Abstract
While much research has been done on the domestic determinants of alliance institutionalization, there has been a neglect of the effect of domestic politics, by which we mean contestation between political actors in the same country. We hypothesize that the ideology of the parties governing countries negotiating the terms of security relationships will affect their preferences over the degree and kind of institutionalization seen in alliances. Drawing on previous literature, we argue that rightist parties are more sensitive to sovereignty costs and will therefore insist on maintaining more control over policy than their leftist counterparts. They can assert control either by imposing hierarchical forms of institutionalization when they are a stronger party to an alliance or by avoiding institutionalization altogether if they are the weaker party in an alliance. In contrast, we expect leftist parties to be less sensitive to sovereignty costs and generally favorable to more voice-driven, egalitarian institutions that have institutionalized mechanisms for consensus-building, regardless of their country’s relative power position. Combining the ATOP dataset on alliance design with the Parties Manifesto Project, we find broad support for our hypotheses. Our findings indicate that scholars should pay more attention to the internal ideological contestation within countries, making room for domestic political factors that go beyond regime type.
Introduction
When forming alliances, states are faced with a difficult trade-off between effectiveness and sovereignty. More effective alliances are likely those that are less anarchic, meaning they restrict the autonomy of the parties to the arrangement and hinder unilateral actions. Among other things, institutionalized alliances do more to facilitate the careful coordination of policy (Wallander & Keohane, 2002), integrate armed forces so that they can act more cohesively in combat (Weitsman, 2013), and more credibly bind state parties to their commitments so as to facilitate deterrence (Morrow, 2000; Leeds, 2003). These advantages, however, come with ‘sovereignty costs’. Even scholars who contend that alliances can be ‘liberating’ acknowledge that alliance agreements often constrain some forms of behavior, a price states pay to take advantage of the opportunities alliances afford them to pursue valued foreign policy objectives (Palmer & Morgan, 2006: 41). A number of factors likely affect calculations concerning these trade-offs. Stronger states might be less concerned with military effectiveness and more with avoiding entrapment by smaller allies. Smaller states, in contrast, might prefer institutionalization so as to avoid abandonment.
We suggest another important influence on alliance design. Sensitivity to sovereignty costs likely varies across the political spectrum. Those more sensitive to limitations on state autonomy are likely to prefer alliances with very different designs than those who are less sensitive. Based on prior research, we hypothesize that governments led by leftist parties are more likely to accept sovereignty limitations for the added gains of coordination than are conservative parties of the right. Right-wing parties are ideologically less trusting and inclined to guard national sovereignty since they are less willing to place their fate in the hands of others, something true of conservatives generally (Hibbing, Smith & Alford, 2014: 250). States have different preferences for institutional design based on the parties that lead them, differences that cannot be reduced to external features of their environment.
We do not contend that alliances formed by right-of-center governments will always be less institutionalized, however. The alliance literature generally treats institutionalization as a matter of degree, ranging from low to high, thereby missing important qualitative differences in types of institutionalization. More institutionalized alliances can be egalitarian or hierarchical in character. The former seeks to make security policy more effective by providing more ‘voice’ opportunities for the parties involved through provisions requiring regular consultation between military and defense officials as well as the creation of formal organizations. The latter subordinate a weaker party to the direction of a stronger through such features as non-reciprocal basing rights that allow powerful states to position forces on their ally’s territory but not vice versa and sometimes give the stronger ally command authority over the weaker partner’s forces. Hierarchical alliances therefore allow the possibility of a more effective alliance with low sovereignty (although not necessarily low financial) costs for the stronger party but considerably higher costs for the subordinate party.
We hypothesize that hierarchical alliances will be associated with the presence of militarily strong partners led by right-wing parties who wish to avoid sovereignty costs, to which they are more sensitive, by asserting control over subordinates. For weaker states in an alliance, we expect rightist parties will want to avoid hierarchical institutionalization. They lack the leverage to impose terms that would mitigate their sovereignty costs and cannot subordinate others. We also hypothesize that leftist governments in both stronger and weaker parties to an alliance will be positively associated with voice-driven institutional designs that are more egalitarian in character. The left is largely defined ideologically in democracies by its commitment to joint gains and inclusive political participation, tendencies that we expect will manifest themselves in their choice of alliance organization.
