Network Power and Militarized Conflicts

Introduction

The balance of power theory posits that under a relatively equal power balance encompassing two or more states, the states in question will be less likely to go to war (or militarized conflict) with one another. In contrast, when one state is substantially more powerful than another, it may go to war (or militarized conflict) to enhance its power position further. The power balance thesis rests on the logic that victory becomes problematic under a condition of relative power parity and that the resulting uncertainty enhances deterrence and discourages aggression. A power imbalance (i.e., power preponderance) tends to support aggression and weaken deterrence by increasing the probability of success for the stronger state’s use of force. On the other hand, the power preponderance theory posits that when power is roughly equal among the states involved, they may perceive a reasonable chance of winning (i.e., be overconfident about their ability to secure their interests through the use of force). This makes them more willing to take firm stances or escalate tensions, leading to an increased probability of disputes and actual war. In contrast, when one side enjoys a preponderance of power, the outcomes of potential conflicts are clear, and states settle disputes before they escalate to war. This power preponderance thesis holds that the probability of war (or militarized conflict) increases under a condition of relative parity. The logic is that the likelihood of war (or conflict) is greatest when both sides see a prospect for victory, and this condition is met when parity characterizes the power balance. Under power preponderance, the weaker side cannot afford to fight, and the stronger side rarely has to go to war in order to achieve its goals.

The core argument of this article is that a state’s power can be conceptualized by the social network perspective of international system structure. International communication and resource networks take into account each state’s interactions with all other states in the system across different types of network relationships salient to international relations. Accordingly, the measures of structural network power (SNP) are derived by utilizing six international interaction data sets based on two dimensions (communication flow networks and resource flow networks) of international systems (the theoretical and empirical decision rules used with these six international networks data sets are available at the Web site identified in the Notes). Hypotheses relating power balance between states and dyadic militarized interstate disputes are tested using two different estimation methods: (1) logistic regression, with errors clustered on dyads and (2) ReLogit estimation, which is the logistic regression of rare events. ¹ We demonstrate the utility of the new network power concept by applying a social network perspective to international conflicts. ² Hypotheses arising from both balance of power theory ³ and power preponderance theory ⁴ are investigated using an SNP conception derived from the social network perspective. We demonstrate how these new social networkconcepts of international system structure and national power lead to more accurate and powerful empirical models of militarized conflicts than previous ones mainly rooted in attribute logic. The performance of conflict models at the dyadic level using SNP measures is compared against those using attribution-based power measures, ⁵ through nonparametric model discrimination statistics and information criteria measures.

Theoretical Justification for the New Power Measures

This article proposes a new power concept, using the social network perspective, called “structural network power” (SNP) which is defined as the power of an individual state arising from its location within the networks of international relations. ⁶ A state does not maintain its power in isolation from other states nor from the system structure; rather, it has power as a consequence of itsinteractive relations with other states in the system and its structural positions in the networks of relations. ⁷ An SNP concept views an individual state’s power as arising from its positions in different interaction networks of internationalrelations: if it is well positioned, or occupies relatively advantageous positions in networks, it will be powerful. This way of conceptualizing a state’s power accords with social network theorists who believe that the characteristics of social units arise out of structural or relational processes played out among all the units within the network. ⁸ As most social network theorists emphasize, there is a consensus among social network analysts that a positive relationship exists between an actor’s centrality (i.e., holding an advantageous position compared with other actors) and its power within a network. Actors occupying central positions in the network are essentially seen to be powerful due to their greater access to and possible control over the relevant resources. Social network theorists argue that a state with a high centrality (i.e., with an advantageous position) holds a powerful and prominent position in the network. A state with most visibility and centrality is the point where there is most activity in the network and gains influence as a result.

This way of conceptualizing a state’s power concurs with the work of social network theorists who believe that the characteristics of social units arise out of structural or relational processes played out among all the units within the network. The measures of SNPs in this article are derived by utilizing six types of international interaction data sets relating to communication patterns and resource flows: ⁹ (1) diplomatic exchanges; (2) foreign student exchanges; (3) international telecommunications; (4) arms transfers; (5) international exports; and (6) international assistance. This article is based on the theoretical difference of the new measure of national power (structural network power index [SNPI]) to the previous measures of national power, focusing on the correlates of war (COW) material capability index (composite index of national capabilities [CINC]), which is the most widely used measure of national power in the field of international relations. ¹⁰

