The Journal of Conflict Resolution, Vol. 45, No. 3. (Jun., 2001), pp

Transcription

1 The Sources of Threat Credibility in Extended Deterrence Vesna Danilovic The Journal of Conflict Resolution, Vol. 45, No. 3. (Jun., 2001), pp Stable URL: The Journal of Conflict Resolution is currently published by Sage Publications, Inc.. Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers, and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take advantage of advances in technology. For more information regarding JSTOR, please contact support@jstor.org. Thu Jun 28 10:07:

2 The Sources of Threat Credibility in Extended Deterrence VESNA DANLOVIC Department of Political Science Texas AdiM University For deterrence to work, a potential attacker needs to perceive the deterrer's threats as capable and credible for retaliation. The issue of relative capabilities has been extensively analyzed, yet the coreproblemconcerns credible intentions. The strategic literature is largely influenced by commitment theory prescribing various "manipulation of risk strategies intended to enhance threat credibility. A few analysts, however, challenge this view and recognize the importance of "intrinsic interests" in the issues at stake (i.e., inherent credibility), which can limit the effectiveness of costly signals. These two alternative views are quantitatively tested in the cases of major-powers deterrence from 1895 to A new measure of the observable category of inherent credibility is also introduced, consisting of the regional interests of major powers in the area of dispute. The results indicate strong and significant effects of regional interests on both the onset and outcomes of extended-immediate deterrence, thus suggesting relevant theoretical and policy implications. For several decades during the cold war, the "unrestricted globalism'' of containment policy left a strong imprint on U.S. strategic thought.' The question of whether and what national interests the United States, or any other power for that matter, might have in different parts of the world was not generally raised. Instead, it was replaced by the concern that the failure to stand firm against an opponent in one region would reflect negatively on the U.S. reputation for containment of potential challenges in other parts of the world. Consequently, strategic writings were mainly preoccupied with the effective techniques of signaling the U.S. intention to use force, even when its stakes were low in the particular region. In the post-cold war era, however, the demise of a major global adversary has prompted the question of whether the U.S. national interests and goals should be uniform across all world regions. Predominant strategic theories, largely shaped in the context of superpower deterrence, do not always have well-developed and ready answers to this important policy question. 1.The term is attributed to Walter Lippmann, known for his early criticism of the globalist approach in U.S. containment policy (see Steele 1980). AUTHOR'S NOTE: An earlier version of this article was presented at the annual meeting of the International Studies Association, Los Angeles, California, March I am grateful to Frank Zagare for his valuable comments and constant encouragement, and to Paul Huth for his detailed and perceptive suggestions that improved this article considerably. I also thank graduate assistants Lucy Xie and Edward Yang for their help with dataentry work and Belinda Bragg for her editorial assistance. My deep thanks to Michelle Taylor- Robinson for her helpful advice. Responsibility for any errors and interpretations is entirely my own. JOURNAL OF CONFLICTRESOLUTION, Vol. 45 No. 3, June O 2001 Sage Publications

3 342 JOURNAL OF CONFLICT RESOLUTION The occasional confusion about foreign policy goals seems then to be yet another legacy of the cold war era. Phrases such as "conceptual poverty," "from containment to confusion," or "paradigm lost" are consequently found as descriptors of the U.S. foreign policy in recent critical literature (e.g., Haas 1995; Nye 1999; Posen and Ross 1997). As Albert Einstein (1950, 113) remarked in a slightly different but still related context, "Perfection of means and confusion of goals seem, in my opinion, to characterize our age." Current world events require, therefore, a fresher look and a reinterpretation of a number of old premises inherited from the past. This study points to one of the ways to rethink current and past events as they emerge in the relations between major powers. It also represents an attempt to bring back into focus the long neglected question of the role of national interests in deterrence theory. RESEARCH PROBLEM Most theories of international conflict and deterrence that we explore and implement today were formulated or reformulated in the nuclear context of the cold war. They were meant to provide a logical road map for the age in which world stability and survival appeared to rest on "the irrationality of the rational" logic of nuclear threats. The archetypal situation was one of two nuclear states facing each other with threats of mutual annihilation if one or the other launched an attack. If either side's home territory was attacked, the response would be less unpredictable since one's own survival would be at stake. The precarious nature of nuclear deterrence arose, however, when threats were expected to deter attacks on a third nation, usually the deterrer's ally or protcgc. In the context of mutually assured destruction (MAD), the inherent value of a protcgc for the deterrer's national interests becomes a secondary question as this value could not possibly compensate for the costs the deterrer would have to sustain if the threat were carried out. Consequently, the problem of nuclear threats primarily concerns their credibility, which "depends on being willing to accept the other side's retaliatory blow. It depends on the harm he can do, not on the harm we can do" (Kahn, cited in Freedman 1989, 134). To solve the credibility problem (i.e., to make irrational nuclear threats look rational)? attention centered on the techniques of signaling intentions. This new focus, in turn, gradually overshadowed the issue of whether the protcgc was vital for the deterrer's national interests in the first place. Hence, the deterrer of the nuclear age faced the difficult task of persuading its adversary that it would be willing to endure costs disproportionate to the issues at stake in every instance when the deterrer's commitments to its allies were challenged. Otherwise, its reputation for strong resolve would be compromised, and the adversary was 2. The term rational,as used here, is identical to the notion of "instrumental" rationality, which postulates that, if all other conditions permit, actors are expected to act according to their preferences (Zagare 1990).An irrational actor, therefore, deliberately chooses a course of action that is inconsistent with its preferences. In the situation of nuclear balance, the second-strike capabilities of both sides would ensure that any nuclear attack is also suicidal for the attacker. It would thus be irrational for a deterrer to carry out its nuclear threat.

4 Danilovic/SOURCES OF THREAT CREDIBILITY 343 believed to interpret the deterrer's threats as bluffs in other places as well. To prevent such a chain of events, triggered by the failure to stand firm, the deterrer had to effectively signal its strong commitment to retaliate, even at the cost of its own annihilation. Would the United States be willing to trade Washington or New York for Paris or Bonn? This was not just a rhetorical question Charles de Gaulle asked at the height of the cold war. A number of modern strategic studies, borne out of this puzzle, are still in the process of solving it. The questions being asked, as well as the suggested answers carefully analyzed, ponder the most effective ways to signal foreign policy interests, especially when those interests are weak. Sometimes it appears to be forgotten, however, that the preoccupation with costly signals and reputation was predicated on a very specific idea, i.e., that it was not rational for an adversary to believe that the deterrer's interests were worthy of retaliation. It is clear why this was the quintessential premise of nuclear deterrence analysis. It is questionable, though, whether it is useful as a general theory of international conflict. Nevertheless, it finds advocates, both in theory and policy, even for nonnuclear deterrence encounters. Moreover, it seems that a reputation for standing firm itself often becomes the issue of national interest. Such an approach, then, can almost completely marginalize the critical question of politics, that is, what political goals are at stake in specific situations, apart from the reputational issue. This study starts with a distinction between the two main approaches to the issue of threat credibility-commitment theory and the theory of inherent credibility. It then proceeds with an extension of the idea of inherent credibility, identifying it with the traditional notion of vital national interests. Previous work on extended deterrence, as the most common and precarious form of deterrence between major powers, focused on the isolated value of the protcgc to the potential defender. This study suggests that a defender's vital national interests may be better understood by looking at the regional context. The test is subsequently conducted to compare the empirical validity of the two views of credibility issues. The empirical analysis uses a new data set of the cases of extended-immediate deterrence between major powers for the period between 1895 and It also introduces a newly developed measure of regional interests, which is examined by using the original data set of major powers' stakes in world regions during the entire observed period. After presenting the empirical findings, this article concludes with a discussion of theoretical and policy implications. THREAT CREDIBILITY AND RESOLVE The traditional view of the necessary conditions for successful deterrence is that a potential attacker must perceive the deterrer's threats as sufficiently capable and credible for retaliation. Simply put, the potential attacker needs to be convinced that the deterrer can and will execute its threats if the attack occur^.^ Although major powers can achieve and often do have the capability necessary for effective threats, their core 3. For an extensive treatment of these requirements, see Kaufmann (19.56). Schelling (1960). and Zagare (1990).

5 344 JOURNAL OF CONFLICT RESOLUTION problem is conveying their willingness to execute threats. As Kaufmann (1956) and many others have indicated, the crucial requirement for a policy of deterrence lies in the realm of intentions. Furthermore, an adversary needs to be convinced that these intentions are credible. As previously pointed out, two different schools of thought subsequently emerged from this concern with threat credibility: (1) commitment theory and (2) theory of inherent credibility. Although the fundamental differences between these theories are rarely registered in the literature, their logic roughly corresponds to Huth's (1997) distinction between "strong-interdependence-of-commitments" and "case-specific" ~redibility.~ Commitment theory can be traced to the pioneering work of Schelling, Kahn, and others. Since it is the single most influential theory of deterrence, it should be no surprise to learn that it has received much more attention in the literature than theories that focus on inherent credibility. Summaries of this classical approach to deterrence can be found in Young (1967), where it is labeled "manipulative bargaining theory," and Zagare (1996) and Zagare and Kilgour (2000), where it is called "decision-theoretic deterrence theory." A COMMITMENT THEORY In his seminal study The Strategy of Conflict (1960), Schelling essentially argued that the problem of nuclear threat credibility should be solved through refinements of signaling techniques that manipulate the levels of risk. One of the major concerns for maintaining a reputation for strong resolve comes from the understanding that commitments and events are seen as interdependent (Schelling 1966,55-59). In this view, the interdependence of events takes two forms. First, commitments and events are considered to be horizontally (spatially) "coupled," implying that a failure to react to challenges in one region would diminish others' perceptions of one's resolve to honor its commitments in other areas as well. Domino theory that guided the cold war containment policy was premised on this understanding of international events as spatially interdependent. The other form concerns the issue of past behavior and the consistency of strong resolve over time, indicating what we may call the vertical (temporal) interdependence of commitments. The Munich analogy figures prominently as its metaphorical version that has driven a number of foreign policy decisions. Schelling and other scholars in this tradition argue for the validity of both these approaches, succinctly summarized as follows: "Few parts of the world are intrinsically worth the risk of serious war by themselves...but defending them or running risks to protect them may preserve one's commitments to action in other parts of the world at later times" (Schelling 1966, 124). 4. Commitment theory, as defined here, is also partly analogous to Huth's notion of "qualifiedinterdependence-of-commitments." For the purpose of this study, I do not separate this literature from the research on interdependent commitments, either cross-sectional or temporal, because both approaches find their original source in the research tradition of solving the credibility problem of "irrational" threats (Schelling 1960, 1966). However, a detailed survey of the literature with a different theoretical aim than here, such as that done by Huth (1997), could indeed validly point to many fine differences between them.

