Allying to Win

Alliances have long been recognized as a key component of the study of international security. Realists in particular view alliances as one of two dynamics conditioning the structure of world power.Footnote 5 Countries can balance ‘internally’, forging military might through their own economic capacity and national determination, or they can contrive ‘external’ balances through foreign alliances. Nations blessed with neither size nor prosperity depend as a matter of course on informal or formal alignments to pursue the national interest.Footnote 6 Large or capable nations have the luxury of arming, but building a more powerful military is expensive and can raise considerable difficulties with other nations, where uncertainty and internal balancing invoke effects of the security dilemma.

Waltzian ‘defensive’ realism in particular emphasizes the need for states to be circumspect about arming.Footnote 7 States are not free from harm if their quest for power causes other nations to form opposing coalitions.Footnote 8 Yet, it is not obvious from Waltz's theory how balancing can be generated adequately among egoistic states. While balancing may well be sufficient to achieve the social good of international stability, egoistic states are presumably better off letting others do the ‘heavy lifting’ needed to constrain common enemies.Footnote 9 Just as nations can free-ride within alliances, they can presumably also free-ride by not joining balancing coalitions. Since security and particularly stability cannot be denied to states that do not participate in balancing coalitions, and since balancing requires social action to be realized, defensive realism is in the odd position of juxtaposing anarchy and individual rationality with a theory that predicts collective action to provision a public good.

Power-seeking states can presumably pursue alliances for private benefit, but the strongly zero-sum nature of competition in offensive realism poses other challenges.Footnote 10 The neo-mercantilist critique of liberal trade theory is precisely that concerns about relative gains stymie co-operation under anarchy.Footnote 11 Nations that cannot co-operate over commerce because some participants will get more must similarly find it difficult to co-operate over security, where relative gains concerns are even more intense. To prefer to ally, a power-seeking sovereign and its prospective partner must each expect to obtain more power from an alliance, which of course should not be possible.Footnote 12 The realist preoccupation with zero-sum competition means that allies should seek to shirk alliance costs, while hoarding benefits. Powerful states should seek to capture any surplus, or compel corresponding transfers from allies, since any benefit foregone is a future disadvantage in the zero-sum competition of world affairs. As Niou et al. note, victory can be particularly hazardous for an alliance.Footnote 13

As Powell and Snidal explain, differences between relative and absolute gains diminish as the number of participating states increases.Footnote 14 Two countries can fail to trade if each insists on receiving a majority of the surplus. Add a third relative gains egoist and all nations can prefer obtaining some benefit from commerce, rather than accepting reversion to a status quo that makes two of the three nations strictly worse off. This also implies that alliances are possible in an n-state world, but relative gains concerns reassert themselves if the effects of power cannot be completely internalized by the alliance. Defeating an enemy helps not just states that actively participate in a contest, but any other nations threatened by the defeated state. Allies gain to the extent that victory can be converted into private benefits (plunder, territory, trade), but lose when victory is costly and where benefits are public (global or regional stability or security, norm enforcement). This difference between economic co-operation (where benefits are largely internalized) and security co-operation (where benefits may be public) means that explaining alliances among power maximizers is arguably more challenging than explaining the presence of international trade.

Even if egoistic states can find common ground to ally, it is far from clear who they should choose as partners. With dozens of nations in the international system, there exist thousands of potential alliance coalitions. Less than 1 per cent of the potential dyadic pairings contain an actual alliance. This seems extraordinarily choosy of countries if alliances are simply aggregating capabilities. With almost all potential alliance combinations unfulfilled, any state with narrow security interests would do well to cast its net farther afield. If instead, as seems apparent, alliances are formed among a much more restricted set of potential partners, then some additional factor beyond security or capability aggregation must be driving the high degree of selectivity in forming alliance dyads.

By the same measure, it seems very unlikely that alliance decisions are driven by uniform considerations across countries. Waltz's own theoretical framework of internal versus external balancing suggests that the decision to ally cannot be driven by power aggregation alone.Footnote 15 Sorokin offers a constrained-optimization model of the choices states make in constructing a security plan that mixes arms and alliances: ‘[W]hen a state relies on its own arms, it decides whether and in what way to use them; when it relies on allies, it may have access to a larger pool of capabilities, but it sacrifices control’.Footnote 16 If states all share the same security preferences, then it is not clear why they should differ in their response to the tradeoff between capabilities and control.