The social network approach to national power is responsive to the criticism that the previous concept of an individual state’s national power such as the CINC is an isolated concept, isolated both from other system members and from international system structure. First, because it conceptualizes a state’s national power on the basis of what it possesses, it is isolated from other states in the system: what state A possesses does not take account of what any other system members possess. The perspective that power is by definition a relative concept is not new. Many scholars have emphasized that the definition of power is essentially based upon relations and should be conceptualized as relational. For example, Dahl defines power as the ability to get others to do what they otherwise would not do; ¹¹ Morgenthau’s classical realism asserts that power should be understood as “control” over actors. ¹² Baldwin identifies several principles of power analysis and emphasizes that “power is a relational concept” and that “it refers to a relationship between two or more people, not to a property of any one of them.” ¹³ Lasswell and Kaplan define power relationally, not as a simple property. ¹⁴ Pruitt posits that international theorists have placed too much emphasis on the resources of nations as the basis of their power. ¹⁵ As Baldwin and others correctly point out, international theorists have long been criticized for their failure to define power in relational terms. ¹⁶ For example, during the Cold War era, some viewed power as a zero-sum concept, so that any power increase by the Soviet Union translated to a power decrease of the United States. However, this relative power concept by realists has been treated only as a measurement issue rather than a conceptual one (e.g., the power balance variable in dyadic conflict studies, measured by the difference between state A’s power and state B’s power). Even worse, this approach used in dyadic studies ignored information from outside the dyad: behavior in a dyad is studied as if it were a closed system, when of course each dyad is embedded in a network of other international relations. A social network view of power is closer to the relational power concept of “the ability to get people to do what I want” (as in Dahl, Morgenthau, or others) than to the one that is based on a state’s attributes.

Second, the conception of national power is also disconnected from international system structure. Neorealists, such as Waltz, assert that the international system is affected by the distribution of power among major powers. However, the causal arrows could go in both directions: units (or characteristics of units such as their power) affect international system structure and international structure affects units (and their characteristics). ¹⁷ Waltz himself emphasizes that causation runs not only from international structure to interacting units but from units to structure. However, the existing method of conceptualizing national power has been unable to incorporate the influence of international system structure and its characteristics over the power of its member states. A network view of power implies that an individual state’s power derives not from what it possesses but rather from how it connects or interacts with other system members in the networks of relations. Cartwright posits, “When an agent, O, performs an act resulting in some change in another agent, P, we say that O influences P. ¹⁸ If O has the capacity of influencing P, we say that O has power over P.” In other words, he argues that power is specific to each dyadic relationship. The main difference between the concepts is in how one conceptualizes this influencing capacity. In the attribute power concept, the capacity comes from the properties of a state’s own resources (i.e., its control over internal, domestic resources). In the network power concept, this capacity comes from how a state interacts with other system members (i.e., its control over external interactions). Oppenheim distinguishes power as property versus power as relation. ¹⁹ Hart and Schmidt discuss power as control over resources versus power as control over actors. ²⁰ In other words, the two approaches provide different ways of looking at the influencing capacity and, as a result, they tap two different aspects of national power. In brief, this understanding of the “relational” aspect of national power, focused on control over actors, is not new in the field of international relations.

In this article, we focus on several different dimensions of SNP that arise from a state’s position in different types of interaction networks of the international system. These dimensions are based on five different types of interaction data (i.e., degree, betweenness, flow-betweenness, coreness, and ego network brokerage). There is a consensus among social network analysts that there is a positive relationship between an actor’s centrality and its power within a network. Actors occupying central positions are viewed as powerful due to their greater access to and possible control over relevant resources. Each of the five different centralities that have been developed by social network theorists emphasizes different aspects of SNP, and we argue that we should consider all five when we conceptualize a state’s national power.

SNP Based on Degree Centrality

Viewing a state’s SNP as arising from its degree centrality treats each state’s structural power as deriving from its total number of direct connections to other states in the network. A state with a high level of degree centrality (i.e., many direct ties to other states) holds a powerful (or influential), prestigious (or prominent), or advantaged position in the network. It is the most visible actor in the network, and therefore it is “where the action is” in the network. ²¹ Compared with the state that holds a less powerful or prestigious position in the network (with its limited number of direct ties to other network members), a highly degree-central state (1) is less dependent on other states because it has many alternative ways to get what it needs; (2) has more access to the resources of the network since it has more ties to other states; and (3) usually holds and benefits from a third-party or deal-maker position in exchanges among other states in the network because it has many ties with other states. ²² For example, the Cold War matrices of arms transfers show that the Soviet Union had a higher level of degree centrality in the arms transfer network than Bulgaria. The structural position of the Soviet Union enabled it to (1) be less dependent on other states for its export and import needs; (2) have more access to the arms resources available within the network; and (3) benefit from advantageous third-party positions in the exchanges of arms.

