¹ George, Alexander L. and Smoke, Richard, Deterrence in American Foreign Policy (New York: Columbia University Press, 1974), p. 11.Google Scholar

² Theorists as far back as Thucydides dealt with this question, though they did not use the specific term and did not attempt to integrate their ideas into a systematic body of knowledge.

³ There is some debate over whether formal deterrence theory requires that states actually make rational cost-benefit calculations or whether they only act as if they behaved rationally. See de Mesquita, Bruce Bueno, The War Trap (New Haven, Conn.: Yale University Press, 1981), pp. 29–33Google Scholar; Achen, Christopher H. and Snidal, Duncan, ‘Rational Deterrence Theory and Comparative Case Studies’, World PoliticsGoogle Scholar, forthcoming

⁴ Classical deterrence theory includes the following: Kaufmann, William W., ‘The Requirements of Deterrence’ (monograph from the Center of International Studies, Princeton, 1954)Google Scholar; Kissinger, Henry H., Nuclear Weapons and Foreign Policy (New York: Council on Foreign Relations, 1957)Google Scholar; Brodie, Bernard, ‘The Anatomy of Deterrence’, World Politics, 11 (1959)CrossRefGoogle Scholar; Wohlstetter, Albert, ‘The Delicate Balance of Terror’, Foreign Affairs, 37 (1959), 211–34CrossRefGoogle Scholar; Schelling, Thomas C., The Strategy of Conflict (Cambridge: Cambridge University Press, 1960)Google Scholar; Ellsberg, Daniel, The Crude Analysis of Strategic Choices (Rand P-2183, Santa Monica, Calif., 1960)Google Scholar; Snyder, Glenn H., Deterrence and Defense (Princeton, NJ: Princeton University Press, 1961).CrossRefGoogle Scholar

⁵ See George, and Smoke, , Deterrence in American Foreign PolicyGoogle Scholar; Lebow, Richard Ned, Between Peace and War (Baltimore: Johns Hopkins University Press, 1981)Google Scholar and ‘Deterrence Reconsidered: The Challenge of Recent Research’, Survival, 27 (1985), 20–9Google Scholar; Steinbrunner, John, ‘Beyond Rational Deterrence’, World Politics, 28 (1976), 223–45CrossRefGoogle Scholar; Jervis, Robert, ‘Deterrence and Perception’, International Security, 7 (1982/1983), 3–30CrossRefGoogle Scholar; Jervis, Robert, Lebow, Richard Ned and Stein, Janice, Psychology and Deterrence (Baltimore: Johns Hopkins University Press, 1985)Google Scholar; and several essays in Stern, Paul, Axelrod, Robert, Jervis, Robert and Radner, Roy, eds, Perspectives on Deterrence (New York: Oxford University Press, 1989).Google Scholar

⁶ On the use of events data see Kegley, Charles W. Jr et al. , eds, International Events and the Comparative Analysis of Foreign Policy (Columbia: University of South Carolina Press, 1975).Google Scholar

⁷ Although these hypotheses are ‘consistent’ with the rational and unitary actor assumptions of classical deterrence theory, they have not been formally derived from that theory, and in fact are more specific than those suggested by classical deterrence theory. Although classical deterrence theory provides a framework for the analysis of interests, power and costs, it rarely specifies precisely what those interests are. The quantitative empirical literature on deterrence goes beyond classical theory by attempting to evaluate the relative importance of various Realpolitik variables, along with some others, in determining when deterrence works.

⁸ Russett, Bruce M., ‘The Calculus of Deterrence’, Journal of Conflict Resolution, 7 (1963), 97–109.CrossRefGoogle Scholar

⁹ Specific studies will be cited as they are examined. Because many of these studies were not designed specifically and solely to answer the theoretical question of the conditions affecting the success or failure of deterrence, my analysis of the bearing of these studies on that question should not be interpreted as a judgement of their overall merit.

