Defense Pacts

Defense Pacts: A Prescription for Peace?1

Jesse C. Johnson and Brett Ashley Leeds

Rice University

Do military alliances lead to peace or to war? Research has suggestedthat defensive alliances to potential targets deter dispute initiation(Leeds 2003b:427). This would seem to suggest that forming defensivealliances is a good policy prescription for those seeking to encouragepeace. Yet, some argue that even if defense pacts have a deterrenceeffect, defense pacts may also have other effects that increase milita-rized conflict in the international system. Specifically, defense pacts mayencourage member states to initiate and ⁄ or escalate disputes. In ananalysis covering the period from 1816 to 2000, we evaluate these threepotential effects of defense pacts—deterrence, initiation, and escalation.We find support for the hypothesis that defensive alliances deter theinitiation of disputes but no evidence in support of the claims thatstates with defensive allies are more likely to initiate disputes in theinternational system or that targets with allies are more likely torespond to dispute initiation with further militarization. We concludethat defensive alliances lower the probability of international conflictand are thus a good policy option for states seeking to maintain peacein the world.

Introduction: Security Institutions and the Quest for Peace

One of the most important goals of international relations scholars is to identifycost-effective foreign policy actions that can reduce the probability of militarycombat. War and militarized disputes short of war are directly costly in lives andtreasure and have additional opportunity costs in lost trade and development.Fortunately, militarized conflict is rare, but given its enormous cost, it remainsworthy of considerable scholarly attention.

What has been particularly frustrating to scholars wishing to influence policy,however, is that many of the factors that we know are associated with anincreased probability of war are not easily manipulable by policymakers. There isconsensus in the literature that stronger states are less likely to be attacked, butincreasing military capabilities dramatically through economic development, pop-ulation growth, building arms, and adding military personnel is a very costlyproposition. Democratic states are less likely to fight one another, but democrati-zation (of one’s own state or a potential adversary’s) is also a difficult process.

1 Previous versions of this paper were presented at the annual meeting of the International Studies Association,February 17–20, 2010, New Orleans, Louisiana, and at the North American meeting of the Peace Science Society,Chapel Hill, North Carolina, November 20–22, 2009. The authors thank Daina Chiba, Songying Fang, Doug Gibler,Michaela Mattes, Cliff Morgan, Gary Uzonyi, and participants in the Rice University International Relations Work-shop for helpful comments on this research. Data for replication and all additional analysis referenced in the textand footnotes are available at http://www.ruf.rice.edu/~leeds.

doi: 10.1111/j.1743-8594.2010.00122.x� 2010 International Studies Association

Foreign Policy Analysis (2011) 7, 45–65

Thus, it is not surprising that policymakers are attracted to the potential paci-fic effects of international security institutions. More easily manipulable thanregime type or power, alliances are a potentially useful policy tool for reducingthe probability of militarized disputes. But whether states can design alliancesthat make them safer without unintentionally increasing incentives for membersto initiate or escalate militarized conflicts is currently unclear to scholars. Whileevidence supports the claim that states with allies committed to defend them inthe event of attack are less likely to be the targets of militarized attack (Leeds2003b), studies have also shown that alliance membership may encourage statesto initiate disputes (Palmer and Morgan 2006) or to escalate conflicts to higherlevels (Smith 1995; Senese and Vasquez 2008).

In this study, we first test the hypothesis that defensive alliances to potentialtargets deter dispute initiation on a larger spatial-temporal domain than wasexamined in previous studies. We find support for this hypothesis. We then testhypotheses regarding the probability that states with defensive alliances are morelikely to initiate disputes and that targets with defensive alliances are more likelyto resist when challenged, opening the possibility of escalation to war. We findno support for these additional hypotheses. If anything, our results all aim in thedirection of peace—states with defensive allies are less likely to be the targets ofdisputes; states with defensive allies are no more likely to initiate disputes; andtargets with allies are no more likely to resist when challenged. In view of theseresults, we feel increasingly comfortable suggesting that defensive alliances pro-mote peace. While we believe that future research should aim to be more precisein specifying the particular conditions under which alliances are likely to be mosteffective at promoting peace, we see these results as a helpful first step thatshould be relevant to important current policy debates.

The Dangerous Side Effects of Defense Pacts

Recent theoretical research points out that war is an inefficient way to divideresources. If both sides knew what the outcome of a war would be, they wouldprefer to reach a peaceful division of their assets that would satisfy both sideswithout paying the costs of war. Incomplete information and commitment prob-lems, however, can prevent states from reaching a successful prewar bargain.When the leader of a challenger state believes that the status quo distribution ofbenefits does not reflect what he could claim through force, he will demand con-cessions from his target, with a threat of force should the demand be rejected.In other words, the challenger state will initiate a militarized interstate dispute(MID). If the leader of the target state believes that granting the concessions willproduce a preferable settlement to what could be gained through fighting, thetarget will concede the demand and peace will ensue. Alternatively, if the targetbelieves he can achieve a preferable outcome (relative to the settlement pro-posed by the challenger) through bearing the costs of war, he will resist thedemand, raising the possibility of escalation to war (Fearon 1995; Powell 1999).

How do alliances fit into the war story described above? Alliances are formalagreements among states to cooperate militarily in the event of conflict. Whilealliances vary in the actions they require of member states, a subset of alli-ances—defense and offense pacts—require member states to join conflicts involv-ing their allies when the casus foederis of the treaty is invoked. While alliances arenot always reliable, both theoretical and empirical studies suggest that challeng-ers should expect most states to fulfill their alliance commitments (Morrow1994; Smith 1995, 1998; Fearon 1997; Leeds, Long, and Mitchell 2000; Leeds2003a). Thus, defensive and offensive alliances provide ex ante information toboth potential challengers and potential targets that conflicts are unlikely toremain bilateral. Leaders of states with allies committed to assist them should

46 Defense Pacts and Peace

expect a higher probability of success in war, and leaders who expect their oppo-nents to gain assistance from allies should expect a lower probability of successin war. All else equal, potential targets with allies committed to assist themshould be less likely to be targets of MIDs, and potential challengers with alliescommitted to assist them should be more likely to initiate militarized disputes.Leeds (2003b) provides evidence in support of these hypotheses using data cov-ering the period from 1816–1944, and we test these hypotheses again on a largerspatial-temporal domain.

In recent years, offense pacts (that is, alliances that promise assistance in con-flicts initiated by a member state) have become rare. When states promise oneanother active military assistance in conflict, they most often limit their commit-ments to cases in which their partners are attacked. According to the AllianceTreaty Obligations and Provisions (ATOP) data set, during the period from 1815to 2003, 71% of alliances requiring active military assistance are limited to defen-sive situations only, that is, cases in which the sovereignty or territorial integrityof a member state is attacked. For alliances formed between 1945 and 2003, thatproportion rises to 85%, and to 100% for alliances formed after 1989 (Leedsand Mattes 2007). For those who seek policy prescriptions that make militarizedconflict less likely, this appears to be good news given the evidence that defen-sive alliances deter dispute initiation (Leeds 2003b).

Yet, some have questioned whether defensive alliances may have some addi-tional effects that embolden alliance members and increase the likelihood ofwar. First, some scholars have suggested that while defensive alliances may helpstates avoid being attacked, they may encourage them to go on the attack. Sec-ond, scholars have suggested that when challengers are not deterred and initiatedisputes against states with allies, target states may be particularly intransigent,leading to a higher probability of dispute escalation. Thus, while the deterrenceeffect of defensive alliances increases peace, the overall pacific effect of defensepacts may be tempered by an initiation effect and ⁄ or an escalation effect.

We consider the potential for an initiation effect of defensive alliances first.There are at least three different causal mechanisms that have been proposedthat could lead states with defensive allies to go on the attack. First, members ofdefensive alliances may make demands of other states believing that their allieswill help them in any conflict that results despite the fact that the conditions donot meet the specific casus foederis of the alliance agreement. Essentially, this is amoral hazard or ‘‘entrapment’’ argument—states may behave in a more riskymanner because of their increased protection (for example, Snyder 1984). Giventhat it is sometimes difficult to pinpoint who started a conflict (and bothsides have incentives to blame their adversaries for initiating the dispute), thisargument is intuitively plausible.