Network power and other forms of power, such as material power, might be correlated, but, in general, one does not theoretically cause the other. However, we should be able to find such cases in which states can increase their power by exploiting and enhancing their network positions. For example, returning to our Cold War example, the network position of the Soviet Union enabled it to hold a more powerful position compared with other Eastern European countries. If one of the Eastern European countries, Bulgaria for example, decided not to import arms from the Soviet Union, the Soviet Union had many other places to export their arms. However, if the Soviet Union decided not to export arms to Bulgaria, Bulgaria (with its limited number of arms sources) might have been unable to find other alternatives from which to import their arms. Also, if the Soviet Union wanted to interact with Albania, for instance, it was simply able to do so. However, if Bulgaria wanted to interact with Albania, it could do so only by way of the Soviet Union. The Soviet Union (as a leader of the Council for Mutual Economic Assistance [COMECON]) might have discouraged such a bilateral agreement between Bulgaria and Albania with the purpose of dictating their relationship and of exploiting its network position to enhance its power (in its relationships with the two countries specifically and in the region of Eastern Europe in general). This example demonstrates the way states might increase their power by exploiting and/or enhancing their network positions.

SNP Based on Betweenness Centrality

An approach to SNP focused on betweenness centrality treats each state’s structural power as arising from its position on the geodesics (minimal length paths) that connect to other nodes in the network. A state with a high betweenness centrality (standing on many geodesics) holds a powerful or prestigious position in the network. ²³ Many other states depend on it to make connections with other states in the system. The betweenness centrality conceptualizes the degree to which a state plays the role of a “broker” or “gatekeeper,” with a potential for control over other states in the network. ²⁴ Centrality is also interpreted as the extent to which a state controls communication between other pairs of states in the system. ²⁵ For example, the Cold War matrices of arms transfers show that Italy played an important intermediary role among the three sets (or blocs) of states centered on the Soviet Union, the United States, and the United Kingdom; the same role was played by Indonesia in the international arms transfer network of 1960. During the Cold War, the data show that the Soviet Union had a higher betweenness centrality in the arms transfer network than Bulgaria. The Soviet Union was standing on the geodesic paths connecting many pairs of communist states (as a leader of the COMECON), while there were no direct connections between Bulgaria and many other communist states. Thus, if the Soviet Union wanted to interact with, say, Albania, it was able to do so. However, if Bulgaria wanted to interact with Albania, it could do so only by way of the Soviet Union.

SNP Based on Flow-Betweenness Centrality

Treating SNP as arising from flow-betweenness centrality implies measuring each state’s structural power based on its position on both the direct and the indirect paths that connect two other states in the network. As a modification of Freeman’s original conceptualization of betweenness, flow-betweenness is focused on the notion that the actors will use all paths available to them to connect to other actors (not only the shortest geodesic paths); as Stephenson and Zelen point out, there is no reason to believe that interactions between a pair of states occur only on the shortest path. ²⁶ This approach assumes that states use each pathway that connects them in proportion to the length of that pathway and that states that are “between” other states are able to translate their broker roles into power. A state with high flow-betweenness centrality (standing on many direct and indirect paths) holds a powerful or prestigious position in the network because it can affect so many interaction channels. Flow-betweenness centrality enriches the conceptualization of betweenness centrality. Suppose, for example, that two states, South Korea and North Korea, want to exchange arms transfers and that the direct geodesic path between them is blocked by China (say, to maximize its arms transfers by its separate connections to South Korea and to North Korea by blocking the direct connection between the two Koreas). If there is another pathway to connect them, such as through Russia, they will be likely to use it, in spite of the fact that it is longer and also subject to disruption. States interact with other network members using both direct and indirect pathways, and, from a social network perspective, a state that holds a more advantageous position (standing on many direct and indirect paths among network members) holds more structural social network power (in terms of flow-betweenness centrality).

SNP Based on Core Centrality

Viewing a state’s SNP as arising from its core centrality treats each state’s structural power on the basis of its degree of coreness compared with the other states in the network. Based on the concepts of a core–periphery structure (i.e., a dense and cohesive core and a sparse and unconnected periphery), and of a core–periphery relationship (the former exploiting the latter), this conceptualization of core centrality is an extension of Wallerstein’s dichotomous (core–periphery) or trichotomous (core–semiperiphery–periphery) typology that has been used in many studies of world-system theory. ²⁷ It measures “coreness” as a continuous variable (a high score on coreness represents a highly core state; a low score on coreness represents a highly peripheral state). World-system theorists claim that a state’s degree of coreness is strongly related to its power. With this approach, we can now not only partition states into different groups (core, semiperiphery, and periphery) but also differentiate the within-group members (who is the most/least powerful within each group).

