Scholarly article on topic 'Wars of the World: Evaluating the Global Conflict Structure During the Years 1816-2001 Using Social Network Analysis'

Wars of the World: Evaluating the Global Conflict Structure During the Years 1816-2001 Using Social Network Analysis Academic research paper on "Computer and information sciences"

CC BY-NC-ND
0
0
Share paper
Keywords
{"social network analysis" / "historic conflicts ;dynamic analysis" / "network visualization"}

Abstract of research paper on Computer and information sciences, author of scientific article — Olga Levina, Robert Hillmann

Abstract This paper explores the application of dynamic network analysis techniques to the area of historical martial conflict analysis. The goal is to provide input for the area of quantitative historic analysis, specifically the study of armed conflicts. Concepts from (dynamic) network analysis literature are applied to derive information on historical conflict development as well as on the definition of actor roles involved in the conflicts. Thus, network centrality measures as well as dynamic properties related metrics are examined towards their relevance for the identification of military conflict mechanics.

Academic research paper on topic "Wars of the World: Evaluating the Global Conflict Structure During the Years 1816-2001 Using Social Network Analysis"

Available online at www.sciencedirect.com

ScienceDirect

Procedia - Social and Behavioral Sciences 100 (2013) 68 - 79

8th Conference on Applications of Social Network Analysis - ASNA 2011

Wars of the world: Evaluating the global conflict structure during the years 1816-2001 using Social Network Analysis

Olga Levinaa, Robert Hillmanna

a Berlin Institute ofTechnology, Department ofSystemsAnalysis and IT, Franklinstr. 28/29,10587Berlin, Germany

Abstract

This paper explores the application of dynamic network analysis techniques to the area of historical martial conflict analysis. The goal is to provide input for the area of quantitative historic analysis, specifically the study of armed conflicts. Concepts from (dynamic) network analysis literature are applied to derive information on historical conflict development as well as on the definition of actor roles involved in the conflicts. Thus, network centrality measures as well as dynamic properties related metrics are examined towards their relevance for the identification of military conflict mechanics.

© 2013 The Authors. Published by Elsevier Ltd.

Selectionand/orpeer-reviewunderresponsibilityofDr.ManuelFischer

Keywords: social network analysis; historic conflicts;dynamic analysis; network visualization.

1877-0428 © 2013 The Authors. Published by Elsevier Ltd. Selection and/or peer-review under responsibility of Dr. Manuel Fischer doi: 10.1016/j.sbspro.2013.10.700

1. Introduction

Historical analysis is a method that often uses qualitative analysis of written documents, personal statements or artifacts (Gradner, 2006). On the other hand, techniques relying on historical data can also be applied for the review ofhistorical events. Quantitative historical analysis uses databases and statistical data analysis to derive insights on the past events or structures. Statistical analysis has already been used to describe and evaluate historical events (see e.g., (Wilkinson, 1980)). Also mathematical approaches has been applied to derive dynamics and structure of historical development on a global scale (Grinin, 2006). (Flint, 2009) use social network analysis in the context of political geography and conflicts, while (Maoz, 2010) introduces social network analysis in the context of international collaboration networks.

Here we propose the use of social network analysis toolset for the analysis of relations and roles between countries involved in martial conflicts in a specified time interval and on global scale. Social network analysis focuses on the relations between network elements using network theory and thus providing quantitative results for further interpretation and comparison. Application of this research approach from the social sciences is likely to extend the method set used for analyzing historical events by delivering insights on roles, event structure as well as their structural development over time.

Results of this paper include a visualization approach for the interrelation structures between global conflict participants, a qualitative approach to martial conflict analysis based on the social network analysis techniques, as well as the set of metrics that can be applied for historical analysis of martial conflicts.

The paper is therefore structured as follows. First, the original dataset used for the analysis as well as its interpretation as a network is described. Then, assumptions as well as the goal of analysis are presented and are followed by the description of applied method and metrics used to derive results. These are presented and discussed in the final sections of the paper.