¹⁰ Morgan, Patrick M., Deterrence: A Conceptual Analysis (Beverly Hills, Calif.: Sage, 1977), Chap. 2.Google Scholar

¹¹ Thucydides, , History of the Peloponnesian War, trans. Warner, Rex (New York: Penguin Books, 1954), V/89.Google Scholar

¹² The importance of the balance of interests at stake in a conflict is emphasized by George and Smoke in their concept of asymmetry of motivation and by Jervis in his concept of ‘intrinsic interests’. Both the probability of victory or defeat and the value of those outcomes have been integrated into an expected utility theory of war by Bueno de Mesquita. See George, and Smoke, , Deterrence in American Foreign PolicyGoogle Scholar; Jervis, Robert, ‘Deterrence Theory Revisited’, World Politics, 31 (1979), 314–17CrossRefGoogle Scholar; de Mesquita, Bueno, The War Trap.Google Scholar

¹³ Dyadic-level studies of the ‘parity hypothesis’ and the ‘power preponderance hypothesis’ focus on the relationship between the distribution of capabilities between pairs of states and the frequency of war between them, and are not concerned with the question of whether one state is even considering aggression against another. Thus they are more relevant to the question of general deterrence than immediate deterrence, and for this reason will not be examined here. For studies supporting the preponderance hypothesis see Garnham, David, ‘Dyadic International War, 1816–1965’, Western Political Quarterly, 29 (1976), 231–42CrossRefGoogle Scholar; Organski, A. F. K. and Kugler, Jacek, The War Ledger (Chicago: Chicago University Press, 1980), Chap 1.Google Scholar The bulk of the evidence, however, runs against the preponderance hypothesis. See Ferris, Wayne, The Power Capabilities of Nation States (Lexington, Mass.: Lexington Books, 1973)Google Scholar; Siverson, Randolph M. and Sullivan, Michael P., ‘The Distribution of Power and the Onset of War’, Journal of Conflict Resolution, 27 (1983), 473–94CrossRefGoogle Scholar; and Narroll, Raoul, Bullough, Vern L. and Naroll, Frada, Military Deterrence in History: A Pilot Cross-Historical Survey (Albany: State University of New York Press, 1974).Google Scholar

¹⁴ Zinnes, Dina A., North, Robert C. and Koch, Howard E. Jr, ‘Capability, Threat and the Outbreak of War’ in Rosenau, , ed., International Politics and Foreign Policy (New York: Free Press, 1961), 469–82.Google Scholar

¹⁵ Zinnes, , North, and Koch, , ‘Capability, Threat, and the Outbreak of War’, p. 473.Google Scholar

¹⁶ Zinnes, , North, and Koch, , ‘Capability, Threat, and the Outbreak of War’, p. 4.Google Scholar Note that unless the qualifier ‘sufficiently’ is operationally defined, its inclusion seriously reduces the explanatory and predictive power of the hypothesis by making it nearly non-falsifiable. These findings gain further support from North's subsequent study, which uses events' data on military actions (n = 354) as well as perceptual data from the documents. North, Robert C., ‘Perception and Action in the 1914 Crisis’, Journal of International Affairs, 21 (1967), 103–22.Google Scholar

¹⁷ For reviews of the 1914 studies see Jervis, Robert, ‘The Costs of the Quantitative Study of International Relations’, in Knorr, Klaus and Rosenau, James N., eds, Contending Approaches to International Politics (Princeton, NJ: Princeton University Press, 1969), pp. 177–217Google Scholar; Hilton, Gordon, ‘The 1914 Studies: A Reassessment of the Evidence and Some Further Thoughts’, Peace Research Society (International Papers), 13 (1970), 117–41Google Scholar; Hoole, Francis W. and Zinnes, Dina A., eds, Quantitative International Politics (New York: Praeger, 1976), Part V, pp. 347–459.Google Scholar See also my additional comments in ‘Quantitative Studies of Deterrence Success and Failure’ in Stern, et al. , eds, Perspectives on Deterrence.Google Scholar