One possibility is that alliance selectivity and differential responses to the alternatives of arming and allying simply reflect the skewed distribution of power. Few states are sufficiently capable to add significant military potential to an alliance. Major and regional powers should be much in demand as allies, and they are. However, large, prosperous countries are also relatively intensive internal balancers, typically spending a higher proportion of gross domestic production (GDP) on defence than smaller states.Footnote 17 Weak nations constitute the majority of states, and the majority of alliance partners, as Altfeld, and later Morrow, point out.Footnote 18 Asymmetric alliance ties are much more common than pairings of roughly comparable states. Morrow explains these asymmetric alliances in terms of complementary ‘alliance goods’. Weaker states want security while capable partners seek autonomy (i.e., influence). This conception and the supporting evidence reveal a more varied set of state objectives. Nations are not only power or security seeking, but are often willing to ‘trade’ security for other objectives, though doing so presumably dilutes a capable nation's ability to protect its own territory.

While the notion of alliance goods is informative, our ability to predict which countries ally remains limited. Knowing that alliances more frequently involve unequal powers does not tell us which of the many possible unequal alliances are most likely to form. Any attempt to explain the origins of alliances must ultimately confront the question of state preferences. Indeed, power may be instrumental; leaders perhaps view power or security not unlike how consumers or firms view income. The objective in accumulating these assets has less to do with each as an end in its own right and more to do with exercising power or security as the medium through which political goals can be realized. Powerful nations are not fortunate simply because they are powerful, but because power allows them to create a world that more nearly suits their interests or preferences. As such, we must know something of national preferences or interests to understand international behaviour.

Finding Common Ground

Smith, following Bueno de Mesquita, claims that alliances form between ‘friends’.Footnote 19 Alliances are costly, with the costs conditioned to a considerable extent by a state's choice of partners.Footnote 20 Countries with similar political, economic, social or ethnic characteristics should be able to exercise influence or maintain security with the mildest tradeoffs in terms of compromised policies or abrogated autonomy. Patrons that share similar interests with their protégés can be more confident that their allies will not drag them into unwanted conflicts, while protégés need worry less about being exploited or abandoned by their patrons to the degree that each shares share similar preferences.Footnote 21

If friends ally, what is the basis for friendship in international affairs? A multitude of issues could conceivably generate affinities or animosities among states. For example, scholars have debated whether trade follows the flag, or whether alliance ties reflect existing trading relationships.Footnote 22 Others claim that foreign policy preferences, and international cleavages, reflect ethnic affinities or differences or cultural biases endemic to diasporas in domestic politics.Footnote 23 Rather than propose a particular hierarchy of preferences, we adopt a dimension along which affinities and cleavages generally are believed to exist, and then discuss how these interests respond to an evolving environment.

The most prominent research programme relaxing the assumption that states are all uniform in their objectives and interests involves the democratic peace. If democracies are less likely to fight with each other than with non-democracies, perhaps democracies also show a special affinity for other democracies in forming formal national security bonds. Similar preferences could derive from selection; elites or populations in democracies share in common the decision to create and sustain popular rule. Alternatively, the norms or institutions of democracy may be responsible for fashioning common foreign policies, or a natural compatibility may form from like regimes. Democracies arguably face fewer compromises in forging common security bonds. As such, democracies would seem to constitute ‘friends’ in the very sense proposed by Bueno de Mesquita, Smith and rationalist alliance theories generally.Footnote 24 Certainly, the popular conception, championed by leaders like Woodrow Wilson and Franklin Delano Roosevelt and epitomized by institutions like the North Atlantic Treaty Organization, is that democracies naturally gravitate towards one another as allies.

Surprisingly, a preference among democracies for democratic allies is far from an established fact. A number of studies examine how regime type affects alliance choice. Walt argues that states with similar political characteristics should prefer to co-ally, but finds no support for this argument in a study of thirty-six alliances.Footnote 25 Walt however selects on the dependent variable and blurs the distinction between informal alignments and formal alliances, making it difficult to evaluate his conclusions.Footnote 26 In contrast, Siverson and Starr find that states tend to change their alliance ties after regime change, though the substantive effect of regime transition is quite small.Footnote 27

Siverson and Emmons offer a direct test of the impact of joint democracy on decisions to ally.Footnote 28 They find differing results in analyses of different time periods.Footnote 29 On balance, however, they conclude that democracies prefer democratic allies. Thompson confirms the results of Siverson and Emmons by using a different dataset of regime type and a longer time frame.Footnote 30 Both studies base their conclusions on a comparison of the proportion of alliances between democracies. Simon and Gartzke find instead that differing regime types prefer to co-ally.Footnote 31 They point out problems in the inferences made by Siverson and Emmons and by Thompson, as the availability of alliance partners is constrained by the distribution of regime types and by the sample properties of states seeking to ally. By segmenting the alliance data into regime type categories, Simon and Gartzke find that alliances between like regimes are much less frequent than alliances between regimes of differing regime type. Lai and Reiter examine ten hypotheses related to regime type and alliance formation drawn from three theoretical perspectives: constructivism, economic interdependence and credible commitments.Footnote 32 The study controls for a variety of possible confounding covariates, such as culture, threat, trade and learning. The authors find that states with similar regime types are more likely to co-ally, but that democracy per se is not unique as a determinant of nations’ alliance choices.

