reliable. Indeterminate Research Designs A common approach to evaluating the credibility of nuclear threats is to conduct in-depth case studies of crises (almost always involving the United States) in which nuclear weapons appeared to play a significant role. Bundy (1984), Betts (1987), Halperin (1987), and Trachtenberg (1991), for example, all consider Cold War crises in which the use of nuclear weapons was threatened or discussed. 7 While these studies all reach somewhat different conclusions, their methodological assumption is the same: to understand the outcomes of nuclear crises, we must focus our attention on nuclear crises. Yet this kind of research design cannot provide us with an adequate test of the hypotheses described above, for three reasons. First, while an exclusive focus on nuclear crises can help evaluate the historical role of nuclear weapons in individual episodes, it does not offer inferential leverage for ascertaining the relative effects of nuclear weapons across crises. In other words, studies adopting this 7 Other examples of a case-based approach to evaluating the effectiveness of nuclear threats include George and Smoke (1974); Blechman and Hart (1982); Foot (1988); George and Simons (1994); Soman (2000); Art and Cronin (2003). 10

13 approach cannot say whether coercive threats from nuclear states are more effective, on average, than threats from non-nuclear states. Without knowing how often non-nuclear threats succeed, we cannot conclude whether nuclear weapons help, hurt, or have no impact on the effectiveness of threats overall. The common conclusion that nuclear weapons are not useful for compellence is particularly problematic, since several studies have concluded that compellence rarely succeeds in general (George and Simons 1994; Art and Cronin 2003). The conclusion of some scholars that nuclear blackmail appears to be ineffective therefore might simply be a reflection of the overall difficulty of compellence without comparing nuclear threats to non-nuclear threats, we cannot be sure. A second problem with many narrative-based studies of nuclear coercion is that they tend to focus on crises that are deemed to be historically significant in other words, crises that resulted in wars, war-scares, or otherwise became protracted affairs. The problem is that high-profile crises also tend to be crises in which threats were not very successful, since a successful threat would ideally bring a crisis to an end before it escalated and captured public attention. Studying only well-known crises might therefore cause unsuccessful threats to be overrepresented in the study sample, in turn biasing the results toward the conclusion that nuclear weapons cannot compel. Third, qualitative studies of nuclear compellence usually emphasize episodes in which nuclear weapons appeared to play a prominent role, either because nuclear forces were alerted or because the possibility of nuclear attack was mentioned by leaders or in the media. Yet many scholars argue that the coercive value of nuclear weapons in crises persists even when their use is not explicitly threatened. If nuclear weapons indeed influence the outcomes of crises even when they are not invoked, then studying only crises in which nuclear weapons played a high-profile role might not offer an adequate measure of their utility. 8 8 To be sure, the drawbacks described above are not necessarily inherent to qualitative research designs; our point here is only that they happen to be common problems in the qualitative literature about nuclear coercion. 11

14 Inappropriate Quantitative Data A more recent development in the study of nuclear coercion is the use of quantitative data to ascertain the correlation between nuclear weapons and crisis outcomes. For example, Beardsley and Asal (2009b) employed quantitative data from the International Crisis Behavior (icb) dataset to evaluate whether nuclear weapon states win crises more often than non-nuclear states, finding that states are indeed more likely to back down and make concessions if they are facing a nuclear adversary. They conclude that nuclear weapons are useful in coercive diplomacy and that they improve the bargaining leverage of their possessors (296-97). Several other studies have used similar data to evaluate the relationship between nuclear weapons and crisis victories (e.g., Gelpi and Griesdorf 2001). Existing quantitative studies, however, actually have little to say about the effectiveness of compellent threats from nuclear-armed states. Recent research has revealed two serious problems with the datasets they use, limiting their ability to generate inferences about coercive threats. Both problems stem from the fact that the datasets typically used in these analyses were designed to assess the causes and frequency of conflict not the effectiveness of threats. First, the datasets used in these studies do not actually contain very many compellent threats. Downes and Sechser (2010) have found that about four-fifths of the crisis observations in the icb dataset involve no compellent threats; most crises in the dataset were triggered instead by unilateral raids, border clashes, and other military events in which no threats were ever issued. The Militarized Interstate Dispute (mid) dataset, also commonly used to evaluate crisis outcomes, fares even worse: the analysis by Downes and Sechser showed that only about one-tenth of the disputes in this dataset contain a compellent threat. Moreover, since these datasets do not distinguish the cases containing threats from those that do not, researchers cannot readily exclude inappropriate cases from an empirical analysis. A second problem is that the datasets conflate military and coercive outcomes. To determine whether nuclear-armed states make more effective threats, one naturally needs to know the outcomes of threats that are made. Datasets such as the icb and mid datasets 12

