Alliance treaty obligations and war intervention

Introduction

Understanding whether military alliances lead to peace or to war is the focus of a large amount of international relations research. For a long time this research provided inconsistent findings. More recently, however, scholars have developed a theoretical framework for understanding the relationship between alliances and war that has been consistent with empirical evidence. This research suggests that alliances influence states’ decisions to initiate militarized disputes because they provide information about other states’ likelihood of intervening in a potential war (Fearon, 1997; Morrow, 1994; Smith, 1995, 1998a). Furthermore, alliances provide different information depending on their content; defense pacts signal that the members will receive assistance if they are attacked and offense pacts signal that the members will receive assistance if they attack (Leeds, 2003). As a result, defensive alliances deter dispute initiation, but offensive alliances encourage dispute initiation.

There is empirical evidence that suggests these different alliance obligations—defense and offense—systematically influence potential challengers’ decisions to initiate militarized disputes (Johnson and Leeds, 2011; Leeds, 2003). However, there is not yet evidence that suggests these different alliance obligations influence states’ war intervention decisions. That is, we do not know if defense pacts increase the likelihood of alliance members joining the target of a war and offense pacts increase the likelihood of members joining the initiator of a war. Testing these hypotheses is an important next step in evaluating the informational perspective of alliances.

In this study I develop a research design that allows me to estimate the effects of defense and offense alliance obligations on states’ war intervention decisions in 95 wars. The results provide strong evidence that defense obligations increase the likelihood that alliance members join the target of a war and that offense obligations increase the likelihood that alliance members join the initiator of a war. These findings provide additional support for the informational theory of alliances. Both of these results hold even after controlling for factors that are expected to influence alliance formation and war such as capabilities, regime similarities and distance as well as various unmeasured factors. The findings of this study corroborate a growing body of research on the role of alliances in conflict processes and contribute to a more comprehensive accounting of the effects of alliances on war intervention.

I proceed as follows. First, I review the informational theory of alliances and describe the implications of the theory that have been tested and identify two untested implications. Second, I describe the research design that I use to test the main hypotheses of this study. Third, I present the results of the empirical analysis. The final section features concluding remarks.

Alliances, information and war

There is a long tradition in international relations research of explaining the relationship between alliances and war. Early attempts to explain this relationship focus on the systemic level and the “balance of power” (e.g. Gulick, 1955; Waltz, 1979). This approach suggests that states use alliances to reduce war in the international system and preserve their independence by aggregating their capabilities to create an even distribution of power. When power is evenly distributed, states are expected to exhibit more restraint, which results in a more stable international system. However, large-N empirical tests of the balance of power approach do not produce consistent findings (Levy, 1981; Ostrom and Hoole, 1978; Small and Singer, 1969).

As a result, more recent attempts to understand the relationship between alliances and war have shifted away from the systemic level and have focused on how alliances may influence a potential challenger’s decision to initiate a militarized dispute against a potential target at the dyad level. The shift in focus has been accompanied by an emphasis on the information that military alliances provide to potential challengers. Alliances supply information to potential challengers about other states’ participation in a potential war. Formal models of alliance formation (Morrow, 1994; Smith, 1995, 1998a) predict that being allied to a war participant increases the probability that a state will intervene because forming an alliance changes the incentives of states deciding whether to intervene in an ongoing war (Morrow, 2000).

Alliances are able to change the incentives of states deciding to intervene through two mechanisms. ¹ First, alliances increase states’ ability to fight together (Morrow, 1994). Many allies engage in prewar military coordination that creates a more unified and effective fighting force. ² For example, some alliances, such as the 1950 Arab Collective Defense Pact, require official contact among the military forces of the members even during peacetime. ³ Other alliances, such as the 1955 Warsaw Pact, feature an integrated command among the allies during both wartime and peacetime. ⁴ Presumably, this prewar coordination improves the probability that the allies win wars they fight. ⁵ Therefore, if states are more likely to join wars as the marginal effect of their intervention on their ally’s chance of winning increases, then one would expect that a state would be more likely to intervene in the presence of an alliance agreement than in the absence of one (Morrow, 2000).

