ABSTRACT Title of Document: FOREIGN MILITARY INTERVENTIONS IN CIVIL CONFLICTS, 1946-2002 Pelin Eralp Wolak, Doctor of Philosophy, 2014 Directed By: Professor Jonathan Wilkenfeld Department of Government and Politics Empirical evidence shows that foreign military interventions in civil conflicts on the side of the government or opposition are frequent and they have significant political and economic impacts on both the intervening states and the target states. While many recent quantitative studies have examined the impact of foreign military interventions on the dynamics and outcomes of civil conflicts, similar attention has not been paid to the factors that motivate foreign powers to intervene in intrastate disputes. Most of the theoretical insight on the causes of military intervention comes from earlier qualitative studies that analyze the foreign policy decision making of interveners in detail. In contrast, the small amount of quantitative research conducted on this topic focuses more on the attributes of the civil conflict that attract foreign military intervention. The purpose of this study is to analyze the causes of military interventions from a foreign policy decision making perspective which has been neglected in current quantitative studies. In order to identify the factors that motivate state leaders to use military intervention as a foreign policy instrument, this dissertation examines the international and domestic sources of foreign policy decision making through a modified realist framework. Hypotheses are tested against a novel dataset that includes both actual and potential interveners in all civil conflicts between 1946 and 2002. Sub-sample analyses are also conducted for major powers, democracies and autocracies to understand the relative importance of international, domestic and contextual factors on the intervention decisions of different types of states. The empirical findings show that the strategic significance of the conflict state, interventions by rivals or allies, and domestic considerations of leaders play a more critical role than the attributes of the civil conflict when foreign powers are deciding whether and on whose side to intervene in a civil conflict. While these empirical findings provide an improved understanding of the rationale behind foreign military interventions in civil conflicts, this dissertation also contributes theoretically to the current literature by bringing back the much needed foreign policy decision making perspective into the study of interventions. FOREIGN MILITARY INTERVENTIONS IN CIVIL CONFLICTS, 1946-2002 By Pelin Eralp Wolak Dissertation submitted to the Faculty of the Graduate School of the University of Maryland, College Park, in partial fulfillment of the requirements for the degree of Doctor of Philosophy 2014 Advisory Committee: Professor Jonathan Wilkenfeld, Chair Professor Emeritus George Quester Professor Paul Huth Professor Mark Lichbach Professor Gary LaFree ? Copyright by Pelin Eralp Wolak 2014 ii Dedication Dedicated to my parents Sunay and Bulent, my sister Aylin, and my husband Jeff, without whom this dissertation could not have been written. iii Acknowledgements This dissertation could not have been completed without the support of several people. First, I would like to acknowledge my committee; Professors Jonathan Wilkenfeld, George Quester, Mark Lichbach, Paul Huth and Gary LaFree for their support. In particular, I would like to thank my advisor, Jonathan Wilkenfeld, for his continuous encouragement. He has been an excellent mentor from whom I have learned a lot and I am privileged to be one of his graduate students. I also would like to acknowledge my family who have provided so much moral and financial support since the beginning of my graduate studies. I am extremely lucky to have a wonderful family and a loving husband who believe in me and cheer for me at every stage of my life. iv Table of Contents Dedication ..................................................................................................................... ii Acknowledgements ...................................................................................................... iii Table of Contents ......................................................................................................... iv List of Tables ............................................................................................................... vi Chapter 1: Introduction ................................................................................................. 1 The Significance of Foreign Military Interventions ................................................. 1 Overview of Study .................................................................................................... 5 Outline of Chapters ................................................................................................. 11 Chapter 2: Review of Intervention Literature ............................................................. 13 Early Literature on Foreign Military Interventions ................................................ 13 Contemporary Literature on Foreign Military Interventions .................................. 26 The Effects of Foreign Military Interventions on Civil Conflicts ...................... 27 The Causes of Foreign Military Interventions in Civil Conflicts ....................... 36 Where is Progress Needed? .................................................................................... 57 Chapter 3: Theory ....................................................................................................... 59 ?Modified Realist Framework? for Foreign Policy Analysis ................................. 59 Why are Civil Conflicts Significant to Other States? ............................................. 63 International Sources of Military Intervention ....................................................... 65 Strategic Significance of the Conflict State ........................................................ 66 Strategic Significance of the Civil Conflict ........................................................ 76 Control Variables ................................................................................................ 81 Domestic Sources of Military Intervention............................................................. 83 Contextual Sources of Military Intervention .......................................................... 92 Conclusion .............................................................................................................. 98 Chapter 4: Research Design ...................................................................................... 100 Dataset................................................................................................................... 100 Dependent Variables ............................................................................................. 106 Independent Variables .......................................................................................... 107 International Sources of Intervention ................................................................ 107 Control Variables .............................................................................................. 110 Domestic Sources of Intervention ..................................................................... 110 Contextual Sources of Intervention .................................................................. 111 Statistical Methods ................................................................................................ 114 Other Methodological Issues related to the Research Design .............................. 117 v Chapter 5: Empirical Findings ................................................................................. 122 Empirical Findings from the Full Sample ............................................................. 122 Statistical Models .............................................................................................. 122 Interpretation of Results .................................................................................... 132 Discussion of Findings from the Full Sample................................................... 138 Empirical Findings for Major Powers................................................................... 144 Statistical Models .............................................................................................. 145 Interpretation of Results .................................................................................... 153 Discussion of Findings from the Major Power Sub-Sample ............................ 158 Conclusion ............................................................................................................ 160 Chapter 6: Comparing Democracies and Autocracies ............................................. 163 Empirical Findings for Democracies .................................................................... 164 Statistical Models .............................................................................................. 165 Interpretation of Results .................................................................................... 173 Discussion of Findings for Democracies .......................................................... 177 Empirical Findings for Autocracies ...................................................................... 182 Statistical Models .............................................................................................. 182 Interpretation of Results .................................................................................... 190 Discussion of Findings for Autocracies ............................................................ 194 Conclusion ............................................................................................................ 198 Chapter 7: Conclusion.............................................................................................. 204 Review of the Study .............................................................................................. 204 Summary Findings and Theoretical Implications ................................................. 207 International Sources of Intervention ................................................................ 207 Domestic Sources of Intervention ..................................................................... 210 Contextual Sources of Intervention .................................................................. 213 Policy Implications ............................................................................................... 214 Future Research Directions ................................................................................... 217 Appendix A: Civil Conflicts 1946-2002 ................................................................... 220 Appendix B: Military Interventions in Civil Conflicts ............................................. 227 Appendix C: Variable Descriptions and Coding Rules ............................................ 252 Appendix D: Relogit Models .................................................................................... 258 Bibliography ............................................................................................................. 262 vi List of Tables Table 4.1: Trends in Military Interventions in Civil Conflicts ................................. 105 Table 4.2: Military Interventions according to their Targets in Civil Conflicts ....... 106 Table 4.3: Descriptive Statistics ............................................................................... 113 Table 4.4: Independent variables and their hypothesized relationship with dependent variables .................................................................................................................... 121 Table 5.1: Model 1 - Censored Probit Results for Full Sample ................................ 125 Table 5.2: Model 2 - Censored Probit Results for Full Sample ................................ 126 Table 5.3: Model 3 - Censored Probit Results for Full Sample ................................ 127 Table 5.4: Summary of Support for Hypotheses ...................................................... 128 Table 5.5: Changes in the Predicted Probability of Intervention .............................. 134 Table 5.6: Changes in the Predicted Probability of Government Intervention ......... 137 Table 5.7: Military interventions according to power status of interveners ............. 145 Table 5.8: Military interventions by each major power ............................................ 145 Table 5.9: Model 1 - Censored Probit Results for Major Powers ............................. 146 Table 5.10: Model 2 - Censored Probit Results for Major Powers ........................... 147 Table 5.11: Model 3 - Censored Probit Results for Major Powers ........................... 148 Table 5.12: Summary of Support for Hypotheses .................................................... 150 Table 5.13: Changes in the Predicted Probability of Intervention ............................ 155 Table 5.14: Changes in the Predicted Probability of Government Intervention ....... 156 Table 6.1: Observations according to the regime type of potential interveners ....... 165 Table 6.2: Military interventions according to the regime type of interveners ........ 165 Table 6.3: Model 1 - Censored Probit Results for Democratic states ....................... 167 vii Table 6.4: Model 2 - Censored Probit Results for Democratic states ....................... 168 Table 6.5: Summary of Support for Hypotheses ...................................................... 169 Table 6.6: Changes in the Predicted Probability of Intervention .............................. 174 Table 6.7: Changes in the Predicted Probability of Government Intervention ........ 176 Table 6.8: Model 1 - Censored Probit Results for Autocratic states ........................ 183 Table 6.9: Model 2 - Censored Probit Results for Autocratic states ........................ 184 Table 6.10: Summary of Support for Hypotheses .................................................... 186 Table 6.11: Changes in the Predicted Probability of Intervention ............................ 191 Table 6.12: Changes in the Predicted Probability of Government Intervention ....... 193 Table 6.13: Comparing substantive effects of variables in selection equations ....... 199 Table 6.14: Comparing substantive effects of variables in outcome equations ........ 200 Table D.1: Relogit Results for Full Sample .............................................................. 258 Table D.2: Relogit Results for Major Powers ........................................................... 259 Table D.3: Relogit Results for Democratic States .................................................... 260 Table D.4: Relogit Results for Autocratic States ...................................................... 261 1 Chapter 1: Introduction The Significance of Foreign Military Interventions Although sovereignty and non-intervention in the domestic affairs of other states are fundamental principles of the international system, external states frequently intervene militarily on the side of government or rebels during civil conflicts. A brief look at the major intervention data collection projects reveals the frequency of foreign military interventions since the end of World War II. According to the International Military Intervention Dataset, approximately 650 military interventions were undertaken by major and minor powers during the Cold War period. Between 1990 and 2005, close to 350 foreign military interventions took place which actually indicates that intervention rates increased during the post-Cold War period (Pickering and Kisangani 2009). Of the 138 civil conflicts identified by Regan for the post-World War II period, 85 of them experienced approximately 180 foreign military interventions on behalf of government or opposition forces (Regan 1996). As several international relations scholars have pointed out, foreign military intervention dates all the way back to the Peloponnesian Wars when Athens and Sparta intervened in the internal conflicts of other city states and supported opposing domestic factions (Jentleson and Levite 1992; Young 1968). Morgenthau describes intervention as an ?ancient and well-established instrument of foreign policy? (1967, p. 425) while Young underlines that intervention in the domestic affairs of other states ?has been a recurrent fear of the history of international politics? (1968, p. 2 177). Foreign military interventions, particularly during periods of domestic instability, have been a persistent feature of world politics. The continuing practice of military interventions by major and minor powers alike in the current international system despite the existence of stronger norms of state sovereignty demands a closer look at the rationale behind this foreign policy behavior. In addition to their persistency and frequency as a foreign policy instrument, military interventions usually have significant political, economic and social impacts on both the intervening states and the target states. Even though some military interventions can last only weeks such as the U.S. military interventions in Guatemala (1954) and the Dominican Republic (1965), others can last for years. The U.S. intervention in Vietnam, Soviet intervention in Afghanistan, Cuban intervention in Angola and Syrian intervention in Lebanon are examples of protracted military interventions that had serious political, economic consequences and high human costs for the parties involved. Military interventions are risky policies and they can make civil conflicts longer, more violent and more difficult to resolve. The domestic and international impacts of military interventions for both the intervening state and the conflict state also require a thorough understanding of this phenomenon. Another pressing reason for analyzing foreign military interventions during periods of domestic turbulence has to do with the salience of civil conflicts in the world. Since the end of World War II, intrastate conflicts rather than interstate conflicts have been the most frequent and deadly form of armed conflict in the world 3 according to major conflict datasets in the literature.1 The end of the Cold War and superpower rivalry along with the outbreak of civil wars with devastating human costs in Bosnia, Burundi and Rwanda generated particular interest from international relations scholars and the foreign policy community towards the study of civil conflicts in the early 1990s. In the past 20 years, there has been a significant increase in the number of academic studies examining the causes and dynamics of civil conflicts which are crucial for conflict prevention and conflict resolution purposes. Since external states intervene frequently on the side of either government or opposition during civil conflicts as observed by Regan and other scholars, understanding the impact of third party intervention on the dynamics, escalation and termination of civil conflicts has become even more compelling in the literature. A critical part of understanding the effect of foreign military interventions on civil conflicts lies in examining the causes of this phenomenon by analyzing interveners? rationale for choosing this course of action and supporting one of the domestic disputants against the other. Unfortunately, the last two decades of quantitative research has examined this important foreign policy behavior mostly in an indirect fashion. By putting the civil conflict at the center of analysis, the majority of recent empirical research studied the attributes of civil conflicts that may or may not attract foreign military interventions. In other words, the recent literature neglected the decision making calculus of the intervener and focused more on the contextual and structural factors related to the civil conflict phenomenon that motivate interveners. 1 Correlates of War Project and UCDP/PRIO Armed Conflict Data both show that the number of civil conflicts is dramatically higher than interstate wars. Intrastate disputes constitute approximately 80 percent of all armed conflicts since 1945. 4 In order to explore the rationale behind military interventions, one has to make the potential intervener the theoretical focus and examine why some states decide to intervene militarily in other states? civil conflicts by looking closely at the critical factors that influence potential intervener?s decision making process. Examining international and domestic considerations of state leaders lies at the core of foreign policy analysis. Of course, foreign policy decisions are also influenced by contextual factors including the attributes of the target country and the civil conflict, but putting these attributes at the center of study especially at the expense of international and domestic concerns of potential interveners is problematic. It is the goal of this study to examine the reasons behind foreign military interventions by applying a theoretical framework that puts the potential intervener at the center of analysis while paying due attention to the civil conflict phenomenon foreign policy decision makers are reacting to. The past two decades saw a dramatic increase in the number of quantitative studies that examine various aspects of civil conflicts including the impact of foreign military interventions on the dynamics and outcomes of civil conflicts. However, the relative paucity of quantitative research on the causes of foreign military interventions is problematic. On the other hand, there are a number of comparative and single case studies conducted by international relations scholars between the 1970s and early 1990s that provide an in-depth analysis of the intervention decisions of foreign powers, particularly the United States and the Soviet Union, in civil conflicts around the world. The important theoretical insights provided by these earlier qualitative studies need to be examined in large-N studies to be able to arrive 5 at some generalizations regarding the rationale behind foreign military intervention. Therefore, the purpose of this dissertation is to examine the factors that motivate foreign powers to intervene militarily in civil conflicts on behalf of government or opposition in a large-N study. By putting the intervener at the center of theoretical and empirical analysis and bringing the attention back to the foreign policy decision making calculus of the intervener, this dissertation aims to fill an important vacuum in the current quantitative research on interventions. Overview of Study This research investigates the rationale behind foreign military interventions and identifies the critical factors that motivate states to intervene militarily on the side of government or opposition in another state?s civil conflict. The conceptual definition of foreign military intervention used in this study is the one developed by Rosenau. Military intervention is defined as ?convention-breaking? military activities in the internal affairs of a foreign country directed at ?changing or preserving the structure of political authority in the target state? and these activities must be finite and transitory; an action is not considered intervention if it becomes a permanent feature of the relationship (Rosenau 1969, pp. 161-162). Neutral military interventions conducted by third parties, including the ones by international or regional organizations or coalitions of states, are excluded from this definition. Neutral interventions can be described as conflict resolution efforts aimed at ending the civil conflict and bringing peace without supporting one domestic disputant militarily against the other. Biased military interventions, on the other hand, are different in their political nature and should be studied separately. Thus, the 6 underlying theoretical question driving this study is: why do some states prefer to intervene and pick one side against the other and consequently become a party to the conflict? In essence, this dissertation argues that the decision to intervene and the decision to choose one side over the other are closely linked. In order to understand the rationale behind biased military interventions, one has to analyze the linkage between the decision to intervene and the side to support. These two decisions are generally intertwined and the relationship between them has to be accounted for both theoretically and methodologically. This dissertation aims to contribute to the international relations literature on foreign military interventions by building a much needed theoretical bridge between current quantitative research on interventions in civil conflicts and earlier qualitative studies that analyze military interventions from a foreign policy decision making perspective. While recent studies have been ambitious in providing generalizations about various aspects of civil conflicts that attract foreign military intervention, they have moved away from an actor-centric theoretical framework and examined the causes of interventions indirectly. This research uses a theoretical framework that puts the potential intervener at the center of study by analyzing the international and domestic factors that affect the foreign policy decision making calculus of state leaders. If one assumes that state leaders are rational actors who make cost-benefit analyses and compare the expected utility of different foreign policy options before choosing a particular course of action, then the theoretical task is to examine the critical factors determining the costs and benefits of intervention versus non- 7 intervention; but this task has not been done adequately within contemporary empirical research. Prominent scholars of foreign policy have called for the need to open up the ?black box? and argued that domestic factors are critical in shaping foreign policy behavior (Wilkenfeld et al. 1980, Putnam 1988). Putnam persuasively argued that state leaders are concerned with domestic and international pressures simultaneously (1988). Since then many foreign policy scholars have used theoretical frameworks that combine structural and strategic factors from international and domestic levels of analysis to study foreign policy behavior. This type of theoretical rigor is not adequately applied in current quantitative studies of foreign military interventions. Therefore, this study applies a theoretical framework that examines both international and domestic sources of foreign policy behavior. It then derives theoretical arguments about the critical domestic and international factors that influence the cost-benefit calculation of leaders which consequently produces the foreign policy decision about whether or not to intervene militarily on the side of the government or opposition in a civil conflict. The first major contribution of this dissertation is bringing back the much needed theoretical framework of foreign policy analysis into the study of military interventions and identifying empirical patterns through a more suitable theoretical perspective. Second, this theoretical framework is also useful for exploring how international and domestic sources of foreign policy influence intervention behavior of different types of states. Although empirical evidence shows that both major and minor powers as well as democratic and autocratic states have engaged in military 8 interventions in civil conflicts, it might be wrong to assume that they are all motivated by the same set of factors in a similar manner. The actor-centric theoretical framework used in this study can help identify whether certain critical international and domestic factors have the same effect on foreign policy choices of different types of states. While it is desirable to arrive at some generalizations regarding the rationale behind foreign military interventions, one also needs to check the robustness of such generalizations. Analyzing sub-samples of potential interveners that might be classified in the same group due to some important shared characteristic is an effective way to check how robust such generalizations are. Thus, the hypotheses derived from the theoretical framework will first be tested against all states in the international system to uncover universal patterns. Then, these hypotheses will be tested against three important sub-samples: major powers, democracies and autocracies. These sub-samples are chosen based on international relations scholars? particular interest in understanding the relationship between conflict involvement and power status as well as regime type of states. The purpose of examining these sub- samples is threefold. The first purpose is to check the robustness of findings from all states and determine the level of generalizability about the causes of foreign military interventions. The second purpose is to examine whether and to what extent major powers, democracies and autocracies are motivated by the same set of international and domestic factors when they intervene militarily. Finally, analyzing the similarities and differences between different types of states is useful for assessing the utility of the theoretical framework applied in this study. 9 The hypotheses will be tested in a dataset that includes all potential interveners in all civil conflicts between 1946 and 2002. Potential interveners include all states in the international system except the conflict state. Civil conflicts are drawn from the UCDP/PRIO Armed Conflict Dataset (Gleditsch et al. 2002). Each observation is a pair of potential intervener and conflict state. Both potential interveners and conflict states are observed from the year the civil conflict starts until the year it ends. Hence, the data structure is time-series cross-sectional which makes it a dynamic analysis instead of a static one capturing the changes in the potential intervener?s international and domestic situation. Both of the dependent variables analyzed in this study are dichotomous. The first dependent variable measures whether or not the potential intervener has actually intervened militarily in any year during the life of the civil conflict. If military intervention has taken place, then the second dependent variable looks at whether the foreign state intervened on the side of the government or rebels. Military interventions are operationalized as the supply of troops, military equipment, intelligence and logistical support, air or naval support to one of the sides in civil conflict. The data on military interventions are compiled from six different intervention datasets in the literature. Thus, the military intervention data used in this study is a synthesis of different datasets and can be described as an updated collection of military intervention data which have also been cross-checked and supplemented by multiple sources to study foreign military interventions as thoroughly as possible. As mentioned above, this dissertation argues that the decision to intervene and on whose side to intervene are intertwined. Even though the theoretical analysis in 10 this study claims that these two decisions are closely linked and their interdependence is important for understanding the rationale behind military interventions; applying the right methodology not only can account for this relationship, but it can also test if this relationship really exists. Choosing the right method to test all aspects of the theory is crucial for empirical studies. This dissertation uses a selection model which can test the interdependence between the two dependent variables and account for its effects if it exists. Given that both dependent variables are binary, the most suitable statistical technique for this study is censored probit. As a type of selection model, censored probit can test if outcomes of interest are linked. If they are found to be linked, then censored probit takes into account the level of interdependence when examining the substantive impact of independent variables on the outcome. The first part of the censored probit analysis, which is called the selection stage, will test hypotheses related to the first dependent variable. The second part of the censored probit analysis, which is called the outcome stage, will test hypotheses related to the second dependent variable. The research design of this study has several advantages over the majority of the large-N studies on military interventions in civil conflicts. First, the time-series cross-sectional data structure of this research makes the analysis a dynamic one which is superior to static analyses because it enables one to analyze the impact of changing conditions over time. Second, this research design avoids the selection bias problem that some of the quantitative studies in the intervention literature suffer from. Aside from a number of recent quantitative studies, the majority of these studies have examined the causes of foreign military interventions in a methodologically flawed 11 way due to selecting on the dependent variable. In other words, by testing hypotheses only against cases of intervention these studies have arrived at biased results. The selection bias problem can only be avoided by testing hypotheses against cases of intervention by actual interveners and cases of non-intervention by potential interveners. Including actual interveners as well as potential interveners in the research design solves the problem of selection bias and provides more accurate results. Finally, a censored probit model makes this research design fully equipped to analyze both the decision to intervene and on whose side to intervene while also examining the interdependence between these two outcomes of interest. Selection models have rarely been utilized in the intervention literature, therefore the use of a censored probit model makes this a more comprehensive study of foreign military interventions. The significance of foreign military interventions in world politics is demonstrated by their frequency, persistency as well as their impact on both intervening states and conflict states. This dissertation aims to make theoretical and methodological contributions to the study of foreign military interventions in civil conflicts and advance the current state of knowledge about this critical foreign policy behavior. Outline of Chapters This study will proceed as follows. Chapter 2 provides a review of the literature on foreign military interventions by summarizing its major contributions as well as its theoretical and methodological limitations. Chapter 3 discusses the theoretical framework and develops arguments regarding the rationale behind 12 external military interventions. Hypotheses about the international and domestic factors that shape foreign policy decisions of state leaders are derived from these theoretical arguments. Chapter 4 describes the data collected for this research project and explains the methodology used to test the hypotheses. Operationalization of variables is also discussed in this chapter. Chapters 5 and 6 present the results of the quantitative analyses and discuss the findings in light of theoretical arguments made in Chapter 3. Chapter 7 summarizes key findings and their implications for theory and policy. It also makes suggestions for future research. 13 Chapter 2: Review of Intervention Literature This chapter provides an overview of the literature on foreign military interventions in civil conflicts. The first part of the chapter discusses some of the influential works in the early literature while the second part reviews the significant works in contemporary literature. In addition to surveying the historical evolution of the literature, the chapter also groups these influential works according to the major question driving research and the theoretical and methodological approaches being used. Thus, the purpose of this chapter is to review the major theoretical and empirical contributions of the literature as well as its limitations to determine the areas where progress in needed. Early Literature on Foreign Military Interventions Significant international events have always shaped research interests of international relations scholars. In the case of intervention literature, it was the increasing political ramifications of the Vietnam War in the early 1960s that triggered an academic interest in understanding the relationship between foreign intervention and dynamics of civil conflict. The emergence of the literature on military interventions in civil conflicts can be traced back to an influential volume published in 1964 which was a collection of essays on international aspects of civil conflicts written by prominent foreign policy scholars. This volume discussed the internal attributes of civil conflicts to understand their internationalization process (Rosenau ed. 1964). Although the focus was on the 14 impact of domestic conflicts on international politics, some of the essays drew attention to the subject of foreign interventions. Rosenau argued that international repercussions, including foreign military interventions, of civil conflicts vary according to the scope, duration and type of civil conflict (1964, pp. 61-80). Modelski hypothesized that military intervention is most likely to occur during periods of extreme polarization between domestic disputants because that is when internal parties perceive their survival to be at stake and hence make appeals for external assistance to change the balance of power (1964, pp. 14-44). Finally, Kaplan reflected on the relationship between the type of international system and the likelihood of foreign intervention in civil conflicts, essentially looking at the issue from the potential intervener?s perspective. He argued that bipolar international systems create more incentives for states to intervene than balance-of-power systems (1964, pp. 92- 121). As the U.S. involvement in the Vietnam War escalated in the late 1960s, a number of scholars started to examine the rationale behind military interventions and focus directly on the foreign policy decision making of the intervener. Morgenthau argued that the logic behind interventions remains rooted in the pursuit of national interest and interventions by the U.S. and the U.S.S.R. in domestic conflicts of weaker states could be conceived as competition for influence on particular governments. Rather than ?confronting each other openly and directly?, the U.S. and the U.S.S.R. were competing with each other indirectly through interventions (Morgenthau 1967, p. 428). Morgenthau?s view of interventions as decisions guided by national interest was criticized by Rosenau for not providing a ?meaningful? and 15 scientific explanation of interventions, because national interest was a vague term that did not specify the conditions causing states to intervene (Rosenau 1969, pp. 157- 158). According to Rosenau, ?the factors that foster, precipitate, sustain, channel, constrain, and/or curb intervention? had not been explored scientifically and the ?lack of definitional clarity? in the literature regarding the concept of intervention was partly responsible for impeding the scientific analysis of this empirical phenomenon (1969, pp. 150, 154). Rosenau argued that by defining intervention as ?any action whereby one state has an impact upon the affairs of another?, many scholars were not distinguishing between intervention and other types of influence processes (1969, p. 153). Hence, the first step was to identify the characteristics of intervention that distinguish it from other forms of state interactions and develop an operational definition in order to study this empirical phenomenon scientifically. He posited three criteria that classify state behavior as an intervention: it had to be ?convention- breaking?, ?authority-oriented? and ?finite and temporary? (1969, pp. 161-162). Thus, according to Rosenau, an intervention takes place when the intervening state makes a sharp break with the prevailing manner of relating to the target state and directs its behavior at changing or preserving the authority structure of the target state. However, the ?study of intervention is the study of the unconventional in international politics and, since unconventionality becomes conventional the longer it persists?, interventions have to be finite and temporary (1969, p. 162). These criteria, Rosenau claimed, distinguish intervention from other types of influence processes and facilitate scientific inquiry as well as theory-building by providing an operational 16 definition of intervention. Accordingly, he then focused on building a ?theory of intervention? (1969, p. 165). In order to accomplish his goal of theory building, Rosenau took a methodical approach and questioned the relative importance of individual, bureaucratic, governmental, societal and systemic variables in explaining intervention. He quickly eliminated societal variables arguing that although public opinion can sometimes constrain foreign policy behavior, in the case of intervention the public is rarely concerned about foreign authority structures or ?organized? enough to press for unconventional modes of foreign policy behavior (1969, p. 166). Roseanu also discarded government structure as a source of intervention because both democracies and autocracies had carried out interventions since the end of World War II. Given the small, if any, role played by societal and governmental variables, Rosenau argued that intervention decisions result mainly from the ?perceptions, calculations, and decisions? of state leaders and ?dynamics of bureaucratic structures (1969, p. 166). Although the decision making calculus of top leaders and their bureaucratic staff was important for explaining interventions, one had to also take into account the international context since intervention decisions were not taken in a vacuum. Thus, Rosenau claimed that the structure of the international system, the level of ideological rivalry and the stability of nations within the system were three crucial systemic variables that affect the probability of intervention. He hypothesized that both bipolarity and intense ideological rivalry in the international system increased the likelihood of foreign interventions. Finally, he asserted that state leaders were sensitive to the stability of foreign governments. As the stability of foreign 17 governments decreased, state leaders would be more likely to resort to unconventional modes of foreign behavior to avoid threats or to seize opportunities from unstable situations (1969, pp. 165-169). In short, Rosenau claimed that intervention behavior was a function of international systemic variables and the decision making calculus of top leaders. Rosenau never conducted an empirical analysis of his theoretical arguments, yet his conceptual and theoretical insights influenced the evolution of intervention literature significantly. A number of scholars have used his operational definition of intervention in their theoretical and empirical analyses. In fact, this dissertation also uses Roseanu?s definition. Others have built on his theoretical insights and tested some of his hypotheses empirically. Thus, the conceptual and theoretical beginnings of the intervention literature can be traced back to Rosenau?s influential work. By emphasizing the importance of conceptual clarity and scientific inquiry, Rosenau?s work provided the early foundation for systematic and scientific analysis of interventions in contemporary international relations literature. Mitchell?s study was another important early theoretical work which examined the factors that increase the probability of intervention in civil conflicts (Mitchell 1970). Following Rosenau?s advice on systematic inquiry, Mitchell argued that a theoretical framework for investigating foreign interventions had to look at four major categories of factors: factors within the conflict state, factors within the intervening state, factors within the international system and linkages between the intervening state and conflict state. He claimed that previous studies on intervention, including those by Rosenau and Kaplan, did not pay enough attention to the linkages 18 between the intervening state and the conflict state. Therefore, Mitchell focused on analyzing the nature of linkages that increase the probability of interventions. He identified two types of linkages: transactional and affective. The first type included economic, military, educational and political linkages between the intervener and target country, while the second one involved ideological, religious and ethnic ties. He argued that a third party intervening in a civil conflict ?is already committed, in some quite significant way, to an elite, a class, a social group, or a set of values held by such groups within the disrupted state? and intervention can be viewed as ?an extension of an already existing commitment? (1970, pp. 186-187). He then hypothesized that the probability of intervention increases as these linkages become stronger and more important. Similar to Rosenau, Mitchell also did not test his hypotheses empirically. However, his theoretical arguments influenced the intervention literature in three critical ways. First, by emphasizing transnational factors, he prompted future scholars to study the impact of affective factors, particularly ethnic ties. Second, he drew attention to the dyadic linkages between the intervening state and conflict state which led future works to examine the role of dyadic relations as a source of intervention behavior. Third, he encouraged scholars to look at domestic factors within the intervening state, unlike Rosenau who excluded societal factors from his theory of interventions. The academic interest in understanding the causes of foreign military intervention was becoming particularly apparent in the increasing number of case studies. Some of these case studies adhered to a historical approach. Others aimed at providing theoretical insights about the causes of interventions through an in-depth 19 empirical analysis of a particular case and hence fulfilled Rosenau?s wish for scientific inquiry to some extent. Regardless of whether they took a diplomatic history approach or a case-study approach for the purpose of developing testable hypotheses, this growing intellectual desire to understand foreign military interventions was partly caused by international political events. Intervention was rapidly becoming a key concept for understanding superpower competition and the Cold War order. For example, Hoffman argued that the stability of nuclear deterrence left superpowers ?ample room for interventions aimed at changing the international milieu by affecting the domestic political make-up of other countries? (Hoffman 1984, p. 18). Qualitative studies looking at superpower interventions such as the U.S. interventions in Indonesia (James and Sheil-Small 1971), the Dominican Republic (Lowenthal 1972), Vietnam (Gurtov 1974; Blaufarb 1977; Karnow 1983; Cable 1986; Dietz 1986), El Salvador (Baloyra 1982), Thailand (Girling 1981), Cambodia (Vickery 1984) and Soviet interventions in Czechoslovakia, Afghanistan (Paul 1971; Schmid 1985) emphasized the role of different factors including bipolarity, rivalry, ideology, credibility, prestige as well as strategic and military considerations among others. Superpowers were not the only ones intervening in civil conflicts during the Cold War. European powers also displayed a diverse history of interventions during this period. Some of these interventions could not be fully explained in terms of Cold War politics and bipolarity because they were also related to the process of decolonization and efforts to redefine metropolitan relations with former colonies. Hence, some case studies aimed at explaining military interventions by Britain (Van 20 Wingen and Tillema 1980), France (Corbett 1972; Cronje 1972; Nweke 1976; Stremlau 1977), Belgium (Helmreich 1976) and other European powers. There were also interventions conducted by smaller powers that scholars examined in detailed case studies such as interventions by Syria in Lebanon (Rabinovich 1979; Dawisha 1980; Deeb 1980), Uganda and Libya in Sudan (Howell 1978), and Cuba in Angola and Ethiopia (Durch 1978; Valenta 1978). Compared to the fast accumulation of qualitative works, the development of quantitative literature on interventions was slower in the 70s and 80s. This was partly due to the lack of comprehensive datasets on military interventions in the literature. However, a number of quantitative studies during this period stand out as early attempts to identify empirical patterns related to foreign military interventions in civil conflicts. Analyzing a dataset that included both violent and nonviolent domestic conflicts between 1960 and 1967, Pearson found that violent conflicts were more likely than nonviolent conflicts to attract foreign military interventions, especially interventions on behalf of the government (Pearson 1974a). He also concluded that military interventions, including pro-government ones, prolong the conflict and increase the intensity of violence which was also observed by Gurr and Duvall in their study on civil conflicts (Gurr and Duvall 1973). Hence, Pearson claimed that military intervention is a foreign policy instrument used by both ?large and small powers interested in preserving rather than destroying a target government?, which ironically leads to longer and more violent internal conflicts (Pearson 1974a, p. 286). 21 In another study, Pearson examined the relationship between military interventions and geographic proximity using a dataset which included all foreign military interventions between 1948 and 1967 (Pearson 1974b). This dataset included military interventions during civil conflicts as well as in their absence, because Pearson?s goal was to analyze the monadic use force. He defined military intervention as the movement of troops or military forces of one country across the border of another independent country; thus, military interventions in civil conflicts were only a subset of his dataset. Pearson?s study had some interesting findings regarding geography and military interventions. Large powers were more likely to intervene in distant countries than in nearby countries and the majority of those distant interventions were pro-government. Small and middle powers, on the other hand, were more likely to intervene close by and these interventions were generally hostile to the government, except during civil conflicts. Power disadvantage did not necessarily deter friendly or hostile interventions in nearby states as well. Finally, his data did not show a strong relationship between contiguity and probability of military intervention, especially during civil conflicts. In other words, proximity and being in the same region had a more significant impact on the likelihood of military interventions than contiguity (Pearson 1974b). Even though Pearson?s main motivation was to analyze the monadic use of force by states and examining interventions during civil conflicts was only a part of his research agenda, his studies are some of the earliest examples of quantitative research on interventions in civil conflicts. In fact, Pearson directed his efforts into further data collection and created one of the most comprehensive datasets on military 22 interventions in the literature. The ?International Military Intervention (IMI) Dataset? includes all military interventions in the world between 1946 and 1988 (Pearson and Baumann 1993). Using this new dataset, Pearson and his colleagues described some of the patterns in military intervention behavior of states between 1946 and 1988. For instance, they found that small powers carried out more interventions during this period than superpowers and major powers combined. The majority of the interventions during this period were hostile to the target government, but there was a contrast between small and major powers in terms of the side they support. Most small power interventions were hostile to the target government, while most major power interventions were pro-government (Pearson, Baumann and Pickering 1994). Even though this was mostly a descriptive empirical study which did not employ rigorous statistical techniques utilizing the dataset, Pearson?s dataset is used quite frequently in contemporary quantitative intervention literature. In fact, the IMI dataset has been updated and now covers military interventions between 1989 and 2005 as well (Kisangani and Pickering 2008). An important quantitative study that set out to test Modelski?s and Mitchell?s hypotheses on interventions was Rasler?s longitudinal study of Syrian intervention in Lebanon (Rasler 1983). The variations in Syrian foreign policy during the Lebanese civil war enabled Rasler to apply a time-series analysis. Regarding Modelski?s hypothesis on extreme polarization between domestic disputants, she found that the timing of Syrian interventions was not associated with extreme levels of polarization in the Lebanese conflict. She also did not find strong cooperative ties between Lebanese and Syrians that would support Mitchell?s hypothesis on transnational 23 linkages and intervention. Although this was not a broad cross-national analysis, Rasler?s longitudinal study was one of the first quantitative studies to test some of the earlier hypotheses on intervention. Duner?s study was an empirical analysis of 62 military interventions in seven civil wars that occurred in the 1970s (Duner 1983). By distinguishing between levels of military intervention, he was able to add some interesting results to Pearson?s earlier findings. Duner examined both high level military intervention such as troop movement and combat involvement and low level military intervention such as military training and arms supply, in contrast to Pearson who only looked at high level military intervention. Similar to Pearson, Duner found that the majority of external interventions were carried out by less developed countries rather than industrialized countries and they were also more likely to intervene at high levels. Vicinity was another important factor, as more than half of the interveners were neighboring states. Duner also observed that if one side was supported from outside, the other side also received external military support. Finally, Duner found that low levels of military interventions, especially arms supply, were more frequently used than high levels of military intervention; but when several states intervened on opposing sides in a civil war, the level of intervention escalated from low to high. Even though Duner?s analysis consisted of 62 interventions and he described the lack of comprehensive datasets on military interventions as an obstacle to quantitative research, his study was quite significant because he drew attention to the strategic context of civil conflicts with multiple interveners and their relationship with each other. 24 Tillema?s study reinforced Pearson?s and Duner?s observations about how diffuse the practice of military intervention is by using a new dataset that included 591 foreign military interventions in both interstate and intrastate conflicts between 1945 and 1985 (Tillema 1989). He claimed that traditional theories of power politics would expect that intervention would be mostly used by great powers since they are disproportionately powerful and can resort to intervention more often than less powerful states. However, he found that even though great powers have each intervened more frequently than other states, as a group they were responsible for a small portion of interventions. Tillema?s dataset was an important contribution to the intervention literature which was developing very slowly in terms of conducting large-N cross-national research. The two intervention datasets developed by Pearson and Tillema helped change that in the 90s, but during the 70s and 80s the literature was dominated by case studies. As the Cold War was ending, the literature on military interventions in civil conflicts had made the following progress after Rosenau?s call for a systematic and scientific study of intervention. First, qualitative research was the dominant methodology. Single case studies as well as comparative case studies provided detailed explanations of many foreign interventions carried out by superpowers, great powers as well as minor powers. Second, these in-depth studies frequently traced the causes of interventions to the international system, geopolitics and decision making calculus of top leadership within states similar to what Rosenau hypothesized. The specific factors described as affecting intervention decisions included bipolarity, superpower rivalry, strategic political and military considerations of leaders, 25 economic gains, credibility, resolve, colonial ties and proximity. In many respects, the study of interventions remained rooted in the realist school of thought. Although ideological motivations (promotion of democracy versus promotion of communism) could be categorized as normative considerations, ideology was generally mixed in with realist concerns of power and security and was subordinated to strategic factors. Third, societal and institutional factors were not examined adequately. For instance, the impact of regime type and institutional constraints on intervention behavior of states were rarely analyzed. Despite Mitchell?s theoretical arguments about transnational linkages, affective ties between the intervener and target state were also neglected by most scholars, except a few who paid attention to the relationship between ethnic ties and external involvement in civil conflicts (Suhrke and Noble, 1977; Rothschild 1981). Thus, the dominant theoretical approach among qualitative studies was realist and it was frequently combined with rational actor assumption to analyze factors conducive to intervention such as opportunities, threats, incentives and capabilities from the perspective of the intervener. Despite their theoretical insights and detailed descriptions of specific interventions, case studies have limitations when it comes to providing generalizations. The systematic and scientific inquiry into interventions is constrained by their limited generalizability. However, as discussed above, the development of quantitative literature was slow mainly due to a lack of comprehensive datasets. Moreover, scholars did not use sophisticated statistical techniques and generally revealed descriptive patterns about the intervention behavior of large and small powers with the data they had available. In fact, Tillema acknowledged that ?mere 26 enumeration? of foreign military interventions ?raises as many questions as it answers? because observing such empirical patterns still cannot explain ?what accounts for widespread willingness and ability to resort to military force?? (1989, p. 187). Hence, he argued that military interventions required further systematic study because the common characteristics among large and small interveners were still unknown (1989). Nevertheless, these quantitative works inspired a number of scholars who were determined to study interventions using more rigorous techniques in the 90s. Contemporary Literature on Foreign Military Interventions The end of the Cold War and the rise in the number of intrastate conflicts in the early 90s (UCDP/PRIO Armed Conflict Dataset 2010) led to increasing academic interest in civil conflicts. As intrastate conflicts rose to the forefront of the international agenda, more scholars also started to focus on the role of third party interventions in civil conflicts. It is useful to review the last two decades of literature on interventions by categorizing studies according to the two major themes that have shaped research since the end of the Cold War. The first group of studies examines the impact of third party interventions on the dynamics and outcomes of civil conflicts, while the second group focuses on the conditions for third party interventions. Even though these two research agendas are linked and both are necessary for a comprehensive understanding of interventions, contemporary studies tend to examine the causes and consequences of external interventions in civil conflicts separately. Therefore, it is useful to group studies according to the major 27 research question they are answering and evaluate the accomplishments and shortcomings of the current literature accordingly. It should be noted here that the following sections do not review studies on neutral third party interventions. Even though a number of studies have analyzed the role of neutral third party interventions in civil conflicts since the end of the Cold War, they are not included here because of this dissertation?s focus on biased military interventions. The Effects of Foreign Military Interventions on Civil Conflicts One of the scholars who looked at the impact of foreign interventions on conflict outcomes in the 90s was Regan and his works have influenced the current intervention literature significantly. First and foremost, the two research questions he analyzed in his 1996 and 1998 articles were quite influential in shaping the contemporary research agenda. His first article analyzed the impact of interventions on conflict outcomes, while the second one examined the conditions that lead to external military interventions (Regan 1996, 1998). These two research questions Regan analyzed had a big impact on shaping the two branches of the contemporary quantitative research that aim to understand the causes and consequences of military interventions. Second, Regan used a new dataset that he constructed and this dataset on interventions in civil conflicts has become one of the most frequently used datasets in the literature. Third, despite their major contributions to the literature, both of these studies also had important shortcomings and raised as many questions as they answered. Hence, Regan?s early works triggered an important dialogue among scholars who are interested in understanding the causes and effects of interventions. 