Military alliances have long been recognized as an important policy strategy for states to counter international threats. Rather than relying only on one's own military capabilities to deter threats, states can band together and aggregate their capabilities, thus increasing their deterrent potential and gaining from economies of scale in the provision of defense. Even in cases of issue linkage, in which one state may offer policy concessions in return for a promise of military support, at least one party is motivated by the desire to respond to military threat.
Empirical studies, however, have not consistently found evidence in support of this basic theoretical proposition.Footnote 1 In part, this may be explained by the difficulty in measuring external threat. It may also be explained, however, by a failure to recognize that many threats are not external to the state. States have long used alliances not only to respond to threats from other states, but to respond to threats from internal and transnational groups. As far back as 1833, for instance, Prussia, Russia, and Austria signed a set of agreements including promises of support against “revolutionaries,” particularly in the Polish provinces.Footnote 2
Not all alliances include explicit promises to intervene militarily if one or more parties is attacked. Instead, many alliances include only promises to consult and coordinate policy with one's allies in the face of a threat or to refrain from participating in or aiding attacks on one's allies. Many alliances even combine these different obligations, which suggests they each serve a particular purpose. The prevalence of consultation pacts that promise communication and policy coordination has been a puzzle for alliance theorists,Footnote 3 since it would appear these weaker agreements are less powerful deterrents than defense commitments. Thus far, explanations for consultation pacts have focused on domestic political considerations.Footnote 4 We suggest an additional possibility that relates to the leader's threat environment.
We argue that alliance formation is motivated both by external threat from other states and also by internal threats that make civil conflict more likely. We also argue that different threats motivate different alliance obligations. While external threats are most related to the formation of alliances that include obligations for defense, internal threats are related to the formation of flexible pacts including consultation obligations that could lead to a variety of different forms of support. Additionally, internal threats with the greatest potential to become internationalized are related to the formation of alliances with neutrality and/or nonaggression promises, which make it more difficult for opposition groups to gain international support. States create security portfolios that include alliances best suited to respond to the different types of threats that they face.
To evaluate this argument, we examine a broad sample of countries from 1946 to 2009. To test our claims about the threat of international conflict and defense pacts, we use a measure of external threat that accounts for the military capabilities of countries that, based on their location, status, and policy positions, are most likely to be perceived as threatening by the state in question. To test our claims about the threat of civil conflict, we focus on a specific form of internal threat: the existence of politically relevant groups excluded from state power. Political exclusion has been shown to have a robust relationship to the outbreak of civil war,Footnote 5 and scholars have also found that civil conflict is more likely to become internationalized when such groups have connections to transnational ethnic kin.Footnote 6 Thus, according to our theory, the presence of excluded groups should lead states to seek allies, and excluded groups with links to transnational ethnic kin should cause leaders to seek to limit the international support available to their internal rivals. Our analysis finds support for all our hypotheses.
In the twenty-first century, states continue to plan for interstate conflict and protect themselves accordingly, but many also feel acute threats from other sources, both internally and transnationally, and are creating security policies that are flexible enough to address these multiple threats. Our understanding of security policy must evolve to account for this. We demonstrate here that a more complete understanding of alliance formation and design requires taking into account these multiple sources of threat.