A similar picture emerges from the examination of other alliance-related variables. Several studies have found that joint democracy is associated with increased alliance durability, but these studies have reached differing conclusions about the monadic effect of democracy.Footnote 33 Gaubatz, for example, finds that democracy is not a statistically significant predictor of alliance durability, while Reed concludes that alliances with more democracies are more durable. Work on the relationship between democracy and victory has reached similarly divergent findings, with Reiter and Stam concluding that the relationship is not a function of the tendency for democracies to come to one another's aid, while Choi concludes that democracies make far more effective partners in war.Footnote 34 Similarly, while Leeds finds that democratic states are more likely to uphold alliance commitments, Gartzke and Gleditsch use data from a broader time series to argue that informational and institutional features of domestic politics make democracies less reliable alliance partners.Footnote 35 The relationship between regime type and alliances is thus less clear than the friendship logic suggests.

The End of a Friend

Perhaps one reason that it has proven difficult to document a connection between regime type and alliance affinity is that the salience of regime type as a motive for friendship has changed over time. Affinities and antagonisms change with the proximity and intensity of threats. The old adage that the enemy of an enemy is a friend speaks to the shifting relational nature of politics under anarchy. At times, as with the alliance between the Western powers and the Soviet Union in the Second World War, strategic realities produce incentives to co-operate in the absence of similar underlying preferences. If two dissimilar nations ally against an opponent whose preferences are even more divergent from either ally, then one still must reference preferences to explain a core link in the chain. Moreover, as the contrast between Anglo-American trust and Western–Soviet mistrust makes clear, alliances of conviction will typically prove more durable and more effective than alliances of convenience.

The lack of robust evidence for a connection between democracy and alliance choice could be explained if the forces of friendship among democracies have yet to gather sufficient momentum. A number of scholars have argued that democracies have created a collective security community whose ‘tightness’ can be expected to grow as the number and strength of states in the community increases.Footnote 36 Collective security implies that democratic alliance ties should multiply as more democracies enter the international system. Yet, while democracies do appear to be protecting one another informally,Footnote 37 it is not clear that there is an increasing tendency for democracies to co-ally.Footnote 38

Another possibility presents itself if we address the enemy-of-an-enemy logic a bit more earnestly. The argument is not that friendship causes alliances, but that enmity causes states to look to ally with partners that are relatively friendly. By seeking to characterize the supply of allies, while ignoring demand, the collective security perspective is missing half of the equation. Existing rational choice approaches do consider the role of threats, but prominent interpretations appear to ignore the dynamism inherent in this conception of friendship. If alliances are formed substantially in the presence of enemies, then who one's friend might be depends on the nature of one's enemies. Yet, if states are friends or enemies only in relative terms, then affinity or enmity must also be contextual, at least in part. Changes in the availability of affinities or animosities among states should lead states to reconsider which states constitute enemies, and which friends. Put simply, the incentives for any two states to identify as friends and to ally should be considered temporary.

By construction, as more states convert to democracy, fewer must remain as autocracies, reducing the number of unlike regimes. The decline in the size or strength of threats should weaken the bonds uniting democratic states. This effect is enhanced by the fact that democracies tend to cluster geographically.Footnote 39 Enemies are abundant or absent (or at least distant), alternately heightening or diminishing the need for strong security arrangements among friends.Footnote 40

We can take this argument even further by recognizing that regime type is only one dimension along which states might find common identity/difference and a reason to co-operate. Constructivist theories emphasize the role of the ‘other’ in coalescing social preferences.Footnote 41 Having a threat that is clearly opposed to the interests or actions of a given group of individuals or countries creates a catalyst for the formation or resurrection of a particular identity. During the Super Bowl, many Americans find themselves coalescing or dividing up along fan loyalty lines. At other times, cleavages appear across class, race, partisanship or geography. For most people, the salience of the Super Bowl identity is temporary, with other identities resurfacing not long after the game ends.

The recognition that identities are fleeting, or at least contextual and temporary, can have important implications for the study of international relations. Democracy was the rallying cry of Western nations in the twentieth century. This may continue, but as more nations adopt institutions of popular suffrage, being a member of the democratic club is no longer quite so exclusive. Pundits have in recent years begun to distinguish between different kinds of democracy. ‘Illiberal democracies’ are said to be different, a distinction that effectively bisects the inclusiveness of the democratic label.Footnote 42 While elections have been an important unifying principle, popular rule can still lead to leaders that are highly unpopular elsewhere. An election victory by Hamas in 2006 in the Gaza Strip led to, if anything, worse relations between the United States and the Palestinian Authority than had previously been the case under the autocratic Arafat regime.Footnote 43 Similar transitions have occurred in modern European history. Pan-Slavism swept the Balkans at the beginning of the twentieth century, as national groups united in opposition to domination by the Ottoman and Austro-Hungarian empires. The Pan-Slavist movement yielded Yugoslavia in the wake of the First World War. Yet, pan-Slavism turned out to be fragile. With the passing of Josip Tito and the rise of post-Cold War geo-politics in the 1990s, Yugoslavia fractured along other identity lines.