¹⁸ Maoz, Zeev, ‘Resolve, Capabilities, and the Outcomes of Interstate Disputes, 1816–1976’, Journal of Conflict Resolution, 27 (1983), 195–229.CrossRefGoogle Scholar The Militarized Interstate Dispute project (MID) was designed to help answer the question of why some interstate disputes escalate to war while others do not. The project and the data set it has generated are summarized in Gochman, Charles S. and Maoz, Zeev, ‘Militarized Interstate Disputes, 1816–1976’, Journal of Conflict Resolution, 28 (1984), 585–616.CrossRefGoogle Scholar This data set includes 960 cases of militarized interstate disputes from 1816 to 1976.

¹⁹ I use ‘capability/threat model’ to refer to Maoz's ‘threat model’. Maoz suggests another capability model, a power transition model, for which the key independent variable is change in relative military capabilities. This is a key variable in the process leading to war, but is not normally included in the theoretical or empirical literature on deterrence and will not be examined here. See Levy, Jack S., ‘Declining Power and the Preventive Motivation for War’, World Politics, 40 (1987), 82–107.CrossRefGoogle Scholar Note that whereas the North project focused on perceptions of capabilities, Maoz focuses on ‘objective’ capabilities.

²⁰ For a discussion of the capability data see Singer, J. David, ‘Reconstructing the Correlates of War Data Set on Material Capabilities of States, 1816–1985’, International Interactions, 14 (1988), 115–32.CrossRefGoogle Scholar

²¹ There may be some questions regarding Maoz's resolve measure, which is based on the relative levels of hostility reached during the dispute and on the extent to which one side undertakes ‘incidents’ to maintain the intiative during the dispute.

²² Maoz, , ‘Resolve’, p. 221Google Scholar; de Mesquita, Bueno, The War Trap.Google Scholar An important exception is that capability ratios do have a significant impact on the outcome of disputes and wars between major powers, though it is not indicated in the article whether the resolve indicators have an even greater impact. Maoz cautions, however, that whether this imbalance of resolve is real or whether it derives from the deceptive manipulation of risks by initiators needs to be analysed.

²³ George, and Smoke, , Deterrence in American Foreign PolicyGoogle Scholar; Lebow, , Between Peace and WarGoogle Scholar; Jervis, , Lebow, and Stein, , Psychology and Deterrence.Google Scholar Blechman and Kaplan also find, in their study of twelve cases since the Second World War in which the United States threatened the use of force against the Soviet Union, that demonstrating resolve is more important for a successful outcome than regional or global military superiority (Blechman, Barry M. and Kaplan, Stephen S., Force without War: US Armed Forces as a Political Instrument (Washington, DC: Brookings Institution, 1978)).Google Scholar

²⁴ Schelling, Thomas C., Arms and Influence (New Haven, Conn.: Yale University Press, 1966)Google Scholar, Chap. 2; Baldwin, David, ‘Power Analysis and World Polities’, World Politics, 31 (1979), 162–94.CrossRefGoogle Scholar

²⁵ Maoz, Zeev, Paths to Conflict: International Dispute Initiation, 1816–1976 (Boulder, Colorado: Westview, 1982)Google Scholar, Appendix II.

²⁶ Wayman, Frank J., Singer, J. David and Goertz, Gary, ‘Capabilities, Military Allocations and Success in Militarized Disputes’, International Studies Quarterly, 27 (1983), 497–515.CrossRefGoogle Scholar Note that Kennedy, Paul makes the same argument in The Rise and Fall of the Great Powers: Economic Change and Military Conflict from 1500–2000 (New York: Random House, 1988).Google Scholar

²⁷ de Mesquita, Bueno, The War Trap, pp. 140–5.Google Scholar The argument that overall military capabilities have only a secondary impact on immediate deterrence would seem to be strengthened by the fact that most empirical studies of the relationship between the dyadic balance of military capabilities and war support the parity hypothesis rather than the power preponderance hypothesis (see fn. 13). The relationship between these studies, and our ability to make inferences, is complicated by the fact that they focus on slightly different sets of cases. The latter set of studies focuses on the direct relationship between pairs of states, whereas the Russet-Huth analyses focus on a more restricted set of cases involving the existence of (1) a prior threat and (2) one which is directed against a third party. Further research is clearly needed on the relationship between these different dimensions of deterrence.