2. Research Design

Before the historical conflicts were analyzed, existent data set was interpreted as a network and converted into a network processing format. Original dataset was constructed by Wimmer and Min (2009) as a contribution to quantitative study of wars concerning the location and purpose of conflicts around the world. The authors developed a dataset including wars fought during the years 1816 until 2001 based on the canonical Correlates of War (COW) dataset (Ghosn, 2004). They enlarged the dataset by adding armed conflicts as a result of using the established COW standard definition in any area governed by states3. Instead of analyzing state actors as units, the authors defined territories as their units of analysis, i.e. fixed geographic units of observation that conform to the grid of states in 2001. Additionally, this dataset distinguishes between intra-polity wars, i.e. fights for domestic powers, and inter-polity wars, i.e. enlargement of power relative to others Wimmer and Min (2009).

Furthermore, Wimmer and Min (2009) tested an institutionalist theory (see e.g., Peters 2005) of war using this dataset and showed empirically that the type of wars fought in a territory depends on whether it is governed as a modern nation-state, an imperial dependency, or the territory is the center of an empire.

" Wimmer and Min (2009) expand the standard state definition by defining it as a "centralized political organization that includes at a three-level administrative-political hierarchy" according to Murdock and White (1969); Mueller (1999).

To enable dynamic network analysis of this dataset, it was transformed into an event-based network data model. As a differentiation against other structural data models, this approach has a strong focus on network dynamics, thus highlighting temporal characteristics of conflicts. The generic structure of the data model is centered around the data type network based on Trier (2008).

1.1. Towards a Network

In general, the term network is defined as a set of nodes and a set of communication events, so called linkevents, among them. The main difference between the classic link-based and the linkevent-based approach is the fact that the data model does not store network links but instead, the collection of linkevents is aggregated to links among nodes. Furthermore, the data model includes various attributes or properties for nodes and events.

In the given context each region or state from the original dataset is interpreted as a node in the network. The conflicts or wars are interpreted as linkevents among the network nodes. A variety of data fields from the original data is stored as attributes in the present data set. The "Location" field of the source dataset is stored as a node attribute, whereas "War Name", "Main and Sub-Type" as well as "War Number" are stored as linkevent attributes. In addition, each linkevent includes a timestamp, in this specific case the starting year of the war. The linkevents in the data model are directed by including information of senders and recipients.

Due to different historical and political interpretations of the starting or triggering actors of the conflicts, each participating state or region involved in a conflict is interpreted here as sender and recipient of the corresponding linkevent at the same time. As a consequence, the involved nodes form a total connected component and the network links are undirected. Furthermore, the network graph is weighted due to the fact that nodes can be involved in more than one war.

Fig. 1. Conflicts around the World (node size represents number of direct contacts)13 bThe network in figures 1, 2 and 4 was mapped on the world atlas from: (World, 2007).

The temporal resolution of the source data is based on years. To consider the duration of wars, for each year of a specific conflict, a separate linkevent is created. The intra-polity conflicts or civil wars are characterized as self-links with sender and recipient being identical. As a direct consequence, although the data source only contains active nodes, the exclusively involvement in civil wars leads to the existence of isolated nodes in the resulting network.

The annotation of time and duration of wars and the aggregation of linkevents to links among nodes allows for an explicit time-frame based analysis. In this case, the analysis window is defined with start and end point and the evaluation only considers linkevents or wars that happen in between (see figure 2).

Fig. 2. Time-frame view on the conflict network (here: Vietnam War 1960-1965)

The resulting network is captured in figure 1 and consists of 200 nodes (countries or regions) connected with 1666 linkevents aggregated to 1748 links. It has a reach of 73.07%, a diameter of 6 and an average path length of 2.39. Furthermore, it contains 32 isolated nodes and covers a time span of 185 years starting in the year 1816 and ending in2001.

1.2. Assumptions and Goals ofAnafysis

The goal of analysis here is the application of dynamic social network analysis techniques to historical events in order to derive a new perspective on the described structures. Network analysis concentrates on relations between elements and does not focus on the elements themselves. Thus, applying social network analysis in this historical context can provide additional insights into conflict structures as well as on the dynamic changes in the behavior of its elements over time.