28 This section only discusses Regan?s 1996 article because his second article falls within the group of studies that analyze the causes of interventions, therefore it will be reviewed in that section. Using an original dataset that included foreign military as well as economic interventions in civil conflicts between 1944 and 1994, Regan examined the conditions under which biased interventions were successful (Regan 1996). He operationalized success as the cessation of military hostilities for at least 6 months, because he argued that the goal of intervention was to stop the fighting. Regan did not deny that states could have various goals when they intervene in civil conflicts, but he argued that the first step in achieving those other goals was to stop the fighting. Hence, Regan?s study analyzed the impact of biased interventions on civil conflict outcomes but the particular outcome he was interested in was the cessation of hostilities (Regan 1996). Regan found that interventions that include a mix of military and economic strategies were more effective than either alone. In other words, the probability of success increases when an external state carries out both military and economic interventions. Compared to the use of military or economic interventions alone, mixed interventions also had the highest probability of ending hostilities regardless of the identity of the intervener. Both major and minor powers were more likely to stop the fighting in civil conflicts when they used mixed interventions. Regan?s results also showed that supporting the government over the opposition increases the chances of success. However, regardless of the side they support, biased interveners increased the level of violence in civil conflicts. Since his results did not indicate that the 29 characteristics of the dispute, such as its intensity or type, influence the cessation of hostilities significantly; Regan concluded that the characteristics of the intervention were more critical in stopping the fighting. In short, mixed interventions that combine military and economic strategies and interventions on the side of the government increased the likelihood of success significantly according to his results (Regan 1996). Regan?s new dataset as well as his findings regarding the impact of biased interventions on the cessation of hostilities in civil conflicts attracted a lot of scholarly attention and motivated others to examine the effect of interventions on conflict dynamics. However, his article also stimulated an important debate among scholars about the goals of interveners. By defining success as the cessation of hostilities, Regan assumed that states intervene to end the fighting as soon as possible and that they measure success in such terms. Thus, he argued that the main goal of intervention is to stop the fighting. Regan?s argument about the goal of biased interventions has been challenged by many scholars. Some were against such a generalization and claimed that biased interventions can have other goals instead of ending the fighting. Others criticized his formulation of success as the cessation of hostilities and suggested that success should be defined as the fulfillment of particular goals set by the intervener and the analyst should first identify those goals in order to measure success. Finally, some scholars argued that the goals of intervention depend on the reasons that lead a state to intervene in the first place; hence it is premature or even misleading to make an assumption about goals before analyzing the causes of interventions. In short, Regan?s assumptions triggered an important conversation 30 about the goal of biased interventions which in turn influenced the theoretical approaches some scholars took in analyzing the impact of interventions on civil war dynamics as well as in examining the causes of interventions. Regan?s study also raised serious methodological questions as a result of sampling on instances of intervention. Regan tested his hypotheses about success against data that only included actual interveners. Since the decision to intervene might be due to intervener?s expectation of success; when data on potential interveners is left out, the analyst is essentially examining the determinants of success by sampling on interveners who might have expected to be successful in the first place. The decision to intervene imposes a selection bias problem on studies that measure the effectiveness of interventions on ending civil conflicts or on achieving some other outcome, such as negotiated settlements, specified by the analyst. If states choose not to intervene because they expect to be ineffective or unsuccessful in achieving that outcome, then the analyst is practically testing his hypotheses against cases that might have expected to be successful and effective. Regan?s study was one of the first in the literature to demonstrate this selection bias problem. Even though Regan fixed this problem in a future study (Lemke and Regan 2004); there are still plenty of studies that suffer from a selection bias problem in the intervention literature. Moreover, this selection bias problem shows up in different forms depending on the research question and the way it is examined. In majority of the cases, selection bias problem can be easily fixed by including potential interveners and by using appropriate statistical techniques. It was criticisms raised against 31 Regan?s study that led some scholars to take the selection bias problem seriously and use the appropriate methodology. Since the late 1990s, an increasing number of studies have explored the effect of foreign military interventions on the duration and intensity of civil conflicts, while others have examined whether biased military interventions lead to negotiated settlements or military victory by one side. Some have combined both dimensions and analyzed the effect of interventions on conflict duration and outcome simultaneously. The majority of studies that examine the relationship between military interventions and duration have arrived at similar conclusions. For instance, Elbadawi and Sambanis (2000) and Regan (2002) found that external military interventions increase civil war duration regardless of the side they support. Likewise, the results of Balch-Lindsay and Enterline?s study (2000) showed that one-sided or biased interventions increase conflict duration but balanced interventions, in which multiple external states intervene to support each side, lead to even longer conflicts by creating stalemates and making it harder for one side to win. Cunningham?s study (2006) supported the finding that civil conflicts with multiple interveners have longer durations but also showed that they are more difficult to resolve through negotiations. Analyzing the effect of interventions based on the identity of interveners, Akcinaroglu and Radziszewski (2005) found that military interventions by rival states lead to prolonged fighting, while interventions by nonrivals are associated with shorter wars. Moreover, the probability of negotiated settlement is also lower when a rival state intervenes on the side of rebels. One of the few studies that reached a 32 different conclusion was by Collier, Hoeffler and S?derbom (2004). Their research indicated that military interventions on the side of rebels can shorten civil conflicts, while pro-government interventions have no significant effect on duration. Most studies, however, seem to agree that biased military interventions, particularly countering interventions, tend to prolong civil conflicts. There is less empirical consensus in the literature regarding the impact of biased military interventions on conflict outcomes. For example; Mason, Weingarten and Fett (1999) found that biased military interventions lead to an increased probability of military victory by the supported side and a decreased probability of a negotiated outcome. Balch-Lindsay, Enterline and Joyce (2008), on the other hand, described the effect of biased interventions on conflict outcomes as somewhat unclear. According to their study, biased military interventions can shorten conflict duration by decreasing the time until the supported group achieves military victory; but they can also decrease conflict duration by increasing the probability of negotiated settlements. Balanced or countering military interventions, on the other hand, prolong conflicts and decrease the likelihood of a negotiated settlement while increasing the probability of government victory. In Gent?s study, the results were also quite different. His study indicated that rebel-biased interventions increase the probability of rebel victory as well as the probability that a conflict ends in a negotiated settlement, regardless of whether or not there is a countering intervention on the government side. In contrast, government-biased interventions do not have a statistically significant effect on any of the outcomes. Gent argued that this was not a 33 surprising finding since external states are more likely to intervene when rebels are stronger (Gent 2008). The assumption at the core of Gent?s theoretical approach was different than the assumption Regan made about biased military interventions and Gent?s study demonstrated how different assumptions can affect theory building. He disagreed with Regan that external states intervene to end the fighting as soon as possible. Instead, Gent argued that the main goal of third parties is to influence civil conflict outcomes and a theoretical approach based on this assumption leads to an improved understanding of the effects of military interventions. Since the main goal of potential interveners is to influence the conflict outcome, Gent posited that the probability of military intervention both on the side of government and on the side of rebels increases when rebels are stronger. Pro-government military interventions are more likely because stronger rebel groups pose a legitimate threat to the government. Governments can generally defeat weaker rebel groups without outside support, so external states are more likely to intervene to shift the outcome of the conflict towards government victory when there is a stronger rebel group. Since the main effect of biased military intervention is to change the balance of power in favor of the supported side, Gent expects that third parties would use interventions for that purpose. Similar logic applies to military interventions on the side of rebels. External states biased towards rebels are more likely to intervene when rebels are stronger and have a greater chance at defeating government forces. Given that biased interventions on both sides are more likely when rebels are strong, pro-government interventions seem to be less effective than pro-rebel interventions. In other words, government- 34 biased third parties intervene in the ?toughest cases?, while rebel-biased third parties intervene in the ?most favorable cases? which explains the different results in Gent?s statistical analysis (2008, p. 730). Another scholar who questioned the assumptions being made in existing studies about the goals of biased military interventions was Cunningham (2010). He argued that the relative lack of empirical consensus regarding the impact of interventions on the termination of civil conflicts was partly caused by the assumptions scholars made. According to Cunningham, existing studies assumed that external states intervene either to help one side achieve a military victory or to contribute to the resolution of the conflict through a negotiated settlement. He argued that external states can also intervene to pursue objectives other than ending the conflict. In other words, when external states become involved in civil conflicts, they can have an independent agenda which is separate from the goals of domestic disputants. The results of Cunningham?s study indicated that when third parties intervene in a civil conflict not to end the conflict but to pursue an independent agenda, they make civil conflicts much longer and more difficult to resolve (Cunningham 2010). Thus, according to Cunningham, the findings in the literature that external military interventions prolong civil conflicts were caused by the subset of military interventions in which the intervener had an independent agenda. By showing that external states can intervene to advance their specific objectives, not necessarily to resolve the conflict or to help one side win; Cunningham?s study also drew attention to the critical role of assumptions regarding intervener goals for 35 understanding the effect of biased interventions on civil conflict dynamics and outcomes. As the studies reviewed above demonstrate, the literature has made some progress in terms of analyzing the impact of military interventions on civil conflict dynamics and outcomes. However, not all scholars seem to agree on the goals of biased interveners when they are theorizing about the effects of military interventions on conflict dynamics and termination. Some of the contradictory findings in the literature are partly caused by different assumptions scholars make about intervener?s goals. Yet, findings such as biased military interventions by rival states or by external states that have an independent agenda leading to substantially longer civil conflicts imply that examining the causes of interventions is quite critical for understanding the consequences of interventions. Studies conducted by Akcinaroglu and Radziszewski (2005), Gent (2008) and Cunningham (2010) demonstrate the need to look at the goals of interveners more closely. If the goals of interveners affect the consequences of intervention as some scholars argue, then one has to carefully examine the reasons that motivate a state to intervene in the first place. The final section of this chapter will review studies that focus on the causes of intervention. Since the purpose of this dissertation is to understand why foreign powers intervene in civil conflicts and how they choose sides, the next section is particularly important for situating this study in the intervention literature. 36 The Causes of Foreign Military Interventions in Civil Conflicts As mentioned in the previous pages, the second branch of the contemporary intervention literature aims to understand the causes of and conditions for third party interventions in civil conflicts. Scholars have examined the factors that motivate external states to intervene in civil conflicts from three different theoretical perspectives. The first one views intervention as a response to civil conflict conditions and explores the attributes of the conflict that increase the likelihood of external involvement. The second approach views intervention as a foreign policy instrument and analyzes state level and international level factors that cause potential interveners to use this foreign policy tool in civil conflicts. The third approach can be described as the combination of the first two, because it examines both the characteristics of the potential intervener and the civil conflict as well as the interactions between the two. These three approaches can be distinguished from each other by identifying what the study puts the main theoretical and empirical emphasis on. The first approach, conflict-oriented approach, puts the theoretical and empirical emphasis on the civil conflict whereas the second one, actor-oriented approach, puts the intervener at the center of theoretical and empirical study. The third approach, combination approach, makes the intervener theoretically central similar to the second approach but pays more attention to the attributes of the civil conflict during the theory building phase. The third approach is still at a nascent stage within the current literature and there are only a handful of studies that can be categorized as using this approach. In many ways, this newest approach is a reaction to the 37 shortcomings of the conflict-oriented approach. It is also an attempt by scholars, who are critical of the conflict-oriented approach, to incorporate the strengths of the actor- oriented research into a new analytical approach for studying the causes of interventions quantitatively. While not all studies are entirely pure examples of each approach, the dominant perspective used by the scholar can still be identified based on the theoretical and empirical focus of the study. Thus, it is a useful way to review the studies in this section and evaluate their contributions. Many empirical studies in contemporary literature have applied the first theoretical approach, which this dissertation describes as the conflict-oriented approach, to study foreign military interventions in civil conflicts. These studies analyze the conditions under which civil conflicts experience interventions. Investigating various contextual and structural factors related to the civil conflict and/or conflict state is common within this approach. By examining the characteristics of the civil conflict that increase the probability of foreign military interventions, these studies are essentially interested in understanding ?what happens to? the civil conflict (Findley and Teo 2006, p. 828). In other words, the primary theoretical focus is on the civil conflict and/or civil conflict state rather than the intervener. Nevertheless, studies using the conflict-oriented approach have provided useful insights about some of the factors that motivate foreign powers to intervene in civil conflicts and discovered some important empirical patterns about intervention behavior of states. 38 One of the best examples of the conflict-oriented approach is Regan?s 1998 study which looks at the conditions that motivate third parties to intervene in civil conflicts. As mentioned previously, this is the other early study conducted by Regan which influenced subsequent quantitative works on intervention due to its strengths as well as its weaknesses. Regan (1998) argues that hypotheses about conditions that increase the probability of intervention in civil conflicts can be derived from two dominant theories in international relations, realism and idealism. First, he hypothesizes that the greater the number of countries bordering the conflict state, the greater the probability of intervention. Conflicts in contiguous states increase threat to national security due to spillover effects while proximity decreases the costs of intervention according to Regan. Second, the probability of intervention decreases as the intensity of civil conflict increases because violent conflicts require greater and costlier military commitment to end the fighting which Regan claims is the goal of interveners. Third, he posits that the likelihood of intervention increases when there is a humanitarian crisis involving large numbers of refugees because it generates humanitarian concerns. Regan?s final hypothesis is that interventions in civil conflicts should be more likely during the Cold War than post-Cold War period because incentives to intervene for geostrategic reasons decrease as the system becomes less polarized (Regan 1998). The hypotheses on contiguity, intensity of the conflict and Cold War capture realist motives according to Regan while the hypothesis on humanitarian concerns focuses on idealist motives of states. 39 These hypotheses are tested against the original dataset from Regan?s 1996 study which includes military and economic interventions in 138 civil conflicts between 1944 and 1994. Regan finds that an increase in the intensity of the conflict and in the number of contiguous countries decreases the probability of military intervention; while Cold War and the existence of a humanitarian crisis increase the likelihood of intervention (Regan 1998). This study is an important step towards finding some empirical patterns about intervention behavior and Regan contributes to systematic and scientific progress of the field, as Rosenau hoped for, by testing his hypotheses against a large number of civil conflicts over a long period of time. However, similar to his 1996 study, this research also suffers from some serious theoretical and methodological problems. First, looking at the findings of his study Regan concludes that neither realism nor idealism is able to explain interventions adequately. The problem with this conclusion is that the four hypotheses Regan derives are not capable of testing the explanatory power of idealism and realism entirely. These hypotheses are simply inadequate to do such a big task because they leave out many other strategic or moral concerns of states that can be identified as realist or idealist. Hence, Regan?s conclusion about the explanatory power of dominant theories regarding intervention should be questioned. Second, and more importantly, instead of examining the motives of interveners, Regan analyzes the civil conflict conditions that increase the probability of intervention. He completely neglects the characteristics of the intervener by framing his research question from the perspective of the civil conflict and by theorizing about the attributes of the conflict that attract third party 40 intervention. His analytical framework is ill-suited to answer why external states intervene in civil conflicts, because it essentially asks under what conditions civil conflicts experience military interventions. As a result, his methodological approach is also inappropriate to examine why external states intervene because the unit of analysis in his study is civil conflict. As Regan himself admits, the theoretical and empirical focus of the study shifts from the intervener to the civil conflict when the latter is the unit of analysis (Regan 1998, p. 768). Finally, similar to his previous work, this study also suffers from selection bias problem due to the exclusion of potential interveners from the analysis. Regan?s study is an example of the conflict-oriented approach that contributes indirectly to the study of the causes of military interventions. There are many other scholars who hypothesize about the types of conflicts that attract foreign interventions and shed light on some of the factors that motivate interveners. One particularly important theoretical argument about interveners? motivation comes from scholars of ethnic conflict. Scholars interested in the role of ethnicity in international relations have shown that ethnic affinities can motivate external states to intervene in conflicts with an ethnic component (i.e. Heraclides 1990; Gurr 1992; Roy 1997). In several different studies, Carment and James and their colleagues argue that ethnic conflicts make third party involvement, including biased military interventions, more probable when ethnic ties exist between external states and the target state (Carment and James 1995; Carment, James and Rowlands 1997; Carment et al. 2006). These arguments are consistent with Mitchell?s hypothesis on affective linkages between interveners and 41 targets. However, these scholars are theoretically and substantively interested in ethnic conflicts and they argue that ethnic conflicts are prone to experiencing interventions because of their ethnic component. Given that the primary theoretical and empirical goal is to understand the impact of ethnicity on international and domestic politics, third party intervention is a good example to test their hypotheses on ethnicity. Several studies demonstrate the theoretical and empirical purpose of examining interventions particularly well. For instance, Saideman argues that ethnic composition within states influences their foreign policies toward ethnic conflicts in other states and intervention policies generally demonstrate this influence. Since politicians care about gaining and maintaining office, state leaders pay attention to their constituents? preferences and ethnic identities can shape those preferences. If ethnic ties with an outside group determine foreign policy preferences of constituents, then those ethnic ties will also influence state leaders? foreign policy decisions. Thus, domestic politics plays an important role in the ethnic ties argument because leaders seeking support at home take into account the ethnic composition of the population and their constituents? ethnic ties when formulating policies, such as intervention, toward an ethnic conflict (Saideman 1997, 2002). Similarly, Carment and James show how leaders often choose to intervene in ethnic conflicts on behalf of combatants that share ethnic ties with their constituents in order to maintain domestic political support (Carment and James 1996). However, leaders also consider the costs of intervention when choosing the type of intervention strategy according to Carment and James. In order to demonstrate solidarity with an 42 outside ethnic group and to appease their domestic constituents, state leaders can use diplomatic interventions which are less costly than military interventions. In fact, the authors claim that leaders opt for military interventions when costs and benefits are highly concentrated which is more likely to happen in authoritarian states where one ethnic group is dominant or alternatively when the cost of military intervention is relatively low. For example, it is less costly to intervene militarily on behalf of stronger ethnic groups. Since the cost of intervention depends partly on the strength of the group supported, stronger ethnic groups are more likely to receive external military support according to Cetinyan as well (2002). Finally, Carment and James argue that military interventions in ethnic conflicts are also more likely when ethnic affinities are backed by instrumental motives such as international and regional strategic considerations of leaders (Carment and James 1996). Another scholar analyzing what types of third party intervention ethnic conflicts attract is Khosla (1999). She focuses particularly on ethnic conflicts in developing countries and uses MAR data which contains information about various types of third party support to minority groups involved in ethnic conflicts between 1990 and 1998. Khosla finds that military interventions were the most common type of third party involvement followed by diplomatic and economic interventions in these ethnic conflicts. The majority of these military interventions were undertaken by contiguous states and regional powers. Neighbors and regional powers intervened militarily in support of ethnic groups in sixty percent of their interventions. In contrast, major powers generally preferred non-military interventions (Khosla 1999). 43 Khosla?s study does not analyze the impact of ethnic ties on third party intervention; instead, she is interested in examining who intervenes in ethnic conflicts and how. Ethnic conflicts have attracted a lot of attention from scholars who aim to understand how ethnicity influences domestic and international politics. However, their studies have also contributed to the intervention literature by demonstrating how ethnic ties can influence the decision making of potential interveners. In fact, one can argue that their studies put the potential intervener more at the forefront of analysis than Regan does by paying attention to the domestic politics of potential interveners. However, these studies are still limited in providing a complete understanding of the rationale behind military interventions. First, external states also intervene in non- ethnic civil conflicts. Second and more importantly, these scholars do not examine the impact of other factors that originate from different international, dyadic and state level sources on intervention behavior of states. As Carment and James admit, there are other motives besides ethnic ties that lead to military interventions (Carment and James 1996). Another recent important study that focuses mainly on the attributes of the conflict to analyze intervention behavior of states is Koga?s. Although she uses a conflict-oriented approach, her study pays attention to the regime type of potential interveners. She examines whether certain characteristics of the civil conflict affect military interventions by democracies and autocracies differently (Koga 2011). Hence, Koga?s study is theoretically more refined than Regan?s because she argues that various attributes of the conflict affect the probability of intervention by democracies and autocracies differently. Second, by including potential interveners in 44 her analysis, Koga?s study avoids selection bias problem faced by Regan and provides more reliable results. Koga argues that democracies are more likely to intervene when there is public support for the decision because democratic leaders rely on reelection to stay in power. Additionally, democratic leaders are more concerned about the success or failure of military interventions than autocratic leaders because they are afraid of losing their jobs due to policy failure. Thus, democratic leaders are more selective when they intervene and pay more attention to the likelihood of success than autocratic leaders who can maintain their positions by satisfying members of their winning coalition with private goods (Koga 2011). After drawing these theoretical arguments from the ?selectorate theory? developed by Bueno de Mesquita et al. (2004), Koga hypothesizes that the existence of ethnic ties increases the probability of military intervention by democracies but not by autocracies. She also expects that the existence of lootable natural resources, such as secondary diamonds, in a civil conflict state increases the likelihood of military intervention by autocracies because autocratic leaders can exploit these resources to provide private goods. On the other hand, lootable resources in conflict state do not affect the probability of military intervention by democracies. She also posits that in order to increase their likelihood of success, democracies are more likely to intervene on the side of rebels when they are stronger. This last hypothesis is based on Gent?s study which argues that stronger rebels increase the probability of military intervention on both government and rebel side (Gent 2008). However, Koga does not expect that democracies will intervene on the side of government when 45 rebels are stronger because pro-government interventions are more likely to fail under such circumstances. Yet, democracies are also unwilling to intervene when rebels are weaker because governments can defeat weaker rebels without external help. Hence, she cannot make a prediction about when democracies are more likely to intervene on the side of governments which is problematic. Based on her argument about democracies? concern for success, Koga?s theoretical framework expects democracies either to intervene on the side of rebels when they are stronger or not intervene at all when rebels are weak. Thus, her theoretical framework is not able to explain why democracies intervene on the side of governments. This is both a function of her conflict-oriented theoretical approach which limits factors shaping the intervention decision to the attributes of the conflict and also a result of her attempt to measure probability of success in a rather one-dimensional way. Koga tests her hypotheses using Regan?s military intervention data in civil conflicts between 1945 and 1999. The results show support for all three hypotheses. Lootable resources only increase the probability of military intervention by autocracies while ethnic ties only make it more likely for democracies to intervene. She also finds that stronger rebels increase the likelihood of intervention by democracies on the rebel side but has no significant impact on pro-government interventions. As mentioned above, her study cannot conclude much about pro- government interventions. Koga?s statistical results are more reliable than Regan?s because she includes all potential interveners in her analysis. Her theoretical approach is also more nuanced because she distinguishes between democracies and autocracies. Even though Koga includes the regime type of the intervener to examine intervention 46 decisions, she still views intervention as a response to civil war conditions and excludes many critical international factors that affect potential interveners? decision making calculus. Koga?s study demonstrates clearly that not all studies using the conflict- oriented approach neglect the intervener completely. Scholars studying ethnic conflicts pay attention to ethnic politics within the intervener state while others like Koga pay attention to intervener?s regime type. Inclusion of one or two attributes of the intervener, however, does not change the fact that civil conflict conditions drive their analytical approach. Some studies, on the other hand, put the intervener at the center of theoretical and empirical analysis and view intervention as a foreign policy instrument. This second approach, which focuses on the decision making calculus of the intervener, can be described as the actor-oriented approach. Studies applying an actor-oriented approach analyze state level and international level factors that cause potential interveners to use this foreign policy tool in civil conflicts. Similar to the case studies conducted in early intervention literature, these contemporary studies make the intervener?s decision theoretically and methodologically central. In fact, some of the best examples of this approach are still found among qualitative works. In a comparative study of superpower interventions in the Middle East and Africa during the Cold War, Feste argues that American and Soviet military interventions in civil conflicts were driven by bipolarity, rivalry and geopolitical security pursuit. Other factors that motivated superpower interventions included location, alliance commitments, economic ties and strategic significance of the civil conflict country 47 (Feste 1992). Scott also uses a comparative case study method and examines U.S. interventions in Angola, Cambodia, Mozambique and Nicaragua during Reagan?s presidency (Scott 1996). The Reagan administration ?authored a policy initiative to take advantage of the rise of several anti-Soviet insurgencies around the world and provide the rebels with American assistance? which ?became known as the Reagan Doctrine? (Scott 1996, p. 14). According to Scott, the ?primary concern with the targets of [intervention] strategy? in the Third World did not originate from ?the issues or events inside? the civil conflict state; instead the Reagan doctrine used interventions to confront the Soviet Union and to challenge Soviet advances (Scott 1996, p. 225). Both of these actor-oriented studies draw attention to the strategic nature of intervention decisions and demonstrate with comparative case studies how interventions were taken as reactive responses to each other. Such strategic interactions between potential interveners have not been analyzed adequately in the quantitative intervention literature. An actor-oriented approach is used less frequently in current quantitative literature. One of the best quantitative works using this approach is by Yoon who examines the causes of U.S. military and nonmilitary interventions in third world civil conflicts between 1945 and 1989 (Yoon 1997). Since Yoon uses an actor-oriented approach, his study focuses on strategic, economic and domestic factors that motivate the U.S. to intervene in these intrastate conflicts. Strategic factors that increase the likelihood of U.S. intervention include intervention by the U.S.S.R or its allies, proximity of the conflict country and its strategic importance measured by the amount of American military assistance. Economic interests of the U.S. that affect its 48 intervention probability are measured by U.S. foreign investment in the conflict country and the level of trade between the two countries. In terms of the domestic context, Yoon hypothesizes that the U.S. is less likely to intervene during an election year. Moreover, based on the arguments of diversionary theory, he expects that the U.S. should be more likely intervene when it is experiencing domestic economic difficulties measured as the sum of unemployment and inflation rates (Yoon 1997). The factors analyzed by Yoon include some of the same critical factors emphasized in the comparative case studies of Feste and Scott. The results of Yoon?s study show that intervention by a Soviet ally and the existence of a communist domestic opponent in a civil conflict increase the probability of U.S. intervention substantially. In contrast, other strategic interests such as proximity of the conflict country or its importance measured by the extent of U.S. military assistance have only a small impact. There is also no support that Soviet intervention leads to U.S. intervention. Yoon also finds that the U.S. is less likely to intervene in an election year and when its economy is worsening. Finally, economic interests seem to have no impact on the U.S. decision to intervene. According to Yoon, the findings indicate that strategic issues such as intervention by Soviet allies, ideology and rivalry are compelling factors behind U.S. interventions while domestic issues are the constraining factors (Yoon 1997). These empirical results demonstrate the importance of understanding the strategic interests of interveners as well as the strategic interactions between potential interveners which is only possible by giving due attention to the actor making the foreign policy decision. 49 By using an actor-oriented approach and putting the intervener at the center of study, all three scholars draw attention to the strategic issues that affect intervener?s decision making. Unlike the conflict-oriented approach that provides only suggestive evidence about the motivations of interveners, the second approach uses a foreign policy perspective and analyzes the causes of intervention more directly. Therefore, this approach is able to provide more insight about the rationale behind military interventions. The actor-oriented approach views the intervention decision as originating from international influences, strategic issues and domestic constraints. Even though international and domestic level factors should be central to the analysis of intervention decision, scholars still need to take into account how contextual conditions of the civil conflict shape interveners? decision making. Civil conflicts are dynamic processes and ?potential interveners undertake evaluations of the changing civil war context? (Findley and Teo 2006, p. 829). Contextual factors such as duration, intensity, military capabilities of domestic opponents can affect interveners? decisions. As Findley and Teo also argue, neither the actor-oriented nor the conflict- oriented approach is ?complete on its own? (2006, p. 829). Combination of the two provides a better understanding of the rationale behind interventions. However, one should be careful while integrating these two approaches in a theoretical framework to study interventions. The potential intervener should still be theoretically and methodologically central as in the case of actor-oriented studies, but the capacity of the theoretical framework to analyze intervention decisions should be enhanced by including the critical contextual factors related to the civil conflict. 50 Only a few recent quantitative studies have applied what this dissertation describes as the ?combination approach? and the first application of this approach comes from Regan again. In an attempt to address some of the theoretical and methodological criticisms he received, Regan re-examined the conditions that increase the success of foreign interventions using the same intervention dataset from his 1996 study (Lemke and Regan 2004). In order to fix the selection bias problem Regan faced in his previous study, the authors included potential interveners in their empirical analysis in addition to the actual interveners in 138 civil conflicts examined in the dataset. They also used a selection model which is the most suitable method to study the success of actual interveners. The first equation of their selection model examined the factors that increase the probability of intervention in a dataset that includes both actual and potential interveners. Then, in the second equation of the model, the authors analyzed the variables that increase the probability of success after potential interveners are censored. The use of a censored probit model with all potential interveners included in the first stage of analysis enabled the authors to avoid selection bias problem and provide more reliable findings in the second stage of their statistical analysis on the determinants of success. However, it is the first stage of their selection model analyzing the conditions for intervention that is critical for this literature review. The analytical framework Lemke and Regan used to examine the factors that increase the probability of intervention contained components of both the actor-oriented and conflict-oriented approach. Therefore, this study can be considered as a precursor to the current applications of the combination approach in the literature. 51 Similar to what Yoon did in his actor-oriented study, Lemke and Regan (2004) tried to evaluate the strategic importance of the conflict country for potential interveners and found that contiguity, alliances and colonial ties increase the probability of intervention. They also included power status and regime type of potential interveners in their analysis. The results showed that major powers were more likely to intervene in civil conflicts than minor powers but regime type had no significant impact on intervention behavior. In terms of the attributes of the civil conflict, the authors found that high numbers of casualties and refugees increased the probability of intervention. Although their theoretical framework was not as sophisticated as Yoon?s, Lemke and Regan paid attention to the regime type of external states and to some dyadic factors that increase the strategic significance of the conflict country for the potential intervener. As a result of incorporating some elements of the actor-oriented approach, this study affected several important quantitative studies that came after it. These later works have further improved the theoretical application of the combination approach. The first quantitative study to integrate actor-oriented and conflict-oriented approaches in a theoretically refined manner is conducted by Findley and Teo (2006). Their theoretical framework resembles the one used by actor-oriented studies in two ways. First, they adopt an interest-based explanation of intervention similar to scholars like Feste and Yoon who rely on realist school of thought to determine the key interests of potential interveners. Second, Findley and Teo pay attention to the strategic environment in which potential interveners observe each other?s behavior toward the civil conflict. As discussed previously, strategic relations between 52 potential interveners are central to actor-oriented studies. Thus, the authors examine whether an external state is more likely to intervene militarily when its ally or rival intervenes in a civil conflict. The findings show that military intervention by a rival state supporting one side in a civil conflict increases an external state?s probability of intervention on the opposite side. Military intervention by an ally, on the other hand, does not seem to increase the probability of intervention by an external state on the same side. In fact, external states are more likely to intervene on government side when an ally intervenes on the side of opposition (Findley and Teo 2006). In order to gauge the interests of the potential intervener in the civil conflict state, Findley and Teo look at alliance ties and rivalry relations between these two states as well. They find that an external state is more likely to intervene on the opposition side when the conflict state is a rival. However, alliance ties with the conflict state increases the probability of military intervention on both sides (Findley and Teo 2006). Although the relationship between alliance ties and military intervention is somewhat unclear, especially compared to the results regarding rivalry; this is the only quantitative study that takes into account such strategic relations between potential interveners and civil conflict states as well as between third parties themselves. It is also one of the few studies which uses these strategic relations to predict on whose side intervention will take place. Findley and Teo also argue that dyadic relations between the external state and the conflict state affect the probability of military intervention. Similar to Lemke and Regan (2004), they find that colonial ties and contiguity increase the probability of military intervention while joint democracy decreases it. Since they are applying a 53 combination approach, the authors also examine how various attributes of the civil conflict influence the likelihood of military interventions. The results show that intervention is positively associated with increasing casualty and refugee levels. However, the relationship between type of civil conflict and military intervention is ambiguous because both types of civil conflicts they analyze, ethnic and ideological, appear to increase the probability of biased military interventions (Findley and Teo 2006). Findley and Teo deserve credit for applying a unified theoretical framework and making the potential intervener theoretically and methodologically central to their study. However, their study has some weaknesses. Despite their claim to use an interest-based explanation, they still fall short on measuring the strategic significance of the civil conflict state from the perspective of the potential intervener. Rivalry and alliance relations are extremely important but there are additional strategic reasons which are worth investigating. Second, Findley and Teo do not examine the domestic sources of the intervention decision adequately. Even though they view intervention as a foreign policy tool, they do not analyze the impact of critical domestic factors on foreign policy decision making. Finally, this is one of the few studies that distinguish between interventions for or against the government while examining the causes of intervention. While this is an important contribution of this study, the authors only make a theoretical distinction between pro-government and pro-rebel interventions in the context of alliance and rivalry relations. Unfortunately, they do not theorize about how other factors included in their analysis affect interveners? decision on which side to support. 54 Another scholar who combines actor-oriented and conflict-oriented approaches is Kathman (2010, 2011). In two different recent studies, Kathman examines the relationship between geography and potential interveners? decision making. His first study examines the causes of biased military interventions in civil conflicts by neighboring countries. According to Kathman, it is important to study the causes of these interventions separately because neighbors are responsible for a disproportionately large number of biased interventions in civil conflicts. He argues that neighbors are mostly concerned about the contagion effects of civil conflicts as a result of which the probability of intervention by contiguous states increases as their ?risk of hostility infection? from the civil conflict increases (Kathman 2010, p. 996). To test this hypothesis, Kathman first measures the infection risk score of each neighboring country. The infection risk score is made up of two major components. The first component measures the intensity of the civil conflict while the second component examines several factors (i.e. regime type, ethnic heterogeneity, GDP per capita) that predict domestic instability within the neighboring country (Kathman 2010, p. 996). After combining these two major components, Kathman predicts an infection risk score for each neighboring country and includes this as the main explanatory variable in his analysis. He also includes alliance ties and capability ratio between the civil conflict state and neighbors in his statistical analysis as well as various attributes of the conflict similar to the empirical analyses of Lemke and Regan (2004) and Findley and Teo (2006). The results indicate that infection risk score has the largest impact on the probability of military intervention by neighbors. 55 Developing an infection risk score is certainly a useful way to evaluate the possibility of contagion for each neighboring state. However, Kathman does not provide a satisfactory theoretical explanation for why higher infection scores lead to biased military interventions. Neighboring countries which are vulnerable due to high infection scores may prefer neutral interventions such as mediation or other conflict resolution efforts without getting involved militarily. Even if neighbors do intervene militarily, infection risk score cannot explain on whose side (government or opposition) they choose to intervene. Kathman?s study is successful in putting the theoretical emphasis on potential interveners, which are neighbors in this particular study; but he fails to explain theoretically how infection risk score translates into pro- government or pro-rebel military interventions. Kathman?s second study examines the causes of military interventions by all potential interveners, not just neighbors, using a similar but much broader infection risk analysis (Kathman 2011). Like his previous study, Kathman develops an infection risk score for each state neighboring the civil conflict state. However, the purpose of this particular study is to understand how regional concerns of potential interveners influence their probability of intervention in a civil conflict; so he uses infection risk scores to calculate contagion risk scores for the entire region surrounding the civil conflict. After that, Kathman measures the value of a particular region for each potential intervener. In other words, not all regions have the same value for each potential intervener; therefore, the probability of intervention depends not only on the regional contagion risk but also on the value of the region. The value of a region is measured by potential intervener?s proximity to the region as well as its 56 alliance and trade ties with the states located in that region. Kathman combines regional contagion scores with three regional value scores (regional alliance score, regional trade score, regional proximity score) for each potential intervener and constructs three different independent variables. Similar to his first study on neighboring interventions, Kathman finds that these new composite variables measuring regional concerns of potential interveners have the most substantial effect on increasing the probability of military intervention (Kathman 2011). Kathman?s second study is even more original than his first one. It is probably the only quantitative study that includes regional motivations of external states in a statistical analysis by using a sophisticated method to measure regional scores for each potential intervener. However, these regional scores still cannot explain how potential interveners choose sides or why they prefer biased interventions to neutral ones. In fact, the existence of regional concerns can make it more likely to resort to neutral conflict resolution efforts such as regional or international peacekeeping operations. Kathman?s study fails to address these theoretical concerns. Finally, both of his studies do not take into account other critical strategic issues within the civil conflict environment that affect potential intervener?s decision making. Even though Kathman is correct in arguing that dyadic relations between external states and civil conflict states can be inadequate for understanding regional concerns of potential interveners; he fails to include critical strategic factors, such as rivalry, that might be more useful for understanding why states resort to biased military interventions in the first place. In other words, the additional explanatory power gained by regional variables is questionable. 57 By combining actor-oriented and conflict-oriented approaches in a way that puts the potential intervener at the center of theoretical and methodological analysis, Kathman, Findley and Teo make significant contributions to the literature. These recent studies demonstrate that the most promising way to study the rationale behind military interventions is by making potential interveners the primary focus of study while paying due attention to the civil conflict phenomenon they are reacting to. Despite the contributions made by these three recent studies, there are still a lot of theoretical and methodological improvements to make within the quantitative literature. The scientific and systematic study of interventions calls for more progress in order to understand the causes of foreign military interventions. Where is Progress Needed? In 1969, Rosenau argued that ?the factors that foster, precipitate, sustain, channel, constrain, and curb intervention? had not been explored scientifically (1969, p. 150). Due to an increase in the number of quantitative studies on interventions, the scientific and systematic study of interventions has particularly improved in the past two decades. Scholars analyzing the causes and consequences of intervention have made more progress particularly with respect to the latter. There are several key reasons why less progress is made towards understanding the causes of foreign military interventions. First and foremost, the number of quantitative studies analyzing the causes of military interventions is smaller compared to the amount of research on the consequences of interventions. The dearth of quantitative studies analyzing the causes of military interventions is quite troubling, especially because understanding the 58 causes of intervention is important for studying its consequences as exemplified by disagreements on interveners? goals. More quantitative research is simply needed to discover empirical patterns regarding the rationale behind military interventions. Second, the two main approaches used to analyze the causes of interventions need to be synthesized to provide a better theoretical understanding of this phenomenon. As Rosenau said, theory-building is critical for systematic and scientific study of interventions. However, one should be careful while integrating these two approaches into one theoretical framework. The potential intervener should still be theoretically and methodologically central as in the case of actor-oriented studies, but the capacity of the theoretical framework to analyze intervention decision should be enhanced by including the critical contextual factors related to the civil conflict. A handful of recent studies have demonstrated that synthesis is possible and offers promising results, but their shortcomings also indicate that more work is still needed on that front. Third, it is problematic that studies analyzing the causes of biased interventions tend to bypass the question of how interveners choose sides. On whose side external states intervene is closely related to why they intervene. Only a few studies have looked at this relationship, but not in a systematic or consistent way. At the very least, it is an empirical question that needs to be analyzed to understand the rationale behind military interventions. The lack of systematic research by intervention scholars on this issue hampers theoretical and empirical progress as well. In the following chapters, this study aims to address the theoretical and empirical shortcomings discussed throughout the literature review and summarized briefly above. 