Abstract
The systemic offense–defense theory argues that the security dilemma and the risk of war become doubly severe in offense-dominant eras in the state system. However, the theory assumes in support of its main argument that wars are shorter when offense has the advantage. This article empirically tests the expected connection between the systemic offense–defense balance and war duration. A statistical analysis of wars 1817–1992 disconfirms the theory’s expectations. The article then draws different conclusions about the severity of the security dilemma when offense is dominant: both arms racing and the fear of aggression that the security dilemma thrives on should be less severe than offense–defense theorists assume.
Why are some wars longer than others? And how does the expected length of wars influence the security dilemma? The subject of the duration of wars has not received as wide coverage as the study of the causes of war, as scholars perhaps fear that they might confer legitimacy on limited wars (Fox 1970). However, understanding the causes of war duration should not be less important considering the often high costs of especially long military confrontations. Even the risk of war can be related to the expected war duration. In his seminal article “Cooperation under the Security Dilemma,” which Fearon calls “one of the most influential articles on international relations in the past 25 years” (1997, 1) and started an ongoing debate about the offense–defense theory, Jervis (1978) makes far-reaching claims about both war duration and the nature of the security dilemma 1 when offense has the advantage. 2 In brief, he argues that wars are shorter and it is impossible to create security without threatening others when attacking is easier than defending. I believe Jervis is wrong on both accounts.
The main prediction of the offense–defense theory concerns the frequency of war. The general expectation is that more wars are started when offense is perceived to dominate and conquest is thought to be easy. Either aggression for some national interest, such as profit, is the underlying reason for a war that is believed to be quickly won, or the fear of another state taking advantage of the offensive advantage prompts a preemptive strike. Thus, in support of the logic behind this main hypothesis, the systemic offense–defense theorists also present an auxiliary hypothesis that relates to the duration of the armed conflict: wars are assumed to be shorter when attacking is easier than defending. The assumption of the attacker being able to quickly override the defender is crucial to the theory. If wars cannot be expected to be shorter, there are fewer offensive windows of opportunity and the rationale for initiating more wars when offense is dominant would diminish.
Arguments about the expected length of wars are prevalent in the literature. Jervis argues about offense dominance that “[w]hen there are incentives to strike first, a successful attack will usually so weaken the other side that victory will be relatively quick, bloodless, and decisive” (Jervis 1978, 189). Van Evera, in turn, holds that “aggression brings larger rewards at lower cost” when offense has the advantage (1999, 123). In same manner, Glaser and Kaufmann argue that “war will be quick and decisive and therefore profitable” (1998, 48). Similar hypotheses are also suggested by Bloch (1899, xxx–xxxi), Liddell Hart (1932, 72–73), Wright (1965, 673), Gilpin (1981, 61–62), and Quester (1977, 9).
Not only the frequency of wars is expected to depend on the expected length of wars as also arms racing and the fear of aggression that the security dilemma thrives on are exacerbated (Jervis 1978). Jervis’s systemic conception of the offense–defense balance makes a very clear connection: as offense-dominant military technology decreases the expected length of wars both the frequency of wars and the security dilemma in the state system are intensified. Thus, we have two hypotheses to test:
Hypothesis 1: Offense dominance in the state system leads to shorter wars on average.
Hypothesis 2: Offense dominance in the state system makes the security dilemma more severe.
However, there is also another side to the story. Offense dominance has never been absolute or close to absolute to allow an aggressor to swiftly override the defender, irrespective of the balance of military power. The ability to quickly defeat the enemy, even when offense weapons are more efficient, is limited by permanent (weather, climate, and terrain), temporary (technology and training), and institutional (norms and international law) factors (Nilsson 2010). Moreover, if the attacker can quickly seize a piece of territory when offense is dominant, also the defender can use the same offensive weapons systems and tactics to take it back equally easily in a tactical counter offense as it is hard to defend territory. 3 Thus, if the defender has strategic depth where to absorb the initial thrust, the decisive and swift victory may not materialize and the war can be long.
Stam (1999, 144–45) suggests that limited territorial conquest might be appealing and possible even when defense is dominant if there is an element of surprise. If such territorial gains are made, it is easy to defend them. However, if surprise is coupled with a lack of strategic depth, wars may be short and decisive even when defense is dominant. Unless there are misperceptions about the chances of victory that unduly raise the belligerents’ war aims, even a negotiated outcome should be easy to reach as the expected cost of continued warfare outweigh the expected costs of a mutually acceptable peace treaty when defending is easier than attacking. Also commitment issues should be easy to solve when defense is dominant.
In consequence, if there are such countervailing tendencies, systemic offense–defense balance can be expected to have no impact at all on war duration. Furthermore, arms racing and the fear of aggression that the security dilemma thrives on would logically be less severe than Jervis (1978) assumes if wars are not shorter on average when offense is dominant. Thus, we can suspect that the null hypotheses cannot be rejected:
Null Hypothesis 1: Offense dominance in the state system is not associated with war duration.
Null Hypothesis 2: Offense dominance in the state system is not associated with the severity of the security dilemma.
