National Leaders, Political Security, and International Military Coalitions

Transcription

1 National Leaders, Political Security, and International Military Coalitions Scott Wolford University of Texas at Austin Emily Hencken Ritter University of California, Merced Abstract Building coalitions in international crises improves military prospects but requires policy concessions to compensate partners, and we show that a national leader s job security affects her willingness to make this tradeoff. First, politically insecure leaders are more willing to form coalitions than secure leaders, making concessions to partners to bolster their chances of victory so that they may improve their chances of surviving defeats in office. Second, politically insecure leaders are also less selective in their choice of partner, as their willingness to make larger policy concessions leads them to form coalitions with states of increasingly divergent foreign policy preferences. We find empirical support for these hypotheses in a dataset of leader-partner crisis dyads from Draft version; comments welcome, but please do not cite without permission. Previous versions of this paper were presented at the Annual Meeting of the Midwest Political Science Association, 2011, Chicago, IL, the Annual Meeting of the International Studies Association, 2010, Montreal, QC, the Annual Meeting of the Southern Political Science Association, 2012, New Orleans, LA, the 2013 Texas Triangle Conference at Texas A&M University, and the Princeton University Conference on Theoretical and Quantitative Studies in International Relations, May Thanks to Amber Curtis for research assistance and to Phil Arena, Bear Braumoeller, Giacomo Chiozza, Justin Esarey, Ben Fordham, Hein Goemans, Ashley Leeds, and Clifton Morgan for helpful comments and suggestions. Parts of the data collection were supported by a Faculty Research Grant from the Colorado European Union Center of Excellence at the University of Colorado.

2 When do national leaders build military coalitions, and with whom do they choose to partner when they do? Coalitions have been at the center of some of the most visible and consequential conflicts of the last six decades, from Korea in the 1950s to Bosnia to Kosovo in the 1990s to two American-led wars against Iraq, and in general multilateral wars are longer, bloodier, and more destructive than bilateral conflicts. When a coalition forms in a particular crisis, at least two states make joint threats of war should their demands go unmet, but despite the ubiquity and infamy of military coalitions, we know little about the crisis-specific conditions that lead particular states to enter into these cooperative arrangements. Most explanations for military cooperation are cast at the international level, often through the study of alliances, in which states make formal, ex ante commitments to cooperate during future hostilities. Long-term or structural factors like military power (Morrow 1994), domestic institutions (Gibler and Wolford 2006, Lai and Reiter 2000), historical experience (Reiter 1996), and geopolitical interests (Gibler and Rider 2004, Morrow 1991) may be important for the choice of allies prior to conflict. In fact, only about 75% of coalitions since 1946 involve nonallied states (Wolford 2014), suggesting that the processes by which states build alliances and crisis-specific coalitions are quite distinct. Once embroiled in a crisis, short term factors are likely to play a role, from operational needs (Kreps 2011) to the support of international organizations (Chapman 2011, Tago 2007) to domestic politics, from economic concerns to a leader s political security in office. In particular, the expected outcome of an interstate crisis can affect an incumbent leader s ability to remain in political power, putting him at risk of replacement in the wake of defeat. In this paper, we ask how an incumbent s political security affects decisions over whether and with whom to build coalitions in international crises. We analyze a theoretical model in which a national leader, concerned over both the national 1

3 interest and her own political survival, must choose whether to act alone in an international crisis or to propose a coalition to a potential partner. Taking on a partner improves her chances of success in the crisis, enabling her to avoid defeats that threaten her hold on power (cf. Richards et al. 1993, Smith 1996, 1998), but it also requires that she compensate the partner for its cooperation in the costly endeavors of crisis bargaining and, potentially, war. These foreign policy concerns weigh against her domestic political situation; a leader who is secure in power can more easily weather defeats in crises and hold on to her position than a leader made less secure by her domestic prospects. Faced with this tradeoff, we show that politically insecure leaders are both more likely to build coalitions than their secure counterparts and less selective in their choice of partner, willing to take on more diverse that is, more expensive partners as they become increasingly desperate to bolster their chances of political survival. We use new data on coalition participation in international crises from 1951 to 2001 (Wolford 2014) to assess our predictions over coalition formation, showing that the decision to build coalitions and the choice of partner are both conditioned by a leader s job security. Specifically, we show that politically secure incumbents are less likely to take on coalition partners than their insecure counterparts, who are uniquely likely to build coalitions. We also show that Insecure leaders are also more willing to accept partners with foreign policy preferences that diverge from their own. Thus, in contrast to other theories that link political insecurity with the pursuit of private over public interests (Downs and Rocke 1994, Smith 1996, 1998, Tarar 2006, Young 2009), we link it to what the public at large and many scholars consider desirable policy forming coalitions to achieve foreign policy outcomes rather than going it alone (Naím 2009, Nye 2002, Pouliot 2011). More broadly, this implies an association between autocratic leaders, who are generally secure in office, and unilateral action, as well as a similar relationship 2