59 Chapter 3: Theory It is the purpose of this chapter to present a unified theoretical framework that combines actor-oriented and conflict-oriented approaches. A synthesis of these two dominant approaches used in the literature offers a promising way to make theoretical and empirical progress towards an improved understanding of the rationale behind foreign military interventions in civil conflicts. However, the synthesis should be carried out in a certain way. Since military intervention is a foreign policy instrument, this unified theoretical framework should be able to conduct a foreign policy analysis which requires looking at the international and domestic sources of foreign policy decision making. This would enable the analyst to examine the critical international and domestic factors that produce the intervention decision and determine the factors that motivate state leaders to intervene militarily on government or opposition side in some civil conflicts but not in others. This unified model should also pay attention to the civil conflict phenomenon foreign policy decision makers are reacting to. The attributes of the civil conflict state and the contextual factors related to the changing civil conflict environment should be incorporated into the theoretical framework as they relate to the decision making calculus of the intervener. ?Modified Realist Framework? for Foreign Policy Analysis Any foreign policy analysis should start with identifying the main concerns of state leaders who make these foreign policy decisions. The underlying assumption of this study is that state leaders are rational actors and their primary concern is to stay 60 in power. Therefore, state leaders want to pursue foreign policies which benefit their own position in the domestic setting as well as their countries? position in the international realm. As Putnam (1988) argues, leaders are concerned with domestic and international pressures simultaneously while making foreign policy decisions. Although leaders want to pursue foreign policies which are advantageous to their country?s international standing, they also want to avoid foreign policies which would threaten their own domestic standing since their primary objective is to stay in power. As rational actors who compare the expected utility of different foreign policy options before choosing one, state leaders consider the costs and benefits of any foreign policy decision in both domestic and international terms. In other words, both domestic and international conditions affect the cost-benefit analysis of state leaders with respect to different foreign policy options. That is why scholars studying foreign policy decisions advocate examining both international and domestic sources of foreign policy behavior (i.e. Putnam 1988; Fearon 1994; Kapstein 1995; Stam 1996). Given that state leaders are rational actors who want to stay in power and their foreign policy behavior originates from domestic and international sources; the theoretical task is to examine the critical international and domestic factors affecting cost-benefit calculation of leaders and identify which ones lead to positive expected utility from military intervention and cause leaders to choose this course of action. In order to accomplish this task, this dissertation employs a theoretical framework called ?modified realism? developed by Huth (1998) because it is consistent with the theoretical assumptions of this study regarding foreign policy behavior of states. 61 As rational actors, state leaders want to pursue foreign policies that are advantageous to their countries? political, military and economic standing in the international system. According to the realist school of thought which interprets national interest in terms of security and power, foreign policy decisions are shaped by international security and power considerations of states. This study posits that realism is particularly relevant for analyzing foreign policy decisions involving the use of military force. In fact, many studies discussed in the literature review chapter have used a realist framework while explaining intervention decisions (i.e. Feste 1992, Scott 1996, Findley and Teo 2006). Although this study finds realism useful for analyzing interventions, it also recognizes the theoretical weakness of realism in accounting for domestic sources of foreign policy behavior. Huth?s ?modified realism? attempts to address this weakness and offers a promising way to incorporate domestic politics into realism (1998). In Huth?s modified realist framework, state leaders are ?concerned with promoting the international security of their country? but as rational actors they are also ?careful to pursue security policies that do not undermine their domestic political position? (1998, p. 746). Huth argues that realism ?can be strengthened by theorizing about how domestic political concerns of state leaders may have systematic and consequential effects on their foreign policy decisions? (1998, p.745) and he outlines a modified realist framework in which ?both domestic political concerns and international power considerations determine foreign policy choices? (1998, p. 746). According to Huth?s modified realist framework, domestic politics can be connected to international security concerns of leaders to explain their foreign policy 62 decisions in three specific ways. First, leaders need to ?balance the pursuit of security interests abroad with the domestic politics of building ? political support behind? foreign policy decisions (1998, p. 746). In other words; while deciding among foreign policy options, leaders have to consider which of these options would receive support from important domestic constituents and also enable them to confront domestic opposition effectively. Second, ?a leader?s desire to hold onto power can create incentives to pursue foreign policies that the leader believes will enhance his domestic political position? (1998, p. 746). Third, ?the simultaneous pursuit by state leaders of domestic political and international security may create potential trade-offs between the resources available ? to support both sets of goals? and may constrain leaders? foreign policy options because of the ?high priority attached by leaders to domestic political survival? (1998, p. 747). By specifying these three systematic linkages between international and domestic considerations of leaders, Huth?s modified realist framework describes how domestic politics can affect the cost- benefit analysis of leaders with respect to different foreign policy options. Compared to traditional realism, modified realism is much more capable of accomplishing the task of identifying critical factors that affect leaders? cost-benefit analysis in both domestic and international settings. As Huth puts it, ?what seems like a rational decision? in international politics from a realist perspective may not be domestically feasible from a modified realist perspective (1998, p. 745). The systematic linkages between domestic and international politics described by modified realism make it a more suitable theoretical approach for foreign policy analysis. Therefore, this dissertation uses modified realism to hypothesize about the 63 critical domestic and international factors that motivate leaders to intervene militarily in civil conflicts and relies on modified realist arguments about the linkages between domestic and international concerns of leaders to explain the hypothesized relationship. At this point, the study is equipped with a theoretical framework which can account for domestic and international sources of foreign policy behavior. However, before hypothesizing about the international and domestic factors that affect leaders? decision on whether and on whose side to intervene in civil conflicts, one must also consider the role of civil conflicts in international politics. What makes civil conflicts significant in international relations and how do foreign leaders perceive civil conflicts in other states? Why are Civil Conflicts Significant to Other States? Civil conflicts, especially when they escalate to civil wars, can have devastating human costs. The tragic humanitarian consequences of civil conflicts make them dire events for the international community. However, the significance of civil conflicts for state leaders elsewhere comes from their potential impact on international politics. According to Rosenau, foreign leaders are ?constantly alert to any sudden changes that may alter the personnel and orientations of foreign governments? because ?changes in authority structures? can create ?radical transformations in the international system? (1969, p. 168). Due to their potential to affect international security and power considerations of other states, civil conflicts are significant events for foreign leaders. 64 As Rosenau argues, ?top officials everywhere? are ?sensitive to the stability of foreign governments? and consequently pay attention to civil conflicts in the world (1969, pp. 168-169), but that does not mean that civil conflicts influence international security interests of all states similarly. For some external states, a civil conflict might pose a direct or indirect threat to their security interests while for others it might offer an opportunity to improve their position in the international system. In other words, civil conflicts can affect security interests and/or relative power calculations of external states in different ways by creating threats and opportunities due to their direct and indirect repercussions on international politics. Although all foreign leaders pay attention to the civil conflicts in the international system, only some of them choose to devise a foreign policy response to a particular civil conflict. Since foreign policy behavior has international and domestic sources, foreign leaders who decide to intervene militarily on the government or rebel side may perceive the threats or opportunities posed by the civil conflict not only in international terms but also in domestic terms. Thus, modified realism is a more suitable theoretical framework to examine the intervention decision of leaders because realism only explains how leaders perceive the threats and opportunities posed by the civil conflict with respect to international security and power interests. Scholars using the actor-oriented approach as well as the combination approach (i.e. Findley and Teo 2006) have derived interveners? interests from realism which interprets costs and benefits only in international terms. Modified realism can also account for leaders? perception of the civil conflict in terms of their own domestic political standing. 65 As mentioned before, this study argues that combining actor-oriented and conflict-oriented approaches provides an improved understanding of the rationale behind military interventions in civil conflicts. Similar to actor-oriented studies, this study will focus on the potential intervener and it will hypothesize about the critical international and domestic factors that motivate some states to intervene on the government or rebel side in a civil conflict. Then, it will hypothesize about the key attributes of the civil conflict which also affect foreign leaders? decision making. All of these hypotheses will be derived from a modified realist framework which adapts well to a combination approach that makes potential intervener and its decision making theoretically central. International Sources of Military Intervention This section examines the critical international factors that are likely to affect leaders? decision making regarding whether and on whose side to intervene militarily in a civil conflict. Although all of these factors originate from international sources such as dyadic relations with the conflict state or relations with other interveners involved in the civil conflict; hypothesized relationships derived from modified realism will facilitate a discussion of how these international factors affect the cost- benefit analysis of leaders in terms of their international and domestic concerns. Since this dissertation also argues that on whose side external states intervene is closely related to why they intervene in a civil conflict, hypotheses will also address this relationship. 66 Strategic Significance of the Conflict State Foreign leaders are particularly attentive to civil conflicts unfolding in countries that have strategic significance for their own country, because the ?strategic value? of the conflict state is ?critical in determining whether important security interests are at stake? (Huth 1998, p. 750). In other words, leaders? perception of the threats and opportunities created by the civil conflict situation is closely related to the strategic significance of the conflict state. Political, military and economic relations with the conflict state and the existence of perceived strategic interests in that country determine the strategic value of the conflict state for leaders. The strategic value of the conflict state is crucial for leaders? cost-benefit analysis regarding whether and on whose side to intervene in the civil conflict. Since leaders want to pursue foreign policies which are advantageous for their own position in the domestic setting as well as their countries? position in the international realm, military intervention can have important international and domestic benefits when the conflict state is strategically significant. The strategic value of the conflict state also makes it easier for leaders to justify the domestic and international costs of military intervention. As Huth argues, leaders ?should be in a stronger domestic political position to build a coalition in support of? military intervention when important domestic and international security interests are at stake (Huth 1998, p. 750). Military interventions to reduce threats or to seize opportunities from unstable civil conflict situations are more likely to receive domestic support when expected domestic and international benefits are high due to the strategic significance of the conflict state. 67 Having domestic support for military intervention decisions, especially from key constituencies, is important regardless of the regime type of the state because even in the most dictatorial regimes leaders do not rule alone. As a result of competitive elections, political leaders? survival depends on the majority of the selectorate in democracies. An autocratic leader?s political survival, on the other hand, depends on the support of a smaller portion of society which generally consists of political and military elites. Compared to democracies, autocracies are considered to have a much smaller winning coalition whose support the leader has to obtain to remain in office (Bueno de Mesquita et al. 2003). Thus, despite the difference in the size of their winning coalitions, both democratic and autocratic leaders need domestic approval for their foreign policy decisions. Since both democratic and autocratic leaders pay attention to the strategic value of the conflict state when deciding whether and on whose side to intervene, international factors that determine strategic significance of the conflict state for foreign leaders will be discussed one by one. Rivalry with Conflict State: States engaged in interstate rivalry with the conflict country have a particular interest in manipulating the civil conflict to their advantage. Salehyan argues that external states may even instigate rebellion and instability in a rival state to undermine governments (2007). Hence, a civil conflict that is already unfolding in a rival state creates an opportunity for leaders to further weaken or even replace the government. 68 States want to enhance their own position in the international system especially with respect to their rivals (Snidal 1993). When a state is experiencing a civil conflict, its external rival is likely to support rebels because pro-rebel military intervention offers an opportunity to further destabilize or remove a rival government and to change regional or international power balance in its favor. Given the potential security benefits of military intervention, it is easier for leaders to justify the costs and to receive domestic support for their foreign policy decision. Both the public and the elites tend to have a negative image of the rival state and should be more likely to support their leaders in order to receive the potential domestic and international security benefits of military intervention. In fact, for external states interested in destabilizing or removing a rival government, pro-rebel military intervention might even be a less costly foreign policy option than interstate war. Since rivalry affects leaders? decision on both whether and on whose side to intervene, two hypotheses are formulated. H1a: The probability of military intervention by an external state increases when the conflict state is a rival. H1b: An external state is more likely to intervene on the side of the rebels when the conflict state is a rival. Involvement in a Militarized Interstate Dispute with Conflict State: Rivalry partially accounts for hostile relations between states in the international system. States can have enmity towards each other even though their relationship does not meet the criteria of interstate rivalry. If states are engaged in military interstate disputes, they also look for opportunities to weaken each other in 69 order to protect their own security interests. Therefore, leaders are alert to civil conflicts in states with which they have military disputes. An external state is more likely to provide military support to rebels if it is involved in a military dispute with the government of the conflict state. Similar to the rivalry situation described above, leaders are more likely to find domestic support for their foreign policy decision to intervene because of the potential international and domestic security benefits of weakening or removing a hostile government. Hence, the following hypotheses are proposed. H2a: The probability of military intervention by an external state increases if it is involved in a militarized interstate dispute with the conflict state. H2b: An external state is more likely to intervene on the side of the rebels when it is involved in a militarized interstate dispute with the conflict state. Alliance Ties with Conflict State: Civil conflicts in countries with friendly governments can threaten security interests of external states as well, especially when there are alliance ties. Military intervention becomes a plausible option for leaders who want to prevent the removal of an ally and to protect their diplomatic and military interests in the target state that can come with formal alliance ties. When civil conflicts threaten the stability and/or survival of allied governments, external states may also be concerned about international audience costs of non-intervention. Since allies and adversaries are important audiences due to credibility concerns; by intervening, an external state may want to signal to its allies that it will make good on its promises to defend them. 70 Leaders can justify a pro-government military intervention from both international and domestic politics perspective. Defense of allies, prevention of adverse strategic changes due to overthrow of allied governments, protection of diplomatic interests and credibility concerns are some of the strategic considerations that help leaders present a strong case to the public to justify the need for military intervention. H3a: The probability of military intervention by an external state increases when there are alliance ties with the conflict state. H3b: An external state is more likely to intervene on the side of the government when there are alliance ties with the conflict state. Historical Ties with Conflict State: Historical ties between countries are also critical when leaders are deciding whether and on whose side to intervene in a civil conflict. Former colonial powers interact frequently with their former colonies and continue to maintain political and economic interests in those countries. Colonial ties are often included in international studies to capture the political relations between certain pairs of states; however they are not the only type of historical relationship states can have with each other. Disintegration of states due to secession also causes newly formed states to have historical ties with each other that shape their political, economic and societal relations. Thus, foreign leaders may attach strategic significance to a conflict country due to their continued political and social interactions with the government, elites or communal groups in the target state which can originate from different types of historical ties. These can be used as credible arguments by leaders to build domestic 71 support behind military intervention decisions, especially given that some segments of public might be willing to help societies they have historical ties with. When historical ties exist, leaders are more likely to support the government side because they would prefer maintaining the historically hard-won relations with the target government instead of jeopardizing the status quo by supporting the rebels. Since the public is also more likely to favor supporting the government and protecting the current status quo, leaders are more likely to intervene militarily to help the government. H4a: The probability of military intervention by an external state increases when there are historical ties with the conflict state. H4b: An external state is more likely to intervene on the side of the government when there are historical ties with the conflict state. Geographic Proximity to Conflict State: As discussed in the literature review, empirical evidence shows that states intervene frequently in their neighbors? civil conflicts and there are numerous reasons for neighboring interventions that have been discussed by scholars. Essentially, all neighboring countries are strategically significant for leaders because of their proximity. The potential for conflict spillover across territorial borders, threat to stability in border regions and the possibility of refugees create serious problems for leaders who are concerned about domestic and international security of their own countries. Dangers of conflict spillover can motivate leaders to intervene militarily in neighboring civil conflicts and leaders are likely to receive domestic support for their decision because of legitimate threats to national security. 72 Although contiguity increases the probability of military intervention, its effect on which side leaders choose to support in the civil conflict is not so straightforward. As conflict scholars have argued, neighbors frequently interact with each other but greater interaction between contiguous states also increases the likelihood of interstate conflict (Bremer 1992; Diehl 1985; Gleditsch and Singer 1975). In other words, neighbors can have friendly or hostile relations with each other due to their frequent interactions. Therefore, leaders? decision regarding which side to support in the civil conflict depends on whether they have friendly or hostile relations with the neighboring government. For instance, if the neighboring state is an interstate rival, then leaders are more likely to intervene on the side of rebels. Alliance ties or strong trade relations with the neighboring state, on the other hand, would motivate leaders to intervene on the side of government. Consequently, only one hypothesis is proposed in the case of contiguous states. H5: The probability of military intervention by an external state increases if it is contiguous to the conflict state. Proximity is another geographic factor that increases the strategic value of the conflict state for foreign leaders. States that are in the same geographic region with the conflict state are generally concerned about regional spillover effects of the civil conflict as Kathman argues (2011). Regional instability caused by the civil conflict and potential future changes in the regional balance of power due to change of governments can also affect national security and relative power concerns of external states. In short, being in the same region with the conflict state also increases an external state?s likelihood of military intervention. Leaders can justify the costs of 73 military intervention to domestic audience based on these regional security and power interests. Similar to contiguity, the impact of geographic proximity on leaders? decision regarding which side to support in the civil conflict is not so clear-cut. Leaders can intervene on the side of government if they have friendly relations or alliance ties with the conflict state. They can also intervene on the government side if their goal is to prevent the rise of a new radical government that can disrupt the regional status quo. On the other hand, civil conflicts nearby also offer an opportunity to leaders who want to weaken or remove a hostile regime. In such cases, leaders would be more likely to side with the rebels. Thus, only one hypothesis is formulated. H6: The probability of military intervention by an external state increases if it is in the same region with the conflict state. Resource Wealth of Conflict State: Foreign leaders may attach strategic significance to a conflict country due to its resource wealth. Qualitative studies have shown that resource wealth of a conflict country can attract third party military intervention because foreign powers are more likely to intervene in order to gain access to the natural resources. Resource looting in the conflict country can also decrease the costs of military intervention (Dashwood 2000; Ross 2004a, 2004b). For instance, Ross argues that Liberia intervened on the side of the rebels during Sierra Leone?s civil war to gain access to the diamond fields (Ross 2004a). Although there is some empirical evidence from qualitative studies, the relationship between resource wealth and military intervention has not been tested adequately in large-N studies. 74 One can argue that since resource wealth of a conflict country increases the tangible benefits of military intervention, external states who want to exploit these natural resources are more likely to intervene in the civil conflict and leaders can justify the domestic costs of intervention with those tangible benefits. However, resource wealth of a country can be lootable or non-lootable depending on the type of natural resources found in the country. Lootable resources, such as secondary diamonds, can easily be exploited or smuggled. Non-lootable resources, on the other hand, are very difficult to exploit because they require expensive equipment, financial investment, long-term access to a conflict state?s territory in order to receive the tangible benefits (Lujala et al. 2005). Thus, the type of natural resources in a conflict country should factor into the cost-benefit analysis of external states who want to benefit from these resources as a result of military intervention. If an external state?s goal is to make a profit from lootable resources such as secondary diamonds, then it is more likely to intervene on the side of the rebels as in the case of Liberia. During the military intervention, interveners? forces can loot and smuggle such resources easily. If an external state?s goal is to make a profit from non- lootable resources such as oil and gas, then it is more likely to intervene on the government side because governments tend to have control over non-lootable resources. The probability of receiving economic benefits from non-lootable resources both during and after the civil conflict is higher when the external state intervenes militarily on the side of the government. To sum up, even though the resource wealth of the conflict country might motivate external states to intervene, on whose side they intervene depends on the type of natural resources in the country. 75 H7a: The probability of military intervention by an external state increases if the conflict state has lootable resources. H7b: An external state is more likely to intervene on the side of the rebels when the conflict state has lootable resources. H8a: The probability of military intervention by an external state increases if the conflict state has non-lootable resources. H8b: An external state is more likely to intervene on the side of the government when the conflict state has non-lootable resources. Trade Ties with Conflict State: Strong trade ties with the conflict state also increase its strategic value for foreign leaders. Trade is a good indicator of the extent of economic interdependence between pairs of states. Moreover, economic interdependence is positively associated with compatibility and peaceful relations between states in the conflict literature (Gleditsch 2007). Therefore, external states which have strong trade relations with the conflict state would be more likely to intervene on the side of government in order to protect their economic interests and trade relations. To receive domestic approval for their decision, leaders can build a domestic coalition from the key constituent groups whose economic interests are likely to be affected due to ongoing instability or potential for regime change in the conflict state. H9a: The probability of military intervention by an external state increases as trade ties with the conflict state increase. H9b: An external state is more likely to intervene on the side of the government as trade ties with the conflict state increase. 76 Strategic Significance of the Civil Conflict Other international factors besides the strategic value of the conflict country can motivate foreign leaders to intervene in civil conflicts as well. Leaders? perception of the threats and opportunities created by the civil conflict may not necessarily be a function of the strategic value of the conflict state but it may also be related to the strategic significance of the civil conflict environment. Foreign powers sometimes become militarily involved in a civil conflict in response to another state?s military intervention. In other words, military intervention can be used as a strategic foreign policy instrument to destroy, balance or enhance another state?s influence in the civil conflict environment. In fact, civil conflicts are often multiparty environments. Empirical evidence shows that more than half of the civil conflicts in the post-WWII period involved multiple interveners (Findley and Teo 2006, p. 829). Therefore, it is important to understand how previous interventions by other states affect leaders? perception of the strategic significance of the civil conflict and factor into the cost-benefit calculation of leaders when deciding whether and on whose side to intervene. Some scholars argue that during the Cold War, the U.S. sometimes intervened in civil conflicts as a reaction to the Soviet intervention rather than due to the strategic value of the conflict country. For instance, according to Scott, the U.S. had ?little or no strategic? interest in Afghanistan until the late 1970s, but the Soviet intervention in 1978 ?triggered a change in the U.S. assessment? (Scott 1996, p. 40- 41). President Carter commented that the Soviet intervention ?placed the Soviet Union within aircraft striking range of the vital oil resources of the Persian Gulf; it 77 threatened a strategically located country, Pakistan; and it posed the prospect of increased Soviet pressure on Iran and on other nations of the Middle East? (1996, p. 43). He warned that ?an attempt by any outside force to gain control of the Persian Gulf region will be regarded as an assault on the vital interests of the United States of America? (1996, p. 43). Hence, the U.S. decision to intervene militarily on the side of the Afghan rebels was purely a reaction to the Soviet intervention which had made the conflict in Afghanistan strategically significant for the United States. Reactionary military interventions are neither unique to superpowers nor to the bipolar international system of the Cold War. As mentioned above, empirical evidence shows that many civil conflicts experience multiple interventions. States generally observe each other?s behavior and may resort to reactionary interventions to strengthen or reduce each other?s influence on civil conflict dynamics or outcomes. In fact, one of the early empirical studies has noted that if one domestic disputant is supported by an outside state, the other side tends to receive external military support as well (Duner 1983). Despite this early observation of the interdependent aspect of interventions, there have actually been few attempts to model the strategic context of civil conflicts. Findley and Teo (2006) show that states balance the role of their rivals in civil wars by intervening on the opposite side, but alliances do not always lead to interventions on the same side. Gent (2007) suggests that major powers usually undertake opposing interventions in civil wars in order to affect the course of events in their favor. To sum up, foreign leaders who attach strategic value to a conflict country are aware of the presence and interests of other external states involved in the civil 78 conflict and factor those into their decision making. However, a civil conflict can also gain strategic significance because of its potential to affect international security interests of an external state due to other countries? interventions in the civil conflict. Under such circumstances, an external state?s decision on whether and on whose side to intervene depends on the relations it has with other interveners in the civil conflict. Specifically, leaders? perception of whether they have divergent or convergent strategic interests with the other interveners is likely to affect their decision making most. Rivalry with Previous Intervener: Rivals closely observe each other in the international system and are often suspicious of the other?s intentions and actions. Since rivals generally perceive their interests as opposing, they try to reduce each other?s influence on international political events and shape them in line with their own strategic preferences. Civil conflicts are one of those international events where rivals might engage in opposing interventions. Goertz and Diehl (1995) argue that rivals continue a conflictual relationship over a long period of time without a clear-cut victory or defeat. Therefore, incremental changes in power can change the stalemate between rivals and create opportunities to supersede the other. Intervention in a civil conflict can be an attempt by rivals to expand their influence because it might provide them with an advantage to disrupt the status quo in the future. Thus, it can be argued that states are more likely to intervene in civil conflicts in which their rival has previously intervened and back opposing sides to balance their rival?s impact. 79 When leaders perceive a conflict state to be strategically significant, military intervention by a rival state poses a direct threat to their domestic and international interests. The logical response would be to intervene on the opposing side to destroy their rival?s influence on the civil conflict environment and try to prevent an outcome that would risk their national security or damage their strategic interests in the conflict state. As demonstrated in the case of Afghanistan, states can also intervene militarily in reaction to a rival?s intervention. In such cases, states want to shape the conflict dynamics in their own favor and prevent their rival from achieving its preferred outcome which might potentially threaten their security and have adverse effects on the status quo. Military intervention can also be used strategically to weaken a rival state indirectly through the destruction of the domestic opponent supported by the rival intervener in the civil conflict. In short, strategic considerations might necessitate siding with whichever domestic party best supports the potential intervener?s divergent interests with its rival. Two hypotheses are proposed. H10a: The probability of military intervention by an external state increases if its rival has already intervened in the civil conflict. H10b: An external state is more likely to intervene on the opposite side that its rival intervened on. Alliance with Previous Intervener: Civil conflicts can become strategically significant for external states not only based on divergent interests, but also due to the convergent interests they hold with previous interveners. While two states with divergent interests want to influence the conflict outcome in opposing ways and try to mute each other?s influence; states with 80 convergent interests may want to cooperate and increase the probability of securing a preferred outcome that serves their common interests. In some ways, this is similar to the idea of bandwagoning in the international relations literature. States are most likely to hold convergent interests with their allies in the international system. Hence, military interventions by allies can affect leaders? decision making on whether and on whose side to intervene in a civil conflict. Since alliances reflect common interests as well as responsibilities to protect each other against threats, foreign leaders may intervene in a civil conflict on the same side that their ally intervened on. By providing military support to the same domestic party, an external state may want to multiply the impact of its ally?s efforts and achieve a mutually beneficial outcome. However, an external state might also intervene militarily to protect an ally whose security is threatened by a civil conflict nearby. Kathman (2011) claims that third parties can intervene in a civil conflict to protect their allies from the contagion effect of the civil conflict. This study?s argument is slightly different than Kathman?s claim because it expects states to intervene in a civil conflict only if their ally has already intervened on one side to protect their national security and strategic interests. In other words, external states intervene to provide additional military support to the domestic side their ally has already intervened on. Thus, external states can intervene alongside their allies in a civil conflict either to protect their allies and themselves from direct and indirect threats, respectively, or to seize advantages from civil conflicts that their ally has already become involved in by supporting their efforts. The following hypotheses are formulated: 81 H11a: The probability of military intervention by an external state increases if its ally has already intervened in the civil conflict. H11b: An external state is more likely to intervene on the same side its ally intervened on. Control Variables Strategic value of the conflict state and the strategic significance of the civil conflict are major factors affecting leaders? cost-benefit analysis regarding whether and on whose side to intervene in a civil conflict. However, there are several other international factors that are likely to influence state leaders? decision making. Involvement in an Interstate War: When a state is already involved in an international conflict, it has less military capabilities and resources to allocate to other conflicts. Military involvement in additional conflicts becomes more costly and risky under such situations. Domestic and international security concerns of leaders force them to refrain from military interventions in civil conflicts. H12: The probability of military intervention by an external state decreases if it is involved in an interstate war. Power Status: The power status of a state influences its ability and willingness to intervene. Major powers have a higher probability of intervening in civil conflicts because they have more military resources to commit and they can extend their military influence globally (Pearson, Baumann, and Pickering 1994). Major powers are also expected to 82 be more willing to intervene in civil conflicts because of their vested interests in different parts of the world. H13: The probability of military intervention by an external state increases if it is a major power. Relative Capabilities: Not only the power status of potential interveners, but also the power differentials between states can affect the likelihood of military intervention. Realists argue that the ratio of capabilities factor into states? decision making calculus when using military force. Moreover, states can also attempt to increase the probability of achieving their preferred outcome by employing force mainly against weaker conflict states. Leaders may also avoid interventions in stronger conflict states due to military setbacks because that might hurt their own domestic position. Therefore, it can be expected that leaders are less likely to undertake military interventions as relative capabilities start to favor the conflict state. H14: The probability of military intervention by an external state increases as the ratio of capabilities with the conflict state increases. International System: The structure of the international system can also affect the probability of military interventions. Starting with the earliest works in the intervention literature, many studies have argued that interventions in civil conflicts should be more frequent in bipolar systems than in balance-of-power systems because bipolarity creates more incentives for intervention (i.e. Kaplan 1964, Rosenau 1969, Regan 1998, Kathman 2011). Given the intense global competition between the two blocs during the Cold 83 War, both superpowers and their allies intervened in civil conflicts to keep the adversary from gaining additional influence. The zero-sum environment of the Cold War increased incentives to intervene for strategic purposes. With the end of the Cold War and bipolar competition, some of these geostrategic incentives disappeared. Therefore in the post-Cold War international setting, the probability of military intervention in civil conflicts is expected to be lower. H15: The probability of military intervention by an external state is higher during the Cold War compared to the post-Cold War period. Domestic Sources of Military Intervention This section examines the critical domestic factors that are likely to affect leaders? decision making regarding whether and on whose side to intervene militarily in a civil conflict. Although all of these factors originate from domestic sources, modified realism facilitates an analysis of how these domestic factors affect the cost- benefit analysis of leaders by articulating the linkages between international and domestic concerns. Regime Type: State leaders want to pursue foreign policies which are advantageous to their country?s international standing but also want to avoid those which would threaten their domestic standing since their primary objective is to stay in power. As argued before, having domestic support for military intervention decisions is important regardless of the regime type. However, institutional differences in democracies and autocracies determine the size of the winning coalition whose support leaders need to remain in office. As a result of competitive elections, political leaders? survival 84 depends on a large segment of the population in democratic states. An autocratic leader?s political survival, on the other hand, depends on the support of a smaller portion of society which generally consists of political and military elites. Thus, autocracies have a much smaller winning coalition compared to democracies (Bueno de Mesquita et al. 2003). Democratic leaders aim to pursue foreign policies which are supported by a large number of citizens because they rely on re-election to remain in power. Democratic leaders are also more concerned about the success of their policies since they can lose their jobs in the case of policy failure. Hence, military intervention in civil conflicts by democracies is more likely when it is consistent with the preferences of a significant portion of the public and its outcome is likely to satisfy them. Due to the larger size of their winning coalition compared to autocracies, democracies are likely to be more selective when deciding to intervene. Autocracies, on the other hand, are less likely to be selective when intervening in civil conflicts because political survival of autocratic leaders depends on a much smaller segment of public. Autocratic leaders can remain in power as long as they have the support of their small winning coalition which they typically maintain by providing private goods. Even after a policy failure, autocratic leaders can satisfy the members of their winning coalition with private goods to retain their domestic position. Therefore, the following hypothesis is proposed: H16a: Democracies are less likely to intervene militarily in civil conflicts than autocracies. 85 Although democratic leaders are more likely to be selective when they intervene in civil conflicts, it is important to analyze whether democracies have a stronger tendency to support the government side over the rebels when they decide to intervene. Some of the earlier studies suggest that democracies may be more interested in preserving governments rather than challenging them (Pearson 1974a, 1974b). Hence, the following hypothesis will also be tested: H16b: Democracies are more likely to intervene on the government side than are autocracies. Regime Similarity: Regime similarity can also affect leaders? foreign policy decision making. Scholars have argued that democracies form a security community and positively identify their national security with each other (i.e. Doyle 1983a, 1983b, 1986; Russett 1993; Owen 1994). In fact, Hermann and Kegley claim that democracies used interventions in interstate and intrastate conflicts ?primarily for defensive purposes to ensure that countries did not slip from the democratic camp? (Hermann and Kegley 2001, p. 242). Lemke and Regan (2004), on the other hand, find that jointly democratic dyads are no more or less likely than other dyads to experience interventions during civil conflicts. Although there are mixed findings regarding intervention behavior of democracies towards their counterparts in civil conflicts; empirical studies show that while democracies do not feel compelled to militarily assist other democracies in interstate disputes, they almost never enter the conflict on opposing sides (i.e. Maoz and Russett 1993; Oneal and Russett 1997; Dafoe et al. 2013). 86 The impact of regime similarity on states? foreign policy behavior can be noticed not only in democracies, but also in autocracies. Huth argues that leaders of states with similar regimes ?are less likely to view one another as security threats because they share common interests in preserving political stability at home that, in turn, should lead these states to adopt less conflictual foreign policies toward one another. The legitimacy and survival of regimes at home can be enhanced by political allies abroad supporting one another? (Huth 1998, p. 751). Due to their shared interest in protecting domestic norms of governance, leaders can intervene militarily on the government side in a civil conflict to help a counterpart keep its regime intact. If the conflict state has a different political system, then leaders would be more likely to intervene on the side of rebels to undermine a government with a different regime type. Two hypotheses are proposed. H17a: The probability of military intervention by an external state increases if it has the same regime type as the conflict state. H17b: An external state is more likely to intervene on the government side if it has the same regime type as the conflict state. Ethnic Ties with Conflict State: As discussed in the literature review, scholars have argued that ethnic ties with a conflict state can also motivate leaders to intervene in civil conflicts. For instance, Saideman (1997, 2002) asserts that if ethnic ties with an outside group determine foreign policy preferences of important constituents, then those ethnic ties will also influence leaders? foreign policy decisions because they want to stay in office. Since ?politically relevant supporters ? are a crucial concern for 87 policymakers?, leaders take into account important constituents? ethnic ties to outside groups when formulating policies, such as intervention in a civil conflict (Saideman 2002, p. 32). Based on the existing research, it can be argued that leaders are more likely to intervene militarily when their key constituents share ethnic ties with the majority of the population in the conflict state. Since democratic leaders? survival depends on a large segment of the population, the key constituents for democratic leaders are likely to be the dominant ethnic group and the second largest ethnic group in the country. While an autocratic leader?s political survival depends on the support of a smaller portion of society which generally consists of political and military elites, it is highly likely that at least some members of his winning coalition will be from the dominant ethnic group or the second largest ethnic group in society. Heger and Salehyan argue that ?ethnic nepotism heavily shapes the contours of the winning coalition? in both democracies and autocracies (2007, p. 391). Thus, it can be expected that if the two largest ethnic groups in society have ties to the majority of the population in the conflict state, then leaders? decision making regarding military intervention will be affected by the existence of these affective linkages. The following hypotheses are proposed regarding the intervention decision. H18a: The probability of military intervention by an external state increases if it has the same dominant ethnic group as the conflict state. 88 H18b: The probability of military intervention by an external state increases if its dominant ethnic group is the second largest ethnic group in the conflict state and its second largest ethnic group is the dominant ethnic group in the conflict state. These hypotheses do not address leaders? decision on whose side to intervene because data are not available in the literature to analyze that. However, given the previous empirical findings on the role of ethnic ties in foreign policy making, it is a critical domestic variable to include in the theoretical model. Two additional theoretical arguments can also be made about its particular significance for the intervention decision. First, since leaders are interested in satisfying their domestic constituents? policy preferences, by intervening leaders want to convey to the public that they care about protecting ethnic kins abroad who are threatened by a civil conflict. As Huth argues, ?a leader?s desire to hold onto power can create incentives to pursue foreign policies that the leader believes will enhance his domestic political position? (1998, p. 746). Leaders would most likely side with the domestic disputant that is favored by their ethnic kins in the conflict country. Therefore, the ethnic ties argument is more significant for the intervention decision because if it is found to influence leaders? decision to intervene, then they are expected to intervene militarily on the side favored by their ethnic kins instead of supporting the domestic party which threatens them. Second, these hypotheses will be tested against all types of civil conflicts, not just ethnic conflicts. If the security of ethnic kins is threatened due to an ideological conflict, then interveners are expected to side with the domestic party which is supported by their ethnic kins. In other words, these hypotheses do not focus on the ethnic identity of the ?domestic disputants?, rather they focus on the 89 ethnic identity of largest groups in the conflict state. Hence, the hypotheses aim to test whether leaders have a higher likelihood of intervening to protect ethnic kins abroad threatened by any type of conflict, regardless of whether they are a party to the conflict or not. Based on these theoretical justifications, it is critical and necessary to analyze the impact of ethnic ties on leaders? decision to intervene. Domestic Unrest: The systematic linkages described by modified realism between domestic politics and foreign policy decision making help identify another critical domestic variable that can affect leaders? cost-benefit analysis regarding intervention. As mentioned above, Huth argues that ?the simultaneous pursuit by state leaders of domestic political and international security may create potential trade-offs between the resources available ? to support both sets of goals? and may constrain leaders? foreign policy options because of the ?high priority attached by leaders to domestic political survival? (1998, p. 747). When there is domestic unrest and instability, leaders are more likely to be constrained in their foreign policy options. Domestic crises such as coup attempts, strikes or anti-government demonstrations can force national leaders to concentrate on domestic issues, instead of international events, in order to eliminate threats to their domestic political survival. Additionally, leaders are more likely to direct the resources of the state to suppressing domestic challenges instead of using them in foreign military interventions. Thus, it can be argued that leaders are less likely to intervene in civil conflicts elsewhere when they are faced with domestic unrest in their own country because leaders would direct their efforts and resources primarily to consolidating their power in the domestic setting. 90 Diversionary theory, on the other hand, argues that both democratic and autocratic leaders can attempt to draw attention away from domestic problems by using low levels of military force internationally (Levy 1989, Gelpi 1997, DeRouen 2000). With the goal of influencing domestic politics, a state leader can use the international employment of military force to rally the fractious public around his leadership. Hence, leaders can strategically resort to foreign policy behavior that involves the use of military force to divert the public?s attention away from domestic problems and focus it on international issues that require national unity. Based on the logic of diversionary theory, one can also argue that leaders would be more likely to intervene militarily in a civil conflict as a diversionary foreign policy behavior. However, the literature on diversionary theory remains divided. In cross- national studies, Gelpi (1997) and Davies (2002) conclude that democracies are more likely to divert than autocracies, while the findings by Miller (1999) and Enterline and Gleditsch (2000) demonstrate the opposite. Another disagreement in the literature is over whose attention leaders are trying to divert from domestic crises. Morgan and Bickers (1992) argue that elites, not the public, are the only domestic group that can cause democratic leaders to engage in diversionary foreign policy behavior. Kisangani and Pickering (2007) find that elite unrest compels leaders both in democracies and mixed regimes to resort to diversionary behavior. Moreover, studies frequently assume that ?diversionary force tends to be hostile in intent and is typically launched over high politics issues? but Kisangani and Pickering?s empirical results suggest that leaders generally use ?benevolent missions launched for low politics issues? such as ?humanitarian relief operations and peaceful missions? for 91 diversionary purposes (2007, pp. 281, 295). ?Belligerent operations launched for high politics reasons? including ?intervention to take side in a domestic dispute? or ?to change political regime in the target country? and issues that involve ?regional power balances, stability? are excluded from diversionary foreign policy behavior of leaders (Kisangani and Pickering 2007, pp. 281, 284-285, 295). Kisangani and Pickering also conclude that while democracies and mixed regimes use diversionary force for benevolent missions, autocracies do not seem to ?externalize their domestic problems? at all (2007, p. 295). These are not the only contradictory findings in the literature on diversionary behavior, but they are the most relevant ones for this study. Given the important theoretical arguments and mixed empirical findings regarding the impact of regime type on the diversionary use of force, this study expects elite unrest to have a different effect on the intervention behavior of democracies and autocracies. Mass unrest, on the other hand, is expected to have similar effects on democratic and autocratic leaders. Since the literature suggests that democratic leaders might respond to elite unrest by using diversionary force, the following hypotheses are proposed: H19a: The probability of military intervention by a democratic state increases if it is experiencing elite unrest domestically. H19b: The probability of military intervention by an autocratic state decreases if it is experiencing elite unrest domestically. Based on the systematic linkage identified by modified realism which argues that domestic problems may constrain leaders? foreign policy options by threatening their domestic political survival, this study posits that both democratic and autocratic 92 leaders are less likely to intervene when they are faced with mass unrest. The following hypothesis is proposed in the case of mass unrest: H20: The probability of military intervention by an external state decreases if it is experiencing mass unrest domestically. Although the impact of elite and mass unrest will also be analyzed on all states in the full sample, the sub-sample analyses conducted for democratic and autocratic states will be more appropriate for testing these hypotheses and understanding how elite and mass unrest affect the intervention decisions of democratic and autocratic leaders specifically. Thus, in addition to providing a robustness check, sub-sample analyses are useful for comparing the impact of critical international and domestic factors on the intervention behavior of different types of states. Contextual Sources of Military Intervention This study argues that combining actor-oriented and conflict-oriented approaches provides an improved understanding of the rationale behind military interventions in civil conflicts; therefore, this final section examines the impact of contextual factors originating from the civil conflict environment on foreign leaders? decision making. As Findley and Teo also argue, ?conceptualizing the civil [conflict] itself as context integrates the foreign policy aspect of intervention with considerations for relevant ground conditions? (2006, p. 829). Hypotheses about critical contextual conditions which are most likely to factor into foreign leaders? cost-benefit analysis regarding whether and on whose side to intervene are presented in this section. 