Second, audience costs can be imposed on a government for not honoring an alliance commitment (Fearon, 1994; Smith, 1998b). Both international and domestic actors can punish a government for defecting. International actors may consider a state to be unreliable if it does not honor the alliance agreement, which would make it difficult for the state to form alliances or other cooperative agreements in the future. ⁶ Domestic actors may disapprove of a government that made an alliance commitment and did not follow through because they think the violation damages the state’s honor or reputation. ⁷ Consequently, the audience costs, both international and domestic, imposed on a government for violating an alliance agreement make it more likely that alliance agreements will be honored and that allies will intervene in wars on behalf of their partners (Smith, 1998a).

Building on this perspective, Leeds (2003) points out that different alliances provide different information depending on their promised obligations. ⁸ Defense obligations are promises alliance members make to provide military assistance to members who are attacked. Offense obligations, which are far less common, are promises alliance members make to provide military assistance to members who initiate attacks. ⁹ Thus, potential targets that are members of defense pacts and potential challengers that are members of offense pacts are expected to be more likely to receive assistance in a war.

These expectations regarding war intervention lead directly to expectations regarding dispute initiation. If potential challengers are less likely to initiate disputes against potential targets that they are unable to defeat in war, then potential challengers should be less likely to initiate disputes against potential targets that are members of defense pacts. This is because these potential targets are expected to gain assistance in a war and have a higher probability of defeating the potential challenger. This generates a deterrence effect for members of defense pacts. On the other hand, potential challengers that are members of offense pacts should be more likely to initiate disputes. This is because these potential challengers are expected to gain assistance in a war, making it more likely that they will defeat the potential target in a war. Leeds (2003) tests these hypotheses and finds evidence that defensive alliances to potential targets decrease the probability of dispute initiation, but that offensive alliances to potential challengers increase the probability of dispute initiation. ¹⁰ These findings provide support for the informational perspective of alliances.

Even though there is empirical evidence that suggests that these different alliance obligations influence potential challengers’ decisions to initiate militarized disputes, there is no systematic evidence that these different alliance obligations influence states’ war intervention decisions. Previous research, however, has provided some information about the relationship between alliances and war intervention. For example, we know that alliances are honored most of the time. Leeds et al. (2000) revisit earlier studies of alliance reliability and report that, when the specific obligations of alliances members are taken into account, alliances are honored approximately 75% of the time. While this suggests that allies often intervene in wars when their treaty obligation requires them to, it does not tell us whether they are more likely to intervene in wars than similar states that are not obligated by an alliance treaty to assist a war participant. In other words, this result does not tell us that allies’ war intervention behavior is significantly different than non-allies war intervention behavior, which is a central implication of the informational theory of alliances.

In addition, several studies provide some evidence that suggests allies are more likely to fight together in conflicts than non-allies (Altfeld and Bueno de Mesquita, 1979; Huth, 1998; Siverson and King, 1979; Werner and Lemke, 1997). However, the purpose of these studies is not to test the implications of the informational theory of alliances and as a result they have several limitations. First, these studies do not separate out the different effects of the various alliance obligations. If alliances are expected to provide different information depending on their content, we need a research design that allows different obligations to have different effects on states’ war intervention behavior. That is, we need a research design that distinguishes between the effects of defensive and offensive obligations on states’ war intervention behavior. In previous studies, these obligations are considered interchangeable. Second, previous research does not consider the specific promises made in the alliance agreements and includes alliances that are not applicable to certain wars because their obligations are conditional on conflicts with particular adversaries, in specific locations, or some other limitation on the casus foederis. ¹¹ In other words, previous analyses include many alliances that provide no information to potential war participants and should have no effect on states’ war intervention behavior. Finally, the research designs of prior studies do not distinguish between joining the initiating and target states. If alliance obligations are expected to affect intervening on behalf of the initiator and target of a war differently, we need a research design that allows the effect of the explanatory variables to vary across joining the initiator and target of a war.

In sum, existing studies do not provide a direct test of the implications of the informational theory of alliances for war intervention. Therefore, testing the following two hypotheses is an important next step in evaluating the theory: ¹²

Hypothesis 1: A state will be more likely to intervene on behalf of the initiator of a war if it has an alliance with an offensive obligation to the initiator.

Hypothesis 2: A state will be more likely to intervene on behalf of the target of a war if it has an alliance with a defensive obligation to the target.

In addition to evaluating previously untested hypotheses of the theory regarding alliance obligations and war intervention, testing these hypotheses is also relevant for our understanding of alliance obligations and interstate dispute onset. As discussed above, Leeds (2003) argues that alliance obligations affect militarized dispute onset because of their expected effects on states’ war intervention behavior. The following analysis will evaluate whether the basis for arguing that alliance obligations affect the occurrence of militarized disputes is supported empirically.