4 between popular or otherwise domestically secure democratic leaders and unilateralism. The most popular leaders in democracies should be those least likely to seek out partners when they confront international opponents. As debate continues over the international costs and benefits of American multilateralism in the academic literature and in popular discourse (Naím 2009, Nye 2002, Tucker 1999), it may be equally important to consider how domestic political incentives may work against a given conception of the national interest. Military Coalitions and Domestic Politics Leaders who would build coalitions must ask themselves two important questions: with whom do they wish to cooperate (if anyone), and does this cooperation come at an acceptable price? Short of a perfect alignment of preferences, answering the second question in the affirmative is not easy: states can differ in their valuation of the status quo or the stakes of the crisis, willingness to bear costs and tolerate risks, as well as preferred bargaining and escalation strategies. Just as they do when considering potential allies, coalition-builders consider both military capabilities and other contributions like staging or transit rights when identifying potential partners (see Morrow 1991, Snyder 1997), but ensuring cooperation in costly endeavors like crisis bargaining and war may require side payments and/or policy concessions that they would, all else equal, prefer not to make (Kreps 2011, Papayoanou 1997, Riker 1962). In this section we discuss the goods traded in such intra-coalitional bargaining, and we consider how a leader s private interest in political survival might condition her approach to managing this tradeoff. We define a coalition as two or more states that make a joint threat to use force in a specific international crisis (see also Wolford 2014). The coalition can be revisionist or status quo, its 3

5 threat can be explicit or implicit, and the crisis may or may not escalate to war; the key element is a joint threat of military action should its demands be resisted. At a minimum, this requires (i) the existence of a crisis or dispute and (ii) multiple states taking the same side before the crisis escalates to war. 1 In practice, the numerous states threatening Iraq with war lest it withdraw from Kuwait in early 1991, NATO s combined demand that Serbian forces withdraw from Kosovo under pain of air strikes, and the joint French-Chadian threat of war over Libyan support of insurgents in 1983 all qualify as coalitions. However, the American threat to invade Haiti to restore President Aristide, despite diplomatic and institutional support, remains unilateral under our definition, since no other states contributed militarily. 2 The potential benefits of acting with coalition partners are considerable. Coalitions can increase a state s available military power (Morgenthau 1967, Walt 1987, Waltz 1979), which may derive from the strict aggregation of capabilities as well as contributions such as basing, overflight, or transit rights all of which can act as force multipliers without directly adding to material capabilities. Collaboration can also lower the costs of coercion through burden-sharing and specialization (see Kimball 2010, Morrow 1993). To secure these benefits, though, an erstwhile coalition-builder must secure the cooperation of partners that may not share her geopo- 1 Coalition is often used interchangeably with alliance (see, e.g. Resnick 2010/2011, Walt 1987), but the two are analytically and empirically distinct. Alliances are formal promises to cooperate to form coalitions in the event of war (Morrow 2000), but whether these promised coalitions form depends on the activation of their treaties and fulfillment of their terms. Indeed, many alliances are never invoked. While some coalitions do form to honor formal alliance commitments, a large majority include nonallied states (Wolford 2014), and even some coalitions of allies fight together out of interests other than those that bind them as allies for example, American-led wars against Iraq and the NATO interventions in Kosovo and Libya. Therefore, while alliances are peacetime promises to form coalitions that may or may not be fulfilled (Leeds, Long and Mitchell 2000), coalitions are those states that do take the same side in a crisis, whether or not a prior commitment obligates them to do so. In fact, successful defensive alliances may deter many crises in the first place (Leeds 2003), meaning that the sample of observed coalitions should be particularly unlikely to contain precisely those alliances that work best: reliable ones (cf. Smith 1995). Put simply, alliance ties are neither necessary nor sufficient for the formation of coalitions. 2 But see Kreps (2011), whose concept of coalitions codes this crisis as formally multilateral due to institutional authorization but no effective military contributions from other states. 4

6 litical interests, taste for risk, or resolve over the issue, requiring compensation in the form of side payments or policy concessions over the management of the crisis. This compensation comes in a variety of forms, from compromises over war aims and escalation levels to post-crisis influence over the fruits of victory or policy outcomes. Several American-led coalitional efforts since the Cold War are instructive. The United States calibrated its war aims in Operation Desert Storm, agreeing not to topple Iraq s government in 1991, lest it forego critical Saudi staging support; ultimately, the US limited its goals to the restoration of Kuwaiti sovereignty and the degradation of Iraqi military capabilities (Bush and Scowcroft 1998, pp. 313, 491). 3 Likewise, American decisionmakers agreed to limit escalation in the Berlin Crisis of , opting against sending ground forces across the intra-german border (Aono 2010, pp. 326, 334), striking a delicate balance between convincing allies that it would not be rash, even as the Soviet Union had to be persuaded that it just might be (Freedman 2000, p. 93). In other cases, would-be coalition-builders decide that some contributions are not worth paying the necessary compensation. The United States turned down the chance to open a northern front in the Iraq War of 2003, which could have forced a division of Iraqi forces and quickened their collapse, rather than promise Turkey the share of the postwar spoils of victory it would have required in return: the right to enter and establish order in Iraqi Kurdistan (Gordon and Trainor 2006, pp. 42, 115). Instead, the US offered Turkey $6 billion in aid, which proved insufficient to overcome parliamentary opposition (Keegan 2004, p. 138). Thus, coalition formation requires compromise among diverse interests, and a potential coalition-builder weighs these sacrifices against the increased military power it might enjoy by securing a partner s cooperation. How states and their leaders manage this tradeoff in the con- 3 See also Atkinson (1993, pp. 298, 299) and Kreps (2011, p. 27). 5