SNP Based on Ego Network Brokerage Centrality

Defining SNP in terms of ego network (or egonet) brokerage centrality involves measuring each state’s structural power based on its possibilities for brokerage among the states within its own ego network. This treatment involves two separate network concepts. First, the concept of brokerage has been defined as a process “by which intermediary actors facilitate transactions between other actors lacking access to or trust in one another.” ²⁸ Burt and Galaskiewicz and Krohn define brokers as “actors who simultaneously send and receive resources from different parts of the network in which they are embedded.” ²⁹ This concept of brokerage has been studied in both theoretical and empirical social network research: ³⁰ this research emphasizes linking the ability to broker negotiation or resource flows to perceived power or influence. Gould and Fernandez identify five qualitatively different roles of the broker: ³¹ (1) as a local broker or coordinator (e.g., the Federal Reserve Bank as a clearinghouse for all the private banks in a major city); (2) as a cosmopolitan or itinerant broker (e.g., a stockbroker as a mediator among clients, buyers, and sellers); (3) as a gatekeeper or representative (e.g., the broker as a gatekeeper for his or her political party can decide whether to grant other party members access to an outsider in a rival party; (4) as a representative for other party members who can decide whether to establish contact with an outsider in a rival party; ³² and (5) as a liaison to link distinct groups (e.g., agents in the publishing or entertainment industries).

To illustrate the brokerage concept, suppose that South Korea has a tie to the United States, and that the United States has a tie to Afghanistan, but that South Korea has no direct tie to Afghanistan. In this triad of relationships (the United States, South Korea, and Afghanistan), the United States (as a local broker or coordinator, as a cosmopolitan or itinerant broker, as a gatekeeper, as a representative, or as a liaison in the relationship between South Korea and Afghanistan) can play a brokerage role in case South Korea needs to reach Afghanistan. If we extend this brokerage role of the United States to all the member states within the United States’ ego network, we can depict the United States’ SNP in terms of egonet brokerage centrality. Take, for example, the two large ego networks during the Cold War Era (one led by the United States and the other by the Soviet Union) and a small number of groups whose members did not belong to either. The ego in each of the two ego networks has played a number of roles as a broker: coordinating and mediating transactions among states within the egonet (e.g., the United States coordinating or mediating the foreign policies among the Organization of American States [OAS] members); deciding, as a gatekeeper, whom to invite as its egonet members or whether to grant its member states access to an outsider (e.g., during the Cold War era, the United States deciding who can belong to the OAS or granting its egonet member[s] permission, in an implicit or explicit way, to interact with the Soviet egonet member[s]); establishing, as a representative, a relationship with an outsider (e.g., the United States establishing, as a representative of its egonet members, a relationship with outside states such as China); working as an agent to mediate transactions between two outside groups (e.g., the United States mediating among the members of different non-aligned groups, such as the Non-Aligned Movement [NAM] groups). From the social network perspective, this egonet brokerage represents a more localized SNP compared with other types of SNP that use all the interactions within the whole network.

Dyadic Analysis of Militarized Conflicts

Data

As noted above, Oneal and Russett are leading proponents of the dyadic study of international conflicts, and we use their data set as the baseline of this study. ⁴³ This data set was merged with the data set of each state’s SNP measures that we have created using six types of interaction network data among all states in the international system. Based on aforementioned six types of interaction data, we have derived each state’s scores on different aspects of SNP.

Estimation Methods

The main estimation method in the dyadic study is the logistic regression assuming clustering on dyads. To check the robustness of results from the empirical tests, we employ an additional estimation method: ReLogit estimation. ReLogit analysis is based on the corrections of the biases in the logit estimations of rare events. ⁵¹ The findings from additional sensitivity analysis are generally with our main analysis, and therefore the discussions of the results in the following section are based on the results from our main method of analysis (logistic regression analysis clustered on dyads) in Table 1 . Table 2 displays the results from our sensitivity analysis, ReLogit estimation. To add the substantive meanings from the logistic results in Table 1, Table 3 displays the results from the marginal effects analysis for the main independent variables of the study, the balance of SNP, using the Clarify program. ⁵² This program, using stochastic simulation techniques, first “draws simulations of the main and ancillary parameters from their asymptotic sampling distribution,” and second “converts the simulated parameters into substantively interesting quantities, such as predicted values, expected values, or first differences.” We choose to simulate 1000 sets of parameters from our models to produce the predicted values for each of our models to be tested.