Research design

This study evaluates the effect of different alliance treaty obligations on states’ war intervention decisions. Therefore, the standard dyad-year or directed dyad-year research design is not appropriate. The war intervention opportunity is the unit of analysis for this study. A war intervention opportunity is a case in which a state has an opportunity to intervene in an ongoing war on the side of the initiator or target, or to remain neutral. All independent states in the international system at the onset of a war are considered to have a war-intervention opportunity. ¹³

To identify wars I utilize data from the Correlates of War (COW) project. ¹⁴ The COW project defines a war as an armed conflict between two or more members of the interstate system involving at least 1000 battle deaths per year (Sarkees and Wayman, 2010). The data allow me to analyze all states’ war intervention decisions from 1816 to 2007, which includes 95 wars producing 7829 war intervention opportunities. ¹⁵ Given this sample of war intervention opportunities, I code each state’s war intervention decision and a variety of variables, including the potential intervener’s alliance obligations, that are expected to affect a state’s war intervention decision. I describe how I code each variable in detail below. Table 1 provides basic summary statistics for all of the variables. ¹⁶

OPEN IN VIEWER

Table 1.

Summary of variables

	Mean	Standard deviation	Minimum	Maximum
War intervention	0.0314972	0.2325957	0	2
Offensive obligation to the initiator	0.0015263	0.0390407	0	1
Defensive obligation to the target	0.0162342	0.1263839	0	1
Distance from target	8.108645	0.8682152	3.78419	9.421168
Potential contribution to initiator	0.1038443	0.167888	0.0000007	0.8610793
Potential contribution to target	0.1134496	0.1695813	0.0000263	0.9348571
Initiator’s probability of winning	0.5936941	0.3172484	0.0418741	0.9977058
Regime difference with initiator	7.445955	6.212455	0	20
Regime difference with target	7.526433	6.12091	0	20
Observations		7207

Control variables

In addition to including indicators for the alliance obligations of the potential intervener, I also estimate models that include a number of control variables. More specifically, I attempt to control for factors that may influence both alliance formation and war intervention decisions. Including these variables in the model increases confidence that the estimated relationships between alliance obligations and war intervention are not spurious. I describe each of the variables included below. ²⁴

First, I include a variable that measures the potential joiner’s distance from the war. It is more costly for states to mobilize troops as distance increases and states further away will tend to have less interest in the outcome of the war. Therefore, I expect that as the distance of the potential intervener increases it will be less likely to intervene on either side in the war. Additionally, states that are further away from a war are less likely to be allied to the war participants since they will have a harder time fighting alongside them and have less interest in the outcomes of their disputes. I take this into account by including a measure of the natural log of the capital-to-capital distance between the potential intervener and the target. ²⁵ This measure assumes that the fighting is taking place on the target’s territory. I use the natural log because I expect the marginal impact of distance to decrease as the distance becomes larger.

Second, I include two variables that measure the potential intervener’s ability to influence the outcome of the war. Given that war intervention is costly, a state will be less likely to join a state in war if it will make no significant difference to that state’s probability of winning. Importantly, these states will also be less desirable alliance partners. However, when a state can significantly improve another state’s probability of winning, it will be more likely to intervene on that state’s behalf and be more likely to enter into alliances. To take this into account, I include a variable that measures the potential joiner’s contribution to the initiator’s probability of winning the war and a variable that measures the potential joiner’s contribution to the target’s probability of winning the war. Both variables are calculated by subtracting the war participant’s bilateral probability of winning the war from its probability of winning with the potential joiner on its side. More specifically, I measure the potential joiner’s contribution to the initiator using this expression,

( c a p I + c a p J ) ( c a p I + c a p J + c a p T ) − c a p I ( c a p I + c a p T )

and the potential intervener’s contribution to the target with this expression,

( cap T + cap J ) ( cap T + cap J + cap I ) − cap T ( cap T + cap I )

where cap_I is the initiator’s capabilities, cap_T is the target’s capabilities, and cap_J is the potential joiner’s capabilities. I use the composite index of national capabilities (CINC) data of the COW project to measure the actor’s capabilities (Singer et al., 1972). Both variables are always between 0 and 1 where values closer to 1 indicate more of a potential contribution to the outcome of the war.