The grounding assumption here is that the specific roles and behavioral patterns in the analyzed conflict structure can be assigned to participating countries using network analysis. The following possible actor roles were suggested for analysis and assigned to accordant network metrics:

• Conflict-prone countries —»possible identification using the degree centrality metric;

• Countries that are autonomous in waging of conflicts —possible identification using the closeness centrality metric;

• Countries that are prone to conflict initiation —possible identification using the local clustering coefficient;

Other metrics that are often used in the context of social network analysis were considered during the analysis. The metric of a node's brokering activity (Trier, 2007) can be used to identify countries that actively influence relations within the conflict development; betweenness centrality is often used for analysis of e.g. criminal interrelations (see e.g. (Sparrow, 1991; Xu, 2005)). In this context, this metric will be considered for identifying countries that might be relevant for the distribution of conflicts among active nodes as the linking relations in this network represents involvement in military conflicts.

An additional assumption was made considering the conflict structure development. The effect of globalization can be seen from the evolution of certain network metrics as globalization is assumed to derive from augmented information, i.e., here conflict relation, exchange between countries. Thus, it affects conflict design and composition. To analyze this assumption network density, diameter and link strength metrics were visualized according to their temporal development.

1.3. Metrics Used

To identify distinctive network features and roles of network actors, different metrics from the social network analysis toolset on network as well as on actor level were applied. Known from the analysis and visualization of criminal networks first of all the centrality measures (Freeman, 1978) were calculated. These metrics were developed and are used to define roles of individuals in the network concerning their importance. Hereby, high centrality values indicate the level of importance of a node in a network (Wasserman, 1994).

As there is no single feature of an actor that implies its absolute importance within the network, several criteria for measuring the importance of a node exist and thus can be measured with respect to different network or nodal properties. (Freeman, 1978) suggested the degree, betweenness and closeness centrality measures for describing the importance or influence of nodes.

Degree centrality C'd describes the number of links of a particular node, indicating that the importance is assigned here to the node with the highest number of contacts. In the context of conflict analysis, contacts indicate the involvement in conflicts with numerous partners or opponents.

The variable n represents the number of nodes in the network and d(ni; nj) the geodesic path between the two nodes and di the degree of node i. N is the number of nodes in the network.

Betweenness centrality C'B indicates the number of shortest paths (geodesies) passing through a specific node (Freeman, 1978; Wasserman, 1994; Xu, 2005). Here the importance of the node is based on its potential to control information flow and its distribution within the network. High betweenness centrality values thus indicate a node with controlling power on the information flow. This metric can be used as an indicator of the importance of a node for effective communication within or operation of a network (Sparrow, 1991). Put in the conflict analysis context, actors with high betweenness centrality values can be considered to have a significant influence on the informational and thus conflict alliance structures in the network. Removing a node with high betweenness centrality value will result in less efficient communication between several nodes.

C'D(ni))= dt/(N-l)

C'B (n_i)= (2^k"b]«(nl))/(n2-3n+2), with bf(nd= (gj^(ni))/gjk

gjk(ni) in equation 2 is the set of paths that go through ni; gk is the number of shortest paths between nodes nj and nk.

Closeness centrality C'c metric calculates the sum of the distance between a specific node and every other node in the network (Xu, 2005). Thus, the value is inverse proportional to the distance between related nodes. Nodes with small closeness centrality may be considered as autonomous, as they are only marginally involved into network interactions. In the context of conflict analysis, countries with small values of the closeness centrality metric can be regarded as being rarely involved in conflicts with multiple participants.

C'c(nO = (n-iyG"Md(ni,nj)) (3)

As an additional metric, local clustering coefficient measures the interaction of nodes within an egonetwork including transitive connections, and can therefore be useful for the identification of sub-grouping in the network. The degree of connectivity of the group of nodes around a specific node (ego) is measured by the clustering coefficient. A node with a small clustering coefficient value can be thus interpreted as the key element (here: conflict initiator) in its network.

Connectivity is a measure for the network robustness, i.e. reaction to structural changes within a graph. It is often used to define the number of nodes that are crucial for network cohesiveness, so that isolation of these nodes will disrupt the network. The measure can also be used to quantify the strengths of interactions of an actor. Density is defined as the ratio of links present in the network and the maximum number of possible links (Wasserman, 1994). Thus, it can also be used to refer to the stability of the network with respect to structural changes. A small density indicates that elimination of only few nodes can lead to network dispersion.