Alliances As a Response to External and Internal Threat
An alliance is a formal agreement among independent states to cooperate militarily in the event of conflict. A subset of alliances, defense pacts, require member states to assist their partners militarily in the event of a threat to a member's sovereignty or territorial integrity. Consultation pacts require member states to communicate and work toward coordinating their responses to crises with the potential to result in military conflict, but they do not specifically commit members to active military support. Neutrality and nonaggression pacts require member states to refrain from military conflict with one another, either by denying assistance to an ally's adversary or forgoing direct attacks on an ally, respectively. When states form alliances, they may choose to include one or more of these different obligations. Out of 745 alliances recorded in version 4.01 of the Alliance Treaty Obligations and Provisions (ATOP) data set, 351 (47%) include two or more types of commitments. States often have multiple alliances at the same time; in the period from 1815 to 2016 the average number of alliances that a state holds in a given year is 2.7, and the maximum number is fifty-three.Footnote 7
Traditionally, we have assumed that most threats to sovereignty and territorial integrity derive from other states in the international system, and thus that the purpose of alliances in general and defense pacts in particular is to increase the power of a member state's deterrent threat. A strong predictor of alliance formation, therefore, should be international threat. Yet many studies have had trouble demonstrating this relationship consistently.Footnote 8 This may be in part due to the fact that international threats are not the only threats leaders consider when forming alliances. Barnett and Levy argue, for example, that threats to domestic political stability motivate states to seek security cooperation.Footnote 9
When states face threats from other states, the logic behind a defense pact is clear. Because alliances are costly to form, maintain, and break, allies tend to support one another if war occurs,Footnote 10 and because challengers prefer to target adversaries who will not receive outside assistance, defense pacts deter aggression.Footnote 11 While many scholars have criticized the so-called capability aggregation model of alliances, primarily by pointing out that strong states often make defense commitments to weaker states who do not improve their security and that, in some cases, states trade security promises for concessions on other issues,Footnote 12 even when alliance formation is accomplished through issue linkage, the core motivation for at least one state must be to enhance its ability to deter threats. Otherwise, states would form agreements that did not promise military support. Johnson demonstrates that states that feel more threatened are willing to offer greater concessions to obtain an ally.Footnote 13
Defense pacts may not be the most appropriate response to threats from domestic groups. Defense pacts often specifically limit military assistance to cases of external aggression, making them inapplicable to internal security threats. For example, the 1992 Tashkent Treaty states that “in case one of [the] States Parties is subjected to an aggression by any state or a group of states, this shall be considered as an aggression against all the States Parties of the present Treaty.”Footnote 14 When Kyrgyzstan requested assistance from Russian troops to quell ethnic violence in June 2010, Russia refused, arguing the pact was created to deal with external aggression, not internal conflict.Footnote 15
Although defense obligations are occasionally worded broadly enough to allow for activation under conditions of internal threat, states may be wary of such commitments for fear that this would invite foreign interference in their domestic affairs or jeopardize their claims to legitimacy. Scholars have found only mixed evidence that military interventions on behalf of the incumbent increase the odds of government victory in civil conflict. On the contrary, direct military intervention can undermine the government's legitimacy by calling into question its ability to govern independently, sparking a nationalist backlash that unifies and emboldens the rebels and encourages more extreme tactics.Footnote 16 Foreign troops may also be less effective at countering guerilla forces because they are not trained to fight such unconventional campaigns and lack the local knowledge critical to identifying insurgents.Footnote 17 Hence, while sometimes leaders may invite their allies’ direct military intervention in civil conflicts, they may want considerable control over when and if allied troops arrive and prefer more subtle forms of support that preserve their legitimacy and claims to independence.
Alliances that include consultation obligations seem especially appropriate to these goals. Consultation obligations commit members to consult with an eye toward coordinated action during crises that could result in military conflict. This action can take a variety of forms, including those that would be particularly useful to states facing internal threats, such as information exchange, arms provisions, military training, assistance in preventing groups from establishing bases in neighboring territories, and prevention of funding and the transport of provisions to enemy nonstate actors. For instance, a 1947 consultation obligation between Iraq and Jordan spells out “If circumstances should necessitate the adoption of joint measures or actions to suppress disturbances or disorders, the two [parties] shall consult with each other concerning the policy of co-operation which shall be followed for this purpose.”Footnote 18
It may also be the case that the actions a leader would like support for with regard to internal enemies are not defensive, but offensive. Offense pacts, at least publicly revealed ones, have all but disappeared from the landscape of international politics. According to the ATOP data set, version 4.01, there have been no alliances that obligate states to assist one another in an offensive conflict (one that does not involve an attack on the territorial integrity or sovereignty of the member states) since 1990.Footnote 19 We seem to have developed a norm that prevents states from declaring openly aggressive intentions that cannot be framed in a defensive manner. As a result, states with intentions to engage in conflict that is not motivated primarily by defense may create more flexible commitments that do not restrict support to a defensive conflict. Consultation obligations again fit this requirement.