15 provide some of this information but, crucially, they do not distinguish between crisis victories achieved by brute force from those achieved through successful coercive diplomacy. For instance, one of the most statistically influential observations in Beardsley and Asal s (2009b) study Winning With the Bomb, which finds that nuclear states prevail in crises more often, is the 1991 Gulf War. The icb dataset records the episode as a crisis victory for the United States and each of its coalition partners, on the grounds that the U.S.-led coalition ultimately won the war. Yet the compellent threat associated with this crisis was a clear failure: the ultimatum demanding Iraq s evacuation from Kuwait was flatly rejected, thus necessitating the war in the first place. This case like the dozens of other military victories in the dataset should not be classified as a success for coercive diplomacy since the central purpose of making a threat is to achieve one s objectives without large-scale military action. At best, existing quantitative studies can help ascertain whether nuclear states are more likely to win the wars that they fight but not whether their threats are more effective. Research Design We aim to correct these problems by using a new quantitative dataset to evaluate the comparative effectiveness of threats made by nuclear-armed states. 9 The Militarized Compellent Threats (mct) dataset contains information about 194 interstate compellent threats that is, episodes in which one or more challengers issued a compellent demand against a target and threatened to use force if it did not comply comprising 225 challenger-target dyads overall. The dataset, which spans the years , contains both well-known superpower 9 The central drawback of a quantitative approach, naturally, is that it can only identify correlations between the possession of nuclear weapons and the outcomes of threats. Yet the alternative likely would be no better in this regard. Proponents of a case-study approach have noted that information about foreign leaders crisis decision-making processes is extremely rare. Thus, even in a qualitative study, the effects of nuclear threats can be inferred only from circumstantial evidence and correlations between the threats and the two sides concessions (Betts 1987, 216). 13

16 crises as well as more obscure disputes between small states. 10 Likewise, it includes crises in which nuclear weapons played a central role (e.g., the Cuban missile crisis) as well as episodes in which the possibility of nuclear attack was never mentioned (e.g., the 1994 Haitian crisis). The structure of the dataset is ideal because it helps resolve the two inferential problems described in the previous section. First, the dataset contains threats made by nuclear states as well as those made by non-nuclear challengers. 11 This is essential because the relative advantages of nuclear weapons can only be determined by comparing the success rates of nuclear and non-nuclear challengers. Second, the dataset contains only compellent threats, defined as interstate demands to change the status quo which are backed by the threat of military force (Sechser 2011). The dataset consciously excludes military clashes, raids, and wars in which compellent threats were not made. The dataset therefore allows us to evaluate coercive diplomatic successes (as distinct from military victories), thus providing a more valid assessment of the extent to which nuclear weapons make compellent threats more effective. Dependent Variable The dependent variable measures the target s level of compliance with the challenger s demands. Compellence success is a dichotomous variable that is coded 1 if the target voluntarily complies with all demands of the challenger and 0 otherwise. 12 We observe success in roughly 35 percent of the cases in the dataset. This is noteworthy because it indicates 10 The dataset does not include compellent demands made by states already at war. Thus we do not evaluate the hypothesis that nuclear weapons help states secure more favorable settlement terms during wartime negotiations. 11 In the mct dataset, 49 of 225 challengers possessed nuclear weapons at the time they made threats (21.7%). 12 We code this variable using the compliance variable in the mct dataset, which does not code threats as successful if the challenger had to use large-scale military force in order to achieve its desired outcome. 14