6 Danilovic /SOURCES OF THREAT CREDIBILITY 345 As for the nature of threats, Schelling's approach develops an entire range of techniques for signaling strong commitments. The key idea is that "it does not always help to be, or to be believed to be, fully rational, coolheaded, and in control of oneself or one's country" (Schelling 1966,37). It is, consequently, assumed that the deterrer benefits by leaving the impression that the process is not completely controllable (Schelling 1960, ; Schelling 1966, 35-91). Besides recklessness and unpredictability, such as compulsive or unreliable behavior (Schelling 1966,36-43) or the "threat that leaves something to chance" (Schelling 1960, ), the deterrer's threat can be greatly enhanced by costly signals. This notion is closely related to Schelling's understanding of deterrence as "competition in risk-taking." "It [competition in risk-taking] involves setting afoot an activity that may get out of hand, initiating a process that carries some risk of unintended disaster" (Schelling 1966,91; see also, Schelling 1960, ). Deterrence theorists who subscribe to the signaling approach generally examine and refine a number of ideas generated from Schelling's works on "the art of manipulation" (e.g., Fearon 1997; Jervis 1970; Powell 1990). Fearon's (1997) recent formal analysis, for instance, adds further refinements by distinguishing between the "tie-hands" and "sink-cost" strategies. He generally finds that signaling states do better by tying their hands, i.e., by creating costs that would be paid ex post if they fail to honor their commitments, rather than by using a sink-cost strategy (p. 82). The tie-hand strategy is especially linked to the issue of ex post costs arising from the actions of domestic political audiences; consequently, further formal stylizations add more insights into the credibility problem as a function of domestic audience costs (e.g., Fearon 1994a; Schultz 1998). On the other hand, the sink-cost strategy is costly ex ante for the deterrer. "Leaders might...sinkcosts by taking actions such as mobilizing troops or stationing large number of them abroad that are financially costly ex ante" (Fearon 1997,82). It is the sink-cost strategy, therefore, that fits more closely the classical commitment theory as defined here. As already discussed, its central idea is that events are interdependent rather than discrete. For this reason, costly threats such as troop mobilization or even limited interventions are deemed necessary to project a reputation for strong resolve and prevent domino effects elsewhere. This is exactly what Fearon (1997) identified as a sink-cost strategy as distinct from the one that ties hands through the ex post domestic audience costs. To what degree is a commitment theory empirically valid? It has certainly been influential in politics; a number of strategic thinkers observed that during the cold war, the United States was indeed greatly concerned with its reputation for fulfilling its commitments. According to several historical analyses, this concern was found to be so strong that it was often given primacy regardless of the specific contextual issue involved in the crisis (Betts 1987; George and Smoke 1974). It is interesting that the same studies did not find other major players following a similar logic (e.g., Betts 1987, 13; see also Mercer 1996), not even the Soviet Union (Hopf 1994; Jervis 1985) or an ally such as France (Weinstein 1969). The approach of other major powers to deterrence appeared to be rather context dependent, fitting more the idea of what was inherently at stake in particular situations. The concern with honoring commitments,

7 346 JOURNAL OF CONFLICT RESOLUTION irrespective of the particular context, seemed to play only a secondary role in their behavior. Besides the question of the degree to which states follow costly signaling prescriptions, there is also an empirical question of whether such strategies are effective deterrents. The empirical findings are not generally positive in this respects5 In their comparative case studies, George and Smoke (1974) did not find that communicating signals were visibly affecting potential atta~kers.~ As for the effect of past behavior on the chances for successful deterrence in current or future conflicts, Huth (1988) reported mixed findings. "Weakness in one situation did undermine a state's future credibility....but firmness in one situation did not necessarily enhance a state's credibility in another" (p. 81). Based on these findings, it is plausible to suggest that the question of whether vital interests are at stake needs to be carefully examined first before any costly signals are sent to an adversary. Nevertheless, the concern with international reputational costs appears to guide the U.S. policy in a number of post-cold war crises as well, including the military interventions in the post-persian Gulf Iraq, Somalia, Haiti, and former Yugoslavia. For example, American credibility was listed at the top of the administration's reasons for the invasion of Haiti: "First is the essential reliability of the United States and the international community...we must make it clear that we mean what we say. Our actions in Haiti will send a message far beyond our region-to all who seriously threaten our interests."' The bombing of Iraq in December 1998 was also justified on similar grounds: "The U.N.'s credibility is on the line, and I think U.S. credibility as well."' Even the U.S. show of force in the 1996 Taiwanese crisis was reportedly interpreted as a bluff, designed only to demonstrate its strong resolve to the world (Blechman and Wittes 1999, 24; see also Layne 1997, 108). The Munich analogy and domino theory-policy metaphors for the idea that events are interdependent, consequently requiring the display of strong commitment to act even in peripheral areas to contain the instabilities in central areas-were thus often invoked to justify the post-cold war intervention^.^ The resulting criticism of "unin- 5. A number of studies concern the effects of bargaining strategies on deterrence outcomes, but most of the examined strategies (e.g., tit-for-tat, firm-but-flexible bargaining, etc.) arenot specifically intended to change the adversary's perception about the deterrer's rationality or, alternatively, to deliberately increase the deterrer's costs for not honoring its commitments. Research on these strategies (e.g., Leng 1993) is, therefore, distinct from the commitment theory as defined here. 6.A few game-theoretic analyses also question the logical validity of this argument. Carlson (1993, for instance, found an inverse relationship between cost tolerance and escalatory behavior. As the disparity between the players' - cost tolerances increases, the lower cost-tolerant actor is more likely to escalate the conflict. 7. The quote comes from the address of Anthony Lake, National Security Adviser at the time, delivered to the Council on Foreign Relations on September 12, 1994 (cited in McManus 1994, Al). 8. U.S. Defense Secretary William S. Cohen, cited in Gellman and Graham (1998, A01). 9. As an illustration, in her interview to the New York TimesMagazine,Secretary of State Madeline Albright famously delineated her foreign policy outlook as being driven by the Munich analogy: "My mind-set is Munich; most of my generation's is Vietnam" (Sciolino 1996,67). Since 'the lesson of Munich is to get in" and "that of Vietnam is to stay out," the distinction between "Vietnam group" and "Munich group" is sometimes applied to demarcate two groups of advocates in the Clinton administration (e.g., see Sciolino 1996; Dobbs 1999). The Vietnam group was represented by General Colin Powell, later succeeded by General Henry H. Shelton, and the crux of policy disagreement concerned the idea of interdependent commitments in the postxold war world. As reported by the observers, "led by Gen. Henry H. Shelton,

8 ---- Danilovic/SOURCES OF THREAT CREDIBIWTY 347 hibited interventionism" largely reflected concerns about military overstretch in the areas of dubious interests for the U.S. national security. If these Teas are categorized into the hierarchy of threats, descending from "A" to " C lists in their gravity, it has become "striking," according to the critics, "how the 'C list' has come to dominate today's foreign policy agenda" (Nye 1999,27). Instead of predominant concern with reputation and credibility for its own sake, this literature suggests the need for "grand strategy" and "selective engagements." This alternative view puts forth the notion of national interests at the core of any potential intervention, which corresponds closely to the notion of inherent credibility in deterrence studies. A THEORY OF INHERENT CREDIBILITY More than 40 years ago, Kaufmann (1956,20) recognized that "there must be some relationship between the value of the objective sought and the costs involved in its attainment. A policy of deterrence which does not fulfill this requirement is likely to result only in deterring the deterrer." In the 1960s and 1970s, several "third-wave" deterrence theorists (see J ~N~s 1979) raised doubts about the validity of the theory of costly signaling and commitments. Its suitability was questioned for both nuclear and conventional threats. Maxwell (1968) was one of the first to observe that it was difficult to understand why it would be rational for a state to engage itself in a nuclear war over an issue it did not consider worth a war "simply in order to support the credibility of its commitment to other objectives it did consider worth waging a war. To do so would mean deliberately exchanging a risk of nuclear war, even if a very high one, for a certainty of nuclear war" (p. 12). The most important inference drawn from this criticism is that "if interest is substituted for 'commitment,' the argument for interdependence immediately becomes less plausible" (p. 19). Even Schelling, as a "first-wave" theorist, was aware of the importance of the inherent stakes or interests that the opponents had in deterrence situations. He originally made the distinction between "warnings," intended to convey a deterrer's "true incentive structure," and "threats" conveying the intention of changing it (Schelling 1960, , n.5). Because he focused on threats and not warnings, the credibility of deterrence has been mostly understood in terms of the deterrer's effectiveness in projecting its resolve to change its own "true" or "intrinsic" values placed on the prot6g6, especially when they are inherently weak. This is exactly what Kaufmann called "deterring the deterrer" rather than deterring the potential attacker. Third-wave theorists, in fact, restated this distinction but shifted their focus to the sources of Schelling's warnings. Following Maxwell's (1968) forceful essay about the hazardous aspects of commitment theory, Weinstein (1969) distinguished between "situational" and "nonsituational" concepts of commitments. The former was inherent to the situation, regardless of verbal or nonverbal signals. Weinstein argued that it was precisely the implicit situational commitments that were better predictors of behavior than verbal (nonsituational) commitments, because they were dictated by national interests involved in the given crisis. Jervis (1970) also made a distinction between chairman of the Joint Chiefs of Staff, the commanders challenged in particular the 'domino theory' being pressed in interagency discussions by Secretary of State Madeline K. Albright" (Graham 1999, A01).