OPEN IN VIEWER

Table 1.

Logit Analysis of the Onset of Militarized Interstate Disputes, 1950–1992

	Model 1 (CINC)	Model 2 (GNP)	Model 3 (Degree)	Model 4 (Betweenness)	Model 5 (Flow-Betweenness)	Model 6 (Coreness)	Model 7 (Egonet Brokerage)	Model 8 (SNPI)
Balance of Structural Network Power	−0.121 ** (0.047)	−0.116 ** (0.045)	−0.097 *** (0.026)	−0.118 *** (0.025)	−0.153 *** (0.028)	0.010 (0.063)	−0.058 *** (0.017)	−0.161 *** (0.047)
Level of Economic Interdependence	−19.982 * (11.927)	−24.769 * (13.248)	−21.296 * (12.382)	−26.922 * (13.854)	−27.241 * (13.686)	−16.988 (11.605)	−20.349 * (12.105)	−21.712 * (12.380)
Level of Democracy	−0.049 *** (0.013)	−0.047 *** (0.013)	−0.047 *** (0.013)	−0.045 *** (0.013)	−0.045 *** (0.013)	−0.050 *** (0.013)	−0.048 *** (0.013)	−0.047 *** (0.013)
Shared Alliance Ties	−0.431 ** (0.155)	−0.421 ** (0.156)	−0.400 ** (0.155)	−0.435 ** (0.156)	−0.387 * (0.154)	−0.413 ** (0.156)	−0.397 ** (0.155)	−0.414 ** (0.156)
Contiguity	2.550 *** (0.207)	2.530 *** (0.203)	2.514 *** (0.208)	2.503 *** (0.199)	2.490 *** (0.201)	2.564 *** (0.208)	2.522 *** (0.207)	2.525 *** (0.208)
Distance	−0.581 *** (0.078)	−0.569 *** (0.075)	−0.555 *** (0.079)	−0.554 *** (0.071)	−0.559 *** (0.073)	−0.575 *** (0.079)	−0.559 *** (0.079)	−0.554 *** (0.079)
Major Power	1.735 *** (0.166)	1.867 *** (0.186)	1.693 *** (0.165)	1.845 *** (0.169)	1.831 *** (0.168)	1.683 *** (0.158)	1.669 *** (0.164)	1.748 *** (0.167)
Peace Year	−0.341 *** (0.034)	−0.341 *** (0.034)	−0.342 *** (0.034)	−0.338 *** (0.033)	−0.340 *** (0.034)	−0.345 *** (0.034)	−0.343 *** (0.034)	−0.344 *** (0.034)
Spline1	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)
Spline2	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)
Spline3	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)
Constant	−0.410 (0.655)	−0.433 (0.627)	−0.549 (0.670)	−0.363 (0.606)	−0.357 (0.618)	−0.636 (0.670)	−0.559 (0.666)	−0.531 (0.667)
N	282,230	254,579	282,505	282,505	282,505	282,505	282,505	282,505

Note: Robust standard errors appear in parentheses below the coefficient estimates. All significant tests are two-tailed. Model 1 (CINC), dispute model using the Correlates of War CINC power index; Model 2 (GNP), dispute model using Gross Domestic Product; Model 3 (Degree), dispute model using the degree aspect of Structural Network Power Index; Model 4 (Betweenness), dispute model using the betweenness aspect of Structural Network Power Index; Model 5 (Flow-Betweenness), dispute model using the flow-betweenness aspect of Structural Network Power Index; Model 6 (Coreness), dispute model using the coreness aspect of Structural Network Power Index; Model 7 (Egonet Brokerage), dispute model using the egonet brokerage aspect of Structural Network Power Index; Model 8 (SNPI), dispute model using the composite Structural Network Power Index.

^* p < .10.

^** p < .01.

^*** p < .001.

OPEN IN VIEWER

Table 2.