Third, I also control for the intitiator’s probability of winning the war. As discussed above, some early research on alliances and war intervention argues that states will ally with and join the weaker side to balance against strong states. If this is the case, states should be less likely to join the initiator and more likely to join the target as the initiator’s probability of winning the war increases. The opposite relationship would support the notion that states tend to bandwagon against weaker states. To measure the initiator’s probability of winning I use the ratio of the initiator’s military capabilities to the sum of the initiator’s and target’s military capabilities. I use the target and initiator’s CINC scores to measure their capabilities (Singer et al., 1972). This variable is always between 0 and 1 where values closer to 1 indicate higher probabilities of the initiator defeating the target in war.

Finally, I include two variables that measure the regime difference between the potential intervener and the war participants. Research suggests that states that have similar regime types are more likely to align in international conflicts (Werner and Lemke, 1997). There is also research that demonstrates that states with similar regime types are more likely to ally with each other (e.g. Lai and Reiter, 2000). ²⁶ Therefore, to control for this potentially confounding factor I include one variable that measures the potential intervener’s regime difference with the initiator and another variable that measures the potential intervener’s regime difference with the target. Similar to Werner and Lemke (1997), the variables are created by taking the absolute difference of the potential intervener’s regime score and the initiator’s score and the absolute difference of the potential intervener’s regime score and the target’s score. To measure regime type I use the score on the Polity2 variable from the PolityIV project (Marshall et al., 2014). ²⁷ Higher values on these variables indicate more regime dissimilarity.

Results

The statistical results are reported in Table 2. Model 1 does not include the control variables and model 2 includes the control variables. ²⁹ The coefficient estimates in the first column for both models are the effects of the variables on joining the initiator compared with remaining neutral. The coefficient estimates in the second column for both models are the effects of the variables on joining the target compared with remaining neutral.

OPEN IN VIEWER

Table 2.

Multinomial logit regressions of war intervention

	Model 1		Model 2
	Join initiator	Join target	Join initiator	Join target
Offensive obligation to the initiator	6.59 **	−8.24	5.68 **	−8.46
	(0.80)	(488.83)	(0.89)	(344.89)
Defensive obligation to the target	1.29	3.72 **	0.90	3.30 **
	(0.73)	(0.25)	(0.75)	(0.29)
Distance from target			−1.16 **	−0.76 **
			(0.13)	(0.11)
Potential contribution to initiator			1.80	0.01
			(1.10)	(1.05)
Potential contribution to target			−0.14	2.65 **
			(1.01)	(0.59)
Initiator’s probability of winning			0.95	0.67
			(0.65)	(0.58)
Regime difference with initiator			−0.03	0.04 *
			(0.02)	(0.02)
Regime difference with target			0.10 **	−0.00
			(0.02)	(0.02)
Constant	−5.08 **	−4.84 **	2.29 *	−0.12
	(0.15)	(0.13)	(1.05)	(0.94)
Observations	7829		7207
Log-likelihood	–718.34		–598.72
AIC	1448.67		1233.45

Standard errors in parentheses.

*

p < 0.05, ^** p < 0.01.

With respect to Hypothesis 1, the results indicate that a state will be more likely to intervene on behalf of the initiator of a war if it has an offensive obligation to the initiator. This is indicated by the significant positive coefficients in the first column of both models associated with the variable that codes whether the potential intervener had an offensive obligation to the initiator of the war. To provide substantively meaningful quantities, I present Figure 1. It provides predicted probabilities of a state intervening on behalf of the initiator of a war with 95% confidence intervals. ³⁰ The black density shows the estimated predicted probability of intervening on behalf of the initiator when the potential intervener does not have an offensive obligation to the initiator and and the gray density shows the estimated predicted probability of intervening on behalf of the initiator when the potential intervener has an offensive obligation to the initiator. Figure 1 shows that the predicted probability of a state intervening on behalf of the initiator of a war significantly increases when it has an offensive obligation to the initiator of the war. The mean predicted probability is 0.003 [0.002, 0.005] when a state does not have an offensive obligation to the initiator and it is 0.452 [0.134, 0.783] when a state does have an offensive obligation to the initiator. In other words, the predicted probability of a state intervening on behalf of the initiator of a war is 0.449 greater for a state with an offensive obligation to the initiator than for a state without one. This results in a 150% (=0.449/0.003) increase in the probability of joining the initiator of a war.

OPEN IN VIEWER

Figure 1.

Predicted probabilities of joining the initiator.