17 that it is far from impossible for states to effectively employ compellent threats, even if compellence is indeed harder than deterrence. Independent Variables We code three primary independent variables to empirically evaluate the relationship between nuclear weapons and successful coercive diplomacy. The dichotomous variables challenger s nuclear weapons and target s nuclear weapons indicate the nuclear weapons status for each compellence dyad member. Following standard practice in the literature (e.g., Beardsley and Asal 2009b), both of these variables are coded 1 if a state possesses at least one nuclear weapon in a given year and 0 otherwise. The relationship between a challenger s possession of nuclear weapons and coercive diplomacy, however, might be conditional on whether the target state also possesses the bomb, as we noted above. 13 We therefore include an interaction term (challenger s nuclear weapons target s nuclear weapons) to account for this possibility. To determine the first year that a state possessed nuclear weapons, we consult datasets compiled by Jo and Gartzke (2007) and Singh and Way (2004). Control Variables We control for the same confounding factors as Beardsley and Asal (2009b). 14 First, states that possess superior levels of power might have an easier time coercing their adversaries, regardless of whether they possess nuclear weapons. We include two variables in the sta- 13 In particular, the benefits that a challenger obtains from possessing the bomb might diminish if the target is also a nuclear power, other things being equal. For example, to the extent that nuclear weapons are effective instruments of coercion, U.S. threats against Brazil should in principle be more successful than similar threats against India. 14 Note, however, that not all variables are identical to those used by Beardsley and Asal since we are relying on a different dataset. 15

18 tistical model to control for this possibility. The variable capability ratio measures the proportion of material capabilities controlled by the challenger in each dyad. 15 In addition, challenger superpower is a dichotomous measure that is coded 1 if the challenging state is either the United States or the Soviet Union (from 1918 to 1991) and 0 otherwise. This indicator of military strength is important because there might be something unique about a superpower that is not captured by the material capability ratio. Second, several scholars highlight the importance of the relative stakes in explaining crisis outcomes. States with critical interests at stake should be more tolerant of costs and less likely to back down without a fight. In one respect, states are unlikely to make costly threats for marginal returns on trivial matters, so it is plausible to presume that all of the issues in the dataset are important to the challengers. Not all threats were made over equally salient issues, however. Issues related to territory and leadership, in particular, tend to be more important to states than matters of policy and ideology. Challengers therefore might have a harder time succeeding when they demand that the target cede disputed land or remove its leader from power, compared to threats over trade policy or other relatively minor issues. Salience is a dichotomous variable that is coded 1 if the challenger made a demand over territory or leadership and 0 otherwise. 16 Third, the target s resolve could be weakened as a result of actions taken by the challenger during a crisis. We include a variable measuring whether the challenger employed demonstrations of force or conspicuous military mobilizations during a threat episode This variable uses data from version 3.01 of the Correlates of War National Military Capabilities Index (Singer et al. 1972). 16 By this definition, roughly 66 percent of the compellence cases in the mct dataset involve highly salient issues. 17 Here we depart slightly from Beardsley and Asal (2009b). Since we code cases as failures if the challenger engages in military force, we do not include an independent variable to indicate whether violence was employed by the challenger since this variable would be perfectly correlated with the dependent variable. 16

19 Demonstration is a dichotomous variable that is coded 1 if force was used in this manner and 0 otherwise. 18 This variable accounts for the possibility that challengers might be able to signal their willingness to use force and therefore raise the expected costs of resistance for target states. Finally, a challenger s threat history could influence coercive diplomacy outcomes. Frequent threats could suggest highly contested, unresolved issues between states, implying that threats in the status quo are less likely to succeed. If this is true, a history of threat-making would be negatively associated with coercive diplomacy success. On the other hand, it is also plausible that repeated threats may indicate that a challenger has identified an issue on which the target is vulnerable, and for which threats are more likely to be effective. In other words, it could be the case that challengers are emboldened by past successes (Horowitz 2009). We therefore include the variable challenger s previous threats that measures the average number of compellent threats a challenger has made from 1918 to year t. Method An exclusive focus on threat outcomes could result in strategic selection bias because challengers likely avoid making threats when they expect failure. Addressing the dependence between threat initiation and demand outcome is especially important when trying to understand how nuclear weapons influence coercive diplomacy. Challengers that possess nuclear weapons can inflict catastrophic damage on targets if a crisis escalates to war. Targets therefore may acquiesce to a challenger s wishes early in order to prevent a simmering dispute from escalating to a crisis. Simply put, nuclear challengers may not have to issue formal threats in order to achieve their preferred outcomes. If this is true, nuclear weapons would aid coercive diplomacy even though we would not necessarily observe a relationship between possessing the bomb and successful compellent threats. We address this issue by simultaneously estimating the selection and outcome processes 18 This variable is obtained from the mct dataset. 17