Combining the Alliance Treaty and Obligations and Provisions (ATOP) dataset (Leeds et al., 2002) with that of the Parties Manifesto Project (Volkens et al., 2014), we find broad support for our hypotheses. Alliances in which the stronger partner is led by a right-wing government are more likely to be hierarchical in character but those in which the weaker partner is led by a right-wing government are less likely to be so. Alliances that include left-wing governments are more egalitarian and voice-driven rather than uninstitutionalized, regardless of whether the country is weaker or stronger than its partner.
Theoretical review
Alliance trade-offs
‘Institutionalization’ refers to the formalization of interstate relationships via explicit rules establishing specific behaviors that political actors must adopt under defined conditions (Koremenos, Lipson & Snidal, 2001), thereby reducing institution members’ freedom of action. Institutionalized alliances are not ‘anarchic’ relationships in which each state retains full rights of control over its actions in scenarios not specified by the alliance agreement, as well as the right to unilaterally determine whether a given situation is covered by the alliance’s terms (Lake, 1996; 1999).
States institutionalize alliances to make them more effective. By coordinating defense expenditures, states can promote specialization and avoid duplication. By consulting on defense policy, partners can coordinate on policy and bring maximum pressure to bear on adversaries. By creating an integrated command structure, they can more effectively act in times of crisis and war. By specifying a more binding commitment to collective defense, they might make such a war less likely in the first place. However, all of this comes with potential costs of various kinds. Snyder (1984) theorizes that allies worry that their partners might abandon them if an armed conflict were to occur, or entrap them in an armed conflict that they otherwise would have avoided. An institution meant to bind allies together to lower the probability of abandonment increases the chances of entrapment, and vice versa. While alliance institutions can increase efficiency by allowing states to play to their comparative advantages (Wallander & Keohane, 2002), such specialization also raises states’ exit costs by making it more difficult for them to provide for their own security without allied support. Furthermore, if institutionalization enhances an alliance’s military effectiveness, it could exacerbate the problem of moral hazard by increasing a state’s expected probability of winning an armed conflict. This can incentivize states to take riskier, more aggressive actions towards adversaries than they otherwise would have (Benson, 2012), raising the prospect of entrapment.
Even if abandonment and entrapment are relatively rare outcomes, sovereignty costs associated with ‘entangling’ institutions are ubiquitous (Kim, 2011). In institutions that create fora for joint decisionmaking, a key element of institutionalization, policy might often veer from one’s ideal point. It is commonly observed that international organizations of all types and in all issue areas develop informal norms of unanimity. The social pressures to compromise are intense so that even in organizations with voting rules that protect states from doing something they do not want to do, there are still significant sovereignty costs (Buzan, 1981; Footer, 1996; Heisenberg, 2005). Leeds & Savun (2007: 1129) speculate that institutionalized alliances may be terminated earlier than uninstitutionalized ones because the constraints of institutionalization lead to more disagreements among members over the onerous terms of their relationship. Different types of institutionalization leave states more or less vulnerable to these sovereignty costs.
Typology of institutionalization: Degree or kind?
Our typical understanding of alliance institutionalization is one of degree rather than kind. For instance, as operationalized by Leeds & Anac (2005) using the ATOP data, alliance institutionalization is coded into three ordinal levels. At the highest level it is sufficient for allies to do one of the following: operate an integrated military command with partners during both peacetime and wartime; conduct a common defense policy; or allow joint troop placements, mutual exchange of bases, or establish bases on the territory of an allied state. Alliances at the next lower level of institutionalization feature none of these obligations. Instead, they require members to do at least one of the following: have their military officials participate in regular planning and coordination activities with allies during peacetime; create a formal military organization to coordinate plans and behavior; provide training and/or technology for the military of other parties; allow for the subordination of one military to another during conflict; or specify military contribution levels from each ally in the event of conflict. If an alliance features none of the above obligations, it is not considered institutionalized at all.
Within this ordinal conceptualization, however, we see institutionalizations of very different qualitative types. Integrated military commands, the subordination of one party’s defense policy to another, and non-reciprocal basing rights imply institutionalization of a hierarchical type in which one alliance partner is dominant over another and exercises greater control over its policies and military decisions. In contrast, the common coordination of military plans through institutionalized cooperation implies a more egalitarian and multilateral process in which even smaller countries are given opportunities for ‘voice’ (Ruggie, 1992; Ikenberry, 2001; Risse-Kappen, 1995). Building voice opportunities into an alliance reduces the centralization of decisionmaking and control exercised by the stronger state. This indicates the stronger ally is more trusting, as it demonstrates less concern pertaining to the uncertainty about its partner’s future behavior and any potential enforcement problems (Koremenos, Lipson & Snidal, 2001: 787–792). Although this raises transaction costs within the alliance by obligating the stronger state to partake in more discussions with its partner, it can also increase alliance cohesion by allowing the weaker state to influence the stronger ally’s decisions; the weaker party trades less of its autonomy for the security the stronger state provides (Morrow, 1991).