7 text of specific crises is not well understood. Most explanations of military cooperation are cast at the question of alliance formation, focusing on factors such as military power (Morgenthau 1967, Waltz 1979), security-autonomy tradeoffs (Morrow 1991, 1993), and the desire to limit costs and share resources (Kimball 2010). Though many arguments about alliance formation apply in principle to coalitions as well, an alliance treaty s concern with future conflict leads it to deal almost exclusively with long-term stable or structural factors in the form of power, geopolitical interests, or domestic institutions. The crisis-specific nature of decisions over coalition formation suggests a role for short-term considerations affected by features of the target, the international environment, and a leader s domestic political situation at the time of the crisis. Kreps (2011), for example, links two crisisspecific factors the immediacy of threats and operational needs to American choices over unilateral and multilateral action. Additionally, both Tago (2007) and Chapman (2011) show that the support of international organizations makes states more likely to join coalitions in particular crises. Taking a similar approach, we argue that an additional short-term factor a national leader s domestic political security should shape both the desire for partners and the willingness to compensate them, because political survival is in many cases linked to leaders performance in international disputes. National leaders make ultimate decisions over war and peace, and an extensive literature explores the relationship between an incumbent s desire to stay in office and the use of military force. Scholars continue to debate the general relationship between conflict outcomes and political survival (Bueno de Mesquita et al. 2003, Chiozza and Goemans 2004, Debs and Goemans 2010), but having entered a conflict, leaders are on average rewarded for victory and punished 6

8 for defeat (Bueno de Mesquita et al. 2003, Colaresi 2004, Croco 2011). 4 Leaders may be punished for incompetence signaled by failed foreign policies (Johns 2006, Richards et al. 1993, Smith 1996) or more easily toppled if defeat compromises the repressive apparatus (Debs and Goemans 2010), which makes participating in an international crisis personally risky. As such, leaders are most likely to initiate crises when they are secure in power (Chiozza and Goemans 2003, 2011), hoping to minimize the risks that they will be subsequently turned out of office. In other words, those leaders best able to weather defeats due to competent management of domestic policy or institutional protections appear most likely to use military force. The risk that defeat can result in the loss of office should also influence a leader s decisions once engaged in a crisis: if crisis outcomes affect political survival, then leaders have strong private incentives to avoid defeat. One way in which they can do so is to take on a coalition partner, which holds out the promise of increasing the odds of victory. However, if building coalitions requires concessions in return for improved military prospects, then political survival incentives may affect a leader s willingness to make just such a trade, especially if she wishes to demonstrate competence with a foreign policy success. Thus, even as leaders have public incentives to maximize their country s share of the international pie, they may be tempted to pursue their private goals of political survival, making ostensibly unnecessary policy concessions to coalition partners in order to maximize their chances of retaining office. Given these incentives, when will they successfully build coalitions, and when will they act unilaterally? In the following section, we specify a theoretical model designed to answer this question. 4 Croco (2011) shows that this is true only for culpable leaders, i.e. those leaders who started the war or have a close political tie to the leader in office at the beginning of the war. Since we look at coalition formation before any escalation to war, our theory and empirics both focus on culpable leaders. 7

9 A Model of Political Security and Coalition Building Suppose that the incumbent leader of state L, who we call l, has a policy dispute with some other state over which she can either initiate a crisis or tolerate the status quo. Next, let her payoffs consist of two components: national and personal interests. While she cares about the substance of the policy dispute (national interest), she also cares about her own political survival (personal interest), which is a function of the competence she demonstrates, or fails to, in the crisis (see Richards et al. 1993, Smith 1996, 1998). Once engaged in a crisis, she decides whether to propose a coalition to a potential partner state, P, or to act unilaterally against the target state. Taking on a partner can improve l s military prospects, but the partner must also be compensated to ensure its cooperation, which requires side payments or other political concessions that come at the expense of the national interest. Before discussing the sequence of play and the details of the players utility functions, we should mention some notable characteristics of the game. The incumbent l pursues two goods: the distributive outcome of the crisis, which we call the national interest (η), and her personal benefits from surviving in political power (ρ). For a given outcome, her expected utility is the product of her national and personal interests, or u l (η l,ρ l ) = η l ρ l. 5 Thus, while she is attentive to the inefficiency of war to achieve benefits at the national level (Fearon 1995), she also values the potential benefits of war at the personal level (Chiozza and Goemans 2004). To keep matters simple, however, we assume that P is a unitary state that values only the national interest, such that u P (η P ) = η P, while the target is wholly nonstrategic its military strength helps shape the crisis outcome, as it would if it were a strategic player, but we abstract away from 5 This is a Cobb-Douglas representation of leaders attempting to maximize their enjoyment of two goods through the pursuit of foreign policy. For a similar example, see Palmer and Morgan (2006). 8