93 Regime Type of Conflict State: The first contextual factor is the regime type of the conflict state. It has already been hypothesized in the previous section that an external state is more likely to intervene on the government side if it has the same regime type as the conflict state. However, it is still important to analyze if democracies are more or less likely to attract military interventions than autocracies because there are contradictory findings in the literature. Hermann and Kegley (1995, 1996) find that the targets of intervention are disproportionately autocracies. Democracies are less likely to be the targets of military intervention irrespective of the political system of the intervener. These results are also supported by Kathman?s study (2011). On the other hand, Lemke and Regan (2004) observe that democracies involved in civil conflicts are more likely to experience foreign intervention. In short, intervention studies seem to diverge on the impact of the conflict state?s regime type on foreign military interventions. Since it is critical to understand whether external states, irrespective of their regime type, are less likely to interfere in the domestic affairs of democracies; it is necessary to include the regime type of the conflict state in the model. The inclusion of this contextual factor will also complement the findings from previous hypotheses because it will provide a thorough understanding of the relationship between military intervention and regime type by analyzing whether democracies are more resistant to interventions than autocracies and if this resistance decreases the likelihood of interventions on the government side as suggested by Hermann and Kegley (1996). This study proposes the following hypotheses: 94 H21a: The probability of military intervention by an external state decreases if the conflict state is a democracy. H21b: An external state is less likely to intervene on the government side if the conflict state is a democracy. Type of Civil Conflict: The second contextual factor that is likely to affect leaders? decision making is the type of civil conflict. It can be argued that certain types of civil conflicts are more salient to potential interveners than others and are more likely to attract military interventions. As discussed in the literature review chapter, studies that examine the role of ethnic politics in international relations argue that ethnic conflicts are particularly prone to interventions. However, findings from recent studies indicate that ethnic conflicts are not more likely to attract military interventions than other conflicts. Using Regan?s dataset that distinguishes between ethnic and ideological conflicts, three important intervention studies conclude that both types of conflicts appear to increase the probability of biased military interventions (Lemke and Regan 2004, Findley and Teo 2006, Gent 2008). However, Gent (2008) also adds that government-biased intervention is more likely in ideological conflicts than in identity conflicts whereas rebel-biased intervention is equally likely in both types of conflicts. Since recent findings could not find much difference in the intervention behavior of states towards ethnic and ideological conflicts, this study aims to shed light on how type of conflict can affect intervention likelihood by distinguishing between civil conflicts fought over government or territory which has not been analyzed adequately yet. Classification of civil conflicts into governmental and territorial disputes is also 95 more promising for arriving at conclusive results because ethnic conflicts can be related to governmental and/or territorial issues; therefore, the typology used in this study is more accurate. This study argues that governmental conflicts are more prone to foreign military interventions than territorial conflicts because foreign leaders are much more sensitive to the possibility of government change or regime change in other states as Rosecrance (1963) and Rosenau (1969) have argued. Since both democratic and autocratic leaders would be more alert to the possibility of a government change in a country, they would be more likely to intervene in governmental conflicts than in territorial conflicts. It can also be hypothesized that these interventions are more likely to happen on the government side because generally more states are interested in the stability of nations in the international system and want to maintain the status quo rather than facing unsettling changes. This prediction is also in line with Gent?s finding regarding ideological conflicts. H22a: The probability of military intervention by an external state increases if the civil conflict is fought over government. H22b: An external state is more likely to intervene on the government side when the civil conflict is fought over government. Conflict Intensity: In addition to the contextual factors discussed above, there are also changing contextual conditions of the civil conflict that affect foreign leaders? decision making. Intensity of a conflict is one those changing contextual conditions that potential interveners pay attention to. As the conflict escalates and battle deaths increase, 96 leaders are more likely to be alarmed by the situation. Worsening conditions pose bigger and more credible threats to other states, especially to those which have strategic interests in the conflict state. Growing intensity of the conflict may also trigger interventions that motivate reactionary interventions by additional states. To sum up, this study argues that as the intensity of a conflict increases, leaders are more likely to perceive the situation as posing threats or opportunities and consequently they would be more likely to intervene. However, conflict intensity can increase the probability of foreign military intervention on either side. Therefore, conflict intensity is not expected to have a discernible impact on whose side an external state intervenes. The only hypothesis that can be formulated is the following: H23: The probability of military intervention by an external state increases as the intensity of the civil conflict increases. Refugees: Another changing contextual condition related to the civil conflict is the number of refugees. A growing number of refugees creates social and economic problems for neighboring countries but it also indicates that the civil conflict is becoming more violent and destructive. In addition to the number of battle deaths caused by the civil conflict, foreign leaders are also concerned about the increasing number of refugees because it is another sign that the civil conflict is escalating. Therefore, as the number of refugees originating from the conflict country increases, external states are more likely to intervene. Similar to the situation with conflict intensity, increasing number of refugees can increase the probability of foreign 97 military intervention on either side. Hence, it is not expected to have a discernable impact on whose side an external state intervenes. Only one hypothesis is proposed. H24: The probability of military intervention by an external state increases as the number of refugees resulting from the civil conflict increases. Relative Capabilities of Opposition: The last contextual factor that is likely to affect foreign leaders? decision making is the military capabilities of the opposition relative to the government in the civil conflict. Since foreign military intervention changes the balance of capabilities between domestic disputants, leaders pay attention to this contextual factor while deciding whether and on whose side to intervene. Gent (2008) argues that external states are more likely to become involved in civil conflicts when military intervention has the greatest marginal effect on producing the outcome preferred by the third party. Based on this logic, government and rebel biased interventions should be more likely when the government is facing a stronger opposition. When there is a weak rebel group, the government is likely to win the conflict without foreign military support. However, strong rebel groups provide a credible threat to the survival of the government. External states that are sympathetic towards the government are more likely to intervene militarily to help the government when it is fighting a stronger rebel group. On the other hand, external states that are biased in favor of rebels should also be more likely to intervene when the rebel group is stronger because they would want to increase the probability of rebel victory. Following Gent?s argument, this study expects the probability of foreign military interventions on both government and opposition side to increase as the relative 98 capabilities of the opposition increase. Since stronger opposition increases intervention probability on both sides, only the following hypothesis is proposed: H25: The probability of military intervention by an external state increases as rebels? fighting capacity against the government increases. Conclusion The purpose of this chapter was to present a unified theoretical framework that combines actor-oriented and conflict-oriented approaches in the intervention literature. Since military intervention is a foreign policy instrument, the synthesis has been conducted by making potential intervener?s foreign policy decision making theoretically central to the study. Hypotheses on international and domestic factors that are likely to affect leaders? intervention decision have been derived from modified realism because the modified realist framework is able to account for both international and domestic sources of foreign policy decisions as well as the systematic linkages between them. Attention has also been paid to the contextual conditions related to the civil conflict that influence foreign leaders? intervention decision. Hypotheses were formulated about the contextual factors that are most likely to be relevant for leaders? decision making calculus a result of using a unified theoretical approach. The hypotheses derived from the theoretical framework in this chapter will be tested against all states in the international system to uncover universal patterns about the rationale behind foreign military interventions. Then, these hypotheses will be tested against three important sub-samples which are major powers, democracies and 99 autocracies to understand the relative importance of critical factors for each group of state and to check the robustness of empirical results. Research design of this study and the methods used to test the hypotheses are the subject of the following chapter. 100 Chapter 4: Research Design The purpose of this chapter is to explain the research design of this study. The first part of the chapter will discuss the cases included in the dataset, data collection and coding procedures, as well as the operationalization of the dependent and independent variables. After presenting the data, this chapter will explain the statistical models employed to test the hypotheses. Other important methodological issues related to the research design will also be covered in the second part of the chapter. Dataset The dataset used in this study includes actual and potential interveners in all civil conflicts between 1946 and 2002. Potential interveners consist of all states in the international system, except the conflict state.2 Civil conflicts are drawn from the UCDP/PRIO Armed Conflict Dataset version 4-2010 (Gleditsch et al. 2002). Each observation is a pair of actual or potential intervener and conflict state. Intervener- conflict state dyads are observed from the year the civil conflict starts until the year it ends. Thus, the data format is time-series cross-sectional which captures the changes in variables on a yearly basis. 2 Micro states that have a population of less than 500,000 people are also excluded from the group of potential interveners in the international system. Exclusion of micro states is quite common among conflict datasets in the literature. 101 The UCDP/PRIO dataset uses a lower threshold of battle deaths to identify civil conflicts in the world as opposed to other civil conflict datasets that use much higher thresholds and consequently exclude many cases of intrastate disputes. For instance, the COW project only covers civil wars and requires a minimum 1000 battle deaths per year for a conflict to qualify as a case of civil war (Sarkees and Wayman 2010). Regan requires a total of 200 fatalities in order to include a civil conflict in his dataset (Regan 2000). UCDP/PRIO, on the other hand, has a threshold of 25 battle deaths per year which makes it possible to analyze the variation in intervention behavior of foreign powers in civil conflicts with different levels of violence. Thus, the opportunity to test the hypotheses in this study against a larger number of cases that vary in conflict intensity is the primary reason for selecting the UCDP/PRIO dataset. The UCDP/PRIO dataset defines an armed conflict as ?a contested incompatibility that concerns government and/or territory where the use of armed force between two parties, of which at least one is the government of a state, results in at least 25 battle-related deaths? (UCDP/PRIO Codebook version 4-2010, p. 1). There are four types of armed conflicts in the UCDP/PRIO dataset: extra-systemic such as anti-colonial wars, interstate, intrastate and internationalized intrastate conflicts. This study includes only the last two categories of conflicts from the UCDP/PRIO dataset that began between 1946 and 2002. The UCDP/PRIO dataset considers all armed conflicts within one country over separate territories as distinct conflicts and all armed conflicts over government as the same conflict. Even though the dataset makes a distinction between civil 102 conflicts over government and over different territories within the same country, the termination dates of armed conflicts are still problematic as a result of this coding rule. For example, all governmental disputes, such as coups, that are 10 years apart or armed conflicts triggered by different rebel groups regarding the same territory that occur 20 years apart are treated as part of the same conflict. In order to distinguish between conflicts within a country over government or over the same territory which are quite apart in time and to be able to treat them as separate cases of civil conflicts in the dataset, this dissertation developed several coding rules to identify end dates for conflicts within a country. The conflict termination criteria developed in this study to deal with the problem of recurring conflict in the UCDP/PRIO dataset are based on the following rules. If a peace agreement or complete military victory ends the conflict and the conflict is inactive for at least three consecutive years, then it is considered terminated. If the conflict restarts within three years of the peace agreement or military victory, then it is considered as the continuation of the same conflict. If it restarts after three years, then it is coded as a new conflict. A conflict can also terminate as a result of inactivity despite the lack of a clear outcome. If no peace agreement or complete military victory occurs but the conflict becomes inactive meaning there are no battle-related deaths taking place for at least five years, then it is also considered terminated. If the conflict restarts within a period of five years, then it is considered the same conflict. When the conflict restarts after a period of five years of inactivity, it is coded as a new conflict. 103 The choice of three year or five year intervals depending on the conflict situation was made for two reasons. First, there are quite a number of civil conflicts in the UCDP/PRIO dataset that restart after two years despite the existence of a peace agreement or a complete victory by the government. Hence, two years seem to give enough time to the same rebel organizations to remobilize after a government victory or after a peace agreement they are not satisfied with. Moreover, these rebel groups tend to be supported by the same external parties when the conflict restarts in two years, which indicates that interveners also view it as a continuation of the same conflict. Therefore, the second reason has to do with the perception of potential interveners. The data patterns reveal that three or five year intervals are long enough periods for potential interveners to perceive it as a new conflict with different dynamics and ground conditions instead of continuation of the same conflict that is experiencing a lull or remobilization period during which rebels continue to receive external military support. Thus, the conflict termination criteria in this study have been developed carefully by paying attention to both conflict dynamics and potential interveners? perception of the conflict. Three and five year periods are sufficiently cautious guidelines for determining whether or not a civil conflict is terminated as well as for deciding whether an intervention occurs in a new conflict or during the same conflict. As a result of applying the conflict termination criteria developed in this study to the UCDP/PRIO dataset, 232 civil conflicts are identified between 1946 and 2002. Appendix A provides a complete list of these civil conflicts. In order to identify the instances of biased military interventions in these 232 civil conflicts, this study utilized different datasets. Since some of these data sources 104 were incompatible with each other as a result of having different formats and operationalization rules, a number of coding and merging procedures were developed in order to compile the intervention cases. This study defines military intervention as ?convention-breaking? military activities in the internal affairs of a foreign country directed at ?changing or preserving the structure of political authority in the target state? which are finite and transitory (Rosenau 1969, pp. 161-162). Military interventions are operationalized as the supply of troops, military equipment, intelligence and logistical support, air or naval support to one of the sides in a civil conflict. The cases of military interventions that fit the definitional criteria of this study were identified from six different datasets that provide information on interventions. Depending on the time period and the types of conflicts covered in each dataset, intervention cases were cross-checked in at least two of these data sources. Whenever there was an inconsistency between two datasets or when an intervention instance was included only in one dataset, additional sources were checked to decide if a case qualifies for inclusion according to the definitional criteria of this study. The UCDP/PRIO dataset identifies the states that actively support one of the sides in a civil conflict with troops. The information on these biased military interventions is provided within the category of internationalized civil conflicts. However, the UCDP/PRIO dataset excludes other types of military interventions that are included in this study. In order to collect information on all cases and types of military interventions, this study also looked at four intervention datasets that are frequently used in the literature. These four data sources are Regan?s dataset on military interventions in intrastate conflicts between 1945 and 1999 (Regan 2000), 105 Tillema?s dataset on foreign overt military interventions between 1945 and 1985 (Tillema 1989), Pearson and Baumann?s international military intervention (IMI) dataset between 1946 and 1988 (Pearson and Baumann 1993), and the updated IMI dataset by Kisangani and Pickering which covers international military interventions between 1989 and 2005 (Kisangani and Pickering 2008). Finally, external support data from the 2001-2003 release of the Minorities at Risk Dataset were also used to supplement those four intervention datasets (Minorities at Risk 2009). Thus, the military intervention data used in this study is a synthesis of different datasets in the literature and can be described as an updated collection of military intervention data which have been cross-checked and supplemented by multiple sources to study foreign military interventions as thoroughly as possible. Table 4.1 shows the number of civil conflicts that experienced foreign military interventions while Table 4.2 presents the total number of military interventions in the dataset according to their targets. In addition, Appendix B provides a complete list of all military interventions in the dataset. A civil conflict can have multiple military interventions by different states during a conflict year as shown in Appendix B. Table 4.1: Trends in Military Interventions in Civil Conflicts Total # of civil conflicts between 1946-2002 232 100% Civil conflicts with military intervention 142 61% Civil conflicts without military intervention 90 39% 106 Table 4.2: Military Interventions according to their Targets in Civil Conflicts Total # of military interventions between 1946-2002 2,285 100% Military interventions on government side 1,056 46% Military interventions on rebel side 1,229 54% As mentioned above, the dataset used in this study includes both the actual and the potential interveners in the 232 civil conflicts between 1946 and 2002. Potential interveners are included in the dataset to avoid the selection bias problem some quantitative studies in the literature suffer from. Examining causes of third party interventions by testing hypotheses against cases of interventions is methodologically flawed due to selecting on the dependent variable. A research design that tests hypotheses against cases of intervention by actual interveners as well as cases of non-intervention by potential interveners solves this selection bias problem and provides more accurate results. Given that each pair of actual/potential intervener and conflict state is observed from the year the civil conflict starts until the year it ends, the total number of observations in the dataset is 209,685. Dependent Variables Since this dissertation examines the factors that affect external states? decisions regarding whether and on whose side to intervene militarily in civil conflicts, there are two dependent variables in this study and both of them are dichotomous. The first dependent variable, intervention, measures whether or not the external state intervened militarily at a given year during the life of the civil conflict. If a potential intervener carried out a military intervention in a conflict state in that dyad year, then intervention is coded as 1 for that dyad year. When the external state 107 intervenes in multiple years while the conflict is continuing, each of those dyad years receives a coding of 1. Table 4.2 shows that 2285 observations received a coding of 1 for the first dependent variable among the 209,685 dyad years in the dataset. The second dependent variable, intervention side, measures whether the external state intervened on the side of the government or rebels. If the first dependent variable equals 1 indicating that military intervention took place in that particular dyad year, then the second dependent variable is coded as well. Interventions on the government side are coded as 1 while interventions on the rebel side are coded as 0. According to Table 4.2, 1056 military interventions were on the side of the government out of the 2285 military interventions in the entire dataset. Appendix C provides detailed descriptions, coding rules and sources for both the dependent variables and the independent variables which are discussed briefly in the following section. Independent Variables Independent variables that are used in this study are discussed in four separate subsections. International Sources of Intervention This subsection covers the independent variables which are used to test the hypotheses on the international sources of military intervention. Three different variables are constructed to examine the role of interstate rivalry between an external state and a conflict state as well as between interveners themselves. Rivalry is a dichotomous variable that measures whether rivalry exists 108 between the external state and the conflict state in a given dyad year to test hypotheses 1a and 1b. Rival intervention, on the other hand, indicates whether a rival of the external state intervened militarily in the conflict state during that particular year. A third variable also examines whether the rival intervention took place on the rebel side or not. These two variables are used to test hypotheses 10a and 10b. The source for rivalry data is the New Rivalry Dataset, 1816-2001 (Klein, Goertz and Diehl 2006). Hypotheses 2a and 2b are examined by another dichotomous variable that measures whether a militarized interstate dispute exists between the external state and conflict state in a given dyad year. The data source for this variable is the COW Militarized Interstate Dispute Dataset v4.01 (Ghosn, Palmer and Bremer 2004). Similar to rivalry, hypotheses on alliance ties are tested with three different variables. Alliance is a dichotomous variable indicating whether there is a formal alliance between the external state and the conflict state in a dyad year. Defense pacts, neutrality pacts, non-aggression pacts and ententes are considered as alliances. Ally intervention measures whether an ally of the external state intervenes militarily in the conflict state in that year. Finally, a third variable looks at whether the ally intervenes militarily on the government side or not. The last two variables are used to test hypotheses 11a and 11b. The source for alliance data is the COW Formal Alliances Dataset v4.1 (Gibler 2009). Hypotheses 4a and 4b on the historical ties between the external state and the conflict state are tested with a dichotomous variable that equals 1 if the conflict state was a former colony of the external state. However, it can also be coded as 1 if the conflict state seceded from that potential intervener. The data 109 source used to code this variable is the ICOW Colonial History Dataset v0.4 (Hensel 2009). Two different variables are employed to test hypotheses 5 and 6 which examine the impact of geographic proximity on the probability of military intervention. The first one, contiguous, is a dichotomous variable that equals 1 if the potential intervener and the conflict state are separated by land or river border according to the COW Direct Contiguity Data v3.1 (Stinnett et al. 2002). The second variable, same region, indicates whether the potential intervener and the conflict state are in the same geographical region as defined by the COW Project. These five geographical regions are Africa, Asia, Americas, Europe and Mideast. Lootable and non-lootable resources in the conflict state are measured by two proxy variables that are used for testing hypotheses 7 and 8. The existence of secondary diamonds is used as a proxy for lootable resources in the conflict state and these data are taken from the International Peace Research Institute (PRIO) Diamond Dataset (Gilmore, Gleditsch, Lujala and Rod 2005). The existence of petroleum deposits, on the other hand, is used as a proxy for non-lootable resources in a conflict state. Both oil and gas deposits are considered as petroleum deposits according to the PRIO Petroleum Dataset v1.2 (Lujala, Rod and Thieme 2007). Finally, to examine the impact of trade ties on military intervention, a continuous variable measuring the amount of trade per dyad year between the potential intervener and the conflict state is constructed from the available data in the COW Dyadic Trade Dataset v2.01 (Barbieri, Keshk and Pollins 2009). Variables ?flow 1? and ?flow 2? in the dyadic trade dataset reports the imports of state A from 110 state B and the imports of state B from state A respectively. The sum of those two variables gives the total amount of trade between a potential intervener and a conflict state per dyad year in current U.S. millions of dollars. Control Variables This study also controls for several important international factors that are likely to influence the intervention behavior of states. The first control variable measures whether or not a potential intervener is involved in an interstate war in a given dyad year. The data source for this variable is the COW Militarized Interstate Dispute Dataset v4.01. Power status is a dichotomous variable which equals 1 if the potential intervener is identified as a major power by the COW Project. This variable is generated by the EUGene software (Bennett and Stam 2000). Capabilities of the external state relative to the conflict state are measured by a continuous variable that is generated by calculating the ratio of potential intervener?s CINC (Composite Index of National Capability) score to the conflict state?s CINC score for each dyad year. The CINC scores of countries are taken from the COW National Material Capabilities Dataset v4.0 (Singer 1987). Finally, Cold War is a dichotomous variable coded as 0 after 1989. Domestic Sources of Intervention This subsection discusses the independent variables which are used to test the hypotheses on the domestic sources of military intervention. Regime types of potential interveners are measured using the Polity IV Project (Marshall, Jaggers and Gurr 2011). Democratic intervener is coded as 1 if it has a 111 polity2 score of seven or higher. Two different variables are used to test the hypotheses on regime similarity. Joint democracy equals 1 when both the potential intervener and the conflict state in the dyad have a polity2 score of seven or higher. Joint autocracy, on the other hand, is coded as 1 when both the potential intervener and the conflict state in the dyad have a polity2 score of six or less. Two dichotomous variables are employed to test hypotheses 18a and 18b which examine the impact of ethnic ties on the probability of military intervention. If the external state and the conflict state have the same dominant ethnic group, then the first ethnic ties variable is coded as 1. The second ethnic ties variable equals 1 when the dominant ethnic group in the external state is the second largest ethnic group in the conflict state and the dominant ethnic group in the conflict state is the second largest ethnic group in the external state. Both of these variables are constructed from Gartzke and Gleditsch replication data (2006). Finally, two variables that are used to test the hypotheses regarding domestic unrest are composed from the Domestic Conflict Event Data section of the Cross- National Time-Series Data Archive (Banks 2008). Elite unrest consists of government crises and/or purges while mass unrest includes general strikes, riots, anti-government demonstrations, and/or civil conflicts. Both variables are measured dichotomously where 1 indicates that some type of domestic unrest which qualifies as elite or mass unrest took place in the potential intervener state in that dyad year. Contextual Sources of Intervention The last subsection discusses the independent variables which are employed to test the hypotheses on the contextual sources of military intervention. 112 A conflict state is considered democratic if it has a polity2 score of seven or higher according to the Polity IV Project (Marshall, Jaggers and Gurr 2011). The type of civil conflict is also a dichotomous variable which is coded as 1 when it is governmental and as 0 when it is a territorial conflict based on the UCDP/PRIO Armed Conflict Dataset (version 4, 2010). Battle deaths measure the number of battle-related fatalities in the conflict state per dyad year and the data source for this variable is the PRIO Battle Deaths Dataset v3.0 (Lacina and Gleditsch 2005). Refugees, another discrete variable, indicate the number of refugees originating from the conflict state per conflict year. These data are generated from the United Nations High Commissioner for Refugees Statistical Online Population Database. The last contextual variable related to the civil conflict conditions is the fighting capacity of rebels relative to the government in the conflict state. This is an ordinal variable with three categories identified as low, moderate or high fighting capacity rated relative to government forces in the conflict state. These data are taken from the Non-State Actor Data v3.3 (Cunningham, Gleditsch and Salehyan 2012). Table 4.3 presents descriptive statistics for all variables. 113 Table 4.3: Descriptive Statistics Dependent Variables N Mean St. Dev. Min Max Intervention 209,685 .011 .104 0 1 Intervention side 209,685 .005 .0708 0 1 Independent Variables N Mean St. Dev. Min Max International Rivalry 209,685 .014 .115 0 1 Rival intervention 209,199 .040 .195 0 1 Rival intervention on rebel side 209,199 .018 .132 0 1 Militarized dispute 206,085 .010 .099 0 1 Alliance 209,685 .047 .211 0 1 Ally intervention 209,685 .086 .280 0 1 Ally intervention on govt. side 209,685 .072 .258 0 1 Historical ties 209,685 .006 .075 0 1 Contiguous 209,685 .031 .172 0 1 Same region 209,685 .202 .402 0 1 Secondary diamonds 209,685 .379 .485 0 1 Petroleum 209,685 .799 .401 0 1 Trade 209,685 134.889 930.722 .0005 148,503 Domestic Democratic intervener 206,047 .316 .465 0 1 Joint democracy 205,235 .085 .279 0 1 Joint autocracy 205,235 .507 .499 0 1 Ethnic ties 1 209,653 .023 .151 0 1 Ethnic ties 2 209,653 .019 .136 0 1 Elite unrest 205,942 .181 .385 0 1 Mass unrest 209,685 .397 .489 0 1 Contextual Democratic target 208,863 .261 .439 0 1 Conflict type 209,685 .285 .451 0 1 Battle deaths 209,685 2,977 13,019 13 350,000 Refugees 209,685 198,897 601,026 1 6,339,096 Rebel fighting capacity 209,685 1.252 .472 1 3 Control Variables War involvement 209,685 .038 .192 0 1 Major power 209,685 .040 .197 0 1 CINC ratio 209,685 -.777 2.330 -9.08 8.637 Cold War 209,685 .596 .491 0 1 114 Statistical Methods The goal of this dissertation is to understand the rationale behind military interventions by examining the factors that affect states? decisions regarding whether and on whose side to intervene in civil conflicts. As discussed in the theory chapter, some of the factors that are hypothesized to influence leaders? decision making on whether or not to intervene are also expected to affect their decision on whose side to intervene. In other words, this study argues that these two decisions are closely linked and examining their interdependence is important from a theoretical standpoint. Since choosing the right method to test all aspects of theory is critical for empirical research, this study applies a statistical model that can examine the relationship between these two decisions while also testing the hypotheses derived from the theoretical framework. The relationship between the two outcomes of interest in this study can be examined by a selection model which can simultaneously estimate the two equations on the decision to intervene and on whose side to intervene. If the selection model finds a systematic linkage between the two dependent variables, then it provides correct estimates for the second equation by taking this relationship into account (Reed 2000, Lemke and Reed 2001). Given that both dependent variables are dichotomous, the appropriate statistical technique for this study is censored probit. Censored probit is the equivalent of a Heckman selection model but in a censored probit model, there is a probit model in the selection equation and a probit model in the outcome equation. Selection models, such as censored probit, are used to detect selection bias problems 115 and to provide unbiased estimates for the outcome equation if selection bias exists. It is important to clarify what the cause of selection bias actually is. In studies that use selection models, such as this one, observations are not randomly selected into the second equation since the second dependent variable is only observed if the first dependent variable is observed. In other words, there is a selected or censored sample for the second equation. The ?nonrandom aspect of the sample is what is commonly misunderstood? as the selection bias problem (Sartori 2003, p. 114). Sartori argues that this nonrandom aspect of the sample does not on its own bias the estimation of the outcome equation (2003, p. 114). The most important selection bias problem has to do with the correlation between the error terms of the outcome and selection equations. When the error term of the outcome equation is correlated with the error term of the selection equation, it means there is a systematic relationship between the two dependent variables and hence the two equations have to be estimated simultaneously in a selection model. If the outcome equation is estimated separately from the selection equation, it produces biased and inconsistent results (Heckman 1979, Sartori 2003). In short, correlation between the errors of two equations is the major cause of selection bias problems because it indicates that the selection process and the outcome process are not independent of each other. By using a selection model, one can simultaneously analyze two equations and estimate the rho which is the correlation between the errors of two equations. If rho is not statistically significant, then errors are uncorrelated and the two dependent variables can be analyzed in two separate equations. However, if errors are correlated then the two outcomes are systematically related which makes the use of a selection 116 model necessary. In other words, a statistically significant rho indicates that the factors that influence the first outcome are correlated with those that influence the second outcome and estimating the second equation separately would lead to biased results. Selection model corrects for this correlation while estimating the outcome equation (Greene 2003, pp. 781-785). Reed notes that estimating the statistical link between the two error terms allows one to correct for the direct and indirect influence of the selection process on the outcome process (Reed 2000). Therefore, analyzing on whose side third parties intervene separately may lead to biased results and inaccurate inferences if there is a statistical link between the error terms of the selection and outcome equations. To the degree that common factors influence both decisions, studying them separately will lead to biased results (Clark and Reed 2003, p. 82). This study already posits that the decisions on whether and on whose side to intervene are linked. Thus, estimating the correlation between the error terms of two equations will clarify this relationship and ?hint at the strength of the actual link between? the two dependent variables (Reed 2000, p. 86). In light of the theoretical arguments made in this study as well as the methodological issues discussed above, the most appropriate statistical technique for modeling the probability of intervention and the side on which intervention occurs is censored probit. The first equation of the censored probit analysis, which is called the selection stage, will test the hypotheses related to the probability of intervention. The second equation of the censored probit analysis, which is called the outcome stage, will test the hypotheses related to the target of intervention. Censored probit will simultaneously estimate the impact of independent variables on the two binary 117 dependent variables and will also report a rho parameter indicating the correlation across the dependent variables? disturbance terms. If the rho parameter is statistically significant, censored probit is clearly the most suitable method to estimate the direct effects the variables have on the dependent variable in each equation, but also the indirect effects they may have on the second dependent variable by affecting the probability of intervention. A censored probit model makes this research design fully equipped to analyze both the decision to intervene and on whose side to intervene while also examining the interdependence between these two outcomes of interest. Since it is reasonable to suspect that the two decisions are not independent of each other, the rare utilization of selection models in the intervention literature is surprising. The use of a censored probit model is one of the strengths of this research design compared to other studies on interventions. Censored probit analysis will be used to test the hypotheses regarding the decision to intervene and on whose side to intervene against all third parties in the dataset. After testing hypotheses against all observations, this study will also employ censored probit analyses to examine different subsamples in the dataset. These additional analyses will be used to check the robustness of results and to understand the relative importance of critical factors on the intervention behavior of major powers, democracies and autocracies. Other Methodological Issues related to the Research Design The research design of this study has several advantages over the majority of the large-N studies on military interventions in civil conflicts. First, the time-series 118 cross-sectional data format enables one to analyze the impact of changing conditions over time. Second, the censored probit model is useful for examining the interdependence between the two dependent variables. However, there are some additional methodological issues that need to be addressed since the data structure is time-series cross-sectional with a binary dependent variable which Beck et al. refer to as BTSCS data (Beck, Katz and Tucker 1998). BTSCS data tend to suffer from heteroskedasticity and temporal dependence which threaten proper inference (Beck and Katz 1995; Beck, Katz and Tucker 1998). In order the deal with the problem of heteroskedasticity, this study uses Huber-White robust standard errors clustered on potential intervener-conflict state dyads. Using robust standard errors clustered over dyads addresses the problem of heteroskedasticity and the fact that observations for the same potential intervener- conflict state dyad may not be independent. As Doyle and Sambanis note, ?clustering allows one to relax the assumption of independence among? same dyad observations (Doyle and Sambanis 2000, p. 786). Observations in BTSCS data are also likely to be temporally dependent which violates the independence assumption and leads to biased results as Beck et al. demonstrate (Beck, Katz and Tucker 1998). Therefore, it is necessary to control for time dependence in BTSCS analysis. This study uses the correction method developed by Beck et al. (1998) to control for time dependence. All analyses include spell years and cubic splines as functional forms for the baseline duration dependency. Spell years variable counts how many years have passed since the previous military intervention for the same dyad. This correction method takes 119 account of the fact that the probability of intervention depends both on whether or not there was a previous intervention by the same third party state and on how many years have passed since the previous intervention. Spell years and cubic splines in the dataset are created by the BTSCS software program provided by Tucker (1999). Another statistical concern involves missing data. Data collected for the majority of the independent variables were available for all 209,685 dyad years while data collected for the remaining variables, except two of them, were available for almost 98 percent of the observations. The two variables with a lot of missing data were trade and refugees. Data on trade were available for 165,981 dyad years which is approximately 80 percent of the dataset. Since trade is a critical variable in the model, list-wise deletion would cause losing 20 percent of the data in the empirical analysis. Furthermore, list-wise deletion would also introduce sample selection bias and produce biased estimates because trade data were not available for a particular group of countries and were not missing randomly. Therefore, this study used the mean substitution method to replace the missing trade data for 20 percent of the observations. In the case of refugees, only 67 percent of the observations were available. In order to avoid list-wise deletion, this study also replaced the missing values for refugee data with the mean value for 33 percent of the observations. However, given that one third of the refugee data is substituted with the mean value, this study uses the refugees variable in only one model to examine its impact. The number of refugees is used as an indicator of conflict intensity for civilian population. Since the number of battle deaths is another important indicator of conflict intensity, there is no reason to include refugees in all models unlike trade which is a critical 120 variable that evaluates the impact of economic relations between a potential intervener and a conflict state in the theoretical model. Therefore, trade is included in all models while the refugees variable is only included in one model to examine its impact. Attention has also been paid to avoiding multicollinearity in the models. Two separate models assess the impact of regime type and regime similarity on the dependent variables because the high correlation between the two prevents them to be included in the same model. The first model includes regime type of the potential intervener and regime type of the conflict state. The second model examines the impact of regime similarity by including joint democracy and joint autocracy as two separate variables in the analysis. Finally, some of the independent variables are lagged one year to ensure temporal order in all empirical analyses. In addition to Appendix C, results tables in Chapters 5 and 6 also provide information on which variables are lagged in the analyses. Table 4.4 provides a summary of the hypotheses by listing the independent variables and their hypothesized relationship with dependent variables which will be analyzed in the next two chapters. Chapter 5 will test these hypotheses in the entire dataset as well as in the sub-sample of major powers. Chapter 6 will present a comparative analysis and test these hypotheses against democratic and autocratic states. After presenting the results from statistical analyses, both chapters will discuss the significance of empirical findings with respect to the theoretical arguments made in Chapter 3. 121 Table 4.4: Independent variables and their hypothesized relationship with dependent variables Hyp. Independent Variables Dependent Variables International Intervention Government side H1a, 1b Rivalry + - H10a Rival intervention + na H10b Rival intervention on rebel side na + H2a, 2b Militarized dispute + - H3a, 3b Alliance + + H11a Ally intervention + na H11b Ally intervention on govt. side na + H4a, 4b Historical ties + + H5 Contiguous + na H6 Same region + na H7a, 7b Secondary diamonds + - H8a, 8b Petroleum + + H9a, 9b Trade + + Domestic H16a, 16b Democratic intervener - + H17a, 17b Joint democracy + + H17a, 17b Joint autocracy + + H18a Ethnic ties 1 + na H18b Ethnic ties 2 + na H19 Elite unrest - na H20 Mass unrest - na Contextual H21a, 21b Democratic target - - H22a, 22b Government conflict + + H23 Battle deaths + na H24 Refugees + na H25 Rebel fighting capacity + na Control Variables H12 War involvement - na H13 Major power + na H14 CINC ratio + na H15 Cold War + na 122 Chapter 5: Empirical Findings This chapter is divided into three sections. First section presents the results of statistical analyses that were performed using the entire dataset while the second section reports the results from the statistical analyses conducted in the sub-sample of major powers. Results are interpreted in light of the hypotheses postulated in Chapter 3. The final section concludes with a comparative analysis of the empirical findings from the first two sections and discusses their theoretical and substantive significance. Empirical Findings from the Full Sample This section focuses on the empirical findings from the statistical analyses that were conducted using the entire dataset. The following sub-sections present the statistical results, interpret their statistical significance and discuss their theoretical and substantive implications. Statistical Models As mentioned in Chapter 4, three different models are used to test the hypotheses. In order to avoid multicollinearity problems, regime type and regime similarity variables are examined in two different models. The first model includes the regime type variables for the third party state and the conflict state while the third model replaces these variables with the regime similarity variables. Finally, the second model adds the refugees variables to the first censored probit model. Since one third of the refugees data is substituted with the mean value for missing data, this study takes a cautious approach and runs a separate model with the refugees variable. 123 These three censored probit models are presented in Tables 5.1, 5.2 and 5.3. As a result of randomly missing observations, the number of dyad years used in the selection equation of each censored probit model is 199,180 which constitute 95% of the dataset. The number of uncensored observations, which refer to the number of dyad years with intervention, is 2223 in each outcome equation. Hence, 97% of the intervention years are used while estimating the outcome equations. Examination of the correlation matrix from each analysis shows little reason for concern about multicollinearity among the explanatory variables in each censored probit model, with the highest absolute level of correlation occurring between the variables conflict type (as indicated by government conflict) and battle deaths at 0.523. Since the rest of the correlations are much smaller, there is no evidence of significant multicollinearity among the independent variables. In each of the censored probit analyses, the model chi-square rejects the null hypothesis that all coefficients in the model are zero. In the first model, the correlation between the two equations, rho, is 0.65 which is greater than a zero correlation. The hypothesis that rho equals zero is rejected at the 1% level based on the Wald test of independent equations. The statistically significant rho in the first analysis shows that the two dependent variables are linked and the use of a censored probit model is justified. The statistical findings regarding the correlation between the two equations are very similar in the other two models. Since the findings show the existence of selection bias and indicate that the two decisions are linked, the use of a censored probit model is appropriate for this study. In other words, running a separate model to analyze on whose side states intervene will lead to biased results if it is 124 estimated separately from the selection equation that analyzes the decision to intervene. This study argues that the decisions on whether to intervene and on whose side to intervene are closely related and some of the same factors that influence foreign leaders? decisions to intervene also influence which side they support. Given that many of the independent variables that are hypothesized to affect the second dependent variable also appear in the selection equation, the coefficients of the variables in the outcome equation are affected by the presence of the same variables in the selection equation. The statistically significant rho in all three models already proves the linkage between the two dependent variables but the value of rho is also useful for substantive interpretation of the coefficients in the outcome equation. The indirect impact of variables in the selection equation must be taken into account while interpreting the coefficients of those same variables in the outcome equation. Rho facilitates the calculation of this indirect impact and allows one to interpret coefficients correctly in the outcome equation. In short, the direct and indirect effect of explanatory variables on the second dependent variable can only be estimated in a censored probit analysis that takes into account the statistically significant linkage between the dependent variables. Hence, the use of censored probit is not only appropriate but also necessary to understand the substantive impact of the variables. The substantive impact of the variables will be discussed after the results from models 1, 2 and 3 are interpreted in terms of the support they provide to the hypotheses. 125 Table 5.1: Model 1 - Censored Probit Results for Full Sample Selection Equation Outcome Equation Dependent Variables Intervention Intervention on govt. side Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ .506**** (.132) -.613*** (.238) Militarized dispute ~ .588**** (.130) -.213* (.128) Alliance ~ .055 (.085) .457*** (.153) Historical ties .796**** (.182) .822**** (.253) Contiguous .754**** (.108) - Same region .401**** (.078) - Secondary diamonds .075 (.066) -.354** (.171) Petroleum .192*** (.064) .411*** (156) Trade (logged) ~ .045**** (.013) .207**** (.032) Rival intervention ~ .678**** (.078) - Rival intervention on rebel side ~ - .908**** (.167) Ally intervention ~ .285**** (.078) - Ally intervention on govt. side ~ - .621*** (.243) Democratic intervener ~ -.263**** (.063) .029 (.207) Ethnic ties 1 .378**** (.116) - Ethnic ties 2 .407*** (.133) - Elite unrest ~ -.163**** (.046) - Mass unrest ~ -.140**** (.042) - Democratic target ~ -.057 (.083) -.581*** (.212) Government conflict .051 (.065) -.006 (.161) Battle deaths (logged) ~ .154**** (.015) - Rebel fighting capacity ~ -.190**** (.047) - War involvement ~ -.192*** (.063) - Major power .521**** (.088) - CINC ratio (logged) ~ .166**** (.016) - Cold War .005 (.046) - Spell years -.921****(.045) -.418***(.158) Constant -3.014**** (.134) -2.963**** (.465) N Total obs. = 199180 Uncensored obs. = 2223 Wald Chi-square = 361.99**** Rho = .648 (.147) Wald test of independent equations (rho = 0), Chi-square = 9.23*** Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 126 Table 5.2: Model 2 - Censored Probit Results for Full Sample Selection Equation Outcome Equation Dependent Variable Intervention Intervention on govt. side Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ .493**** (.130) -.613*** (.233) Militarized dispute ~ .575**** (.131) -.217* (.128) Alliance ~ .068 (.085) .463*** (.152) Historical ties .777**** (.183) .810**** (.252) Contiguous .741**** (.108) - Same region .398**** (.077) - Secondary diamonds .085 (.066) -.345** (.171) Petroleum .155** (.064) .378*** (.152) Trade (logged) ~ .051**** (.013) .210**** (.032) Rival intervention ~ .673**** (.077) - Rival intervention on rebel side ~ - .907**** (.167) Ally intervention ~ .268**** (.076) - Ally intervention on govt. side ~ - .605*** (.242) Democratic intervener ~ -.265**** (.062) .029 (.205) Ethnic ties 1 .378**** (.115) - Ethnic ties 2 .428**** (.133) - Elite unrest ~ -.161**** (.046) - Mass unrest ~ -.136**** (.042) - Democratic target ~ -.137 (.084) -.529** (.217) Government conflict .013 (.064) -.035 (.165) Battle deaths (logged) ~ .137**** (.016) - Refugees (logged) ~ .060**** (.011) - Rebel fighting capacity ~ -.177**** (.047) - War involvement ~ -.189*** (.063) - Major power .516**** (.088) - CINC ratio (logged) ~ .166**** (.017) - Cold War .020 (.046) - Spell years -.928****(.045) -.431***(.156) Constant -3.513**** (.168) -3.307**** (.545) N Total obs. = 199180 Uncensored obs. = 2223 Wald Chi-square = 370.52**** Rho = .660 (.144) Wald test of independent equations (rho = 0), Chi-square = 9.72*** Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 127 Table 5.3: Model 3 - Censored Probit Results for Full Sample Selection Equation Outcome Equation Dependent Variable Intervention Intervention on govt. side Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ .531**** (.134) -.561*** (.223) Militarized dispute ~ .580**** (.130) -.178 (.126) Alliance ~ .048 (.085) .345** (.151) Historical ties .768**** (.108) .974**** (.202) Contiguous .750**** (.182) - Same region .424**** (.079) - Secondary diamonds .071 (.066) -.351** (.162) Petroleum .191*** (.064) .390*** (.154) Trade (logged) ~ .040*** (.013) .225**** (.033) Rival intervention ~ .705**** (.081) - Rival intervention on rebel side ~ - .817**** (.178) Ally intervention ~ .264**** (.077) - Ally intervention on govt. side ~ - .650*** (.227) Joint democracy ~ .058 (.149) .584* (.325) Joint autocracy ~ .178*** (.065) .519**** (.160) Ethnic ties 1 .349*** (.117) - Ethnic ties 2 .407*** (.136) - Elite unrest ~ -.150**** (.046) - Mass unrest ~ -.149**** (.042) - Government conflict .048 (.064) -.011 (.161) Battle deaths (logged) ~ .148**** (.015) - Rebel fighting capacity ~ -.201**** (.047) - War involvement ~ -.173*** (.064) - Major power .532**** (.088) - CINC ratio (logged) ~ .158**** (.017) - Cold War .017 (.045) - Spell years -.920****(.045) -.411***(.149) Constant -3.130**** (.149) -3.689**** (.412) N Total obs. = 199180 Uncensored obs. = 2223 Wald Chi-square = 365.11**** Rho = .653 (.145) Wald test of independent equations (rho = 0), Chi-square = 9.59*** Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 128 Table 5.4: Summary of Support for Hypotheses Hyp. Independent Variables Dependent Variables International Intervention Government side H1a, 1b Rivalry supported supported H2a, 2b Militarized dispute supported weakly supported H3a, 3b Alliance not supported supported H4a, 4b Historical ties supported supported H5 Contiguous supported na H6 Same region supported na H7a, 7b Secondary diamonds not supported supported H8a, 8b Petroleum supported supported H9a, 9b Trade supported supported H10a Rival intervention supported na H10b Rival intervention on rebel side na supported H11a Ally intervention supported na H11b Ally intervention on govt. side na supported Domestic H16a, 16b Democratic intervener supported not supported H17a, 17b Joint democracy not supported weakly supported H17a, 17b Joint autocracy supported supported H18a Ethnic ties 1 supported na H18b Ethnic ties 2 supported na H19 Elite unrest supported na H20 Mass unrest supported na Contextual H21a, 21b Democratic target not supported supported H22a, 22b Government conflict not supported not supported H23 Battle deaths supported na H24 Refugees supported na H25 Rebel fighting capacity opposite found na Control Variables H12 War involvement supported na H13 Major power supported na H14 CINC ratio supported na H15 Cold War not supported na Table 5.4 summarizes the support the results lend to each hypothesis postulated for both dependent variables to make discussion easier. The table presents whether or not each hypothesis is supported based on the statistical significance and direction of the independent variables across all three models. If a variable is 129 statistically significant but its effect is the opposite of what was hypothesized, then Table 5.4 reports that finding as well. As shown in Table 5.4, a very large portion of the hypotheses are supported for both dependent variables in all of the three censored probit analyses conducted by using the full sample. The hypotheses that receive support are the same across all three models. The particular findings regarding the hypotheses on regime type, refugees and regime similarity reported in the table come from Models 1, 2 and 3 respectively. Finally, a couple of the hypotheses are categorized as receiving relatively weaker support due to the statistical significance of results at the 90% level. The rest of the hypotheses are supported at higher significance levels as can be seen in Tables 5.1, 5.2 and 5.3. The majority of the hypotheses on the international sources of intervention decision receive support. Factors related to the strategic significance of the conflict state such as rivalry and disputes with the conflict state, historical ties, geographic proximity, trade ties and the existence of petroleum in the conflict state increase the probability of military intervention. Foreign powers are also more likely to become involved militarily when their rivals or allies intervene in a civil conflict. As hypothesized, all of these international factors also influence how third parties choose sides in a civil conflict when they intervene. Among the hypotheses that are not supported by the results, one of them is the existence of alliance ties between the external state and the conflict state. Alliance ties do not seem to influence the intervention decision of leaders, but they have a significant effect on the second dependent variable. In other words, even though alliance ties may not motivate 130 leaders to intervene in a civil conflict, they increase the probability of intervention on the government side if an external state decides to intervene, most likely due to other international and domestic factors that receive support in the selection equation. Similarly, the existence of secondary diamonds in the conflict state does not affect the probability of intervention but it does decrease the probability of intervention on the government side (and increase the likelihood of pro-rebel intervention) if a third party intervenes in the civil conflict. Hypotheses regarding the domestic sources of intervention decisions receive more support in the selection equation compared to the outcome equation. Ethnic ties with the conflict state and domestic unrest influence leaders? decision on whether or not to intervene as hypothesized in Chapter 3. Democracies are also less likely to intervene in civil conflicts compared to autocracies as expected. On the other hand, regime type does not seem to affect leaders? decision on whose side to intervene because democratic leaders are not more likely to intervene on the government side than autocracies. Finally, regime similarity increases the probability of intervention for autocratic states but the same cannot be said for democracies since regime similarity does not seem to motivate democratic leaders to intervene in civil conflicts. However, when democratic leaders decide to intervene in a civil conflict, probably due to reasons other than regime similarity, they are more likely to intervene on the government side if the conflict state is democratic. Since autocracies are also more likely to support the government side when the conflict state is autocratic, it can be concluded that regime similarity has a more consistent effect with respect to leaders? 