Gortzak, Haftel, and Sweeney (2005) have tested the connection between the offense–defense balance and the frequency of wars. In this article, I will first statistically analyze the claim about the assumed connection between the offense–defense balance and war duration and then return to the results’ implications for the security dilemma.
Offense–defense theorists have made arguments about the nature of the offense–defense balance in recent history. Therefore, we can empirically test whether Jervis (1978) and the others are correct in arguing that offense dominance is associated with shorter war duration than defense dominance. I use hazard analysis to evaluate the effect of offense–defense balance on war duration. Hazard analysis has also been referred to in the literature as survival, transition, duration, failure time, or reliability analysis. Many social science research questions have an interest in the duration of events and lead to the consideration of hazards models. For example, if something such as a war persists, what variables are associated with the risk of it subsequently ending? The strength of hazards models for an analysis of war duration is that they can analyze variables that assume different values across the span of the observed time period (Box-Steffensmeier and Jones 2004, 95).
The variable of interest here is the offense–defense balance. The inherent problem troubling offense–defense theorists has clearly been their varying classification schemes in their search to characterize some periods in the international state system as offense or defense dominant (Levy 1984, 234; Mearsheimer 1983, 25). I therefore use four different codifications of the offense–defense balance since 1815 (Adams 2003–2004; Quester 1977; Van Evera 1998; Jervis 1978) in order to test the offense–defense theorists’ “commonsense” assumption, which suggests that offense dominance makes wars quick.
Offense–defense balance refers to the relative ease of attack (Jervis 1978, 178) or the cost ratio of offense and defense (Glaser and Kaufmann 1998, 46). Most offense–defense theorists argue that the main variable affecting the balance is military technology, tactical mobility being the primary determinant of offense dominance. Also, most critics of the theory have focused on this core version of the theory and whether it is possible to categorize military technology as offensive or defensive. For example, tanks provide the mobile firepower necessary for Blitzkrieg, but at the same time, the defender can use the same weapon to quickly respond to attacks along the defensive line (Mearsheimer 1983, 25-26). According to Levy (1984, 235), the concept of offense–defense balance then becomes too vague to be useful in theoretical analysis.
Lynn-Jones (1995) responds that the difficulty of classifying weapons is not detrimental to the offense–defense theory as it focuses on the relative costs of offensive and defensive strategies. Moreover, some weapons are clearly easier to use for offensive purposes and even tanks that can be used for both offensive and defensive purposes have made Blitzkrieg more efficient (1995, 674–76). Thus, a change in the weapons employed should change the appeal of attacking. Glaser and Kaufmann’s (1998) respond to criticism by adding several more indicators of the offense–defense balance, such as the size of the forces, but this makes the theory dyadic rather than systemic. Moreover, adding more variables can make classification schemes even more ambiguous and contestable (Betts 1999, 187). As Biddle acknowledges, “many doubt whether the balance can be operationalized in any meaningful way” (2001, 742).
The most complete work on war duration thus far is that of Bennett and Stam (1996). Thus, to perform an adequate test of the offense–defense balance’s potential impact on how long wars last, the best way is to use Bennett and Stam’s variables as statistical controls. 4 I use the same hazard analysis technique with time-varying covariates, as these data include variables with annually measured values, and the Weibull specification, which parameterizes the hazard rate as not constant and allows for both positive and negative duration dependence. The model is fitted in accelerated failure time metric.
These data include the same interstate wars started between 1817 and 1992 as in the updated model by Bennett and Stam (2006) and is given in Appendix A. War duration is measured in months. The data generally follow Small and Singer’s (1982, 65–66) procedures, which identify the starting and ending dates of war by a combination of when actual continued fighting began and ended and information about declarations of war and signed armistices. If there is a considerable difference between actual fighting and the legal dates of declarations and treaties, priority is given to when actual battle occurred.
The hazard analysis is made with data on the systemic concept of offense–defense balance that is advocated by most offense–defense theorists. I estimate four separate models with explanatory variables from Adams (2003–2004), Quester (1977), Jervis (1978), and Van Evera (1998), which are given in Appendix B. Adams classifies various periods as offense, defense, or deterrent dominant. For the sake of comparison and following the analysis of the frequency of wars by Gortzak, Haftel, and Sweeney (2005), offense dominance is coded as 1 and defense and deterrence dominance as 2. Adams argues that the lacking defensive lethality and improved mobile artillery made the balance offense dominant from 1800 to 1849. Between 1850 and 1933, improved firearms, defensive fortifications, transportation, and communications gave rise to defense dominance. However, tanks and aircraft made offense again dominant from 1934 to 1945. Nuclear weapons made the balance deterrence dominant after World War II (WWII).
Van Evera uses three categories to characterize the offense–defense balance (military factors) among great powers in Europe. To increase the number of wars in the analysis in accordance with Gortzak, Haftel, and Sweeney (2005), all applicable interstate wars are considered. The offense–defense balance is codified as follows: offense dominance 1, intermediate values 2, and defense dominance 3. Van Evera argues that the disappearance of mass armies made the balance defense dominant from 1815 to 1855. Railroads and small arms, favoring the defense, and the return of mass armies made the balance cut both ways from 1856 to 1871. Van Evera characterizes the period between 1872 and 1918 as defense dominant, but the Blitzkrieg doctrine combining armor and infantry created offense dominance till 1945. Since then, nuclear weapons have created defense dominance.