Policymakers have sometimes expressed concerns about the possibility ofentrapment. For example, France and Russia formed a military alliance in 1893,specifically aimed at deterring German attack. The agreement states ‘‘If Franceis attacked by Germany, or by Italy supported by Germany, Russia shall employall her available forces to attack Germany. If Russia is attacked by Germany, orby Austria supported by Germany, France shall employ all her available forces tofight Germany’’ (Hurst 1972:668). Yet, Russian leaders delayed making such anexplicit formal commitment for some time, out of fear that a promise of Russiansupport might lead France to provoke a war with Germany to regain Alsace-Lorraine (Langer 1951:8; Taylor 1954:345; Michon 1969:17, 61; Snyder 1997:113). The Russian czar explicitly warned the French ambassador of the need forFrance to avoid pursuing ‘‘revenge’’ shortly before accepting the agreement(Langer 1951:48; Taylor 1954:345).

Alternative arguments that suggest defense pacts may lead to an increase inmilitarized disputes also exist. A second argument suggests that defensive

47Jesse C. Johnson and Brett Ashley Leeds

alliances may reduce the resources that states must devote to preserving theirindependence and territorial integrity. To the extent that their defense burdenis reduced, states may choose to use the newly available resources to pursueaggressive action in the international system. Palmer and Morgan (2006), forinstance, argue that states pursue two foreign policy goals—maintenance andchange. In other words, states want to invest in keeping the things they like inthe international system and in changing the things they do not like. At anygiven time, a state is investing some portion of its foreign policy resources inmaintenance and the remainder in change, and the relative investment is deter-mined by relative preferences for the two goods, resource endowment, and theefficiency by which the two goods can be produced. If a defensive alliance allowsa state to invest fewer resources in defending itself (that is, producing main-tenance), the state might choose to use those newly freed resources to supportpolicies of change, which could involve initiating disputes against outside states.2

This argument differs from the moral hazard argument because there is noexpectation of allied assistance in the offensive endeavors but is similar to themoral hazard argument in predicting that states with allies committed to defendthem are more likely to initiate interstate disputes against outside states. In amonadic analysis, Palmer and Morgan find that joining a new alliance increasesthe probability a state initiates a new MID against some other state (2006:158).3

Just as we find historical evidence that policymakers have worried aboutentrapment, we also find anecdotal evidence that policymakers have expressedconcern about the effects of defensive alliances on the willingness of membersto initiate aggression in areas where they might not expect allied support. Uponwitnessing the formation of the same 1893 Franco-Russian alliance, British offi-cials feared that greater security in Europe might lead both France and Russia tobecome more aggressive in their pursuit of objectives outside of Europe, particu-larly French claims in Africa and Russian claims in Asia (Langer 1929:198, chap-ter 12; Michon 1969: 61). In fact, it was in 1894, shortly after signing the alliancewith Russia, that France created an independent Ministry of Colonial Affairs withTheophile Delcasse, an advocate of imperialist expansion, at its helm. France didpursue an aggressive colonial policy in the mid- to late 1890s, leading to disputeswith Britain in both the Far East and Africa (for example, Aldrich 1996; Kaiser2000:293).

A third and final argument linking defense pacts to increased conflict suggeststhat forming alliances may be part of a process that leads to increased militariza-tion and insecurity, which in turn make dispute initiation more likely. Seneseand Vasquez (2008) see alliance formation as one of a series of ‘‘power politics’’steps that states may use when dealing with issues in dispute. Every action thatmakes one state more secure will lead its rivals to feel more insecure and torespond with increased hostility, including counter-alliances, arms spending, ordispute initiation. The dyad ends up in a conflict spiral, each reacting to theother, that increases the probability that hard-liners take control of the statesand the states eventually end up in war. Thus, while alliances may not immedi-ately lead to war, alliance formation is dangerous because it contributes to a cli-mate of militarization that has a higher probability of eventually resulting in theuse of force. While Senese and Vasquez’s theory does not specify which side willinitiate the dispute, nor does it distinguish between types of alliances, Senese

2 Palmer and Morgan (2006) argue that alliances can be change-producing or maintenance-producing depend-ing on the relative capabilities of the state and its ally. To the extent that defensive alliances are limited to protect-ing the sovereignty and territorial integrity of alliance members, however, we believe they are more likely toenhance maintenance rather than change.

3 The Palmer and Morgan test does not distinguish between types of alliances. Ententes, neutrality ⁄ nonaggressionpacts, and defense pacts are all included in their analysis (2006; see also Morgan and Palmer 2003).


and Vasquez do expect the presence of alliances to be associated with a highernumber of militarized disputes and wars.

While the causal mechanisms are different, all three of these arguments sug-gest that defensive alliances may have indirect, sometimes unintended, effectsthat increase rather than reduce militarized conflict. While states that are mem-bers of defense pacts may be less likely to find themselves the target of a MID,they may be more likely to initiate interstate disputes. If there is evidence to sup-port this initiation effect, it should temper our enthusiasm for defensive alliancesas tools of peace.

Another concern arises when we consider the cases in which alliances do notdeter attacks. Even if defense pacts deter the initiation of some MIDs, nothing inthe theory suggests that all potential challengers are deterrable. While targetswith allies committed to defend them are less likely to be challenged than targetswithout defensive allies, some may still find themselves facing challengerdemands, either because challengers doubt the reliability of allies or becausechallengers believe their probability of success in a multilateral war is sufficientto justify making a demand. How will targets respond? Some scholars claim thattargets who expect allied support will be more likely to escalate the disputerather than making concessions (Smith 1995). Because targets with allies believetheir probability of winning is higher than in a bilateral setting, they are danger-ously intransigent. Thus, when defensive alliances fail to deter dispute initiation,the resulting conflicts may be particularly severe, as target resistance is a neces-sary step on the path to war. This escalation effect, too, might call into questionthe overall effects of defense pacts on peace.

In 1911, for instance, Germany initiated a crisis with France in Morocco.France intended to exert colonial control over Morocco, and Germany wasunwilling to allow France to expand her influence in the Mediterranean withoutappropriate compensation. The Russians worried that their French allies wererisking war by rejecting the substantial demands Germany made for territorialcessions in Africa. Indeed, while the French and Germans eventually agreedupon an exchange of colonial properties and avoided war, Russia was concernedthat the promise of support against Germany in Europe that was included in the1893 Franco-Russian alliance caused France to take too firm a position in bar-gaining. The result, the Russians feared, could be an unintended war (Michon1969:195; Snyder 1997:285).

In addition to historical anecdotes such as this, there is some systematic evi-dence that seems to link alliances to a higher probability that disputes escalateto war. Senese and Vasquez (2008) report that conflicting dyads in which one orboth states have outside allies are more likely to end up in war. While Seneseand Vasquez do not match specific alliances to wars directly, they do provide twodifferent statistical analyses that associate the tendency for states to have outsideallies with the tendency for their disputes to escalate to war. First, Senese andVasquez show that dyads that have spent a larger proportion of their conflict his-tory with both states or one state involved in an outside alliance are more likelyto end up in war at some point during their history than states that spend moretime without outside allies (2008: chapter 5).4 Second, Senese and Vasquez showthat if participants in a MID have outside allies, they are more likely to experi-ence war sometime within the next five years (although not necessarily owing tothe escalation of this particular MID) than if participants in a MID do not have

4 Senese and Vasquez’s (2008) alliance dominance measure includes defense pacts, nonaggression ⁄ neutralitypacts, and ententes, but only with ‘‘politically relevant’’ states. They examine each MID that the dyad is involved into see whether one or both sides were members of outside alliances. Whichever alliance status is the modal valuefor a dyad’s MIDs throughout its history is coded as its ‘‘dyad-history’’ value and used to predict whether the dyadever experiences war.


allies (2008: chapter 7). Neither of these analyses provides a direct test of theinfluence of alliances on the decisions of targets to escalate disputes, but if alli-ances do make targets more likely to escalate disputes, the resulting aggregatepatterns could be consistent with Senese and Vasquez’s results.