131 decision on whose side to intervene. In terms of the decision on whether or not to intervene, regime similarity matters only for autocracies. Hypotheses related to the contextual sources of intervention receive the least amount of support in all three models. Regime type of the conflict state does not affect the intervention decision as democracies are not more or less likely to attract military intervention compared to autocracies. However, democratic conflict states are less likely to experience pro-government intervention than autocratic conflict states if intervention does occur. The only exception to this finding seems to happen when interveners are democratic because joint democracy increases the likelihood of pro-government intervention as mentioned above. The type of civil conflict also does not matter for the decision on whether and on whose side to intervene. The results show that the impact of rebel fighting capacity on the probability of intervention is the opposite of what was expected theoretically. As rebels? fighting capacity increases relative to the government forces, foreign leaders are less likely to intervene in civil conflicts. The only contextual variables that affect the intervention probability in the hypothesized way are the number of battle deaths and refugees per conflict year. Both of these contextual variables have a positive relationship with the probability of intervention. Finally, all control variables except one have the hypothesized relationship with the intervention decision. Involvement in an interstate war constrains military intervention behavior of states. As expected, increasing relative capabilities of third parties also increase their intervention likelihood. However, Cold War does not have 132 a statistically significant impact on the probability of intervention according to the results from all three models. As an additional robustness check, this study also employed rare events logit to analyze the first dependent variable. Intervention occurs in less than 2% of the observations in the dataset. King and Zeng show that standard logit or probit estimation tends to underestimate the probability of rare events (King and Zeng 2001). Rare events logit is a correction method developed by these scholars for rare events data. The differences between the rare events logit results and the results from the selection equation of the censored probit analysis are negligible with all variables significant at same levels. The results from rare events logit analyses which are consistent with the results from the first stage of censored probit analyses provide additional confidence in the findings on the hypothesized relationships. Appendix D provides the results from relogit models. Interpretation of Results After reviewing the censored probit results in terms of the support they provide to the hypotheses, it is now useful to discuss the substantive impact of the statistically significant variables on the two dependent variables. In a linear model, the marginal effect of x on y is the same as the effect of a one unit increase in x on y which is due to the linearity of the model. However, this is not the case in a binary response model due to the non-linearity of the model. A marginal effect in a binary response model is the effect of a very small change in x on the probability of y=1. This is rarely what one wants to know about. Instead, it is more useful to know how the probability of y=1 changes when x increases by one unit or some number of units. 133 Therefore, it makes more sense to calculate first differences in this study because a first difference is just the change in the probability that y=1 associated with some unit change in x. Tables 5.5 presents the changes in the predicted probability of intervention=1 when the variable of interest increases by one unit or some number of units depending on the nature of the variable. The table indicates the unit change for each independent variable while calculating the first differences. First differences are only calculated for the statistically significant variables from all three models but in order to avoid repetition they are only computed once. The changes in the probability of intervention reported in Table 5.5 are also calculated by holding all continuous variables at their means and all dichotomous and categorical variables at their modes in the respective model they were computed in. Table 5.6 indicates the changes in the predicted probability of pro-government intervention=1 when the variable of interest changes by one unit or some number of units as reported in the table. First differences for the second stage of censored probit analysis are calculated in the same manner for the most part. However, the substantive impact of statistically significant variables in the outcome equation is the sum of their direct and indirect effects on the probability of observing pro-government intervention if those variables also appear in the selection equation. The substantive impact of independent variables reported in Tables 5.5 and 5.6 will be discussed separately for each dependent variable. 134 Table 5.5: Changes in the Predicted Probability of Intervention When this variable moves from ? to ? in Change in probability of intervention 95% Confidence Interval (unless indicated by *) Model 1: Rivalry 0 ? 1 16.46% [8.62% 24.30%] Militarized dispute 0 ? 1 17.23% [9.59% 24.87%] Historical ties 0 ? 1 19.64% [10.37% 28.91%] Contiguous 0 ? 1 19.12% [12.21% 26.03%] Same region 0 ? 1 14.17% [7.48% 20.86%] Petroleum 0 ? 1 7.08% [2.64% 11.52%] Trade p25 ? p75 Trade min ? max 0.83% 1.45% [0.27% 1.39%] [0.53% 2.37%] Rival intervention 0 ? 1 18.31% [10.78% 25.84%] Ally intervention 0 ? 1 12.44% [5.86% 19.02%] Democratic intervener 0 ? 1 -10.62% [-16.51% -4.73%] Ethnic ties 1 0 ? 1 13.35% [5.26% 21.44%] Ethnic ties 2 0 ? 1 14.39% [6.31% 22.47%] Elite unrest 0 ? 1 -5.64% [-9.35% -1.93%] Mass unrest 0 ? 1 -4.19% [-7.12% -1.26%] Battle deaths p25 ? p75 Battle deaths min ? max 3.47% 5.02% [1.69% 5.25%] [2.28% 7.76%] Rebel fighting capacity 1 ? 2 Rebel fighting capacity 1 ? 3 -4.68% -6.84% [-7.53% -1.83%] [-10.19% -3.49%] War involvement 0 ? 1 -7.14% [-11.72% -2.56%] Major power 0 ? 1 16.73% [7.37% 26.09%] CINC ratio p25 ? p75 CINC ratio min ? max 4.37% 6.21% [1.95% 6.79%] [2.44% 9.98%] Model 3: Joint autocracy 0 ? 1 7.43% [3.19% 11.67%] Model 2: Refugees p25 ? p75 Refugees min ? max 1.18% 2.14% [0.55% 1.81%] [1.02% 3.26%] Notes: This table reports the first differences for statistically significant variables from the selection equations in Tables 5.1, 5.2 and 5.3. The probabilities are calculated by holding all continuous variables at their means and all dichotomous and categorical variables at their modes. * indicates 90% confidence interval. Table 5.5 shows that international factors in general have the largest substantive impact on the probability of military intervention compared to domestic and contextual factors. Historical ties with the conflict state and contiguity increase 135 the predicted probability of intervention more than 19 percent. Previous military interventions by rival states increase the probability of intervention by 18 percent as a result of changing the strategic significance of the civil conflict for third parties, while previous military interventions by allies increase the probability of intervention by 12 percent. Hostile relations with the conflict state also play a critical role in the intervention decision of leaders because the existence of militarized disputes and rivalry have the next largest effects. In addition, major powers are almost 17 percent more likely to intervene than minor powers. On the other hand, international factors associated with the economic importance of the conflict state such as trade ties and the existence of petroleum have much smaller effects on military intervention probability compared to the rest of the international factors. Among the domestic factors, ethnic ties have the largest impact on the likelihood of intervention compared to regime characteristics of third parties and the existence of domestic unrest which might constrain foreign policy decision making. Ethnic ties with the dominant or the second largest ethnic group in the conflict state increase the probability of military intervention by 13 to 14 percent. Democracies are 10.6 percent less likely to intervene than autocracies while regime similarity with the conflict state increases intervention probability of autocratic third parties by 7.4 percent. Finally, elite unrest constrains leaders? intervention decision more than mass unrest based on the findings. However, domestic unrest in general has the smallest impact, with -4 to -5.6 percent, on the intervention decision of leaders with respect to other domestic factors. In comparison petroleum, which is one of the less crucial 136 factors among international sources of intervention, increases intervention probability by 7 percent. Contextual factors related to the changing conditions of the civil conflict have a smaller impact than domestic factors on the probability of intervention. As the rebel fighting capacity relative to the government forces changes from low to moderate, the likelihood of military intervention decreases by 4.7 percent. However, when the rebel fighting capacity is high, intervention probability decreases by almost 7 percent. This negative relationship is the opposite of what was expected. Number of battle deaths and refugees per year are the only two civil conflict attributes that increase the probability of intervention among all contextual factors. As the number of battle deaths change from minimum to maximum, the probability of intervention increases by 5 percent while this change is only 2 percent in the case of refugees. Control variables also have a relatively important impact. Involvement in an interstate war decreases intervention probability by 7 percent while increasing relative capabilities have a positive relationship with the military intervention behavior of states. As the ratio of capabilities goes from minimum to maximum, intervention likelihood rises by more than 6 percent. Table 5.6 indicates that international factors also have the largest substantive impact on whose side foreign powers intervene. Intervention by a rival state on the rebel side increases pro-government military intervention probability by 40 percent. On the other hand, intervention by an ally on the government side increases pro- government intervention likelihood more than 20 percent. It can be concluded that previous interventions by rivals or allies in a civil conflict clearly have an important 137 effect on whose side foreign powers intervene. The existence of historical ties also has a large impact as it increases pro-government intervention by 34 percent. The last international factor with a sizable effect on the side decision is rivalry with the conflict state because it decreases pro-government intervention by 23 percent whereas alliance ties increases the probability of intervening on the government side by 15 percent. Table 5.6: Changes in the Predicted Probability of Government Intervention When this variable moves from ? to ? in Change in probability of govt. intervention 95% Confidence Interval (unless indicated by *) Model 1: Rivalry 0 ? 1 -23.12% [-36.98% -9.26%] Militarized dispute 0 ? 1 -9.34% [-18.02% -0.66%] * Alliance 0 ? 1 15.31% [6.42% 24.20%] Historical ties 0 ? 1 34.28% [16.74% 51.82%] Secondary diamonds 0 ? 1 -12.62% [-21.96% -3.28%] Petroleum 0 ? 1 14.17% [4.49% 23.85%] Trade p25 ? p75 Trade min ? max 6.83% 8.76% [3.17% 10.49%] [4.58% 12.94%] Rival intervention rebel side 0 ? 1 40.67% [24.80% 56.54%] Ally intervention govt. side 0 ? 1 23.73% [9.35% 38.11%] Democratic target 0 ? 1 -20.46% [-31.36% -9.56%] Model 3: Joint democracy 0 ? 1 19.18% [7.62% 30.74%] * Joint autocracy 0 ? 1 16.72% [9.88% 23.56%] Notes: This table reports the first differences for statistically significant variables from the outcome equations in Tables 5.1, 5.2 and 5.3. The probabilities are calculated by holding all continuous variables at their means and all dichotomous and categorical variables at their modes. * indicates 90% confidence interval. Domestic factors related to regime characteristics have the second largest effect on the intervention side decision. Regime similarity with the conflict state increases pro-government intervention by 17 percent for autocracies and by 19 percent for democracies. In addition, democratic conflict states are 20 percent less 138 likely to experience pro-government interventions during civil conflicts and are more likely to experience pro-rebel interventions. Resource wealth of the conflict state does not seem to have a critical impact on the decision to intervene. However, lootable and non-lootable resources clearly matter in terms of which side to support once leaders decide to intervene in a civil conflict. The existence of secondary diamonds decreases the probability of pro- government intervention by more than 12 percent while petroleum in the conflict state increases the likelihood of pro-government intervention by 14 percent. Therefore, the type of resources in a conflict state have a significant impact on the side decision even though resource wealth itself may not motivate third parties to intervene in a civil conflict. Finally, trade ties have the smallest substantive impact on the probability of intervention as reported in Table 5.5, but increasing trade ties with the conflict state has a bigger impact on the probability of supporting the government side once a military intervention decision is taken. Discussion of Findings from the Full Sample The findings from the censored probit analyses show strong support for a majority of the hypotheses on the international and domestic sources of intervention derived from the modified realist framework. The findings from the full sample are important theoretically and empirically because some of these international and domestic factors have not been adequately analyzed before in quantitative studies of intervention. This is one of the few quantitative studies that extensively examines the role of critical international and domestic factors on the intervention decision as a result of focusing on the intervener?s foreign policy decision making through a 139 modified realist framework. While the analyses provide an improved understanding of the particular international and domestic factors that motivate leaders to intervene militarily in civil conflicts, they also provide novel findings on others. Finally, by using a combination approach this study is able to show how contextual factors related to civil conflict attributes matter for potential interveners in comparison to the international and domestic sources of foreign policy decision making. The findings show significant support for most of the hypotheses on the international sources of intervention and some of these have not been tested before. According to Tables 5.5 and 5.6, international factors related to the strategic significance of the conflict state such as historical ties, contiguity and hostile relations with the conflict state are quite critical for the intervention decision. While several previous studies have found that historical ties and contiguity increase the probability of intervention, only one previous study analyzed the impact of rivalry on intervention probability (Findley and Teo 2006). The findings here show that both rivalry and other hostile relations with the conflict state, such as the existence of militarized disputes, can increase the likelihood of intervention dramatically. Quantitative studies also have not analyzed the strategic nature of intervention decisions sufficiently. In comparison, the strategic environment of civil conflicts has been explored extensively in case studies. Findley and Teo?s study (2006) is one of the few quantitative studies that analyzes how interventions can occur as reactive responses to previous interventions. This study also shows that interventions by rivals and allies have a substantial impact on the intervention decision of foreign leaders by increasing the strategic significance of the civil conflict. Therefore, both the findings 140 related to the strategic significance of the conflict state and the strategic reactions of interveners to each other make this analysis quite critical. Moreover, by analyzing the decisions on whether and on whose side to intervene in a selection model, this study is able to show how the relative effects of certain international factors differ for each outcome. One of the striking examples from Tables 5.5 and 5.6 is alliance ties. Two previous studies have found that alliance ties increase the probability of intervention (Lemke and Regan 2004, Findley and Teo 2006). This study, on the other hand, finds that while alliance ties are important for choosing sides in a civil conflict, they do not necessarily increase the probability of intervention. In other words, alliance ties do not seem to motivate states to intervene in a civil conflict. The use of a selection model to examine the relationship between the decisions on whether and on whose side to intervene offers a more nuanced understanding of how alliance ties affect intervention behavior. Similarly, the economic significance of the conflict state has a greater impact on how third parties choose sides than on their probability of intervention. As Table 5.6 indicates, the type of resources in a conflict state can affect a third party?s decision on whose side to intervene. Both the findings on trade ties and resource wealth of the conflict state contribute to the current literature by examining the economic significance of the conflict state for potential interveners which have not been addressed adequately in previous studies. This study shows that while economic factors are not as critical as the other international factors when states are deciding to intervene, they can have a significant impact on how states choose sides. 141 Domestic factors have the next largest effect after international factors according to the censored probit analyses from the full sample. Ethnic ties with the conflict state population seem to motivate foreign leaders more than other domestic factors while deciding to intervene militarily. While confirming the previous findings in the ethnic conflict literature on the role of ethnic ties in intervention decisions, this finding also shows that leaders seeking support at home take into account the ethnic composition of the population and their constituents? ethnic ties when formulating foreign policies as modified realism expects. Although it has a smaller substantial effect than ethnic politics, regime type of a country also proves to be quite significant for understanding intervention behavior. As hypothesized, democracies are less likely to intervene in civil conflicts than autocracies; but when they do intervene, democracies are more likely to intervene on behalf of democratic governments instead of against them. This is one of the first studies which finds that democracies are less likely to intervene in civil conflicts. Lemke and Regan (2004) find that regime type has no significant impact on intervention behavior whereas according to Kathman (2011) democracies are more likely to intervene in civil conflicts. Additionally, this is also one of the few studies that shows how regime similarity can play a crucial role when third parties choose sides in a civil conflict. For instance, similar to Lemke and Regan (2004), this dissertation finds that joint democracy does not affect the probability of military intervention. However, this study demonstrates that joint democracy increases the probability of intervening on the government side when democracies do intervene which is an important finding 142 about the impact of regime similarity on choosing sides in a civil conflict. Moreover, this study also contributes to the literature by showing that joint autocracy increases both the probability of intervention and the probability of intervening on the government side. Finally, Lemke and Regan (2004) find that democracies involved in civil conflicts are more likely to experience foreign intervention while Kathman (2011) argues that democracies are less likely to attract military interventions. The censored probit analyses from the full sample lead to a different conclusion about the relationship between the regime type of the conflict state and the probability of intervention. This study shows that democratic conflict states are no more or less likely to experience military interventions than autocracies. However, the findings from both the outcome and selection equations suggest that foreign powers are less likely to intervene on the side of the government when the conflict state is democratic unless the intervener is democratic itself. In other words, regime similarity appears to motivate democracies to intervene on the side of government. In short, the censored probit analyses from the full sample make important contributions to the nascent debate between regime type and intervention behavior in civil conflicts by offering new as well as nuanced findings. Contextual factors related to civil conflict attributes have a much smaller substantive impact on the probability of intervention compared to international and domestic factors which also indicate that an actor-oriented approach is more useful for understanding the rationale behind foreign military interventions. The only two contextual attributes that influence intervention likelihood in the hypothesized way 143 are conflict intensity indicators. The results show that increasing number of battle deaths and refugees also increase the probability of intervention confirming previous findings by Lemke and Regan (2004) and Findley and Teo (2006). However, the substantive impact of these changing attributes of the civil conflict is less than almost all of the other factors in the model. The most important contextual factor turns out to be the fighting capacity of rebels relative to government forces, but the results demonstrate the opposite of Hypothesis 25 which was formulated based on Gent?s argument (2008). Foreign powers are less likely to intervene when rebels? fighting capacity starts to balance or surpass government?s fighting capacity. This is a surprising finding which might indicate that foreign powers perceive military interventions to be more costly when the civil conflict has the potential to experience a stalemate situation. The findings from the full sample demonstrate that control variables such as power status, ratio of capabilities, or involvement in another interstate war also influence leaders? decision making on intervention. However, bipolarity of the Cold War period does not have an impact on the intervention behavior of states which is an unexpected finding. In general, control variables have a bigger substantive impact on the decision to intervene than contextual variables which are derived from the conflict-oriented approach in the literature. The censored probit analyses conducted by using the entire dataset provide important results which lend support to many of the hypotheses in this study. Although these are important findings about the factors that motivate states to intervene militarily in civil conflicts, it is crucial to check the robustness of these 144 results in various sub-samples of the dataset. Moreover, while it is desirable to arrive at some generalizations regarding the rationale behind foreign military interventions, it is useful to analyze whether or not the same factors influence the intervention behavior of different types of states similarly. In other words, testing these hypotheses against different sub-samples of the dataset and comparing the substantive effect of variables on intervention behavior of different types of states facilitates a better understanding of this foreign policy behavior. As mentioned before, the robustness of findings from the entire dataset will be checked in three sub-samples. The following section tests the hypotheses of this study against major powers and the last section concludes this chapter by comparing the findings. Empirical Findings for Major Powers This section focuses on the empirical findings from the statistical analyses that were conducted within the sub-sample of major powers. The following sub-sections present the statistical results and interpret their statistical significance as well as substantive implications. Major powers consist of the United States, U.S.S.R/Russia, United Kingdom, France and China. The total number of observations for major powers in the dataset is 7468 and as a group, they are responsible for 35% of the military interventions according to Table 5.7. The number of interventions by each major power is shown in Table 5.8. 145 Table 5.7: Military interventions according to power status of interveners Table 5.8: Military interventions by each major power Major Power # of interventions % United States 371 45.9% U.S.S.R./Russia 183 22.6% United Kingdom 39 4.8% France 86 10.6% China 130 16.1% Total 809 100% Statistical Models The same three models are also used to test the hypotheses against major powers. These three censored probit models are presented in Tables 5.9, 5.10 and 5.11. As a result of randomly missing observations, the number of dyad years used in the selection equation of each censored probit model is 7317 which constitute 98% of the major power sub-sample. The number of uncensored observations, which refers to the number of dyad years with intervention, is 799 in each outcome equation. Thus, almost 99% of the major power interventions are used while estimating the outcome equations. Finally, examination of the correlation matrix shows no evidence of significant multicollinearity among the explanatory variables in each censored probit model. Total # of military interventions between 1946-2002 2285 100% Military interventions by major powers 809 35% Military interventions by minor powers 1,476 65% 146 Table 5.9: Model 1 - Censored Probit Results for Major Powers Selection Equation Outcome Equation Dependent Variable Intervention Intervention on govt. side Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ -.065 (.166) -.874* (.524) Militarized dispute ~ .592**** (.172) .062 (.286) Alliance ~ .599**** (.134) .805*** (.302) Historical ties .649**** (.189) .966**** (.236) Contiguous .566* (.316) - Same region .115 (.150) - Secondary diamonds .109 (.114) -.440 (.301) Petroleum .207* (.120) .513** (.226) Trade (logged) ~ .123**** (.031) .276**** (.058) Rival intervention ~ .607**** (.106) - Rival intervention on rebel side ~ - .636**** (.174) Ally intervention ~ .363**** (.111) - Ally intervention on govt. side ~ - .398* (.227) Democratic intervener ~ .080 (.121) -.243 (.215) Ethnic ties 1 -.440 (.387) - Ethnic ties 2 -.049 (.232) - Elite unrest ~ -.072 (.076) - Mass unrest ~ -.158*** (.060) - Democratic target ~ -.081 (.169) .038 (.241) Government conflict .067 (.112) -.088 (.207) Battle deaths (logged) ~ .147**** (.029) - Rebel fighting capacity ~ -.238*** (.092) - War involvement ~ -.011 (.083) - CINC ratio (logged) ~ .305**** (.042) - Cold War .173** (.089) - Spell years -1.326****(.099) -1.199****(.187) Constant -3.261**** (.416) -3.885**** (.859) N Total obs. = 7317 Uncensored obs. = 799 Wald Chi-square = 354.40**** Rho = .831 (.067) Wald test of independent equations (rho = 0), Chi-square = 11.83**** Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 147 Table 5.10: Model 2 - Censored Probit Results for Major Powers Selection Equation Outcome Equation Dependent Variable Intervention Intervention on govt. side Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ -.065 (.162) -.891* (.510) Militarized dispute ~ .593**** (.172) .053 (.291) Alliance ~ .601**** (.132) .815*** (.298) Historical ties .642**** (.189) .954**** (.236) Contiguous .558* (.316) - Same region .108 (.149) - Secondary diamonds .114 (.114) -.441 (.301) Petroleum .181 (.119) .485** (.224) Trade (logged) ~ .126**** (.032) .278**** (.058) Rival intervention ~ .597**** (.105) - Rival intervention on rebel side ~ - .632**** (.174) Ally intervention ~ .356**** (.110) - Ally intervention on govt. side ~ - .386* (.231) Democratic intervener ~ .091 (.119) -.242 (.215) Ethnic ties 1 -.456 (.386) - Ethnic ties 2 -.013 (.233) - Elite unrest ~ -.072 (.075) - Mass unrest ~ -.159*** (.060) - Democratic target ~ -.040 (.173) .067 (.243) Government conflict .049 (.114) -.105 (.209) Battle deaths (logged) ~ .138**** (.028) - Refugees (logged) ~ .030* (.018) - Rebel fighting capacity ~ -.233*** (.092) - War involvement ~ -.011 (.083) - CINC ratio (logged) ~ .303**** (.042) - Cold War .180** (.088) - Spell years -1.325****(.098) -1.193****(.187) Constant -3.486**** (.485) -4.086**** (.950) N Total obs. = 7317 Uncensored obs. = 799 Wald Chi-square = 352.84**** Rho = .822 (.067) Wald test of independent equations (rho = 0), Chi-square = 11.76**** Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 148 Table 5.11: Model 3 - Censored Probit Results for Major Powers Selection Equation Outcome Equation Dependent Variable Intervention Intervention on govt. side Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ -.064 (.165) -.918* (.527) Militarized dispute ~ .605**** (.176) .130 (.292) Alliance ~ .579**** (.133) .779*** (.294) Historical ties .664**** (.187) 1.036**** (.226) Contiguous .535* (.302) - Same region .124 (.148) - Secondary diamonds .109 (.114) -.461 (.299) Petroleum .204* (.120) .501** (.222) Trade (logged) ~ .129**** (.033) .297**** (.059) Rival intervention ~ .597**** (.104) - Rival intervention on rebel side ~ - .629**** (.180) Ally intervention ~ .371**** (.110) - Ally intervention on govt. side ~ - .436* (.233) Joint democracy ~ .004 (.178) .279 (.255) Joint autocracy ~ .209* (.124) .497** (.239) Ethnic ties 1 -.424 (.383) - Ethnic ties 2 -.041 (.230) - Elite unrest ~ -.081 (.076) - Mass unrest ~ -.149*** (.060) - Government conflict .067 (.111) -.079 (.212) Battle deaths (logged) ~ .148**** (.029) - Rebel fighting capacity ~ -.229*** (.091) - War involvement ~ -.009 (.084) - CINC ratio (logged) ~ .305**** (.042) - Cold War .171** (.088) - Spell years -1.320****(.098) -1.194****(.185) Constant -3.433**** (.409) -4.433**** (.826) N Total obs. = 7317 Uncensored obs. = 799 Wald Chi-square = 358.48**** Rho = .837 (.059) Wald test of independent equations (rho = 0), Chi-square = 11.92**** Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 149 In each of the censored probit analyses, the model chi-square rejects the null hypothesis that all coefficients in the model are zero. In the first model, the correlation between the two equations, rho, is 0.83 which is far greater than a zero correlation. The hypothesis that rho equals zero is rejected at the 0.1% level based on the Wald test of independent equations. The statistical findings regarding the correlation between the two equations are very similar in the other two models. The statistically significant rho across all three models show that the two dependent variables are linked and the use of a censored probit model is necessary in the case of the major power sub-sample as well. In order to make discussion easier, Table 5.12 summarizes the support the results lend to each hypothesis postulated for both dependent variables. The table presents whether or not each hypothesis is supported based on the statistical significance and direction of independent variables across all three models. Some of the hypotheses are also categorized as receiving relatively weaker support due to the statistical significance of results at the 90% level. The rest of the hypotheses are supported at higher significance levels which can be seen in Tables 5.9, 5.10 and 5.11. As shown in Table 5.12, a smaller number of hypotheses receive support in the major power sub-sample compared to the full sample. However, some of the statistical findings which support or reject the hypotheses for major powers are actually consistent with the results from the full sample. In other words, even though the differences between the two sets of findings are immediately noticeable since 150 fewer hypotheses are supported in the case of major powers, results from the major power sub-sample share some important similarities with the full sample as well. Table 5.12: Summary of Support for Hypotheses Hyp. Independent Variables Dependent Variables International Intervention Government side H1a, 1b Rivalry not supported weakly supported H2a, 2b Militarized dispute supported not supported H3a, 3b Alliance supported supported H4a, 4b Historical ties supported supported H5 Contiguous weakly supported na H6 Same region not supported na H7a, 7b Secondary diamonds not supported not supported H8a, 8b Petroleum weakly supported supported H9a, 9b Trade supported supported H10a Rival intervention supported na H10b Rival intervention on reb. side na supported H11a Ally intervention supported na H11b Ally intervention on govt. side na weakly supported Domestic H16a, 16b Democratic intervener not supported not supported H17a, 17b Joint democracy not supported not supported H17a, 17b Joint autocracy weakly supported supported H18a Ethnic ties 1 not supported na H18b Ethnic ties 2 not supported na H19 Elite unrest supported na H20 Mass unrest not supported na Contextual H21a, 21b Democratic target not supported not supported H22a, 22b Government conflict not supported not supported H23 Battle deaths supported na H24 Refugees weakly supported na H25 Rebel fighting capacity opposite found na Control Variables H12 War involvement not supported na H14 CINC ratio supported na H15 Cold War supported na 151 There are remarkable similarities between the major power sub-sample and the full sample in the case of hypotheses on international sources of intervention while the most significant differences between the two samples are in the case of domestic sources of intervention. Almost all of the same hypotheses on the international sources of intervention receive support in the major power sub-sample and in the full sample. The major differences between the two sets of findings are related to rivalry-alliance relations and geographic proximity with the conflict state. While rivalry with the conflict state increases the probability of intervention in the full sample as hypothesized, it is not a significant factor motivating major powers to intervene in a civil conflict. Instead, alliance with the conflict state seems to be more important for major powers than rivalry. Even though alliance and rivalry relations have different impacts in the full sample and in the major power sub-sample, these factors affect the decision on whose side to intervene similarly in both samples as indicated by Tables 5.4 and 5.12. The second major difference in the major power sub-sample has to do with geographic proximity. Being in the same region with the conflict state does not affect the probability of major power intervention while this is a statistically significant factor in the full sample. Given that major powers have the capacity to intervene in civil conflicts taking place in different parts of the world, this is not a surprising finding. Finally, the hypothesis on secondary diamonds is not supported in both samples, but the existence of secondary diamonds is also not important when major powers choose sides in a civil conflict whereas in the full sample secondary diamonds matter for the side decision. 152 There are also important similarities between the two sets of findings with respect to the contextual and control variables. The only two contextual variables that seem to increase the probability of major power intervention are the number of battle deaths and refugees per conflict year which is consistent with the findings from the full sample. Although the hypothesis on rebel fighting capacity is not supported in the major power sample, the results are still consistent with the full sample because increasing rebel fighting capacity decreases the likelihood of major power intervention as well. The hypothesis on the ratio of capabilities is supported in both sets of analyses, but involvement in an interstate war does not have a statistically significant impact on the intervention behavior of major powers probably due to their greater ability compared to minor powers to become militarily involved in more than one conflict. The most important difference between the major power sub-sample and the full sample has to do with the impact of Cold War on intervention probability. Cold War increases the probability of military intervention in the case of major powers whereas Cold War does not have a statistically discernible effect in the full sample. Thus, the results from the major power sub-sample show that the bipolar structure of international system influenced the intervention behavior of major powers more than the rest of the countries. The role of domestic variables in the intervention decision of major powers seems to be very different. In the full sample, all of the domestic explanatory variables except joint democracy had a statistically significant impact on the likelihood of military intervention whereas in the case of major powers only two hypotheses on domestic sources receive support. Joint autocracy increases the 153 probability of major power intervention while mass unrest decreases it. In terms of the second dependent variable, joint autocracy also has a positive relationship with pro-government intervention. Since none of the other explanatory factors receive support aside from these two domestic variables, it appears that domestic sources of foreign policy behavior are less critical for the intervention decision of major powers compared to international factors. However, it is still necessary to compare the substantive impact of these variables before arriving at such a conclusion. Interpretation of Results After reviewing the censored probit results from the major power sub-sample in terms of the support they provide to the hypotheses, it is now useful to discuss the substantive impact of statistically significant variables on the two dependent variables. Table 5.13 presents the changes in the predicted probability of intervention=1 when the variable of interest increases by one unit or some number of units as indicated in the table. First differences are calculated for the statistically significant variables from all three models, but in order to avoid repetition they are only computed once. Similarly, Table 5.14 presents the first differences for statistically significant variables from the outcome equations. As in the full sample, most of the same international factors have the largest substantive effect on the intervention decision of major powers according to Table 5.13. Historical ties increase the probability of major power intervention by approximately 24 percent while intervention by a rival state raises it by almost 22 percent. Both of these findings are consistent with the findings from the full sample but the substantive effects of these international factors are higher for major powers. 154 Alliance with the conflict state increases the likelihood of military intervention by 21 percent which is an important finding in the case of major powers because alliance does not have a statistically significant effect in the full sample. Militarized dispute with a conflict state and contiguity have the next largest substantive impact as can be seen in Table 5.13. However, even though militarized dispute with the conflict state increases the probability of intervention by 20 percent, it does not have a discernible effect on major powers? decision on whose side to intervene in a civil conflict as indicated by Table 5.12. Finally, interventions by allies have a smaller impact, less than 10 percent, on the intervention decision of major powers than interventions by rivals. The two contextual variables, battle deaths and refugees, have the smallest substantive effect on the probability of major power intervention along with trade ties. On the other hand, the third contextual variable decreases the probability of intervention by 6 percent as rebels? fighting capacity relative to the government forces increases. According to Table 5.13, rebel fighting capacity has a greater impact on the intervention decision of major powers than domestic variables because joint autocracy increases intervention likelihood by less than 6 percent while mass unrest decreases it by 3.4 percent. Therefore, compared to the results from the full sample, domestic factors have far less impact on the intervention probability of major powers in terms of both the number of statistically significant domestic variables as well as their relative substantive effect. 155 Table 5.13: Changes in the Predicted Probability of Intervention When this variable moves from ? to ? in Change in probability of intervention 95% Confidence Interval (unless indicated by *) Model 1: Militarized dispute 0 ? 1 20.04% [9.11% 30.97%] Alliance 0 ? 1 20.78% [9.85% 31.71%] Historical ties 0 ? 1 23.72% [11.64% 35.80%] Contiguous 0 ? 1 16.21% [3.56% 28.86%] * Petroleum 0 ? 1 5.56% [1.39% 9.73%] * Trade p25 ? p75 Trade min ? max 0.97% 2.12% [0.45% 1.49%] [1.08% 3.16%] Rival intervention 0 ? 1 21.59% [10.72% 32.46%] Ally intervention 0 ? 1 9.63% [4.84% 14.42%] Mass unrest 0 ? 1 -3.43% [-5.04% -1.82%] Battle deaths p25 ? p75 Battle deaths min ? max 1.49% 2.77% [0.86% 2.12%] [1.95% 3.59%] Rebel fighting capacity 1 ? 2 Rebel fighting capacity 1 ? 3 -4.41% -6.03% [-7.76% -1.06%] [-9.67% -2.39%] CINC ratio p25 ? p75 CINC ratio min ? max 5.28% 7.87% [2.36% 8.20%] [3.22% 12.52%] Cold War 0 ? 1 4.74% [2.25% 7.23%] Model 3: Joint autocracy 0 ? 1 5.82% [1.68% 9.96%] * Model 2: Refugees p25 ? p75 Refugees min ? max 0.48% 1.03% [0.22% 0.74%] * [0.44% 1.62%] * Notes: This table reports the first differences for statistically significant variables from the selection equations in Tables 5.9, 5.10 and 5.11. The probabilities are calculated by holding all continuous variables at their means and all dichotomous and categorical variables at their modes. * indicates 90% confidence interval. The ratio of capabilities has a bigger impact on the intervention probability of major powers compared to its impact in the full sample. In fact, this control variable has the largest substantive effect after the group of international variables with the most critical impact on military intervention behavior. Finally, the probability of major power intervention is almost 5 percent higher during the Cold War period. Although this may not be a large substantive effect, it nevertheless indicates that the 156 structure of the international system plays a role in the intervention decision of major powers. There are some interesting results in terms of the substantive effects of variables on choosing sides in a civil conflict. Table 5.14 shows that historical ties with the conflict state increase the probability of intervening on the government side by more than 42 percent. After historical ties, rivalry and alliance with the conflict state have the highest substantive impact on the side decision. Although rivalry does not affect the probability of major power intervention, it does decrease the probability of intervening on the government side by almost 37 percent once major powers decide to intervene in a civil conflict. Alliance with the conflict state, on the other hand, increases the likelihood of pro-government intervention by 32 percent. Table 5.14: Changes in the Predicted Probability of Government Intervention When this variable moves from ? to ? in Change in probability of govt. intervention 95% Confidence Interval (unless indicated by *) Model 1: Rivalry 0 ? 1 -36.62% [-64.08% -9.16%] * Alliance 0 ? 1 32.03% [17.95% 46.11%] Historical ties 0 ? 1 42.48% [22.36% 62.60%] Petroleum 0 ? 1 13.79% [4.62% 22.96%] Trade p25 ? p75 Trade min ? max 3.86% 5.93% [1.54% 6.18%] [2.87% 8.99%] Rival int. rebel side 0 ? 1 21.36% [10.24% 32.48%] Ally int. govt. side 0 ? 1 9.97% [1.28% 18.66%] * Model 3: Joint autocracy 0 ? 1 12.84% [4.16% 21.52%] Notes: This table reports the first differences for statistically significant variables from the outcome equations in Tables 5.9, 5.10 and 5.11. The probabilities are calculated by holding all continuous variables at their means and all dichotomous and categorical variables at their modes. * indicates 90% confidence interval. 157 In addition to the strategic significance of the conflict state, strategic interactions with the other interveners also have a critical impact on how major powers choose sides in a civil conflict. Previous intervention by a rival state on the rebel side increases the probability of intervening on the government side by more than 20 percent which is more than twice the size of the substantive impact of pro- government interventions by allies. Similar to the results from the full sample, economic significance of the conflict country has a more crucial impact with respect to the decision on whose side to intervene than on the probability of intervention in the major power sub-sample. For instance, trade ties have one of the smallest substantive impacts on the probability of major power intervention but it increases the probability of intervening on the government side by almost 6 percent as the amount of trade reaches maximum. However, this is still a smaller substantive effect compared to the larger influence of other international factors. Table 5.14 indicates that the existence of petroleum in the conflict state can increase the probability of pro- government intervention by almost 14 percent, which is a quite significant substantive effect. As mentioned in the previous sub-section, the only regime related variable that has a critical impact on the side decision of major powers is joint autocracy. Autocratic major powers are almost 13 percent more likely to support the government side when they intervene in a civil conflict taking place in an autocratic state. In fact, among the variables related to regime type, joint autocracy is the only one that has an impact on major powers? intervention decisions which is quite different than the findings in the full sample. 158 Discussion of Findings from the Major Power Sub-Sample The censored probit analyses conducted in the major power sub-sample provide an improved understanding of the factors that influence major powers? decisions on whether and on whose side to intervene in civil conflicts. First and foremost, international factors play a more dominant role in the decision making of major powers compared to domestic and contextual factors. The strategic importance of the conflict state and interventions by rivals or allies which alter the strategic significance of the civil conflict as perceived by the major power have the largest substantive impact on the intervention probability of major powers. There are numerous historical examples where one can notice the impact of historical ties, rivalry and alliance relations with other interveners as well as economic factors on the intervention behavior of major powers in civil conflicts. For instance, France has frequently intervened militarily in the civil conflicts of its former colonies including Mauritania, Morocco, Lebanon and Chad. All of these military interventions were also on the government side as hypothesized. However, some of these examples show that there were other factors at play in addition to historical ties. In both Chad and Mauritania, French military interventions also took place in reaction to Libya?s military interventions on the rebel side because Libya was an interstate rival of France. Moreover, France has important trade ties with some of these countries, particularly Chad. Historical ties were also important in Russia?s military intervention on the government side in Tajikistan?s civil conflict between 1992 and 1998. However, Russia?s pro-government intervention was also motivated by its alliance with the government as well as its strong trade ties with the country which 159 also happens to have petroleum. All of these factors predict Russia?s intervention on the government side. Another example is Britain?s military intervention on the government side in Malaysia during 1963-1966. But in addition to historical ties, pro- rebel intervention by Indonesia which was an inter-state rival of Britain was another motivating factor for this military intervention. Of course, alliance and rivalry relations with other interveners had a significant impact on intervention decisions of the superpowers during the Cold War as documented in many case studies. Contiguity is also important for major powers? intervention behavior as demonstrated by Russia?s interventions in Afghanistan, Georgia and Iran or China?s interventions in Laos, Myanmar and India. Among the international factors that affect the probability of major power intervention, perhaps the more surprising finding is the importance of alliance ties with the conflict state which is not a significant factor in the full sample analysis. While rivalry with the conflict state does not necessarily motivate major powers to intervene, alliance ties play a critical role for major powers while deciding to intervene in a civil conflict. Protecting allied governments appears to be a more important motivation than removing rival governments. However, as the findings from the outcome equation show rivalry can be a deciding factor while choosing sides after major powers decide to intervene. This finding also shows the important contribution of selection models to the study of interventions because variables which may not be critical in the selection equation may have an important impact on how states choose sides in a civil conflict. 160 Another important finding from the major power sub-sample is the smaller effect of domestic factors on intervention probability of major powers. Joint autocracy and mass unrest are the only significant domestic variables in this sub- sample. Thus, international factors, including control variables such as the ratio of capabilities and Cold War, are more crucial for understanding intervention decisions of major powers. Conclusion Overall, the findings from the full sample and major power sub-sample show significant support for a number of the hypotheses derived from modified realism. However, there are both important similarities and differences between the two sets of analyses. While the similarities indicate the robustness of some findings, the differences show that sub-sample analyses can be quite useful for a nuanced understanding of the intervention behavior of different types of states. In both sets of analyses, international variables are the most important factors motivating states to intervene in civil conflicts compared to domestic, contextual and control variables. Historical ties, contiguity, interventions by rivals or allies, militarized dispute with the conflict state have the most significant impact on the probability of military intervention. However, rivalry and alliance with the conflict state appear to have different impacts in the full sample and in the major power sub- sample. The same group of international factors also influences how states, including major powers, choose sides in civil conflicts. Another similarity has to do with the role of contextual variables in both sets of analyses. Changing conflict intensity, as measured by battle deaths and refugees, affects intervention likelihood in both 161 samples, but its effect is much smaller compared to other variables. The only exception is the effect of rebel fighting capacity on intervention probability because this variable is even more critical than some domestic considerations of leaders, such as mass or elite unrest, for deciding whether or not to intervene in a civil conflict. The biggest difference between the two sets of analyses comes from the domestic sources of intervention. Ethnic ties, regime related variables and both types of domestic unrest influence the probability of intervention in the full sample. In fact, ethnic ties have a quite large substantive effect on the intervention decision. There are also important findings from the full sample about the intervention propensity of democracies compared to autocracies. Moreover, regime characteristics of both the intervener and the conflict state are the most crucial group of factors for choosing sides in a civil conflict after the international factors. The major power sub-sample shows that the only regime related factor that matters for the intervention decision is joint autocracy. Autocratic major powers, which are U.S.S.R/Russia and China, are more likely to intervene on the side of autocratic governments in civil conflicts while joint democracy does not have the same impact for the other three major powers. However, ethnic ties and elite unrest do not seem to affect the intervention probability of any of the major powers. Thus, analyses from the major power sub-sample conclude that domestic factors do not play a critical role in the intervention decision of major powers and this finding proves how useful sub-sample analyses can be. Additional benefits of analyzing the major power sub-sample include discerning the impact of Cold War on major powers? probability of intervention in civil conflicts. Finally, both sets of analyses demonstrate the close relationship 162 between the decisions on whether and on whose side to intervene and show how same factors can have different substantive impacts on each dependent variable. The use of selection models also facilitates a nuanced understanding of certain factors, such as the economic significance of the conflict state, for the intervention decision in both the full sample and in the major power sub-sample. To sum up, the similarities between the full sample and major power sub- sample indicate the robustness of some findings. The differences, on the other hand, provide an improved understanding of some factors that motivate major powers particularly. The study also aims to explore the similarities and differences between democracies and autocracies which have not been analyzed adequately in the literature. Moreover, since categorizing states based on regime type splits the dataset into two sub-samples, the analyses in the next chapter will be even more useful for checking the robustness of findings from the full sample. 163 Chapter 6: Comparing Democracies and Autocracies This chapter examines the similarities and differences between democratic and autocratic third parties regarding their decisions on whether and on whose side to intervene in civil conflicts. While the analyses from the full sample indicate that regime type and regime similarity influence states? intervention behavior, they are not able to show whether the remaining international, domestic and contextual factors affect the intervention decisions of democratic and autocratic leaders differently. The motives of potential interveners based on their regime type have not been explored sufficiently in the literature. In fact, there are only a small number of quantitative studies that specifically examine how democracies and autocracies differ in terms of their intervention behavior in civil conflicts (i.e. Hermann and Kegley 2001, Koga 2011). Therefore, a comparative analysis of the factors that motivate democracies and autocracies to intervene militarily in civil conflicts can provide important theoretical and empirical insights. The purpose of this chapter is twofold. The first purpose is to provide a thorough understanding of the factors that affect intervention decisions of democratic and autocratic leaders. A comparative analysis will shed light on the use of this foreign policy instrument by democracies and autocracies which has not been adequately analyzed in the context of civil conflicts yet. The second purpose is to check the robustness of findings from the full sample in two sub-samples. Since categorizing states based on their regime type splits the dataset into two sub-samples, 164 the analyses in this chapter will be particularly useful for checking the robustness of findings from the full sample. Split sample analyses are becoming more common in the literature for doing a robustness check and for comparing the effect of predictors in theoretically important sub-samples of the data (i.e. Doyle and Sambanis 2000, Heger and Salehyan 2007, Kathman 2010). This chapter is divided into three sections. The first section presents the results of the statistical analyses that were performed using the sub-sample of democratic third parties while the second section reports the results from the statistical analyses conducted in the sub-sample of autocratic third parties. Results are interpreted in light of the hypotheses postulated in Chapter 3 as well as in terms of their substantive effects. The final section of the chapter concludes by discussing the theoretical significance of the findings from the first two sections. Empirical Findings for Democracies This section focuses on the empirical findings from the statistical analyses that were conducted within the sub-sample of democracies. The following sub-sections present the statistical results, interpret their statistical significance and discuss their theoretical and substantive implications. Table 6.1 shows the number of observations in the dataset based on the regime type of potential interveners while Table 6.2 presents the number of military interventions conducted by democratic and autocratic states. 