Quester uses two categories, and some years are indeterminate. Offense dominance is coded as 1, indeterminate offense–defense balance is coded as 2, and defense dominance as 3. He argues that the balance favored the defense from 1815 to 1862 as there was no mass mobilization and railroads and machine guns were introduced. The period between 1919 and 1945 was indeterminate because increased mobility and firepower improved both offense and defense. The US nuclear monopoly from 1946 to 1949 and a brief break from the balance of terror, 1955–60, were exceptions to the defense dominance created by the spread of nuclear weapons after WWII.
Jervis’s offense dominance is coded as 1 and defense dominance as 2. Jervis is not precise about the periodization of the offense–defense balance. However, I use Gortzak, Haftel, and Sweeney’s (2005) interpretation of Jervis’s text as the data. Bismarck’s wars of conquest showed that the balance was offense dominant from 1864 to 1871. The period between 1872 and 1918 was defense dominant, but from 1919 to 1973, the balance favored the offense again because of tanks and tactical airpower. However, since then affordable antitank and antiaircraft weapons have made the balance defense dominant.
Control variables come from Bennett and Stam (1996, 2006, 2009). Bennett and Stam include in their analysis four dummy variables codifying combinations of the strategy of both the attacker (offensive) and the defender (defensive), which they observe historically. Possible strategies are “maneuver” if Blitzkrieg is used, “attrition” if states fight meeting engagements against each other, and “punishment” if civilians are the principal target and guerilla warfare is used. 5 Terrain is a dummy variable, where 0 stands for open terrain and 1 for impassable terrain. To measure the interaction of terrain and strategy, Bennett and Stam multiply a single ordinal scaled strategy variable by terrain. Balance of forces is the ratio of the largest side’s total capability to all the belligerents’ total capability. Correlates of War (COW) capability scores are discounted according to the distance from a state to the battlefield. Total military capabilities use COW national capabilities measures of both sides’ total military personnel in millions. Total population measures the states’ total population as indicated in the same data set. Bennett and Stam also calculate the population ratio of the larger side to that of the smaller side.
The difference in the quality of the military forces is estimated by dividing a state’s military expenditure by the number of military personnel and then creating a ratio of the superior side’s quality to that of the inferior side. Surprise is a measure of strategic surprise during any time of the war. It ranges from 0 (no or symmetrical surprise) to 1 (large and asymmetrical surprise). Issue salience is coded as 0 (salient to neither side), 1 (salient to one side), or 2 (salient to both sides) using Holsti’s (1991) categorization. A measure of the repressiveness of the governments is obtained by summing the repressiveness measures of each side using Polity II data set’s competitiveness of participation variable, which varies from −5 (significant and regular political competition) to −1 (no significant opposition activity permitted). Bennett and Stam also construct a democracy variable by summing the democracy value of each side using Polity II data set’s institutionalized democracy variable, which ranges from 0 (high level of democracy) to 10 (low level of democracy).
Previous disputes are measured with the help of the COW-militarized interstate disputes data set by counting the average number of disputes lasting at least thirty days in the ten years before each war between all pairs of states on the opposing sides. The total number of disputes is then divided by the number of states in each war. The number of states indicates how many states were involved in the war based on the COW interstate war data set. 6
There is no single best measure of fit in hazards models. Still, in accordance with Bennett and Stam (2009), I estimate PRE% (a proportional reduction in error as a proportion of actual war duration) as a measure of how well the different models fit the data. This provides an intuitively appealing measure of model fit. PRE% is calculated by first estimating a constant-only model and summing the absolute prediction error across all wars. I then estimate the complete models with control variables and different measures of offense–defense balance and sum the absolute prediction errors across all wars. PRE is obtained by subtracting a complete model’s prediction error from the constant-only model’s prediction error and dividing the results by the constant-only model’s prediction error. Finally, PRE% is arrived at by dividing the sum of prediction error by the actual war duration. 7
The main result of the analysis in Table 1 indicates that there is no statistically significant association between systemic offense–defense balance and war duration. Positive coefficients indicate longer war duration and negative ones shorter war duration. There are some changes if we compare the full models including the four measures of systemic offense–defense balance to Bennett and Stam’s (2006) analysis. The models including Quester's and Jervis’s measures of the offense–defense balance indicate that strategic surprise is significantly associated with shorter war duration. The impact of terrain also gained statistical significance in both models, whereas issue salience and the number of states lost statistical significance in Quester’s model. In Adams’s model, terrain times strategy lost statistical significance, while the same happened in Van Evera’s model with repression and the number of states.
Hazard Analysis, the Effect of Offense-Defense Balance on War Duration
Note: OADA = offensive attrition, defensive attrition; OADM = offensive attrition, defensive maneuver; OADP = offensive attrition, defensive punishment; OPDA = offensive punishment, defensive attrition. Coefficients reported. Standard errors in parentheses. All significance tests one-tailed. Calculations made with STATA 11.2.
*p < .10. **p < .05. ***p < .01.