10 Figure 1: Crisis initiation, political survival, and coalition formation crisis u i (status quo) join u i (coalitional crisis) l crisis l s [0,1] P refuse u i (bilateral crisis) unilateral u i (bilateral crisis) other decision-making elements of its role in the crisis. As shown in Figure 1, the game begins as l chooses whether to tolerate the status quo or to initiate a crisis. If she accepts the status quo without dispute, she receives a payoff for the national interest in her country s current share of the issue in question η l = q, the total possible value of which we normalize to one, q [0,1]. In addition, l receives a personal payoff ρ l, in that she receives either 0 for losing office or 1 for retaining it as a function of her probability of political survival if the status quo stands, θ. We use θ to represent the incumbent s baseline probability of remaining in power absent participation in a crisis, which can depend on her time in office to date, the frequency of previous leadership turnover, the costs of replacing her, and the success of her domestic policy agenda. Her personal payoff for accepting the status quo is thus ρ l = θ 1 + (1 θ) 0 = θ. The potential partner receives its own payoff for the status quo, u P (status quo) = βq, where β [, ] is a bias term representing the extent to which P shares l s foreign policy preferences; as β increases, P s preferred status quo division resembles l s, but as β decreases, its preferences are increasingly opposed to l s. Should l initiate a crisis, she must choose whether to act unilaterally, which leads to a bilateral crisis with the target state, or to propose a coalition with P. Whether she acts unilat- 9

11 erally or with a partner, we represent the crisis as a costly contest that resolves the international issue probabilistically in favor of either l or the target state, where the victorious side sets its preferred policy. Should l act unilaterally, the crisis entails upfront costs c L > 0, such as the expense of military deployments, opportunity costs of lost investment, as well as casualties, deaths, and destruction. The probability with which l succeeds in overturning the status quo and receiving her preferred policy, p L (0,1), depends on both her state s military capabilities and her own competence (see, inter alia, Banks 1990, Fearon 1995, Powell 1999, Richards et al. 1993, Smith 1998). 6 Therefore, l s national interest payoff in a bilateral crisis is η l = p L (1) + (1 p L ) 0 c L = p L c L, where success means that l s country enjoys all the benefits of revising the status quo at no cost of concessions to ensure P s cooperation. Her personal payoff, ρ l, depends on the competence she demonstrates or fails to in the crisis, such that she is rewarded with retaining office if successful, but she is punished for failures. We assume that a successful crisis ensures l s political survival, but her baseline job security mediates the effects of failure. If the crisis ends in L s defeat, she survives in office with probability ρ(θ) (0,θ], which is a function of her baseline job security, θ. Formally, she is less likely to survive a defeat than if she were to avoid the crisis in the first place, such that ρ(θ) < θ, but she can weather foreign policy failures more easily the more secure in power she is otherwise, or ρ(θ)/ θ > 0. 7 Therefore, her personal payoff is ρ l = ( p L + (1 p L )ρ(θ) ). We assume that the stakes of the international crisis over which l has a national interest be they territory, influence, policy, etc. have both a public (non-excludable) and a private (excludable) component. Normalizing the stakes to a total value of one, both l and P have prefer- 6 For a similar representation of a reduced-form distributive contest, see Gilligan, Johns and Rosendorff (2010). 7 Note that ρ(θ) < θ also implies that, while positive by construction, ρ(θ)/ θ is also less than one. 10

12 ences over the substance of the policy outcome. First, there are non-excludable benefits, in that P may care about the location of L s border with the target state or whatever policies the target might agree to change, regardless of whether it participates. We represent the relative size of this public component as γ [0,1]. On the other hand, by not participating in the crisis, P foregoes enjoyment of the private benefits that accrue to participants, the relative share of which is (1 γ), such as control over strategies and aims in the crisis and ensuing conflict, influence over the revised status quo, and even shares of the spoils of victory. Thus, if she acts alone, l receives the entirety of the public and private components of the good in the instance of victory, while P receives a payoff for the public component of the outcome, or u P (bilateral crisis) = p L βγ, excluded as it is from any share of the private component. If l wishes to form a coalition, she proposes a division of the private benefits of victory to compensate P for its cooperation (cf. Riker 1962), which it accepts to enter the crisis in coalition with L or rejects such that the crisis remains bilateral. She proposes a division s [0,1] of the private benefits to P, such that, if the coalition is successful, her country retains s(1 γ) and P receives (1 s)(1 γ). P s acceptance affects l s military prospects in two ways. First, capability aggregation or force multiplication weakly increases her chances of success, such that she succeeds in the crisis with probability p LP p L. Second, burden- or cost-sharing weakly lowers the costs of the crisis even if P does not make a substantial military contribution, such that the public costs of the crisis are c LP c L. Thus, her national-interest payoff ( ) is η l = p LP γ + s(1 γ) clp, while her personal payoff reflects her increased chances of victory in the crisis, such that ρ l = ( p LP + (1 p LP )ρ(θ) ). Should P accept l s proposal, it receives ( ) u P (coalitional crisis) = p LP βγ + (1 γ)(1 s) cp, where c P > 0 represent its own upfront costs for participating in the crisis. Finally, if P rejects l s proposal, the game proceeds as a bilateral 11