20 using a bivariate probit model commonly known as a Heckman selection model (Heckman 1979). Two probit models are jointly estimated. The predicted probabilities from the first model (threat initiation) are transformed to the inverse Mills ratio, which is included as an independent variable in the second equation (threat outcomes). The sample for our analysis of threat initiation includes all politically relevant directed dyad-years. 19 This approach is commonly used in political science to account for strategic selection effects (e.g., Reed 2000; Von Stein 2005; Allee and Huth 2006). 20 Selection models depend on well reasoned exclusion restrictions. In other words, there must be at least one variable that is included in the threat initiation equation but excluded from the threat outcome equation. The excluded variable(s) should be related to threat initiation but unrelated to the ultimate outcome of interest. Six variables are included in the selection (initiation) equation but not the censored (outcome) model. Contiguity is a dichotomous variable that is coded 1 if the challenger and the target share a land border and 0 otherwise. The variable foreign policy similarity, calculated with the EUGene program (Bennett and Stam 2000), measures the similarity of states alliance portfolios, and rivalry indicates whether the states in a dyad are enduring rivals (Diehl and Goertz 2000). The other three excluded variables control for temporal dependence in the cross-section time-series data. Following the advice of Carter and Signorino (2010), we include a variable measuring the number of years that pass between threats (time), its square (time 2 ), and its cube (time 3 ). 19 Politically relevant dyads constitute all country pairs where a threat might conceivably be made. This method excludes dyads in which threats backed by the credible use of military force are implausible due to the distance between countries and/or the challenger s relatively minimal military capabilities (e.g., the Burkina Faso-Peru dyad). 20 This is also the approach of Beardsley and Asal (2009b). 18

21 Non-Nuclear Nuclear Challenger Challenger Threat Success 70 (39.3%) 8 (17.0%) Threat Failure 108 (60.7%) 39 (83.0%) Total 178 (100.0%) 47 (100.0%) note: χ 2 = 8.17 (p = 0.004). No Nuclear Nuclear Monopoly Monopoly Threat Success 72 (39.3%) 6 (14.3%) Threat Failure 111 (60.7%) 36 (85.7%) Total 183 (100.0%) 42 (100.0%) note: χ 2 = 9.47 (p = 0.002). Table 1. Nuclear weapons and compellent threat outcomes. Empirical Findings We begin by conducting a simple cross-tabulation analysis to expose underlying patterns in the dataset. Table 1 reveals an interesting and unexpected relationship between nuclear weapons and coercive diplomacy outcomes. Challengers that possess the bomb appear to be less likely to make successful compellent threats overall. Notice that we observe success in 17 percent of the cases when the challenger possesses the bomb, compared to 39 percent of the cases when the challenger is non-nuclear. Of the 225 dyad-year observations in our sample, only eight involve successful compellent threats from nuclear weapons states. The negative relationship between threat success and the challenger s possession of nuclear weapons is statistically significant at conventional levels (p<0.001). The second part of Table 1 sheds light on how nuclear asymmetry influences coercive diplomacy outcomes. The findings demonstrate that accounting for the target s possession of the bomb does not alter the aforementioned relationship. Nuclear-armed challengers with a dyadic monopoly in other words, whose opponents do not have nuclear weapons are considerably less likely to issue successful threats than all other challengers. While revealing, the cross-tabulation analysis does not allow us to make definitive con- 19

22 clusions. We now turn to the multivariate statistical tests to determine if the relationships change once we account for confounding variables and the factors that motivate challengers to select themselves into a crisis. The findings are displayed in Table Model 1, which treats nuclear weapons possession as a dichotomous variable, represents the core test of the three hypotheses. Challenger s nuclear weapons is negative and statistically significant, while target s nuclear weapons is positive and significant. These findings tell us both that a challenger s possession of the bomb decreases the likelihood of a successful compellent threat when the target is non-nuclear, and that non-nuclear challengers are more likely to succeed against nuclear-armed targets. When facing non-nuclear targets, a nuclear challenger s compellent threat is 71 percent less likely to succeed than a threat from a non-nuclear challenger. The interaction term is negative and statistically significant at the 99 percent level, indicating that nuclear weapons appear to hurt challengers slightly more when the target is also a nuclear power. Nuclear challengers are 73 percent less likely than non-nuclear challengers to prevail against nuclear-armed targets. Thus, a challenger s possession of nuclear weapons has a substantively similar effect on the likelihood of compellence success regardless of the target s nuclear status. Overall, the results clearly indicate that nuclear weapons are not associated with successful compellent threats. Indeed, possessing the bomb actually appears to work against challengers. Three factors unrelated to nuclear weapons appear to influence compellent threat success. First, challengers with greater conventional capabilities are less likely to make successful threats. Second, demonstrations of force orchestrated by the challenger also increase the likelihood of successful compellence. Third, states that have made more threats in the past are more likely to succeed in the present, although this result appears less statistically reliable than the other two relationships. The other controls do not achieve conventional levels of statistical significance in the outcome model. It is possible that these results are an artifact of Model 1 s particular measure of nuclear 21 Replicating the models in Table 2 using uncensored probit analysis has no effect on the substance of these results. 20