9 348 JOURNAL OF CONFLICT RESOLUTION "signals" and "indices," analogous to Schelling's threats and warnings, respectively. Snyder and Diesing (1 977) followed with their distinction between "inherent bargaining power," on one hand, and "actual bargaining power" and "bargaining skill," on the other. The latter were a function of a challenger's perceptions and the deterrer's manipulation of its own "inherent resolve." Yet, the former represents the "inherent" resolve that "sets a certain limit on what the parties can achieve by manipulative tactics" (p. 190). Although Jervis's work on deterrence escapes easy classification into one or another theory, he did make the argument that "indices" made the greatest impact on potential initiators (1970, 18, 65). This argument was stated even more strongly by George and Smoke (1974,559), after their careful examination of a number of deterrence cases in American foreign policy: "The fact of the matter is that the task of achieving credibility is secondary to and dependent upon the more fundamental question regarding the nature and valuation of interests." Nevertheless, most deterrence theorists did not move beyond these basic conceptual distinctions. Consequently, deterrence theory continued to view the credibility of deterrence in terms of effective manipulation (Schelling's threats, Jervis's signals, Weinstein's nonsituational commitments, and Snyder and Diesing's bargaining skills). In a series of quantitative studies of deterrence, Huth and Russett (1984) revived the interest in the issues raised by the third-wave theorists. Their results provided further justification for a better understanding of the "issues at stake," which referred to a less manipulable inherent interest in the issue of dispute. Their findings were not uniform for all indicators of interests at stake, but were nonetheless critical for the further advancement of deterrence theory. In general, they provided empirical validation for the importance of inherent interests, apart from manipulative strategies, to the calculus of deterrence. THE GEOPOLITICS OF COMMITMENTS In extended deterrence, a state's (defender's) willingness to become involved in conflict depends on the strength of its interest in the third party (protkgk), threatened by another power (challenger). Previous empirical deterrence research (e.g., Huth 1988; Huth and Russett 1984) focused primarily on the defender's ties with a particular nation (protege) to measure the issues at stake in extended deterrence. The notion of inherent credibility of extended threats can be reinterpreted, however, to reflect the importance of the ties between a major power and the entire region where the protege is located. Although a particular state may not have great significance for a major power, it may still be important if it is located in the region of critical strategic importance for that major power's interests. Regional aspects of major power rivalry have rarely been subjected to a rigorous theoretical or empirical analysis. Although there is almost no theoretical framework available for their examination, these issues are partly related to the idea of "shatterbelts" (Cohen 1973) and a territory's "relational importance" (Goertz and Diehl 1992), or to the earlier version of "spheres of interest" in historical and geo-

10 Danilovic /SOURCES OF THREAT CREDIBILITY 349 political writings (e.g., Taylor 1954). In the international relations literature, Morgenthau's (1948) and Gilpin's (198 1) works figure prominently for their acknowledgment that spheres of interest policy often plays an important part in major power behavior. Yet, an exact causal relationship between this politics and international conflict has been left underspecified. The notion from the deterrence literature about inherent credibility or, alternatively, the issues at stake, can be tied to traditional scholarship about the regional aspects of major power rivalry. Specifically, a major power's stakes in an entire region of conflict could be considered as a main ingredient of the inherent credibility of its extended threat. Regional stakes can, in turn, be interpreted as a reflection of the tightness of relations between a major power and most states in the region. These relations can take many forms, usually manifested through alliances, colonies of the past, or high volumes of mutual trade, arms transfers, and other military, diplomatic, and economic transactions. The regional salience variable then simply refers to the less manipulable interests or stakes that forge a major power's willingness to carry out its threat. This variable might be considered to pick up where the third-wave deterrence theorists left off in their conceptual discussions of "inherent credibility" (i.e., George and Smoke 1974; Jervis 1970; Snyder and Diesing 1977). It also expands, both conceptually and theoretically, the notion of the "issues at stake," empirically explored in quantitative research on deterrence (Huth 1988; Huth and Russett 1984). The general assumption is that the more salient an area is for a major power, the more likely it is to become involved in a crisis arising in the area. In the context of extended deterrence, it follows that the more salient the area is for a major power, the more likely it is that the power will be willing to defend a third party from another major power. The willingness of the state to carry out its threats when challenged, the core of threat credibility issues, can thus be tied to the regional salience of a crisis for that state. Furthermore, the logic of the argument can be extended to the claim that if an area is highly salient for more than one major power (i.e., it is an area of "conflicting interests," also known as a "grey area"), the likelihood of a more serious conflict between major powers increases. Once again, this argument is not at odds with the intuitive claims made by a number of diplomatic historians and geopolitical writers. It is also consistent with the current literature that criticizes interventions based on reputational reasons and instead emphasizes the need for a "grand strategy" as a guiding principle for the post-cold war policy. It is the strategy that "would be based on the belief that concrete vital interests should determine U.S. commitments (rather than credibility determining commitments and commitments, in turn, determining interests)" (e.g., Layne 1997,115; see also Johnson 1994; Haas 1995; Nye 1999). Note that regional stakes can also help us identify an observable aspect of inherent resolve in extended deterrence. HYPOTHESES We can now turn to the formulation of the principal arguments of each theory into testable hypotheses. As already pointed out, there is almost no rigorous analysis of the

11 350 JOURNAL OF CONFLICT RESOLUTION conditions prompting an attempt at extended-immediate deterrence to occur in the first place. Huth's (1994) examination of this question is a rare exception and provides the starting point for future research. The results of his analysis indicated that a defender was likely to attempt extended deterrence if the protcgc was an ally, geographically proximate, and, surprisingly, dissimilar from the defender in its domestic regime type. The amount of foreign trade between the defender and the protcgc and the protcgc's strategic natural resources did not have any significant effect on the probability of attempted deterrence. This study expands this type of analysis by focusing on the effects of regional stakes on the probability of the defender's attempt at extendedimmediate deterrence. The above discussion of the geopolitical aspects of inherent stakes in deterrence situations suggests several hypotheses about the onset (hypothesis 1) and outcomes (hypotheses 2-3) of extended-immediate deterrence: Hypothesis 1:The stronger the defender's stakes in the region of potential conflict, the more likely it is that this defender will attempt extended-immediate deterrence. Hypothesis 2: Either the defender or the challenger is less likely to acquiesce to the other side's threats if its stakes are strong in the protcgc's region. Hypothesis 3: War is likely to occur if the challenger and the defender have equally strong regional stakes in the protcgt's region. Note that these hypotheses represent a variant of the argument about the importance of inherent credibility or intrinsic interests for deterrence success. The third hypothesis is not grounded as much in deterrence literature but instead draws from the conventional wisdom of traditional historical and geopolitical texts. Nevertheless, it is consistent with the argument stressing the significance of inherent credibility in accounting for deterrence outcomes. This set of hypotheses is also interesting to put in the context of a recent theoretical approach highlighting selection effects on deterrence (e.g., Fearon 1994a, 1994b). The approach to selection effects concurs with the empirical literature on deterrence (e.g., Huth and Russett 1984) that general deterrence is likely to succeed if the balance of interests and forces favors the defender. In this respect, hypothesis 1 accommodates both approaches. However, the argument concerning selection effects reverses the "mainstream" expectations about the effects of these factors on immediate deterrence. If the general deterrence fails, it indicates a strongly resolved challenger, willing to select itself into the conflict despite possible differences in capabilities and observable interests in the protcgc. For this reason, these variables are expected to be related to general deterrence success but immediate deterrence failure (Fearon 1994b, 258). Hypotheses 2 and 3 are clearly anticipating the opposite results. A nation with stronger regional stakes, taken as an observable measure of interests, is also expected to prevail in immediate deterrence crises. Since the empirical test in the present analysis does not incorporate the challenger's ex ante resolve, then, according to the argument on selection effects results, the estimated coefficients are likely to reflect the selection bias. This bias should, in turn, attenuate and may even reverse the expected relationship between regional interests and immediate deterrence outcomes. In other words, if the