While we find the logic behind an initiation and an escalation effect of defen-sive alliances plausible and we believe testing for these effects is important forjudging the efficacy of alliances as a tool to promote peace, we also believe thatthere may be countervailing effects of alliances that limit their potential to encour-age the initiation and escalation of disputes. Several scholars have argued thatallies are in a unique position to restrain their partners from engaging in aggres-sive international actions. Snyder (1984, 1997) and Pressman (2008), for instance,argue that alliances produce norms and institutions through which members canexert influence over one another’s foreign policy. These authors suggest that themechanisms of information sharing and norms of consultation that developwithin alliances increase the ability of states to convince their allies to changecourse. In addition, a credible threat to withhold support or to terminate the alli-ance can be a potent bargaining tool under some circumstances. Gelpi (1999)argues that allies have both the incentive and the ability to serve as successfulmediators for disputes between their partners and outside states; because allieshave the ability to influence an eventual war outcome by withholding or offeringsupport, they are in a position to compel concessions from both sides.

There is reason to believe that states may sometimes wish to use these tools ofinfluence to limit the probability that their allies initiate and escalate disputes.First, and most obviously, states may wish to avoid being drawn into militarizedconflicts by their allies’ actions. But even if the conflict would not invoke allianceobligations and the state would not need to participate in a dispute initiated orescalated by a partner, the state might prefer that a partner avoid wastingresources on other conflicts and instead shepherd resources for conflicts thatmight emerge that are more germane to the state’s own interests. When thesefactors outweigh the benefits a state stands to gain from an ally’s aggression orresistance, the state may attempt to convince its ally to refrain from initiating dis-putes or resisting demands.

The 1893 Franco-Russian alliance also provides us with anecdotes of restraint.When France became involved in heated negotiations with Germany during thesecond Morocco crisis in 1911, Russia reportedly recommended that Franceaccept a less favorable deal with Germany (Snyder 1997:285) and even encour-aged France to ask Russia to mediate the dispute (Michon 1969: 195). Similarly,when the crisis over Austria’s annexation of Bosnia threatened to erupt into aEuropean war, France strongly encouraged Russia to moderate her position(Michon 1969:183–4; Snyder 1997:280) and tried to convince Austria’s allyGermany to agree to mediate the crisis jointly (Taylor 1954:453–5). In bothinstances, the crises were resolved without war.

Whether a state can restrain an ally depends on relative bargaining powerwithin the alliance. States that are relatively less dependent on an alliance aremore likely to have a credible threat to abandon an ally and thus have more con-trol over their ally’s behavior. Whether a state wishes to restrain an ally dependson the extent to which the state shares its ally’s interests in altering or preservingthe status quo and the costs the state expects from its partner’s participation inwar (Snyder 1984, 1997). Thus, none of the theorists who argue that states canand sometimes do restrain their allies claims that states always wish to restraintheir partners nor that they always succeed when they attempt to restrain theirpartners. Theories of allied restraint do, however, present an interesting counter-balance to claims that alliances encourage aggression in their members and dam-pen our concerns that the side effects of defense pacts lead unequivocallytoward conflict.


In sum, we believe that both theory and existing empirical evidence supportthe claim that defense pacts deter the initiation of MIDs against members;defensive alliances have a deterrence effect. Yet, whether the overall effect ofdefense pacts is pacifying seems questionable, because plausible arguments andcounter-arguments exist regarding the effects of defense pacts on incentives formembers to initiate and escalate disputes, and some existing empirical evidencesupports the claim that alliances are associated with more militarized disputesor wars. We believe that in some cases, defense pacts may have unintendedside effects that produce militarized conflict, but we also believe that thesecases are counterbalanced by cases in which allies successfully restrain the initi-ation and escalation of conflict. The question then becomes as follows: whicheffect dominates in the empirical data? Over a wide spatial-temporal domain,have defense pacts been associated with an increase in dispute initiation bymembers or the escalation of disputes to war? If so, this should temper ourenthusiasm for defense pacts as a prescription for peace. On the other hand, ifthere is limited evidence of negative side effects, we should feel confident pro-moting defense pacts as a tool to deter the initiation of militarized disputes.With this in mind, we test the following three hypotheses, associated respec-tively with the deterrence effect, the initiation effect, and the escalation effectof defense pacts:

Hypothesis 1: (Deterrence effect): A potential challenger is less likely to initiate amilitarized interstate dispute if the potential target has a relevant defensive alliance.

Hypothesis 2: (Initiation effect): A potential challenger is more likely to initiate amilitarized dispute if the potential challenger has a defensive alliance.

Hypothesis 3: (Escalation effect): A target of a militarized interstate dispute is more likelyto resist if the target has a relevant defensive alliance.

Research Design: Deterrence and Initiation Effects

Our first two hypotheses (regarding the deterrence and initiation effects) con-cern the initiation of MIDs. We evaluate these hypotheses on a sample of alldirected dyads in the interstate system as defined by the Correlates of War pro-ject over the 1816–2000 time period, which coincides with the availability of dataon our dependent variable. Our dependent variable is operationalized using theCorrelates of War MID data set and, specifically, Zeev Maoz’s dyadic version ofthe data set (Ghosn, Palmer, and Bremer 2004; Maoz 2005). A MID occurs when-ever one state threatens, displays, or uses force against another state. We limitour consideration to initial participants in the MID; we do not include joiners.5

The dependent variable for our analysis is coded one if the potential challengerinitiates a dispute against the potential target in the directed dyad-year and zerootherwise. Our sample includes a total of 1,077,992 directed dyad-years and2,354 directed dyad-years (0.22% of the observations) in which the challengerinitiates a dispute against the target.

Our first key independent variable is designed to evaluate the deterrenceeffect of defensive alliances. We employ a dichotomous variable that codes

5 We follow Maoz’s determination of which dispute dyads involve a ‘‘principal initiator’’ interacting with a‘‘principal target.’’ We do not include joiners in our analysis of the deterrence effect because joiners (on either theside of the initiator or target) usually make an active choice to participate in a dispute. Failed deterrence is not thecause of their participation in the dispute. We do not include joiners in our analysis of the initiation effect becausewe believe that the decision for an ally to join an ongoing dispute may be driven directly by alliance commitmentsthat obligate the state to join disputes involving an ally and not an emboldening effect that encourages the initia-tion of new disputes.


whether the potential target had a defensive alliance that is applicable to a dis-pute with the potential challenger to test for this relationship. We code this vari-able using information from the ATOP data set (Leeds, Ritter, Mitchell, andLong 2002; Leeds and Mattes 2007).

The ATOP data set provides information about the obligations of states aswritten in their alliance agreements. Alliances include one or more of five differ-ent obligations—defensive support, offensive support, neutrality, nonaggression,and ⁄ or consultation. This study is primarily concerned with alliances thatinclude provisions for defensive support, which we refer to as defense pacts.According to the ATOP coding rules, defense pacts are alliances that include ‘‘apromise of active military support in the event of attack on the sovereignty orterritorial integrity of one or more alliance members’’ (Leeds and Mattes 2007:189).6

Defense pacts do not always obligate all members to defend one another in allcircumstances, however. Forty-five percent of defense pacts in the ATOP data setspecify certain conditions under which the defensive obligation becomes opera-tive, with the most common restriction being a restriction to a specific adversary,a restriction found in 31% of all defense pacts. In addition, some defense pactsare asymmetric in which one partner makes no commitment to defend the otherin return for defensive support. To the extent that the parties to an alliancehave made it clear that their alliance would not be invoked by an attack by aparticular potential adversary against a particular member, that adversary shouldnot be deterred by the existence of the alliance.

We use several coding rules to determine which defense pacts should beexpected to exhibit a deterrent effect in which potential conflict dyads. First, weeliminate all secret defensive alliances. If the potential challenger is unaware ofthe alliance, then we would not expect the potential challenger to be deterred.7

Second, we code asymmetric defensive alliances as having a potential deterrenteffect only for alliance members who can expect to receive defensive supportaccording to the provisions of the agreement, and not for those who promise toprovide support to others but do not receive similar promises themselves. Third,defensive alliances that are conditional on specific adversaries are coded asrelevant only in cases in which the potential challenger is an adversary thatwould invoke the alliance commitment.8 In other words, if a potential target is amember of a defensive alliance in a given year that is conditional upon a specificadversary, then the potential target is coded as having a defensive alliance in anydirected dyad including a potential challenger identified in the context of thealliance agreement, but not in its other dyadic relations. So, for instance, the1925 alliance between France and Czechoslovakia requires the two member statesto defend one another in the event of an unprovoked attack by Germany, butnot under other conditions. Thus, both France and Czechoslovakia are coded ashaving an ally committed to defend them in the Germany–France and Germany–Czechoslovakia directed dyads, but not (for example) in the Russia–France orRussia–Czechoslovakia directed dyads. Fourth, we code alliances that are limited

6 Any alliance that includes a defense provision is included in this analysis regardless of its other content. Forexample, if an alliance agreement includes provisions for both defense and neutrality or for defense, neutrality,and consultation, it is coded as a defense pact for the purpose of this study.