13 crisis and payoffs are identical to the case in which l chooses to act alone in the first place. 8 With preferences and the sequence of moves defined, we can summarize the incumbent s payoff function over the game s three possible outcomes as θ q u l (ρ l,η l ) = ( pl + (1 p L )ρ(θ) ) ( ) p L c L ( plp + (1 p LP )ρ(θ) ) ( ) p LP (γ + s(1 γ)) c LP if status quo if bilateral crisis if coalitional crisis, and the potential partner s as βq u P (η P ) = p L βγ if status quo if bilateral crisis ( ) p LP βγ + (1 γ)(1 s) cp if coalitional crisis. These payoff functions make clear the strategic tension at the heart of coalition-building. For P, joining the coalition is costly, and it may not value the stakes as highly as l; thus, it requires some compensation to make participation worthwhile. For her part, l would like to enhance her chances of success, both to pursue the national interest and to secure her own political survival, but this comes at the cost of political concessions that she would rather not make. How, then, does the incumbent evaluate and make such tradeoffs? When will she compensate a reluctant partner for its cooperation, when will she act unilaterally, and when might she be 8 It is plausible that l might pay some kind of political cost for failing to form a coalition, but since complete information dictates that she knows ex ante whether a coalitional proposal will succeed or fail, she would avoid failed coalitional proposals under the same conditions she does in the present version of the model, leading to no changes in the substantive results. Therefore, we opt for the simpler specification here. 12

14 deterred from initiating crises in the first place? The following section shows that the answers to these questions turn on both the international environment and the leader s political security. Theoretical Analysis In this section, we answer two questions. First, once involved in a crisis, when will the incumbent and potential partner build a coalition, and when will she act unilaterally? Second, how does the availability of a coalition partner affect her willingness to initiate or select into crises? This allows us to generate predictions over the formation of coalitions as well as develop expectations about the selection process that generates the sample of crises used in our empirical models. Proposition 1 summarizes each player s strategy in the Subgame Perfect Equilibrium, outlining the conditions under which (1) l initiates a crisis or accepts the status quo, (2) l proposes a coalition or acts alone, and (3) P accepts or rejects coalition proposals. Proposition 1. The following strategies constitute the unique Subgame Perfect Equilibrium. When β > β LP, l initiates a crisis and proposes s = s LP when q < q LP ; when q q LP, she tolerates the status quo. When β β LP, l initiates a crisis and acts unilaterally when q < q L ; when q q L, she tolerates the status quo. P accepts any s max{s LP,0} and rejects otherwise. See appendix for proof. To focus on the process of coalition formation, we first discuss those cases in which l has selected into a crisis, then follow by discussing how anticipated behavior in the coalition-formation subgame affects l s selection into crises in the game s first move. 13

15 Political Survival, Side Payments, and Coalition Formation When will a leader and a potential partner successfully form a coalition? P accepts a coalition proposal when u P (coalitional crisis) u P (bilateral crisis), such that l can only propose to keep s max{ (1 (1 β)γ)p LP βγp L c P (1 γ)p LP,0} s (1) for herself and still secure P s cooperation. As shown in Inequality (1), the required side payment increases in P s costs for participating (c P ), and it decreases in β, or the extent to which P s preferences align with the leader s. Since l wishes to pay as little as possible and keep the maximum for her country, she meets P s acceptance constraint at equality. As stated in Proposition 1, l will propose this side payment, successfully building a coalition, when P s preferences are sufficiently similar to her own, or when β > 1 γ ( clp c L + c P (1 ρ(θ))(p ) L c L ) 1 β LP. (2) p LP c L ρ(θ) + (1 ρ(θ))p LP The reasoning behind this result is straightforward: when leader and partner have similar foreign policy preferences (high β), P receives more from the public benefit of victory, it requires less in the way of compensation to join the coalition and ensure that victory, and l is happier to take on a partner when doing so does not require large side payments. All else equal, states building coalitions are more likely to take on partners who share their preferences, as opposed to states with more divergent interests (low β) that would require more substantial compensation in return for cooperation. Implication 1. A leader is more likely to form a coalition in a crisis with a given partner as the 14