If alliances arise in world politics in response to an enemy or an ‘other’, then victory is just the prelude to some new conflict along lines that have yet to assert themselves. In the midst of the Cold War, it made perfect sense to arm the Afghan Mujahideen in their struggle against the Soviet Union. With the collapse of the Soviet Union, the basis for common interest between Washington and fundamentalist Islamists was eradicated. Animosities that had remained dormant when both groups faced a common foe reasserted themselves. The galvanizing effect of the 9/11 terrorist attacks was to make it seem as if differences between the West and fundamentalist Islam were ancient, deep and immutable. Today, the idea of finding common purpose with the Mujahideen appears bizarre, just as current policies regarding as yet unanticipated foes that seem entirely justified in today's political climate might someday prompt congressional inquiries or blue-ribbon fact-finding panels.

Modelling Evolving Alliance Friendships

Below we use a simple formal model to help illustrate our argument. While past studies have frequently relied on game theory to model alliance behaviour, our interest in changing systemic dynamics, with numerous actors, militates against such an approach. A fully game-theoretic model would impose strong rationality assumptions (given the requirement that actors anticipate the possible alliance decisions of all other actors in the system), making such an approach intractable for our purposes. For this reason, we instead rely on an agent-based model.Footnote 44 The goal of this model is to determine whether a simple set of assumptions about actor characteristics and interactions produces a consistent pattern of behaviour as attributes of the system in which actors operate change, even while allowing for variation on a number of relevant variables. More specifically, we are interested in how the probability that similar-regime actors ally is affected by changes in the system in which they interact, most notably the proportion of system actors that are democratic.

In the model, a set of actors with defined capabilities and with preferences that can diverge on multiple dimensions interact with one another to produce policy outcomes, from which they derive utility. Policy outcomes in each interaction are determined by the relative capabilities of the two sides; capable actors are able to produce policy outcomes closer to their ideal point, generating greater utility for themselves and correspondingly lower utility for their opponent. An individual actor's utility is thus its summed utility from interactions with every other actor in the system.

Alliances permit actors to aggregate capabilities and thus potentially to produce preferable policy outcomes in interactions with actors outside the alliance. Alliances also have a downside, however. Alliance partners must agree on a common policy objective – what we call the alliance's ‘effective ideal point’ – that, unless allies have identical preferences, necessarily means that outcomes will differ from the ideal point of at least some members of the alliance. In other words, the benefits associated with the alliance's greater capabilities are at least partially offset for actors by the need to compromise with other alliance members about the policies the alliance will pursue.

This inherent tradeoff to alliances is an essential feature of the model, and is thus worth illustrating with a simple example. Figure 1 represents an alliance decision in a simple system consisting of three actors, A, B and C, with ideal points in two dimensions (labelled for simplicity as regime and economic) corresponding to their location in the Cartesian space, and with A having half the capabilities of either B or C. Consider A's decision about whether to form an alliance with either B or C. In the status quo situation of no alliances, A's interaction with C produces a policy outcome at the point labelled ‘A vs. C outcome’, which lies on the line between A's ideal point and C's ideal point, closer to C because of C's greater strength. Assuming perfect cumulativity of resources, an alliance with B by contrast would shift the policy outcome in A's interaction with C to the point labelled ‘A & B vs. C outcome’.Footnote 45 The addition of B's strength allows A to pull the policy outcome further from C's ideal point than otherwise would be possible; the tradeoff is that the outcome is pulled not towards A's preferred outcome but towards a different point determined by the combined preferences and relative capabilities of A and B. In this case, B's proximity and capabilities mean that this tradeoff benefits A. By contrast, A is unwilling to enter an alliance with C. Such an alliance, while it would have strengthened A against B, poses an unacceptable cost in terms of deviation from A's ideal point. As is evident in the figure, the point ‘A & C vs. B outcome’ is substantially further from A's ideal point than is the policy outcome (‘A vs. B outcome’) that would arise in the absence of an alliance. In the cases discussed below, of course, there are substantially more actors. In making alliance decisions in a complex world, actors must weigh the consequences of alliance formation for interactions with all other actors in the system.