12 Danilovic /SOURCES OF THREAT CREDIBILITY 351 selection bias literature is correct, then the present empirical analysis should make it harder to validate hypotheses 2 and 3. As previously discussed, a commitment theory focuses primarily on consistently resolute responses and costly signals as solutions to the problem of threat credibility. Two major hypotheses can be derived specifically from this argument: Hypothesis 4: The defender's resolve in its past deterrence encounters is likely to affect the outcome of its current deterrence encounters. The more resolute the defender's past behavior, the more likely the other side will acquiesce to the defender's current threats. Hypothesis 5: The defender's less costly threats ("cheap talk") are more likely to result in deterrence failure than threats issued as "costly signals." Before these hypotheses are empirically tested, the principal variables of the two competing approaches to the issues of credibility and resolve must first be operationalized. RESEARCH DESIGN DETERRENCE The above hypotheses present alternative explanations for the onset and outcomes of extended deterrence between major powers. As already described, in extended deterrence, one major power (defender) attempts to deter another (challenger) from starting or continuing a conflict involving a third party (protege). Deterrence cases. The cases of general deterrence failure include all instances in which at least one major power (challenger) upsets the status quo by becoming involved in a conflict with a third state (protege). Because general deterrence failure is a necessary but not a sufficient condition for immediate deterrence to occur, cases of immediate deterrence are identified only if the defender demands that the challenger pull out from the conflict with the protcg6." As deterrent threats can be manifested both verbally and behaviorally, the range of these moves includes (1) the threat of force (including the threat to blockade, occupy territory, declare war, or use force), (2) the display of force (alert, mobilization, show of force), or (3) the use of force (blockade, occupation, seizure, limited use of force, or war)." Based on these operational rules and after consulting an extensive number of historical sources, a data set of all cases of major power deterrence was compiled. During the period from 1895 to 1985, there were 105 cases of general deterrence failures between major powers that did not 10. These operational criteria are consistent with the standard definitions in the literature: "Immediate deterrence concerns therelationship between opposing states where at least one side is seriously considering an attack while the other is mounting a threat of retaliation in order to prevent it. General deterrence relates to opponents who maintain armed forces to regulate their relationship even though neither is anywhere near mounting an attack" (Morgan 1983,30) This typology is based on the coding rules of the Militarized Interstate Disputes (MID) data project (see Gochman and Maoz 1984).

13 352 JOURNAL OF CONFLICT RESOLUTION escalate into immediate deterrence crises (i.e., there was no other major power reacting as the third party's defender against the challenger). On the other hand, a total of 44 cases of general deterrence failures escalated into extended-immediate deterrence (E1D).12 Some of the EID cases had more than one challenger, defender, or protcg.5, which resulted in a total of 70 cases of EID dyads. Deterrence outcomes. Once general deterrence between major powers fails-that is, conflict breaks out between at least one major power (challenger) and a third nation (protcg.5)-it can either escalate into immediate deterrence between the challenger and another major power (defender) or not. If there is at least one other major power that would aid the prottgc as its defender, then the analysis differentiates four possible outcomes of immediate deterrence between the defender and the challenger: (1) challenger's acquiescence to the defender's demands to pull out from the conflict against the prottgc (AcqCh), (2) defender's acquiescence to the challenger's perseverance in its conflict with the defender's protcgc (AcqDef), (3) compromise between the challenger and the defender in which each side achieves some goals while yielding on other issues without any use of force (compromise), and (4) war in which both sides use force against each other as a means of resolving the dispute (war). The operational rules for determining deterrence outcomes are, therefore, based on the combination of two factors: the extent to which each side's demands are met and the degree of escalation in the use of force.13 Most studies have considered only two outcomes, either deterrence success or failure, which are differentiated mainly by the use of force. Still, to consider deterrence outcomes with greater precision, the above four outcomes are introduced instead. This expanded view of deterrence outcomes has important substantive implications. It reveals, for instance, that there can be two forms of deterrence failure, one violent (war) and the other peaceful (defender's acquiescence). Clearly, these and other subtleties, such as the possibility of compromise as a midpoint between success and failure, are obscured in a simple dichotomized approach to deterrence outcomes. EXPLANATORY VARIABLES Both approaches to threat credibility consider capabilities as another important requirement for deterrence success. While the measure of capabilities can safely be 12. The data set also expands and refines the set of deterrence cases provided by Huth and Russett (1984). Major powers are identified following Singer and Small (1982). The complete list of all cases of general deterrence failures, whether escalating into extended-immediate deterrence (EID) crises or not, and consulted historical sources is available on request. For the list of EID cases and their comparison with previous deterrence data sets, see Danilovic (2001). 13. For the sake of the parsimony, often required for larger data sets, the operational rules do not include a stalemate. Like compromise, stalemate has often been neglected in empirical analyses, and its inclusion needs to be a part of the agenda for future research. Empirically, however, the inclusion of stalemate would not significantly alter thedataset usedin this analysis. There wereonly a few cases that might be considered stalemates, including the Manchurian Evacuation, Anglo-Russian frictions in Central Asia, and 1939 Nomonhan. Because the coded outcomes of these crises are not uniform (respectively, defender's acquiescence, compromise, and war), it can be safely concluded that stalemate is also not merged with any particular outcome category under the current coding.

14 Danilovic /SOURCES OF THREAT CREDIBILITY 353 assumed to be similar for both models, their different indicators of threat credibility are critical for comparing the relative validity of each argument. Relative capabilities. It is standard in the literature to measure the relative power variable as the ratio of one nation's capability to the sum of both nations' capabilities. Relative Capability = [Defender's Capability/ (Defender's Capability x Challenger's Capability)] x 100. The value ranges from 0 (defender's absolute power inferiority) to 100 (defender's absolute power superiority), with the exact midpoint (50) indicating power parity between the two sides.14 Measuring regional salience. The salience of geopolitical regions (see Appendix for their classification and composition) for each major power can be measured by looking at the tightness of political, military, and economic linkages between a major power and aregion.i5 It is very similar to Huth and Russett's (1984) operationalization of the individual ties between the defender and its protcgc in extended-immediate deterrence. These ties may be forged through colonial possessions, alliances, diplomatic exchange, arms sales, foreign trade, investments, and so on. Some principal modes of linkages among states change and ultimately erode over time, but others emerge and strengthen in importance. During the past 100 years, for instance, colonial ties reached both their peak and rapid collapse. Similarly, through the middle of the 20th century, diplomatic exchanges of permanent missions stood as a symbol of the reputational status granted to different states, but their post-world War I1 explosion eroded their previous significance. On the other hand, ties established through less formal channels (e.g., arms sales, foreign trade, and investment) have become more salient since the middle of the 20th century, if not earlier. Typically, the forging of alliances was and continues to be a principal means for major powers to secure their foothold in various corners of the world. Hence, any measure of the stakes that major powers have in world regions needs to be adapted to the variable nature of the linkages they form with other states. Moreover, as no single type of linkage predominates, this measure must be composite. For these reasons, one needs to use partially different indicators of regional salience for a longer time span as that from 1895 to While alliances are included throughout the entire time span ( ), colonial possessions and diplomatic exchanges are added to the alliance bonds only for the period from 1895 to They are replaced by the level of foreign trade for the period from 1945 and Foreign trade is also 14.Dataon power capabilities were obtained from the Correlates of War(C0W) National Capabilities Data Set. 15. Geographic proximity suggests itself as another apparent indicator, but previous empirical research did not confirm a significant difference between the decisions of major powers regarding long-distance wars compared with neighborhood wars. Bueno de Mesquita (1981,166). for instance, found that major powers tended "to initiate a disproportionate share of long-distance wars. This indeed proves to be true, with 89 percent of such wars being initiated by major powers, compared to only 40percent of neighborhood wars."

15 354 JOURNAL OF CONFLICT RESOLUTION included for the period from 1920 to Because the composite score of regional salience is an additive index, it is constructed as a sum of the components and then averaged out by dividing the sum by the number of included components. As for the individual components specifically, alliance bonds were measured in terms of the regional distribution of alliances in a major power's alliance portfolio." The measurement proceeded in several stages. First, if there was an alliance between that major power and other states, the variable was coded 1 ;all other dyadic relations between that major power and other states without alliance ties were coded 0. Then, the number of alliances between the major power and every state in the region was added into the regional sum (e.g., let us label this figure as Xi for a region i, Xo for a region o, etc., to calculate the alliance portfolio of a major power X). Finally, the obtained regional sums were calculated as a percentage share of all the alliances concluded between that major power and all other states in the world. The sum of the percentage shares of all regions in a major power's alliance portfolio should then yield 100% (i.e., Xi + Xo +... = 100%). Diplomatic exchanges were measured in a similar way. The number of diplomatic missions received by a major power from all independent states in a particular region was first computed into the regional sum, which was then converted into the percentage share of all diplomatic missions received by that major power.'' Since there is no comprehensive data set available for colonial possessions and foreign trade, an original data set had to be created. For colonial possessions, the square miles of colonial territory controlled by a major power were accrued for each region and then converted into a percentage of the total colonial holdings of that major power. The regional score of amajor power's foreign trade was constructed in the same way as the other three components of the regional salience variable. That is, the amount of trade between a major power and a particular region was expressed in terms of its percentage of the total trade between that major power and the rest of the world. The calculation of regional shares of a major power's foreign trade and diplomatic exchange followed the similar additive logic as was used for alliances and diplomatic exchanges." The composite scores were calculated for each region's salience for all major powers from 1895 to As stipulated in hypotheses 2 and 3, the empirical analysis examines the effects of both the challenger's and defender's regional stakes on deterrence outcomes. 16. The data on the number and types of alliances between major powers and other states were collected from the original data set by Small and Singer (1969) and as updated by Oren (1990). 17.The data on permanent diplomatic missions are provided by the University of Michigan data set of the COW project, conducted by Singer and Small. As it was done for alliances, the diplomatic mission variable was first coded as a binary variable, with the value of 1 if there was a diplomatic exchange between a major power and another state; otherwise, it was coded The New Encyclopaedia Britannica (1998) and Townsend (1941) were the primary sources of information for the square miles figures of colonial areas, supplemented by a number of additional sources. The main sources for the figures on foreign trade between major powers and each independent state were the annual statistical reports on world trade issued by the League of Nations, the United Nations, and the International Monetary Fund. Detailed information on all sources consulted for major powers' colonies and their individual trade with other independent states is available on request.