Sensitivity (ReLogit) Analysis of the Onset of Militarized Interstate Disputes, 1950–1992

	Model 1 (CINC)	Model 2 (GNP)	Model 3 (Degree)	Model 4 (Betweenness)	Model 5 (Flow-Betweenness)	Model 6 (Coreness)	Model 7 (Egonet Brokerage)	Model 8 (SNPI)
Balance of Structural Network Power	−0.121 ** (0.047)	−0.116 ** (0.045)	−0.095 *** (0.026)	−0.118 *** (0.025)	−0.153 *** (0.028)	0.011 (0.063)	−0.058 *** (0.017)	−0.160 *** (0.047)
Level of Economic Interdependence	−19.025 (11.926)	−23.810 * (13.247)	−20.333 (12.381)	−26.017 * (13.853)	−26.331 * (13.686)	−15.979 (11.604)	−19.376 (12.104)	−20.754 * (12.380)
Level of Democracy	−0.049 *** (0.013)	−0.047 *** (0.013)	−0.047 *** (0.013)	−0.044 *** (0.013)	−0.044 *** (0.013)	−0.050 *** (0.013)	−0.048 *** (0.013)	−0.047 *** (0.013)
Shared Alliance Ties	−0.431 ** (0.155)	−0.421 ** (0.156)	−0.400 ** (0.155)	−0.435 ** (0.156)	−0.387 * (0.154)	−0.413 ** (0.156)	−0.396 ** (0.155)	−0.414 ** (0.156)
Contiguity	2.548 *** (0.207)	2.529 *** (0.203)	2.513 *** (0.208)	2.502 *** (0.199)	2.489 *** (0.201)	2.562 *** (0.208)	2.521 *** (0.207)	2.524 *** (0.208)
Distance	−0.580 *** (0.078)	−0.568 *** (0.075)	−0.554 *** (0.079)	−0.553 *** (0.071)	−0.558 *** (0.073)	−0.574 *** (0.079)	−0.558 *** (0.079)	−0.553 *** (0.079)
Major Power	1.734 *** (0.166)	1.865 *** (0.186)	1.692 *** (0.165)	1.844 *** (0.169)	1.829 *** (0.168)	1.682 *** (0.158)	1.668 *** (0.164)	1.747 *** (0.167)
Peace Year	−0.341 *** (0.034)	−0.341 *** (0.034)	−0.342 *** (0.034)	−0.337 *** (0.033)	−0.340 *** (0.034)	−0.345 *** (0.034)	−0.343 *** (0.034)	−0.343 *** (0.034)
Spline1	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)	−0.002 *** (0.000)
Spline2	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)	0.001 *** (0.000)
Spline3	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)
Constant	−0.412 (0.654)	−0.435 (0.627)	−0.551 (0.670)	−0.364 (0.606)	−0.358 (0.618)	−0.638 (0.670)	−0.560 (0.666)	−0.533 (0.667)
N	282,230	254,579	282,505	282,505	282,505	282,505	282,505	282,505

Note: Robust standard errors appear in parentheses below the coefficient estimates. All significant tests are two-tailed.

^* p < .10.

^** p < .01.

^*** p < .001.

OPEN IN VIEWER

Table 3.

Marginal Effects Analysis of the Onset of Militarized Interstate Disputes: Changes in Predicted Probabilities of Dispute Onset from Fitted Logit Models

	Minimum	Mean (Baseline)	Maximum	Min ⇒ Mean	Mean ⇒ Max	Min ⇒ Mean (Reduction, %)	Mean ⇒ Max (Reduction, %)	Min ⇒ Max (Reduction, %)
Model 1	0.00092	0.00077	0.00031	−0.00015	−0.00045	−20.0	−59.2	−66.0
Model 2	0.00098	0.00077	0.00034	−0.00021	−0.00043	−27.6	−56.1	−65.6
Model 3	0.00094	0.00077	0.00039	−0.00018	−0.00038	−22.9	−49.2	−58.7
Model 4	0.00116	0.00077	0.00028	−0.00039	−0.00048	−50.9	−62.8	−75.3
Model 5	0.00111	0.00076	0.00019	−0.00035	−0.00057	−46.4	−75.3	−83.2
Model 7	0.00091	0.00077	0.00042	−0.00015	−0.00035	−19.4	−45.2	−54.1
Model 8	0.00098	0.00077	0.00019	−0.00020	−0.00058	−26.1	−74.9	−80.1

Note: The marginal analyses were calculated using Clarify software.

Results and Discussion

Hypotheses 1 and 1’ test the arguments advanced by “balance of power” theorists that the power parity of states in a dyad leads to a decreased probability of dispute onset and by “power preponderance” theorists that power preponderance leads to a decreased probability of dispute onset, using the new concept of SNP. Table 1 displays the results from the logistic regression of the onset of dyadic dispute: the logistic results for six models that consist of five different aspects and their composite index. The results from the logistic regressions clustered by dyads (Table 1) strongly support the arguments of power preponderance theorists. In all but one model specification, the coefficient on the Balance of SNP is negative and statistically significant, suggesting that the more skewed the distribution of SNP, the less likely a militarized interstate dispute. In the other model specification, the coefficient is statistically indistinguishable from zero. The results from sensitivity analyses all support the power preponderance theory.