7 We remove only alliances that include provisions requiring members to keep the full existence of the alliancesecret, not those that have some public and some secret content. We are aware that some, or even most, of thesealliances may become known to (or at least highly suspected by) potential adversaries. Because of this, we also ana-lyze a model that includes alliances with provisions requiring the signatories to keep the alliance secret in our testof the deterrence effect. Our conclusions are not affected by the inclusion of secret alliances.

8 Some cases involve difficult coding decisions because the specification of adversaries who invoke the alliancecommitments is ambiguous. For example, some examples of phrasing that required special consideration are‘‘European great powers,’’ ‘‘Balkan states,’’ or ‘‘imperialist states.’’ We provide coding notes describing thesecoding decisions with our replication data set.


to a particular ongoing conflict as applicable only in those directed dyad-years inwhich the potential challenger is involved in the specified conflict on the oppo-site side. Fifth, we code all remaining defensive alliances as applicable to disputeswith all potential challengers in a given year.9

Most of the remaining defensive alliances are unconditional and would beinvoked by any potential conflict. Some, however, are conditional on conflictoccurring in a specific location, for example on the home territory of the alli-ance members and not on colonial territories or military outposts. When con-flicts occur, it is easy to determine whether the casus foederis for the alliance wasinvoked or not based on the location of the attack. The problem is that inmost dyad-years, no conflict occurs, but conflict might have occurred in the loca-tion specified in the treaty. Because it is possible that conflict could haveoccurred in the specified location with any potential challenger, we code thealliance as potentially relevant in relations with all potential challengers. Insome cases, this means we will attribute failed deterrence to alliances that werenot relevant to the conflict. For example, Argentina initiates a militarized dis-pute against Great Britain that becomes the Falklands War. Great Britain didhave a number of defensive alliances in operation at the time, but none wereinvoked by attack on its territory in the Falklands. Yet, if Argentina hadattacked Great Britain in its home territory, the alliances would have beeninvoked. Thus, for 1982 (and the surrounding years), Great Britain is consid-ered to have defensive allies who might contribute to deterring an Argentineattack. While this coding rule does add some noise to our data, we believe it isthe most theoretically justified systematic rule we can choose. Implementingthe full set of coding rules described in the last two paragraphs produces585,467 directed dyad-years (54.31% of the observations) in which the potentialtarget had a defensive alliance that was applicable to a dispute with the poten-tial challenger.10

Our second key independent variable allows us to evaluate the initiation effectof defense pacts. Here, we utilize a dichotomous variable, also drawn from theATOP data set, that codes whether the potential challenger had any allies com-mitted to defend her under any conditions in the given year. We do not distin-guish between conditional and unconditional alliances in this analysis. Becausethe arguments supporting an initiation effect do not depend on an ally having atreaty obligation to join a conflict, we see no reason to distinguish betweendefense pacts based on the conditions under which they can be invoked. Secretalliances are also included in the analysis. While secret alliances may beconcealed from potential adversaries, they are known to the alliance membersand thus can influence the conflict initiation behavior of alliance members.Asymmetric alliances, however, are coded as influencing the initiation decisions

9 For multilateral alliances, defensive obligations that are not conditional upon a specific adversary are appliedto potential conflicts among members. That is, we code potential targets as having allies committed to defend themagainst other members of a defensive alliance. Members of NATO, for instance, are coded as having defensive alliesin their relations with other members of NATO. Our reported results, however, are not sensitive to this coding deci-sion. We performed a number of robustness checks including (in separate analyses): (i) dropping directed dyads inwhich both states are members of the same defensive alliance from the sample; (ii) coding all directed dyads inwhich both states are members of the same defensive alliance as having no allies committed to defending them;and (iii) controlling for shared defensive alliance membership in our analysis. In no case, did our assessment ofsupport for our hypotheses change.

10 There are differences over time in the proportion of directed dyads in the international system in which thepotential target has an ally committed to its defense. In a sample covering 1816–1945, 14.31% of directed dyad-yearsinclude a potential target with a relevant defensive ally, whereas in a sample covering 1946–2000, 64.10% of direc-ted dyad-years include a potential target with a relevant defensive ally. Large, long-lasting multilateral alliancesformed during the Cold War (for example, NATO, the Rio Pact ⁄ OAS, the Arab Collective Defense Pact, and theWarsaw Pact) have resulted in a much higher proportion of states in the international system belonging to one ormore defense pacts.


of alliance members for whom the alliance offers defensive support only, andnot for those members who promise to defend others but receive no promise ofdefensive support from their partners. In 592,261 directed dyad-years (54.94% ofthe observations), the potential challenger has at least one ally committed todefend her if she is attacked.

In coding both of these independent variables (the potential target’s alliancestatus and the potential challenger’s alliance status), we are careful not to attri-bute the initiation of disputes to agreements that were not in effect at the timeof the dispute. Our data are aggregated by year, so in most cases, if an alliance isin effect for any part of a year, that year is coded as an alliance year. For years inwhich MIDs begin, however, we remove any alliances that were (i) formed afterthe beginning of the MID or (ii) terminated before the beginning of the MID,because such alliances could not have caused or deterred dispute initiation.Thus, alliances that are formed in response to a MID do not influence ourestimates.

To understand the contribution of alliances to deterring or encouraging theinitiation of conflict, it is important to control for other factors that determinethe baseline probability of conflict within a dyad. There are many dyads in ourdata set that would be unlikely to experience a MID under any circumstancesowing to factors unrelated to alliances. We use a number of control variables tocapture an underlying probability of conflict that then allows us to judge moreappropriately the impact of defense pacts on deterrence and conflict initiation.To the extent that some of these control variables could be related as well to thepropensity to form defense pacts, excluding them from the model will bias ourresults.

We include four variables in our model that have been shown consistently tobe related to the probability of MID initiation.11 The first is a variable that codesthe potential challenger’s likelihood of defeating the potential target in aconflict. If a potential challenger is unlikely to defeat the potential target in aconflict, then the potential challenger should be less likely to initiate a disputeagainst the potential target. To operationalize the potential challenger’s likeli-hood of defeating the potential target in a conflict, we use the ratio of thepotential challenger’s military capabilities to the sum of the potential chal-lenger’s and potential target’s military capabilities. As this variable approachesone, the potential challenger is more likely to defeat the potential target in aconflict, and thus, we expect a positive relationship between this variable andMID initiation. The states’ Correlates of War CINC scores are used as measuresof their capabilities (Singer, Bremer, and Stuckey 1972).

The second variable is the natural log of the capital-to-capital distance betweenthe potential challenger and the potential target. As the distance between thepotential challenger and the potential target increases, the potential challengershould be less likely to initiate a dispute because conflict with more distant statesis more costly and states further apart have fewer issues to dispute. We expect anegative relationship between distance and MID initiation. The data on distancewere collected with the use of the EUGene data generation program (Bennettand Stam 2000).

The third variable measures how similar the potential challenger’s interestsare to those of the potential target. If the two states have similar interests, thepotential challenger will be less likely to initiate a dispute. We use Signorino andRitter’s S-score to measure this variable (1999). Higher values of the S-scoreindicate more similar interests, so this variable should be negatively related toMID initiation.

11 EUGene software version 3.201 is used to generate this set of variables (Bennett and Stam 2000).


Fourth, we include a measure of joint democracy. Numerous studies posit anddemonstrate that two democracies are less likely to enter into a dispute thanother dyads (for example, Russett and Oneal 2001). To measure joint democ-racy, we use a dichotomous variable that codes whether both states receive ascore of 6 or higher on the Polity2 variable from the PolityIV data set (Marshalland Jaggers 2002). We expect joint democracy to be negatively related to disputeinitiation.