In 1977, for example, Mozambique signed a consultation pact with the Soviet Union. While the government of Mozambique was ideologically compatible with the Soviet bloc, it had been committed to pursuing its own path.Footnote 20 As a result, the western media was surprised by this commitment.Footnote 21 In 1979, a civil war officially began between the government and the RENAMO rebel group. The USSR provided significant support through training, aid, equipment, and military advisors.Footnote 22
In some cases, a goal of a government facing internal threat is to obtain support from allies in deterring and/or fighting rebels, and we have argued that consultation pacts are a useful tool for accomplishing this goal. Leaders may also wish to use alliances to deny international support to a rebel group by forming nonaggression and/or neutrality pacts. Rebels may have incentives to seek state sponsors, and other states may have incentives to undermine the sovereignty and internal control of a government.Footnote 23 Nonaggression pacts include promises not to use force against other alliance members, and neutrality pacts require a broader commitment to take no action that advantages a partner's adversaries. During the Mozambique civil war in 1984, Mozambique and South Africa signed the Nkomati Accord in which the two sides agreed not to support one another's rebels, with extensive detail of actions that the governments would and would not take.Footnote 24 Sudan and South Sudan signed a similar agreement in 2012. Such pacts seek to deny support to nonstate actors who might conduct their operations or receive support from neighboring states.
A variety of circumstances might lead a state facing internal threat to seek to form alliances including nonaggression and/or neutrality provisions. First, a state might seek such assurances from a state that might be a natural ally of the rebel group, for instance, because it shares ethnic characteristics or ideological commitments. While another state government might have preferences aligned with a potential rebel group, that state might also prefer regional stability and be willing to make a commitment not to support a rebel group in order to obtain a reciprocal commitment or in order to benefit from the advantages associated with peace. This is what likely motivated the Nkomati Accord. Second, a state might seek such assurances from a state that is facing similar potential challenges from internal opposition groups to prevent collaboration between their rebel groups. Third, a state might be motivated to form nonaggression and neutrality pacts with nearby states to prevent rebels from forming outposts and supply chains through proximate territories.
Internal conflict can escalate to interstate war if third parties intervene militarily to help rebels or if the government retaliates against foreign sources of support. In fact, many militarized interstate disputes arise from issues related to civil war through these mechanisms of intervention and externalization.Footnote 25 Jenne shows that the potential for external support emboldens rebels and affects the demands that potential rebel groups make.Footnote 26 We argue that a key goal of leaders when they anticipate internal threat is to try to avoid internationalization by breaking the connections between foreign governments and rebel groups—that is, to keep the threat internal and avoid creating new external threats. Thus, when a leader fears internal threats from groups deemed likely to receive international support, he or she should be especially motivated to add alliances with neutrality and/or nonaggression obligations to the state's security portfolio.
In sum, we contend that a more complete explanation of alliance formation and the full content of state security portfolios requires that we take into account both external and internal sources of threat. States may form multiple alliances or alliances with multiple obligations to deal with different threats. We argue that the goal of alliances is sometimes to gain direct support in limiting internal threats and to prevent those internal threats from becoming internationalized. We also argue that different threats lead to different types of alliance obligations, helping to resolve a long-standing puzzle.Footnote 27 While states facing external threats will be likely to form alliances with defense obligations, states facing internal threats will be likely to form alliances including consultation obligations. Moreover, states facing internal threats with high potential for internationalization will be more likely to form alliances promising neutrality and/or nonaggression than those not facing such threats. The question then becomes how can we identify instances in which state leaders believe the potential for conflict is high and design their security policies accordingly?
Identifying Threat
A major challenge for evaluating our argument is that to do so requires measuring external and internal threat. Alliances will sometimes be formed in anticipation of future conflict, even when no militarization has yet occurred, and if the alliance is successful at deterring adversaries, we may never observe overt conflict. As such, any measure of threat that depends upon observing militarized engagement is insufficient; we need to include threats that could have become militarized and did not.Footnote 28
We measure external threat based on a combination of power and interests rather than on past history of militarized disputes. We use an updated version of the measure developed by Leeds and Savun,Footnote 29 which is a sum of the composite index of national capabilities (CINC) of all system members in a state's threat environment. A system member is considered to be in a state's threat environment if it is not an ally and is either a rival, contiguous neighbor with a dissimilar foreign policy orientation, or a major power with a dissimilar foreign policy orientation. Hence, this measure captures the factors that are likely to precipitate international conflict.