Whereas the hierarchical form of institutionalization delegates sovereignty to a more powerful state or states, the voice-driven design pools it. This does not mean, however, that institutionalization of the more egalitarian type does not entail dangers. Even for weaker countries, who might gain greater voice than they would in comparison to a hierarchical alliance, there are sovereignty costs inherent in such an institution. The core element of multilateral forms of organization is that they entail mutual rather than asymmetrical vulnerability. Both sides are putting their trust in the other side, and both sides can harm each other. This type of institutionalized alliance rests on trust on both sides, whereas hierarchical alliances primarily require trust from the weaker partner as its fate is in the stronger’s hands.
Sovereign sensitivities and ideology
States jealously guard their sovereignty. However, to the extent that we observe substantial heterogeneity in alliance design, it suggests that they do not always do so to the same extent. This raises the question of when states will make the decision to pool and perhaps even delegate control over important policy matters to an n-adic collective body. Part of the answer partially lies in broader structural and systemic features. Weaker states might be persuaded to delegate control over their armed forces to a powerful patron because they cannot defend themselves alone. The patron might agree to common defense policymaking, thereby restraining its ability to act unilaterally, in exchange for forward bases that it needs to project power (Lake, 1999).
We expect, however, that alliance design also has an important domestic component. The political ideology of states’ governing parties influences numerous forms of behavior in international politics (Arena & Palmer, 2009; Grieco, Gelpi & Warren 2009; Potrafke, 2009; Rathbun, 2004; Hanania, 2019). Changes in the core domestic constituencies of democratic governments affect alliance duration (Leeds, Mattes & Vogel 2009), though the question of which domestic groups are more or less prone to join alliances has received less systematic attention. We expect that different political actors weigh trade-offs differently depending on how sensitive they are to the loss of sovereignty and how wary they are of opportunistic behavior by their allies. Sovereignty sensitivity is a function of this distrust. The more willing states are to allow foreign actors influence over matters that were previously decided unilaterally, the more likely they will be to reap the gains from coordination that institutionalized alliances provide.
Conservative sensitivity to national sovereignty is long documented. Rathbun (2012) argues that left-leaning parties in democratic societies are more multilateralist in their foreign policy preferences than right-leaning parties, something evident in how they choose to design international institutions. Relative to conservative Republicans, he shows that the liberal wing of the Democratic Party in the United States has consistently been more willing to enter into agreements with binding institutionalized forms, larger numbers of partners, and provisions that compromise autonomy and sovereignty. Even conservatives with an international outlook favor unilateralism so as to avoid potential opportunistic exploitation (Nau, 2013), preferring to cooperate with smaller sets of partners who are known quantities. This is apparent in public opinion surveys as well, which show that conservatives in the USA view international organizations such as the UN much less favorably than their liberal counterparts (Gries, 2014: ch. 10). This phenomenon is not uniquely American. Hooghe, Marks & Wilson (2002: 981) observe that as EU integration progressed, mainstream conservative parties in Europe placed national sovereignty, unilateralism, and opposition to pressures from international organizations at the center of their political programs. A large literature on foreign policy belief systems finds that conservatives oppose ‘cooperative internationalism’, a general disposition towards the collective solution of problems found to predict foreign policy attitudes on specific issues (Wittkopf, 1990).
Why is this the case? We expect it is because right-leaning individuals and parties are more fearful of being taken advantage of by alliance partners and therefore insist on maintaining control. The right is less trusting than the left, consistently attuned to how others might take advantage of them, as well as being more threat-focused, convinced that the world is a dangerous place (Duckitt, 2001). Jost et al. (2007) argue that the right is driven primarily by an ‘existential motive’ to reduce uncertainty and threat. The right has a more pessimistic view of human nature (Conover & Feldman, 1981; Tomkins & Izard, 1965). Some even argue that conservatives’ lack of trust has a biological basis (Hibbing, Smith, & Alford, 2014).