In addition, following Leeds (2003b), we include two additional alliance vari-ables that have previously been shown to affect the initiation of MIDs. Weinclude one variable that codes whether the potential challenger has an offen-sive alliance that is applicable to a dispute with the potential target and one var-iable that codes whether the potential challenger has a neutrality pact that isapplicable to a dispute with the potential target. Challengers who expect sup-port from outside states and challengers who expect their targets not to receivesupport (because states have promised to remain neutral should conflict arise)may believe they are more likely to succeed in accomplishing their goalsthrough the threat or use of force and will be more likely to initiate a dispute.Thus, these variables should be positively related to dispute initiation. Informa-tion from the ATOP data set and a coding procedure similar to the coding pro-cedure described earlier is used to generate these variables (see also Leeds2003b).

To estimate the deterrence and initiation effects of defense pacts, we use aprobit model that includes all of the variables described earlier in the systematiccomponent of the model. Furthermore, to take into account any possible tempo-ral dependence in our data, we employ the strategy suggested by Carter andSignorino (2010). That is, we include a variable that codes the number of yearsthat have elapsed since the last conflict in the dyad as well as the square and thecube of that variable in our estimation.

Empirical Results: Deterrence and Initiation Effects

Table 1 presents the results of our analysis of dispute initiation. Consistent withtheoretical expectations and previous empirical research, we find that defensivealliance agreements have a deterrence effect. That is, a potential challenger willbe less likely to initiate a dispute against a potential target with a relevant defen-sive alliance than a potential target without one.12 Figure 1 provides predictedprobabilities of dispute initiation with 95% confidence intervals for an averagedyad in which the potential target has a relevant defensive alliance and for anaverage dyad in which the potential target does not have a relevant defensive alli-ance. The predicted probability of dispute initiation is approximately 20% lowerin an average dyad in which the potential target has a relevant defensive alliancethan it is in an average dyad in which the potential target has no relevant defen-sive alliance.13

Thus, our model reveals a statistically significant deterrence effect, suggestingthat defensive alliances promote peace. But what does our analysis say about thepotential additional effects of defensive alliances that may increase conflict? Wefind no evidence that defensive alliances encourage states to initiate disputes.Table 1 indicates that the coefficient for our variable capturing whether thepotential challenger has one or more allies committed to defend her is negative

12 To check the robustness of our results and ensure that they are not being driven by a small number of large,long-lasting alliances, we dropped observations in which the target was a member only of NATO, the OAS and ⁄ or theRio Pact, or the Arab Collective Defense Pact—four alliances that account for a significant proportion of our alliance-dyad-year observations. Our interpretation of results remains unchanged.

13 We generate predicted probabilities using Clarify (King, Tomz, and Wittenberg 2000) holding the values ofall other independent variables constant at their mean values.


and statistically significant. States with allies committed to defend them are nomore likely to initiate disputes than states without allies committed to defendthem and, if anything, are less likely to initiate disputes.

Because this finding goes against some established arguments in past litera-ture, we were concerned that multicollinearity or post-treatment bias might bemasking the true effect of challenger defense pacts on conflict (King and Zeng2007). In other words, we thought it might be possible that some of our controlvariables (for instance, defense pacts to the target) might be correlated with, oreven caused in part by, the challenger defense pact variable. To be confidentthat this estimate is not a result of multicollinearity or post-treatment bias, weestimated a number of simpler models. For example, we estimated a modelincluding only challenger defense pacts as an independent variable, a modelincluding only challenger defense pacts and distance as independent variables,and models using a variety of other combinations of control variables. In none

TABLE 1. Probit Analysis of Dispute Initiation 1816–2000

Deterrence effect: potential target has a relevant defensive alliance )0.062 (0.015)***Initiation effect: potential challenger is a member of a defensive alliance )0.068 (0.016)***Potential challenger has a relevant offensive alliance 0.276 (0.029)***Potential challenger has a relevant neutrality pact 0.315 (0.025)***Challenger’s likelihood of winning 0.161 (0.022)***Distance )0.384 (0.006)***Similarity in alliance portfolios )0.455 (0.041)***Joint democracy )0.155 (0.027)***Constant 0.827 (0.068)***Observations 1,077,992

(Notes. Standard errors in parentheses; ***p < .001, **p < .01, *p < .1; peace years, (peace years)2, (peace years)3

included in estimation).

FIG 1. Predicted Probabilities of Dispute Initiation(Notes. These densities are 1,000 simulated conditional probabilities of a potential challenger initiat-ing a dispute. The black density represents the estimated probability of a potential challenger initiat-ing a dispute when the potential target does not have a relevant defensive alliance, and the graydensity represents the estimated probability of a potential challenger initiating a dispute when thepotential target has a relevant defensive alliance. The conditional probabilities were calculated usingsimulated coefficients from our model of dispute initiation and setting the covariates at values for anaverage directed dyad-year.)


of these simpler models was the estimated coefficient for the variable indicatingwhether a potential challenger had at least one defensive ally positive and statisti-cally significant, although in many of the simpler models, the coefficient wasnegative and statistically insignificant. We are convinced that in our sample, thepotential initiation effect of defensive alliances does not undermine the notionthat defensive alliances promote peace.

The rest of the estimates in Table 1 are consistent with previous research. Sim-ilar to Leeds (2003b), we find that potential challengers with relevant offenseand neutrality pacts are more likely to initiate disputes; the estimated coefficientsfor both of these variables are positive and statistically significant. This is consis-tent with the idea that the effect of alliances on conflict is conditional upon thecontent of the alliance. As we expected, we find that potential challengers areless likely to initiate disputes when they are unlikely to win, they have interestssimilar to the target, the target is far away, and both actors are democratic.14

The estimated coefficients for the variable representing the power ratio betweenchallenger and target are positive and statistically significant, whereas the esti-mated coefficients for variables representing distance, similarity in alliance port-folios, and joint democracy are negative and statistically significant.

The fact that defense pacts have a positive deterrence effect and no discern-ible initiation effect suggests that alliances can be a useful policy tool for promot-ing peace. For policy-making purposes, however, it is interesting to know notonly whether particular variables make militarized conflict less likely but also therelative effects of different manipulable factors. Presumably, policymakers prefera stronger reduction in conflict propensities when it is available at reasonablecost. Thus, we compare the reduction in the probability of dispute initiationassociated with defense pacts to those associated with changes in power ratiosand regime type.

Recall that calculations based on an average dyad reveal that a state with adefensive ally is approximately 20% less likely to be the target of a MID than asimilar state without a defensive ally. Compare this to the effects of changing thepower balance between a potential challenger and a potential target. The pre-dicted probability of dispute initiation in an average dyad in which the potentialtarget is three times stronger than the potential challenger is 13% lower than inan average dyad in which the potential target’s power is approximately equal tothat of the potential challenger.15 Thus, to achieve a 20% reduction in the prob-ability of being the target of a militarized dispute by developing one’s own mili-tary capacity is probably out of reach for many countries; such a reduction in theprobability of conflict requires a very large change in the power relationshipbetween potential challenger and potential target. Joint democracy does have astronger conflict-reducing effect. The predicted probability of dispute initiationin an average dyad that is jointly democratic is approximately 40% lower than inan average dyad that is not jointly democratic, approximately double the averagereduction achieved by a defensive alliance. Democratization, however, of eitherone’s own state or a potential adversary’s (or both) is a difficult and expensiveproposition, especially in the short term.

Our analysis of dispute initiation does not reveal evidence that defensive alli-ances increase conflict. Yet, Palmer and Morgan (2006) provide statistical evi-dence of a relationship between alliance formation and conflict. Why are ourresults different? There are three primary research design differences that we

14 We also estimated a model that included a control variable for contiguity. Contiguity is positively related toMID initiation in our analysis, which is consistent with existing research, but the inclusion of contiguity as a controlvariable does not change our interpretation of results regarding our key independent variables.