²⁸ Karsten, Peter, Howell, Peter D. and Allen, Artis F., Military Threats: A Systemic Historical Analysis of the Determinants of Success (Westport, Conn.: Greenwood, 1984).Google Scholar I have examined the Karsten, Howell and Allen study in more detail in my ‘Quantitative Studies of Deterrence Success and Failure’, in Stern, et al. , eds, Perspectives on DeterrenceGoogle Scholar, Chap. 6. Note that Karsten, Howell and Allen examine military threats in general rather than just deterrent threats.

²⁹ Karsten, , Howell, and Allen, , Military Threats, pp. 54, 67–70, 73, 83, 110.Google Scholar The authors also conclude that neither the clarity of threats nor attempts to fine-tune threats have much impact on the outcome of the dispute. Whether clear and specific threats are more effective than more ambiguous and general threats is an important question. The Karsten, Howell and Allen findings draw some support from work on Leng's Behavioral Correlates of War Project, which involves events data analyses of interstate crisis behaviour. Leng finds that the more specific the threat the greater the likelihood of a defiant response, for some degree of vagueness facilitates the target's compliance with the threat by avoiding the appearance of being bullied. Leng, Russell J., ‘When Will They Ever Learn? Coercive Bargaining in Recurrent Crises’, Journal of Conflict Resolution, 27 (1983), 379–419CrossRefGoogle Scholar, and ‘Reagan and the Russians: Crisis Bargaining Beliefs and the Historical Record’, American Political Science Review, 78 (1984), 338–55.CrossRefGoogle Scholar Specific threats may also fail because the target is frequently able to ‘design around’ a deterrent threat, particularly if it has multiple options by which to achieve its objectives. See George, and Smoke, , Deterrence in American Foreign Policy, Chap. 17.Google Scholar

³⁰ Russett, , ‘Calculus of Deterrence’Google Scholar; Russett, , ‘Pearl Harbor: Deterrence Theory and Decision Theory’, Journal of Peace Research, 4 (1967), 89–105CrossRefGoogle Scholar; Huth, Paul and Russett, Bruce, ‘What Makes Deterrence Work? Cases from 1900 to 1980’, World Politics, 36 (1984), 496–526CrossRefGoogle Scholar; Huth, and Russett, , ‘After Deterrence Fails: Escalation to War?’ (paper presented to the Conference on the Risk of Accidental Nuclear War, University of British Columbia, Vancouver, 05 1986)Google Scholar; Huth, , Extended Deterrence and the Prevention of War (New Haven, Conn.: Yale University Press, 1989).Google Scholar I will make some references to Huth's book but will not examine it in detail, for at the time of this writing it has not been published.

³¹ Russett, , ‘Calculus of Deterrence’, p. 97.Google Scholar

³² Russett, , ‘Calculus of Deterrence’, p. 98.Google Scholar

³³ Russett, , ‘Calculus of Deterrence’, pp. 100–6.Google Scholar

³⁴ Russett, , ‘Calculus of Deterrence’, pp. 105–7.Google Scholar

³⁵ For further discussion of this problem, and a simple derivation from the expected utility model of the conditions under which a credible threat will be a necessary and sufficient condition for deterrence, see the critique by Fink, Clinton F., ‘More Calculations about Deterrence’, Journal of the Conflict Resolution, 9 (1965), 54–65.CrossRefGoogle Scholar