16 Danilovic /SOURCES OF THREAT CREDIBILITY 355 Past behavior.following Huth (1988, 69), the defender's past behavior was analyzed for its most recent past case of attempted extended deterrence. I do not include an alternative measure of the most recent past conflict between the defender and the same challenger because almost one half of the cases would entail Russo-American conflicts during the cold war.19 If the most recent deterrence outcome involving the same defender resulted in compromise or the defender's acquiescence, this variable was coded 0, indicating a less resolute defender in its most recent deterrence encounter. If the past outcome resulted in war or the challenger's acquiescence, the variable was coded 1, suggesting a more resolved defender in its recent behavior. If there was a long period between the defender's most recent and current conflicts, i.e., extending beyond 15 years, such historically distant cases of the defender's past behavior were not included in the analysis. For example, U.S. involvement as a defender against Britain and Germany in the Venezuelan crisis of 1902 was the only case of its involvement in extended-immediate deterrence against another major power before its confrontation against Japan in the Shanghai incident of It is reasonable to assume that Japan was not drawing the "lessons" from U.S. past behavior in the Venezuelan crisis of 30 years ago. Costly signals. A straightforward indicator was selected for measuring the defender's threats as costly signals. If the highest hostility level for the defender's behavior in a crisis stage (i.e., before the potential outbreak of war) was costly in terms of delivering the threat, such as troop mobilization or display of force, the variable was coded 1 ("a costly signal"). If the defender's extended threat was a verbal statement without any costly movement of troops or display of force, the variable was coded 0 ("cheap talk").20 Fearon's (1997) concept of "sink-cost" signals, for instance, fits this measure (see above). As already discussed in the theoretical section, the notion of costly signals, as analyzed here, does not include behavior that is intended to create domestic or other audience costs that would be paid ex post. The present analysis is focused on the assumption that the interdependence of events makes costly behavior necessary for deterrence to work (e.g., Schelling 1960, 1966). The alternative "tie-hand" strategy of ex post domestic audience costs is not premised on this idea and, consequently, is not tested here. MODEL For testing hypothesis 1, which predicts attempts at extended-immediate deterrence from the defender's regional stakes, the dependent variable was coded 1 if general deterrence failure led to the rise of an extended-immediate deterrence crisis, and 0 if it did not. Since the variable is binary in this case, binomial logit (or probit) is an appropriate method for testing the hypothesis. The other four hypotheses predict the 19. To be exact, if the latter method were used, the number of extended-immediate deterrence cases would be 42, and 18 (43%) of those 42 cases would represent deterrence encounters between the Soviet Union and the United States from 1945 to I am grateful to Bruce Bueno de Mesquita for pointing out this measuring possibility for the distinction between "cheap talk" and "costly signals."

17 356 JOURNAL OF CONFLICT RESOLUTION outcome of extended-immediate deterrence, which is a categorical variable that takes one of four discrete values (AcqDef, AcqCh, compromise, war). Multinomial logit (MNL) is the method of choice for such a polychotomous nominal (i.e., unordered discrete) variable (Amemiya 198 1, 1517; see also Greene 1997; Long 1997; Maddala 1983). In multinomial logit analysis, the coefficients indicate the predicted marginal effects of each explanatory variable (xi) on the log-odds ratio between two outcomes in each possible pair of deterrence outcomes.21 The log-odds ratio is computed as follows: This expression means that the MNL coefficients indicate the effects of an explanatory variable (xi) on the log-odds ratio of the probabilities of choosing alternative Jrelative to the baseline alternative (outcome) K. There are several advantages to the MNL model, and two in particular are relevant for this analysis. First, like all other discrete choice estimations, multinomial logit is a nonlinear probability model, based on maximum likelihood (ML) estimation. Unlike linear probability models, most commonly estimated by ordinary least squares (OLS), discrete choice models do not assume that each unit change in the independent variable produces a monotonic change in the dependent variable. For example, the effect of the power variable on deterrence outcomes may be greater if two sides are relatively equal rather than highly unequal. Second, besides the nonlinearity assumption common to all discrete choice models, MNL has the unique property of allowing that the relationship between the independent variables and individual deterrence outcomes does not necessarily have the same functional form. For instance, the use of MNL allows us to see if the explanatory factors have separate effects on each of the four outcomes (e.g., positive effects on the likelihood of one set of outcomes but negative or even curvilinear effects on the probability of other outcomes-on these issues, see also Stam ' 2 1. The only disadvantage of the multinomial logit (MNL) model lies in aproperty known as the independence of irrelevant alternatives (IIA). The IIA problem results from the MNL assumption that the relative probability of choosing between two alternatives is unaffected by the presence of additional alternatives. Since the IIA property does not always have to bias the parameter estimates, the Hausman test is the standard procedure for testing whether this property is problematic for a particular empirical analysis. To test for the IIA restriction of the MNL models in this analysis, the Hausman specification test was run for all models. The results consistently showed that IIA was not violated; consequently, the MNL parameter estimates should not be suspect as biased. In other words, these results indicate that none of the four deterrence outcomes can be reasonably substituted with another outcome because they are all reasonably dissimilar. The results of the Hausman tests are available on request. See Hausman and McFadden (1984) for computational details of the specification tests for the IIA property. 22. Also, MNL does not require any assumptions to be made about the ordered nature of deterrence outcomes. This assumption is not necessarily problematic, but it should be handled with care, especially if sensitivity to different rank orderings is observed in the analysis that also uses different explanatory models. Amemiya (1981). Greene (1997). Long (1997), and Maddala (1983) provide extensive surveys of modeling optionsfor llmited dependent variables. Slnce LIMDEP7.0 and stati6.0are the standard software of choice for estimating discrete choice models, both software packages were used to estimate the models in this study.

18 Danilovic /SOURCES OF THREAT CREDIBILITY 357 EMPIRICAL ANALYSIS INHERENT CREDIBILITY MODEL According to hypothesis 1 of the inherent credibility model, states are expected to "select themselves" into foreign crises, such as extended-immediate deterrence encounters, if they have strong interests in the region of potential conflict. To test this argument, the cases of general deterrence failure that never escalated into extendedimmediate deterrence against another power should be compared with those cases in which this escalation indeed occurred. While the identity of the defender is certain for the latter situation, it can be a methodological challenge to identify a potential defender for the former situation because it does not involve an actual defender. Nevertheless, this issue is not problematic here since this study is only concerned with major powers as defenders or challengers. All major powers provide the universe of potential defenders in situations of general deterrence failure, whether these cases escalate into immediate deterrence or not. Consequently, the regional stakes of all other major powers were compared with those of the actual challenger in all cases of general deterrence failure (i.e., when the challenger was involved in conflict against a smaller nation). Each case was disaggregated into the same number of dyadic cases as there were major powers in the international system, minus the challenger. Table 1 presents the test results concerning the potential of regional salience and relative power for explaining the onset of EID. The dependent variable is dichotomous, with a value of 0 indicating the absence of EID in a major-power dyad and a value of 1 marking the onset of EID in the dyad. Binomial logit was accordingly used for estimation. There are two variations of the inherent credibility model presented in Table 1. Model I compares the impact of the challenger's and defender's regional stakes on the probability of escalation. The correlation between the defender's interests and the probability of extended-immediate deterrence is statistically significant and positive. This finding clearly supports the argument regarding the defender's "self-selection" into immediate deterrence based on its regional stakes (hypothesis 1). The coefficient for the challenger's interests, however, is statistically insignificant, though its sign is negative, suggesting that other major powers are less likely to interfere in conflicts between a major power (challenger) and a third state located in the region of the challenger's vital interests. Nevertheless, this finding is not statistically significant, leading to the conclusion that the challenger's regional stakes are not as relevant as the defender's in accounting for the onset of extended-immediate deterrence. Model I1 includes the impact of both relative power and regional interests on the probability of EID occurring.23 As in Model I, the unit of analysis is the major-power dyad. Table 1 shows that the results are statistically significant. A major power with stronger regional stakes than the challenger is likely to act as the defender of the smaller nation against that challenger. Given the high statistical significance of the 23. The estimation of the relative ratios of the defender's and challenger's regional interests follows the same calculating routine as that used for their relative capabilities.

19 358 JOURNAL OF CONFLICT RESOLUTION TABLE 1 The Impact of Regional Interests and Relative Power on the Onset of Extended-Immediate Deterrence (ED) Variable Coeficient Standard Error Model I Constant *** Challenger's regional interests Defender's regional interests 0.049*** Model chi-square (df) = *** (2) Log-likelihood function = n (dyads) = 894 Model I1 Constant *** Defender's relative regional interests 0.029*** Defender's relative power (COW) 0.020*** Model chi-square (db = *** (2) Log-likelihood function = n (dyads) = 874 (missing capability data for 20 dyads) NOTE: COW = Correlates of War. ***p <.O1 (one-tailed). parameters, this finding firmly points to the self-selection effects of the observable variable of the defender's inherent resolve, ifthis variable is interpreted in terms of the defender's and challenger's regional stakes in the disputed area. Since this analysis examines the relative validity of two schools of thought about deterrence credibility, the impact of capabilities, as the second important requirement for deterrence, is not closely examined in the hypotheses. Still, it should be noted that the results show similar self-selection effects for the power variable. The stronger the potential defender relative to the challenger, the more likely it is to become the actual defender of the smaller nation. The results for the outcomes of extended-immediate deterrence are almost reversed regarding the statistical significance of the challenger's stakes (see Model I in Table 2). Note that only EID dyads now become the appropriate unit of analysis for this test. As previously explained, multinomial logit coefficients indicate the effects of independent variables on the log-odds ratios between an outcome J and the baseline outcome K. In this specific case, the coefficients estimate the effects of the variables on three outcomes (defender's acquiescence, challenger's acquiescence, compromise) relative to the baseline category (i.e., war). As Table 2 shows, the coefficients for both the defender's and challenger's regional interests tend to be statistically significant. The negative direction of correlation reveals that either side with strong interests in the region of conflict is more likely to fight rather than yield to the other's demands or even agree to compromise. These results further confirm the strong explanatory power of regional stakes in accounting for the onset and outcome of extended-immediate deterrence. In the context of the selection bias issue, Tables 1 and 2 strongly indicate that