15 These predicted probabilities (and those for joint democracy described below) were generated using Clarify(King et al. 2000) holding the values of all other independent variables constant at their mean values.


suspect are consequential for the inferences drawn.16 First, we employ differentunits of analysis. Palmer and Morgan evaluate whether a new alliance is associ-ated with a state’s propensity to initiate conflict at a monadic level. We insteadinvestigate whether the existence of a defense pact makes a state more likely toinitiate a militarized dispute against a particular target, controlling for factorsthat make a directed dyad more or less likely to engage in conflict. Second, Pal-mer and Morgan examine the effect of the formation of new alliances, whereaswe examine the effect of the existence of alliances. Their alliance variable cap-tures only whether a state acquired one or more new alliance partners in thegiven year, not whether the state had any alliance partners. Thus, many fewerstate-years are coded as alliance years in their analysis than in ours. Third, Pal-mer and Morgan do not distinguish between types of alliances. The formation ofany alliance of any kind, whether the alliance provided for active assistance, isincluded in their analysis.17 We believe that our research design, and particularlyconsidering the full time period of the alliance and limiting the analysis todefense pacts, is more appropriate for evaluating the relationship betweendefense pacts and the propensity to initiate interstate conflict.

To summarize our results so far, we find evidence in support of a deterrenceeffect for defensive alliances but no evidence in support of an initiation effectfor defensive alliance. The formation of defense pacts, therefore, seems to resultin the initiation of fewer MIDs in the international system. But, how do defensepacts affect those disputes that are not deterred? Are disputes involving targetswith allies committed to defend them particularly likely to escalate to moresevere levels? We turn now to this question.

Research Design: Escalation Effect

To estimate the escalation effect of defensive alliances, we need to evaluate thepotential target’s behavior once a dispute is initiated. Therefore, we are inter-ested in the sample of observations where the potential challenger initiated a dis-pute against the potential target. This sample comprises 2,354 of the 1,077,992directed dyad-years from 1816 to 2000.

The argument we are evaluating predicts that states with allies committed toassist them will be less likely to concede demands because of their greater expec-tation of success should the dispute escalate. In other words, targets with relevantdefense pacts may be more likely to respond to dispute initiation by continuingthe dispute and resisting the challenger. The dependent variable we use toevaluate this argument is a dichotomous variable coded one if the target takesmilitarized action in the dispute (that is, responds with a threat, display, oruse of force) according to the COW MID data set, and zero otherwise.18 The tar-get responds with a threat, display, or use of force in 1,115 of the 2,354 directeddyad-years (47.37% of the observations). Target response is a necessary first steptoward the potential escalation of a dispute to higher levels of force and poten-tially war.

The key independent variable we use to estimate the escalation effect of defen-sive alliances is nearly the same variable we use to estimate the deterrence effectof defensive alliances. It is a dichotomous variable that codes whether thepotential target has a defensive alliance that is applicable to a dispute with thepotential challenger. The coding procedure for this variable is described earlier.

16 Palmer and Morgan’s model also includes different control variables than ours does.17 Palmer and Morgan operationalize alliances using the Correlates of War formal alliance data set (Gibler and

Sarkees 2004). The CoW alliance data set includes defense pacts, nonaggression ⁄ neutrality pacts, and ententes.18 In directed dyad-years in which more than one MID occurs, we consider the MID with the highest hostility

level by the target.


The one difference is that in our analysis of the escalation effect, secret defensivealliances are included, because they are known to the target and should there-fore influence the target’s behavior.19 In 1,134 of the 2,354 directed dyad-years(48.17% of the observations) in which a dispute is initiated, the target hasa defensive alliance that could be applicable to a dispute with the potentialchallenger.

In addition to this variable, we also include several other independent vari-ables in the model that may also affect the target’s decision to escalate the dis-pute. First, we include dichotomous variables that code whether the potentialchallenger had any offense or neutrality pacts that were applicable to a disputewith the target. These alliances would make the target less likely to respondmilitarily because they provide information that, all else equal, the target is lesslikely to win a conflict against the challenger. Second, we include a variablethat measures the challenger’s likelihood of defeating the target in a conflict.Targets should be less likely to respond to a challenge militarily as the likeli-hood of the challenger defeating them in a conflict increases. We use the sameoperationalization as described earlier, the ratio of the potential challenger’scapabilities to the sum of the potential challenger’s and potential target’scapabilities.

We also take into account any unmeasured factors that may influence the chal-lenger’s decision to initiate a dispute and the target’s decision to resist by esti-mating a censored probit model in which the dependent variable for theselection stage is the same as the dependent variable for the previous researchdesign, a dichotomous variable that codes whether the potential challenger initi-ates a dispute against the potential target in the directed dyad-year. All of theindependent variables that were used in the previous research design areincluded in the selection stage of the censored probit model with the same oper-ationalizations discussed earlier: whether the potential target has a relevantdefensive ally, whether the potential challenger is a member of a defense pact,whether the potential challenger has a relevant offensive ally, whether the poten-tial challenger is a member of a relevant neutrality pact, the potential chal-lenger’s likelihood of winning, distance between the two states, similarity of thetwo states’ foreign policy interests, a dummy variable indicating whether bothstates are democracies, and the number of years since the last MID in the dyad,plus squared and cubed values of the peace years variable.

We identify the censored probit model using several variables: the challenger’sdefensive alliance membership, distance, S-score, and joint democracy. That is,we include these variables in the selection stage of the censored probit model,but we exclude them from the outcome stage. We contend that these variablesinfluence the challenger’s decision to initiate a dispute but not the target’s deci-sion to resist the challenger.20 As a robustness check, however, we estimate the

19 Removing secret alliances from the test of the escalation effect does not change our results.20 Why should some of these variables influence the challenger’s decision to initiate a dispute, but not the tar-

get’s decision to resist, and thus be included in only the selection and not the outcome stage of our model? Fortwo variables (challenger’s defense pact and distance), we believe the impact is really on the challenger’s decision,much more so than the target’s decision. Because most of the theories of the initiation effect do not rely on theassumption that the ally will help the challenger in the dispute, there is no reason to believe the challenger’sdefense pact status will influence the target’s decision. Because the dispute is likely to be fought on the target’s ter-ritory, distance between the states is much more consequential for the challenger’s decision than for the target’sdecision. Similarity in alliance portfolios is included in the dispute initiation equation because we believe that it isan indicator of whether contentious issues are likely to exist between the two states. Disputes are much less likely tooccur between states that agree on policy. Once a dispute has been initiated, however, we know there is a conten-tious issue, and thus, this variable is no longer relevant. The final variable we leave out of the outcome equation isjoint democracy. While there are theories of target resistance that predict different behaviors depending on theregime type of the challenger, these arguments are not dependent on dyadic democracy.


censored probit model including each one of the identifying variables in the out-come stage of the model in turn, and our inferences do not change.21

Empirical Results: Escalation Effect

Table 2 presents the results of our analysis of target resistance. These results donot support arguments in favor of an escalation effect of defense pacts. Targetsof disputes with relevant defensive alliances are no more likely to resist challeng-ers than targets without relevant defensive alliances. The coefficient on the vari-able is negative and statistically significant with 90% confidence intervals in themodel presented in Table 2. As a robustness check, we estimated a number ofother models including each of the variables in the selection equation of thecensored probit model in the outcome equation of the model. The variable rep-resenting whether a dispute target has a relevant defensive alliance was negativein each case although sometimes did not reach conventional levels of statisticalsignificance. There is no systematic evidence in this sample to suggest that addi-tional effects of defense pacts on conflict behavior undermine the notion thatdefensive alliances promote peace.

With regard to the rest of the estimates in Table 2, there are two new results.Previous research and our estimates from Table 1 support the idea that offenseand neutrality pacts increase conflict by encouraging potential challengers to ini-tiate disputes. Table 2 does not dispute this finding, but it shows that once a dis-pute is initiated, a target will be less likely to resist a potential challenger with arelevant offense or neutrality pact. Therefore, offense and neutrality pactsincrease dispute initiation, but they reduce target resistance.