To identify states that face internal threat, we rely on research that demonstrates that the presence of politically excluded groups increases the probability of civil conflict.Footnote 30 Political exclusion makes civil war more likely through the grievances it creates. Political inequalities between groups generate dissatisfaction with the status quo and resentment toward the incumbent government. These inequalities can then be politicized and used to mobilize excluded groups to take up arms against the government, producing an increased probability of civil conflict. Because such grievances tend to be more intense for larger excluded groups, the threat of internal conflict should also increase with the relative size of excluded groups.
Excluded groups are certainly not the only factor that generates internal threat, but we focus on them here for two reasons. First, the empirical relationship between excluded groups and civil war onset has proven to be quite robust. Studies show that the relationship holds for different model specificationsFootnote 31 and have found support for additional implications of the theoretical argument.Footnote 32 Scholars have even found support for the relationship employing an instrumental variable approach designed to address endogeneity, increasing confidence that political exclusion is causally linked to civil war onset.Footnote 33 Additionally, several qualitative accounts provide further credibility to the finding.Footnote 34
Second, factors that generate internal threat through other pathways, such as opportunity or greed, can affect alliance formation through channels other than internal threat. For example, per capita income, a robust correlate of civil war onset, can contribute to alliance formation since richer countries with more capabilities make more desirable alliance partners and poorer countries less capable of dealing with threats may seek out more alliance partners. Oil production, another common variable used to explain civil war onset, can make a state attractive to defend or obtain concessions from via an alliance. Thus, many of the variables employed in the civil war literature could influence alliance formation through alternate channels, making their coefficients difficult to interpret.Footnote 35 Thus, while our argument is broader—we argue that any threat of civil war encourages these policy responses—we test our argument focusing on indicators for the potential for ethnic conflict.
To identify internal threats that pose the greatest potential for internationalization, we focus on excluded groups with ethnic ties to groups in neighboring states. The presence of excluded groups with transborder ethnic kin (TEK) increases the probability of civil conflict because TEK can enable the excluded group to initiate conflict by providing information, military support, external bases, and supply chains.Footnote 36 Studies also find that rebels with transnational constitutencies are more likely to attract external support and the presence of ethnic kin contributes to state motivations to intervene in an ongoing conflict.Footnote 37 However, as Cederman, Gleditsch, and Buhaug suggested,Footnote 38 only those TEK groups large enough to enhance the rebels’ capabilities and improve the group's chances of winning a war will be able to provide significant support.Footnote 39 Thus, we expect states facing internal threats from excluded groups with relatively large TEK to have greater incentives to form alliances with neutrality and/or nonaggression obligations compared to states that do not face such threats. Again, any potential for internationalization of internal conflict should make nonaggression/neutrality pacts an attractive policy option. We focus on ethnic ties as one indicator of this potential.
Note that we focus here on the motivation of a state to seek an ally—the demand side of alliance formation. We do not explicitly theorize about why outside states are willing to make these commitments. We can speculate about reasons they might—reciprocal promises to help in their own potential internal conflicts, an affinity for the current government, a desire to promote regional stability, receipt of policy concessions or side payments, or influence over the partner government's conflict behavior. Both theoretically and in our research design, however, we take a monadic approach. We assume that those states who want to find allies will be more likely to do so—even if it means they need to offer policy concessions in return—than those who do not see a strong need for an alliance. A monadic research design prevents us from analyzing some interesting questions, namely what sort of allies a state might seek, and what states might be willing to provide the desired alliance under what conditions. Our theory leads to some expectations about these factors, and we imagine this as a profitable direction for future research. Our primary goal here, however, is to understand how an individual state builds a security portfolio rather than to understand the state's relationship with particular external partners. Because allies may be substitutable for one another or complementary to one another, focusing on the probability of dyadic alliance can obscure important monadic tendencies by dramatically increasing the sample size and reducing the independence of observations. While dyadic analyses bring benefits, they also have costs. Thus, in this paper we focus on when states seek allies, and we leave the question of who they choose as their allies for future work. We focus on the following hypotheses:
H1 States facing higher levels of external threat are more likely to form alliances with defense obligations than states facing lower levels of external threat.