30% of the 2285 military interventions were carried out by democracies which make up 32% of the dataset. Autocracies, which constitute 68% of the dataset, were responsible for the remaining 70% of the military interventions according to Table 6.2. 165 Table 6.1: Observations according to the regime type of potential interveners Table 6.2: Military interventions according to the regime type of interveners Statistical Models Two different censored probit models are used to test the hypotheses against democracies in the dataset. None of the models include the regime type variable for potential interveners since they are all democratic. The first censored probit model includes all the remaining explanatory variables in this study except the refugees variable. As a result of substituting one third of the refugees data which were missing with the mean value, this study takes a cautious approach and runs a separate model which adds the refugees variable to the first censored probit model. The two censored probit models are presented in Tables 6.3 and 6.4. Due to randomly missing observations, the number of dyad years used in the selection equation of each censored probit model is 62,768 which constitute more than 96% of the democratic sub-sample. The number of uncensored observations, which refers to the number of dyad years with intervention, is 664 in each outcome equation. Thus, 98% of the interventions by democracies are used while estimating the outcome equations. Finally, examination of the correlation matrix from both analyses shows Total # of observations (excluding missing values) 206,047 100% Observations by democratic states 65,193 32% Observations by autocratic states 140,854 68% Total # of military interventions between 1946-2002 2285 100% Military interventions by democratic states 675 30% Military interventions by autocratic states 1,610 70% 166 little reason for concern about multicollinearity among the explanatory variables in each censored probit model. The highest absolute level of correlation is between the variables conflict type (as indicated by government conflict) and battle deaths at 0.494, while the rest of the correlations are much smaller. In both of the censored probit analyses, the model chi-square rejects the null hypothesis that all coefficients in the model are zero. In the first model, the correlation between the two equations, rho, is 0.81 which is far greater than a zero correlation. The hypothesis that rho equals zero is rejected at the 5% level based on the Wald test of independent equations. The statistical findings regarding the correlation between the two equations are very similar in the second model. Since both analyses show the existence of selection bias and indicate that the two decisions are linked, the use of a censored probit model is appropriate for the sub-sample of democracies as well. Table 6.5 summarizes the support the results lend to each hypothesis postulated for both dependent variables to make discussion easier. The table presents whether or not each hypothesis is supported based on the statistical significance and direction of the independent variables in both models. If a variable is statistically significant but its effect is the opposite of what was hypothesized, then Table 6.5 reports that finding as well. Some hypotheses are also categorized as receiving relatively weaker support due to the statistical significance of results at the 90% level. The rest of the hypotheses are supported at higher significance levels as can be seen in Tables 6.3 and 6.4. 167 Table 6.3: Model 1 - Censored Probit Results for Democratic states Selection Equation Outcome Equation Dependent Variable Intervention Intervention on govt. side Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ .598*** (.225) -.014 (.275) Militarized dispute ~ .624**** (.190) -.515* (.288) Alliance ~ .164 (.141) .145 (.311) Historical ties .782**** (.181) .956**** (.216) Contiguous .652**** (.164) - Same region .355**** (.106) - Secondary diamonds -.230 (.239) -.383 (.251) Petroleum .194* (.108) .020 (.196) Trade (logged) ~ .112**** (.022) .237**** (.047) Rival intervention ~ .570**** (.099) - Rival intervention on rebel side ~ - .858**** (.226) Ally intervention ~ .190* (.110) - Ally intervention on govt. side ~ - .405* (.245) Ethnic ties 1 .237** (.106) - Ethnic ties 2 .334* (.204) - Elite unrest ~ -.038 (.062) - Mass unrest ~ -.083 (.057) - Democratic target ~ .077 (.153) .453** (.216) Government conflict .091 (.116) -.514* (.297) Battle deaths (logged) ~ .152**** (.029) - Rebel fighting capacity ~ -.176** (.083) - War involvement ~ -.232** (.098) - Major power .571**** (.099) - CINC ratio (logged) ~ .240**** (.029) - Cold War -.007 (.074) - Spell years -1.031****(.093) -.808****(.168) Constant -3.758**** (.219) -3.429**** (.608) N Total obs. = 62768 Uncensored obs. = 664 Wald Chi-square = 260.27**** Rho = .814 (.094) Wald test of independent equations (rho = 0), Chi-square = 6.24** Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 168 Table 6.4: Model 2 - Censored Probit Results for Democratic states Selection Equation Outcome Equation Dependent Variable Intervention Intervention on govt. side Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ .602*** (.222) -.014 (.286) Militarized dispute ~ .609**** (.187) -.508* (.285) Alliance ~ .180 (.138) .145 (.317) Historical ties .766**** (.180) .951**** (.218) Contiguous .656**** (.161) - Same region .342**** (.105) - Secondary diamonds -.220 (.233) -.391 (.270) Petroleum .156 (.106) .023 (.185) Trade (logged) ~ .119**** (.022) .238**** (.048) Rival intervention ~ .547**** (.099) - Rival intervention on rebel side ~ - .839**** (.229) Ally intervention ~ .182* (.109) - Ally intervention on govt. side ~ - .391 (.258) Ethnic ties 1 .243** (.114) - Ethnic ties 2 .375* (.209) - Elite unrest ~ -.039 (.062) - Mass unrest ~ -.075 (.056) - Democratic target ~ .149 (.159) .493** (.221) Government conflict .064 (.113) -.503* (.297) Battle deaths (logged) ~ .138**** (.031) - Refugees (logged) ~ .045** (.018) - Rebel fighting capacity ~ -.169** (.083) - War involvement ~ -.225** (.097) - Major power .560**** (.098) - CINC ratio (logged) ~ .241**** (.029) - Cold War -.008 (.073) - Spell years -1.036****(.092) -.800****(.173) Constant -4.135**** (.272) -3.501**** (.745) N Total obs. = 62768 Uncensored obs. = 664 Wald Chi-square = 263.69**** Rho = .819 (.091) Wald test of independent equations (rho = 0), Chi-square = 6.37** Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 169 Table 6.5: Summary of Support for Hypotheses Hyp. Independent Variables Dependent Variables International Intervention Government side H1a, 1b Rivalry supported not supported H2a, 2b Militarized dispute supported weakly supported H3a, 3b Alliance not supported not supported H4a, 4b Historical ties supported supported H5 Contiguous supported na H6 Same region supported na H7a, 7b Secondary diamonds not supported not supported H8a, 8b Petroleum weakly supported not supported H9a, 9b Trade supported supported H10a Rival intervention supported na H10b Rival intervention on reb. side na supported H11a Ally intervention weakly supported na H11b Ally intervention on govt. side na weakly supported Domestic H18a Ethnic ties 1 supported na H18b Ethnic ties 2 weakly supported na H19a Elite unrest not supported na H20 Mass unrest not supported na Contextual H17a, 17b Democratic target not supported supported H22a, 22b Government conflict not supported opposite found H23 Battle deaths supported na H24 Refugees supported na H25 Rebel fighting capacity opposite found na Control Variables H12 War involvement supported na H13 Major power supported na H14 CINC ratio supported na H15 Cold War not supported na Table 6.5 shows that there are a lot of similarities between the democratic sub- sample and full sample in terms of the hypotheses that receive support in the selection equations. On the other hand, more striking differences are observed between the two samples with respect to the outcome equation. A smaller number of hypotheses regarding the decision on whose side to intervene receive support in the sub-sample 170 of democracies which indicates that some of the factors that are critical in the full sample do not play a role when democratic interveners are choosing sides in a civil conflict. All hypotheses on the international sources of intervention decisions receive support in the democratic sub-sample with the exception of alliance ties and the existence of secondary diamonds in the conflict state. Factors related to the strategic significance of the conflict state which include historical ties, contiguity, geographic region, hostile relations with the government, trade ties and the existence of petroleum in the conflict state have the hypothesized relationship with intervention probability. Interventions by rivals and allies also increase the likelihood of military intervention by changing the perceived strategic significance of the civil conflict. Similar to the full sample, alliance ties with the conflict state do not affect the probability of intervention by democracies. However, while alliance increases the probability of intervening on the government side in the full sample, it does not have any effect on how democratic interveners choose sides if they become militarily involved in the civil conflict. The existence of secondary diamonds in the conflict state also does not affect the intervention likelihood of democracies. Moreover, both lootable and non-lootable resources do not seem to be critical when democracies are choosing sides in a civil conflict whereas in the full sample both resources influence the side decision. The only observable effect in terms of the resource wealth of the conflict state is petroleum which increases the probability of military intervention by democratic third parties. Although the findings from the selection equation regarding alliance and resource wealth are consistent with the full sample, the results from the 171 outcome equation are quite different since neither alliance ties nor types of resources in the conflict state appear to be significant for how democracies choose sides when they intervene in a civil conflict. Moreover, rivalry with the conflict state also does not influence the side decision but it increases the probability of intervention in the democratic sub-sample similar to the full sample. The hypotheses on contextual variables which are supported in the full sample also receive support in the democratic sub-sample. Increasing number of battle deaths and refugees increase the probability of intervention by democratic states whereas higher rebel fighting capacity decreases it. Similar to the full sample, regime type of the conflict state does not affect the likelihood of intervention in the democratic sub- sample; but if a democratic state intervenes, it is more likely to support the government side when the conflict state is democratic. Finally, the type of civil conflict also does not have a statistically significant impact on the intervention probability of democracies. However, the type of conflict influences the side decision in a surprising way when democracies intervene in intrastate disputes. Democracies are less likely to intervene on the government side when it is a governmental conflict and more likely to support the government if it is a territorial conflict. Although this relationship is the opposite of what was hypothesized, it is an interesting finding which indicates that democracies are less likely to give military support to rebels fighting for territorial autonomy or secession. One of the biggest differences between the full sample and the democratic sub-sample comes from the domestic sources of intervention. While ethnic ties increase the likelihood of military intervention in both samples, neither type of 172 domestic unrest appears to have an impact on democracies? decision to intervene in a civil conflict. Based on the theoretical arguments in the literature discussed in Chapter 3, mass unrest was hypothesized to constrain democratic leaders? intervention behavior while elite unrest was expected to compel them to resort to diversionary foreign policy behavior. Although neither of these hypotheses is supported in the censored probit analysis, the findings are consistent with Kisangani and Pickering?s (2007) conclusion that democracies do not resort to biased military intervention in civil conflicts as diversionary foreign policy behavior. Kisangani and Pickering argue that democracies resort to benevolent missions abroad only when they face elite unrest domestically, but decisions on military interventions in civil conflicts which the authors categorize as ?high politics? related missions are not influenced by elite or mass unrest (2007, pp. 281, 284-285, 295). Therefore, this study confirms the findings by Kisangani and Pickering (2007) that neither mass nor elite unrest has a significant impact on the probability of military intervention by democracies. The potential impact of domestic unrest on democratic leaders? foreign policy decision making is more likely to be observed on other types of foreign policy behavior as Kisangani and Pickering argue. All of the control variables have the hypothesized effect on the probability of intervention by democratic third parties except Cold War which is not statistically significant in the full sample as well. Involvement in an interstate war constrains military intervention behavior of democracies. Democratic major powers are also more likely to intervene in civil conflicts than minor powers. Finally, increasing ratio 173 of capabilities increases the likelihood of military intervention in the democratic sub- sample. As an additional robustness check, this study also employed rare events logit to analyze the first dependent variable. The results from rare events logit analyses conducted in the sub-sample of democracies are consistent with the results from the first stage of censored probit analyses which provide additional confidence in the findings on the hypothesized relationships. Appendix D provides the results from relogit models. Interpretation of Results After reviewing the censored probit results in terms of the support they provide to the hypotheses in the sub-sample of democracies, it is now useful to discuss the substantive impact of the statistically significant variables on the two dependent variables. Tables 6.6 presents the changes in the predicted probability of intervention=1 when the variable of interest increases by one unit or some number of units depending on the nature of the variable. First differences are only calculated for the statistically significant variables from the first model except for the refugees variable which is computed from the second model. Table 6.7 indicates the changes in the predicted probability of pro-government intervention=1 when the variable of interest changes by one unit or some number of units as reported in the table. The substantive impact of independent variables reported in Tables 6.6 and 6.7 will be discussed separately for each dependent variable. 174 Table 6.6: Changes in the Predicted Probability of Intervention When this variable moves from ? to ? in Change in probability of intervention 95% Confidence Interval (unless indicated by *) Model 1: Rivalry 0 ? 1 17.54% [7.36% 27.72%] Militarized dispute 0 ? 1 18.09% [8.97% 27.21%] Historical ties 0 ? 1 19.86% [10.08% 29.64%] Contiguous 0 ? 1 18.49% [9.41% 27.57%] Same region 0 ? 1 12.78% [5.14% 20.42%] Petroleum 0 ? 1 6.52% [1.25% 11.79%] * Trade p25 ? p75 Trade min ? max 2.24% 3.17% [0.92% 3.56%] [1.43% 4.91%] Rival intervention 0 ? 1 16.94% [7.78% 26.10%] Ally intervention 0 ? 1 6.29% [1.03% 11.55%] * Ethnic ties 1 0 ? 1 9.43% [3.84% 15.02%] Ethnic ties 2 0 ? 1 12.19% [3.57% 20.81%] * Battle deaths p25 ? p75 Battle deaths min ? max 3.40% 4.85% [1.16% 5.64%] [1.98% 7.72%] Rebel fighting capacity 1 ? 2 Rebel fighting capacity 1 ? 3 -3.68% -5.36% [-5.89% -1.47%] [-8.53% -2.19%] War involvement 0 ? 1 -8.97% [-13.82% -4.12%] Major power 0 ? 1 17.02% [9.58% 24.46%] CINC ratio p25 ? p75 CINC ratio min ? max 6.77% 9.90% [3.34% 10.20%] [4.73% 15.07%] Model 2: Refugees p25 ? p75 Refugees min ? max 0.86% 1.41% [0.39% 1.33%] [0.64% 2.18%] Notes: This table reports the first differences for statistically significant variables from the selection equations in Tables 6.3 and 6.4. The probabilities are calculated by holding all continuous variables at their means and all dichotomous and categorical variables at their modes. * indicates 90% confidence interval. Similar to the full sample, international factors have the largest substantive impact on the probability of intervention in the sub-sample of democracies. Historical ties with the conflict state increase the likelihood of military intervention almost 20 percent while contiguity raises it 18.5 percent. Democratic states that have a militarized interstate dispute with the conflict state are also 18 percent more likely to intervene militarily. Rivalry with the conflict state as well as interventions by rival 175 states increase the probability of military intervention by 17.5 percent and 17 percent respectively. Finally, major powers are also 17 percent more likely to intervene than minor powers in the sub-sample of democracies. The substantive effects of international factors in the sub-sample of democracies are quite similar to the results from the full sample presented in Table 5.5. There are only two major differences between the two sets of findings. First, previous interventions by allies increase the probability of intervention by 6 percent in the sub-sample of democracies as opposed to 12 percent in the full sample. Second, the substantive effect of trade ties is twice as much in the sub-sample of democracies which indicates that trade plays a bigger role when democracies are deciding to intervene in civil conflicts. In the full sample, the impact of trade is 1.5 percent whereas in the democratic sub-sample trade ties increase the probability of intervention by more than 3 percent as it changes from minimum to maximum. Petroleum, another variable measuring the economic significance of the conflict state, has similar substantive effects in both samples. Ethnic ties with the dominant and second largest ethnic group in the conflict state increase the probability of intervention by 9 percent and 12 percent respectively. However, compared to the full sample, the substantive effect of ethnic ties is smaller in the sub-sample of democracies. On the other hand, control variables have a larger impact in the case of democracies. For instance, as the ratio of capabilities changes from minimum to maximum, the likelihood of military intervention increases by almost 10 percent. 176 Finally, contextual variables have a smaller substantive effect in the sub- sample of democracies compared to the full sample. When rebel fighting capacity is high, democracies are 5.4 percent less likely to intervene whereas in the full sample this substantive effect is almost 7 percent. The difference between the substantive impacts of conflict intensity indicators in the two samples is quite minor, but it indicates that intervention decisions of democracies are less affected by the number of battle deaths and refugees compared to the full sample. Table 6.7: Changes in the Predicted Probability of Government Intervention When this variable moves from ? to ? in Change in probability of govt. intervention 95% Confidence Interval (unless indicated by *) Model 1: Militarized dispute 0 ? 1 -24.86% [-40.98% -8.74%] * Historical ties 0 ? 1 41.63% [22.39% 60.87%] Trade p25 ? p75 Trade min ? max 7.39% 10.08% [2.87% 11.91%] [4.32% 15.84%] Rival int. rebel side 0 ? 1 35.34% [16.62% 54.06%] Ally int. govt. side 0 ? 1 16.59% [5.45% 27.73%] * Democratic target 0 ? 1 17.74% [6.96% 28.52%] Government conflict 0 ? 1 -19.42% [-32.67% -6.17%] * Notes: This table reports the first differences for statistically significant variables from the outcome equation in Table 6.3. The probabilities are calculated by holding all continuous variables at their means and all dichotomous and categorical variables at their modes. * indicates 90% confidence interval. Table 6.7 shows that international factors also have the largest substantive impact on whose side democracies intervene in civil conflicts. When democratic third parties have historical ties to the conflict state, they are almost 42 percent more likely to intervene on the government side. Democracies are also 35 percent more likely to support the government side militarily when their rivals intervene on the rebel side. Both of these variables have the biggest substantive impact in the full sample as well. 177 Having a militarized dispute with the conflict state decreases pro-government intervention probability almost 25 percent and has the third largest effect in the sub- sample of democracies whereas in the full sample it has one of the smallest effects with less than 10 percent. Contextual variables have the next largest impact in the democratic sub- sample. When the conflict type is governmental, democracies are 19.4 percent less likely to support the government side and more likely to support the rebel side. Another way to state this substantive effect is democracies are 19.4 percent less likely to intervene on the rebel side in territorial conflicts. Finally, the probability of pro- government intervention increases by almost 18 percent when the conflict state is democratic. The remaining two variables with a substantive impact on how democracies choose sides in civil conflicts are interventions by allies and trade ties. Democracies are 16.6 percent more likely to support the government when their allies intervene on the government side. However, as mentioned above, the substantive impact of rival interventions on the side decision is twice as much. Trade has the smallest impact on pro-government intervention with 10 percent, but trade ties still have a bigger impact in the democratic sub-sample compared to the full sample. Discussion of Findings for Democracies The censored probit analyses conducted in the sub-sample of democratic states indicate the robustness of a lot of the findings from the full sample. However, they also provide an improved understanding of the specific factors that influence democratic leaders? decision making on interventions. Since some of the hypotheses 178 in this study have not been tested specifically against democracies before, a few of the findings are quite novel whereas others are theoretically and empirically consistent with the previous findings in the literature. International factors have the most significant impact on intervention decision making in the sub-sample of democracies. Similar to the findings from the full sample, factors related to the strategic significance of the conflict state, which include historical ties, contiguity and hostile relations, motivate democratic leaders to intervene in civil conflicts more than others. The findings in this sub-sample show once more that rivalry and militarized disputes with the conflict state are crucial for intervention decisions of states, including democracies, even though these have not been analyzed adequately in quantitative studies of intervention before. Interventions by rivals also increase democratic leaders? willingness to intervene by making civil conflicts strategically important whereas interventions by allies have a much less critical impact on democracies? intervention probability compared to the full sample. Thus, balancing or restraining a rival?s capacity to affect the dynamics or outcomes of civil conflicts motivates democracies more than supporting allies. However, when democracies intervene in a civil conflict, they tend to bandwagon on the same side with their allies. Some of these factors can be observed in the following examples. One of the countries that provided military support to the Kurdish rebel group, PKK, during the civil conflict in Turkey was Greece which was not only contiguous to Turkey but also had rivalry with this country. Similarly, rivalry with Iraq influenced Israel?s decision to support the Kurdish rebel group, KDP, militarily for more than a decade during the 179 60s and early 70s. Historical ties motivated Italy to intervene militarily on the government side in Somalia in 1982 and Belgium to intervene several times in the DRC/Zaire in the 1960s. Rival military interventions by the U.S.S.R. and Cuba on the government side caused the U.S. to intervene on the rebel side in Nicaragua?s civil conflict throughout the 1980s. In addition to these interventions by rivals, geographic proximity as well as hostile relations with the government of Nicaragua affected the U.S. intervention decision. There are some minor but noticeable differences between the full sample and the democratic sub-sample with respect to the remaining international factors related to the strategic significance of the conflict state. First, trade ties are more critical for democracies? intervention decision making confirming empirical findings on the impact of economic interdependence and trade relations on democracies? behavior (Gleditsch 2002). Compared to the full sample, trade ties have more than twice as much substantive effect on increasing democracies? probability of intervention. Trade ties with the conflict state also increases democracies? probability of intervening on the government side more. Second, the resource wealth of the conflict state does not influence how democracies choose sides in a civil conflict unlike the full sample. However, petroleum increases the probability of military intervention by democracies as it does in the full sample. Koga?s study, the only study which examines the specific impact of resources and ethnic ties on the intervention behavior of democracies and autocracies, concludes that resources do not affect democracies? intervention decision in civil conflicts (Koga 2011). By distinguishing between the decisions on whether and on whose side to intervene, this study is able to show that petroleum actually 180 increases democracies? probability of intervention even though resources do not influence democracies? decision on whose side to intervene. To sum up, the results from the censored probit analysis in this sub-sample provide a better understanding of the role of economic factors in democracies? intervention decision making. The findings regarding the impact of ethnic ties on intervention probability are similar in the full sample and in the sub-sample of democracies. The results also confirm Koga?s finding that ethnic ties make it more likely for democracies to intervene in civil conflicts (2011). Koga?s study does not examine the impact of other domestic factors on intervention behavior whereas this study also examines the impact of domestic unrest on intervention behavior of democracies. The findings support Kisangani and Pickering?s conclusion that neither elite unrest nor mass unrest motivates democratic leaders to intervene militarily in civil conflicts (2007). To put it differently, democracies do not seem to use military interventions in intrastate disputes as a form of diversionary foreign policy behavior. Finally, this study also shows that while regime similarity on its own does not motivate democratic leaders to intervene in civil conflicts, democracies are more likely to support the government side when they do intervene in civil conflicts taking place in democratic states. In addition to the regime type of the conflict state, another contextual factor that affects how democracies choose sides is the type of civil conflict. The impact of this contextual factor can only be observed in the sub-sample analysis since the type of conflict is not a significant variable in the full sample. Democracies are more likely to support the government side in territorial conflicts and more likely to support the rebel side in governmental conflicts. However, other factors such as trade ties or 181 regime type of the conflict state can also influence the side decision. Nonetheless, this finding is quite interesting and can be demonstrated by some examples. The U.S. intervention on the rebel side in Nicaragua, which was mentioned above, can be cited as an example since it occurred in a governmental conflict. Between 1963 and 1966, the U.K. intervened militarily on the rebel side in North Yemen which was experiencing a governmental conflict, but Britain also had a militarized interstate dispute with this country. In 1990, Israel intervened militarily on the side of the Ethiopian government in its territorial conflict with the Eritrean secessionist group, EPLF. However, it is important to note that EPLF was supported militarily by Israel?s rivals Iraq and Syria. Finally, Turkey provided military support to the government of Azerbaijan in 1992 and 1993 during its territorial conflict over Nagorno-Karabakh. In addition to being contiguous to Azerbaijan, Turkey also has ethnic ties with this country. These examples show that while the type of conflict can affect the side decision, it does not motivate democratic states to intervene in civil conflicts on its own which is also demonstrated by the statistical results from the selection equation. While the type of conflict affects democracies? decision on whose side to intervene, the rest of the contextual variables such as battle deaths, refugees and rebel fighting capacity influence their probability of intervention. However, contextual variables seem to have smaller substantive effects in the democratic sub-sample compared to the full sample. Control variables, on the other hand, play a more critical role for democracies. Involvement in an interstate war has a bigger constraining effect 182 on democracies? intervention behavior while relative capabilities increase their intervention probability more significantly compared to the full sample. Overall, the findings provide a more nuanced understanding of the factors behind the intervention decision of democracies. The similarities with the full sample prove that the modified realist framework is still useful for examining intervention decisions of democracies. The differences, on the other hand, indicate that regime type of states has some noticeable impact on their intervention behavior. In order to understand more about the relationship between regime type and intervention behavior, the next section focuses on autocratic third parties. Empirical Findings for Autocracies This section focuses on the empirical findings from the statistical analyses that were conducted within the sub-sample of autocracies. The following sub-sections present the statistical results, interpret their statistical significance and discuss their theoretical and substantive implications. Statistical Models Two different censored probit models are employed to test the hypotheses against autocratic states in the dataset. Similar to the analyses in the sub-sample of democracies, none of the models include the regime related variables for potential interveners since they are all autocratic. The first censored probit model includes all the remaining explanatory variables in this study except the refugees variable. The second model adds the refugees variable to the first censored probit model. The two censored probit models are presented in Tables 6.8 and 6.9. 183 Table 6.8: Model 1 - Censored Probit Results for Autocratic states Selection Equation Outcome Equation Dependent Variable Intervention Intervention on govt. side Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ .505**** (.152) -.765*** (.293) Militarized dispute ~ .572**** (.147) -.245 (.167) Alliance ~ -.045 (.106) .462** (.200) Historical ties -.230 (.407) -.051 (.411) Contiguous .852**** (.118) - Same region .440**** (.093) - Secondary diamonds .180 (.178) -.540** (.220) Petroleum .222*** (.079) .704**** (.209) Trade (logged) ~ .013 (.016) .214**** (.039) Rival intervention ~ .724**** (.097) - Rival intervention on rebel side ~ - .998**** (.206) Ally intervention ~ .337**** (.104) - Ally intervention on govt. side ~ - .891** (.353) Ethnic ties 1 .435**** (.128) - Ethnic ties 2 .479*** (.173) - Elite unrest ~ -.174**** (.051) Mass unrest ~ -.145*** (.051) - Democratic target ~ -.136** (.55) -1.193**** (.280) Government conflict .052 (.076) .262 (.216) Battle deaths (logged) ~ .154**** (.018) - Rebel fighting capacity ~ -.203**** (.056) - War involvement ~ -.073 (.087) - Major power .315** (.142) - CINC ratio (logged) ~ .171**** (.021) - Cold War .001 (.057) - Spell years -1.064****(.059) -.715****(.104) Constant -3.002**** (.165) -2.851**** (.650) N Total obs. = 136412 Uncensored obs. = 1559 Wald Chi-square = 309.36**** Rho = .617 (.112) Wald test of independent equations (rho = 0), Chi-square = 7.22*** Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 184 Table 6.9: Model 2 - Censored Probit Results for Autocratic states Selection Equation Outcome Equation Dependent Variable Intervention Intervention on govt. side Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ .485**** (.150) -.773*** (.287) Militarized dispute ~ .557**** (.148) -.254 (.168) Alliance ~ -.038 (.105) .471** (.201) Historical ties -.290 (.416) -.106 (.432) Contiguous .839**** (.117) - Same region .444**** (.092) - Secondary diamonds .190 (.184) -.534** (.221) Petroleum .187** (.078) .650*** (.205) Trade (logged) ~ .018 (.017) .220**** (.039) Rival intervention ~ .729**** (.096) - Rival intervention on rebel side ~ - 1.018**** (.205) Ally intervention ~ .313*** (.102) - Ally intervention on govt. side ~ - .865** (.356) Ethnic ties 1 .432**** (.127) - Ethnic ties 2 .496*** (.172) - Elite unrest ~ -.173**** (.052) - Mass unrest ~ -.141*** (.051) - Democratic target ~ -.129** (.53) -1.123**** (.289) Government conflict .009 (.075) .194 (.224) Battle deaths (logged) ~ .135**** (.018) - Refugees (logged) ~ .068**** (.014) - Rebel fighting capacity ~ -.187**** (.056) - War involvement ~ -.070 (.087) - Major power .321** (.141) - CINC ratio (logged) ~ .169**** (.021) - Cold War .020 (.058) - Spell years -1.073****(.059) -.720****(.108) Constant -3.574**** (.218) -3.568**** (.747) N Total obs. = 136412 Uncensored obs. = 1559 Wald Chi-square = 312.44**** Rho = .623 (.116) Wald test of independent equations (rho = 0), Chi-square = 7.41*** Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 185 As indicated in Tables 6.1 and 6.2, autocratic third parties make up 68% of the entire dataset and they are responsible for 70% of the 2285 military interventions. Due to randomly missing observations, the number of dyad years used in the selection equation of each censored probit model is 136,412 which constitute almost 97% of the autocratic sub-sample. The number of uncensored observations, which refers to the number of dyad years with intervention, is 1559 in each outcome equation. Thus, 96.8% of the interventions by autocracies are used while estimating the outcome equations. Finally, examination of the correlation matrix shows no evidence of significant multicollinearity among the explanatory variables in each censored probit model. The highest absolute level of correlation is between the variables conflict type (as indicated by government conflict) and battle deaths at 0.531, while the rest of the correlations are much smaller. In both of the censored probit analyses, the model chi-square rejects the null hypothesis that all coefficients in the model are zero. In the first model, the correlation between the two equations, rho, is 0.62 which is greater than a zero correlation. The hypothesis that rho equals zero is rejected at the 1% level based on the Wald test of independent equations. The statistical findings regarding the correlation between the two equations are also similar in the second model. Since both analyses show the existence of selection bias, the use of a censored probit model is justified for the sub-sample of autocracies as well. Table 6.10 summarizes the support the results lend to each hypothesis postulated for both dependent variables to make discussion easier. The table presents whether or not each hypothesis is supported based on the statistical significance and 186 direction of the independent variables in both models. If a variable is statistically significant but its effect is the opposite of what was hypothesized, then Table 6.10 reports that finding as well. Some hypotheses are also categorized as receiving relatively weaker support due to the statistical significance of results at the 90% level. The rest of the hypotheses are supported at higher significance levels as can be seen in Tables 6.8 and 6.9. Table 6.10: Summary of Support for Hypotheses Hyp. Independent Variables Dependent Variables International Intervention Government side H1a, 1b Rivalry supported supported H2a, 2b Militarized dispute supported not supported H3a, 3b Alliance not supported supported H4a, 4b Historical ties not supported not supported H5 Contiguous supported na H6 Same region supported na H7a, 7b Secondary diamonds not supported supported H8a, 8b Petroleum supported supported H9a, 9b Trade not supported supported H10a Rival intervention supported na H10b Rival intervention on reb. side na supported H11a Ally intervention supported na H11b Ally intervention on govt. side na supported Domestic H18a Ethnic ties 1 supported na H18b Ethnic ties 2 supported na H19b Elite unrest supported na H20 Mass unrest supported na Contextual H17a, 17b Democratic target supported supported H22a, 22b Government conflict not supported not supported H23 Battle deaths supported na H24 Refugees supported na H25 Rebel fighting capacity opposite found na Control Variables H12 War involvement not supported na H13 Major power supported na H14 CINC ratio supported na H15 Cold War not supported na 187 Table 6.10 shows that international factors affect autocracies? intervention decision making somewhat differently compared to the full sample as well as the democratic sub-sample. However, a lot of similarities can also be observed between the autocratic and democratic sub-samples. Similar to the full sample and democratic sub-sample, international factors such as hostile relations with the conflict state, geographic proximity, petroleum and interventions by rivals and allies influence the intervention probability of autocracies. On the other hand, alliance ties with the conflict state and secondary diamonds do not affect the probability of intervention by autocracies, which are also consistent with the findings from the full sample and democratic sub-sample. However, historical ties and trade relations with the conflict state also do not play a role when autocracies are deciding to intervene in a civil conflict whereas these two international factors are critical in the other two samples. There are also noticeable differences between the samples with respect to the outcome equation. Historical ties and the existence of militarized disputes with the conflict state are not critical for autocracies when they are choosing sides in a civil conflict. In the democratic sub-sample, both of these factors affect the side decision while alliance or rivalry with the conflict state and the type of resources do not. Similar to the full sample, rivalry and alliance with the conflict state affect how autocracies choose sides. The type of resources in the conflict state also influences autocracies? side decision. As hypothesized, autocracies are more likely to intervene on the rebel side when there are secondary diamonds in the conflict state but they are more likely to support the government side if the conflict state has petroleum. In fact, 188 economic factors in general seem to be more decisive for autocracies when they are choosing sides because trade ties also affect their decision making. The sub-sample analyses show that while there are some similarities between autocracies and democracies, there are also important differences with respect to how certain international factors affect the intervention decision of states based on their regime type. Since these differences cannot be captured in the full sample analysis, sub- sample analyses provide an improved understanding of the similarities and differences between democracies and autocracies. Domestic variables have the hypothesized effect on autocracies as they do in the full sample. Ethnic ties with the population of the conflict state increase the probability of military intervention by autocracies. Overall the findings confirm the ethnic ties argument in the ethnic conflict literature both in the case of democracies and autocracies. Moreover, domestic unrest also affects autocratic leaders? intervention decisions. Unlike in the case of democracies, both mass unrest and elite unrest decrease the probability of intervention by autocracies. In other words, domestic problems seem to force autocratic leaders to concentrate on domestic issues in order to eliminate threats to their own political survival. In fact, Kisangani and Pickering conclude that autocracies do not ?externalize their domestic problems? and tend to use domestic solutions in response to domestic crises (2007, p. 295). This negative relationship between domestic unrest and the probability of military intervention suggests that autocratic leaders focus on suppressing domestic challenges through domestic measures instead of resorting to diversionary foreign policy behavior which confirms Kisangani and Pickering?s findings. 189 The findings regarding the hypotheses on contextual factors in the autocratic sub-sample are similar to the full sample. Increasing number of battle deaths and refugees increases the probability of intervention by autocracies. On the other hand, higher rebel fighting capacity decreases the intervention likelihood. As hypothesized, autocracies are less likely to intervene in a democratic conflict state and when they do intervene, autocracies are less likely to support the government side in democracies. Both of these results from the sub-sample analysis support the findings on joint autocracy from the full sample. Finally, similar to the full sample, the type of conflict does not play a role in the intervention decisions of autocracies. Thus, the type of conflict appears to be critical only for democracies when they intervene in a civil conflict. Only two of the hypotheses on control variables are supported in the autocratic sub-sample. Similar to the democratic sub-sample, as the ratio of capabilities increases the probability of intervention also increases in the case of autocratic third parties. Major powers are more likely to intervene in the sub-sample of autocracies as well. Cold War does not have a statistically significant effect on autocratic states similar to democracies. However, involvement in an interstate war also does not have a significant impact in the autocratic sub-sample whereas this control variable has a constraining effect on democracies? intervention behavior. As an additional robustness check, this study also employed rare events logit to analyze the first dependent variable. The results from rare events logit analyses conducted in the sub-sample of autocracies are consistent with the results from the first stage of censored probit analyses and provide additional confidence in the 190 findings on the hypothesized relationships. Appendix D provides the results from relogit models conducted in this sub-sample. Interpretation of Results After reviewing the censored probit results in terms of the support they provide to the hypotheses in the sub-sample of autocracies, it is now useful to discuss the substantive impact of statistically significant variables on the two dependent variables. Table 6.11 presents the changes in the predicted probability of intervention=1 when the variable of interest increases by one unit or some number of units depending on the nature of the variable. First differences are only calculated for the statistically significant variables from the first model except for the refugees variable which is computed from the second model. Table 6.12 indicates the changes in the predicted probability of pro-government intervention=1 when the variable of interest changes by one unit or some number of units as reported in the table. The substantive impact of independent variables reported in Tables 6.11 and 6.12 will be discussed separately for each dependent variable. Three international factors which have the largest substantive effect on the intervention probability of autocracies are contiguity, interventions by rivals and militarized dispute with the conflict state. These three are also the most influential factors in the full sample after historical ties with the conflict state which is not a statistically significant variable in the autocratic sub-sample. Contiguity increases the probability of military intervention by 20 percent and interventions by rivals increase it by almost 19 percent in the autocratic sub-sample. Hostile relations with the conflict state have the next largest impact on autocracies? intervention probability 191 similar to the democratic sub-sample. Having a militarized dispute or rivalry with the conflict state increases the probability of intervention by 17 percent and 16 percent respectively. Table 6.11: Changes in the Predicted Probability of Intervention When this variable moves from ? to ? in Change in probability of intervention 95% Confidence Interval (unless indicated by *) Model 1: Rivalry 0 ? 1 16.23% [7.67% 24.79%] Militarized dispute 0 ? 1 17.11% [9.08% 25.14%] Contiguous 0 ? 1 20.06% [11.10% 29.02%] Same region 0 ? 1 14.43% [7.35% 21.51%] Petroleum 0 ? 1 7.67% [2.96% 12.38%] Rival intervention 0 ? 1 18.82% [10.43% 27.21%] Ally intervention 0 ? 1 12.89% [6.26% 19.52%] Ethnic ties 1 0 ? 1 13.96% [6.72% 21.20%] Ethnic ties 2 0 ? 1 15.04% [5.86% 24.22%] Elite unrest 0 ? 1 -6.46% [-9.78% -3.14%] Mass unrest 0 ? 1 -4.68% [-7.45% -1.91%] Democratic target 0 ? 1 -4.39% [-7.56% -1.22%] Battle deaths p25 ? p75 Battle deaths min ? max 3.48% 5.15% [1.69% 5.27%] [2.32% 7.98%] Rebel fighting capacity 1 ? 2 Rebel fighting capacity 1 ? 3 -5.42% -7.28% [-8.19% -2.65%] [-10.82% -3.74%] Major power 0 ? 1 12.13% [5.70% 18.56%] CINC ratio p25 ? p75 CINC ratio min ? max 4.62% 5.87% [2.21% 7.03%] [2.59% 9.15%] Model 2: Refugees p25 ? p75 Refugees min ? max 1.36% 2.32% [0.71% 2.01%] [1.18% 3.46%] Notes: This table reports the first differences for statistically significant variables from the selection equations in Tables 6.8 and 6.9. The probabilities are calculated by holding all continuous variables at their means and all dichotomous and categorical variables at their modes. * indicates 90% confidence interval. After these four international factors, ethnic ties have the most critical impact on autocracies? intervention probability in civil conflicts. While ethnic ties are influential in all samples, their biggest impact is in the autocratic sub-sample with 14 192 to 15 percent. The remaining statistically significant international factors have the next largest effects on autocracies? intervention behavior. Being in the same region with the conflict state increases the probability of intervention by 14.4 percent while interventions by allies increase it by almost 13 percent. The substantive effects of these two factors are again larger for autocracies than democracies. However, major powers are only 12 percent more likely to intervene in the autocratic sub-sample compared to 17 percent in the case of democracies. Finally, petroleum in the conflict state also has a higher impact, 7.7 percent, on the probability of military intervention by autocracies than it has on democracies. Contextual factors have a larger effect on autocracies than democracies as well. As the fighting capacity of rebels relative to government forces changes from low to high, autocracies are 7.2 percent less likely to intervene militarily. Autocracies are also 4.4 percent less likely to intervene when the conflict state is a democracy. On the other hand, as the number of battle deaths and refugees reach maximum, the probability of intervention increases by 5.2 percent and 2.3 percent respectively. Although contextual factors have a somewhat bigger impact on autocracies than democracies, the ratio of capabilities has a much smaller effect on autocracies which indicates that relative capabilities play a less significant role in the decision making of autocracies compared to democracies when they intervene in civil conflicts. Finally, both types of domestic unrest have a quite significant constraining effect on the intervention behavior of autocracies. Elite unrest decreases the probability of military intervention by 6.5 percent while mass unrest decreases it by 4.7 percent. The substantive effects of these domestic factors are even higher than 193 most of the contextual factors as well as the ratio of capabilities. In short, domestic factors such as ethnic ties or domestic unrest seem to play a critical role when autocracies are deciding to intervene in a civil conflict. Table 6.12 shows that the regime type of the conflict state is crucial when autocracies are choosing sides in a civil conflict. Autocracies are 48 percent less likely to intervene on the government side when the conflict state is democratic. While this finding again proves the importance of regime similarity when third parties choose sides in a civil conflict, it also indicates that domestic factors in general are quite critical for autocracies? foreign policy decisions. The literature has mostly analyzed the impact of domestic factors on the foreign policy behavior of democracies, but the findings in this study show that foreign policy decision making of autocracies is influenced by domestic sources as well. Table 6.12: Changes in the Predicted Probability of Government Intervention When this variable moves from ? to ? in Change in probability of govt. intervention 95% Confidence Interval (unless indicated by *) Model 1: Rivalry 0 ? 1 -25.46% [-37.44% -13.48%] Alliance 0 ? 1 14.78% [5.53% 24.03%] Secondary diamonds 0 ? 1 -18.53% [-28.39% -8.67%] Petroleum 0 ? 1 22.27% [10.98% 33.56%] Trade p25 ? p75 Trade min ? max 4.94% 7.13% [2.57% 7.31%] [3.52% 10.74%] Rival int. rebel side 0 ? 1 42.31% [22.73% 61.89%] Ally int. govt. side 0 ? 1 37.68% [14.81% 60.55%] Democratic target 0 ? 1 -48.02% [-68.20% -27.84%] Notes: This table reports the first differences for statistically significant variables from the outcome equation in Table 6.8. The probabilities are calculated by holding all continuous variables at their means and all dichotomous and categorical variables at their modes. * indicates 90% confidence interval. 194 The rest of the factors that affect how autocracies choose sides in civil conflicts also demonstrate some of the major differences between autocracies and democracies. Autocracies are 42 percent more likely to intervene on the government side when their rivals intervene on the rebel side while interventions by allies on the government side increase this probability by almost 38 percent. Therefore, interventions by rivals and allies have a significantly larger impact on the side decision of autocracies than democracies. Rivalry and alliance with the conflict state also play an important role when autocracies choose sides in a civil conflict whereas neither of these two factors affects how democracies choose sides. Rivalry decreases the probability of pro-government intervention by 25 percent while alliance increases it by 15 percent. Finally, the type of resources in the conflict state also has a critical effect on autocracies? side decision whereas resources do not factor into the decisions of democracies when they choose sides. Autocracies are 22 percent more likely to intervene on the government side when the conflict state has petroleum and they are 18 percent less likely to support the government side when there are secondary diamonds. Increasing trade ties with the conflict state increases the probability of pro- government intervention but the impact of trade ties is larger for democracies when they are choosing sides in a civil conflict. Discussion of Findings for Autocracies The censored probit analyses conducted in this sub-sample provide a more nuanced understanding of the factors that motivate autocracies to intervene in civil 195 conflicts and facilitate a comparative analysis of the intervention behavior of democracies and autocracies. International factors have the largest effect on autocracies? intervention decision making. With the exception of historical ties, the main international factors that motivate both democratic and autocratic leaders to intervene in civil conflicts are the same. Contiguity, hostile relations with the conflict state as well as rival interventions have some of the largest effects on the probability of intervention by both autocracies and democracies. On the other hand, power differentials seem to play a more important role in the intervention decision making of democracies than autocracies. Democratic states are more likely to intervene in weaker conflict states than autocracies. Interventions by rivals and allies seem to motivate autocratic leaders to intervene in civil conflicts more than power differentials with the conflict state. Some of these factors can be observed in the following examples. Rivalry and the existence of militarized interstate disputes have caused Pakistan to provide military support to various rebel groups in India since the beginning of 90s. Some of these rebel groups are involved in territorial conflicts over Kashmir, Nagaland and Manipur with the government of India. While others like PWG and MCC are fighting for control over the government. Similarly, rivalry and hostile relations led Ethiopia to intervene militarily on the rebel side in Sudan as soon the civil conflict between SPLM and the Sudanese government started in 1983. Ethiopia provided military support to SPLM until the late 1990s. While having a militarized interstate dispute with Ethiopia, Eritrea intervened militarily on the side of OLF and ONLF, two rebel groups fighting with the government of Ethiopia, in 1999. Since this was a part of 196 Eritrea?s military strategy, it only lasted for a year and occurred during its interstate military dispute with Ethiopia. Another important finding from the sub-sample analysis is the significant impact of domestic factors on the intervention decision of autocracies. While ethnic ties also motivate democratic leaders to intervene, ethnicity plays a bigger role when autocracies decide to intervene. Moreover, both elite and mass unrest have a considerable impact on the likelihood of military intervention by autocracies. The findings from the sub-sample analyses suggest that although both democratic and autocratic leaders do not use military interventions in civil conflicts as a form of diversionary foreign policy behavior when they are faced with domestic crises, domestic unrest decreases autocracies? probability of military intervention in civil conflicts. On the other hand, domestic unrest does not affect the probability of intervention by democracies. These results are consistent with Kisangani and Pickering?s study on diversionary behavior of democracies and autocracies (2007). Contextual factors, such as rebel fighting capacity and conflict intensity, have a substantively larger effect on autocracies than democracies as well. The different impact of contextual factors on autocracies and democracies is also noticeable with respect to the regime type of the conflict state. While the regime type of the conflict state does not seem to affect democracies? probability of military intervention, autocracies are less likely to intervene in democratic conflict states. There are even more important differences between democracies and autocracies when they are choosing sides in a civil conflict. Historical ties have the most significant effect on how democracies choose sides whereas the regime type of 197 the conflict state is the most crucial factor for autocracies. Although the regime type of the conflict state also influences the side decision of democracies, this contextual factor has less substantive impact on democracies compared to other factors. While both democracies and autocracies are more likely to intervene on the rebel side if they have hostile relations with the conflict state, alliance ties also matter in the case of autocracies. Finally, interventions by rivals and allies seem to play a bigger role for autocracies than democracies when they are choosing sides. In fact, interventions by rivals and allies have the second largest impact on the side decision of autocracies. For example, despite its rivalry with El Salvador, Honduras intervened on the government side multiple times during the civil conflict in the 1980s. This decision was influenced by pro-government interventions by the U.S. and Venezuela which were both allies of Honduras and pro-rebel intervention by its rival, Nicaragua. Rival intervention by Iran on the government side in the early 1970s also caused Iraq to support the rebel side during the civil conflict in Oman despite its alliance with the government. The existence of petroleum in the conflict state increases the probability of intervention by both democracies and autocracies. However, the type of resources in the conflict state also plays a decisive role when autocracies choose sides in a civil conflict whereas resources do not influence the side decision of democracies. The results from the sub-sample analysis confirms Koga?s finding (2011) that autocracies are more likely to intervene on the rebel side when there are secondary diamonds in the conflict state. In addition, this study finds that autocracies are more likely to support the government side when there are non-lootable resources like petroleum in 198 the conflict state. Even though resources do not affect how democracies choose sides, stronger trade ties with the conflict state motivate democracies more to intervene on the government side compared to autocracies. These findings regarding the different impacts of variables on autocracies demonstrate that sub-sample analyses are quite useful for comparative purposes. Conclusion In order to demonstrate the similarities and differences between democracies and autocracies one more time, the following tables rank the substantive effects of statistically significant variables in each sub-sample. Table 6.13 compares the substantive impact of international, domestic and contextual factors that motivate democracies and autocracies to intervene in civil conflicts. Table 6.14 presents the factors that are critical for democracies and autocracies when they choose sides. As the tables indicate, the sub-sample analyses provide an improved understanding of the substantive effects of explanatory variables and demonstrate the differences between autocracies and democracies much more effectively than the full sample analysis does. However, the sub-sample analyses also show that some of the factors have quite similar effects on democracies and autocracies. Therefore, the overall findings from the full sample analysis are quite robust and provide important insights into the intervention behavior of states even without splitting the dataset into democracies and autocracies. Nevertheless, the additional benefits of sub-sample analyses are indisputable if one is also interested in comparing the intervention behavior of different types of states. 