20 Danilovic /SOURCES OF THREAT CREDIBILJTY 359 TABLE 2 The Inherent Credibility Model of Deterrence (EID) Outcomes, Multinomial Logit Coefficients Defender Challenger Acquiesces Acquiesces Compromise Model Challenger's regional interests *** *** (0.029) (0.029) (0.043) Defender's regional interests * -0.W5** ** (0.033) (0.031 ) (0.037) Constant 1.819* 4.429*** 3.701*** (1.265) (1.181) (1.255) Model 11~ Defender's relative regional interests ** 0.029** (0.017) (0.016) (0.017) Defender's relative power (COW) 0.043** 0.094*** 0.058** (0.031) (0.029) (0.032) Constant *** ** (1.853) (1.813) (1.998) NOTE:E D=extended-immediate deterrence. COW =Correlates of War. Numbers in parentheses are standard errors. a. Waris the baseline outcome. Model chi-square (df)=28.010*** (6). Log-likelihood function = ; n = 70. b. Waris the baseline outcome. Model chi-square (df)= ** (6). Log-likelihood function = ; N= 69 (missing capabilities data for one case). *p <.lo. **p <.05. ***p <.01. (one-tailed tests) regional ties are positively related to both general and immediate deterrence success despite the potential selection bias. That is, the model does not include the challenger's ex ante resolve, but the coefficients for immediate deterrence are still not reversed from those obtained in general deterrence models. Despite some expected attenuation in the estimated coefficients due to selection bias, the findings reported above suggest a favorable balance of forces and interests to have strong deterring effects in both general and immediate deterrence. Model I1 estimates the impact of the relative ratios of the defender's and challenger's regional interests and power capabilities on the probability of En> escalation into war. The findings are significant for most pairs of outcomes. The coefficients' positive signs indicate that a defender with stronger regional interests or power is more likely to prevail over the challenger, prompting the challenger to acquiesce rather than fight. War is also less likely than compromise when there is an asymmetry of regional stakes to the defender's advantage. These results strongly support hypothesis 2, but only partly validate hypothesis 3. Hypothesis 3 implies that we should not expect a defender with stronger stakes either to acquiesce or compromise, whereas the results indicate a statistically significant likelihood for compromise if the alternative choice is war.

21 360 JOURNAL OF CONFLICTRESOLUTION However, under the same conditions of stronger defender's stakes, compromise is not likely to occur when the alternative choice is the challenger's acquiescence, which is, in turn, supportive of hypothesis 3. For the sake of simplicity, Table 2 does not include the odds ratios for the three remaining pairs of outcomes, that is, the probability of the defender's acquiescence relative to the probability of the challenger's acquiescence or compromise. The results indicate that a defender with stronger relative interests or power is less likely to acquiesce than the challenger. The probability of the defender's acquiescence in this situation is also much lower than the likelihood of compromise. A more refined insight into the impact of regional interests is provided by Table 3, which reports predicted probabilities and their marginal change for different values of regional stakes on both sidesz4 Table 3 reveals that for war to occur, the regional interests of both sides must not only be equal but also strong. If the stakes are low for both major powers (e.g., set at their minimum in the example), they are unlikely to fight. Fighting is also unlikely if only one side has high stakes. The probability of war increases substantially only if both sides have equal and strong stakes in the prottgt's region. This pattern clearly confirms one of the key arguments in this study-the geopolitics of commitments often explains why certain deterrence failures lead to war (hypothesis 3). Overall, the results confirm the validity of the argument that a state's decision to join a crisis, in the role of defender, is largely a function of its intrinsic interests (hypothesis 1). There is also convincing evidence that these interests are manifested through the defender's ties with its prottgt's entire region. The variable of regional interests is also a valid predictor of deterrence outcomes, and these findings are consistent with the logic of deterrence theory emphasizing the role of inherent credibility (hypotheses 2 and 3). In sum, the results reported in Tables 1 through 3 confirm the importance of inherent resolve, as identified with regional stakes, for predicting both the onset and outcomes of EID crises. 24.As the MNL coefficients indicate the log-odds ratios, they are useful for indicating the sign and statistical significance of parameter estimates, but they are not always intuitively appealing for estimating the substantive significance of correlations. For assessing the substantive significance of MNL estimates, the probability of having a deterrence outcome Jrather than outcome K (a base category), given a set of values in the explanatory variables (xi), is calculated as follows: Given J + 1 outcomes, J number of equations are estimated to show the effects of the explanatory variables on producing outcome J rather than the baseline outcome K. The probability for choosing the base case is For further details, see Greene (1997) and Long (1997).

22 Danilovic /SOURCES OF THREAT CREDIBILITY 361 TABLE 3 Predicted Probabilities from the Inherent Credibility Model I Regional Interests AcqDef AcqCh Compromise War Challenger Defender P A P A P A P A Symmetric low stakes Minimum Minimum Symmetric high stakes Maximum Maximum Asymmetric stakes Maximum Minimum Minimum Maximum NOTE: Predicted probabilities of deterrence are calculated according to equations (1)and (2). The sum of predicted probabilities for all four deterrence outcomes should yield a total of 100%.The marginal change (A) presents a percentage change in the predicted probabilities from the value above in the table. For example, changing the distribution of regional stakes from symmetric low to symmetric high stakes increases the probability of war by 88.7 percentage points. AcqDef = defender's acquiescence. AcqCh = challenger's acquiescence. TABLE 4 Descriptive Statistics for Past Behavior and Costly Signals DefenderS Past Behavior Resolute? Defender's Signal: Threat Only? Current Outcome No Yes Yes No AcqDef 6 (50.0) 6 (50.0) 9 (75.0) 3 (25.0) Compromise 2 (22.2) 7 (77.8) 7 (58.3) 5 (41.7) AcqCh 6 (26.1) 17 (73.9) 9 (25.7) 26 (74.3) War 4 (50.0) 4 (50.0) 4 (36.4) 7 (63.6) Total 18 (34.6) 34 (65.4) 29 (41.4) 41 (58.6) NOTE: Percentages are in parentheses. AcqDef = defender's acquiescence. AcqCh = challenger's acquiescence. COMMITMENT MODEL Tables 4 and 5 present descriptive statistics and simple cross-tabulations for the key explanatory variables in the "commitment" version of deterrence theory. As reported in Table 4, a defender's past behavior accounts for a large number of cases that result in compromise or the challenger's acquiescence. About three-fourths of conflicts that end in compromise or the challenger's acquiescence involve a defender who previously displayed a high degree of resolve. This finding is also consistent with the results of the multinomial analysis reported in Table 6, which shows that either compromise or the challenger's acquiescence is more likely than war if the defender displayed a strong resolve in its most recent deterrence crisis.

23 362 JOURNAL OF CONFLICTRESOLUTION TABLE 5 A Comparison of the Defender's Past and Current Behavior Nuclear Power? Defender (major power) Defender S Past and Current United United Behavior No Yes France China Russia Kingdom States Total Both resolute (AcqCh (28.1) (47.8) (57.1) (0.0) (14.3) (26.7) (73.3) (38.2) or war) Other (71.9) (52.2) (42.9) (100.0) (85.7) (73.3) (26.7) (61.8) Total (100.0) (100.0) (100.0) (100.0) (100.0) (100.0) (100.0) (100.0) NOTE: Percentages are in parentheses. AcqCh = challenger's acquiescence. Interestingly, though, the cross-tabulated findings reported in Table 4 also indicate that the impact of this variable is indeterminate for war or defender's acquiescence. Table 5 reveals some further interesting patterns. If the deterrence outcome is compared to the outcome of a previous deterrence encounter with the same defender, then the behavioral consistency in the defender's commitment to act resolutely is almost insignificant for the prenuclear age. Among individual major powers, this consistency is apparent only in the case of the United States. Such a trend offers some support for the claim of third-wave deterrence theorists that the preoccupation with interdependent commitments in the temporal sense was almost unique to U.S. policy in the cold war era (George and Smoke 1974; Weinstein 1969; Betts 1987). A defender's behavior as a costly signal is relevant only for predicting the probability of the defender's or the challenger's acquiescence. As indicated in Table 4, about 75% of the EID cases that resulted in the defender's acquiescence included only the defender's verbal threat (cheap talk), whereas almost 75% of the EID cases that ended in the challenger's acquiescence were the cases in which the defender used costly moves (e.g., mobilization, display of force). The results, therefore, provide some support for hypothesis 5. On the other hand, about two-thirds of the wars occurred despite a defender's costly signal. Since this finding is inconsistent with hypothesis 5, it is reasonable to conclude that costly signals are not effective deterrents against resolved challengers. The question of why challengers are resolved to fight, despite the defender's signaling technique, seems to be more validly explained by the inherent credibility model. Note also that Table 6 shows that the impact of this variable is statistically insignificant for the choice between most alternatives. INTRAREGIONALEFFECTS Although the commitment theory of deterrence does not specify the implications of a defender's past behavior in the same region, they are worth exploring as this has been