H2 States with at least one politically excluded group will be more likely to form alliances with consultation obligations than states without any politically excluded groups; this probability should increase for larger excluded groups.
H3 States with at least one politically excluded group with links to a relatively large TEK group will be more likely to form alliances with neutrality and/or nonaggression obligations than states without any politically excluded groups.
Research Design
To test our hypotheses, we utilize a country-year data set and three dependent variables. Our spatial-temporal domain covers 180 countries from 1946 to 2009. Each dependent variable codes whether the state entered into one of the different types of alliance obligations outlined in our hypotheses in the observation year.Footnote 40 Because our theory is about motivation to form alliances, we concentrate on the initial formation decision.Footnote 41 Data for all of the dependent variables are obtained from the Alliance Treaty Obligations and Provisions (ATOP) data set, version 4.01,Footnote 42 which provides detailed information about the obligations in all alliances formed between 1815 and 2016. Importantly for our theory, ATOP does not code only the “highest” obligation of a given alliance, but codes each obligation for every alliance agreement. Sometimes states might use the same alliance agreement to respond to different threats, and in other cases they may create a security portfolio that includes multiple alliances that are complementary to one another.
The first dependent variable codes whether the state entered into an alliance with a defense obligation in the observation year. A defense obligation provides the state with a promise of military assistance from its partner(s) in the event of an attack on the state's sovereignty or territorial integrity.Footnote 43 In 415 country-years in our sample (5%), a state forms an alliance with a defense obligation.
The second dependent variable codes whether the state entered into an alliance with a consultation obligation in the observation year. A consultation commitment is a promise of communication and policy coordination in the event of a crisis with potential for military conflict. States form alliances with consultation obligations just as frequently as alliances with defense obligations. In 510 country-years in our sample (6%), a state forms an alliance with a consultation obligation.
The final dependent variable codes whether the state entered into an alliance with a neutrality or nonaggression obligation in the observation year. Neutrality and nonaggression obligations are similar in that they require the members’ promise not to attack one another. A neutrality obligation, however, requires the members to stay out of conflicts involving their partners and refrain from aiding one another's adversaries.Footnote 44 Neutrality/nonaggression is the most common type of alliance formation in our sample. In 736 country-years (8%) a state forms an alliance with a neutrality or nonaggression obligation.
The key independent variables in our analysis are our measures of external and internal threat. As discussed before, we measure external threat by calculating the sum of the composite index of national capabilities (CINC) of all system members in a state's threat environment at the start of the observation year. A state's threat environment includes all system members that are not allies of the state and are either rivals, contiguous neighbors with a dissimilar foreign policy orientation, or major powers with a dissimilar foreign policy orientation. Allies are identified through the ATOP data set,Footnote 45 and rivals are identified using Thompson and Dreyer's perception-based rivalry measure.Footnote 46 We consider pairs of states with below average S-scores to have dissimilar foreign policy orientations. S-scores are obtained from Chiba, Johnson, and Leeds’ ATOP-based scores.Footnote 47 The resulting variable ranges from 0 to .7 with an average of .31.
The other key variables in our analysis are the indicators of internal threat. We consider internal threat to be present in states with politically excluded groups. The basis for this variable comes from work by Cederman, Gleditsch, and Buhaug and the Ethnic Power-Relations (EPR) Dataset, version 2019.Footnote 48 They code the level of access to state power for each ethnic group within a state.Footnote 49 A group is considered excluded if it has “no regular representation in the executive.”Footnote 50 The most obvious example of this would be a group subject to targeted discrimination with limited political rights, such as the Kurds in Turkey. However, Cederman, Gleditsch, and Buhaug also identify groups blocked through more informal measures, such as Israeli Palestinians.Footnote 51 They also consider groups that have been excluded from the executive but maintain some degree of autonomy separate from the state as excluded.