Fig. 1

An example of alliance decisions in the model

A single run of the model thus consists of several discrete phases. To facilitate understanding, we first describe the model informally, and then present the formal details. Figure 2 presents a graphical summary of the process that is followed in a single iteration of the model.Footnote 46 First, the world is populated with a set of actors with preferences on multiple dimensions, and with capabilities that vary according to a power law distribution. This assumption is useful because it ensures that in the model there are typically a few powerful actors and a much larger number of less capable states, just as in the empirical world.Footnote 47 One dimension is specified to correspond to the actor's regime type, while other dimensions correspond to other non-regime preferences, as with differences of opinion about how to organize the economy or differing preferences about significant political issues like international law, the environment, etc. Those actors with a regime score above a standard threshold are characterized as democracies, while the remainder are autocracies. Note that regime type enters into the model only through actor preferences: all else equal, actors will be happier in their interactions with states of a similar regime than they will be in interactions with states possessing a different regime type, but democracies do not otherwise differ from autocracies.

Fig. 2

Overview of the simulation model

In the second phase, actors form alliances. Each actor, in random order, is given the opportunity to propose an alliance to another actor. In choosing the target of an alliance proposal, the proposer compares the utility that she would gain were the game to proceed to the conflict phase with the current alliance structure to the utility she would gain from entering the conflict phase allied to each remaining actor in the game. She then proposes the alliance that most benefits her, provided that at least one acceptable alliance exists. The target of the proposal likewise compares his expected utility from the current alliance structure to his utility in an alliance with the proposer, and accepts if the latter is greater than the former. Current allies of both the proposer and the target conduct a similar comparison, and have the right to veto any alliance proposal.Footnote 48 If the proposal is rejected, the proposer is given the opportunity to approach another actor. This process continues until either an alliance is formed or the proposer exhausts the list of attractive potential allies, after which a different actor who has not yet served as proposer is given the opportunity to propose an alliance.Footnote 49

In the final (conflict) stage, every actor interacts with all others (both allies and non-allies) to generate local policy outcomes, drawing on their allies’ capabilities in interactions with those outside the alliance.Footnote 50 Actors gain greater utility (or less disutility) from policy outcomes that are closer to their own ideal point. A given actor's overall utility is simply the sum of the individual utilities from all of his interactions. Ultimately, we are primarily interested in the results of the alliance stage of the model – indeed, as Figure 2 makes clear, the conflict phase is not formally included in the model because by the time it arises the alliance decisions that are of interest to us have already been made. However, because actors anticipate the implications of alliance decisions for the conflict phase, it is necessary to capture actors’ expectations about conflict in the model. These anticipations govern actor decisions about making and accepting or rejecting alliance proposals.

We next provide formal details of the model. Let us start with a set A of actors, where each actor i∈A is defined by capabilities c_i and ideal point $$)(--><$> {{S}_i}\, = \,\{ {{r}_i},s_{i}^{\sigma } \} <$><!--$$ , with r, s∈[0,1] and σ∈{1,2,3,4}. Actor capabilities are determined by taking a random draw from a uniform distribution over the [0,1] interval and raising the resulting value to the negative α, where 0≤α≤1 is an exogenous parameter that determines the degree of variation in actor strength, with α = 0 corresponding to uniform capabilities across actors and α = 1 constituting exceptionally high variation. For actor preferences, r_i represents regime type, while $$)(--><$> s_{i}^{\sigma } <$><!--$$ represents preferences in up to four additional issue dimensions. In a given iteration of the model, values for the additional dimensions $$)(--><$> s_{i}^{\sigma } <$><!--$$ are selected randomly from a uniform distribution over the [0,1] interval. To ensure sufficient variation in systemic democracy, a predetermined set of actors in each iteration are assigned to be democracies, with the remainder non-democracies. Given a threshold regime value r_θ separating democracies from non-democracies, regime scores for the democratic states are chosen randomly from a uniform distribution over (r_θ ,1], while non-democracies are similarly assigned a random regime value in the range [0,r_θ ]. Primary analysis and most robustness checks set r_θ = 0.85, which corresponds to 7 on the standard -10 to +10 Polity scale, but similar results follow any threshold value.

Define the subset $$)(--><$> {{F}_i} \subseteq A <$><!--$$ to be the set consisting of actor i and all of i's allies (friends), with ƒ corresponding to the number of members of F. The alliance's combined capabilities are captured by

where 0<k≤1 represents the efficiency with which alliance partners are able to aggregate their collective capabilities.Footnote 51 In primary analysis and most robustness checks, k is set to an arbitrary value of 0.9; as elsewhere, the basic results continue to hold for any value of k such that actors are ever willing to form alliances. This alliance has an effective ideal point $$)(--&#x003E;&#x003C;$&#x003E; S_{{Fi}}^{\ast} \, = \,\{ {{r}_{Fi}},s_{{Fi}}^{\sigma } \} &#x003C;$&#x003E;&#x003C;!--$$ that is the capability-weighted mean of member ideal points. Specifically,

The elements $$)(--&#x003E;&#x003C;$&#x003E; s_{{Fi}}^{\sigma } &#x003C;$&#x003E;&#x003C;!--$$ of $$)(--&#x003E;&#x003C;$&#x003E; S_{{Fi}}^{\ast} &#x003C;$&#x003E;&#x003C;!--$$ are defined analogously. Note that for an actor with no allies, $$)(--&#x003E;&#x003C;$&#x003E; S_{{Fi}}^{\ast} \, = \,{{S}_i} &#x003C;$&#x003E;&#x003C;!--$$ .