16 partner s preferences grow more similar to her own. A closer look at Inequality (2) shows that the incumbent s baseline political security also plays an important role in coalition formation. As θ increases, l is better able to weather defeats in office, and the right side of the inequality becomes greater, making the coalition-formation constraint more difficult to satisfy. Politically secure leaders are in a better position to survive defeats, so they are less compelled to make concessions to acquire coalitional help to win. Politically insecure leaders should be more willing to make side payments than secure leaders, because they are more in need of a military boost to feel comfortable in their post-crisis political survival. Insecure leaders, then, trade an ever larger share of the private element of the outcome, (1 γ), in return for improved chances of victory and thus political survival. Proposition 2. The leader is less willing to take on a partner as her baseline job security increases, because β LP / θ = (p l c L )ρ (θ) γ(p LP (1 ρ(θ))+ρ(θ)) 2 > 0. Implication 2. A politically secure leader is less likely to form a coalition in a crisis than a politically insecure leader. Finally, consider how the effect of l s political security interacts with P s preferences. Figure 2 plots the coalition formation constraint, β LP, as a function of the leader s baseline political security. 9 Below the threshold, β β LP (the region shaded in gray), l is sufficiently secure in office that she acts unilaterally, confident in her probability of surviving a foreign policy failure in office that she would rather not make the side payments necessary to form a coalition. On the other hand, when β > β LP, she is willing to make side payments to boost her chances of 9 Parameters are fixed at γ = 0.5, c P = 0.1, p LP = 0.65, p L = 0.55, c L = 0.2, c LP = 0.1, and ρ = 0.8, imposing the functional form ρ(θ) ρ θ = 0.8 θ. 15

17 Figure 2: Preference similarity, political security, and coalition formation b Ø LP Similarity of preferences between ` and P Coalition Incumbent ` s political security Unilateral 1 q avoiding failure and surviving in office despite the fact that these side payments come principally at the expense of the national interest. Key to this graph, however, is the fact that, as l becomes less politically secure, she also becomes less selective in her choice of partner. Specifically, the range of acceptable partner preferences, and thus tolerable side payments, grows as θ approaches zero, such that l strikes coalitional bargains with partners of ever more divergent preferences as her own political security declines. Implication 3. A politically insecure leader will take on coalition partners with preferences that diverge more from its own preferences than will a politically secure leader. Thus, there appears to be an interactive relationship between political security and preference diversity in the formation of coalitions; coalition builders have a national-interest incentive to minimize the side payments they must make to increase their military prospects in international crises, but as they grow increasingly desperate to preserve their own political sur- 16

18 vival, leaders will become increasingly willing to take on partners of ever more divergent preferences. Therefore, as suggested above, short-term incentives like political security can have an impact on an incumbent s evaluation and pursuit of potential coalition partners, independently of more long-term stable factors such as military power. Coalitions and Crisis Initiation Having established the conditions under which coalitions form, we can now analyze the game s first move: the incumbent s initiation of a crisis aimed at revising the status quo. We focus here on both the availability of a coalition partner and l s baseline political security as factors in the decision, with an eye to improving inferences drawn over coalition formation in the sample of crises analyzed in the following section. Proposition 1 states that l initiates a crisis when her valuation of the status quo is sufficiently low, or q < 1 θ ( pl + (1 p L )ρ(θ) ) (p L c L ) q L (3) when she will act unilaterally in the crisis, and q < 1 θ ( (ρ(θ) 1)pLP ρ(θ) )( c LP + c P + βγp L (βγ + 1)p LP ) qlp (4) when she will go on to form a coalition. In each case, favorable military prospects increase the attractiveness of a crisis, but the military boost provided by a coalition, p LP p L, comes at the expense of compensating the potential partner; formally, the initiation constraint when l forms a coalition, q < q LP, decreases in the partner s costs of participation (c P ) but rises in the dyadic 17

19 Figure 3: Preference similarity, status quo evaluations, and crisis initiation q 1 Incumbent ` s status quo evaluation Crisis q L Crisis q LP b LP 0 b Similarity of preferences between ` and P similarity of preferences between l and P (β). To see how the availability of a partner affects crisis initiation, we compare the constraints defined by Inequalities (3) and (4), also plotted in Figure 3. As stated in Proposition 3, q LP is greater than q L, meaning that an available coalition partner increases the attractiveness of selecting into a crisis, as long as β > β LP, which is precisely the condition supporting coalitionbuilding. The prospect of the increased probability of success in the crisis and reduced costs thereof that accompany a coalition makes a crisis into a more appealing option than tolerating the status quo, even given the price of side payments to secure the partner s cooperation. This is the case because β > β LP, such that the partner does not require high side payments to be convinced to join a coalition. Proposition 3. The leader is more willing to initiate crises when she expects to build a coalition than when she expects to act unilaterally when q LP > q L, which is true as long as β > β LP. See 18

20 appendix for proof. Thus, the availability of cheap coalition partners should make incumbents more willing to initiate crises than they would be if they knew they would have to act alone. 10 Put differently, the very possibility of cooperation is associated with an increase in the chances of military conflict. When partners are available on the cheap, the leader is comparatively more willing to initiate crises, but when coalition building requires substantial compensation, crises are less attractive, and l tolerates the status quo rather than mortgage whatever she might gain in the crisis to ensure P s cooperation. This implies that leaders with available and easily compensated partners will be systematically overrepresented in our sample of crises, and we discuss the implications of this selection process for our empirical model in the following section. Finally, it is worth noting that l s baseline job security has no independent effect on her selection into crises. As we show in the appendix, taking comparative statics over both q L and q LP with respect to θ shows that baseline job security can increase or decrease the attractiveness of initiating a crisis, depending on a complex interaction of l s political sensitivity to failure in the crisis, military prospects, and the amount of P s required compensation. Therefore, while the unambiguous relationship between the availability of an inexpensive coalition partner and the initiation of crises has potentially large effects on inferences drawn from a sample of leaders already in crises, the lack of a consistent relationship between political security and crises should pose less of a problem for drawing inferences from the empirical models we analyze in the following section. 10 See Chapman and Wolford (2010) for an argument that international organizations like the UNSC can also encourage conflict when their support can lower the costs of war for a strategic state intent on winning its approval. 19