199 Table 6.13: Comparing substantive effects of variables in selection equations Democratic States Autocratic States Independent variables % Independent variables % Historical ties 19.86% Contiguous 20.06% Contiguous 18.49% Rival intervention 18.82% Militarized dispute 18.09% Militarized dispute 17.11% Rivalry 17.54% Rivalry 16.23% Major power 17.02% Ethnic ties 2 15.04% Rival intervention 16.94% Same region 14.43% Same region 12.78% Ethnic ties 1 13.96% Ethnic ties 2 12.19% Ally intervention 12.89% CINC ratio 9.90% Major power 12.13% War involvement -8.97% Petroleum 7.67% Ethnic ties 1 9.43% Rebel fighting capacity -7.28% Petroleum 6.52% Elite unrest -6.46% Ally intervention 6.29% CINC ratio 5.87% Rebel fighting capacity -5.36% Battle deaths 5.15% Battle deaths 4.85% Mass unrest -4.68% Trade 3.17% Democratic target -4.39% Refugees 1.41% Refugees 2.32% Alliance - Alliance - Secondary diamonds - Historical ties - Elite unrest - Trade - Mass unrest - War involvement - Cold War - Cold War - Democratic target - Secondary diamonds - Government conflict - Government conflict - Notes: Explanatory variables are ranked according to their substantive effects. - indicates that the variable is not statistically significant. 200 Table 6.14: Comparing substantive effects of variables in outcome equations Democratic States Autocratic States Independent variables % Independent variables % Historical ties 41.63% Democratic target -48.02% Rival int. on rebel side 35.34% Rival int. on rebel side 42.31% Militarized dispute -24.86% Ally int. on govt. side 37.68% Government conflict -19.42% Rivalry -25.46% Democratic target 17.74% Petroleum 22.27% Ally int. on govt. side 16.59% Secondary diamonds -18.53% Trade ties 10.08% Alliance 14.78% Rivalry - Trade ties 7.13% Alliance - Militarized dispute - Secondary diamonds - Historical ties - Petroleum - Government conflict - Notes: Explanatory variables are ranked according to their substantive effects. - indicates that the variable is not statistically significant. This chapter has discussed the differences and similarities in the intervention behavior of democracies and autocracies in detail. Thus, it is more useful at this point to emphasize some of the most striking findings from this comparative analysis that are theoretically and empirically important for the literature on interventions in civil conflicts. First and foremost, many of the hypotheses derived from the modified realist approach are supported by the sub-sample analyses which indicate that modified realism is an appropriate theoretical framework for identifying the international and domestic sources of intervention decision making for both democracies and autocracies. Moreover, it is also capable of integrating contextual factors into the study of interventions. As a result, modified realism facilitates a comparative 201 understanding of the impact of different groups of factors on the intervention behavior of democracies and autocracies. Second, the findings from sub-sample analyses show that the strategic significance of the conflict state plays the most crucial role when foreign leaders decide to intervene in a civil conflict. Geographic proximity, hostile relations with the conflict state and interventions by rivals or allies are some of the key international factors motivating both democratic and autocratic leaders to intervene in civil conflicts. The substantive effects of the rest of the international factors seem to differ for democracies and autocracies, particularly with respect to how they choose sides in a civil conflict. Hence, sub-sample analyses prove extremely useful for observing these differences. Nevertheless, international factors in general play the most decisive role for both democracies and autocracies. Since some of the hypotheses related to rivalry, the strategic impact of other interveners, type of resources and trade have not been adequately tested in quantitative studies before, the findings from the sub- sample analyses provide an improved understanding of the role of these factors in the intervention decisions of democratic and autocratic leaders. Third, domestic factors tend to have the second most significant impact on the intervention decisions of leaders after international factors. The sub-sample analyses show that ethnic ties with the conflict state motivate both democratic and autocratic leaders to intervene in civil conflicts. Elite and mass unrest also constrains autocratic leaders? intervention behavior whereas intervention decisions of democratic leaders are not influenced by domestic unrest. Since regime similarity also affects the probability of intervention by autocracies, it can be concluded that domestic factors 202 are quite important when autocratic leaders are deciding to intervene. The literature would benefit from analyzing the domestic sources of foreign policy decision making not only for democracies but also for autocracies. Fourth, contextual factors seem to play a smaller role than international and domestic factors when states are deciding to intervene in a civil conflict. Although the intensity of the conflict and rebel fighting capacity influence the probability of military intervention by democracies and autocracies, their impact is not as critical as the other factors related to the international and domestic sources of intervention. These findings indicate once more that quantitative intervention studies using a conflict-oriented approach cannot provide a satisfactory explanation of the motives of interveners by focusing on the attributes of the civil conflict. Although contextual factors do contribute to the decision making of potential interveners, the findings in this study demonstrate that they are not as critical as the strategic international and domestic considerations of leaders for the intervention decision. One important exception to this is the regime type of the conflict state because it has a much more significant impact on the intervention behavior of states than other contextual factors. When the conflict state is democratic, autocratic third parties are more likely to intervene on the rebel side while democratic interveners are more likely to support the government side. Ironically, neither regime type nor regime similarity has been studied sufficiently in quantitative intervention studies using a conflict-oriented approach. The findings in this study suggest that regime type of the conflict state and regime similarity should be analyzed more closely in the intervention literature. 203 To sum up, sub-sample analyses provide a thorough understanding of the impact of international, domestic and contextual factors on the intervention behavior of democracies and autocracies. They also show that analyzing leaders? international and domestic considerations which influence their perception of the strategic significance of the conflict state or the civil conflict is more conducive to understanding the rationale behind military interventions than just focusing on contextual factors. Bringing the much needed foreign policy analysis into the quantitative study of interventions leads to important findings in the case of both democratic and autocratic interveners. 204 Chapter 7: Conclusion Review of the Study Back in 1969, Rosenau argued that ?the factors that foster, precipitate, sustain, channel, constrain, and curb intervention? had not been explored scientifically (1969, p. 150). Since then, the majority of the progress towards a scientific understanding of the causes of interventions came from qualitative studies. With regard to foreign military interventions, these case studies offered important theoretical insights on why some countries decided to use this foreign policy instrument in specific civil conflicts, but they fell short on providing generalizations about the rationale behind foreign military interventions. In the late 90s, more scholars have adopted quantitative techniques to study third party interventions in civil conflicts. While most of the recent quantitative studies have examined the impact of foreign military interventions on the dynamics and outcomes of civil conflicts, only a small amount of quantitative research has been conducted on the causes of military interventions. More importantly, the recent quantitative research has neglected the decision making calculus of the intervener and focused more on the attributes of the civil conflict that attract third party intervention. Foreign policy analysis, the dominant analytical framework in earlier qualitative studies, has been minimized in the recent quantitative literature as the theoretical emphasis shifted from the intervener to the civil conflict. The purpose of this study was to advance the current state of knowledge on the causes of biased military interventions by focusing on the foreign policy decision making calculus of the intervener. While the first major contribution of this large-N 205 study was to identify some of the main factors that motivate foreign powers to intervene militarily in civil conflicts, the second major contribution was to bring the much needed theoretical framework of foreign policy analysis into the quantitative study of military interventions. By uncovering the empirical patterns in the rationale behind military interventions through a more suitable theoretical perspective, this study aimed to make theoretical and empirical contributions to the current quantitative literature on military interventions. In order to identify the factors that motivate state leaders to use military intervention as a foreign policy instrument, this dissertation analyzed the international and domestic sources of foreign policy decision making through a modified realist approach (Huth 1998). There are several reasons which make modified realism an appropriate theoretical framework for studying the rationale behind military interventions. First, it is able to account for both international and domestic sources of foreign policy decisions as well as the systematic linkages between them. Second, modified realism can also account for the contextual factors related to the civil conflict that affect the decision making calculus of foreign leaders. Thus, the modified realist framework used in this study was able to synthesize the actor- oriented and conflict-oriented approaches while making the potential intervener theoretically central and incorporating the attributes of the civil conflict that also affect leaders? foreign policy decision making. Hypotheses on the international, domestic and contextual factors that are likely to affect foreign leaders? decisions were derived from modified realism. Since this study also argued that the decisions on whether and on whose side to intervene in 206 a civil conflict are closely related, some of the same factors were hypothesized to influence both decisions. These hypotheses were first tested against a dataset that included both actual and potential interveners in civil conflicts between 1946 and 2002. In order to check the robustness of results and to better understand the relative importance of critical factors for different types of states, these hypotheses were then tested against three important sub-samples: major powers, democracies and autocracies. The empirical findings from the statistical analyses in the full sample provided support for many of the hypotheses derived from modified realism. Some of these findings were quite novel while others were consistent with the previous findings in the literature. The sub-sample analyses, on the other hand, offered an improved understanding of the factors that particularly motivate major powers, democracies and autocracies. The similarities between the sub-sample analyses demonstrated the robustness of the results from the full sample. However, these additional analyses also proved useful for observing the differences between sub-samples that could not be captured in the full sample analysis. The next section summarizes the key findings on international, domestic and contextual factors that ?foster? or ?curb? the use of military intervention in civil conflicts as a foreign policy tool by state leaders (Rosenau 1969, p. 150). The theoretical significance and policy implications of these findings are also discussed in the following sections. The final section outlines future research directions. 207 Summary Findings and Theoretical Implications International Sources of Intervention This study argued that foreign leaders pay attention to the strategic importance of the conflict state for their country as well as the strategic significance of the conflict environment when they are deciding whether and on whose side to intervene in a civil conflict. Hence, hypotheses were postulated about the international factors which make the conflict state or the civil conflict strategically important for foreign leaders. Political, military and economic relations with the conflict state and the existence of perceived strategic interests which can also stem from the military interventions of allies and rivals in the civil conflict country were among these critical international factors. The findings from the full sample demonstrate that the most crucial factors which prompt foreign leaders to intervene in civil conflicts are historical ties, geographic proximity, hostile relations with the conflict state and military interventions by rivals and allies. These same factors also have some of the most critical impact on whose side foreign leaders decide to intervene in a civil conflict. Although these results are mostly confirmed in the sub-sample analyses, some differences are also observed between different types of states. Historical ties do not influence autocratic leaders? decisions on whether and on whose side to intervene in conflicts. In the case of major powers, alliance ties with the conflict state also increase the probability of intervention and the probability of supporting the government side significantly. Thus, major powers are the only group of states which are more likely to intervene in civil conflicts to support their allies. 208 The impact of contiguity and historical ties on leaders? intervention decisions confirms the previous findings in the literature. However, neither the role of hostile relations such as rivalry and militarized disputes with the conflict state nor the reactions of foreign powers in response to interventions by their rivals or allies have been examined adequately in the recent quantitative literature. In contrast, qualitative studies have emphasized such strategic considerations of foreign leaders and argued that states can intervene in reaction to a rival intervention or in order to weaken a rival government. This is one of the few large-N studies demonstrating that these strategic considerations motivate many third party states to intervene in civil conflicts. The findings show that although international economic factors, such as trade ties and resource wealth of the conflict, affect the intervention probability of foreign powers, they play a much smaller role compared to the international factors discussed above. The existence of petroleum in the conflict state motivates foreign powers to intervene in civil conflicts more than trade ties. Secondary diamonds, on the other hand, do not affect the intervention likelihood of foreign powers. The sub-sample analyses provide a more nuanced understanding of the role of trade relations in the intervention decisions of leaders. While stronger trade ties with the conflict state increase the probability of intervention by democracies, they have no impact on the intervention decisions of autocratic leaders. This is also a novel finding in the quantitative literature since trade ties with the conflict state have not been examined extensively before. 209 The sub-sample analyses also demonstrate that economic factors have a more critical impact when foreign powers choose sides in a civil conflict. Autocracies are more likely to intervene on the rebel side when there are secondary diamonds in the conflict state whereas they are more likely to support the government when there is petroleum. On the other hand, the type of resources has no effect on which side democracies intervene but stronger trade ties increase their likelihood of supporting the government. While the relationship between third party intervention and resource wealth of the conflict state has been analyzed in recent quantitative studies, the different effects of resources on democratic and autocratic third parties have not been studied in detail before. Overall, this study offers an improved understanding of how the economic significance of the conflict state affects the intervention decision making of democracies and autocracies. Finally, international factors that were included as control variables in the study also have considerable effects on the intervention decision, particularly in the case of democracies. Involvement in an interstate war constrains the intervention behavior of democracies whereas the intervention decisions of autocracies and major powers are not affected by it. Power differentials are much more important for democracies than autocracies as well which indicates that democracies are more likely to intervene when the conflict state is weaker. In short, control variables have a more significant impact on the intervention decision making of democracies than autocracies which could only be captured in the sub-sample analyses. However, the bipolar structure of the international system does not seem to have an effect on the 210 intervention behavior of democracies and autocracies. The only group of states that had a higher probability of intervention during the Cold War was major powers. Domestic Sources of Intervention This study argued that the regime type of interveners, ethnic ties with the conflict state and domestic unrest are likely to have critical impacts on the intervention decisions of foreign leaders. Among these three factors, ethnic ties have been analyzed the most in the literature and there is more consensus on the role of ethnic ties among scholars. On the other hand, the regime type of interveners and the domestic unrest in the intervener state have not been studied as extensively and the findings from existing studies are mixed. The findings from the full sample indicate that the domestic sources of intervention decisions are quite crucial. As a group, domestic factors play the most significant role in intervention decision making after historical ties, geographic proximity, hostile relations with the conflict state and military interventions by rivals and allies. However, sub-sample analyses indicate that there are important differences between democracies, autocracies and major powers with respect to the substantive impact of domestic factors on their intervention decisions. Therefore, the combined findings from the full sample and sub-sample analyses facilitate an improved understanding of the domestic sources of intervention behavior. One of the most important findings in this study has to do with the regime type of the intervener. Although the impact of regime type on interventions in civil conflicts has not been studied as comprehensively as its effect on interstate conflicts, scholars who examined the role of regime type in third party interventions have 211 arrived at contradictory results. Some scholars conclude that regime type has no significant impact on intervention likelihood (i.e. Lemke and Regan 2004), while others argue that democracies are more likely to intervene in civil conflicts (i.e. Kathman 2011). The findings from the full sample in this study show that democracies are less likely to intervene in civil conflicts than autocracies, which is more consistent with the findings from the conflict literature. Scholars analyzing the impact of regime type on involvement in interstate wars or military disputes have pointed out the cautious and selective attitudes of democratic leaders in initiating wars (Reiter and Stam 1998; Reed and Clark 2000; Clark and Reed 2003; Bueno de Mesquita et. al. 2004). This study indicates that democratic leaders are also more selective than autocratic leaders when they intervene militarily in civil conflicts. This dissertation also shows that regime similarity has a crucial impact on the intervention behavior of states, particularly in the case of autocracies. Although joint democracy does not increase the probability of intervention; when democracies do intervene in civil conflicts, they are more likely to intervene on behalf of democratic governments instead of against them. This finding is consistent with what Hermann and Kegley found in a series of studies on interventions in interstate and intrastate conflicts (1995, 1996, 2001). On the other hand, joint autocracy increases both the probability of military intervention and the probability of supporting the government side. However, the sub-sample analyses also indicate that regime similarity has a more dramatic impact on how autocracies choose sides in a civil conflict compared to democratic third parties. Overall, the results from the statistical analyses shed more 212 light on the role of regime type and regime similarity in intervention decision making than the existing quantitative studies on military interventions in civil conflicts. Ethnic ties with the conflict state are quite critical for both democracies and autocracies. In fact, ethnic ties have the largest impact among the domestic sources of intervention decision. While neither democratic nor autocratic leaders use military interventions as a form of diversionary foreign policy behavior when they are faced with domestic problems, elite unrest and mass unrest curb the use of military intervention in the case of autocracies. These results are consistent with some previous findings in the literature but they also indicate that domestic factors in general, including regime similarity, are quite important for autocracies. The literature has mostly focused on the impact of domestic factors on the foreign policy decision making of democracies, but the findings in this study indicate that foreign policy decision making of autocracies is influenced by domestic sources as well. Finally, the sub-sample analyses show that major powers are quite different than the rest of the states in terms of the effect of domestic factors on their intervention behavior. The role of domestic factors is quite minimal when major powers decide to intervene in civil conflicts. They are the only group of states in which ethnic ties do not affect the intervention decision. Regime similarity also has a relatively smaller effect on the intervention decisions of major powers compared to other states. These findings indicate that strategic international considerations are more critical for major powers when they are deciding to intervene in civil conflicts. 213 Contextual Sources of Intervention As discussed before, existing quantitative studies have analyzed the attributes of the civil conflict more frequently than the international and domestic considerations of interveners in order to understand the causes of military interventions. This study also posited that certain contextual factors are likely to affect the intervention decisions of foreign leaders. However, the findings show that contextual factors have a much smaller impact on the intervention probability of foreign powers compared to international and domestic factors. Foreign powers are more likely to intervene in a civil conflict as the number of battle deaths and refugees increase, but they are less likely to become involved militarily when rebels? fighting capacity increases relative to government forces. These results indicate that states are more likely to avoid civil conflicts that have the potential to experience a stalemate whereas the increasing conflict intensity does not have the same impact on potential interveners. Moreover, the regime type of the conflict state is quite important when third parties choose sides in a civil conflict. Autocratic third parties also pay attention to the regime type of conflict states when they are deciding to intervene. Despite its significance, the regime type of the conflict state is one of the least analyzed contextual factors in the conflict-oriented quantitative studies. One of the conclusions of this study is that while the contextual factors, which have been frequently analyzed in the literature, contribute to the intervention decision making, they are not the most decisive factors for potential interveners. Therefore, 214 applying a foreign policy analysis that focuses on the international and domestic sources of intervention is more productive for theoretical and empirical purposes. Policy Implications Foreign military interventions have been a persistent feature of international politics and they will continue to occur frequently in civil conflicts if measures to prevent them are not taken by the international community. Academic studies on the short and long-term impacts of foreign military interventions show that they can have devastating effects on conflict states. Biased military interventions tend to increase the duration and intensity of civil conflicts significantly according to empirical findings in the literature. Scholars examining the long-term political effects of interventions argue that even military interventions by democratic third parties do not facilitate the process of democratization in conflict states. In fact, they can eventually lead to more authoritarian systems in target states (Bueno de Mesquita and Downs 2006). None of these findings are surprising since biased militarily interventions are not conflict resolution efforts. Foreign leaders interpret the threats or opportunities posed by the civil conflict in terms of their own international and domestic strategic considerations when they provide military support to one of the domestic opponents. As this study shows, some of the main factors that prompt states to intervene on the rebel side are rivalry and militarized disputes with the target state. While biased military interventions will continue to occur in civil conflicts, it is possible to minimize their negative political and social effects on conflict states by reducing their frequency. However, this can only happen if the international community becomes more willing to address this issue and recognizes that biased 215 military interventions make conflicts more intractable. Foreign military interventions, even with the best of intentions such as protecting ethnic minorities, have damaging effects on societies in the short run as well as in the long run. International and regional organizations are involved in different conflict resolution efforts in various civil conflicts around the world. While there has been an increase in the number of multilateral conflict resolution efforts that are neutral in nature, they are not always very effective in resolving civil conflicts. This is partly caused by the inability or the reluctance of the international community to intervene early in civil conflicts and apply effective mediation strategies to resolve the issues between disputants before the conflict escalates. However, it is also caused by the reactions of policy makers towards biased military interventions. The international community tends to be unresponsive towards biased military interventions. Strong reactions to foreign military interventions are quite rare. If the international community wants to resolve civil conflicts, then policy makers also have to find ways to reduce the occurrence of foreign military interventions. Understanding the motives of interveners can be useful for finding effective methods to minimize biased military interventions and for preventing their adverse effects on civil conflicts. This study shows that rivalry and militarized disputes with the conflict state motivate foreign powers to intervene on the rebel side in civil conflicts. If the international community focuses on resolving these interstate disputes through diplomatic methods including mediation, then the frequency of biased military interventions can decrease. Similarly, foreign powers that are likely to intervene due to alliance or historical ties can be persuaded by the international community to 216 participate in neutral peacekeeping efforts. In other words, if the international community can predict which external states are likely to intervene in a civil conflict, they can attempt to prevent these interventions by mediating between rival states or by attempting to persuade potential interveners to participate in neutral conflict resolution operations headed by objective third parties. While these methods might work for some potential interveners, others may not be dissuaded from intervening easily. In such cases, at least three options are available to the foreign policy community. One option is to include potential interveners in the mediation process along with the domestic party they are supporting. Although this might seem unconventional or problematic to other domestic parties involved in the civil conflict, excluding them from the mediation process can be more damaging in the long run. If mediators prefer to use this option, they should pay close attention to the motives of interveners. While interveners with certain motives, such as the protection of an ethnic group in a conflict state, might be eligible for inclusion in the negotiation process, third party states who want to exploit the natural resources of a conflict state would not be eligible. Therefore, the second option for the international community is to prevent biased military interventions by strengthening the international norms against it and by applying effective diplomatic and economic sanctions against interveners. Finally, increasing the quantity and quality of multilateral conflict resolution and conflict prevention efforts is one of the most effective ways to prevent biased military interventions. If the international community becomes more willing to commit resources and take early preventive 217 actions before civil conflicts escalate, then they will be more successful at deterring biased military interventions. Neither the current international norms nor the intensity of conflict resolution efforts are deterring foreign powers from intervening and making civil conflicts more intractable. However, the international community is also not benefitting from the willingness of some states to intervene militarily to protect victims of civil conflicts such as ethnic groups. The resources such third party states want to commit to a civil conflict can be effectively utilized in multilateral peacekeeping missions headed by neutral states. In other words, knowing the motives of potential interveners can be very useful in determining the best method to prevent them from intervening militarily on one side and making the conflict eventually more complicated for the international community to resolve in the long run. Future Research Directions The quantitative literature on the causes of foreign military interventions is still at a developing stage. The number of empirical studies analyzing the motives of interveners is quite small in the literature. More importantly, the existing studies are not using theoretical frameworks which are conducive to foreign policy analysis. By applying a modified realist framework, this study was able to shed more light on the motives of interveners than previous quantitative studies. However, more research still needs to be conducted to understand the motives behind third party interventions and future works should adopt theoretical frameworks that are more suitable to analyzing the foreign policy decision making calculus of potential interveners. 218 One trajectory for future research is to collect more data on other possible causes of biased military interventions and expand the scope of empirical analysis which will allow more empirical patterns to emerge. For instance, while this dissertation included the regime type of the conflict state in the analysis, it did not examine the political ideology of the rebel groups. Currently, data are not available about the political ideologies of opposition groups in all civil conflicts but it is reasonable to expect that foreign states sharing the same political ideology would be more willing to support the opposition militarily. Another promising trajectory for further research is to compare the causes of biased and neutral military interventions. Although neutral military interventions such as peacekeeping missions are conflict resolution efforts, states participating in such operations can have strategic international and domestic considerations motivating them as well. This type of comparative analysis is particularly important for its policy implications. As discussed in the previous section, understanding the motives of interveners can help the international community devise more effective measures to prevent biased interventions. If there are similarities between the motives of biased and neutral interveners, then it might be possible to persuade potential interveners to participate in neutral conflict resolution efforts. Thus, analyzing the factors that motivate states to resort to biased versus neutral military interventions would be fruitful for both theory and policy purposes. Examining the causes of foreign military interventions is also critical for understanding their impacts on civil conflict dynamics and outcomes. Interventions motivated by different reasons are likely to have different impacts on the duration, 219 intensity and outcomes of conflicts. Several studies that distinguish between the motives of interveners have already showed that there is a close relationship between the causes and consequences of interventions (i.e. Akcinaroglu and Radziszewski 2005, Gent 2008, Cunningham 2010). Therefore, another trajectory for future research is to study the effects of biased military interventions based on the particular motives of interveners. This kind of research will provide a more refined understanding of the consequences of interventions and help policy makers devise better strategies to minimize the occurrence of interventions with the most threatening short term and long term effects. This study has attempted to contribute to the scientific and systematic study of interventions, but there are still more questions to be answered in future research. Some of these important questions are outlined above based on their potential contribution to both theory and policy. The challenges posed by civil conflicts are already great and biased interventions complicate them even further. Theoretically driven empirical research on the causes and consequences of biased interventions can help policy makers fine-tune their conflict resolution methods and reduce the adverse effects of interventions on conflict states. 220 Appendix A: Civil Conflicts 1946-2002 ID Conflict Name Conflict Years 100 Afghanistan ? Mujahideen groups, UIFSA 1978-2001 101 Algeria ? GIA, MIA/FIS/AIS 1991-2002 102 Angola ? Unita 1975-1995, 1998-2002 103 Angola ? Cabinda 1991, 1994, 1996-1998, 2002 104 Argentina ? Military faction 1955 105 Argentina ? Military faction 1963 106 Argentina ? ERP 1974-1977 107 Azerbaijan ? Nagorno Karabakh 1992-1994 108 Azerbaijan ? Military faction 1993 109 Azerbaijan ? OPON forces 1995 110 Bangladesh ? JSS/SB 1975-1992 111 Bolivia ? Popular Revolutionary Movement 1946 112 Bolivia ? MNR 1952 113 Bolivia ? ELN 1967 114 Bosnia ? Serbian Republic of Bosnia and Herzegovina 1992-1995 115 Bosnia ? Western Bosnia 1993-1995 116 Bosnia ? Croatian Republic of Bosnia and Herzegovina 1993-1994 117 Burkina Faso ? Popular Front 1987 118 Burundi ? Military faction 1965 119 Burundi ? CNDD, Palipehutu 1991-1992, 1994-2002 120 Cambodia ? KR 1967-1975 121 Cambodia ? KR, KPNLF, FUNCINPEC 1978-1998 122 Cameroon ? UPC 1960-1961 123 Cameroon ? Military faction 1984 124 Central African Republic ? Military faction 2001-2002 125 Chad ? Anti-government groups 1966-1972, 1976-1984, 1986- 1987, 1989-1994, 1997-2002 126 Chile ? Military faction 1973 127 China ? Peoples Liberation Army 1946-1949 128 China ? Taiwanese insurgents 1947 129 China ? Tibet 1950 130 China ? Tibet 1956, 1959 131 Colombia ? FARC, ELN, M-19 1964-2002 132 Comoros ? Presidential guard 1989 221 133 Comoros ? MPA 1997 134 Congo ? Cobras, Cocoyes, Ninjas, Ntsiloulous 1993-1994, 1997-1999, 2002 135 Costa Rica ? National Liberation Army 1948 136 Cote d?Ivoire ? MJP, MPCI, MPIGO 2002 137 Croatia ? Serbian Republic of Krajina 1992-1993, 1995 138 Cuba ? Cuban Revolution Movement 1953 139 Cuba ? 26th of July Movement 1956-1958 140 Cuba ? Cuban Revolutionary Council 1961 141 DRC/Zaire ? Katanga 1960-1962 142 DRC/Zaire ? South Kasai 1960-1962 143 DRC/Zaire ? CNL 1964-1965 144 DRC/Zaire ? Opposition Militias 1967 145 DRC/Zaire ? FLNC 1977-1978 146 DRC/Zaire ? AFDL, MLC, RCD 1996-2001 147 South Yemen ? Yemenite Socialist Party 1986 148 Djibouti ? FRUD 1991-1994, 1999 149 Dominican Republic ? Military faction 1965 150 Egypt ? al-Gamaa al-Islamiyya 1993-1998 151 El Salvador ? Military faction 1972 152 El Salvador ? FMLN 1979-1991 153 Equatorial Guinea ? Military faction 1979 154 Eritrea ? EIJM-AS 1997, 1999 155 Ethiopia ? Military faction 1960 156 Ethiopia ? ELF (Eritrea) 1964-1991 157 Ethiopia ? ALF 1975-1976 158 Ethiopia ? EPRP, EPDM, TPLF 1976-1991 159 Ethiopia ? WSLF (Ogaden) 1976-1983 160 Ethiopia ? OLF 1977-1978, 1980-1981, 1983- 1985, 1987-1991 161 Ethiopia ? OLF 1999-2002 162 Ethiopia ? ALF 1989-1991, 1996 163 Ethiopia ? ONLF (Ogaden) 1996, 1998-2002 164 Ethiopia ? Somali 1996-1997, 1999 165 France ? OAS 1961-1962 166 Gabon ? Military faction 1964 167 Gambia ? NRC 1981 168 Georgia ? Anti-government alliance 1991-1993 169 Georgia ? Abkhazia 1992-1993 170 Georgia ? South Ossetia 1992 171 Ghana ? NLC 1966 222 172 Ghana ? Military Faction 1981 173 Ghana ? Military Faction 1983 174 Greece ? DSE 1946-1949 175 Guatemala ? Military faction 1949 176 Guatemala ? Forces of Carlos Castillo Armas 1954 177 Guatemala ? FAR I, FAR II, EGP, URNG 1965-1995 178 Guinea ? RFDG 2000-2001 179 Guinea-Bissau ? Military Junta 1998-1999 180 Haiti ? Military faction 1989 181 Haiti ? Military faction 1991 182 India ? CPI 1948-1951 183 India ? CPI 1969-1971 184 India ? NNC (Nagaland) 1956-1959, 1961-1968 185 India ? MNF 1966-1968 186 India ? TNV (Tripura) 1978-1988 187 India ? PLA (Manipur) 1982-1988, 1992-2000 188 India ? Punjab/Khalistan 1983-1993 189 India ? Kashmir 1989-2002 190 India ? ABSU (Bodoland) 1989-1990, 1993-2002 191 India ? PWG, MCC 1990-1994, 1996-2002 192 India ? ULFA (Assam) 1990-1991, 1994-2002 193 India ? NSCN (Nagaland) 1992-1997, 2000 194 India ? ATTF, NLFT (Tripura) 1992-1993, 1995, 1997-2002 195 Indonesia ? Republic of South Moluccas 1950 196 Indonesia ? Darul Islam Movement 1953, 1958-1961 197 Indonesia ? OPM 1965, 1967-1969 198 Indonesia ? OPM 1976-1978 199 Indonesia ? Fretilin (East Timor) 1975-1989, 1992, 1997-1998 200 Indonesia ? GAM (Aceh) 1990-1991 201 Indonesia ? GAM (Aceh) 1999-2002 202 Iran ? Republic of Kurdistan 1946 203 Iran ? Republic of Azerbaijan 1946 204 Iran ? KDPI 1966-1968 205 Iran ? KDPI 1979-1988, 1990, 1993, 1996 206 Iran ? APCO 1979-1980 207 Iran ? MEK 1979-1982, 1986-1988, 1991- 1993, 1997, 1999-2001 208 Iraq ? Military faction 1958 209 Iraq ? Military faction 1959 210 Iraq ? Military faction 1963 223 211 Iraq ? KDP 1961-1970, 1973-1993, 1995- 1996 212 Iraq ? SCIRI 1982-1984, 1987, 1991-1996 213 Israel ? PLO groups 1949-1996, 2000-2002 214 Israel ? Hezbollah 1990-1999 215 Kenya ? Military faction 1982 216 Laos ? Pathet Lao 1959-1961, 1963-1973 217 Laos ? LRM 1989-1990 218 Lebanon ? Independent Nasserite Movement 1958 219 Lebanon ? LNM/NSF 1975-1976 220 Lebanon ? LNM/NSF, Amal, NUF 1982-1986, 1989-1990 221 Lesotho ? Military faction 1998 222 Liberia ? Military faction 1980 223 Liberia ? INPFL, NPFL 1989-1995 224 Liberia ? LURD 2000-2002 225 Macedonia ? UCK 2001 226 Madagascar ? Monima National Independence Movement 1971 227 Malaysia ? CPM 1958-1960 228 Malaysia ? CPM 1974-1975 229 Malaysia ? CPM 1981 230 Malaysia ? CCO 1963-1966 231 Mali ? MPA, FIAA 1990, 1994 232 Mauritania ? Polisario 1975-1978 233 Mexico ? EZLN 1994 234 Mexico ? EPR 1996 235 Moldova ? Dniestr Republic 1992 236 Morocco ? Military faction 1971 237 Morocco ? Polisario 1975-1989 238 Mozambique ? Renamo 1977-1992 239 Myanmar ? CPB / ABSDF 1948-1988, 1990-1992, 1994 240 Myanmar ? Arakan 1948-1988, 1991-1992, 1994 241 Myanmar ? Karen 1949-1992, 1995, 1997-2002 242 Myanmar ? Mon 1949-1963 243 Myanmar ? Mon 1990 244 Myanmar ? Mon 1996 245 Myanmar ? Kachin 1949-1950 246 Myanmar ? Kachin 1961-1992 247 Myanmar ? Karenni 1957 248 Myanmar ? Karenni 1992, 1996 224 249 Myanmar ? Shan 1959-1970, 1972-1973, 1976- 1988, 1993-2002 250 Myanmar ? Wa 1997 251 Nepal ? Nepali Congress 1960-1962 252 Nepal ? CPN-M 1996-2002 253 Nicaragua ? Contras/FDN 1978-1979, 1981-1989 254 Niger ? FLAA (Touareg) 1992 255 Niger ? UFRA (Touareg) 1997 256 Niger ? CRA (Air and Azawad) 1994 257 Niger ? FDR, FARS (Toubou) 1996-1997 258 Nigeria ? Military faction 1966 259 Nigeria ? Republic of Biafra 1967-1970 260 Oman ? PFLOAG/PFLO 1972-1975 261 Pakistan ? Mukti Bahini 1971 262 Pakistan ? Baluchi separatists 1974-1977 263 Pakistan ? MQM 1990, 1995-1996 264 Panama ? Military faction 1989 265 Papua New Guinea ? BRA 1989-1990, 1992-1996 266 Paraguay ? Opposition coalition 1947 267 Paraguay ? Military faction 1954 268 Paraguay ? Military faction 1989 269 Peru ? ELN, MIR 1965 270 Peru ? Sendero Luminoso, MRTA 1981-1999 271 Philippines ? HUK 1946-1954 272 Philippines ? CPP 1969-1995, 1997, 1999-2002 273 Philippines ? Mindanao 1970-1990, 1993-2002 274 Republic of Vietnam ? FNL 1955-1964 275 Romania ? National Salvation Front 1989 276 USSR ? Forest Brothers 1946-1948 277 USSR ? LNPA 1946 278 USSR ? BDPS 1946-1948 279 USSR ? UPA 1946-1950 280 USSR ? Republic of Armenia 1990-1991 281 USSR ? Azerbaijani Popular Front 1990 282 Russia ? Parliamentary Forces 1993 283 Russia ? Republic of Chechnya 1994-1996, 1999-2002 284 Russia ? Wahhabi Movement 1999 285 Rwanda ? FPR 1990-1994 286 Rwanda ? PALIR 1997-2002 287 Saudi Arabia ? JSM 1979 225 288 Senegal ? MFDC 1990, 1992-1993, 1995, 1997- 2001 289 Sierra Leone ? RUF, AFRC, Kamajors 1991-2000 290 Somalia ? Military faction 1978 291 Somalia ? SNM, SPM, USC/SNA, SRRC 1982-1984, 1986-1996, 2001- 2002 292 South Africa ? SWAPO 1966-1988 293 South Africa ? ANC 1981-1983, 1985-1988 294 Spain ? ETA 1978-1982, 1985-1987, 1991- 1992 295 Sri Lanka ? JVP 1971 296 Sri Lanka ? JVP 1989-1990 297 Sri Lanka ? LTTE, TELO 1984-2001 298 Sudan ? Anya Nya 1963-1972 299 Sudan ? Sudanese Communist Party 1971 300 Sudan ? Islamic Charter Front 1976 301 Sudan ? SPLM/A, NDA 1983-2002 302 Surinam ? SLA/Jungle Commando 1986-1988 303 Syria ? Military faction 1966 304 Syria ? Muslim Brotherhood 1979-1982 305 Tajikistan ? UTO, Movement for Peace 1992-1996, 1998 306 Thailand ? Military faction 1951 307 Thailand ? CPT 1974-1982 308 Togo ? MTD 1986 309 Togo ? Military faction 1991 310 Trinidad and Tobago ? Jamaat al-Muslimeen 1990 311 Tunisia ? Tunisian Armed Resistance 1980 312 Turkey ? PKK 1984-2002 313 Turkey ? Revolutionary Left 1991-1992 314 Uganda ? Military faction 1971 315 Uganda ? Military faction 1972 316 Uganda ? Military faction 1974 317 Uganda ? UNLF, FUNA, NRA, HSM, UPA 1978-1991 318 Uganda ? UDCA/LRA, ADF 1994-2002 319 UK ? IRA 1971-1991 320 Uruguay ? MLN/Tupamaros 1972 321 Uzbekistan ? IMU 2000 322 Venezuela ? Military faction 1962 323 Venezuela ? Military faction 1992 324 North Yemen ? Opposition coalition 1948 325 North Yemen ? Royalists 1962-1970 226 326 North Yemen ? National Democratic Front 1980-1982 327 Yemen ? Democratic Republic of Yemen 1994 328 Yugoslavia ? Republic of Slovenia 1991 329 Yugoslavia ? Republic of Croatia 1991 330 Yugoslavia ? UCK (Kosovo) 1998-1999 331 Zimbabwe (Rhodesia) ? ZAPU, ZANU, PF 1967-1968, 1973-1979 227 Appendix B: Military Interventions in Civil Conflicts ID Year Conflict State Intervener on government side Intervener on rebel side 100 1978 Afghanistan USSR 100 1979 Afghanistan USSR 100 1980 Afghanistan USSR US, Pakistan, Iran 100 1981 Afghanistan USSR 100 1982 Afghanistan USSR 100 1983 Afghanistan USSR 100 1984 Afghanistan USSR US 100 1985 Afghanistan USSR 100 1986 Afghanistan USSR US 100 1987 Afghanistan USSR 100 1988 Afghanistan USSR 100 1989 Afghanistan US, Pakistan 100 1990 Afghanistan Pakistan 100 1991 Afghanistan USSR Pakistan 100 1992 Afghanistan Russia Pakistan 100 1993 Afghanistan Russia Pakistan 100 1994 Afghanistan Russia Pakistan 100 1995 Afghanistan Russia Pakistan 100 1996 Afghanistan Pakistan 100 1997 Afghanistan Pakistan 100 1998 Afghanistan Pakistan 100 2001 Afghanistan US, UK, Australia, Canada, France, Germany, Italy, Turkey, Japan, Jordan, Netherlands, Poland, Russia 101 1993 Algeria Egypt 101 1996 Algeria Tunisia 101 1997 Algeria Sudan 101 2000 Algeria Morocco 101 2002 Algeria US 102 1975 Angola USSR, Cuba South Africa, US, Zaire, China, North Korea 102 1976 Angola USSR, Cuba South Africa, US, France, Zaire 102 1977 Angola USSR, Cuba South Africa, France, Zaire, Morocco 102 1978 Angola USSR, Cuba South Africa, France, Zaire, Morocco 102 1979 Angola USSR, Cuba South Africa, France, 228 Zaire, Morocco 102 1980 Angola USSR, Cuba South Africa 102 1981 Angola USSR, Cuba South Africa 102 1982 Angola USSR, Cuba South Africa 102 1983 Angola USSR, Cuba South Africa 102 1984 Angola USSR, Cuba South Africa, Zaire 102 1985 Angola USSR, Cuba South Africa, Zaire, US 102 1986 Angola USSR, Cuba South Africa, Zaire, US 102 1987 Angola USSR, Cuba South Africa, Zaire, US 102 1988 Angola USSR, Cuba South Africa, Zaire, US 102 1989 Angola USSR, Cuba, East Germany Zaire, US, Morocco 102 1990 Angola USSR, Cuba Zaire, US, Morocco 102 1991 Angola USSR, Cuba Zaire, US, Morocco 102 1992 Angola Zaire, US, Morocco 102 1993 Angola US, Israel Zaire, Morocco, Congo, Togo 102 1994 Angola Zaire, Morocco, Congo, Togo 102 1995 Angola Zaire, Morocco, Congo, Togo 102 1998 Angola DRC, Congo-Brazzaville Togo, Morocco 102 1999 Angola DRC, Congo-Brazzaville, Namibia Togo, Morocco 102 2000 Angola DRC, Congo-Brazzaville, Namibia Togo 102 2001 Angola DRC, Congo-Brazzaville, Namibia Togo 103 1991 Angola USSR Zaire 103 1994 Angola Congo 103 1996 Angola Congo 103 1998 Angola DRC, Congo-Brazzaville 106 1975 Argentina Chile, Uruguay, Paraguay, Bolivia, Brazil, US 106 1976 Argentina Chile, Uruguay, Paraguay, Bolivia, Brazil, US 106 1977 Argentina Chile, Uruguay, Paraguay, Bolivia, Brazil, US 107 1992 Azerbaijan Turkey Armenia 107 1993 Azerbaijan Turkey Armenia 107 1994 Azerbaijan Armenia 110 1975 Bangladesh India 110 1976 Bangladesh India 110 1977 Bangladesh India 113 1967 Bolivia US Cuba 229 114 1992 Bosnia and Herzegovina Yugoslavia (Serbia) 114 1993 Bosnia and Herzegovina Yugoslavia (Serbia) 114 1994 Bosnia and Herzegovina Yugoslavia (Serbia) 114 1995 Bosnia and Herzegovina Yugoslavia (Serbia) 116 1993 Bosnia and Herzegovina Croatia 119 1991 Burundi Rwanda, Tanzania 119 1995 Burundi France, US 119 1996 Burundi France Zaire, Tanzania 119 1998 Burundi DRC 119 1999 Burundi DRC, Zimbabwe 119 2000 Burundi DRC, Zimbabwe 119 2001 Burundi DRC 119 2002 Burundi DRC 120 1967 Cambodia Vietnam 120 1968 Cambodia Vietnam 120 1969 Cambodia Vietnam 120 1970 Cambodia US, South Vietnam Vietnam 120 1971 Cambodia US, South Vietnam Vietnam 120 1972 Cambodia US, South Vietnam Vietnam 120 1973 Cambodia US, South Vietnam Vietnam 120 1974 Cambodia Vietnam 120 1975 Cambodia US Vietnam 121 1978 Cambodia China Vietnam 121 1979 Cambodia Laos, Vietnam USSR, China 121 1980 Cambodia USSR, Vietnam China, Thailand 121 1981 Cambodia USSR, Vietnam China, Thailand 121 1982 Cambodia USSR, Vietnam China, Thailand, Singapore 121 1983 Cambodia USSR, Vietnam China, Thailand, Malaysia 121 1984 Cambodia USSR, Vietnam China, Thailand 121 1985 Cambodia USSR, Vietnam China, Thailand 121 1986 Cambodia USSR, Vietnam China, Thailand 121 1987 Cambodia USSR, Vietnam China, Thailand 121 1988 Cambodia USSR, Vietnam China, Thailand 121 1989 Cambodia USSR, Vietnam China, Thailand, US, Singapore 121 1990 Cambodia USSR, Vietnam China, US 121 1991 Cambodia USSR China, US 121 1994 Cambodia Thailand 230 122 1960 Cameroon France 124 2001 Central African Republic Libya, Chad 124 2002 Central African Republic Libya Chad 125 1968 Chad France 125 1969 Chad France 125 1970 Chad France Libya 125 1971 Chad France Libya 125 1972 Chad France Libya 125 1976 Chad France 125 1977 Chad France Libya 125 1978 Chad France Libya 125 1979 Chad France US, Sudan, Egypt 125 1980 Chad Libya US, France, Sudan, Egypt 125 1981 Chad Libya US, France, Sudan, Egypt 125 1982 Chad US, France, Sudan, Egypt 125 1983 Chad US, France, Sudan, Egypt, Zaire Libya 125 1984 Chad US, France, Sudan, Egypt, Zaire Libya 125 1986 Chad US, France, Zaire Libya 125 1987 Chad US, France, Zaire Libya 125 1989 Chad Libya, Sudan 125 1990 Chad US, France, Zaire, Iraq Libya, Sudan 125 1991 Chad France 125 1992 Chad France 125 1993 Chad France 125 1994 Chad France 125 1997 Chad France 125 1998 Chad France 125 1999 Chad France 125 2000 Chad France 125 2001 Chad France 125 2002 Chad France 127 1946 China US 127 1947 China USSR 127 1948 China US 127 1949 China US USSR 129 1950 China US 130 1956 China US 130 1959 China US 131 1990 Colombia US 231 131 1991 Colombia US 131 1992 Colombia US 131 1993 Colombia US 131 1994 Colombia US 131 1995 Colombia US 131 1996 Colombia US 131 1997 Colombia US 131 1998 Colombia US 131 1999 Colombia US 131 2000 Colombia US 131 2001 Colombia US 131 2002 Colombia US 134 1997 Congo Angola, Chad 134 1998 Congo Angola, Chad, France 134 1999 Congo Angola, Chad 134 2002 Congo Angola 135 1948 Costa Rica Nicaragua, Honduras Guatemala 136 2002 Cote d?Ivoire France, Angola Burkina Faso, Liberia 137 1992 Croatia Slovenia, Hungary Yugoslavia (Serbia) 137 1993 Croatia Yugoslavia (Serbia) 137 1995 Croatia US Serbian Republic of Bosnia and Herzegovina 140 1961 Cuba US 141 1960 DRC (Zaire) Belgium 141 1961 DRC (Zaire) Belgium 143 1964 DRC (Zaire) Belgium, USA Burundi 143 1965 DRC (Zaire) US Burundi 145 1977 DRC (Zaire) US, France, Belgium, China, Morocco, Uganda Angola 145 1978 DRC (Zaire) US, France, Belgium, China Angola 146 1996 DRC (Zaire) Uganda, Rwanda, Zambia 146 1997 DRC (Zaire) Uganda, Rwanda, Zambia, Angola, Zimbabwe 146 1998 DRC (Zaire) Angola, Chad, Namibia, Zimbabwe, Sudan Rwanda, Uganda, Burundi 146 1999 DRC (Zaire) Angola, Chad, Namibia, Zimbabwe, Sudan Rwanda, Uganda, Burundi 146 2000 DRC (Zaire) Angola, Namibia, Zimbabwe Rwanda, Uganda 146 2001 DRC (Zaire) Angola, Namibia, Zimbabwe Rwanda, Uganda 148 1991 Djibouti France 148 1992 Djibouti France 232 148 1993 Djibouti France 148 1994 Djibouti France 148 1999 Djibouti France, Ethiopia Eritrea 149 1965 Dominican Republic US 150 1993 Egypt US Iran, Sudan 150 1994 Egypt US Sudan 150 1995 Egypt US Sudan 150 1996 Egypt Sudan 150 1997 Egypt US Iran, Sudan 150 1998 Egypt Sudan 152 1980 El Salvador US Nicaragua, Cuba, USSR, Vietnam 152 1981 El Salvador US Nicaragua, Cuba, USSR, Vietnam 152 1982 El Salvador US, Venezuela, Honduras Nicaragua, Cuba, USSR, Vietnam 152 1983 El Salvador US Nicaragua, Cuba, USSR, Vietnam 152 1984 El Salvador US, Honduras Nicaragua, Cuba, USSR, Vietnam 152 1985 El Salvador US, Honduras Nicaragua, Cuba, USSR, Vietnam 152 1986 El Salvador US Nicaragua, Cuba, USSR, Vietnam 152 1987 El Salvador US, Honduras Nicaragua, Cuba, USSR, Vietnam 152 1988 El Salvador US Nicaragua, Cuba, USSR, Vietnam 152 1989 El Salvador US Nicaragua, Cuba, USSR, Vietnam 152 1990 El Salvador US Nicaragua, Cuba, USSR, Vietnam 152 1991 El Salvador US Nicaragua, Cuba, USSR, Vietnam 154 1997 Eritrea Sudan 154 1999 Eritrea Sudan, Ethiopia 156 1973 Ethiopia Syria, Iraq, Libya 156 1974 Ethiopia Syria, Iraq, Libya 156 1989 Ethiopia USSR Sudan 156 1990 Ethiopia USSR, Israel Syria, Libya, Iraq 158 1976 Ethiopia US Sudan 158 1977 Ethiopia USSR, Israel, Libya Sudan 158 1978 Ethiopia USSR, Israel, East Germany Sudan 158 1979 Ethiopia USSR, East Germany Sudan 158 1980 Ethiopia USSR, East Germany Sudan 233 158 1981 Ethiopia USSR, East Germany Sudan 158 1982 Ethiopia USSR, East Germany Sudan 158 1983 Ethiopia USSR, Israel, East Germany Sudan 158 1984 Ethiopia USSR, Israel, East Germany, Cuba Sudan 158 1985 Ethiopia USSR, East Germany, Cuba, North Korea Sudan 158 1986 Ethiopia USSR, East Germany, North Korea Sudan 158 1987 Ethiopia USSR, East Germany, Cuba, North Korea Sudan 158 1988 Ethiopia USSR, East Germany, Cuba, North Korea Sudan 158 1989 Ethiopia USSR, East Germany, North Korea Sudan 158 1990 Ethiopia USSR, Israel Sudan 158 1991 Ethiopia Sudan 159 1976 Ethiopia Cuba 159 1977 Ethiopia Cuba Somalia 159 1978 Ethiopia Cuba Somalia 159 1979 Ethiopia Cuba 159 1980 Ethiopia Cuba 159 1981 Ethiopia Cuba 159 1982 Ethiopia Cuba 159 1983 Ethiopia Cuba 161 1999 Ethiopia Eritrea 163 1999 Ethiopia Eritrea 166 1964 Gabon France 167 1981 Gambia Senegal 169 1993 Georgia Russia 174 1947 Greece US Albania, Bulgaria, Yugoslavia 174 1948 Greece Albania, Bulgaria 174 1949 Greece Albania 176 1954 Guatemala US 177 1966 Guatemala US 177 1971 Guatemala US 177 1975 Guatemala US 177 1976 Guatemala US 177 1977 Guatemala US, Israel, Taiwan, Argentina 177 1978 Guatemala Israel, Taiwan, Argentina 177 1979 Guatemala Israel, Taiwan, Argentina 234 177 1980 Guatemala US, Israel, Taiwan, Argentina 177 1981 Guatemala US, Israel, Taiwan, Argentina 177 1982 Guatemala US, Israel, Taiwan, Argentina, Dominican Republic 177 1983 Guatemala US, Israel, Taiwan, Argentina 177 1984 Guatemala US 177 1985 Guatemala US 177 1986 Guatemala US 177 1987 Guatemala US 177 1988 Guatemala US 177 1989 Guatemala US 177 1990 Guatemala US 179 1998 Guinea-Bissau Guinea, Senegal 179 1999 Guinea-Bissau Guinea, Senegal 186 1983 India Bangladesh 186 1984 India Bangladesh 187 1992 India Bangladesh, Pakistan 187 1993 India Bangladesh 189 1989 India Pakistan 189 1990 India Pakistan 189 1991 India Pakistan 190 1994 India Bhutan 190 1996 India Bhutan 190 2000 India Bhutan 190 2001 India Bangladesh, Bhutan 191 1998 India Pakistan 191 2000 India Pakistan 192 1990 India Bangladesh, Bhutan 192 1991 India Bangladesh, Bhutan 192 1994 India Bangladesh, Bhutan 192 1995 India Myanmar Bangladesh, Bhutan 192 1996 India Bangladesh, Bhutan 192 1997 India Bangladesh, Bhutan 192 1998 India Bangladesh, Bhutan 192 1999 India Bangladesh, Bhutan 192 2000 India Pakistan, Bangladesh, Bhutan 192 2001 India Bangladesh, Bhutan 192 2002 India Bangladesh, Bhutan 235 193 1992 India Bangladesh, Pakistan 193 1993 India Pakistan 193 1996 India Bangladesh, Pakistan, Burma 193 2000 India China 194 2001 India Bangladesh 194 2002 India Bangladesh 196 1953 Indonesia Malaysia, US 196 1958 Indonesia US, Czechoslovakia, USSR 196 1959 Indonesia US 198 1976 Indonesia US 198 1977 Indonesia US 198 1978 Indonesia US 199 1976 Indonesia US 199 1977 Indonesia US 199 1978 Indonesia US 199 1979 Indonesia US 199 1980 Indonesia US 199 1981 Indonesia US 199 1982 Indonesia US 199 1983 Indonesia US 199 1984 Indonesia US 199 1985 Indonesia US 199 1986 Indonesia US 199 1987 Indonesia US 199 1988 Indonesia US 199 1989 Indonesia US 199 1992 Indonesia US 199 1997 Indonesia US 199 1998 Indonesia US 200 1990 Indonesia US 200 1991 Indonesia US 202 1946 Iran USSR 203 1946 Iran USSR 204 1966 Iran Iraq 205 1979 Iran Iraq 205 1981 Iran Iraq 206 1979 Iran Iraq 207 1986 Iran Iraq 207 1987 Iran Iraq 207 1988 Iran Iraq 236 207 1991 Iran Iraq 207 1992 Iran Iraq 207 1993 Iran Iraq 207 1997 Iran Iraq 207 1999 Iran Iraq 207 2000 Iran Iraq 207 2001 Iran Iraq 211 1961 Iraq Iran, Israel 211 1962 Iraq Iran, Israel 211 1963 Iraq Syria Iran, Israel 211 1964 Iraq Iran, Israel 211 1965 Iraq Iran, Israel 211 1966 Iraq Iran, Israel 211 1967 Iraq Iran, Israel 211 1968 Iraq Iran, Israel 211 1969 Iraq Iran, Israel 211 1970 Iraq Iran, Israel 211 1973 Iraq USSR Iran, Israel 211 1974 Iraq USSR Iran, Israel 211 1975 Iraq USSR Iran 211 1983 Iraq Turkey Iran 211 1986 Iraq Turkey Iran 211 1987 Iraq Turkey 211 1989 Iraq Syria 211 1996 Iraq Iran, US 212 1982 Iraq Iran 212 1983 Iraq Iran 212 1984 Iraq Iran 212 1987 Iraq Iran 212 1991 Iraq Iran 212 1992 Iraq Iran 212 1993 Iraq Iran 212 1994 Iraq Iran 212 1995 Iraq Iran 212 1996 Iraq Iran 213 1949 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1950 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1951 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1952 Israel US Jordan, Syria, Egypt, Iran, Iraq 237 213 1953 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1954 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1955 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1956 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1957 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1958 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1959 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1960 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1961 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1962 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1963 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1964 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1965 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1966 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1967 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1968 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1969 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1970 Israel US Jordan, Syria, Egypt, Iran, Iraq 213 1971 Israel US Syria, Egypt, Iran, Iraq 213 1972 Israel US Syria, Egypt, Iran, Iraq 213 1973 Israel US Syria, Egypt, Iran, Iraq 213 1974 Israel US Syria, Egypt, Iran, Iraq 213 1975 Israel US Syria, Egypt, Iraq, Libya, Algeria, South