To code our independent variables, we use the EPR Dataset to identify whether there was any excluded group in the state at the start of the observation year.Footnote 52 We then code whether a large excluded group was present. Following Cederman, Gleditsch, and Buhaug and Buhaug, Cederman, and Gleditsch,Footnote 53 we code the relative size of each excluded group by dividing the population of the excluded group by the sum of the population of the excluded group and the population of the group that holds executive power.Footnote 54 We consider an excluded group to be large when this ratio is above the median, which is approximately 0.04.Footnote 55 Country-years with at least one excluded group above this threshold are considered to have a large excluded group and country-years with excluded groups with populations below this threshold are coded as having small excluded groups.Footnote 56 There are 1,401 (16%) country-years that contain small excluded groups and 4,649 (54%) that contain large excluded groups.Footnote 57
In addition to identifying country-years with excluded groups, we also identify countries that exclude groups with transborder ethnic kin (TEK). As discussed earlier, these excluded groups should prompt the formation of alliances with neutrality/nonaggression obligations. However, we expect this to be the case only if the TEK group is large relative to the group in power in the excluded group's state. Relatively small TEK groups will not pose a significant-enough threat to warrant the costs of alliance formation.
We identify this subset of excluded groups using the EPR data,Footnote 58 which identifies all ethnic groups that are in multiple countries and code whether a given excluded group has TEK. The relative size of the TEK group is determined by the population of the TEK group divided by the sum of the population of the TEK group and the population of the group that holds executive power in the excluded group's country.Footnote 59 We consider the group to be large when this ratio is above the median, which is approximately 0.34.Footnote 60 We found that 2,956 (34%) country-years contain an excluded group that has a relatively large TEK group.
Throughout our analysis, we include several control variables that may influence both alliance formation and the primary independent variable. We first control for the status of the state in the international system. Major powers are more likely to form alliances because they have more global interests and are more desirable partners because of their greater capabilities. Importantly, major powers also tend to experience fewer civil conflicts as a result of their relative strength and stability. Thus, we include a dummy variable that codes whether the Correlates of War project considered the state to be a major power in the observation year.Footnote 61
We also account for a state's number of neighbors. States with more neighbors tend to have more alliance partners because their shared borders create greater opportunities for external threats. They also have more potential partners that they can easily coordinate with militarily. States with multiple borders also have greater potential for civil conflict because of the risk of civil conflicts spilling across borders.Footnote 62 To code this variable we count the number of land borders a state has in a given year according to the COW project.Footnote 63
Finally, we control for any existing alliances a state may have. Because states work to build a security portfolio, alliances can be substitutable for one another, or they can be complementary. We expect that states consider their existing alliances when making decisions about forming new alliances. We include a separate control variable for each type of alliance analyzed. That is, we code whether the state has an existing alliance with defense, consultation, neutrality, or nonaggression provisions according to the ATOP data.Footnote 64
Results
To test the hypotheses we estimate three logit models with robust standard errors, one for each dependent variable.Footnote 65 Table 1 reports the results and substantive effects for our key independent variables are reported in Figures 1 to 3. All of the figures, first differences, and percentage changes were generated using the procedures for obtaining predictions and uncertainty estimates developed by King, Tomz, and Wittenberg.Footnote 66
Predicted probabilities of alliance formation with defense obligations
Predicted probabilities of alliance formation with consultation obligations
Predicted probabilities of alliance formation with neutrality/nonaggression obligations
Logit regressions of alliance formation on threat, 1946 to 2009
Note: Standard errors in parentheses; * p < 0.05, ** p < .01.
Model 1 in Table 1 tests Hypothesis 1. The key variable in this model is the indicator of external threat. The model shows that external threat increases the likelihood that a state forms an alliance with a defense obligation. This relationship can be seen clearly in Figure 1, which plots the predicted probability and 95 percent confidence interval of forming an alliance with a defense commitment across the levels of external threat observed in our data. The figure demonstrates that there is a positive and substantively meaningful relationship between external threat and defense commitments. Increasing external threat by one standard deviation from the mean produces a 59 percent [42, 77] increase in the probability of defense pact formation. This offers strong support for Hypothesis 1.