We believe that target resistance is an important first step in determiningwhich disputes escalate to war. If a target does not respond to dispute initiationwith militarized action, it is hard to imagine how war could result. On the otherhand, while targets respond with force in 47% of dispute initiation cases, only4% of these (93 cases) escalate to war. Thus, as an additional check, we estimatethe same censored probit model replacing target resistance with war as thedependent variable in the outcome equation. We rely on the hostility level cod-ing to determine which disputes in the MID data set end in war (Ghosn et al.2004). The results are similar and appear in Table 3. Cases in which the targetof a dispute has a relevant defensive ally are not more likely to escalate to war;the coefficient on the relevant independent variable is negative and statisticallysignificant.22

Again, our results seem to disagree with some published results, most notablythose presented by Senese and Vasquez (2008), and again differences in theunit of analysis and the types of alliances evaluated might account for the differ-ent conclusions.23 Our evaluation examines how a target responds in a particu-lar MID and whether a particular MID escalates to war. Senese and Vasquez, on

21 Sartori (2003) provides a possible strategy for estimating selection models in which all the variables influenc-ing selection also influence the outcome of interest. Sartori’s model is only appropriate, however, when a strongtheory exists about the sign of the unmeasured factors influencing both selection and outcome. In fact, the signs ofall of the coefficients in the outcome stage of the model are dependent upon the assumption made about the signsof the unmeasured factors. Because we do not feel confident in our ability to identify, and thus theorize about, allunmeasured factors linking the challenger’s decision to initiate a dispute and the target’s decision to resist, we iden-tify our model with variables that we believe are more likely to influence the selection stage than the outcome stagerather than using Sartori’s estimator.

22 This is consistent with results reported by Leeds (2005) on a smaller sample of cases. Leeds (2005) does find,however, that disputes in which the target has allies committed to assist her are more likely to expand beyond theinitial participants and that conflicts that expand are more likely to escalate to war.

23 Our control variables also differ from those of Senese and Vasquez (2008). Senese and Vasquez (2008: 146,192) control for the issue under dispute, prior MIDs ⁄ enduring rivalry, and arms races and not the variables we con-trol for.


the other hand, evaluate whether war occurs any time during a particular timeperiod (either the full dyad-history or within five years after a MID) given thatat some point, a MID occurred among participants with outside allies. Seneseand Vasquez evaluate the effect of alliance with any major or regional powerregardless of the obligations of the agreement; that is, defense pacts, neutral-ity ⁄ nonaggression pacts, and ententes are all included in their analysis (seeGibler and Sarkees 2004). We, on the other hand, do not limit the types ofalliance partners we consider, but we do limit our consideration to defensepacts only. Thus, our analysis is more appropriate for judging specifically the

TABLE 2. Censored Probit Analysis of Dispute Initiation and Resistance 1816–2000

Target resistanceEscalation effect: potential target has a relevant defensive alliance )0.103 (0.048)*Potential challenger has a relevant offensive alliance )0.241 (0.084)**Potential challenger has a relevant neutrality pact )0.261 (0.075)**Challenger’s likelihood of winning )0.304 (0.074)***Constant 1.672 (0.119)***

Dispute initiationDeterrence effect: potential target has a relevant defensive alliance )0.060 (0.015)***Initiation effect: potential challenger is a member of a defensive alliance )0.066 (0.015)***Potential challenger has a relevant offensive alliance 0.276 (0.029)***Potential challenger has a relevant neutrality pact 0.315 (0.025)***Challenger’s likelihood of winning 0.161 (0.022)***Distance )0.383 (0.006)***Similarity in alliance portfolios )0.449 (0.041)***Joint democracy )0.168 (0.027)***Constant 0.818 (0.067)***Rho )0.588 (0.052)***Observations 1,077,992Uncensored observations 2,354


included in dispute initiation estimation stage).

TABLE 3. Censored Probit Analysis of Dispute Initiation and Escalation to War 1816–2000

Escalation to warEscalation effect: potential target has a relevant defensive alliance )0.479 (0.105)***Potential challenger has a relevant offensive alliance 0.252 (0.150)Potential challenger has a relevant neutrality pact )0.467 (0.190)*Challenger’s likelihood of winning )0.126 (0.153)Constant )0.601 (0.298)*

Dispute initiationDeterrence effect: potential target has a relevant defensive alliance )0.062 (0.015)***Initiation effect: potential challenger is a member of a defensive alliance )0.069 (0.016)***Potential challenger has a relevant offensive alliance 0.276 (0.029)***Potential challenger has a relevant neutrality pact 0.315 (0.025)***Challenger’s likelihood of winning 0.161 (0.022)***Distance )0.384 (0.006)***Similarity in alliance portfolios )0.457 (0.041)***Joint democracy )0.156 (0.027)***Constant 0.829 (0.067)***Rho )0.313 (0.096)**Observations 1,077,992Uncensored observations 2,354


included in dispute initiation estimation stage).


influence of defense pacts on the target’s willingness to escalate an individualmilitarized dispute.

So, how can we explain the fact that targets with allies are not more likely torespond militarily to dispute initiation? Presumably, targets with allies are morelikely to believe they can be successful in conflicts than targets without allies, soon first examination, this seems like an odd result (Smith 1995). First, it is possi-ble that only very strong and resolved initiators are not deterred from initiatingdisputes against targets with allies, and targets avoid escalating disputes againstsuch initiators. Our selection model and control variables should capture thisdynamic, however, and thus, we believe this is an insufficient explanation for ourresults. More likely, allies play a useful role in brokering a peaceful deal to keepthese disputes from escalating, as our discussion of the potential restrainingeffect of alliances would suggest (Gelpi 1999). It is also possible, however, thatchallengers adjust their demands and ask for smaller concessions from targetswith allies committed to assist them, increasing the probability that targets con-cede without fighting (Werner 2000; Yuen 2009). Evaluating these potential cau-sal mechanisms for this empirical result is an important direction for futureresearch.

A Policy Prescription for Peace

One of the main goals of international relations scholars should be to identifypolicies that states can adopt that are likely to reduce the probability of milita-rized conflict. In addition to resulting in the direct loss of life, militarized dis-putes have indirect effects on quality of life through the reduction and diversionof productive capacity. Militarized conflict is fortunately quite rare in the interna-tional system, but it is always costly and sometimes devastating. As a result, poli-cies that can be adopted that reduce the risk of militarized conflict, even slightly,are worth a great deal of attention.

Previous research suggested that defense pacts are effective at deterring theinitiation of militarized disputes (Leeds 2003b). Targets that have allies commit-ted to defend them are less likely to find themselves challenged militarily. In thisstudy, we demonstrate that this deterrence effect of defense pacts is robust overa larger spatial-temporal domain, and it compares well to much more costly pol-icy options for reducing the probability of conflict like increasing one’s militarycapabilities or democratizing one’s own or a rival’s country. Yet, if there are addi-tional effects of alliances that increase incentives for member states to initiate orescalate disputes, our enthusiasm for promoting defense pacts as a policy toolfor peace might be dampened. Before proceeding with confidence to recom-mend defense pacts as a policy tool for peace, we felt a need to evaluate whethersystematic patterns exist that suggest that alliances encourage states to initiate orescalate disputes.

We find no evidence that defense pacts produce either initiation or escalationeffects in our large sample, covering all directed dyad-years from 1816 to 2000.While we are not willing to argue that defense pacts never encourage states toinitiate or escalate disputes, we can conclude that as a general policy, defensepacts seem to have more positive than negative effects. Defense pacts are a usefultool for deterrence, and they do not have consistent side effects that underminepeace.

That being said, we believe future research can do a better job of identifyingthe particular conditions under which defense pacts are most likely to encouragepeace without any unintended side effects that lead to more or more severe con-flicts. First, we believe scholars can specify more clearly the conditions underwhich allies have the ability and willingness to restrain their partners from behav-ing aggressively. We have argued that restraint is most likely when the outside


state has a strong bargaining position in the alliance and would find a conflictcostly relative to the value of what could be gained. These are hypotheses thatcan be tested directly in the future.

Second, and relatedly, we believe that a weakness of our analysis is that it doesnot take into account any differences in characteristics of allies or alliances. Wetreat all defensive allies as equivalent and do not examine the effect of strongerallies, more reliable allies, or allies with stronger interests in the issue under dis-pute. We do not distinguish bilateral and multilateral alliances, alliances with dif-ferent levels of peacetime military coordination or institutionalization, oralliances that have been more durable or more recently formed. One can imag-ine that some of these characteristics of allies and alliances might affect thedeterrence effect of defense pacts and any initiation or escalation effect thatexists in limited circumstances. This is a very challenging task to undertake giventhat many states have multiple allies; attributing behavior to the characteristics ofa particular ally or alliance in a large alliance portfolio is very tricky. We dobelieve it is a worthy goal for another study, however.