Metrics from the dynamic network analysis such as brokering activity (Trier, 2007), number of linkevents sent and received as well as metric trends were also considered. Brokering activity of a node indicates its impact on the network structure by counting the number of its activities in a considered time slot, i.e. the metric examines how many nodes would be additionally connected or linked with a shorter geodesic if the observed node would not have existed. As a consequence, this metric indicates the level of responsibility of nodes regarding the current network structure.

To interpret the outcomes of the network analysis a context- specific ontology needs to be created. Countries are interpreted as nodes, whereas links between the nodes, i.e. the relations, represent the involvement in an armed conflict; thus, the flow of information is interpreted here as the spreading of the conflict potential among countries or states.

3. Results

(Wimmer, 2009) derive several conclusions from the dataset of armed conflicts. Based on the COW interpretation of the units of analysis and the number of wars, United Kingdom, France and Russia are identified as the most war-prone states, but they are not the most war-prone territories in the context of inter-polity conflicts Wimmer and Min (2009). This observation is due to the fact that most of the armed conflicts led by these countries were directed against colonized peoples. In contrary (Wimmer, 2009) identified China, India and Ethiopia as territories that are most war-prone for inter-polity conflicts.

In the context of intra-polity wars such as conflicts that were lead for the purpose of gaining domestic power, the COW approach identifies Turkey, Russia, France, the United States of America as well as the

United Kingdom as the most war-prone states. The secessionist and non-secessionist wars were not

differentiated. In comparison, (Wimmer, 2009) identified China, Argentina and Mexico as territories that are most prone to intra-polity conflicts (see also figure 3).

Fig. 3. Geographic distribution of military conflicts according to Wimmer and Min (2009) 1.4. Results of static network analysis

The network analysis approach applied in the present paper considers and analyses relations between states or regions in the context of armed conflicts. As a first result, following observations were made. France and United Kingdom are identified as the countries with the highest values of the three centralities (see table 1 for results overview and metric values) followed by Italy, United States of America, Belgium and Greece with high values in two of the three centrality types. Thus, these countries can be interpreted as war-prone (using the degree centrality measure), significantly affecting the conflict alliances structure (betweenness centrality measure) and having a rather small autonomy in conflict leading (closeness centrality measure).

WorldWarl (1914-1918) World War II (1939-1945)

Korean War (1950-1953) Yugoslav Wars (1998-1999)

Fig. 4. Visualization of the conflict networks on a world mapc

c Historical data are taken according to Wimmer and Min (2009).

Nodes with the highest impact on the network, i.e. with the highest value of the brokering activity indicating the involvement in numerous conflicts, and therefore indicating a significant impact on the network structure, are: United Kingdom, France and Russia. The implication is, that these countries significantly shaped alliances within the network and enabled conflict propagation and are therefore responsible for forming the overall network topology.

These findings are also conforming to the results provided by the COW-approach. Due to the fact that these results were obtained using a simple war count, they do not define the role of these countries relevant to the conflict structure but only identify countries that are most prone to be involved in an armed conflict. Another interesting aspect is that the same countries have the highest value of the connectivity metric in the network. This fact indicates that they obtain the major interaction strength within the network, leading to the assumption of these countries being the most probable combination of participants in an armed conflict. Additionally, the removal of one of those, e.g., United Kingdom, from the network, would lead to the separation of the network and thus to a less effective interconnections of the actors.

Nodes with high value of the number of Unkevents metric (here: France, United Kingdom and China) indicate countries that are involved either in numerous or long wars, whereas the degree centrality metric indicates nodes that were involved in numerous conflicts also considering intra-polity conflicts, i.e., civil

The analysis of the local clustering coefficient shows that Chile, France and United Kingdom can be regarded as possible initiators of martial conflicts in their ego-networks as they show the lowest value of a local clustering coefficient"1.