³⁶ Russett, , ‘Calculus of Deterrence’, p. 98.Google Scholar

³⁷ George, and Smoke, , Deterrence in American Foreign Policy, p. 517.Google Scholar

³⁸ Russett, , ‘Calculus of Deterrence’Google Scholar; Huth, and Russett, , ‘What Makes Deterrence Work’ and ‘After Deterrence Fails’Google Scholar; Huth, , Deterrence and War.Google Scholar

³⁹ Fink, , ‘Calculations’.Google Scholar

⁴⁰ This problem of identifying genuine deterrence success led George and Smoke to restrict their Deterrence study to cases of deterrence failure. This is reasonable given their interest in constructing a typology of deterrence failure, but ultimately the success or failure of deterrence cannot be fully explained without a fully controlled study which includes cases of deterrence success as well as failure.

⁴¹ This problem of identifying genuine deterrence success led George and Smoke to restrict their Deterrence study to cases of deterrence failure. This is reasonable given their interest in constructing a typology of deterrence failure, but ultimately the success or failure of deterrence cannot be fully explained without a fully controlled study which includes cases of deterrence success as well as failure.

⁴² Admittedly, the N of seventeen cases in Russett's ‘Calculus of Deterrence’ is too small to capture all of the advantages of a large-N correlational study, but the number of cases is increased in his later studies.

⁴³ Russett, Bruce M., ‘International Behavior Research Case Studies and Cumulation’, in Haas, Michael and Kariel, Henry S., eds, Approaches to the Study of Political Science (Scranton, Penn.: Chandler, 1970), pp. 425–43.Google Scholar

⁴⁴ Russett, , ‘Pearl Harbor’, pp. 94–6Google Scholar. This argument is reminiscent of Jervis's (1970) distinction between signals and indices.

⁴⁵ Russett, , ‘Pearl Harbor’, pp. 96–9.Google Scholar

⁴⁶ Russett classifies this case as a deterrent failure, but in fact it involved the backfiring of a strategy of compellence as much as the failure of deterrence. The United States attempted not only to deter the Japanese from moving into Southeast Asia but also to compel them to withdraw from China, and used a highly coercive oil embargo to enforce that policy.

⁴⁷ In the nuclear age the option of a direct attack against a major power defender rather than its pawn is unlikely to be very attractive.

⁴⁸ Huth, and Russett, , ‘What Makes Deterrence Work?’, p. 497.Google Scholar

⁴⁹ Singer, , ‘Material Capabilities’.Google Scholar

⁵⁰ Huth, and Russett, , ‘What Makes Deterrence Work?’, p. 514.Google Scholar

⁵¹ Huth and Russelt explain this by arguing that if an adversary threatens a protégé that is formally allied to the defender, that adversary has already decided to stand firm and carry out the threat (‘What Makes Deterrence Work?’, p. 517).Google Scholar

⁵² Huth, and Russett, , ‘What Makes Deterrence Work?’, pp. 516–18Google Scholar. Similarly, Blechman, and Kaplan, , Force without WarGoogle Scholar, find no evidence that US decisions to use force short of war are strongly influenced by aggregate strategic capabilities, and argue that the local balance of conventional power tends to be more important. Organski and Kugler reach similar conclusions from their examination of fourteen cases of deterrence since 1945 (The War Ledger, Chap. 4). They find that nuclear powers have prevailed in only about half of these but that conventional superiority does make a difference. These findings are reinforced in a subsequent study, where Kugler finds that even a nuclear monopoly has brought a favourable outcome only about half the time (Kugler, Jacek, ‘Terror without Deterrence: Reassessing the Role of Nuclear Deterrence’, Journal of Conflict Resolution, 28 (1984), 470–506)CrossRefGoogle Scholar. Weede finds that mutual nuclear deterrence has reduced the risk of war not only between superpowers but also between allies in opposing blocs (Weede, Erich, ‘Extended Deterrence by Superpower Alliance’, Journal of Conflict Resolution, 27 (1983), 231–54).CrossRefGoogle Scholar