As for the three measures of systemic offense–defense balance, only Quester’s shows a modestly statistically significant effect at the .10 level. Yet, the effect is in the opposite direction than what is hypothesized by the offense–defense theorists (Jervis 1978; Van Evera 1999). If we exponentiate the coefficient −.335, we get a hazard ratio of 1.410, which means that the hazard of war termination increases by 41 percent when we move from offense dominance to middle values or from middle values to defense dominance. Thus, Quester’s model seems to indicate that defense-dominant wars are considerably shorter than offense-dominant wars.
While Van Evera's and Adams’s measures of the offense–defense balance were not statistically significant, they point in the same direction as Quester’s measure. Jervis’s measure points in the direction hypothesized by the offense–defense theorists but is far from being statistically significant (p = .41) at an adequate level. If we look at the PRE%, rather than merely PRE, Quester’s model further seems to fit best to the data as its reduction of prediction error relative to the constant-only model, as a proportion of actual war duration, is 72 percent as compared to the 45 percent in Bennett and Stam’s (1996) model. Yet, lacking statistical significance, the results indicate that the first null hypothesis cannot be rejected: taken together, systemic offense dominance does not make wars quicker and systemic offense–defense balance is not associated with war duration. 8 None of the measures reached a significance level of .05. And when the data were tested with a Cox proportional hazards model, not even Quester’s measure of the offense–defense balance reached a significance level of .10.
The fact that we cannot find a statistically significant association between systemic offense–defense balance and war duration, no matter which codification of the balance we use, runs counter to the offense–defense theorists’ “commonsense” assumption of swift and decisive warfare enabled by offensive weapons technology. This possibility is an alternative neglected by offense–defense theory because the assumptions about more frequent warfare and exacerbated security dilemma depend on the expectation of shorter wars in offense-dominant periods.
The results of the hazard analyses call for a theoretical refinement of the offense–defense theory in terms of our understanding of war duration and the security dilemma. First, if offense–defense theorists are correct in arguing that offense dominance is associated with a higher risk of war, many wars must have been started with a misperception of the expected length and costs of war by the decision makers. From a rational choice perspective, initiating a war that one knows will become long and costly is clearly irrational unless the expected benefits of a long war outweigh the costs. Thus, the causal logic of the offense–defense theory should not rely on the argument that wars are shorter when offense is dominant. Instead, perceptions matter more than the actual war duration. Second, if wars are not shorter on average when offense has the advantage, the nature of the security dilemma is different from what Jervis (1978) argues as it depends on the accuracy of the perceptions of the decision makers.
Eight Worlds of the Security Dilemma
The offense–defense theorists argue that shifts in the offense–defense balance offer a powerful explanation for the likelihood of violent conflict and war. In the classical realist tradition, aggression in the face of perceived threats, even distant ones, has been viewed as a natural consequence of states’ rational quest for security in an anarchical state system. For example, Morgenthau argues that preventive war, however “abhorrent in democratic public opinion, is in fact a natural outgrowth of the balance of power” ([1948]1985, 229). While the offense–defense theorists perceive more peaceful interstate relations when defense has the advantage, they also assume that offense dominance exacerbates the security dilemma and increases the risk of war as the expectation of short military campaigns makes aggression more tempting for both gain and security (Jervis 1978; Van Evera 1999).
Yet, the empirical results of this study are clear: wars are not on average shorter when offense has the advantage. This has implications for Jervis’s claim that it is impossible for states to create security without threatening others when offense is dominant in the state system. As wars are not shorter on average, the security dilemma can be alleviated in offense-dominant eras, compared with Jervis’s analysis, by a weaker motivation for both aggression and arms racing.
The security dilemma depicts the realist predicament of states acting in anarchy, and offense–defense theory has developed it into an aid for rational decision making. On the basis of the assumption that general offense dominance in the state system ought to give rise to shorter wars, Jervis (1978) formulated hypotheses about how the systemic offense–defense balance affects the security dilemma. In his seminal article “Cooperation under the Security Dilemma,” Jervis argued that the offense–defense balance, and whether offensive postures can be differentiated from defensive ones, yield different levels of security dilemma. Table 2 depicts what he designates as the “four worlds of the security dilemma” (1978, 211). 9 The explanation here follows Jervis’s account.
Effect of Beliefs about War Duration on Security Dilemma
In the first world, it is impossible to create security without threatening others, as offense is believed to have the advantage and an offensive posture is not distinguishable from a defensive one. Wars are expected to be short and easy to win for the aggressor. Thus, arms races are likely and attacking is the best defense. Even status quo states behave like aggressors and the security dilemma is doubly dangerous. In the second world, the inability to distinguish between defensive and offensive postures creates a security dilemma but, since defense is perceived to have the advantage and wars are expected to be long and hard to win, it is not extreme. In the third world, there are security problems. Because offense is believed to have the advantage, wars are expected to be short and aggression is possible, but war preparations are easy to detect as an offensive posture can be distinguished from a defensive one. Finally, in the fourth world, the security dilemma can be escaped altogether since defense is believed to have the advantage, wars are expected to be long and it is possible to distinguish an offensive posture from a defensive one.