Centrality metrics in social network analysis are used as a rough indicator of social power of a node based on the impact that a node provides to cohesion of the network. Thus, based on the static network analysis results and the analogy of social network terminology, centralities are used here to index the martial power or capital of a node. The martial capital is a term that is defined here in analogy to the social capital definition in social networks. Adopting the social capital definition by (Nahapiet, 1998), the martial capital can be defined as "the sum of the actual and potential resources embedded within, available through and derived from the network of relationships, i.e. conflict involvement, comprising both the network and the assents that may be mobilized through that network". Obtaining the martial capital would facilitate coordination and cooperation within the network in terms of conflict participation, securing the benefits by the membership in the conflict structure5. Introduction of this term is meant to facilitate the discussion of the findings resulting from network analysis and their interpretation.

d Isolated nodes were not considered here.

1 See the social capital definition by Putnam (1995) and Portes (1998) respectively.

Table 1: Results of the network analysis'

Metric

Country 'with the highest metric value (decending)

Semantic interpretation

Betweenness Centrality

Degree Centrality

Closeness Centrality

Number of linkevents per direct contact

Brokering Activity

Connectivity

Local Clustering Coefficient

France (27.76)

United Kingdom (18.87)

Belgium (10.48)

France (48.65)

United Kingdom (47.03)

Italy (39.46)

France (56.48) United Kingdom (55.28) Italy (51.9)

France (2.25)

United Kingdom (2.41)

China(2.96)

United Kingdom (2361) France (1744) Russia (507)

United Kingdom (18.1) France (9.88) Russia (3.77)

Chile (30)

France (32.99)

United Kingdom (33.11)

Potential impact on the information (alliances) structure

Direct involvement in wars

Low autonomy in war waging

High average conflict duration or high number of conflicts with the same actors

Impact on the network structure

Impact on network disruption; measures strength of interaction within the network

Possible initiator of the martial conflict

f Table 1 shows only the first three locations with the highest values of the correspondent metric.

1.5. Results of dynamic network analysis

Thus, the analysis shows that United Kingdom and France are the countries with the highest martial capital and therefore the main actors in the analyzed network. Including dynamic aspects into the analysis provide some further insights on the network structure. Although the average duration of the conflicts led by United Kingdom and France is relatively small, being ca. 2.5 linkevents per direct contact. Further use of the dynamic analysis of the network, i.e. time-frame analysis, allowed visualizing and analyzing conflicts in specified time interval (see figure 4).

Furthermore, selected network characteristics were analyzed in their behavior over time, i.e., in their dynamic development. The diameter, density and link strength trends are visualized in figures 5 and 6. The temporal analysis was conducted here per year as well as using the time-frame functionality to visualize cumulated development of the metrics and therefore the network.

n A s ; / « • • * - ' m i » . 4 : :;

1 \ . .' i »"i * * n-

-11111111111111 ri 111111 n 1111111 m i n 11 n 1111111i iti^i 11 n i n i~m n i 111111111 n 11111i ■ 11 Tliiiiiiiiiiiiiilii U.........

oocococoooixJMooMooMcoMWmmcjimaiffiQmffifficnc^mtriijio^cTi

Fig. 5 Link strength (dashed line), diameter (dotted line) and density distribution per year

Significant increase of the density metric value is observed in the years 1917-18, 1940-1944 as well as during the Korean War (1950-1953). Similar results can be observed for the diameter analysis. Global conflicts are indicated in figure 5 and 6 by the rise of the density metric, whereas density values of 0 indicate rather conflict free years (here the years 1872 and 1888) with no or solely intra-polity wars.

Fig. 6. Cumulated density (dashed line) and link strength- including the trend lines

The cumulative view shows that a decrease of the link strength indicates an increased martial activity, while a rather constant trend of the link strength indicates an overall peaceful climate within the network.

4. Discussion

In this paper social network analysis approach was used to describe and analyze the emerged structure as well as to identify key roles of involved actors in the analyzed network of martial conflicts in the years 1816-2001. Using dynamic network analysis and visualization allowed a differentiated view on the historical events and their analysis. The use of analysis techniques from the domain of dynamic social analysis with its added value to the classic quantitative historic analysis can be considered as a possibility to assign roles in martial conflicts to the involved actors.