21 Empirical Analysis In this section we estimate empirical models designed to analyze the implications derived from our theoretical model, highlighting in particular the independent effects of and the interaction between an incumbent s political security and the preferences of potential partners on the decision to build military coalitions in interstate crises. After translating the logic of the theoretical model into explicit hypotheses over coalition formation and expectations about sample bias, we discuss issues of research design and data collection followed by the results of our analysis. Hypotheses Our theory implies that an interactive relationship between an incumbent leader s political security and preference divergence with a potential partner should help explain both whether and with whom leaders choose to build military coalitions. In this section, we translate the implications of the theoretical model into hypotheses for empirical analysis. First, as indicated by Proposition 2, leaders who are secure in political power should be less likely to form coalitions with any partners, and Proposition 1 indicates a similar unconditional relationship between foreign policy preference divergence and coalition formation. Hypothesis 1. An incumbent leader in a crisis is less likely to form a coalition as her baseline job security increases. Hypothesis 2. An incumbent leader in a crisis is more likely to form a coalition with a potential partner as the similarity of their foreign policy preferences increases. In addition to these bivariate relationships, the theory also suggests a conditional, interactive effect of preference divergence and political security on coalition formation. 20

22 Hypothesis 3. The effect of similarity in foreign policy preferences on coalition formation increases in an incumbent s baseline job security. Though we explicitly argue in the theory that the choice to initiate a crisis is influenced by the leader s expectations, our empirical models focus only on coalition formation in crises that have begun. An alternative would be to use a sample selection model (cf. Heckman 1979, Sartori 2003), estimating first l s entrance into an international crises and then coalition formation in a truncated sample, directly modeling the correlated errors across the two decision stages. However, while crisis involvement is contingent upon coalition formation in general, the process we model in the second stage is the decision to join a particular partner out of many possible ones. The latter is quite different than the former, with the selection equation largely conditioned on the characteristics and relationships of the initiator and the target and the outcome equation including the characteristics and relationships of the crisis actors and a large number of states who may or may not become involved in the crisis. Such differently specified equations are unusual for a selection model, but modeling the crisis choice conditioned only on the coalitional elements from the second equation would leave us with significant omitted variable bias. 11 Thus, we instead use standard probit models on a truncated sample of states involved in international crises, each paired with the states that could be potential partners in a coalition. We cluster standard errors by the observed participant leader identified as l in each dyad. Nevertheless, we can use the theoretical model to generate some expectations over how key independent variables affect the selection process. First, returning to Figure 3, note that the availability of acceptable partners makes incumbents more willing to initiate crises (to the right 11 Indeed, when attempting to use this type of model, the estimated parameter ρ indicated the errors of our two equations were not correlated, though they clearly should be in theory. 21

23 of β LP ) than cases in which they will act unilaterally (to the left of β LP ). As a result, we expect the sample of leaders and potential partners in observed crises will exhibit restricted variation at the lower end of our variable measuring preference diversity. Leaders with few available partners might be underrepresented in the sample, making comparisons with the comparatively overrepresented leaders with viable partners difficult. However, this yields a conservative test; to the extent that we do find the expected relationships, their ability to overcome this attenuation bias should increase confidence in the results. Second, since the theoretical model predicts no consistent relationship between political security and crisis initiation, we do not expect it to have a substantial effect on our ability to draw inferences over the key independent variable in the analysis. Research Design As implied by our hypotheses, our unit of analysis is the leader-partner directed dyad, where the leader is the incumbent of a state involved in an international crisis, and the potential partner state may become involved if it joins the leader in a coalition. Each incumbent that is involved at the outset of a crisis takes a turn in the analysis as the leader (l) making a decision to form a coalition, and each leader is paired with all relevant potential partners (P), i.e. those states with whom the leader could possibly propose to form a coalition. The dependent variable is coalition formation, reflecting the joint decision of the leader to propose a coalition and the partner to accept it. This structure allows us to empirically model each leader s decision to make an offer the potential partner will accept to form a coalition in the context of an international crisis, as well as which partners will receive and accept offers. In the data discussed below, this 22