Moreover, models 2 and 3 show that external threat does not increase the probability that states form alliances with other types of obligations. In fact, the results suggest that external threat is unrelated to the formation of consultation pacts and decreases the probability a state forms an alliance with neutrality/nonaggression obligations. An exclusive focus on external threat may prevent scholars from understanding the full range of alliance obligations.
Model 2 suggests that a focus on internal threat helps us to understand the formation of alliances with consultation obligations. Countries excluding groups from political power are more likely to form alliances that include consultation obligations. Figure 2 reports the predicted probability and 95 percent confidence interval of an alliance formation with consultation obligations for a country with no excluded group, a small excluded group, and a large excluded group. The figure indicates that the presence of an excluded group of any size increases the probability of forming a consultation pact. Compared to no excluded group, the presence of a small excluded group increases the probability of forming a consultation pact by approximately 54 percent (15, 98) and a large excluded group increases the probability of forming a consultation pact by approximately 68 percent (31, 109). This provides strong support for the claim that excluded groups encourage states to pursue consultation pacts. Moreover, the effects are comparable to the impact of external threat on defense pact formation presented in model 1. However, as model 1 shows, the presence of an excluded group, small or large, is not clearly related to defense commitments.Footnote 67 Thus, consistent with our argument, the results indicate that different types of threats motivate different types of alliances.
We find no evidence that larger excluded groups have a bigger effect; first differences indicate that small and large groups have a similar impact on consultation formation. It seems that the desire to prepare for and/or deter internal conflict is sufficient even with small excluded groups for states to seek consultation pacts. While there might be other additional actions that a government will take as the risk increases, this is a security policy that comes into play even with small excluded groups. We find this plausible because not only are there many cases of civil war with small excluded groups, but small excluded groups can also engage in other activities short of civil war, for instance, terrorist acts, that will increase the sense of internal threat that state leaders feel and potentially encourage them to seek support from allies. For dealing with threats short of civil war, troops on the ground are even less useful, and thus the focus on consultation pacts, which can facilitate information exchange, cooperation in surveillance, and supply of equipment and training, is particularly relevant.
Model 3 tests Hypothesis 3 by disaggregating excluded groups into those with large TEK, small TEK, and no TEK and analyzing their impact on the formation of alliances with neutrality/nonaggression obligations. Country-years without any excluded groups are the comparison group for this analysis. Figure 3 reports the predicted probability of formation of an alliance with neutrality/nonaggression obligations and 95 percent confidence interval for each scenario. An examination of first differences demonstrates that the predicted probability of neutrality/nonaggression formation is significantly higher in the presence of an excluded group with large TEK than the other scenarios, but the other scenarios are not distinguishable from one another. Thus, the results indicate that the presence of excluded groups with large TEK encourages states to form alliances with neutrality/nonaggression obligations, but having excluded groups without large TEK does not show a clear relationship with the formation of neutrality/nonaggression pacts—only those internal threats with the greatest potential for internationalization encourage the formation of alliances with neutrality/nonaggression obligations.Footnote 68 Figure 3 shows that excluded groups with large TEK increase the probability of forming an alliance with a neutrality/nonaggression obligation by approximately 65 percent (34, 102) relative to no excluded group, 36 percent (4, 78) relative to excluded groups with no TEK, and 32 percent (4, 93) relative to excluded groups with small TEK. Similar to the substantive effect of excluded groups on consultation obligations, these results are comparable in magnitude to the substantive effect of external threat on defense commitment formation.
The control variables largely perform as expected. Major powers and states with more neighbors are more likely to form all types of alliances. With regard to existing alliances, an existing defense pact is associated with a lower probability of consultation-pact formation, an existing neutrality/nonaggression pact is associated with a lower probability of forming all types of alliance, and an existing consultation pact is related to the formation of additional defense and consultation pacts; alliances can be substitutable or complementary under different circumstances.