Furthermore, the role definition in the network allows statements on potential structure change and behavior of the nodes within the network. Several metrics form social network analysis context have been suggested and applied within this research. Nevertheless, not all of them could be justified and semantically explained in the martial analysis context. Thus, not every considered metric from social analysis, like betweenness centrality and brokering activity, was considered for the final metric set (see table 2). Metrics not included in table 2 still bear analytic potential that has to be examined further.

Table 2 contains a suggestion for network metrics, their semantic interpretation and accordant parameter values in the context of martial conflicts to be used.

Table 2. Suggested metrics for conflict analysis using the social network analysis toolset

Metric Parameter value Semantic Interpretation

Degree Centrality High Direct involvement in wars, i.e. war prone countries

Linkevents per Direct Contacts Number Average conflict duration of the country under analysis

Closeness Centrality High Autonomy in war waging

Local Clustering Coefficient Low Conflict initiation potential

Density Distribution High Globalization in current conflict waging

Link Strength Distribution Trend Decreasing Increased war activities in the network

Assigning and determining roles to actors in the observed conflict also enables statements on their importance for the considered network in terms of their martial activities and influence on the network structure. Importance of the nodes in this context is measured by their martial capital based on their role in conflict initiation, participation and spreading. Application of the dynamic analysis technique allowed definition of the development of the network metrics. Hereby not only observation of the per year development of the network density was enabled, but also the cumulated observation of the metrics' dynamics. This metric representation led to the detection of conflict patterns over the years. Thus, an increase of the density metric indicates a conflict that involved a significant number of countries, whereas the fall of the metric indicates the decrease of conflict intensity in the analyzed time period. Thus, temporal evolution of the network metrics allows the visualization and identification of conflicts with a high impact on the network and therefore helps to identify global conflicts. Trend analysis of density and link strengths exhibits a slope of these metrics over time; this fact supports the thesis of an interconnected world - in terms of martial conflicts. Based on these findings, future research will include further semantic interpretation of the gained insights form the network analysis and their evaluation using other historical datasets. The proposed method can be applied to derive insights on martial conflicts but also on other historical interactions.

Acknowledgement

The authors would like to thank the reviewer for the helpful remarks and analysis as well as the editors and the organizers of the ASNA 2011 conference for the support during the publishing process.

References

Flint, C., Diehl, P., Scheffran, J., Vasquez, J., and Chi, S-H. (2009). Conceptualizing ConflictSpace: Toward a Geography of Relational Power and Embeddedness in the Analysis of Interstate Conflict. Annals of the Association of American Geographers, 99(5), 827-835.

Freeman, L. C. (1978). Centrality in social networks conceptual clarification. Social Networks, 1, 215-239.

Ghosn, F., Palmer, G., Bremer, S. (2004). The MID3 Data Set, 1993-2001: Procedures, Coding Rules, and Description. Conflict

Management and Peace Science, 21, 133-154. Gradner, P. (2006). Historical Analysis The SAGE Dictionary of Social Research Methods. London: SAGE.

Grinin, L., de Munck, V.C. (2006). History& Mathematics: Analyzing and Modeling Global Development. Moscow: Editorial URSS.

Maoz, Z. (2010). Networks of Nations: The Evolution, Structure, and Effects of International Networks, 1816-2001. Cambridge: Cambridge University Press

Nahapiet, J., Ghoshal, S. (1998). Social capital, intellectual capital, and the organizational advantage. Academy of Management Review, 23(2), 242-266.

Sparrow, M. K. (1991). The application of network analysis to criminal intelligence: An assessment of the prospects. Social Networks, 13(3), 251-274.

Trier, M., Bobrik, A. (2007). Analyzing the Dynamics of Community Formation Using Brokering Activities Communities and

Technologies (pp. 463-477). London: Springer Wasserman, S., Faust, K. (1994). Social network analysis: methods and applications: Cambridge University Press. Wilkinson, D. O. (1980). Deadly quarrels: Lewis F. Richardson and the statistical study of war: University of California Press.

Wimmer, A., Min, B. . (2009). The Location and Purpose of Wars Around the World: A New Global Dataset, 1816-2001. International Interactions, 35(4), 390-417. Xu, J., Chen, H. (2005). Criminal network analysis and visualization. Communications ofthe ACM, 48, 100-107.