Given actors’ capabilities and preferences and each alliance's joint capabilities and effective ideal point, one can determine policy outcomes from actor interactions and actor utilities from those policy outcomes. In interactions between non-allies, the policy outcome lies on the line connecting each side's alliance effective ideal point $$)(--&#x003E;&#x003C;$&#x003E; S_{{Fi}}^{\ast} &#x003C;$&#x003E;&#x003C;!--$$ , with the specific point on the line a function of relative capabilities. Formally, each interaction produces a policy outcome O_ij with elements o_ij , where

Interactions with allies by contrast produce outcomes at the alliance effective ideal point $$)(--&#x003E;&#x003C;$&#x003E; S_{{Fi}}^{\ast} &#x003C;$&#x003E;&#x003C;!--$$ . Actor utility from a given interaction is then simply the negative distance from the actor's ideal point to the policy outcome. Formally, actor i's utility from a given set of alliance partners is calculated as:

The first term in this utility function captures the actor's utility from interactions with non-alliance partners, while the second captures utility from the ƒ−1 alliance partners (omitting the actor herself).Footnote 52 In every alliance formation decision, actors compare the utility that they would gain under the current alliance to what they would gain by forming a new alliance. Actors are open to forming a new alliance to the extent that formation of the new alliance raises their utility.Footnote 53

Figure 3 provides a simple illustration of alliance decisions in the model, using a relatively limited number of actors and restricting preferences to two dimensions. Twelve actors are represented by dots, with their regime type captured by their vertical location in the figure, their preferences on the second dimension by their horizontal location, and their capabilities by their size. Thin lines connect the ideal points of actors sharing an alliance. In the figure, the three uppermost actors (D₁, D₂ and D₃) are democracies. The first two are part of an alliance that also includes a non-democracy. D₃, however, holds sufficiently extreme preferences on the second dimension that it did not ally with another democracy (or any other actor). The model thus captures a world in which regime type provides a basis for co-ordination, but not the sole one. A typical run of the model involves several thousand iterations, with each iteration constituting a single round of the process described above. The model produces at least 1,000 observations of potential alliances at each level of systemic democracy. The result is dyad-level data containing actor preferences and capabilities, whether or not two actors were allied, and the proportion of the system that is democratic.

Fig. 3

Representative results from a single iteration ofall the model

Our interest, of course, is in what happens as changes in the attributes of member states change the composition of the international system. By altering the initial distribution of preferences, we are able to observe the effects of a changing system on the probability that similar-regime actors ally. Figure 4 summarizes the simulated probability of an intra-regime alliance for both democracies and autocracies at different levels of systemic democracy, along with lowess-smoothed lines for each regime type. In this run, the threshold value for democracy was set to 0.85, actors held preferences in four dimensions in addition to regime type, capabilities varied across actors to a moderate degree (α = 0.5), and alliances cumulated capabilities imperfectly but relatively efficiently (k = 0.9). As democracies become more common, they become substantially less likely to ally with each other. When they are scarce, democracies benefit disproportionately from co-operation, as the increase in strength is useful in a hostile world, while the loss of efficiency that results from pooling their efforts is small relative to the large value of leveraging combined capabilities to confront the large differences that democracies initially have with most opponents. As democracy proliferates, however, the decline in external threats makes the internal differences among democracies more salient. In contrast, autocracies become substantially more likely to ally as the system becomes more democratic. This basic dynamic should hold for any characteristic that provides a basis for distinguishing among states, such as monarchy or economic structure. Changing aspects of the model such as the number of actors, the way in which actor capabilities are distributed, how efficiently capabilities cumulate within an alliance, or the threshold regime value for democracy can alter both the probability of an alliance and the slope of each line, but the basic comparative static result from Figure 4 (captured by the sign of the slope) is robust to changes in model specification. Indeed, the relationship is stronger in most robustness checks.Footnote 54

Fig. 4

Probability that similar-regime actors ally at different values of systemic democracy

It is difficult to imagine at this point what might replace democracy as a critical organizing identity in world affairs, just as it must have been difficult in the early 1940s to imagine that China would first become one of the chief adversaries of the United States, then later an important counterbalance to Soviet power in Asia and the Pacific region, and finally a fast-rising global economic power. While it may not be possible to anticipate the nature of future cleavages – we could not test such claims even if we were inclined to make them – one can more easily surmise how the evolution of identity is likely to affect existing alliance structures. If the salience of democracy declines as a discriminating cue or identity with an increase in the number of democracies, then we should expect to see a decline in the tendency for democracies to co-ally. This prediction further helps to explain why the relationship between democracy and alliance choice is weaker empirically than liberal theory predicts. When democracies are scarce – i.e., when a democratic security community is at its weakest – democracies should tend to ally together at much higher rates than autocracies.