Yet, this representation of how perceptions of or beliefs about the offense–defense balance and ability to distinguish postures influence the security dilemma can be called the worlds of misperception because Jervis’s analysis of the security dilemma depends on underestimated war duration when offense has the advantage. Based on the empirical results of this study indicating that offense–defense balance has no effect on war duration, it is possible to make a different analysis of the security dilemma. We can analytically compare Jervis’s four worlds, which rely on empirically misperceived war duration, with what happens if more accurate beliefs about war duration inform decision making.
The arrows in Table 2 represent the changing security dilemma when decision makers become more aware that the offense–defense balance is not associated with war duration. All four worlds become different, and thus, we have possibly eight unique scenarios or worlds of security dilemma depending on the beliefs guiding decision making. In consequence, we can change Jervis’s account of the realist security dilemma when decision makers perceive the risks associated with warfare when offense is dominant.
Offense–defense theorists argue that offense-dominant wars are started with the expectation of a swift victory (Jervis 1978; Van Evera 1999). However, as offense dominance does not decrease the average length of wars, the risk of a longer and therefore also often a more costly war increases. The incentive to start a war decreases also because long war duration is associated with an increased risk of the initiator losing (Slantchev 2004). The risk of ending up in a longer and therefore more costly war than expected by Jervis (1978) and Van Evera (1999), which also increases the risk of the initiator losing, should diminish states’ expected utility of starting a war. This should logically also make the security dilemma less severe and make it easier to create security without threatening others in offense-dominant eras than Jervis’s (1978) assumes. Moreover, the security dilemma should be more severe when defense is dominant.
As the expected utility of starting a war is not greater when offense has the advantage, the security dilemma in Jervis’s first world would not be doubly dangerous, as Jervis expects, but alleviated compared with Jervis’s account. With the expectation of longer wars, the strategy of creating security with expansion (and even wars for gain) would be seen in many cases as not more likely to succeed when compared with defense-dominant eras. Offense-dominant weapons would not be perceived as equally threatening if it is realized that offense dominance is not associated with shorter war duration. In consequence, with correct beliefs, the spiral of action and counteraction causing arms racing, which the realist security dilemma thrives on, would be weakened and less likely to lead to war than Jervis assumes.
With correct beliefs, arms do not lack utility. Si vis pacem, para bellum (if you wish for peace, prepare for war), a central realist dictum, would remain as compelling as before because the deterrent mechanism of potentially long and costly wars depends on the prospect of the defender being able to convince the potential aggressor of its holding-on power, which can only be achieved by adequate military capability. And yet, the expected utility of starting a war would often be diminished in contrast to Jervis’s account because of the uncertainty of knowing whether the offense-dominant weapons systems would actually guarantee a swift victory or a victory at all. There is no assurance that the attacker will, with only a relatively small capability advantage, be able to win the war, or that the attacker will, with much more troops, do so more quickly and at a smaller cost than usual.
In the third world, there is possibly no security dilemma because signaling of intent is possible and offensive postures are easily detected. Yet, there are security problems as “aggression is possible, and perhaps easy” (Jervis 1978, 213). However, with correct beliefs about war duration, the general risk of war for the sake of security or profit is diminished compared with Jervis’s account. Aggression is less tempting because long wars are often costly and long wars increase the risk of the aggressor losing.
The empirical results suggest that Jervis has not only exaggerated the severity of the security dilemma when offense is dominant but also understated its severity in defense-dominant eras. In the second and fourth worlds, perceptions of defense dominance would not discourage aggression as effectively as Jervis argues. At the minimum, awareness that wars are not longer on average when defense is dominant should make the security dilemma more severe compared with Jervis’s account.
In the end, the severity of the security dilemma depends on the correctness of the decision makers’ beliefs and can be characterized as a continuum. If decision makers correctly believe that the offense–defense balance has no effect on war duration, the first and the second worlds will be similar, and the same goes for the third and fourth worlds. Thus, the impact of the offense–defense balance on the security dilemma can nearly disappear. Only the ability to distinguish an offensive posture from a defensive one, which can depend on the nature of the weapons systems, can then be argued to have an impact on the severity of the security dilemma.
Force Ratios and First-Move Advantage when Offense Is Dominant
Offense–defense theorists see the presence of offense-dominant weapons systems as extremely dangerous for international security (Jervis 1978; Van Evera 1999). However, with correct beliefs about war duration, the risk of war should not be as acute as claimed. The alleviated security dilemma, when offense is dominant, compared to a situation where decision makers believe that fighting a war will not take a long time, can be illustrated also with how changes in the required force ratios to swiftly win a war change the impetus for arms racing and how the diminished likelihood of leaders seeing a first-move advantage lowers the risk of arms racing developing into open war. The offense–defense theory expects that the required ratio of forces in the attacker’s favor, if the attacker is to swiftly win the war, changes when offense has the advantage. It is assumed that the attacker using offense-dominant technology and tactics does not need to have a massive overall advantage in numbers in order to be able to push through the enemy lines and swiftly win the war (Lieber 2000, 74). Not only is it possible to win with a smaller advantage in troop numbers: When offense has the advantage, it is impossible for states of equal size to enjoy high level of security simultaneously; arms races will be intense because when one country adds forces its adversary will have to make a larger addition to restore its ability to defend. (Glaser and Kaufmann 1998, 47-48)
The logical result, assuming rational decision making, is a need to quickly respond to changes in the military balance of power. And since more defensive weapons are required to counter the offensive ones, compared with defense dominance, the response cannot be mild. This leads to arms racing. Arms races fuel the security dilemma, which, according to Jervis, “can not only create conflicts and tension but also provide the dynamics triggering war” (1976, 67).