In sum, our results suggest that to understand the variety of alliances in the international system and how states build their individual security portfolios, one must consider internal as well as external threats. External threat and existing alliance theory is helpful for understanding the formation of defense pacts, but the formation of other types of alliances, such as consultation and neutrality/nonaggression pacts, is better understood once we also incorporate factors related to potential domestic conflict.
Conclusion
Military alliances play a central role in international relations and are often understood as responses to threat. We argue that leaders may seek alliances in response to threats from other states and from groups within their polity. Taking into account threats from nonstate actors helps us to understand the broader portfolio of security policies that states adopt. While states facing higher levels of external threat are more likely to form alliances with defense obligations, states with excluded groups are more likely to form alliances with consultation obligations, which offer the potential for flexible, more varied forms of support in different situations. States anticipating the possibility of internationalized civil conflict also tend to form alliances with neutrality and/or nonaggression obligations. An empirical analysis of alliance formation from 1946 to 2009 supports these claims.
Our argument has implications for the study of both alliances and civil war and highlights several promising directions for future research. Alliance scholars have found consultation pacts that exclude defense obligations confusing because they seem to offer “half-hearted commitments.” The existing explanations for the formation of consultation pacts tend to focus on domestic political dynamics.Footnote 69 Our argument suggests another possible explanation—that the threats these alliances are aimed at may make different forms of assistance and flexibility in future forms of coordination valuable.
Additionally, a large body of research investigates the deterrence effect of alliances on international conflict.Footnote 70 We suggest an additional focus should be on the potential for deterring civil conflicts. Governments that secure allies may be less likely to experience armed rebellion. Cunningham shows this dynamic occurs for states in hierarchical relationships with the United States,Footnote 71 and our work suggests existing alliance data can be used to extend these analyses to non-US cases.
Our research also suggests that previous research may underestimate the pacifying effect of nonaggression pacts. Mattes and Vonnahme find that nonaggression pacts reduce conflict among signatories.Footnote 72 Our argument suggests that nonaggression pacts may also limit civil conflict. If domestic groups believe they are less likely to receive outside support, they may be less likely to rebel; nonaggression pacts may reduce both international and civil conflicts. This also highlights an overlooked and potentially normatively unappealing feature of alliances; they make it harder for excluded groups to challenge their political status.Footnote 73
Our research also connects to a growing literature on rebel diplomacy that highlights how rebel groups seek out support for their campaigns.Footnote 74 Given our focus on how governments respond to potential conflicts with politically excluded groups and TEK, future research should investigate how rebels’ ability to garner international support is affected by alliance networks.
Finally, our research contributes to ongoing efforts to understand the path from civil to interstate conflict. Previous work finds that allies of a government involved in civil war are more likely to intervene on the government's behalf compared to other potential interveners,Footnote 75 but these works treat existing alliances as indications that a potential intervener has a stake in the outcome rather than as a treaty designed at least in part to deal with internal threats. Our results suggest more work should be done to examine the link between existing alliances and intervention in civil wars.
While interstate conflict clearly remains a concern of state leaders in the twenty-first century, increasingly security policies are being designed around concerns about threats from nonstate actors, both domestic and transnational. As a result, our research on such important policy matters as defense spending, weapons technology, and grand strategies must take into account more than traditional state-to-state conflict. This research is a step toward doing so and demonstrates that we increase our understanding when we consider both external and internal threat as motivations for security policy.
Data Availability Statement
Replication files for this research note may be found at <https://doi.org/10.7910/DVN/5MP5RJ>.
Supplementary Material
Supplementary material for this research note is available at <https://doi.org/10.1017/S0020818321000151>.
Acknowledgments
A previous version of this paper was presented at the annual meeting of the International Studies Association in Toronto, 27–30 March 2019. We thank Kathleen Cunningham, Daina Chiba, Matt Fuhrmann, Michaela Mattes, Eric Min, Sara Polo, Liana Eustacia Reyes, and Clayton Thyne for advice, and Lars-Erik Cederman and Luc Girardin for providing data.