Third, our analysis here does not consider the effects of defense pacts on con-flict expansion and diffusion. Existing evidence suggests that states are morelikely to join conflicts involving their allies (for example, Werner and Lemke1997; Huth 1998), that alliances are agents of conflict diffusion (Siverson andStarr 1991), and that disputes involving states with allies are more likely toexpand beyond the original participants (Leeds 2005). In cases in which deter-rence fails and the target resists, past research suggests that disputes involvingstates with allies are less likely to remain bilateral affairs. The effects of allianceson the decisions of states to join conflicts involving their allies and the corre-sponding effects of conflict expansion on conflict severity are worthy of furtherresearch.

Future research should be aimed, therefore, at making our policy prescriptionsmore specific. We believe that this research does provide a helpful first step andclearly suggests that alliances can be tools for peace, a finding that is relevant tosuch current policy issues as the debate over whether NATO membership shouldbe extended to Ukraine and Georgia. Our analysis shows that defense pacts deterthe initiation of disputes against their members and do not seem to have consis-tent effects that encourage members to behave aggressively, either by initiatingnew disputes or by escalating those that are not deterred.

References

Aldrich, Robert. (1996) Greater France: A History of French Overseas Expansion. London: MacMillanPress.

Bennett, D. Scott, and Allan C. Stam. (2000) EUGene: A Conceptual Manual. InternationalInteractions 26 (2): 179–204.

Carter, David B., and Curtis S. Signorino. (2010) Back to the Future: Modeling TimeDependence in Binary Data. Political Analysis 18 (3): 271–292.

Fearon, James. (1995) Rationalist Explanations for War. International Organization 49 (3): 379–414.Fearon, James. (1997) Signaling Foreign Policy Interests: Tying Hands versus Sinking Costs. Journal

of Conflict Resolution 41 (1): 68–90.Gelpi, Christopher. (1999) Alliances as Instruments of Intra-Allied Control. In Imperfect Unions:

Security Institutions over Time and Space, edited by Helga Haftendorn, Robert O. Keohane, andCeleste A. Wallander. New York: Oxford University Press, pp. 107–140.

Ghosn, Faten, Glenn Palmer, and Stuart Bremer. (2004) The MID3 Data Set, 1993–2001: Proce-dures, Coding Rules, and Description. Conflict Management and Peace Science 21 (2): 133–154.

Gibler, Douglas M., and Meredith Sarkees. (2004) Measuring Alliances: The Correlates of WarFormal Interstate Alliance Data set, 1816–2000. Journal of Peace Research 41 (2): 211–222.

Hurst, Michael Ed. (1972) Key Treaties of the Great Powers 1814–1914, Vol. 2. New York: St. Martin’sPress.


Huth, Paul K. (1998) Major Power Intervention in International Crises, 1918–1988. The Journal ofConflict Resolution 42 (6): 744–770.

Kaiser, David. (2000) Politics and War: European Conflict from Philip II to Hitler. Cambridge: HarvardUniversity Press.

King, Gary, Michael Tomz, and Jason Wittenberg. (2000) Making the Most of Statistical Anal-yses: Improving Interpretation and Presentation. American Journal of Political Science 44 (2):347–361.

King, Gary, and Langche Zeng. (2007) When Can History Be Our Guide? The Pitfalls of Counter-factual Inference. International Studies Quarterly 51 (1): 183–210.

Langer, William Leonard. (1929) The Franco-Russian Alliance 1890–1894. Cambridge: HarvardUniversity Press.

Langer, William L. (1951) The Diplomacy of Imperialism 1890–1902. New York: Alfred A. Knopf.Leeds, Brett Ashley. (2003a) Alliance Reliability in Times of War: Explaining State Decisions to

Violate Treaties. International Organization 57 (4): 801–827.Leeds, Brett Ashley. (2003b) Do Alliances Deter Aggression? The Influence of Military Alliances

on the Initiation of Militarized Interstate Disputes. American Journal of Political Science 47 (3):427–439.

Leeds, Brett Ashley. (2005) Alliances and the Expansion and Escalation of Militarized InterstateDisputes. In New Directions for International Relations, edited by Alex Mintz, and Bruce Russett.Lanham: Lexington Books, pp. 117–134.

Leeds, Brett Ashley, Andrew G. Long, and Sara McLaughlin Mitchell. (2000) Re-EvaluatingAlliance Reliability: Specific Threats, Specific Promises. Journal of Conflict Resolution 44 (5): 686–699.

Leeds, Brett Ashley, and Michaela Mattes. (2007) Alliance Politics During the Cold War: Aberra-tion, New World Order, or Continuation of History? Conflict Management and Peace Science 24 (3):183–199.

Leeds, Brett Ashley, Jeffrey M. Ritter, Sara McLaughlin Mitchell, and Andrew G. Long.

(2002) Alliance Treaty Obligations and Provisions. International Interactions 28 (2): 237–260.Maoz, Zeev. 2005. Dyadic MID Dataset (version 2.0). Available at http://psfaculty.ucdavis.edu/

zmaoz/dyadmid.html (Accessed November 12, 2010).Marshall, Monty, and Keith Jaggers. 2002. Polity IV Project: Dataset Users Manual. Available at

http://www.cidcm.umd.edu/inscr/polity/index.htm (Accessed November 12, 2010).Michon, Georges. (1969) The Franco-Russian Alliance 1891–1917. New York: Howard Fertig.Morgan, T. Clifton, and Glenn Palmer. (2003) To Protect and to Serve: Alliances and Foreign

Policy Portfolios. Journal of Conflict Resolution 47 (2): 180–203.Morrow, James D. (1994) Alliances, Credibility, and Peacetime Costs. Journal of Conflict Resolution 38

(2): 270–297.Palmer, Glenn, and T. Clifton Morgan. (2006) A Theory of Foreign Policy. Princeton: Princeton

University Press.Powell, Robert. (1999) In the Shadow of Power: States and Strategies in International Politics. Princeton:

Princeton University Press.Pressman, Jeremy. (2008) Warring Friends: Alliance Restraint in International Politics. Ithaca: Cornell

University Press.Russett, Bruce, and John Oneal. 2001. Triangulating Peace. New York: W. W. Norton.Sartori, Anne E. (2003) An Estimator for Some Binary-Outcome Selection Models without

Exclusion Restrictions. Political Analysis 11 (2): 111–138.Senese, Paul D., and John A. Vasquez. (2008) The Steps to War: An Empirical Study. Princeton:

Princeton University Press.Signorino, Curtis S., and Jeffrey M. Ritter. (1999) Tau-b or Not Tau-b: Measuring the Similarity

of Foreign Policy Positions. International Studies Quarterly 43 (1): 115–144.Singer, J. David, Stuart Bremer, and John Stuckey. (1972) Capability Distribution, Uncertainty,

and Major Power War, 1820–1965. In Peace, War, and Numbers, edited by Bruce Russett. BeverlyHills: Sage, pp. 19–48.

Siverson, Randolph M., and Harvey Starr. (1991) The Diffusion of War. Ann Arbor: University ofMichigan Press.

Smith, Alastair. (1995) Alliance Formation and War. International Studies Quarterly 39 (4): 405–425.Smith, Alastair. (1998) Extended Deterrence and Alliance Formation. International Interactions 24

(4): 315–343.Snyder, Glenn H. (1984) The Security Dilemma in Alliance Politics. World Politics 36 (4): 461–495.Snyder, Glenn H. (1997) Alliance Politics. Ithaca: Cornell University Press.


Taylor, A. J. P. (1954) The Struggle for Mastery in Europe 1848–1918. New York: Oxford UniversityPress.

Werner, Suzanne. (2000) Deterring Intervention: The Stakes of War and Third-Party Involvement.American Journal of Political Science 44 (4): 720–732.

Werner, Suzanne, and Douglas Lemke. (1997) Opposites Do Not Attract: The Impact of DomesticInstitutions, Power, and Prior Commitments on Alignment Choices. International Studies Quarterly41 (3): 529–546.

Yuen, Amy. (2009) Target Concessions in the Shadow of Intervention. Journal of Conflict Resolution 53(5): 745–773.


Defense Pacts

Documents

good policy prescription

defensive allies

good policy option

anincreased probability

likely torespond

stronger states

theinitiation of disputes

democratic states