Yet, if offense dominance does not make it easier to bring a war to a quick end, it suggests that there are no dramatic changes in the required power ratios to swiftly win the war. Thus, a potential defender does not need to fear that a small increase in another state’s military capabilities will increase the attacker’s chances of quickly running over the defender more when offense is dominant than in defense-dominant eras. The result is that the impetus to swiftly respond in kind to other states arming when offense has the advantage is not as great as Jervis (1978) assumes. In contrast to a situation where decision makers believe that fighting a war will not take a long time, this alleviated fear reduces the perceived need of immediate and across-the-board reciprocation that causes arms racing.
While the fear that the security dilemma thrives on is being alleviated in offense-dominant periods, the security dilemma will not totally disappear, however, since we cannot expect correct beliefs always to be prevalent. Neither can we assume that the expected utility of starting a war is always less than the expected utility of staying at peace as the causes of war can be many and are not always related to the expected probability of winning or the costs brought about by the length of the war. For example, Lieber (2007) suggests that the German leaders did not initiate World War I (WWI) out of fear that their enemies would move first, and they were well aware that the coming war would almost certainly be long and bloody. Similarly, great changes in a potential enemy’s military capabilities will always incite fear, regardless of whether offense dominance is limited or close to absolute.
Nevertheless, not only the risk of arms racing becomes smaller when the general utility of quickly and massively responding to changes in the balance of power diminishes. The risk of arms racing spiraling into open war also becomes smaller when “the advantages of striking first” (Jervis 1976, 67) start fading away. Perceived first-move advantage, which Van Evera (1999) argues is a function of the offense–defense balance and can create a need of preemptive military action, becomes smaller with correct beliefs about war duration when offense is dominant.
Beliefs about the first-move advantage have been argued to lie behind, for example, the escalation of the pre-WWI crisis into open war (Van Evera 1984, 1985; Snyder 1984). Still, if wars are on average not shorter when offense is dominant, there is no reason to expect the aggressor to be more likely to win the entire war during the first battles or to be able to use the initial gains after a surprise attack to swiftly roll over the rest of the enemy territory. Also, as Slantchev (2004) has shown, long war duration is associated with a decrease in the aggressor’s chances of winning. If wars are known to be longer than expected by the offense–defense theorists and the aggressor is known to be less likely to win when wars become long, the perceived first-move advantage also becomes smaller when offense is dominant. Thus, even in the presence of arms racing, the risk of war preparations inexorably leading to open war is diminished compared to a situation where decision makers believe that wars are short on average.
The differences between correct beliefs and the overappreciated first-move advantage that creates a need of preemptive action can be exemplified with a simple game of prisoner’s dilemma. The game illustrates also how differences in beliefs create a varying need to swiftly and resolutely respond in kind to small increases in a potential enemy’s armaments. The prisoner’s dilemma builds on a story of two suspected criminals. The police do not have enough evidence to convict any of them to a long prison term without one squealing on the other. If one squeals (defects), the other gets twenty years in prison and the squealer goes free. If both squeal, both get ten years in prison. If no one squeals but both cooperate, both get only one year in prison.
If false beliefs about the first-move advantage or force ratios guide decision making when offense is dominant, the prisoner’s dilemma looks like the preceding story, which is illustrated in Table 3. The unique equilibrium for this game is a Pareto-suboptimal solution, as rational choice leads both states to defect. Because of the risk of the other state being able to use a small military advantage to swiftly win a war, the possible outcome of not starting to arms race when the other increases its military capabilities is a defeat or even a loss of sovereignty (a long prison sentence of twenty years). Similarly, because of the risk of the other state attacking first, the possible outcome of not seeking to attack first is a defeat or loss of sovereignty.
Prisoner’s Dilemma when False Beliefs Guide Decision Making
Note: Order of preferences: DC > CC > DD > CD.
Therefore, even if cooperation by not arms racing or attacking first would result in the best outcome for both states (the shortest prison term of only one year), no state will risk that a potential enemy will, with small increases in armaments or by attacking first, markedly increase its chances of swiftly winning. Thus, both will choose to arms race and to seek attacking first if arms racing is perceived to increase the risk of an impending attack. This choice of mutual defection is based on the basic realist assumption that state survival is the foremost national interest. Jervis writes that “states that seek security may believe that the best, if not the only, route to that goal is to attack and expand” (1976, 63).