23 Logged, Continuous IV, Dichotomous IV Ordered IV Continuous IV Continuous IV Deliverable Only Compellent Threat Success challenger s nuclear weapons (0.257) (0.055) (0.000) (0.035) (0.262) target s nuclear weapons (1.516) (0.190) (0.003) (0.132) (1.422) challenger s nuclear weapons target s nuclear weapons (1.661) (0.041) (0.000) (0.017) (1.564) capability ratio (0.389) (0.372) (0.375) (0.37) (0.388) challenger superpower (0.283) (0.302) (0.323) (0.304) (0.287) challenger s previous threats (0.119) (0.115) (0.122) (0.115) (0.121) salience (0.199) (0.197) (0.202) (0.196) (0.2) demonstration (0.201) (0.200) (0.202) (0.2) (0.202) constant (0.599) (0.595) (0.614) (0.593) (0.611) Censored Observations Compellent Threat Initiation challenger s nuclear weapons (0.093) (0.019) (0.000) (0.012) (0.095) target s nuclear weapons (0.219) (0.046) (0.000) (0.030) (0.221) challenger s nuclear weapons target s nuclear weapons (0.259) (0.010) (0.000) (0.004) (0.268) capability ratio (0.106) (0.106) (0.105) (0.106) (0.103) challenger superpower (0.092) (0.096) (0.091) (0.094) (0.096) challenger s previous threats (0.043) (0.043) (0.043) (0.043) (0.043) contiguity (0.108) (0.105) (0.101) (0.104) (0.104) foreign policy similarity (0.014) (0.014) (0.014) (0.015) (0.014) rivalry (0.107) (0.108) (0.109) (0.108) (0.109) time time time constant Uncensored Observations 130, , , , ,214 note: Robust standard errors in parentheses. p<0.01, p<0.05, p<0.10. Table 2. Heckman probit estimates of compellent threat success. 21

24 status. It could be the case that the size of one s nuclear arsenal, rather than the mere fact of nuclear possession, influences compellent threat outcomes. 22 A threat from Pakistan, for instance, might have a different effect than a threat issued by North Korea against the same target, other things being equal, since Pakistan s arsenal is considerably larger. We thus consider four alternative ways of measuring nuclear status 23 in order to capture greater variation in arsenal sizes. 24 First, we employ an ordered independent variable that codes nuclear status on a 7-point scale, with 0 indicating a non-nuclear state and 6 indicating a state possessing more than 1,000 nuclear weapons (Model 2). 25 Second, we count the raw number of nuclear weapons in a state s arsenal (Model 3). Third, we calculate the natural logarithm of each state s raw arsenal size to account for potentially diminishing marginal returns to arsenal expansion (Model 4). As Table 2 shows, however, none of these alternative measures change the thrust of our main findings. A fourth alternative measurement of the key independent variable is motivated by the observation that merely possessing nuclear weapons does not necessarily imply the ability to 22 Recent research shows that a state s nuclear posture how it operationalizes its nuclear capabilities has differential effects on militarized conflict (Narang 2010). In future research, we plan to explore whether nuclear postures influence the success of compellent threats. 23 Note that the alternate methods for operationalizing nuclear status in Models 2 through 5 modify both challenger s nuclear weapons and target s nuclear weapons, as well as the interactions with these variables. 24 We obtained information on arsenal sizes from a variety of sources. Data for the de jure nuclear powers was obtained from the Nuclear Notebook compiled by the National Resources Defense Council (Norris and Kristensen 2006). The de facto nuclear weapons states India, Israel, Pakistan, North Korea, and South Africa generally do not publish data on their arsenal sizes. We obtained weapons data for these cases by consulting a variety of historical sources on each state s nuclear program. 25 The other codings are as follows: (1) 1-10 nuclear weapons, (2) nuclear weapons, (3) nuclear weapons, (4) nuclear weapons, (5) 500-1,000 nuclear weapons. 22