Yemen 213 1976 Israel US Egypt, Iraq, Libya 213 1977 Israel US Syria, Iraq, Libya 213 1978 Israel US Syria, Iraq, Libya, Algeria, South Yemen, USSR, Cuba 213 1979 Israel US Syria, Libya, Yemen, USSR, Pakistan, South 238 Yemen 213 1980 Israel US Syria, Libya, Algeria, USSR 213 1981 Israel US Libya 213 1982 Israel US Syria, Libya, Algeria, South Yemen, USSR, China, North Korea 213 1983 Israel US Syria, Libya 213 1984 Israel US Syria 213 1985 Israel US Syria 213 1986 Israel US Syria 213 1987 Israel US Syria 213 1988 Israel US Syria 213 1989 Israel US Syria, Lebanon 213 1990 Israel US 213 1991 Israel US 213 1992 Israel US 213 1993 Israel US Syria, Iran 213 1994 Israel US Syria, Iran 213 1995 Israel US Syria 213 1996 Israel US 213 2000 Israel US 213 2001 Israel US Syria, Iran 213 2002 Israel US Syria, Iran 214 1990 Israel US Syria, Iran 214 1991 Israel US Syria, Iran 214 1992 Israel US Syria, Iran 214 1993 Israel US Syria, Iran 214 1994 Israel US Syria, Iran 214 1995 Israel US Syria, Iran 214 1996 Israel US Syria, Iran 214 1997 Israel US Syria, Iran 214 1998 Israel US Syria, Iran 214 1999 Israel US Syria, Iran 216 1959 Laos US, Thailand, South Vietnam North Vietnam, USSR 216 1960 Laos US, Thailand, South Vietnam USSR, China, North Vietnam 216 1961 Laos US, Thailand, South Vietnam USSR, China, North Vietnam 216 1963 Laos US, Thailand, South Vietnam USSR, North Vietnam 216 1964 Laos US, Thailand, South Vietnam USSR, North Vietnam 239 216 1965 Laos US, Thailand, South Vietnam USSR, North Vietnam 216 1966 Laos US, Thailand, South Vietnam USSR, North Vietnam 216 1967 Laos US, Thailand, South Vietnam USSR, North Vietnam 216 1968 Laos US, Thailand, South Vietnam USSR, North Vietnam 216 1969 Laos US, Thailand, South Vietnam USSR, North Vietnam 216 1970 Laos US, Thailand, South Vietnam USSR, North Vietnam 216 1971 Laos US, Thailand, South Vietnam USSR, North Vietnam 216 1972 Laos US, Thailand, South Vietnam USSR, North Vietnam 216 1973 Laos US, Thailand, South Vietnam USSR, North Vietnam 217 1989 Laos USSR 217 1990 Laos USSR 218 1958 Lebanon US, UK Syria 219 1975 Lebanon Syria, Israel 219 1976 Lebanon Syria, Israel Libya 220 1983 Lebanon Israel, US, France Syria 220 1984 Lebanon US, France 220 1985 Lebanon Syria 220 1989 Lebanon Syria Iraq 220 1990 Lebanon Algeria, Syria 221 1998 Lesotho Botswana, South Africa 223 1989 Liberia US Cote d?Ivoire, Burkina Faso, Libya 223 1990 Liberia US, Senegal, Nigeria, Ghana Cote d?Ivoire, Burkina Faso 224 2000 Liberia Guinea 224 2001 Liberia Guinea 224 2002 Liberia Guinea 225 2001 Macedonia Bulgaria, Ukraine Kosovo 227 1958 Malaysia UK 227 1959 Malaysia UK 227 1960 Malaysia UK 228 1974 Malaysia UK, Australia, New Zealand 229 1981 Malaysia UK, Australia, New Zealand 230 1963 Malaysia UK, Australia Indonesia 230 1964 Malaysia UK, Australia Indonesia 240 230 1965 Malaysia UK, Australia Indonesia 230 1966 Malaysia UK, Australia Indonesia 231 1990 Mali Libya 231 1994 Mali Libya 232 1975 Mauritania Morocco Algeria 232 1976 Mauritania Morocco, France Algeria, Libya 232 1977 Mauritania Morocco, France Algeria, Libya 232 1978 Mauritania Morocco, France Algeria, Libya 235 1992 Moldova Romania 237 1975 Morocco US, France, Saudi Arabia, Mauritania Algeria, Libya 237 1976 Morocco US, France, Saudi Arabia, Mauritania Algeria, Libya 237 1977 Morocco US, France, Saudi Arabia, Mauritania Algeria, Libya 237 1978 Morocco US, France, Saudi Arabia, Mauritania Algeria, Libya 237 1979 Morocco US, France, Saudi Arabia, Mauritania Algeria, Libya 237 1980 Morocco US, France, Saudi Arabia Algeria, Libya 237 1981 Morocco US, France, Saudi Arabia Algeria, Libya 237 1982 Morocco US, France, Saudi Arabia Algeria, Libya 237 1983 Morocco US, France, Saudi Arabia Algeria, Libya 237 1984 Morocco US, France, Saudi Arabia Algeria, Libya 237 1985 Morocco US, France, Saudi Arabia Algeria, Libya 237 1986 Morocco US, France, Saudi Arabia Algeria, Libya 237 1987 Morocco US, France, Saudi Arabia Algeria, Libya 237 1988 Morocco US, France, Saudi Arabia Algeria, Libya 237 1989 Morocco US, France, Saudi Arabia Algeria, Libya 238 1977 Mozambique USSR, Cuba Rhodesia 238 1978 Mozambique USSR, Cuba Rhodesia 238 1979 Mozambique USSR, Cuba Rhodesia 238 1980 Mozambique USSR, Cuba Rhodesia/Zimbabwe, South Africa 238 1981 Mozambique USSR, Cuba South Africa 238 1982 Mozambique USSR, Cuba, Zimbabwe, Tanzania South Africa 238 1983 Mozambique USSR, Cuba, Zimbabwe South Africa 238 1984 Mozambique USSR, Cuba, Zimbabwe, UK South Africa 238 1985 Mozambique USSR, Cuba, Zimbabwe, Tanzania, UK South Africa 238 1986 Mozambique USSR, Cuba, Zimbabwe, Tanzania, UK, Italy South Africa, Malawi 238 1987 Mozambique USSR, Cuba, Zimbabwe, South Africa, Malawi 241 Tanzania, UK, Italy 238 1988 Mozambique USSR, Cuba, Zimbabwe, Tanzania, UK, Italy, France, East Germany South Africa 238 1989 Mozambique USSR, Cuba, Zimbabwe, UK, Italy, France, East Germany South Africa 238 1990 Mozambique USSR, Cuba, Zimbabwe, UK, Italy, France South Africa 238 1991 Mozambique USSR, UK, Italy, France South Africa 238 1992 Mozambique UK, Italy, France 239 1948 Myanmar China 239 1949 Myanmar China 239 1950 Myanmar China 239 1951 Myanmar China 239 1952 Myanmar China 239 1953 Myanmar China 239 1954 Myanmar China 239 1955 Myanmar China 239 1956 Myanmar China 239 1957 Myanmar China 239 1958 Myanmar China 239 1959 Myanmar China 239 1960 Myanmar China 239 1961 Myanmar China 239 1962 Myanmar China 239 1963 Myanmar China 239 1964 Myanmar China 239 1965 Myanmar China 239 1966 Myanmar China 239 1967 Myanmar China 239 1968 Myanmar China 239 1969 Myanmar China 239 1970 Myanmar China 239 1971 Myanmar China 239 1972 Myanmar China 239 1973 Myanmar China 239 1974 Myanmar China 239 1975 Myanmar China 239 1976 Myanmar China 239 1977 Myanmar China 239 1978 Myanmar China 239 1979 Myanmar China 242 239 1990 Myanmar China 239 1991 Myanmar China 239 1992 Myanmar China 239 1994 Myanmar China 240 1991 Myanmar China 240 1992 Myanmar China Libya 240 1994 Myanmar China 241 1989 Myanmar China India 241 1990 Myanmar China 241 1991 Myanmar China 241 1992 Myanmar China 241 1995 Myanmar China 241 1997 Myanmar China 241 1998 Myanmar China 241 2000 Myanmar China, India, Pakistan 241 2001 Myanmar China 241 2002 Myanmar China, India 243 1990 Myanmar China 244 1996 Myanmar China 246 1961 Myanmar China 246 1962 Myanmar China 246 1963 Myanmar China 246 1964 Myanmar China 246 1965 Myanmar China 246 1966 Myanmar China 246 1967 Myanmar China 246 1968 Myanmar China 246 1969 Myanmar China 246 1970 Myanmar China 246 1971 Myanmar China 246 1972 Myanmar China 246 1973 Myanmar China 246 1974 Myanmar China 246 1989 Myanmar China India 246 1990 Myanmar China 246 1991 Myanmar China 246 1992 Myanmar China 248 1992 Myanmar China 248 1996 Myanmar China 249 1993 Myanmar China 249 1994 Myanmar China 243 249 1995 Myanmar China 249 1996 Myanmar China 249 1997 Myanmar China 249 1998 Myanmar China 249 1999 Myanmar China 249 2000 Myanmar China, India, Pakistan Thailand 249 2001 Myanmar China 249 2002 Myanmar China, India 250 1997 Myanmar China 252 1997 Nepal India 252 2000 Nepal UK 252 2001 Nepal UK, India 252 2002 Nepal UK, US, India 253 1978 Nicaragua Cuba, Panama, Venezuela 253 1979 Nicaragua Cuba, Panama, Venezuela 253 1981 Nicaragua USSR, Cuba US, Honduras 253 1982 Nicaragua USSR, Cuba, East Germany US, Honduras 253 1983 Nicaragua USSR, Cuba, East Germany US, Honduras 253 1984 Nicaragua USSR, Cuba, East Germany US, Honduras 253 1985 Nicaragua USSR, Cuba, East Germany US, Honduras 253 1986 Nicaragua USSR, Cuba, East Germany US, Honduras 253 1987 Nicaragua USSR, Cuba, East Germany US, Honduras 253 1988 Nicaragua USSR, Cuba, East Germany US, Honduras 253 1989 Nicaragua USSR, Cuba, East Germany US, Honduras 255 1997 Niger China, Libya 257 1997 Niger China, Libya 259 1967 Nigeria USSR, UK, Egypt 259 1968 Nigeria USSR, UK, Egypt France 259 1969 Nigeria USSR, UK, Egypt France 259 1970 Nigeria USSR, UK, Egypt France 260 1972 Oman UK, Jordan, Iran, Saudi Arabia South Yemen, China, Iraq 260 1973 Oman UK, Jordan, Iran, Saudi Arabia South Yemen, China, Iraq, USSR 260 1974 Oman UK, Jordan, Iran South Yemen, China, Iraq 260 1975 Oman UK, Jordan, Iran, US South Yemen, China, Iraq 261 1971 Pakistan India 244 262 1974 Pakistan Iran Afghanistan 262 1975 Pakistan Iran Afghanistan 262 1976 Pakistan Afghanistan 262 1977 Pakistan Afghanistan 264 1989 Panama US 265 1989 Papua New Guinea Australia 265 1990 Papua New Guinea Australia 265 1992 Papua New Guinea Australia 265 1993 Papua New Guinea Australia 265 1994 Papua New Guinea Australia 265 1995 Papua New Guinea Australia 265 1996 Papua New Guinea Australia 270 1985 Peru US 270 1987 Peru US 270 1990 Peru US 271 1950 Philippines US 271 1951 Philippines US 271 1952 Philippines US 271 1953 Philippines US 271 1954 Philippines US 272 1969 Philippines US 272 1970 Philippines US 272 1971 Philippines US 272 1972 Philippines US 272 1973 Philippines US 272 1974 Philippines US 272 1975 Philippines US 272 1976 Philippines US 272 1977 Philippines US 272 1978 Philippines US 272 1979 Philippines US 272 1980 Philippines US 272 1981 Philippines US 272 1982 Philippines US 272 1983 Philippines US 272 1984 Philippines US 272 1985 Philippines US 272 1986 Philippines US 272 1987 Philippines US 272 1988 Philippines US 272 1989 Philippines US 245 272 1990 Philippines US 272 1991 Philippines US 272 1992 Philippines US 272 1993 Philippines US 272 1994 Philippines US 272 1995 Philippines US 272 1997 Philippines US 272 1999 Philippines US 272 2000 Philippines US 272 2001 Philippines US 272 2002 Philippines US 273 1970 Philippines US Malaysia 273 1971 Philippines US Malaysia 273 1972 Philippines US Malaysia, Libya 273 1973 Philippines US Malaysia, Libya 273 1974 Philippines US Malaysia, Libya 273 1975 Philippines US Malaysia, Libya 273 1976 Philippines US Malaysia, Libya 273 1977 Philippines US Malaysia, Libya 273 1978 Philippines US Malaysia, Libya 273 1979 Philippines US Malaysia, Libya 273 1980 Philippines US Malaysia, Libya 273 1981 Philippines US Malaysia, Libya 273 1982 Philippines US Malaysia, Libya 273 1983 Philippines US Malaysia, Libya 273 1984 Philippines US Malaysia, Libya 273 1985 Philippines US Malaysia, Libya 273 1986 Philippines US Malaysia, Libya 273 1987 Philippines US Malaysia, Libya 273 1988 Philippines US Malaysia, Libya 273 1989 Philippines US 273 1990 Philippines US 273 1993 Philippines US 273 1994 Philippines US Libya, Pakistan 273 1995 Philippines US 273 1996 Philippines US 273 1997 Philippines US 273 1998 Philippines US 273 1999 Philippines US 273 2000 Philippines US 273 2001 Philippines US 246 273 2002 Philippines US 274 1960 Republic of Vietnam North Vietnam, China, USSR 274 1961 Republic of Vietnam US North Vietnam, China, USSR 274 1962 Republic of Vietnam US North Vietnam, China, USSR 274 1963 Republic of Vietnam US North Vietnam, China, USSR 274 1964 Republic of Vietnam US, New Zealand, Australia, South Korea North Vietnam, China, USSR 285 1990 Rwanda DRC, France, Belgium Uganda 285 1991 Rwanda France, Belgium Uganda 285 1992 Rwanda France, Belgium Uganda 285 1993 Rwanda France, Belgium Uganda 285 1994 Rwanda France Uganda 286 1998 Rwanda DRC 286 1999 Rwanda DRC 286 2000 Rwanda Kenya DRC 286 2001 Rwanda DRC 286 2002 Rwanda Burundi DRC 287 1979 Saudi Arabia US, France 288 1990 Senegal US Libya, Iraq, Guinea Bissau 288 1992 Senegal US 288 1993 Senegal US 288 1995 Senegal US 288 1997 Senegal US 288 1998 Senegal US 288 1999 Senegal Guinea Bissau 288 2000 Senegal US 288 2001 Senegal US, France 289 1991 Sierra Leone Liberia, Libya 289 1992 Sierra Leone Liberia 289 1993 Sierra Leone Liberia 289 1994 Sierra Leone Liberia 289 1995 Sierra Leone Liberia 289 1996 Sierra Leone Liberia 289 1997 Sierra Leone Liberia 289 1998 Sierra Leone Liberia, Libya, Burkina Faso 289 1999 Sierra Leone Liberia, Libya, Burkina Faso 289 2000 Sierra Leone UK Liberia 291 1982 Somalia US, Italy Ethiopia 247 291 1983 Somalia US Ethiopia 291 1984 Somalia US Ethiopia 291 1986 Somalia US Ethiopia 291 1987 Somalia US Ethiopia 291 1988 Somalia US Ethiopia 291 1989 Somalia US Ethiopia 291 2002 Somalia Eritrea, Djibouti, Saudi Arabia, Libya Ethiopia 292 1966 South Africa Zambia, USSR 292 1967 South Africa Zambia, USSR 292 1968 South Africa Zambia, USSR 292 1969 South Africa Zambia, USSR 292 1970 South Africa Zambia, USSR 292 1971 South Africa Zambia, USSR 292 1972 South Africa Zambia, USSR 292 1973 South Africa Zambia, USSR 292 1974 South Africa Zambia, Angola, USSR 292 1975 South Africa Zambia, Angola, USSR, China, North Korea 292 1976 South Africa Zambia, Angola, USSR, China, Cuba 292 1977 South Africa Zambia, Angola, USSR, Cuba 292 1978 South Africa Zambia, Angola, USSR, Cuba 292 1979 South Africa Zambia, Angola, Cuba 292 1980 South Africa Angola, Cuba 292 1981 South Africa Angola, Cuba 292 1982 South Africa Angola, Cuba 292 1983 South Africa Angola, Cuba, China 292 1984 South Africa Angola, Cuba 292 1985 South Africa Angola, Cuba 292 1986 South Africa Angola, Cuba 292 1987 South Africa Angola, Cuba 292 1988 South Africa Angola, Cuba 293 1981 South Africa Zambia, Angola, Tanzania, Mozambique, USSR, Cuba, East Germany 293 1982 South Africa Zambia, Angola, Tanzania, Mozambique, USSR, Cuba, East Germany 293 1983 South Africa Zambia, Angola, Tanzania, Mozambique, USSR, Cuba, East Germany 293 1985 South Africa Zambia, Angola, Tanzania, 248 USSR, Cuba, East Germany 293 1986 South Africa Zambia, Angola, Tanzania, USSR, Cuba, East Germany 293 1987 South Africa Zambia, Angola, Tanzania, USSR, Cuba, East Germany 293 1988 South Africa Zambia, Angola, Tanzania, USSR, Cuba, East Germany 294 1987 Spain France 294 1991 Spain France 294 1992 Spain France 295 1971 Sri Lanka India 296 1989 Sri Lanka India 296 1990 Sri Lanka India 297 1984 Sri Lanka Israel India 297 1985 Sri Lanka Israel, Pakistan, UK India 297 1986 Sri Lanka Israel, Pakistan India 297 1987 Sri Lanka Israel, Pakistan, India 297 1988 Sri Lanka India 297 1989 Sri Lanka India 297 1990 Sri Lanka India 297 1999 Sri Lanka US 297 2000 Sri Lanka US, India 298 1963 Sudan Israel, Uganda 298 1964 Sudan Israel, Uganda 298 1965 Sudan Israel, Uganda 298 1966 Sudan Israel, Uganda 298 1967 Sudan Israel, Uganda 298 1968 Sudan Israel, Uganda 298 1969 Sudan Israel, Uganda 298 1970 Sudan USSR Israel, Uganda 298 1971 Sudan USSR Israel, Uganda 298 1972 Sudan Israel, Uganda 300 1976 Sudan Libya 301 1983 Sudan US, Egypt Ethiopia, Libya 301 1984 Sudan US, Egypt Ethiopia, Libya 301 1985 Sudan US, Egypt, Libya, Iraq Ethiopia 301 1986 Sudan US, Egypt, Libya, Oman Ethiopia 301 1987 Sudan US, Egypt, Libya, Iraq Ethiopia, Israel 301 1988 Sudan US, Egypt, Libya, Iraq Ethiopia 249 301 1989 Sudan Libya, Iraq, China Ethiopia, Uganda 301 1990 Sudan Iraq Ethiopia, Uganda 301 1991 Sudan Libya, Iran Ethiopia, Uganda 301 1992 Sudan Libya, Iran, China Uganda 301 1993 Sudan Iran, China, Pakistan Uganda 301 1994 Sudan Iran, Iraq, South Africa Uganda, Israel, Egypt, Saudi Arabia 301 1995 Sudan Iran, Iraq, South Africa Uganda, Ethiopia, Eritrea 301 1996 Sudan Iran Uganda, Ethiopia, Eritrea 301 1997 Sudan Iran Uganda, Ethiopia, Eritrea 301 1998 Sudan Iran Uganda, Ethiopia, Eritrea 301 1999 Sudan Iran Uganda, Ethiopia, Eritrea 301 2002 Sudan Eritrea 304 1979 Syria USSR 304 1980 Syria USSR 304 1981 Syria USSR 304 1982 Syria USSR 305 1992 Tajikistan Russia, Uzbekistan Afghanistan 305 1993 Tajikistan Russia, Uzbekistan, Kazakhstan Afghanistan 305 1994 Tajikistan Russia Afghanistan 305 1995 Tajikistan Russia Afghanistan 305 1996 Tajikistan Russia Afghanistan 305 1998 Tajikistan Russia, Uzbekistan 307 1974 Thailand US China, North Vietnam 307 1975 Thailand US China, North Vietnam 307 1976 Thailand US China, Vietnam, Laos 307 1977 Thailand US China, Vietnam, Laos 307 1978 Thailand US China, Vietnam, Laos 307 1979 Thailand US China 307 1980 Thailand US 307 1981 Thailand US 307 1982 Thailand US 308 1986 Togo France, Zaire Ghana 310 1990 Trinidad and Tobago Libya 311 1980 Tunisia France 312 1984 Turkey US, Iraq Syria 312 1985 Turkey US Syria 312 1986 Turkey US Syria, Iran 312 1987 Turkey US Syria, Iran 312 1988 Turkey US Syria, Iran 312 1989 Turkey US Syria, Iran 250 312 1990 Turkey US Syria 312 1991 Turkey US Syria 312 1992 Turkey US Syria 312 1993 Turkey US Syria, Greece 312 1994 Turkey US Syria, Greece 312 1995 Turkey US Syria, Greece 312 1996 Turkey US Syria, Greece 312 1997 Turkey US Syria, Greece 312 1998 Turkey US Syria, Greece 312 1999 Turkey US Syria, Iran 312 2000 Turkey US 312 2001 Turkey US 312 2002 Turkey US 313 1991 Turkey US Syria 313 1992 Turkey US, Syria 314 1971 Uganda Tanzania 315 1972 Uganda Libya Tanzania 317 1978 Uganda USSR Tanzania 317 1979 Uganda Libya Tanzania 317 1980 Uganda Tanzania Sudan 317 1981 Uganda Tanzania, North Korea, Cuba, Ethiopia Libya 317 1982 Uganda UK, North Korea Libya 317 1983 Uganda UK, North Korea 317 1984 Uganda UK, North Korea 317 1985 Uganda UK, North Korea, Egypt Libya 317 1986 Uganda UK, Tanzania Sudan 317 1987 Uganda USSR, Libya 317 1988 Uganda Libya 318 1994 Uganda Sudan 318 1995 Uganda Sudan 318 1996 Uganda US Sudan, Zaire 318 1997 Uganda US, DRC Sudan 318 1998 Uganda US, DRC Sudan 318 1999 Uganda US, UK Sudan 318 2000 Uganda Sudan 318 2001 Uganda Sudan 319 1985 United Kingdom Libya 319 1986 United Kingdom Libya 319 1987 United Kingdom Libya 319 1988 United Kingdom Libya 319 1991 United Kingdom Libya 251 321 2000 Uzbekistan Kyrgyzstan Afghanistan 325 1962 North Yemen Egypt Jordan, Saudi Arabia 325 1963 North Yemen Egypt UK 325 1964 North Yemen Egypt UK 325 1965 North Yemen Egypt UK 325 1966 North Yemen Egypt UK 325 1967 North Yemen Egypt, USSR 325 1968 North Yemen Egypt, USSR 325 1969 North Yemen Egypt Saudi Arabia 325 1970 North Yemen Egypt Saudi Arabia 326 1980 North Yemen South Yemen 326 1981 North Yemen South Yemen 326 1982 North Yemen South Yemen 327 1994 Yemen Iraq Saudi Arabia 330 1998 Yugoslavia Albania 330 1999 Yugoslavia Albania, Belgium, Canada, Czech Republic, Denmark, France, German Federal Republic, Greece, Hungary, Iceland, Italy, Luxembourg, Netherlands, Norway, Poland, Portugal, Spain, Turkey, UK, US 331 1967 Zimbabwe (Rhodesia) South Africa USSR, Cuba, Zambia 331 1968 Zimbabwe (Rhodesia) South Africa USSR, Cuba, Zambia 331 1973 Zimbabwe (Rhodesia) South Africa USSR, China, Cuba, Zambia 331 1974 Zimbabwe (Rhodesia) South Africa USSR, China, Cuba, Zambia 331 1975 Zimbabwe (Rhodesia) South Africa China, Mozambique, Tanzania, Zambia 331 1976 Zimbabwe (Rhodesia) South Africa China, USSR, Cuba, Mozambique, Tanzania, Zambia, Botswana 331 1977 Zimbabwe (Rhodesia) China, USSR, Cuba, Mozambique, Tanzania, Zambia, Angola 331 1978 Zimbabwe (Rhodesia) China, USSR, Cuba, Mozambique, Tanzania, Zambia, Angola, Libya 331 1979 Zimbabwe (Rhodesia) China, USSR, Cuba, Mozambique, Tanzania, Zambia, Angola 252 Appendix C: Variable Descriptions and Coding Rules Identifiers State A: refers to third party states which include both actual and potential interveners State B: refers to states experiencing a civil conflict Dependent Variables Intervention: dichotomous measure of whether or not state A intervenes militarily in state B. Intervention consists of military aid and/or the use of military force by a third party state in support of either the government or the opposition in a civil conflict. Military interventions include the supply of troops, military equipment, intelligence and logistical support, air or naval support to one of the sides in civil conflict. 0 = No military intervention by state A in state B during year t 1 = Military intervention by state A supporting government or rebels in state B during year t Intervention Side: dichotomous measure of whether military intervention occurs on government or rebel side. This variable is only coded for third parties that actually intervene in state B. 0 = Military intervention by state A on rebel side in state B at year t 1 = Military intervention by state A on government side in state B at year t Main sources for dependent variables: 1) Regan?s dataset on military interventions in intrastate conflicts, 1945-1999 2) Tillema?s dataset on foreign overt military interventions, 1945-1985 3) Pearson and Baumann?s dataset on international military interventions, 1945-1988 4) Kisangani and Pickering?s dataset on international military intervention, 1989- 2005 5) UCDP/PRIO Armed Conflict Dataset v4-2010: Variables ?sideA2nd? and ?sideB2nd? identify the states that actively support one of the sides in a conflict with troops. 6) Minorities at Risk Project: External Support Variables Independent Variables a) International Variables Rivalry: dichotomous variable indicating whether rivalry exists between third party state and conflict state. 0 = No rivalry between state A and state B at t-1 1 = Rivalry between state A and state B at t-1 253 Rival intervention: dichotomous variable indicating whether a rival of state A intervenes militarily in state B. 0 = No military intervention in state B by a rival of state A at t-1 1 = Military intervention in state B by a rival of state A at t-1 Rival intervention on rebel side: dichotomous variable indicating whether a rival of state A intervenes militarily on rebel side in state B. 0 = No military intervention on rebel side in state B by a rival of state A at t-1 1 = Military intervention on rebel side in state B by a rival of state A at t-1 Source for rivalry variables: James Klein, Gary Goertz and Paul Diehl (2006), ?The New Rivalry Dataset 1816-2001?. Alliance: dichotomous variable indicating whether there is a formal alliance between third party state and conflict state. Defense pacts, neutrality pacts, non-aggression pacts and ententes are considered as alliances. 0 = No alliance between state A and state B at t-1 1 = Alliance between state A and state B at t-1 Ally intervention: dichotomous variable indicating whether an ally of state A intervenes militarily in state B. 0 = No military intervention in state B by an ally of state A at t-1 1 = Military intervention in state B by an ally of state A at t-1 Ally intervention on government side: dichotomous variable indicating whether an ally of state A intervenes militarily on government side in state B. 0 = No military intervention on government side in state B by an ally of state A at t-1 1 = Military intervention on government side in state B by an ally of state A at t-1 Source for alliance variables: Douglas Gibler (2009), Correlates of War (COW) Project - Formal Alliances Dataset v4.1 Militarized dispute: dichotomous variable indicating whether a militarized interstate dispute exists between third party state and conflict state. An interstate dispute can have five different hostility levels according to MID. The study excludes interstate disputes with hostility levels coded as 0 (no hostility) and 1 (no militarized action) and includes those that are coded as 2 (threat to use force), 3 (display of force), 4 (use of force) and 5 (war) according to MID. 0 = No militarized dispute between state A and state B at t-1 1 = Militarized dispute between state A and state B at t-1 Source: COW Project - Militarized Interstate Dispute (MID) Dataset v4.01 Historical ties: dichotomous variable indicating whether historical ties exist between state A and state B. The existence of historical ties indicates either that state B was a former colony of state A or that state B seceded from state A. 0 = No historical ties between state A and state B 254 1 = Historical ties between state A and state B Source: Paul Hensel (2009), Issue Correlates of War (ICOW) Project - Colonial History Dataset v0.4 Contiguous: dichotomous variable indicating whether ?type 1? contiguity relationship as determined by the COW Project exists between state A and state B. Type 1 contiguity means that state A and state B are separated by land or river border. 0 = state A and state B are not contiguous 1 = state A and state B are contiguous by land or river Source: COW Project ? Direct Contiguity Data v3.1 Same region: dichotomous variable indicating whether state A and state B are in the same geographical region as defined by the COW Project. There are five geographical regions: Africa, Asia, Americas, Europe and Mideast. 0 = state A and state B are not in the same region 1 = state A and state B are in the same region Source: This variable was generated by EUGene software (Bennett and Stam 2000). Secondary diamonds: dichotomous variable indicating whether state B has secondary diamond deposits. 0 = no secondary diamond deposits in state B 1 = secondary diamond deposits in state B Source: International Peace Research Institute (PRIO) ? Diamond Dataset http://www.prio.no/Data/Geographical-and-Resource-Datasets/Diamond-Resources Petroleum: dichotomous variable indicating whether state B has petroleum deposits. 0 = no petroleum deposits in state B 1 = petroleum deposits in state B Source: PRIO ? Petroleum Dataset v1.2 http://www.prio.no/Data/Geographical-and-Resource-Datasets/Petroleum-Dataset Trade: continuous variable measuring the amount of trade per dyad year between state A and state B in current US millions of dollars. Variables ?flow 1? and ?flow 2? in dyadic trade dataset reports the imports of state A from state B and the imports of state B from state A respectively. Trade is the sum of those two variables at t-1. Source: COW Project ? Dyadic Trade Dataset v2.01 b) Domestic Variables Democratic intervener: dichotomous variable indicating whether state A has a polity2 score of 7 or higher. 0 = state A?s polity2 score is less than 7 at t-1 1 = state A?s polity2 score is 7 or higher at t-1 255 Joint democracy: dichotomous variable indicating whether both state A and state B have a polity2 score of 7 or higher. 0 = state A and/or state B score less than 7 on polity2 scale at t-1 1 = state A and state B score 7 or higher on polity2 scale at t-1 Joint autocracy: dichotomous variable indicating whether both state A and state B have a polity2 score of 6 or less. 0 = state A and/or state B score higher than 6 on polity2 scale at t-1 1 = state A and state B score 6 or less on polity2 scale at t-1 Source for regime type variables: Polity IV Project Ethnic ties1: dichotomous variable indicating whether state A and state B have the same dominant ethnic group. 0 = state A and state B do not have the same dominant ethnic group 1 = state A and state B have the same dominant ethnic group Ethnic ties2: dichotomous variable indicating whether the dominant ethnic group in state A is the second largest ethnic group in state B and the dominant ethnic group in state B is the second largest ethnic group in state A. 0 = the dominant ethnic group in State A is not the second largest ethnic group in state B and vice versa 1 = the dominant ethnic group in State A is the second largest ethnic group in state B and vice versa Source for ethnic ties variables: Erik Gartzke and Kristian Gleditsch (2006). ?Identity and Conflict: Ties that Bind and Differences that Divide? European Journal of International Relations 12(1): 53-87. Replication data at http://privatewww.essex.ac.uk/~ksg Elite unrest: dichotomous variable indicating whether government crises and/or purges took place in state A. 0 = no elite unrest in state A at t-1 1 = elite unrest in state A at t-1 Mass unrest: dichotomous variable indicating whether general strikes, riots, anti- government demonstrations and/or civil conflicts took place in state A. 0 = no mass unrest in state A at t-1 1 = mass unrest in state A at t-1 Source for domestic unrest variables: Arthur Banks (2008). Cross-National Time- Series Data Archive, Domestic Conflict Event Data section. 256 c) Control Variables Major power: dichotomous variable indicating whether state A is categorized as major power by the COW Project. 0 = state A is not a major power 1 = state A is a major power Source: This variable was generated by EUGene software (Bennett and Stam 2000). CINC ratio: continuous variable measuring relative capabilities of state A to state B per dyad year by calculating the ratio of state A?s CINC (Composite Index of National Capability) score to state B?s CINC score at t-1. Source: COW Project ? National Material Capabilities Dataset v4.0 War involvement: dichotomous variable indicating whether state A is involved in an interstate war. The highest level of hostility (level 5) in MID dataset is categorized as an interstate war. 0 = state A is not involved in an interstate war at t-1 1 = state A is involved in an interstate war at t-1 Source: COW Project - Militarized Interstate Dispute (MID) Dataset v4.01 Cold War: dichotomous variable indicating whether state A-state B dyad year is between 1946 and 1989. 0 = post-Cold War dyad year 1 = Cold War dyad year d) Contextual Variables Democratic target: dichotomous variable indicating whether state B has a polity2 score of 7 or higher. 0 = state B?s polity2 score is less than 7 at t-1 1 = state B?s polity2 score is 7 or higher at t-1 Source: Polity IV Project Government conflict: dichotomous variable indicating whether the type of civil conflict in state B is governmental or territorial. 0 = territorial conflict 1 = governmental conflict Source: UCDP/PRIO Armed Conflict Dataset v4-2010 Battle deaths: discrete variable indicating the number of battle deaths in state B per conflict year. Source: PRIO Battle Deaths Dataset v3.0 257 Refugees: discrete variable indicating the number of refugees in state B per conflict year. Source: United Nations High Commissioner for Refugees (UNHCR), UNHCR Statistical Online Population Database Rebel fighting capacity: ordinal variable indicating the fighting capacity of rebels rated relative to government forces in state B. 1 = low 2 = moderate 3 = high Source: David Cunningham, Kristian Gleditsch and Idean Salehyan (2012). Non- State Actor Data v3.3 http://privatewww.essex.ac.uk/~ksg/eacd.html 258 Appendix D: Relogit Models Table D.1: Relogit Results for Full Sample Model 1 Model 3 Dependent Variable Intervention Intervention Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ 1.056**** (.283) 1.113**** (.289) Militarized dispute ~ 1.143**** (.288) 1.126**** (.289) Alliance ~ .117 (.180) .081 (.181) Historical ties 1.662**** (.369) 1.572**** (.367) Contiguous 1.533**** (.217) 1.567**** (.216) Same region .848**** (.181) .903**** (.181) Secondary diamonds .227 (.152) .225 (.151) Petroleum .405*** (.144) .412*** (.143) Trade (logged) ~ .106**** (.029) .095**** (.029) Rival intervention ~ 1.348**** (.161) 1.390**** (.169) Ally intervention ~ .576**** (.171) .538*** (.171) Democratic intervener ~ -.573**** (.140) - Ethnic ties 1 .742*** (.260) .691*** (.261) Ethnic ties 2 .889*** (.310) .906***(.316) Elite unrest ~ -.466**** (.096) -.438**** (.097) Mass unrest ~ -.274*** (.094) -.282*** (.095) Joint autocracy ~ - .434*** (.151) Joint democracy ~ - .243 (.318) Democratic target ~ -.128 (.193) - Government conflict .076 (.148) .074 (.148) Battle deaths (logged) ~ .342**** (.034) .328**** (.034) Rebel fighting capacity ~ -.441**** (.107) -.465**** (.107) War involvement ~ -.493**** (.132) -.469**** (.132) Major power 1.000**** (.189) 1.008**** (.189) CINC ratio (logged) ~ .369**** (.035) .357**** (.035) Cold War .039 (.107) .070 (.106) Spell years -2.200****(.104) -2.199****(.103) Constant -6.017**** (.301) -6.334**** (.339) N 199180 199180 Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 259 Table D.2: Relogit Results for Major Powers Model 1 Model 3 Dependent Variable Intervention Intervention Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ -.158 (.336) -.139 (.329) Militarized dispute ~ .950*** (.369) .969*** (.379) Alliance ~ 1.128**** (.236) 1.078**** (.233) Historical ties 1.140*** (.367) 1.177*** (.364) Contiguous 1.609* (.878) 1.503* (.814) Same region .120 (.307) .111 (.298) Secondary diamonds .267 (.220) .274 (.221) Petroleum .306* (.163) .306* (.162) Trade (logged) ~ .236**** (.065) .248**** (.067) Rival intervention ~ 1.194**** (.205) 1.168**** (.201) Ally intervention ~ .520**** (.134) .535**** (.138) Democratic intervener ~ .155 (.231) - Ethnic ties 1 -1.050 (.895) -.987 (.904) Ethnic ties 2 -.057 (.521) -.110 (.517) Elite unrest ~ -.276 (.260) -.344 (.336) Mass unrest ~ -.219*** (.82) -.182*** (.69) Joint autocracy ~ - .466* (.262) Joint democracy ~ - .046 (.340) Democratic target ~ -.118 (.321) - Government conflict .096 (.234) .101 (.230) Battle deaths (logged) ~ .275**** (.056) .274**** (.056) Rebel fighting capacity ~ -.411** (.181) -.396** (.179) War involvement ~ -.057 (.185) -.032 (.190) CINC ratio (logged) ~ .553**** (.084) .550**** (.084) Cold War .308** (.136) .310** (.137) Spell years -2.547**** (.208) -2.530**** (.207) Constant -5.920**** (.834) -6.267**** (.797) N 7317 7317 Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 260 Table D.3: Relogit Results for Democratic States Model 1 Model 2 Dependent Variable Intervention Intervention Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ 1.350*** (.526) 1.360*** (.517) Militarized dispute ~ 1.257*** (.435) 1.250*** (.428) Alliance ~ .300 (.284) .345 (.280) Historical ties 1.555**** (.359) 1.536**** (.356) Contiguous 1.491**** (.373) 1.452**** (.377) Same region .528**** (.141) .482**** (.136) Secondary diamonds -.392 (.252) -.377 (.256) Petroleum .413* (.242) .337 (.242) Trade (logged) ~ .262**** (.052) .278**** (.053) Rival intervention ~ 1.117**** (.202) 1.068**** (.205) Ally intervention ~ .407* (.245) .398* (.241) Ethnic ties 1 .594** (.249) .578** (.252) Ethnic ties 2 .721* (.436) .743* (.438) Elite unrest ~ -.069 (.151) -.081 (.151) Mass unrest ~ -.119 (.134) -.106 (.134) Democratic target ~ .227 (.319) .386 (.331) Government conflict .172 (.254) .126 (.250) Battle deaths (logged) ~ .311**** (.062) .279**** (.068) Refugees (logged) ~ - .101*** (.038) Rebel fighting capacity ~ -.349** (.166) -.330** (.154) War involvement ~ -.506** (.216) -.486** (.214) Major power 1.228**** (.222) 1.177**** (.226) CINC ratio (logged) ~ .517**** (.067) .522**** (.067) Cold War -.012 (.191) -.024 (.187) Spell years -2.205**** (.194) -2.200**** (.191) Constant -7.778**** (.525) -8.606**** (.607) N 62768 62768 Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 261 Table D.4: Relogit Results for Autocratic States Model 1 Model 2 Dependent Variable Intervention Intervention Independent Variables Coefficients (std. errors) Coefficients (std. errors) Rivalry ~ 1.099**** (.327) 1.066**** (.321) Militarized dispute ~ 1.073**** (.320) 1.034**** (.323) Alliance ~ -.130 (.240) -.100 (.237) Historical ties -.794 (.858) -.893 (.900) Contiguous 1.676**** (.239) 1.630**** (.237) Same region .993**** (.212) .994**** (.208) Secondary diamonds .453 (.399) .458 (.397) Petroleum .487*** (.176) .409** (.174) Trade (logged) ~ .027 (.036) .041 (.036) Rival intervention ~ 1.464**** (.207) 1.467**** (.204) Ally intervention ~ .687*** (.231) .634*** (.223) Ethnic ties 1 .864*** (.283) .846*** (.280) Ethnic ties 2 1.045*** (.403) 1.070*** (.398) Elite unrest ~ -.411**** (.111) -.391**** (.111) Mass unrest ~ -.331*** (.115) -.339*** (.115) Democratic target ~ -.265** (.108) -.268** (.109) Government conflict .100 (.176) .015 (.171) Battle deaths (logged) ~ .343**** (.040) .297**** (.040) Refugees (logged) ~ - .165**** (.038) Rebel fighting capacity ~ -.480**** (.130) -.447**** (.129) War involvement ~ -.207 (.184) -.206 (.181) Major power .608** (.301) .598** (.299) CINC ratio (logged) ~ .380**** (.044) .373**** (.044) Cold War .025 (.131) .088 (.137) Spell years -2.575**** (.139) -2.592**** (.139) Constant -5.982**** (.368) -7.387**** (.537) N 136412 136412 Notes: Robust standard errors clustered on dyads are reported in parentheses. *p? .10 **p? .05 ***p? .01 ****p? .001 (two-tailed tests) ~ variables are lagged one year. The coefficients for cubic splines are not reported. 262 Bibliography Akcinaroglu, Seden and Elizabeth Radziszewski. 2005. ?Expectations, Rivalries, and Civil War Duration.? International Interactions 31(4): 349-374. Balch-Lindsay, Dylan and Andrew J. Enterline. 2000. ?Killing Time: The World Politics of Civil War Duration.? International Studies Quarterly 44: 615-642. Balch-Lindsay, Dylan, Andrew J. Enterline, and Kyle A. Joyce. 2008. ?Third party intervention and Civil War Process.? Journal of Peace Research 45(3): 345-363. Baloyra, Enrique. 1982. El Salvador in Transition. Chapel Hill: University of North Carolina Press. Banks, Arthur S. 2008. Cross-National Time-Series Data Archive. Jerusalem, Israel: Databanks International. Barbieri, Katherine, Omar M. G. Keshk, and Brian Pollins. 2009. ?Trading Data: Evaluating Our Assumptions and Coding Rules.? Conflict Management and Peace Science 26(5): 471-491. Beck, Nathaniel and Jonathan N. Katz. 1995. ?What to do (and not to do) with Time-Series Cross-Section Data.? American Political Science Review 89(3): 634-647. Beck, Nathaniel, Jonathan N. Katz, and Richard Tucker. 1998. Taking Time Seriously: Time-Series-Cross-Section Analysis with a Binary Dependent Variable.? American Journal of Political Science 42(4): 1260-1288. Bennett, D. Scott and Allan Stam. 2000. ?EUGene: A Conceptual Manual.? International Interactions 26: 179-204. Blaufarb, Douglas. 1977. The Counterinsurgency Era: U.S. Doctrine and Performance, 1950 to the Present. New York: Free Press. Bremer, Stuart A. 1992. ?Dangerous Dyads: Conditions Affecting the Likelihood of Interstate War, 1816-1965.? Journal of Conflict Resolution 36(2): 309-341. Bueno de Mesquita, Bruce, Alastair Smith, Randolph M. Siverson, and James D. Marrow. 2003. The Logic of Political Survival. Cambridge, Mass.: MIT Press. Bueno de Mesquita, Bruce, James D. Marrow, Randolph M. Siverson, and Alastair Smith. 2004. ?Testing Novel Implications from the Selectorate Theory of War.? World Politics 56(3): 363-388. 263 Bueno de Mesquita, Bruce and George Downs. 2006. ?Intervention and Democracy.? International Organization 60(3): 627-649. Cable, Larry E. 1986. Conflict of Myths: The Development of American Counterinsurgency Doctrine and the Vietnam War. New York: New York University Press. Carment, David and Patrick James. 1995. ?Internal Conflicts and Interstate Ethnic Conflict: Toward a Crisis-Based Assessment of Irredentism.? Journal of Conflict Resolution 39(1): 82-109. Carment, David and Patrick James. 1996. ?Two-Level Games and Third-Party Intervention: Evidence from Ethnic Conflict in the Balkans and South Asia.? Canadian Journal of Political Science 29(3): 521-554. Carment, David, Patrick James and Dane Rowlands. 1997. ?Ethnic Conflict and Third Party Intervention: Riskiness, Rationality and Commitment.? In Enforcing Cooperation: Risky States and Intergovernmental Management of Conflict, eds. Gerald Schneider and Patricia A. Weitsman. New York: St. Martin?s Press. Carment, David, Patrick James and Zeynep Taydas. 2006. Who Intervenes? Ethnic Conflict and Interstate Crisis. Columbus: The Ohio State University Press. Cetinyan, Rupen. 2002. ?Ethnic Bargaining in the Shadow of Third-Party Intervention.? International Organization 56(3): 645-677. Clark, David H. and William Reed. 2003. ?A Unified Model of War Onset and Outcome.? The Journal of Politics 65(1): 69-91. Collier, Paul, Anka Hoeffler and Mans S?derbom. 2004. ?On the Duration of Civil War.? Journal of Peace Research 41(3): 253-73. Corbett, Edward M. 1972. The French Presence in Black Africa. Washington, D.C.: Black Orpheus. Correlates of War Project: Direct Contiguity Data v3.1. Data available at: http://www.correlatesofwar.org Correlates of War Project: Dyadic Trade Dataset v2.01. Data available at: http://www.correlatesofwar.org Correlates of War Project: Formal Alliances Dataset v4.1. Data available at: http://www.correlatesofwar.org 264 Correlates of War Project: Militarized Interstate Disputes (MID) Dataset v4.01. Data available at: http://www.correlatesofwar.org Correlates of War Project: National Material Capabilities Dataset v4.0. Data available at: http://www.correlatesofwar.org Correlates of War Project: COW Wars v.4.0, 1816-2007. Data available at: http://www.correlatesofwar.org Cronje, Susan. 1972. The World and Nigeria. London: Sidwick & Jackson. Cunningham, David E. 2006. ?Veto Players and Civil War Duration.? American Journal of Political Science 50(4): 875-892. Cunningham, David E. 2010. ?Blocking resolution: How external states can prolong civil wars.? Journal of Peace Research 47(2): 115-127. Cunningham, David E., Kristian Skrede Gleditsch and Idean Salehyan. 2012. Non-State Actor Data v3.3. Data available at: http://privatewww.essex.ac.uk/~ksg/eacd.html Dafoe, Allan, John R. Oneal and Bruce Russett. 2013. ?The Democratic Peace: Weighing the Evidence and Cautious Inference.? International Studies Quarterly 57(1): 201-214. Dashwood, Hevina S. 2000. Zimbabwe: The Political Economy of Transformation. Toronto, Canada: University of Toronto Press. Davies, Graeme A. 2002. ?Domestic Strife and the Initiation of International Conflicts.? Journal of Conflict Resolution 46(5): 672-692. Davis, David R. and Will H. Moore. 1997. Ethnicity Matters: Transnational Ethnic Alliances and Foreign Policy Behavior. International Studies Quarterly 41:171-184. Dawisha, Adeed I. 1980. Syria and the Lebanese Crisis. New York: St. Martin?s Press. Deeb, Marius. 1980. The Lebanese Civil War. New York: Praeger. DeRouen, Karl Jr. 2000. ?Presidents and the Diversionary Use of Force: A Research Note.? International Studies Quarterly 44(2): 317-328. Diehl, Paul F. 1985. ?Contiguity and Military Escalation in Major Power Rivalries, 1816-1980.? The Journal of Politics 47(4): 1203-1211. 265 Dietz, Terry. 1986. Republicans and Vietnam, 1961-1968. New York: Greenwood Press. Doyle, Michael W. 1983a. ?Kant, Liberal Legacies, and Foreign Affairs.? Philosophy & Public Affairs 12(3): 205-235. Doyle, Michael W. 1983b. ?Kant, Liberal Legacies, and Foreign Affairs, Part 2.? Philosophy & Public Affairs 12(4): 323-353. Doyle, Michael W. 1986. ?Liberalism and World Politics.? American Political Science Review 80(4): 1151-1169. Doyle, Michael W. and Nicholas Sambanis. 2000. ?International Peacebuilding: A Theoretical and Quantitative Analysis.? American Political Science Review 94(4): 779-801. Duner, Bertil. 1983. ?The Many-Pronged Spear: External Military Intervention in Civil Wars in the 1970s.? Journal of Peace Research 20(1): 59-72. Durch, William J. 1978. ?The Cuban Military in Africa and the Middle East: From Algeria to Angola.? Studies in Comparative Communism 11(1-2): 72-86. Elbadawi, Ibrahim A. and Sambanis, Nicholas. 2000. External intervention and the duration of civil wars. World Bank Policy Research Working Paper 2433. Enterline, Andrew J. and Kristian S. Gleditsch. 2000. ?Threats, Opportunity, and Force: Repression and Diversion of Domestic Pressure, 1948-1982.? International Interactions 26(1): 21-53. Fearon, James D. 1994. ?Domestic Political Audience and the Escalation of International Disputes.? American Political Science Review 88(3): 577-592. Feste, Karen A. 1992. Expanding the Frontiers: Superpower Intervention in the Cold War. New York: Praeger. Findley, Michael G. and Tze Kwang Teo. 2006. ?Rethinking Third-Party Interventions into Civil Wars: An Actor-Centric Approach.? Journal of Politics 68(4): 828-837. Gartzke, Erik and Kristian Gleditsch. 2006. ?Identity and Conflict: Ties that Bind and Differences that Divide? European Journal of International Relations 12(1): 53-87. Replication data available at: http://privatewww.essex.ac.uk/~ksg/ Gelpi, Christopher. 1997. ?Democratic Diversions: Government Structure and the Externalization of Domestic Conflict.? Journal of Conflict Resolution 41(2): 255-282. 266 Gent, Stephen E. 2007. ?Strange Bedfellows: The Strategic Dynamics of Major Power Military Interventions.? Journal of Politics 69(4): 1089-1102. Gent, Stephen. 2008. ?Going in When it Counts: Military Intervention and the Outcome of Civil Conflicts.? International Studies Quarterly 52(4): 713-735. Ghosn, Faten, Glenn Palmer, and Stuart Bremer. 2004. ?The MID3 Data Set, 1993- 2001: Procedures, Coding Rules, and Description.? Conflict Management and Peace Science 21:133-154. Gibler, Douglas M. 2009. International Military Alliances, 1648-2008. Washington, D.C.: CQ Press. Gilmore, Elisabeth, Nils Petter Gleditsch, Paivi Lujala and Jan Ketil Rod. 2005. ?Conflict Diamonds: A New Dataset.? Conflict Management and Peace Science 22(3): 257-292. Girling, John. 1981. Thailand: Society and Politics. Ithaca, N.Y.: Cornell University Press. Gleditsch, Kristian Skrede. 2002. All Politics is Local: The Diffusion of Conflict, Integration, and Democratization. Ann Arbor, MI: University of Michigan Press. Gleditsch, Kristian Skrede. 2007. ?Transnational Dimensions of Civil War.? Journal of Peace Research 44(3): 293-309. Gleditsch, Nils Petter and J. David Singer. 1975. ?Distance and International War, 1816-1965.? In Proceedings of the International Peace Research Association, Fifth General Conference, ed. M. R. Khan. Oslo: International Peace Research Association. Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg and Havard Strand. 2002. ?Armed Conflict 1946-2001: A New Dataset.? Journal of Peace Research 39(5): 615-637. Goertz, Gary and Paul F. Diehl. 1995. ?Taking Enduring Out of Enduring Rivalry: The Rivalry Approach to War and Peace.? International Interactions 21: 291-308. Greene, William H. 2003. Econometric Analysis, 5th ed. Upper Saddle River, N.J.: Prentice Hall. Gurr, Ted Robert and Raymond Duvall. 1973. ?Civil Conflict in the 1960s: A Reciprocal Theoretical System with Parameter Estimates.? Comparative Political Studies 6(2): 135-170. 267 Gurr, Ted Robert. 1992. ?The Internationalization of Protracted Communal Conflicts Since 1945: Which Groups, Where and How.? In The Internationalization of Communal Strife, ed. Manus I. Midlarsky. New York: Routledge. Gurtov, Melvin. 1974. The United States Against the Third World: Anti-Nationalism and Intervention. New York: Praeger. Heckman, James J. 1979. ?Sample Selection Bias as a Specification Error.? Econometrica 47(1): 153-162. Heger, Lindsay and Salehyan Idean. 2007. ?Ruthless Rulers: Coalition Size and the Severity of Civil Conflict.? International Studies Quarterly 51(2): 385-403. Helmreich, Jonathan E. 1976. Belgium and Europe: A Study in Small Power Diplomacy. The Hague: Mouton. Hensel, Paul. 2009. Issue Correlates of War (ICOW) Project: Colonial History Dataset v0.4. Data available at: http://www.paulhensel.org/icowcol.html Heraclides, Alexis. 1990. ?Secessionist minorities and external involvement.? International Organization 44(3): 341-378. Hermann, Margaret G. and Charles W. Kegley, Jr. 1996. ?Ballots, a Barrier against the Use of Bullets and Bombs: Democratization and Military Intervention.? Journal of Conflict Resolution 40(3): 436-460. Hermann, Margaret G. and Charles W. Kegley, Jr. 2001. ?Democracies and Intervention: Is There a Danger Zone in the Democratic Peace?? Journal of Peace Research 38(2): 237-245. Hoffman, Stanley. 1984. ?The Problem of Intervention.? In Intervention in World Politics, ed. Hedley Bull. New York: Oxford University Press. Howell, John F. 1978. ?Horn of Africa: Lessons from the Sudan Conflict.? International Affairs 54(3): 421-436. Huth, Paul K. 1998. ?Major Power Intervention in International Crises, 1918-1988.? Journal of Conflict Resolution 42(6): 744-770. James, Harold and Denis Sheil-Small. 1971. The Undeclared War: The Story of the Indonesian Confrontation, 1962-1966. Totowa, N.J.: Rowman & Littlefield. Jentleson, Bruce W. and Ariel E. Levite. 1992. ?The Analysis of Protracted Foreign Military Intervention.? In Foreign Military Intervention: The Dynamics of Protracted Conflict, eds. Ariel E. Levite, Bruce W. Jentleson and Larry Berman. New York: Columbia University Press. 268 Kaplan, Morton A. 1964. ?Intervention in Internal War: Some Systemic Sources.? In International Aspects of Civil Strife, ed. James N. Rosenau. Princeton, NJ: Princeton University Press. Kapstein, Ethan B. 1995. ?Is Realism Dead? The domestic sources of international politics.? International Organization 49(4): 751-774. Karnow, Stanley. 1983. Vietnam: A History. New York: Viking Press. Kathman, Jacob D. 2010. ?Civil War Contagion and Neighboring Interventions.? International Studies Quarterly 54(4): 989-1012. Kathman, Jacob D. 2011. ?Civil War Diffusion and Regional Motivations for Intervention.? Journal of Conflict Resolution 55(6): 847-876. Kegley, Charles W., Jr. and Margaret G. Hermann. 1995. ?Military Intervention and the Democratic Peace.? International Interactions 21(1): 1-21. Keesings World News Archive. www.keesings.com King, Gary and Langche Zeng. 2001. ?Logistic Regression in Rare Events Data.? Political Analysis 9: 137-163. Kisangani, Emizet F. and Jeffrey Pickering. 2007. ?Diverting with Benevolent Military Force: Reducing Risks and Rising Above Strategic Behavior.? International Studies Quarterly 51(2): 277-299. Kisangani, Emizet F. and Jeffrey Pickering. 2008. ?International Military Intervention, 1989-2005.? Inter-University Consortium for Political and Social Research, Data Collection 21282. University of Michigan, Ann Arbor. Khosla, Deepa. 1999. ?Third World States as Interveners in Ethnic Conflicts.? Third World Quarterly 20(6): 1143-1156. Klein, James P., Gary Goertz and Paul F. Diehl. 2006. ?The New Rivalry Dataset: Procedures and Patterns.? Journal of Peace Research 43(3): 331-348. Rivalry dataset available at: ftp://128.196.23.212/rivalry /riv500web.zip Koga, Jun. 2011. ?Where Do Third Parties Intervene? Third Parties? Domestic Institutions and Military Interventions in Civil Conflicts.? International Studies Quarterly 55(4): 1143-1166. Lacina, Bethany and Nils Petter Gleditsch. 2005. ?Monitoring Trends in Global Combat: A New Dataset of Battle Deaths.? European Journal of Population 21(2-3): 145-166. 269 Lemke, Douglas and William Reed. 2001. ?The Relevance of Politically Relevant Dyads.? Journal of Conflict Resolution 45(1): 126-144. Lemke, Douglas and Patrick M. Regan. 2004. ?Interventions as Influence.? In The Scourge of War: New Extensions to an Old Problem, ed. Paul F. Diehl. University of Michigan Press. Levy, Jack. 1989. ?The Diversionary Theory of War.? In Handbook of War Studies, ed. Manus I. Midlarsky. Boston: Unwin Hyman. Lowenthal, Abraham F. 1972. The Dominican Intervention. Cambridge, Mass.: Harvard University Press. Lujala, Paivi, Nils Petter Gleditsch and Elisabeth Gilmore. 2005. ?A Diamond Curse? Civil War and a Lootable Resource.? Journal of Conflict Resolution 49(4): 538-562. Lujala, Paivi, Jan Ketil Rod, and Nadja Thieme. 2007. ?Fighting Over Oil: Introducing a New Dataset.? Conflict Management and Peace Science 24(3): 239-256. Maoz, Zeev and Bruce Russett. 1993. ?Normative and Structural Causes of Democratic Peace, 1946-1986.? American Political Science Review 87(3): 624-638. Marshall, Monty G., Keith Jaggers and Ted Robert Gurr. 2011. Polity IV Project: Political Regime Characteristics and Transitions, 1800-2010. Data available at: http://www.systemicpeace.org/inscr/inscr.htm Mason, David T., Joseph P. Weingarten and Patrick J. Fett. 1999. ?Win, Lose, or Draw: Predicting the Outcome of Civil War.? Political Research Quarterly 52(2): 239-268. Miller, Ross A. 1999. ?Regime Type, Strategic Interaction, and the Diversionary Use of Force.? Journal of Conflict Resolution 43(3): 388-402. Minorities at Risk Project. 2009. ?Minorities at Risk Dataset.? College Park, MD: Center for International Development and Conflict Management. Mitchell, C. R. 1970. ?Civil Strife and the Involvement of External Parties.? International Studies Quarterly 14(2): 166-194. Modelski, George. 1964. ?The International Relations of Internal War.? In International Aspects of Civil Strife, ed. James N. Rosenau. Princeton, NJ: Princeton University Press. 270 Morgan, T. Clifton and Kenneth N. Bickers. 1992. ?Domestic Discontent and the External Use of Force.? Journal of Conflict Resolution 36(1): 25-52. Morgenthau, Hans J. 1967. ?To Intervene or Not to Intervene.? Foreign Affairs 45(3): 425-436. Nweke, Aforka G. 1976. External Intervention in African Conflicts: France and French-speaking West Africa in the Nigerian Civil War, 1967-1970. Brookline, Mass: Boston University Press. Oneal, John R. and Bruce M. Russett. 1997. ?The Classical Liberals Were Right: Democracy, Interdependence and Conflict, 1950-1985.? International Studies Quarterly 41(2): 267-294. Owen, John M. 1994. ?How Liberalism Produces Democratic Peace.? International Security 19(2): 87-125. Paul, David W. 1971. ?Soviet Foreign Policy and the Invasion of Czechoslovakia: A Theory and a Case study.? International Studies Quarterly 15(2): 159-202. Pearson, Frederick S. 1974a. ?Foreign Military Interventions and Domestic Disputes.? International Studies Quarterly 18(3): 259-289. Pearson, Frederick S. 1974b. ?Geographic Proximity and Foreign Military Intervention.? Journal of Conflict Resolution 18(3): 432-460. Pearson, Frederick S. and Robert A. Baumann. 1993. ?International Military Intervention, 1946-1988.? Inter-University Consortium for Political and Social Research, Data Collection 6035. University of Michigan, Ann Arbor. Pearson, Frederic S., Robert A. Baumann, and Jeffrey J. Pickering. 1994. ?Military Intervention and Realpolitik.? In Reconstructing Realpolitik, eds. Frank W. Wayman and Paul F. Diehl. Ann Arbor: The University of Michigan Press. Peceny, Mark. 1999. Democracy at the Point of Bayonets. University Park, PA: The Pennsylvania State University Press. Pickering, Jeffrey and Emizet F. Kisangani. 2009. ?The International Military Intervention Dataset: An Updated Resource for Conflict Scholars.? Journal of Peace Research 46(4): 589-599. PRIO Battle Deaths Dataset 1946-2008, version 3.0. Data available at: http://www.prio.no/Data/Armed-Conflict/Battle-Deaths/ Putnam, Robert D. 1988. ?Diplomacy and domestic politics: the logic of two-level games.? International Organization 42(3): 427-460. 271 Rabinovich, Itamar. 1979. ?The Limits of Military Power: Syria?s Role.? In Lebanon in Crisis: Participants and Issues, eds. P. Edward Haley and Lewis W. Snider. New York: Syracuse University Press. Rasler, Karen. 1983. ?Internationalized Civil War: A Dynamic Analysis of the Syrian Intervention in Lebanon.? Journal of Conflict Resolution 27(3): 421-456. Reed, William. 2000. ?A Unified Statistical Model of Conflict Onset and Escalation.? American Journal of Political Science 44(1): 84-93. Reed, William and David H. Clark. 2000. ?War Initiators and War Winners: The Consequences of Linking Theories of Democratic War Success.? Journal of Conflict Resolution 44(3): 378-395. Reiter, Dan and Allan C. Stam. 1998. ?Democracy, War Initiation, and Victory.? American Political Science Review 92(2): 377-389. Regan, Patrick M. 1996. ?Conditions of Successful Third-Party Intervention in Intrastate Conflicts.? Journal of Conflict Resolution 40(2): 336-359. Regan, Patrick M. 1998. ?Choosing to Intervene: Outside Interventions in Internal Conflicts.? The Journal of Politics 60(3): 754-779. Regan, Patrick M. 2000. Civil Wars and Foreign Powers. Ann Arbor: University of Michigan Press. Data at: http://bingweb.binghamton.edu/~pregan/replicationdata.html Regan, Patrick M. 2002. ?Third-Party Interventions and the Duration of Intrastate Conflicts.? Journal of Conflict Resolution 46(1): 55-73. Rosecrance, Richard N. 1963. Action and Reaction in World Politics. Boston: Little, Brown. Rosenau, James N., ed. 1964. International Aspects of Civil Strife. Princeton, NJ: Princeton University Press. Rosenau, James N. 1969. "Intervention as a Scientific Concept." Journal of Conflict Resolution 13(2):149-171. Ross, Michael L. 2004a. ?How Do Natural Resources Influence Civil War? Evidence from Thirteen Cases.? International Organization 58(1): 35-67. Ross, Michael L. 2004b. ?What Do We Know About Natural Resources and Civil War?? Journal of Peace Research 41(3): 337-356. 272 Rothschild, Joseph. 1981. Ethnopolitics: A Conceptual Framework. New York: Columbia University Press. Roy, Bikash A. 1997. ?Intervention Across Bisecting Borders.? Journal of Peace Research 34(3): 303-314. Russett, Bruce M. 1993. Grasping the Democratic Peace: Principles for a Post-Cold War World. Princeton, N.J.: Princeton University Press. Saideman, Stephen M. 1997. ?Explaining the International Relations of Secessionist Conflicts: Vulnerability versus Ethnic Ties.? International Organization 51(4): 721-753. Saideman, Stephen M. 2002. ?Discrimination in International Relations: Analyzing External Support for Ethnic Groups.? Journal of Peace Research 39(1): 27-50. Salehyan, Idean. 2007. ?Transnational Rebels: Neighboring States as Sanctuary for Rebel Groups.? World Politics 59(2): 217-242. Sarkees, Meredith Reid and Frank Wayman. (2010). Resort to War: 1816-2007. Washington, DC: CQ Press. Sartori, Anne E. 2003. ?An Estimator for Some Binary-Outcome Selection Models Without Exclusion Restrictions.? Political Analysis 11: 111-138. Schmid, A. P. 1985. Soviet Military Interventions since 1945. New Brunswick: Transaction Books. Scott, James M. 1996. Deciding to Intervene: The Reagan Doctrine and American Foreign Policy. Durham: Duke University Press. Singer, J. David. 1987. ?Reconstructing the Correlates of War Dataset on Material Capabilities of States, 1816-1985.? International Interactions 14: 115-132. Snidal, Duncan. 1993. ?Relative Gains and the Pattern of International Cooperation.? In Neorealism and Neoliberalism: the Contemporary Debate, ed. David A. Baldwin. New York: Columbia University Press. Stam, Alan C. 1996. Win, Lose, or Draw: Domestic Politics and the Crucible of War. Ann Arbor: University of Michigan Press. Stinnett, Douglas M., Jaroslav Tir, Philip Schafer, Paul F. Diehl, and Charles Gochman. 2002. ?The Correlates of War Project Direct Contiguity Data, Version 3.? Conflict Management and Peace Science 19(2): 58-66. 273 Stremlau, John J. 1977. The International Politics of the Nigerian Civil War, 1967- 1970. Princeton, N.J.: Princeton University Press. Suhrke, Astri and Lela Garner Noble. eds. 1977. Ethnic Conflict and International Relations. New York: Praeger. Tillema, Herbert K. 1989. ?Foreign Overt Military Intervention in the Nuclear Age.? Journal of Peace Research 26(2): 179-195. Tomz, Michael, Gary King, and Langche Zeng. 1999. RELOGIT: Rare Events Logistic Regression, Version 1.1. Cambridge, MA: Harvard University. http://gking.harvard.edu Tucker, Richard. 1999. BTSCS: A Binary Time-Series-Cross-Section Data Analysis Utility. Version 4.0.4. Cambridge, MA: Harvard University. http://www.fas.harvard.edu/~rtucker/programs/btscs/btscs.html UCDP/PRIO Armed Conflict Dataset version 4-2010. http://www.pcr.uu.se/research/ucdp/datasets/ucdp_prio_armed_conflict_ dataset United Nations High Commissioner for Refugees (UNHCR), UNHCR Statistical Online Population Database. Data extracted: 03/16/2013. Valenta, Jiri. 1978. ?The Soviet-Cuban Intervention in Angola, 1975.? Studies in Comparative Communism 11(1-2): 21-37. Van Wingen, John and Herbert K. Tillema. 1980. ?British Military Intervention after World War II: Militance in a Second-Rank Power? Journal of Peace Research 17(4): 291-303. Vickery, Michael. 1984. Cambodia, 1975-1982. Boston: South End Press. Wilkenfeld, Jonathan, Gerald W. Hopple, Paul J. Rossa and Stephen J. Andriole. 1980. Foreign Policy Behavior: The Interstate Behavior Analysis Model. Beverly Hills, CA: Sage Publications. Yoon, Mi Yung. 1997. ?Explaining U.S. Intervention in Third World Internal Wars, 1945-1989.? Journal of Conflict Resolution 41(4): 580-602. Young, Oran R. 1968. ?Intervention and International Systems.? Journal of International Affairs 22(2): 177-178.