For Hypothesis 2, I find evidence that a state will be more likely to intervene on behalf of the target of a war if it has a defensive obligation to the target. This is indicated by the significant positive coefficients in the second column of both models associated with the variable that codes whether the potential intervener had a defense obligation to the target of the war. As before, I produce a figure that provides substantively meaningful quantities. Figure 2 provides predicted probabilities of a state intervening on behalf of the target of a war with 95% confidence intervals. The black density shows the estimated predicted probability of intervening on behalf of the target when the potential intervener does not have a defensive obligation to the target and and the gray density shows the estimated predicted probability of intervening on behalf of the target when the potential intervener has a defensive obligation to the target. Figure 2 shows that the predicted probability of a state intervening on behalf of the target of a war significantly increases when it has a defensive obligation to the target of the war. The mean predicted probability is 0.005 [0.004, 0.007] when a state does not have a defensive obligation to the target and it is 0.124 [0.073, 0.191] when a state does have a defensive obligation to the target. In other words, the predicted probability of a state intervening on behalf of the target of a war is 0.119 greater for a state with a defensive obligation to the target than a state without one. This results in a 24% (=0.119/0.005) increase in the probability of joining the target of a war.

OPEN IN VIEWER

Figure 2.

Predicted probabilities of joining the target.

Interestingly, the effect of defense pacts on joining targets of war is smaller than the effect of offense pacts on joining initiators of war. While it is beyond the scope of this study to fully evaluate this difference, one possibility is that it is the result of strategic behavior on behalf of war initiators. When states choose to initiate wars they may choose targets with less reliable defensive allies and targets that they believe their offensive allies will intervene against. If this is the case, this may introduce a relatively higher proportion of unreliable allies in the sample of invoked defense pacts, which will weaken the estimated effect of defense pacts on states’ decisions to intervene on the behalf of targets of war. However, this reasoning begs the question of why targets with unreliable allies choose to resist rather than concede and avoid war (e.g. Smith, 1995, 1996). Future research should aim to elucidate this process.

Figures 1 and 2 are consistent with the expectations of the informational theory of alliances. Alliances influence states’ decisions to initiate militarized disputes because they provide information about other states’ likelihood of intervening in a potential war. Potential challengers should be less likely to attack potential targets that are members of defense pacts because these potential targets are more likely to receive assistance in war. Furthermore, potential challengers that are members of offense pacts should be more likely to initiate disputes because these potential challengers are more likely to receive assistance in war. These findings provide additional empirical evidence that suggests the informational theory of alliances is useful for understanding the relationship between alliances and war.

The results of the multinomial logit regressions reported in Table 2 provide support for both hypotheses. Additionally, model 2 provides several other results regarding states’ war intervention behavior. First, states that are further away from wars are less likely to intervene. Second, the potential contribution a state can make to a war participant’s probability of winning only matters for joining the targets of war and not war initiators. States that can increase the target’s probability of defeating the initiator in war more are more likely to join the target. Third, the distribution of power between the initiator and target of a war has no significant effect on states’ war intervention decisions. Finally, the results reveal a more nuanced conclusion about the relationship between regime similarity and conflict joining than previous research (i.e. Werner and Lemke, 1997). The results suggest that states are not more likely to join states with similar regime types in war but states are more likely to intervene and fight against states that have different regime types.

While model 2 attempts to control for potentially confounding variables, there may be important unmeasured factors that are excluded from the analysis. Therefore, to address this concern and evaluate the robustness of the results, I estimate two additional models that include fixed effects. These models control for unit-specific unmeasured factors that could be biasing the results. ³¹ The models are reported in Table 3.

OPEN IN VIEWER

Table 3.

Fixed effect multinomial logit regressions of war intervention

	Model 3		Model 4
	War fixed effects		Potential joiner fixed effects
	Join initiator	Join target	Join initiator	Join target
Offensive obligation to the initiator	6.40 **	−25.69	5.40 **	−10.94
	(1.44)	(1,246,294.16)	(1.17)	(2046.88)
Defensive obligation to the target	1.12	4.24 **	0.48	3.51 **
	(0.89)	(0.59)	(0.84)	(0.39)
Distance from target	−1.72 **	−1.23 **	−1.58 **	−1.15 **
	(0.22)	(0.16)	(0.22)	(0.16)
Potential contribution to initiator	0.79	1.74	−1.93	−0.73
	(1.48)	(1.58)	(1.36)	(1.09)
Potential contribution to target	5.83 **	3.47 **	−4.64 **	0.91
	(1.49)	(0.94)	(1.35)	(0.82)
Initiator’s probability of winning	.	.	0.14	0.18
			(0.73)	(0.65)
Regime difference with initiator	−0.02	0.09 **	−0.06 *	0.03
	(0.03)	(0.03)	(0.03)	(0.02)
Regime difference with target	0.11 **	−0.06 *	0.09 **	−0.02
	(0.03)	(0.03)	(0.03)	(0.02)
Constant	6.62 **	−13.43	−2.54	6.66 **
	(1.65)	(750.15)	(481.06)	(1.76)
Observations	3008		3284
Log-likelihood	–342.19		–430.47
AIC	864.38		1136.95