In contrast, the salience of autocracy as an identity for alliance formation should be on the rise as autocracies shift from being ubiquitous to representing a bare majority of the world's countries.Footnote 55

Research Design and Analysis

We assess the determinants of alliance status between 1816 and 2000. The basic statistical models are based on Lai and Reiter, with changes as discussed below.Footnote 56 Independent variables are lagged by one year to address endogeneity. We correct standard errors for clustering in dyads.

There are a few issues to address before proceeding. First, it is conceivable that changes in alliance patterns at the monadic level might distort a dyadic-level test of regime type and alliance formation. It could be argued, for example, that because democracies are safer today than in the past, they are generally less likely to ally. Since there are more democracies in the world over time, processes that lead democracies to participate in fewer alliances could be misinterpreted as evidence that democracies are becoming less inclined to ally with other democracies. Figure 5 details variation in the proportion of all dyads in the world containing an alliance over time (represented by hollow triangles), and the accompanying relationship for dyads containing at least one democracy (solid circles).Footnote 57 The figure also contains linear trend lines for both sets of alliance probabilities.

Fig. 5

Probability of alliance for democracies and all dyads

As Figure 5 reveals, the propensity to ally varies considerably over time, but there is no overall trend for alliance ties. Pairs of countries are not allying less over time. Democracies, in contrast, are actually increasing their propensity to ally (both with other states and with other democracies). This relationship is the opposite of the one that could conceivably bias findings in favour of our analysis. Dyads containing at least one democracy are more alliance-prone with the passage of time. Part of this relationship is due to the increasing number of democracies in the world. In the early nineteenth century, only the United States is coded by Polity as a democracy. American isolationism explains why dyads containing democracies initially have no alliances. By the early twentieth century, more states are democratic, and alliances involving democracies begin to occur. However, it is not until the Second World War that the United States begins to form its own ‘foreign entanglements’. Since the United States represents such a large and important potential source of democratic alliance partnerships, the unusual American policy of isolationism effectively biases against support for our claims. At the monadic level, democracies are allying more over time, not less.

Secondly, alliance data contain four kinds of observations. States in dyads can be: (1) unallied (‘0’), (2) newly allied (‘1’), (3) ongoing allies (‘1’), and (4) former allies (‘0’). These data do not distinguish between new and old alliances, or between non-allies and former allies. While we are not fundamentally interested in these distinctions here, focusing instead on alliance status, there are potential pitfalls in ignoring these distinctions. In particular, regressions that fail to distinguish between old and new alliances will tend to overestimate the effects of regressors on alliance choice.

The easiest solution to this issue is to include a lagged dependent variable to distinguish between new and ongoing alliances. This is precisely the approach adopted by Lai and Reiter. However, a lagged dependent variable effectively means that the regressors are estimating change in alliance status; alliance onset and termination become equivalent in this type of regression. While not a critical problem – our theory is compatible with alliance change – we consider this less than ideal given our focus on the determinants of alliance status. For example, estimating alliance change would mean that we cannot say anything about the direction of alliance transitions.

Another approach to addressing dependence in the data is to include a count variable for annual observations of no alliance (‘0’). A matrix of lagged dummy variables or splines can then be set up to capture complex dependence structures.Footnote 58 This approach works particularly well with episodic events, such as wars or crises, linked through non-event intervals. While it is conceivable that states, once allied, are more likely to ally again, the primary source of dependence in the alliance data are the persistence of alliance institutions themselves. Unlike the conflict data, long series of 1 s permeate the alliance data. We explored using ‘peaceyear’ counts, matrices or splines of the 0 s in the dependent variable, but as a practical matter these solutions to temporal dependence over-fit the data, preventing the estimator from converging on coefficient estimates.

Bennett and Tarry discuss the tendency for alliances to persist in terms of hysteresis, or institutional inertia.Footnote 59 This is a different process from that observed in episodic conflict. If path dependence in alliance time-series involves institutions, then it is best to address the biasing effect of the persistence of alliance ties. Carter and Signorino offer a practical technique for dealing with path dependence.Footnote 60 We created lagged linear, squared and cubed count variables of the number of years two states have been allies. Alliances tend to persist up to a point, and then the effects of change take a larger role, so that old alliances fail more often than medium-aged alliances.