24 Danilovic /SOURCES OF THREAT CREDIBIWTY 363 TABLE 6 Past Behavior and Costly Signals, Multinomial Logit Coefficients Defender Acquiesces Challenger Acquiesces Compromise Defender's past behavior Defender's costly signal Defender's relative power (COW) Constant NOTE: War is the baseline outcome. Model chi-square (df)= *** (9). Log-likelihood function = ; N=51. Since this analysis encompasses the period from 1895 to 1985, cases of the defender's past behavior occurring prior to 1895 were not included in the data set. Also, as already explained in the body of the text, if there was a period greaterthan 15 years between the defender's previous and current conflicts, the case of its past behavior was not included for lack of reasonable historical proximity. Consequently, the number of observed extended-immediate dyads in this model, which uses the past behavior as an independent variable, dropped from 70 to 51. COW = Correlates of War. *p <.lo. **p <.05. ***p <.01. (one-tailed tests) TABLE 7 The Model of Intraregional Past Behavior, Multinomial Logit Coefficients Defender's intraregional past behavior 3.104** (1.801) Defender's costly signal (1.532) Defender's relative power (COW) 0.100* (0.063) Constant (3.474) Defender Challenger Acquiesces Acquiesces Compromise NOTE: War is the baseline outcome. Model chi-square (df)= *** (9). Log-likelihood function = ; N=33. Note that this model uses thedefender's most recent conflict behavior only within the same region as an independent variable. Unavoidably, this resulted in the number of extended-immediate dyads appropriate to this model being further reduced to 33. COW = Correlates of War. *p <.lo. **p <.05. ***p <.01. (one-tailed tests) only recently suggested in the literature on reputation. Huth (1997,92)suggested that if "reputations do form, they tend to develop on the basis of the past record of interactions between the same defender and potential attacker within the same geographic region." This argument is interesting in the context of the present analysis, which high-

25 364 JOURNAL OF CONFLICT RESOLUTION TABLE 8 Marginal Impact of Past Behavior on the Probability of Deterrence Outcomes Defender Challenger Acquiesces A Acquiesces A Compromise A War A A general signaling model (Table 6) Defender's move as a threat only Defender's resolute past behavior An intraregional signaling model (Table 7)a Defender's intraregional resolute past behavior a. Marginal impact of the defender's move is not calculated for the intraregional version of the signaling model (Table 7) because this variable is not statistically significant. The marginal effect of each explanatory variable is calculated by changing its value from 0to 1 while holding all other variables in the model at their mean (e.g., relative power) or modal value. lights the importance of regional salience but juxtaposes it with the reputational argument. The cited suggestion, however, opens the possibility for a combined effect of these two variables. For exploratory purposes, I ran the test for the cases of the defender's most recent deterrence encounters that occurred in the same regional area as its current conflicts. If the cases used to examine the defender's most recent deterrence encounters are selected from the same regional areas as its current conflicts, then the coefficients for the defender's past behavior indeed become highly statistically significant for most pairs of outcomes. In particular, a challenger was more likely to acquiesce to a defender with a strong past record of honoring its commitments in the same region of conflict. Furthermore, Table 8 indicates that the magnitude of the effect of the defender's past behavior within the same region is substantially stronger than the impact of the defender's past resolve in general. A change in the defender's intraregional past behavior from weak to resolute increases the probability of the challenger's acquiescence in the current crisis by 64.8%, whereas the likelihood of compromise and war is decreased by 31.4% and 27.9%, respectively. These figures substantially improve the magnitude of the effect of past behavior on deterrence outcomes, which produces the effects of less than 10% if not restricted to the behavior within the same region. This finding is certainly worth further exploration because it again validates the significance of regional factors for understanding extended deterrence, but it does not necessarily contradict the reputational argument about the temporal interdependence of events. Still, the finding does question the validity of reputational arguments about the cross-sectional (i.e., cross-regional) interdependence of events, suggesting instead that closer attention be paid to intraregional issues. As for the issue of the defender's use of sink-cost signals, the combined model does not appear to help in any way-the variable is statistically insignificant for all outcomes as alternatives to war.

26 Danilovic /SOURCES OF THREAT CREDIBILITY 365 DISCUSSION AND CONCLUSION It is argued in this study that one of the factors previously ignored in deterrence and conflict studies-regional stakes as a source of inherent credibility of extended threats-is a powerful predictor of the choices that major powers make in their conflicts over other nations. In fact, an even stronger argument is proposed here-that this factor is more powerful than the commonly used factor of threat credibility (i.e., a commitment to stand firm) for understanding the dynamics of extended deterrence. The results of the empirical tests largely validate this claim. Extended-immediate deterrence is likely to be attempted by a power with strong interests in the region of conflict. A deterrer with strong regional stakes is also less likely to acquiesce to another power's demands, while war is more likely to result if both sides have equally strong and competing interests in the prottgt's region. The latter finding is consistent with the conventional wisdom from the traditional geopolitical literature about the precarious role of "grey areas" in major power relations (formore details, see Cohen 1973). The regional interests of major powers are an observable category, and if these interests are indicative of the inherent credibility of threats in extended deterrence, as argued in this study, then the findings about their important role for predicting both the onset and outcomes of extended-immediate deterrence have several important theoretical implications. First, the findings are largely, though not entirely, consistent with previous studies of the effects of individual ties between a defender and its prottgc on deterrence outcomes.25 At the least, the evidence for regional salience in deterrence relations clearly supports the general argument from the quantitative deterrence literature that the issues at stake matter most (Huth and Russett 1984). By the same token, the results are not supportive of an alternative view, developed in the literature on selection effects, that a favorable balance of interests should be associated with general deterrence success, but immediate deterrence failure. Rather, the results indicate the same direction of correlation between regional interests and deterrence outcomes in both stages-general and immediate deterrence alike. Second, the lack of attention granted to less manipulable factors, such as the intrinsic interests on which credibility depends, is unjustified. The presented findings substantiate the importance of intrinsic interests for understanding the dynamics of deterrence, as advocated by the third-wave deterrence theorists (George and Smoke 1974; Jervis 1970; Maxwell 1968; Snyder and Diesing 1977) and also suggested in the later empirical research on "issues at stake" (Huth 1988, 1994; Huth and Russett 1984). This study demonstrates that if these interests or issues are interpreted in terms of the ties between a major power and the prottgt's region, their effects on deterrence attempts and outcomes are quite strong. 25. For instance, Huth and Russett (1984) found that economic ties and arms transfers from a defender to its prottgt tended to have a significant deterrent impact, but the alliance ties between a defender and prottge had surprisingly reverse effects. Since the measure of regional stakes in this study incorporates both alliances and foreign trade, in addition to colonial possessions and diplomatic exchanges for the earlier periods, the analysis of the impact of regional salience does not address the issue of which elements have more deterrent effects than others.

27 366 JOURNAL OF CONFLICTRESOLUTION On the other hand, the findings support only a restricted version of the alternative argument that considers credibility to be a function of an effective signaling of strong resolve. In general, the consistency of resolute responses over time is not confirmed for the behavior of most major powers and certainly not during the prenuclear age. Still, the defender's past behavior is likely to influence the challenger's decision whether to acquiesce or compromise. Furthermore, it is a significant predictor of deterrence outcomes only if the defender's resolve is inferred from its past behavior in the same region. Therefore, the overall findings are mixed in this respect and certainly point to the need for future refinements that would restrict the notion of interdependent commitments to intraregional affairs. As for the deterrer's behavior as a costly signal or cheap talk, the present analysis shows that this variable does not account for the probability of most deterrence outcomes, and certainly not for predicting escalation to war. This result is consistent with the prediction of a formal-theoretic refinement of deterrence analysis, which "demonstrates that in core areas where both players have inherently credible threats, increasing the costs of conflict past a certain point does little to enhance deterrence stability" (Zagare 1990,259). It also lends justification to the policy prescriptions of the "third-wave" deterrence studies, favoring the inductive approach of comparative case studies. "Instead of...emphasizing the critical importance of credibility and signaling to deterrence strategy, theorists would do better to caution that sophisticated opponents will judge credibility on the basis of a more fundamental analysis of the defender's interests" (George and Smoke 1974,560). Despite methodological differences, the quantitative analysis in this study confirms and further refines these conclusions. It is worth remembering that the theoretical focus on reputation and costly signals originated from concern over the irrationality of nuclear threats. If not placed in the right context, this concern can lead to an apolitical approach to deterrence, with the effective signaling of strong resolve itself becoming the interest at stake (see also George and Smoke 1974, 556; Maxwell 1968, 18-19). The focus on signaling techniques has been so influential that it has almost come at the expense of understanding the inherent issues at stake, particularly those related to regional interests. It can potentially obscure, therefore, the important question of what constitutes vital national interests in extended deterrence. Moreover, the preoccupation with a reputation for standing firm, before making a thorough evaluation of what is at stake, might have the opposite effect, inadvertently prompting escalatory behavior. As shown in this analysis, however, it is better for the question of national interests to be clarified first, rather than to have misleading signals sent to the other side. In terms of policy implications for the U.S. post-cold war diplomacy, the analysis indicates that a "micro-management" of particular foreign policy issues needs to be conducted within the framework of a "grand strategy." This kind of strategy would require, among other things, the differentiation of national interests across regions, thus adjusting the cold war policy of "unrestricted globalism" to the new (and old) regional realities.