Whereas the right is scared of danger and disorder, a preference for egalitarianism is the key attribute of the left. More than 50 years ago, Lipset et al. (1954) wrote, ‘By “left” we shall mean advocating social change in the direction of greater equality – political, economic, or social. By “right,” we shall mean supporting a traditional, more or less hierarchical social order, and opposing change towards greater equality’ (1954: 1135). Those on the left have an ‘approach’ orientation: they want to provide for others, which explains their support for state programs to help the most disadvantaged (Janoff-Bulman, Sheikh & Baldacci, 2008). The left is found to identify more strongly with the moral foundations of protecting others from harm as well as ensuring fairness and equality (Graham, Haidt & Nosek, 2009; Schwartz, 1992: 12). Furthermore, liberal and conservative elites’ domestic preferences in the United States at least correspond with their attitudes towards foreign policy (Rathbun, 2007).
Based on these core ideological differences, in combination with the distinction we have made between egalitarian and hierarchical institutionalization, we can develop expectations about the forms of alliance that left- and right-wing governing parties will prefer. Left-wing leaders will feel that they can reap the gains of collective action by limiting their own sovereignty because they are less concerned about how others will take advantage of them. We expect that leftist parties will generally be associated with more egalitarian forms of institutionalization in which all parties have voice opportunities.
This leads us to hypothesize:
Hypothesis 1: Left-wing governments, whether in the weaker or stronger partner, will be positively associated with the presence of institutionalized voice opportunities as compared to non-institutionalized alliances.
For rightist parties, we expect that institutional preferences will depend on their state’s relative power position. If the right-wing politicians of weaker countries want to maintain full sovereignty, they must opt for a non-institutionalized alliance since they cannot subjugate stronger allies. While institutionalized alliances might make abandonment less likely by locking in the more powerful country (Weber, 1997), such delegation of control comes with the possibility of entrapment and high sovereignty costs as one is no longer in control of a matter of supreme national importance. As noted above, we expect that leftist parties, less sensitive to sovereignty costs, are more likely to make such a trade-off.
However, if a distrusting government sensitive to sovereignty costs can attain a hierarchical alliance in which it controls not only its own destiny but also that of others, then it will prefer such an arrangement. Hierarchy is an institutional solution to problems of opportunism, an insight long made by theorists of the firm and later applied to security formation (Lake, 1999; Weber, 1997). Controlling, hierarchical alliance institutions are the natural choice of conservatives. The right resolves the trade-off between personal autonomy and social order in favor of the latter (Feldman, 2003). Left to their own devices, individuals acting freely will not produce a safe and secure social order in conservatives’ minds. They therefore greatly stress on conformity and tradition because diversity and change are seen as threats to social cohesion and stability (Feldman & Stenner, 1997). This explains the stronger support of the right for restrictions on civil liberties, harsher criminal justice and traditional moral norms that restrain individualist impulses (Inglehart & Flanagan, 1987; Kitschelt & McGann, 1997).
In an alliance, though, only a powerful state with more leverage can impose this type of hierarchy. We would therefore expect that right-leaning parties leading more powerful alliance members will be associated with institutionalization of this hierarchical, anti-egalitarian type. Such an arrangement allows the stronger party to reap the gains of specialization and cooperation without the sovereignty costs that loom larger for the rightist parties to an alliance. Hierarchical terms would include non-reciprocal provisions allowing the stronger state to base its forces in its ally’s territory, or placing the weaker state’s forces under the strong state’s command in the event of conflict. Right-wing governments in weaker potential alliance members, however, would oppose such a subordination as they will be less trusting of the more powerful partner. In a way not true of leftist-led countries, our expectations about the preferences of rightist parties in government are therefore contingent on whether they govern a weaker or more powerful state.
Conversely, when it comes to hierarchical forms of institutionalization, leftist governments leading stronger partners should feel it less necessary than rightist parties to assert control, but leftist governments leading weaker partners should be less sensitive than rightist governments when others do so to them since they have less concern about sovereignty costs.
Hypothesis 2: Among stronger alliance partners, right-wing governments will be positively associated with alliances featuring institutionalized hierarchy compared to a non-institutionalized alliance.
Hypothesis 3: Among weaker alliance partners, right-wing governments will be negatively associated with alliances featuring institutionalized hierarchy compared to a non-institutionalized alliance.