Figure 6 reports the estimated effect of the linear and squared alliance count variables, based on regression Model 1.3 in Table 1 (other details are reviewed later). The figure is important because it confirms a quadratic functional relationship in these data and also provides some evidence that our approach addresses the biasing effects of temporal dependence. The horizontal axis is a count of the number of years that have passed since the alliance was formed. The vertical axis measures the probability that an alliance will endure another year, given the current age of the alliance. The cubed term is omitted as this specification over-fits these data. As the estimated relationship reveals, the probability of an alliance in a given year is not independent of the presence or absence of an alliance in previous years. The functional form is clearly quadratic, first increasing and then declining in previous alliance status. While it is impossible to know how a statistical model conforms to the ‘true’ underlying relationship, the linear and quadratic alliance year count variables appear at least partially to address the non-independence of observations in these data.

Fig. 6

The persistence and decay of dyadic alliance ties

Table 1

The Effects of Systemic and Dyadic Regime Type on Alliance Status

Notes: The dependent variable is the alliance status. Significance levels: * 5%, ** 1%, ***0.1%. All tests are two-tailed. † See Mark J. C. Crescenzi and Andrew J. Enterline, ‘Time Remembered: A Dynamic Model of Interstate Interaction’, International Studies Quarterly, 45 (2001), 409–31.

Data

Most of the variables for this study are generated using the EUGene software program.Footnote 61 Other sources are detailed where relevant. Data consist of dyad-year observations over the 1816–2000 period for all dyads in the international system. A STATA ‘do’ file is available from the authors that replicates the analysis. Methodologists also emphasize the value of keeping statistical models simple.Footnote 62 Therefore, we focus on variables that consistently appear in other relevant studies:Footnote 63

• — Alliance Status: Alliance codes the presence of a defence pact, neutrality pact or entente in a dyad using the COW Alliance Dataset.Footnote 64

• — Democracy: We measure democracy using Polity IV data.Footnote 65 We construct monadic values by subtracting Polity AUTOC from DEMOC, adding 10 and dividing by 2. Democracy (Low) reports the lower Polity value in the dyad in a given year, while Democracy (High) reports the higher Polity value in the dyad.

At the system level, Proportion of Democracy (Prop. Dem.) offers a ratio of the number of states surpassing a threshold democracy level (we use seven), divided by the total number of countries in the system in that year, using the Correlates of War list of system members.Footnote 66 A system with no democracies produces a Prop. Dem. score of 0, while having all states as democracies produces a score of 1. In our analyses, we also include an interaction term between Democracy (Low) and Prop. Dem. (and between Democracy (High) and the proportion of the system that is autocratic) to test the claim that the relationship between shared regime type and alliance propensity depends on the broader system environment.

• — Geographic Distance and Contiguity: It is important to include a measure of proximity and distance, given that regime type clusters geographically.Footnote 67 Contiguity is an ordinal variable for six categories of decreasing physical proximity, from shared land border to separated by more than 500 miles of water, either directly or through colonial possessions. Distance is the natural logarithm of the great circle distance between national capitals, or of the closest major cities for some large countries.Footnote 68

• — Capabilities: The realist emphasis on power as a motive for alliance mandates a measure of capabilities. We assess the balance of capabilities in the dyad using the COW Composite Indicators of National Capabilities (CINC) score. CINC is computed as the weighted average of a state's share of total system population, urban population, energy consumption, iron and steel production, military personnel and expenditures. Capability (ratio) measures the CINC owned by the least powerful state, divided by the sum of CINC's in the dyad: CINC_low/(CINC_A +CINC_B ).

• — Major Power Status: Students of international relations often view major powers as different from other states. If in addition to superior capabilities, major powers are more or less likely to be influenced by regime affinities in their alliance choices, then this could bias our results. Using the COW criteria for major power status, we code two dummy variables for whether (1) either dyad member is a major power, or (2) both dyad members are major powers. Major powers are likely to behave differently with other major powers than with minor powers.

• — Threat: Alliances are meant to affect security for participating states. There are two basic ways that threats can manifest. First, there could be a history of conflict within the dyad that interferes with co-operation. Nations that have fought should be less likely to consider each other allies. We follow Crescenzi and Enterline to produce an international interaction score (IIS) that is a summed, depreciated count of militarized dispute behaviour in the dyad.Footnote 69 Secondly, conflict outside the dyad could influence security decisions within the dyad, intensifying incentives to co-ally. We again rely on Crescenzi and Enterline for a measure of the intensity of external threat using a moving average of the count of militarized disputes in which either state in the dyad is a participant. Separate variables for high and low external conflict indicate whether one or both nations are prone to dispute with other states.Footnote 70

• — Alliance Years: We construct linear, squared and cubic count variables representing the number of years two states have been allied as discussed in the research design section.