However, if we change the actors’ beliefs about the cost of abstaining from arms racing and seeking to attack first, when the other state does not abstain, alongside with the actors’ expected benefits of managing to attack first, there are changes in the probability of arms racing and seeking to attack first. These changes can be seen in Table 4. If the defender only gets ten years in prison if it cooperates by not responding to small increases in the enemy’s military capabilities, and the enemy gets the same ten years in prison if it defects by attacking with only a small military capability advantage, as both end up in a war with uncertain length and outcome and high costs, the actors’ preferred course of action changes. Similarly, in case of an arms race, if both a state that does not seek to attack first and a state seeking to utilize a first-move advantage get the same ten years of prison, as both end up in a war of uncertain length and outcome and high costs, the actors’ preferred course of action changes. Since the outcome of attacking first or attacking with only small capability advantage is always ten years and the outcome of not attacking first and not responding to small increases in the enemy’s military capabilities is either one or ten years, the best bet will be not to start arms racing and there is no need to seek attacking first for security reasons.
Prisoner’s Dilemma when Correct Beliefs Guide Decision Making
Note: Order of preferences: CC > DC = CD = DD.
Thus, in Table 4, the decision makers do not believe that the outcome of the war and state security are determined by whether the defender responds to small changes in the balance of power or by who makes the first move. There mutual cooperation is enabled by the defender’s alleviated fear of easily losing sovereignty. The attacker’s expected costs of warfare associated with long war duration are also greater than in Table 3, where a swift and easy victory could be expected to materialize. Nevertheless, if a state does not have territory in which to absorb the first attack, the need for making the first move remains.
Correct beliefs about the lower costs of cooperation, when not swiftly and decisively responding to increases in a potential enemy’s armaments or not seeking to attack first, change the order of preferences for both actors. In Table 3, the actors seek to avoid the worst outcome, where they cooperate by not arms racing or attacking while the enemy defects (CD). They therefore settle for the third-best outcome, mutual defection (DD), by both arming and seeking to use the perceived first-move advantage. Yet, in Table 4, mutual defection by both arming and seeking to attack first (DD) would result in the same outcome of ten years in prison, that is, an unclear result of the war and the need to bear the costs of potentially long warfare, as when only one state defects (DC and CD). When there are no dramatic changes in the force ratios in the attacker’s favor or a first-move advantage that could bring the highest reward of quickly subduing the enemy (go free) or the worst possible outcome of quickly being overrun (twenty years in prison), it is more beneficial for both states not to arms race or seek attacking first and get only one year in prison (CC), that is, solve the dispute with negotiations, than become entangled in a war of uncertain outcome and high costs (ten years in prison).
Thus, assuming correct beliefs, the perceived need for arms racing and seeking to attack first is greatly reduced when offense is dominant. Even without the favorable payoffs in Table 4, as soon as the expected payoff for cooperation starts to increase (a shorter prison term), the risk of arms racing and war diminishes. If war duration is realized to be longer and the probability of the initiator then winning is understood to be lower than Jervis (1978) suggests, the outcome of the game turns out to be mutual cooperation, as no state will believe that they will have to respond to small changes in a potential enemy’s military capabilities or move first in order to preserve their sovereignty.
The fact that offense dominance in the state system is not associated with shorter wars has clear consequences for how the actors should perceive the security dilemma. If decision makers have correct beliefs about the effect of offense–defense balance on war duration, the second null hypothesis that offense dominance in the state system is not associated with the severity of the security dilemma cannot be rejected. Classical realism directly prescribes, and structural realism assumes, rational decision making in response to the problems of anarchy. However, considering this analysis of war duration and the security dilemma, the question is whether the use of force has been too readily advocated as a rational course of action when offense has the advantage. It is possible that offense–defense theory, with its overly optimistic assumptions about the expected length of wars when offense is dominant, has contributed to creating false windows of offensive opportunity. There is no doubt that offense–defense theory has played an important role in policy debates. Some analysts have claimed that the Revolution in Military Affairs has tilted the offense–defense balance toward offense. This conclusion can be questioned, but the offense–defense theory can influence defense and military policy and has been argued to do so, for example, in the United States (Biddle 1996, 2001).
It can be argued that bounded rationality and sufficing (good enough orientation) in decision making under stress prevent behavioral changes. However, rational decision making in response to the challenges of the internationals state system, which is a major tenet of realism, would strive to recognize the hazards of warfare when offense has the advantage. If warfare is not more frequent in offense-dominant eras, as has been claimed by the critics of offense–defense theory (Lieber 2000; Gortzak, Haftel, and Sweeney 2005) in response to Van Evera’s (1998, 1999) opposite claim, decision makers are probably more aware of the risks of warfare than Jervis’s (1978) analysis of the security dilemma and offense dominance assumes. In that case, Table 4 rather than Table 3 already guides many states’ decision making by creating rational incentives to act so as not to fuel arms racing or turn arms racing into open war.
The empirical results have also broader implications beyond the risk of war. Structural realisms expects that states are concerned with relative gains and shun cooperation (Waltz 1959, 198) especially when the cost of using force is low and possible use of force is at issue (Powell 1991). Keohane and Martin argue that the importance of relative gains “is conditional on factors such as . . . whether military advantage favors offense or defense” (1995, 44). However, offense dominance does not need to make states more concerned with relative gains or losses. If wars are neither shorter nor easier to win when offense is dominant and the cost of using force is therefore high, the underlying fear of a potential aggressor being able or willing to challenge other states’ territorial sovereignty should not be greater. Thus, if the expected military scenario is not critical, also states’ possible fear of relative losses is mitigated and international cooperation becomes easier to achieve when offense is dominant. In the same vein, defense dominance is not a magical solution that increases the chances of cooperation.