Analysis
Dependent variables
We utilize Alliance institutionalization, operationalized according to previously described criteria developed by Leeds & Anac (2005), as the dependent variable. However, instead of treating this outcome as ordinal, we analyze it as a three-category nominal variable. What Leeds & Anac (2005) consider ‘category 3’ alliances, based on the criteria enumerated above, we think of as ‘hierarchical’ in character. Of the 166 alliances in the data (discussed below), there are 19 such alliances. They allow for instance the stronger ally control over the weaker: they feature non-reciprocal basing rights, meaning that the stronger state can place troops on the territory of the weaker, but not vice versa. In addition to non-reciprocal basing rights, two of these 19 alliances also provide for the forces of the weaker state to be subordinated to the command of the stronger in the event of conflict. Leeds & Anac (2005) include military subordination as sufficient to warrant a level 2 ordinal coding, a provision we instead consider hierarchical. However, none of the alliances in our data coded as level 2 in Leeds & Anac’s (2005) scheme feature subordination, allowing us to use their categorization without adjustments. Level 2 alliances, of which there are 35 in the data, feature the types of terms we define as being in the ‘voice-driven’ nominal category: consultation and coordination obligations and/or participation in formal organizations created by the alliance. If an alliance features none of the above obligations, it is considered ‘uninstitutionalized’, what Leeds & Anac (2005) define as level 1 (112 cases).
Explanatory variables
Our explanatory variables are categorical measures of democratic alliance members’ political ideology at the time they entered into a bilateral alliance, based on continuous ‘right-left’ (RILE) scores obtained from the Manifesto Project (MP) dataset (Volkens et al., 2014). RILE scores are suitable for comparing democratic governments’ political orientations across states. Although the Database of Political Institutions (Cruz, Keefer & Scartascini, 2018) codes ideology for democratic and non-democratic states, we choose to focus on democracies coded in the MP data primarily for reasons of construct validity (see Online appendix). To be assigned a RILE score, a state had to receive a score of 6 or greater on the polity measure in the Polity IV dataset (Marshall, Jaggers & Gurr, 2002) at the time it entered an alliance. Higher scores indicate greater conservatism, with scores in the data ranging from –31.3 to 38.15 (mean 0.127; sd 16.12). 1 Using the national election-year platforms of political parties that held at least one seat in a state’s national legislature, RILE is constructed by adding the frequency counts of references to numerous issues in party programs which load on to ‘left-wing’ and ‘right-wing’ factors. These issues are particularly well suited to our theory linking ideology to institutional preferences. The left is defined by positive references to social services, education, and labor groups that imply greater egalitarianism (Laver & Budge, 1992: 26–27). The right is defined in large part by its references to social conservatism, defined largely in terms of deference to authority and support for traditional morality. Both are tools used by the right to maintain order, necessary given the lack of trust in their worldview. International egalitarianism is also an element of the RILE score with references to internationalism and other elements of peace and cooperation loading on the left side of the dimension. Positive references to the armed forces, on the other hand, load on the right end of the dimension, an indication of the greater distrust with which the right approaches international affairs. Right-wing items include references to capitalist economics, social conservatism, human rights couched in terms of freedom, and positive references to the military. References that load consistently on the left-wing factor are those to state intervention in the economy, international peace and cooperation, and democracy. The 39 states included in our data when calculating RILE scores are listed in the Online appendix.
The unit of analysis is the bilateral alliance, drawn from the ATOP data. The years of formation for the 166 alliances in our data range from 1946 to 2003. Bilateral alliances without democratic states at the time of formation are excluded, with ‘democratic’ defined as states with a polity score of at least 6. All alliances meeting these criteria within the aforementioned time frame were included regardless of the substance or conditionality of their terms. That means that an alliance with at least one democracy would be included regardless of whether it required its members to actively defend each other – with or without certain specified conditions being satisfied – or (un)conditionally assist each other in offensive military campaigns, or whether the alliance had (un)conditional consultative, non-aggression, and/or neutrality requirements in event of conflict. Though we did not exclude alliances based on their terms, every observation in our data had at least consultation or neutrality requirements; put another way, no alliance was solely a non-aggression pact.