It is clear from all the analyses that when one state in a dyad has more power than the other, they are less likely to be involved in conflicts with each other, supporting the power preponderance-leads-to-peace argument. As power preponderance theorists argue, the weaker state will not try to fight with the stronger state because it will certainly lose, and therefore the stronger state does not have to fight for what it wants. In other words, the power preponderance between states preserves peace since it reduces the uncertainties of winners and losers in a dispute. As there is a highly asymmetric power distribution, a weaker state more likely concedes to a stronger state’s demand since it knows it is less likely to prevail in a crisis bargaining situation. All in all, the results from all sets of dyadic analyses in this study provide the empirical evidence toward supporting the “power preponderance leads to peace” or “power balance leads to conflict” theoretical argument.

To add the substantive meanings from the logistic results in Table 1, Table 3 displays the marginal effects analysis for the main independent variables of the study—the balance of SNP—using the Clarify program. The results show that, for example, using the degree aspect of balance of SNP, the baseline probability of the onset of a dispute, ⁵³ holding all other included variables at their means, is 0.00077. Raising the level of SNP balance from its minimum to the mean, while holding all other variables at their means, reduces the probability of onset by 22.9 percent. Increasing the change to its maximum triggers an additional decrease of 49.2 percent. Increasing the level of the variable from its minimum to its maximum reduces the probability of onset by 58.7 percent, a drop of more than half from the baseline rate in the likelihood that the two states in a dyad will have a dispute. As seen in Table 3, the reduction of predicted probabilities, changing the values for the balance of SNP variable from the minimum to the maximum, while holding all other independent variables at their means, ranges from 54.1 percent to 83.2 percent for the models. The results of the marginal impact analyses show that, in addition to the statistical results from Table 1, there are quite sizable effects of the balance of SNP on the onset of militarized disputes. This finding supports power preponderance theory over balance of power theory.

The results from Table 1 generally support all the other hypotheses on the dyadic dispute onset. Hypothesis 2 predicts that the coefficient on the Level of Interdependence variable will be negative. This hypothesis implies that if a state is economically interdependent with its opponent in a dispute, it is less likely to be involved in a dispute because it fears losing the benefits of trade. The results support this assertion; the estimated coefficient for Level of Interdependence is negative and statistically significant in most of the models. The “democratic peace” pillar of the liberal peace is strongly supported by the data. The Level of Democracy variable is negative as expected and statistically significant at better than the 0.001 level in all models, meaning that democratic states are less likely to be involved in a dispute.

Hypothesis 4 predicts that the coefficient on the Shared Alliance Ties would be negative. The results support this hypothesis; the coefficient on the variable is negative as expected and statistically significant in all models, usually at better than 0.01 level, supporting the realist argument that allies are less likely to be involved in a dispute when they fear losing the security benefit of the alliance. Hypotheses 5–7 are also supported by the regression analysis, supporting the realists’ arguments that (1) the potential for international violence exists when the actor can reach its adversary with military force and (2) major powers have been engaged in more international disputes compared to other states, based on their wide-ranging interests. Hypothesis 5 predicts that a shared common border increases the probability of the onset of dispute. As expected, the coefficient on Contiguity is positive and statistically significant at better than 0.001 level in all models. Hypothesis 6 argues that the geographic distance decreases the probability of the onset of dispute. Again, as expected, the coefficient on Distance is negative and statistically significant. Finally, Hypothesis 7 predicts that when one of the two states in a dyad is a major power, the probability of the onset of dispute is increased. As expected, the Major Power estimated coefficient is positive and statistically significant at better than .001 level in all models.