Footnotes
Appendix A
Interstate Wars in Bennett and Stam’s Updated Data Set
| War Name | Start Year | Length in Months |
|---|---|---|
| Franco-Spanish | 1823 | 4 |
| Mexican-American | 1846 | 22 |
| Austro-Sardinian | 1848 | 16 |
| First Schleswig-Holstein | 1848 | 6 |
| Roman Republic | 1849 | 2 |
| La Plata | 1851 | 12 |
| Crimean | 1854 | 28 |
| Anglo-Persian | 1856 | 6 |
| Italian Unification | 1859 | 5 |
| Italo-Roman | 1860 | 10 |
| Italo-Sicilian | 1860 | 2 |
| Franco-Mexican | 1862 | 58 |
| Second Schleswig-Holstein | 1864 | 6 |
| Lopez | 1864 | 63 |
| Spanish-Chilean | 1866 | 6 |
| Seven Weeks | 1866 | 1 |
| Franco-Prussian | 1870 | 10 |
| Russo-Turkish | 1877 | 9 |
| Pacific | 1879 | 58 |
| Central American | 1885 | 4 |
| Serbo-Bulgarian | 1885 | 3 |
| Sino-Japanese | 1894 | 9 |
| Greco-Turkish | 1897 | 5 |
| Spanish-American | 1898 | 4 |
| Boxer Rebellion | 1900 | 15 |
| Russo-Japanese | 1904 | 16 |
| Central American | 1906 | 3 |
| Central American | 1907 | 11 |
| Italo-Turkish | 1911 | 12 |
| First Balkan | 1912 | 7 |
| Second Balkan | 1913 | 2 |
| World War I | 1914 | 52 |
| Hungarian-Allies | 1919 | 5 |
| Greco-Turkish | 1919 | 41 |
| Russo-Polish | 1920 | 6 |
| Sino-Soviet | 1929 | 4 |
| Manchurian | 1931 | 19 |
| Chaco | 1932 | 36 |
| Sino-Japanese | 1937 | 96 |
| Changkufeng | 1938 | 1 |
| German-Czech | 1938 | 0.033 |
| German-Austrian | 1938 | 0.1 |
| Nomohan | 1939 | 4 |
| Russo-Finnish | 1939 | 4 |
| World War II | ||
| German-Polish | 1939 | 1 |
| German Belgian | 1940 | 0.11 |
| German-Netherlands | 1940 | 0.1 |
| German-Danish | 1940 | 0.033 |
| German-Norwegian | 1940 | 2 |
| German-French | 1940 | 1.5 |
| Italo-Greek | 1940 | 2 |
| Pacific | 1941 | 45 |
| Western | 1942 | 60 |
| Eastern | 1941 | 46 |
| German-Yugoslav | 1941 | 0.33 |
| German-Greek | 1941 | 0.67 |
| Franco-Thai | 1940 | 3 |
| First Kashmir | 1947 | 24 |
| Palestine | 1948 | 8 |
| Korean | 1950 | 36 |
| Russo-Hungarian | 1956 | 1 |
| Sinai | 1956 | 1 |
| Sino-Indian | 1962 | 1 |
| Vietnamese I | 1964 | 121 |
| Second Kashmir | 1965 | 5 |
| Six Day | 1967 | 0.2 |
| Israeli-Egyptian | 1970 | 0.25 |
| Football | 1969 | 0.15 |
| Bangladesh | 1971 | 2 |
| Yom Kippur | 1973 | 3 |
| Turko-Cypriot | 1974 | 1 |
| Vietnamese II | 1975 | 3 |
| Ethiopian-Somalian | 1977 | 8 |
| Ugandan-Tanzanian | 1978 | 6 |
| Iran-Iraq | 1980 | 96 |
| Falklands | 1982 | 3 |
| Israeli-Syrian (Lebanon) | 1982 | 2 |
| Sino-Vietnamese | 1985 | 60 |
| Kuwait War | 1990 | 0.1 |
| Gulf War | 1991 | 2.83 |
Appendix B
Classification of the Systemic Offense–Defense Balance, 1816–1992
| Scholar | Period | Offense–Defense Balance |
|---|---|---|
| Adams (2003-2004) | 1800-49 | Offense |
| 1850-1933 | Defense | |
| 1934-45 | Offense | |
| 1946-92 | Defense (Deterrence) | |
| Quester (1977) | 1815-1918 | Defense |
| 1919-45 | Intermediate (Indeterminate) | |
| 1946-49 | Offense | |
| 1950-54 | Defense | |
| 1955-60 | Offense | |
| 1961-77 | Defense | |
| Van Evera (1998) | 1815-55 | Defense |
| 1856-71 | Intermediate | |
| 1872-1918 | Defense | |
| 1919-45 | Offense | |
| 1946-92 | Defense | |
| Jervis (1978) | 1864-71 | Offense |
| 1872-1918 | Defense | |
| 1919-73 | Offense | |
| 1974-78 | Defense |
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
The author received no financial support for the research, authorship, and/or publication of this article.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