Frequencies of alliances by ideological composition and power distribution
Each alliance in the data contains at least one democracy with a continuous RILE score, whereas in the data, 119 of the alliances contain one state with no RILE score. In 65 of these cases one of the alliance members without a RILE scores is a non-democracy, whereas in the remaining cases one of the democracies did not have any parties with RILE scores in the MP data. Such states were generally newly established democracies. The frequent presence of states without RILE scores in the data makes it difficult to assign continuous ideology scores to the alliances, therefore we use categorical variables. RILE scores of 10 and –10 were used as thresholds to determine whether or not to code a democratic state as having a right-of-center (RoC) or left-of-center (LoC) government, respectively. Scores in between are characterized as ideologically moderate. The 10 and –10 thresholds have been used in previous research examining the relationship between state ideology and conflict (Koch, 2009: 806; Koch & Cranmer, 2007: 321). For every alliance, we include variables on the ideological character of both the stronger and weaker member, given our theoretical expectations. In other words, we have two sets of categorical variables and cluster standard errors as a consequence. Moderate governments, both weak and strong, are used as the comparison category in our multinomial estimations. Relative strength is operationalized using the National Military Capabilities data (NMC; Singer, 1987). Since the alliances in our data are bilateral and each ally has four possible ideological codings (LoC, RoC, no RILE score (NR), or moderate) and two possible relative strength codings (weak or strong), our models include six dummy variables, with moderate as the base category. The distribution of the alliances can be seen in Table I.
Control variables
We include several different sets of control variables when estimating the relationship between states’ political ideology and our dependent variables. First are variables related to threats faced by states in an alliance. States facing high levels of threat may desire greater commitments and more institutionalized alliances from their allies given fears of abandonment; alternatively, states considering an alliance with a country facing many threats may worry about entrapment and have more institutionalized alliances. Because threat should be associated with greater fears of opportunism, we expect that it will be correlated with hierarchical alliance design. Models carry a dummy variable for whether or not an alliance was formed during the Cold War, defined as any year from 1947 to 1989. The Cold War was arguably a time of greater threat, given the salience of nuclear warfare and the decline in international violence since (Goldstein, 2011). Alliances within the data were more institutionalized during the Cold War period, and there were also proportionately fewer RoC regimes appearing in the data during that era. Thus, any negative correlation between rightist ideology and institutionalized alliances may be a result of the relative paucity of right-wing governments in the observations at that time. We also include data on the total number of rivals for each alliance, using the definition and data on rivalry from Klein, Goertz & Diehl (2006). Total rivals is the number of rivals of state A and state B at the time they entered into an alliance with one another. Lastly, we account for the number of militarized interstate disputes, or total MIDs (Palmer et al., 2015) in which member states had been involved in the previous two-year periods before entering the alliance. Total MIDs are highly correlated with Total rivals (r = 0.74). However, the results pertaining to the explanatory variables are not altered by excluding one or the other of these two control variables, nor are they altered by using an alternative measure of threat from Mattes (2012b: 696; results not shown).
The next set of control variables are included to capture the amount of interest overlap that two allied states have with one another when forming an alliance, as indicated by the similarity between their international relationships and preferences. We expect that greater similarity of interests between states would translate to greater levels of trust. This would lessen the need for hierarchy, and perhaps correspond with more voice opportunities within an alliance, since similarity of interests lowers the transaction costs involved with consultation and negotiation. The degree to which allies’ interests are aligned is measured with the global S correlation for each pair (Signorino & Ritter, 1999). This measures the similarity between the states’ alliance portfolios in the year they entered an alliance. Next, we include the distance between allies’ estimated ideal points at the United Nations (UN distance) the year the alliance was formed, based on the analysis of UN voting data by Voeten, Strezhnev & Bailey (2013). These two measures have a moderate negative correlation (r = –0.32). Third, we include a dummy variable capturing whether or not both states in the alliance were democracies (Joint democracy), which was true for 60% of observations. Democracies have contracting advantages that may lower the costs of highly institutionalized agreements (Lipson, 2003).