Standard errors in parentheses.

*

p < 0.05, ^** p < 0.01.

First, one may argue that during certain time periods or wars states were more likely to form alliances and intervene in wars owing to the structure of the international system or the nature of the war. Therefore, I estimate a model including war fixed effects. That is, I include a dummy variable for each war in the data. This model allows me to control for unmeasured factors associated with each war in the data. Model 3 in Table 3 provides the results of this analysis. ³² The results show that even after controlling for unmeasured factors associated with each war, offense obligations increase the likelihood that alliance members will join the initiator of a war and defense obligations increase the likelihood that alliance members will join the target of a war.

Second, factors unique to each potential intervener may be important for alliance formation and war intervention. Therefore, as another robustness check I control for unmeasured factors associated with each potential intervener in the data. For this analysis, I include a dummy variable for each potential intervener in the data. Model 4 in Table 3 provides these results. ³³ The results show that the hypotheses are also supported when I control for unmeasured factors associated with each potential intervener in the data.

The results in Table 3 provide additional support for both hypotheses and reduce concerns regarding omitted variable bias. Additionally, most of the control variables behave similarly to the models without fixed effects: states further away are less likely to intervene, states tend to intervene against states with dissimilar regime types, a state’s potential contribution to the initiator has no effect and the distribution of power between the initiator and target has no effect. However, in the fixed effect models a state’s potential contribution to the target’s probability of winning does not have a consistent effect and there is some support for the claim that states will intervene on behalf of war participants with similar regime types.

Conclusion

A long tradition in international relations research has attempted to understand the relationship between military alliances and war. Early research focused on the systemic level and the “balance of power” but empirical tests produced inconsistent findings, suggesting that a different theoretical framework was necessary to understand the relationship between military alliances and war. As a result, scholars adopted a new theoretical framework that focuses on how alliances may influence a potential challenger’s decision to initiate a militarized dispute against a potential target. This research suggests that alliances influence war through the information they provide to potential challengers. Furthermore, alliances provide different information depending on the obligations of the alliance members. Defense pacts signal that the members will receive assistance in war if they are attacked and offense pacts signal that the members will receive assistance if they go on the attack.

This framework produces implications about potential challengers’ decisions to initiate militarized disputes and other states’ decisions to intervene. Previous research found empirical evidence that suggests that these different alliance obligations systematically influence potential challengers’ decisions to initiate militarized disputes. However, there is no systematic research showing that different alliance obligations influence states’ war intervention decisions. Therefore, in this study I present the first estimates of these different effects of alliance obligations and offer an additional test of the informational theory of alliances.

The empirical analysis produced two key findings that can provide directions for future research. First, I find strong evidence that a state will be more likely to intervene on behalf of the initiator of a war if it has an offensive obligation to the initiator. As a result, potential challengers that are members of offense pacts are more likely to initiate militarized disputes (Leeds, 2003). Offense pacts increase the number of militarized disputes initiated in the world and they expand wars. Therefore, it is important to understand the conditions under which states will make offensive obligations rather than defensive obligations. Recently, offense pacts seem to have become obsolete. States have not made formal offensive obligations in the post Cold War era. ³⁴ It is important to know whether this trend will continue.

Second, I find strong evidence that a state will be more likely to intervene on behalf of the target of a war if it has a defensive obligation to the target. Potential challengers should expect potential targets that are members of defense pacts to be more likely to receive assistance in war. As a result, defense pacts have a deterrence effect (Johnson and Leeds, 2011; Leeds, 2003). Defense pacts reduce the number of militarized disputes in the world but those disputes that defense pacts do not prevent from escalating to war tend to be multilateral wars. Future research can build on this finding by exploring why some defense pacts fail to deter even though they are reliable.