⁵³ Huth, and Russett, , ‘What Makes Deterrence Work?’, pp. 520–3.Google Scholar

⁵⁴ Huth, and Russell, , ‘What Makes Deterrence Work?’, pp. 522–3Google Scholar. They hypothesize, in the ellipses in the quoted passage, that these linkages ‘perhaps give the defender some control to prevent adventurism by a protégé’. Alternative explanations would have to be considered, however, including one based on selection bias, a point to which I will return. I should note that the statement quoted in the text is not technically supported by their analysis. It implies an interaction effect between military alliances and tangible (i.e., economic or political) linkages, but no such term is formally incorporated into their model.

⁵⁵ Some cases were deleted because of the ambiguity regarding whether a prior threat of attack actually existed (and thus whether this was actually a case of immediate deterrence). Other cases were deleted because the intervention of the defender occurred after hostilities (between attacker and protégé) had already escalated to the level of large-scale armed conflict, leading to the classification of the case as compellence rather than deterrence. See Huth, , Deterrence and War.Google Scholar

⁵⁶ The immediate balance corresponds to the ‘existing local’ balance in the 1984 study; the short-term balance to the ‘existing overall’ balance; and the long-term balance to the ‘potential overall’ balance.

⁵⁷ Drawing on some of Leng's early work on the Behavioral Correlates of War project, Huth and Russett have coded the diplomatic and military actions of both defender and attacker for each case of attempted deterrence. Diplomatic strategies are classified as bullying, conciliatory or firm-but-fair. Military actions are classified as either policies of tit-for-tat, strength or weakness, depending on whether they matched, exceeded or failed to match the attacker's level of escalation at each stage in the crisis. These many pairs of actions over the course of the crisis are then aggregated into a single measure reflecting the ‘predominant influence strategy’ of a state on both diplomatic and military dimensions. For a discussion of these classification systems see Leng, Russell J. and Wheeler, Hugh G., ‘Influence Strategies, Success and War’, Journal of Conflict Resolution, 23 (1979), 655–84.CrossRefGoogle Scholar

⁵⁸ Huth, and Russett, , ‘After Deterrence Fails’Google Scholar. Some aspects of the initiator's evaluation of the status quo may be incorporated indirectly through bargaining behaviour and reputation. Bullying strategies are hypothesized to be ineffective precisely because of the diplomatic and domestic political costs of retreating from the status quo. But a loss incurred from a retreat from the status quo is analytically distinct from the value of the status quo itself, so I would argue that the value of the status quo has not been incorporated.

This raises two very interesting theoretical questions. One is whether states evaluate prospective gains and losses in terms of the value or utility of the outcome or final asset position, as classical microeconomic utility theory assumes; or whether calculations are based on the magnitude and direction of the change in utility, as prospect theory assumes. (The more appropriate question would be the relative weights given to final asset positions and to changes in assets or utilities.) A related question (the importance of which would be rejected by expected utility theory) is whether the initiator, having made the threat, conceives of a prospective retreat as a retreat from the status quo or a retreat to the status quo. That is, would a retreat be seen as a loss or the absence of a gain. This question of how the decision is framed may be critical. There is substantial evidence in social psychology that more weight is given to losses than to gains and that individuals are risk-acceptant when faced with losses and risk-averse when faced with gains. See Kahneman, Daniel and Tversky, Amos, ‘Prospect Theory: An Analysis of Decision under Risk’, Econometrica, 47 (1979), 273–91.CrossRefGoogle Scholar

⁵⁹ Huth, and Russett, , ‘After Deterrence Fails’Google Scholar. Only the statistically significant (at p = 0.10) probit coefficients are reported, so I must rely on the authors' interpretation of the results.

⁶⁰ The authors acknowledge that the generalizability of this finding is restricted by the limited number of cases involved (eighteen with nuclear defenders, four with overt nuclear threats).