Lastly we use Mattes’s (2012b: 693) Prior violation variable that captures whether or not the leader of one of an alliance’s members had been in power when that state had previously violated the terms of another alliance (see also Leeds, Mattes & Vogel, 2009). Such unreliability affects states’ demands for assurances through alliance institutionalization, as well as impacting states’ abilities to enter into alliances in the first place (Crescenzi et al., 2012; Gibler, 2008). We expect prior violations to be associated with greater institutional hierarchy. We also interact prior violation with power symmetry, the latter variable based on the alliance members’ status as major or minor powers (Small & Singer, 1982). Symmetry is coded as 1 if both allies are minor powers or both are major powers, and 0 otherwise. Mattes (2012b) finds that, even if a state has previously violated the terms of another alliance, asymmetric alliances it enters into may not be institutionalized. This is because minor powers lack the ability to compel unreliable great powers to make stronger alliance commitments. Conversely, great powers may be able to leverage unreliable minor powers’ dependency on them to secure commitment to an alliance without absorbing the costs of negotiating the terms of an institutionalized security relationship. Along with symmetry, we control for an alliance’s capability ratio, the log of the strongest state’s military capabilities divided by the weaker ally’s capabilities as measured by the NMC data (Singer, 1987). Though there is conceptual overlap between symmetry and capability ratios, the two variables are distinct. Singer & Small (1966) stress that major or minor power status has more to do with collective international perceptions than material capabilities: ‘[t]hough [status] may correlate with certain inherent and objective properties, capabilities, or skills of the actor in question, it need not’. Our data also contain considerable variation in capability ratios within both symmetric and asymmetric alliances: the standard deviation of the logged capability ratio of the strongest to weakest powers in symmetric alliances is 1.127 (min.: 0.081, max.: 6.904) and the standard deviation in asymmetric alliances is 1.649 (min.: 0.050, max.: 7.762).
Estimation
Effect of ideology on alliance institutionalization with baseline of uninstitutionalized
Multinomial logistic regression. Baseline of comparison is ‘uninstitutionalized’. Weak moderate and strong moderate party government are the excluded categories. Coefficients followed by robust standard errors clustered on state dyads. † p < 0.10; *p < 0.05; **p < 0.010; ***p < 0.001. Two observations in Model 2.3 completely determined so standard errors are questionable. A model without control variables, as well as a model including UN distance, produce variance matrices that are non-symmetric or highly singular, preventing the estimation of standard errors, and they are thus excluded from the estimate.
As expected in Hypothesis 3, the presence of weak RoC members was negatively associated with hierarchy in two of the four models. As predicted, rightist parties in government seem to be particularly sensitive to hierarchical alliances when in a weaker position.
The results for strong democracies provide some support for our hypotheses as well. Looking at the relative probability of alliance institutionalization being voice-driven rather than uninstitutionalized, the results show that strong LoC democracies are more likely to be in alliances featuring voice opportunities than are strong moderate democracies. This is consistent with Hypothesis 1. In alliances with NR states, there is a 45% chance of an alliance being voice-driven when it contains a senior, LoC partner, a probability that falls to just 4% with a moderate-led stronger partner. However, alliances with strong RoC members are also more likely to exhibit institutional features enabling consultation and cooperation than are those with strong moderate democracies, a point we return to below.
When it comes to hierarchical alliances, strong right-leaning governments are more predictive than moderate or leftist ones, as expected in Hypothesis 2, but the difference between left and right is not significant. Nevertheless, in every model the difference between the coefficients for the strong LoC and strong RoC variables is in the expected direction: the discrete effects on alliance institutionalization of an alliance featuring a strong LoC member are estimated to be stronger in the voice-driven category, whereas the discrete effects of an alliance having a strong RoC member are stronger in the hierarchical category.
We suspect that these somewhat weaker results might be driven by the lack of a more granular measure of voice opportunities in alliances. We construct a second dependent variable, Voice, a three-level ordinal variable measuring the number of voice opportunities created by an alliance’s terms as indicated by provisions for military consultation and the creation of organizations associated with the alliance agreement. Voice is coded 1 if an alliance does not provide for peacetime military consultation, nor creates any organizations for consultation (65 cases); 2 if the alliance provides for peacetime military consultation or creates one organization (61 cases); and 3 if the alliance provides for peacetime military consultation and creates one or more organizations or does not require peacetime military consultation but creates two organizations (40 cases). This variable allows us to leverage the full range of variation in a way that our other dependent variable does not. We expect alliances with strong left-wing members to be associated with more voice opportunities. Full results are presented in Table III.