All in all, three sets of findings stand out. First, regarding the debate on balance of power and power preponderance theory, our empirical findings strongly support the argument by power preponderance theorists. Using different aspects of SNP (e.g., degree, betweenness, flow-betweenness, coreness, ego-net brokerage, and their composite index) and different estimation methods (logistic regression analysis clustered on dyads and ReLogit analysis), the results consistently support the power preponderance theory—that preponderance of SNP decreases the probability of dispute onset or the more skewed the distribution of SNP, the less likely a militarized interstate dispute. Second, regarding the liberal peace theory, the economic interdependence pillar is weakly supported at the 0.10 significance level compared to that of the democratic pillar at better than the 0.001 level. However, the weak support of the “economic interdependence leads to peace” argument is not surprising on both theoretical and empirical grounds. Many realists theorize that economic interdependence rather increases conflict among states, in which interdependence is a source of friction that can lead to military conflict. However, others argue that economic interdependence and conflict among states are irrelevant, in which interdependence is subordinate to other considerations in determining the incidence of international conflict. Recent empirical work using more advanced econometric methods have found no relationship between economic interdependence and international conflict. ⁵⁴ Finally, the marginal effects analysis using Clarify shows that all the variables of the three main hypotheses in this study have quite sizable and compatible effects; the reductions of the probability of dyadic onset, which is against the baseline rate in the likelihood that the two states will have a dispute, for the Balance of SNP variable range from 54.1 percent to 83.2 percent for six of our models; those for the Level of Interdependence from 74.4 percent to 77.8 percent; and those for the Level of Democracy from 76.0 percent to 85.8 percent. Not only are democratic states or economically interdependent states less likely to be involved in a dispute but also states with sizable power differences. This set of results is consistently supported in all different estimation methods used and with different measures of SNP utilized.

Conclusions

This article examined the determinants of the onset of international conflicts from the social network perspective. In this work, the new social network power concept called SNPI was applied to the previous empirical attempts of balance of power and power preponderance theories at the dyadic level. The focus of hypothesis testing in the empirical section of this article was on how our new social network power conception leads to more accurate and powerful empirical models than previous ones mainly rooted in attribute logic. In other words, this article took existing theories and tested them using our newer network conception of national power, focused on the conceptions of the international system as networks and of a state’s power based on its relational SNP, and was primarily interested in how these newer conceptions lead us to recast a great deal of existing empirical work on the subject. How is the SNP balance between states related to their conflict behaviors for the dyadic conflict analysis? This is the main research question that was examined in this article of social network application to militarized conflicts.

In examining the hypotheses on balance of power theory and power preponderance theory, this study found that, at the dyadic level, the overall results do strongly support power preponderance theory over balance of power theory. The results are clear from all the analyses using two different estimation methods, logistic estimation clustered on dyads and ReLogit estimation, that when one state in a dyad has more power than the other they will be far less likely to be involved in conflicts with each other, supporting the power preponderance-leads-to-peace argument. The marginal impact analysis also shows that the preponderance of SNP has a strong effect on interstate dispute, cutting the probability of a dispute by 54.1–83.2 percent, depending on the model specifications, from the baseline rate. The evidence from nonparametric model discrimination statistics and information criteria measures also shows that those models with network power measures statistically outperform and are preferred over those with attributional power measures such as the COW index and GNP in dyadic conflict onsets studies (more detailed discussion on the performance of conflict models available at the Web site identified in the Notes).

The set of empirical findings from this article provides us some policy implications in the Asia-Pacific region. As many scholars and policy makers argue, one of the most troubled bilateral relationships in the region is that between South and North Korea. We could apply this article’s most important finding, which is supporting power preponderance theory, to the South–North Korean relationship. If South Korea holds a power advantage over North Korea, as seems to be a current situation, military conflict between the two will be less likely to occur. However, if North Korea accumulates more power and, as a result, if the two Koreas share approximately equal power, military conflict between the two will be far more likely to happen, ultimately threatening the regional peace. ⁵⁵ Another important finding from this article, which is supporting liberal peace theory, also has some policy implications. This finding provides us the rational for policies encouraging the spread of democracy and the promotion of economic interdependence among countries in the region. When the majority of countries in the region become democratic and when the majority of countries freely trade with each, there will be less conflicts and peace in the region will be secured.

The policy contribution of this article goes beyond the Korean peninsular and the Asia-Pacific region. Power preponderance theory (or related power transition theory) in international relations argues that when power of one of the dissatisfied major states is matched with that of the most powerful state in the international system, there is a high probability of military conflicts between the two states. The possible military conflicts between the two most powerful states in the system has been continuously discussed, in both academic and political arenas, with regard to the relationship between the United States (hegemonic power) and China (dissatisfied major power). In other words, international relations scholars and policy makers have been worried about the narrowing power gap between the two states and, as a result, the possibility of military conflicts between the two. We think that our findings provide us some policy implications for this relationship in terms of the peace and stability in the world and in Asia-Pacific region in specific. To reduce the possibility of the conflicts between the two most powerful states in the region and in the world (the United States and China) and, as a result, for the peace in Asia and in the World, the policies toward the widening power gap between the two countries should be prescribed and pursued. Also, as liberal peace theory in international relations argue, when the majority of countries in the world becomes democratic and when the majority of countries freely trade with each other, which have been the policies of major powers especially of the United States, there will be less conflicts in the region and in the world, our article prescribes.