⁶¹ One possible explanation is that the defender's bargaining strategy is a function of the extent of defender-protégé bonds (i.e., the stronger the ties the more coercive the bargaining), with the resulting multicollinearity accounting for the drop in significance of defender-protégé bonds. The correlation between these two variables is not reported in the article, though I have been informed (Huth, private correspondence) that correlations between bargaining behaviour and defender-protégé bonds (economic ties and arms transfers) are very low (0.08 and 0.15, respectively), suggesting that multicollinearity is not a problem.

⁶² This finding of the effectiveness of reciprocal strategies is consistent with a growing body of laboratory studies. See Axelrod, Robert, The Evolution of Cooperation (New York: Basic Books, 1984)Google Scholar and numerous studies by other game theorists and social psychologists. It is also consistent with the findings of Leng and his colleagues on the Behavioral Correlates of War project. Leng, and Wheeler, (‘Influence Strategies’)Google Scholar find, in their data-based analysis of twenty crises since 1900, that reciprocating strategies are the most successful, particularly against an adversary employing a bullying strategy. Bullying strategies tend to generate coercive counter-threats and escalation rather than compliance. See also Leng, , ‘Reagan and the Russians’Google Scholar.

⁶³ In ‘What Makes Deterrence Work?’ Huth and Russett defined the defender's reputation in terms of its behaviour in the previous crisis with any adversary, and found that it has no significant impact on the outcome of a current crisis.

⁶⁴ Leng finds that while reciprocating strategies tend to work best, states tend to adopt more coercive bargaining strategies in successive crises against the same adversary. The diplomatic victor in one dispute tends to utilize the same (successful) influence strategy in the next crisis with the same adversary, unless the adversary adopted a more coercive strategy, in which case (and only under such conditions) the previous winner would also adopt a more coercive strategy. The loser, assuming that its diplomatic defeat was due to the failure to demonstrate sufficient resolve, tends to adopt a more coercive strategy in the next crisis. A diplomatic compromise tends to result in more coercive influence strategies by both parties in the next crisis. Crises ending in war tend to result in more coercive strategies in the next crisis unless a state perceives that the war had been ‘unwanted’ (i.e., one's behaviour was overly coercive, leading the adversary to preempt), in which case a more accommodative strategy is adopted. See Leng, ‘When Will They Ever Learn?’

⁶⁵ This finding raises another question. If ‘stalemate’ is defined as a possible outcome of previous crises, utilized in the statistical analysis, and found to be associated with successful deterrence in a current crisis, that is not made explicit in this study. And if outcomes can be adequately measured trichotomously for use as an independent variable predicting to behaviour and outcomes in the next crisis, this new measure should be utilized as the dependent variable in all crises, replacing the problematic success/failure dichotomy. A technical methodological point is also in order. Statistical inference generally requires the independence of cases. If the outcome of one case is affected by the outcome of the previous case, the assumption of independence is violated, and more sophisticated statistical procedures are normally required.

⁶⁶ On the distinction between deterrence by punishment and deterrence by denial see Snyder, Glenn H., Deterrence and Defense (Princeton, NJ: Princeton University Press, 1961).CrossRefGoogle Scholar

⁶⁷ Huth, and Russett, , ‘After Deterrence Fails’.Google Scholar

⁶⁸ Earlier I noted the limitations of such an assumption.

⁶⁹ Because these studies deal only with those cases in which threats of force have already been made, general deterrence has already failed, but that failure does not necessarily imply the failure of immediate deterrence. Thus the conditions for the success of immediate deterrence are not necessarily the same as those for the success of general deterrence, and vice versa. Overall military capabilities may be more important for general deterrence than for immediate deterrence.