This alternate dependent variable Voice helps to further distinguish between alliances with strong LoC and strong RoC members. The former has a strong effect, the latter none compared to a baseline of an alliance whose strongest member is ideologically moderate. This is consistent with Hypothesis 1. As shown in Figure 1 (based on the results in Model 3.5) the discrete effect of adding a strong LoC member to an alliance, holding control variables at the modal or mean values, is to decrease the probability that an alliance has no formal provisions for voice opportunities by about 20%, and increase the probability an alliance has two or more voice opportunities by almost 30%. This creates more confidence that the relationship between ideology and voice-driven design is not curvilinear. Once we create a finer measure, this effect disappears.
Might selection effects be biasing our estimates? When trust between two states is low, cooperation in an alliance may not be possible even with costly institutionalization. We thus would not observe alliance formation between such states. Alternatively, when bilateral trust is very high, alliance formation may not be necessary, as the states in question may believe they can rely on one another for security without expending resources designing formal security agreements or institutions. Either possibility means that the states in the alliances we observe may be more or less trusting of one another than the typical dyad in the international system. For this reason, in the Online appendix we further detail a set of selection models using the 166 alliances we observe along with a larger random sample of dyads whose states are not allied. We find that our substantive results are unaffected when controlling for selection effects.
Turning back to our original multinomial analysis, our expectations were met with respect to our control variables for threat. Both the coefficients for the Cold War period and the total rivals held by members of an alliance in Table II indicate that higher levels of threat, and thus fears of opportunistic behavior by allies, makes hierarchical alliance design more likely relative to an uninstitutionalized alliance. Conversely, the variables we expected to be associated with greater levels of interest similarity and thus trust (global S correlations and joint democracy) were not associated with a greater probability of voice opportunities relative to uninstitutionalized alliances, nor were they associated with a lower probability of hierarchy except for global S in Model 2.2.
Ideology and the number of voice opportunities
Ordered-logistic regression. Weak moderate and strong moderate party government are the excluded categories. Coefficients followed by robust standard errors clustered on state dyads. † p < 0.10; *p < 0.05; **p < 0.010; ***p < 0.001. One observation in Model 3.4 completely determined, standard errors questionable.
In the Online appendix we consider three other questions. First, do leftist and rightist governments have different preferences over institutional design when their partners are democratic as opposed to autocratic? We find that our hypotheses, while broadly confirmed, are truest of jointly democratic alliances. Second, we examine whether militarily stronger democracies are better able to get the types of institutions we theorize they should prefer when their power relative to their ally is greater, or alternatively whether greater power imbalances induce sovereignty concerns among weaker states – those governed by right-wing parties in particular – making any Effect of strong leftist-government presence on alliance voice
Conclusion
The preceding analysis shows the importance of considering the role of domestic politics, specifically party ideology, for explaining variation in alliance design. It also demonstrates the need to investigate the qualitative differences between different types of institutionalization, rather than only conceiving of security arrangements like alliances as ‘more’ or ‘less’ institutionalized in an ordinal sense.
Our study does have an important limitation in that many of the alliances under study included non-democratic states, about whom we have not developed any theoretical expectations. One strategy would have been to think for instance of non-democratic states as arrayed along a left–right ideological spectrum, for instance marked by fascism on the right and communism on the left. Based on our conceptual understanding of these types of regimes from history, however, they fit uneasily on a standard left–right continuum. Fascists are qualitatively, not quantitatively, different from conservative authoritarian governments in their appeal to the masses and aim to upend existing authority structures and replace them with others. Fascists, given their strong corporatism, actually have socialist tendencies, though confined to one’s own nation-state (Mann, 2004). Communist regimes are also qualitatively different from social democratic ones. While both are committed to equality, the former have no conception of individual rights; rather the individual exists merely to serve the community personified by the state (Howard & Donnelly, 1986). How these types of states might approach alliance institutionalization is an open theoretical question. Both are typically premised ideologically on a sense of zero-sum conflict with enemies from within and without. This likely makes both forms of regime less trusting. However, in that communism is an internationalist ideology and fascism a largely nationalist one, communist states might be more willing to countenance voice-driven organization among fellow communist states in a way not true for fascists. We urge others to take up this research agenda, furthering the exploration of how ideology affects alliance choice and design. We are all parties to an alliance committed to a better understanding of security cooperation.
Footnotes
Notes
Replication data
The dataset, codebook, and do-files for the empirical analysis in this article, as well as the Online appendix, can be found at http://www.prio.org/jpr/datasets. The Online appendix is also available on Brian Rathbun’s website: 
.
Acknowledgements
The authors thank the audience at the International Studies Association for helpful comments.