28 Danilovic /SOURCES OF THREAT CREDIBILITY 367 APPENDIX Geopolitical Regions: Classification and Composition West Africa: Benin, Burkina Faso, Cape Verde, Gambia, Ghana, Guinea, Guinea-Bissau, Ivory Coast, Liberia, Mali, Mauritania, Niger, Nigeria, Senegal, Sierra Leone, Togo Central Afn'ca: Cameroon, Central African Republic, Chad, Congo, Gabon, Equatorial Guinea, S5o Tom6 and Principe, Zaire East Africa: Burundi, Djibouti, Ethiopia, Kenya, Madagascar, Rwanda, Somalia, Tanzania Southern Africa: Angola, Botswana, Lesotho, Malawi, Mozambique, Namibia, Swaziland, Zambia, Zimbabwe, South Africa, Uganda South Asia: Afghanistan, Bangladesh, Bhutan, India, Maldive Islands, Nepal, Pakistan, Sri Lanka, (Tibet), USSR: Kazakhstan, USSR: Kirghizia, USSR: Tadzhikistan Southeast Asia: Brunei, Burma, Cambodia, Indonesia, Laos, Malaysia, Philippines, Singapore, Thailand, Vietnama East Asia: China, Japan, North Korea, South Korea, (Manchuria), Mongolia, Taiwan Oceania: American Samoa, Australia, Fiji, Guam, Marianne, Caroline, Marshall Islands, New Zealand, Papua New Guinea, Solomon Islands, Vanuatu, Western Samoa North Asia: USSR: Northern Republics of Russia North America: Canada, United States Central America and the Caribbean: Antigua and Barbuda, Bahamas, Barbados, Belize, Bermuda, British West Indies, Costa Rica, Cuba, Dominica, Dominican Republic, El Salvador, Grenada, Guadalupe, Guatemala, Haiti, Honduras, Jamaica, Martinique, Mexico, Nicaragua, Panama, Puerto Rico, Saint Vincent and the Grenadines, St. Kitts and Nevis, Trinidad-Tobago, Virgin Islands South America: Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Guyana, Paraguay, Peru, Suriname, Uruguay, Venezuela Western Europe: ~ustria,b Belgium, Denmark, France, (West) Germany: Greece (after 1945), Iceland, Ireland, Italy, Luxembourg, Malta, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom Eastern Europe: Albania, Bulgaria, Czechoslovakia, East Germany, (Estonia), Finland, Greece ( ), Hungary, (Latvia), (Lithuania), Poland, Rumania, USSR: Byelorussia, USSR: Moldova, USSR: Russia, USSR: Ukraine, Yugoslavia Maghreb ( ): Algeria, Libya, Morocco, Tunisia Middle East (since 1945): Algeria, Bahrain, Cyprus, Egypt, Iran, Iraq, Israel, Jordan, Kuwait, Lebanon, Libya, Morocco, Oman, Qatar, Saudi Arabia, Sudan, Syria, Tunisia, Turkey, United Arab Emirates, USSR: Armenia, USSR: Azerbaijan, USSR: Georgia, USSR: Uzbekistan, USSR: Turkmenistan, North Yemen, (South) Yemen Middle East ( ):d Bahrain, Cyprus, Egypt, Iran, Iraq, Israel, Jordan, Kuwait, Lebanon, Oman, Qatar, Saudi Arabia, Sudan, Syria, Turkey, United Arab Emirates, Yemen a. As North Vietnam and South Vietnam during the period from 1954 to b. As Austria-Hungary during 1895 to 1914, it was classified in Eastern Europe. c. West Berlin is included. d. The Maghreb states have been included in the Middle Eastern region since 1945.

29 368 JOURNAL OF CONFLICT RESOLUTION REFERENCES Amemiya, Takeshi Qualitative response models: A survey. Journal of Economic Literature 19: Betts, Richard K Nuclear blackmail and nuclear balance. Washington, DC: Brookings Institution. Blechman, Barry M., and Tamara Cofman Wittes Defining moment: The threat and use of force in American foreign policy. Political Science Quarterly 114: Bueno de Mesquita, Bruce The war trap. New Haven, CT Yale University Press. Carlson, Lisa J A theory of escalation and international conflict. Journal of Conflict Resolution Cohen, Saul B Geography andpolitics in a worlddivided. 2d ed. New York: Oxford University Press. Danilovic, Vesna Conceptual and selection bias issues in deterrence. Journal of Conflict Resolution 45: Dobbs, Michael Pitfalls of pendulum diplomacy. The Washington Post, 16 May, B01. Einstein, Albert Out of my later years. New York: Philosophical Library. Fearon, James D. 1994a. Domestic political audiences and the escalation of international disputes. American Political Science Review b. Signaling versus the balance of power and interests: An empirical test of a crisis bargaining model. Journal of Conflict Resolution 38: Signaling foreign policy interests: Tying hands versus sinking costs. Journal of Conflict Resolution 41 : Freedman, Lawrence The evolution of nuclear strategy. 2d ed. New York: St. Martin's. Gellman, Barton, and Bradley Graham Canier, Marines rushed to Gulf. The Washington Post, 11 November, A01. George, Alexander L., and Richard Smoke The deterrence in American foreign policy: Theory and practice. New York: Columbia University Press. Gilpin, Robert War and change in worldpolitics. New York: Cambridge University Press. Gochman, Charles S., and Zeev Maoz Militarized interstate disputes, : Procedures, patterns, and insights. Journal of Conflict Resolution 28: Goertz, Gary, and Paul F. Diehl Territorial changes and international conflict. London: Routledge Kegan Paul. Graham, Bradley Joint chiefs doubted air strategy. The Washington Post, 5 April, A01. Greene, William H Econometric analysis. 3d ed. Upper Saddle River, NJ: Prentice Hall. Haas, Richard N Paradigm lost. Foreign Afairs 74 (1): Hausman, Jerry, and Daniel McFadden Specification tests for the multinomial logit model. Econo- metric~52 (5): Hopf, Ted Peripheral visions: Deterrence theory and American foreign policy in the Third World, Ann Arbor: University of Michigan Press. Huth, Paul Extendeddeterrence and theprevention ofwar. New Haven, CT: Yale University Press When do states take on extended deterrent commitments? Cases from In Reconstructing realpolitik, edited by Frank W. Wayman and Paul F. Diehl, Ann Arbor: University of Michigan Press.,1997. Reputations and deterrence: A theoretical and empirical assessment. Security Studies 7: Huth, Paul, and BNC~ Russett What makes deterrence work? Cases from World Politics 36: Jervis, Robert The logic of images in international relations. Princeton, NJ: Princeton University Press.,1979. Deterrence theory revisited. World Politics 31 : ,1985. Perceiving and coping with threat. In Psychology and deterrence, edited by Robert Jervis, Richard Ned Lebow, and Janice Gross Stein, Baltimore: Johns Hopkins University Press. Johnson, Robert H Improbable dangers: U.S. conceptions of threat in the cold war and after. New York: St. Martin's.

30 Danilovic/SOURCES OF THREAT CREDIBILITY 369 Kaufmann, William The requirements of deterrence. In Militarypolicy and national security, edited by William Kaufmann, Princeton, NJ: Princeton University Press. Layne, Christopher From preponderance to offshore balancing: U.S. grand strategy in the twenty-first century. International Security 22: Leng, Russell J Reciprocating influence strategies in interstate crisis bargaining. Journal of Conflict Resolution 37:3-41. Long, J. Scott Regression models for categorical and limited dependent variables. Thousand Oaks, CA: Sage. Maddala, G. S Limited-dependent and qualitative variables in econometrics. Cambridge, UK: Cambridge University Press. Maxwell, Stephen Rationality in deterrence. Adelphi Paper No. 50. London: Institute for Strategic Studies. McManus, Doyle Clinton's call to arms based on U.S. credibility. Los Angeles Times, 16 September, Al. Mercer, Jonathan Reputation and international politics. Ithaca, NY: Cornell University Press. Morgan, Patrick M Deterrence: A conceptual analysis. 2d ed. Beverly Hills, CA: Sage. Morgenthau, Hans J Politics amongnations: Thestruggle forpower andpeace. New York: Knopf. The new encyclopaedia Britannica th ed. Chicago: Encyclopaedia Britannica. Nye, Joseph, Jr Redefining the national interest. Foreign Affairs 78 (4): Oren, Ido The war proneness of alliances. Journal of Conflict Resolution 34: Posen, Bany R., and Andrew L. Ross Competing visions for US grand strategy. International Securiry 21 (3): Powell, Robert Nuclear deterrence theory: The search for credibility. New York: Cambridge University Press. Schelling, Thomas C The strategy of conflict. Cambridge, MA: Harvard University Press Arms and influence. New Haven, CT: Yale University Press. Schultz, Kenneth A Domestic opposition and signaling in international crises. American Political Science Review 92: Sciolino, Elaine The foreign policy race: Madeline Albright's audition. New York Times Magazine, 22 September: Singer, J. David, and Melvin Small Resort to arms. Beverly Hills, CA: Sage. Small, Melvin, and J. David Singer Formal alliances, : An extension of the basic data. Journal of Peace Research 3: Snyder, Glenn H., and Paul Diesing Conflict among nations: Bargaining, decision making, andsystem structure in international crises. Princeton, NJ: Princeton University Press. Stam, Allan C Win, lose, or draw: Domesticpolitics and the crucible ofwar. Ann Arbor: University of Michigan Press. Steele, Ronald Walter Lippman and the American century. New York: Vintage. Taylor, A.J.P The struggle for mastery in Europe, Oxford, UK: Clarendon. Townsend, Mary Evelyn European colonial expansion since Chicago: Lippincott. Weinstein, Franklin B The concept of a commitment in international relations. Journal of Conflict Resolution 13: Young, Oran R The intermediaries: Third parties in international crises. Princeton, NJ: Princeton University Press. Zagare, Frank C Rationality and deterrence. World Politics 42: Classical deterrence theory: A critical assessment. International Interactions 21: Zagare, Frank C., and D. Marc Kilgour Perfect deterrence. Cambridge, UK: Cambridge University Press.