⁷⁰ This methodological problem is complicated further by the multiple dimensions of military capabilities and the dyadic military balance. Selection bias occurs only if the same dimension of military strength (or any other variable) affects both the initiation of the threat (and thus the case selection) and the decision whether or not to defy the defender's deterrent threat. Selection bias would not arise, for example, if the initial threat were made on the basis of one's overall military strength and if the actual attack were made on the basis of the initiator's expectation of fait accompli based on immediately available military forces. There are good reasons to believe, in fact, that the factors influencing behaviour during a crisis may be different from the factors determining behaviour in the pre-crisis period. Domestic political pressures, vested bureaucratic interests and ego investment deriving from the prior decision to initiate a threat, reputational considerations, and increased propensities to take risks when faced with the alternative of losses from backing down, as well as the motivated biases generated by these considerations, all change the cost-benefit calculus for decision makers once a crisis has arisen.

⁷¹ As noted, Huth and Russett do not include this variable in either ‘What Makes Deterrence Work?’ or ‘After Deterrence Fails’.

⁷² Another clear example of likely selection bias is the finding by Karsten, , Howell, and Allen, (Military Threats, p. 83)Google Scholar that threats tend to fail when the threatener sees itself as stronger and succeed when it perceives itself as the weaker party.

⁷³ Huth, and Russell, , ‘After Deterrence Fails’, p. 17Google Scholar. One possible way to circumvent this problem would be to analyse those cases in which alliances are formed after the initial threat but before an attack. This would not fully deal with the problem of selection bias, however, because the alliance formation would probably be the causal result of the protégé's (or defender's) anticipation of an attack (based on the initiator's prior threat). For a more general discussion of the causal relationship between alliances and war, see Levy, Jack S., ‘Alliance Formation and War Behavior: An Analysis of the Great Powers, 1495–1975’, Journal of Conflict Resolution, 25 (1981), 581–614.CrossRefGoogle Scholar

⁷⁴ Karsten, , Howell, and Allen, , Military Threats, pp. 31–2Google Scholar; Huth, and Russett, , ‘After Deterrence Fails’, p. 524.Google Scholar

⁷⁵ See Achen, Christopher H., The Statistical Analysis of Quasi-Experiments (Berkeley: University of California Press, 1986).Google Scholar

⁷⁶ See George, Alexander L., ‘Case Studies and Theory Development’, paper presented to the second annual Symposium on Information Processing in Organizations, Carnegie Mellon University, 1982.Google Scholar

⁷⁷ George, and Smoke, , Deterrence in American Foreign PolicyGoogle Scholar; George, , ‘Crisis Management: The Interaction of Political and Military Considerations’, Survival (1984), 323–34.Google Scholar

⁷⁸ de Mesquita, Bueno, The War TrapGoogle Scholar; Maoz, , ‘Resolve’Google Scholar; Karsten, , Howell, and Allen, , Military ThreatsGoogle Scholar; Russett, , ‘Pearl Harbor’.Google Scholar

⁷⁹ On the importance of domestic incentives for the use of military force externally, see Lebow, , Between Peace and WarGoogle Scholar, Levy, Jack S., ‘Domestic Politics and War’, Journal of Interdisciplinary History, 18 (1988), 653–73CrossRefGoogle Scholar, and ‘The Diversionary Theory of War’, in Midlarsky, Mantis I., ed., Handbook of War Studies (London: Allen & Unwin, 1989)Google Scholar. For other non-quantitative analyses of motivations to challenge deterrence commitments see George, and Smoke, , Deterrence in American Foreign PolicyGoogle Scholar; Jervis, , Lebow, and Stein, , Psychology and Deterrence.Google Scholar

⁸⁰ Leng, , ‘When Will They Ever Learn?’ and ‘Reagan and the Russians’Google Scholar; Huth, and Russett, , ‘What Makes Deterrence Work?’Google Scholar; Huth, , Deterrence and War.Google Scholar

⁸¹ See Axelrod, , The Evolution of Cooperation.Google Scholar

⁸² Huth, and Russett, , ‘What Makes Deterrence Work?’ and ‘After Deterrence Fails’.Google Scholar