24 yields 148 leaders as crisis participants for the years 1951 to To generate the appropriate dataset, we begin with all participants in international crises, as initiators or targets, using actor-level crisis data from the International Crisis Behavior (ICB) Project (Wilkenfeld and Brecher 2010). Each leader in a crisis has the opportunity to form a coalition or act unilaterally, considering both her domestic political incentives and the characteristics of a given potential partner. We give each participant leader a chance to be observed as the coalition-builder, or l, from the theoretical model s perspective, and pair the observed leader with each potential "partner" state which we identify with Lemke and Reed s (2001) indicator of political relevance using EUGene (Bennett and Stam 2000), then merge them into our list of participant leaders. 13 This yields a directed-dyad dataset such that each leader participating in a crisis has the potential to form a coalition with any of its contiguous states or one of the great powers. 14 The dependent variable is coalition formation. As discussed above, we define a coalition as two or more states engaged in an interstate crisis that make, implicitly or explicitly, a joint threat of war as either initiators or targets. To identify coalitions that form in our sample, we use data collected by Wolford (2014), which relies on information from the ICB Project and secondary sources to code states as being in a coalition when they make joint threats or otherwise expect to act together militarily, not just diplomatically, on the same side at the beginning of a crisis. 12 The panels are unbalanced. The number of complete observations used in each estimated model is reported in Tables 1 and Politically relevant dyads are those including at least one major power or those in which the states are directly or indirectly contiguous (Lemke and Reed 2001, 127). 14 A coalition can include any number of members, and thinking about formation multilaterally (k-adic) can yield different and interesting results as compared to dyadic approaches (Fordham and Poast 2012). We analyze leader-partner directed dyads because it allows us to focus on the characteristics that would make a single state (L) more likely to enter a coalition as a primary decision-maker, as well as the characteristics of particular partners, rather than what size of coalition would be ideal for a given situation. 23

25 The variable Coalition l,p equals 1 if the leader-partner dyad in question forms a coalition in the observed crisis and 0 otherwise. 15 For example, if a leader l is paired with six potential partner states, we observe six leader-partner dyads, but Coalition l,p = 1 only for those dyads containing a partner with which l forms a coalition in the crisis. Hypothesis 1 predicts that a leader will be less likely to form a coalition with any potential partner as she becomes more secure in office; this corresponds to an increase in θ in the theoretical model. We use a measure of a leader s Job Security l that captures the expectation of how secure a leader is in her position, based on observable characteristics of the leader and the state. Conrad and Ritter (2013) replicate and extend Cheibub s (1998) measure of job insecurity from 1951 to 2001, using source data from Bueno de Mesquita et al. (2003). The indicator is the estimated hazard of a given leader losing office as a function of her time in that position to date, the cumulative number of previous leader changes, and the economic status quo. Subtracting the probability of turnover from 1 yields the predicted probability of the leader remaining in power as a function of these observable factors, which we use as the indicator of Job Security l. We also estimate the likelihood that a leader will retain office given a broader range of non-crisis related factors by replicating Chiozza and Goemans s (2004) analysis of leader turnover and using those predicted probabilities as a robustness check, which we report in the tables below. 16 As stated in Hypothesis 2, a leader should be less likely to choose a coalition partner with starkly different foreign policy preferences from her own, given the policy concessions necessary to win such a partner s cooperation. We account for foreign policy preference similarity based on states votes in the United Nations General Assembly (UNGA), which in the aggre- 15 The data excludes states who join coalitions after the initial formation, since the process of adding partners after a coalition has formed is a different decision-calculus (see Wolford 2014). 16 We generate the predicted probabilities of survival for each leader based upon observed characteristics except for crisis or war involvement, which we set to zero for each observation. 24

26 gate serve as a measure of revealed international preferences. Reed et al. (2008) use a state s voting pattern to estimate its ideal point with respect to the international status quo along a one-dimensional scale, ranging from -1 to 1. To account for the differences between a pair of potential coalitional partners, we take the absolute value of the difference between their mean estimated ideal points in the year prior to the crisis under observation. This resultant variable, Preference Divergence l,p, increases as the two states international preferences diverge, ranging from zero (perfect convergence) to a theoretical maximum of two (perfect divergence). This measure allows us to capture how different the two states foreign policy preferences are in general, which approximates their likely divergence over the issue at stake in the crisis. To assess Hypothesis 3, we interact Job Security l with Preference Divergence l,p and report the effect of the interaction along with the effects of the standalone constituent terms. Indicators of the relationship between the leader s state and potential partner state are included to control for elements that could confound the estimated relationship between job security and coalition formation. We created an indicator of Joint Democracy l,p, which equals 1 when both states are democracies, 17 as we expect democracies to be more likely to collaborate with another democracy (Russett and Oneal 2001). 18 The variable Contiguous l,p equals 1 if the members of the dyad share a land border or are separated by no more than 150 miles of water, as coded by Stinnett et al. (2002), which we expect to be positively associated with coalition formation. We include a measure of Relative Power l,p, which measures the national material capabilities of the leader s state as a proportion of the sum of both states capabilities, using the Correlates of War project s National Material Capabilities Dataset, anticipating that leaders 17 The cutoff for Democracy is a Polity IV (Marshall and Jaggers 2009) Democracy score greater than or equal to six. 18 The estimates of Job Security l are not an explicit function of regime type. We thus do not expect to introduce error by controlling for regime type separately. 25