University of Southern California Dissertations and Theses

Forced migration as a cause and consequence of conflict

(USC Thesis Other)

Forced migration as a cause and consequence of conflict

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content University of Southern California Doctoral Thesis Forced Migration as a Cause and Consequence of Con ict Author: Cyrus Mohammadian Supervisor: Dr. Patrick James A thesis submitted in fullment of the requirements for the degree of PhD. from the POLITICAL SCIENCE & INTERNATIONAL RELATIONS FACULTY OF THE USC GRADUATE SCHOOL August 2016 Declaration of Authorship I, Cyrus Mohammadian, declare that this thesis titled, `Forced Migration as a Cause and Consequence of Con ict' and the work presented in it is my own. I conrm that this work submitted for assessment is my own and is expressed in my own words. Any uses made within it of the works of other authors in any form (e.g., ideas, equations, gures, text, tables, programs) are properly acknowledged at any point of their use. A list of the references employed is included. Signed: Date: i \You have to understand, no one puts their children in a boat unless the water is safer than the land." Warsan Shire Abstract The purpose of this dissertation is to study forced migration as a cause and consequence of war-making and state-making in the modern era. I rely on statistical methods as well analyses of current events to show that states and rebels devote tremendous resources to the management of their populations -protecting some while driving others out. Although my research conrms previous studies that identied a link between refugees and transborder instability, my ndings actually push against the emerging consensus that this association is largely driven by refugees who become active participants in warfare, what has been dubbed the \refugee warrior" hypothesis. In contrast, I nd that far from instigating violence, refugees tend to be its most likely targets. This has tremendous implications for the way in which governments manage refugees and how they collaborate with one another over asylum and migrant policy. My results suggest we can do more to limit violence associated with refugees by doing more to guarantee their security. For as long societies wage war, people will be eeing it. Keywords: Refugees, IDP, Forced Migration, Con ict, Civil War, Ethnic- ity, Lebanon Acknowledgements I would like to express my sincere gratitude to my advisor, Dr. Patrick James, without whose guidance, patience, and expertise this dissertation would not have been made possible. I would also like to acknowledge my gratitude to the other members of my committee, Dr. Jeerey M. Sellers and Dr. John P. Wilson, for the assistance they provided at all levels of this research project. Finally, I would like to thank my dear friends and family, whose support and understanding have been vital to my success. iv Contents Declaration of Authorship i Abstract iii Acknowledgements iv Contents v List of Figures viii List of Tables ix Abbreviations x Symbols xi 1 Introduction 1 2 Refugees and the Ethnic Balance of Power 5 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.2 Demographic Pressure and Economic Competition . . . . . . 8 2.2.1 Refugees and Resource Scarcity . . . . . . . . . . . . . 8 2.2.2 Refugees and the Ethnoreligious Balance of Power . . 11 2.3 Research Design, Data, and Methods . . . . . . . . . . . . . . 13 2.3.1 Research Design . . . . . . . . . . . . . . . . . . . . . 13 v Contents vi 2.3.2 Measurement and Data . . . . . . . . . . . . . . . . . 14 2.3.3 Method . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.4.1 Negative Binomial Regression . . . . . . . . . . . . . . 22 2.4.2 Robustness Check . . . . . . . . . . . . . . . . . . . . 25 2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 3 When Ethnic Groups Rebel: Refugees and Transborder Kin 30 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 3.2 Ethnic Armed Rebellion and One-sided Violence . . . . . . . 32 3.3 Coethnic Refugees and Armed Ethnic Rebellion . . . . . . . . 35 3.3.1 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . 36 3.3.2 Methodological Approach and Data . . . . . . . . . . 37 3.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 39 3.4 Coethnic Refugees and One-sided Violence . . . . . . . . . . . 41 3.4.1 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . 41 3.4.2 Methodological Approach and Data . . . . . . . . . . 42 3.4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 4 The Logic of Population Control 53 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 4.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . 56 4.2.1 Ethnicity and Ethnic Civil Wars . . . . . . . . . . . . 56 4.2.2 Civilian Victimization . . . . . . . . . . . . . . . . . . 59 4.2.3 Forced Migration . . . . . . . . . . . . . . . . . . . . . 60 4.3 The Logic of Population Control . . . . . . . . . . . . . . . . 62 4.3.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . 63 4.3.2 Actors and Interests . . . . . . . . . . . . . . . . . . . 65 4.3.3 Stylized Narrative . . . . . . . . . . . . . . . . . . . . 68 4.4 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 4.5 Data and Methods . . . . . . . . . . . . . . . . . . . . . . . . 73 4.5.1 Methodological Approach . . . . . . . . . . . . . . . . 73 4.5.2 Estimation Technique . . . . . . . . . . . . . . . . . . 76 4.5.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.6 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 4.6.1 Organization of Results . . . . . . . . . . . . . . . . . 86 Contents vii 4.6.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 87 4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 5 Conclusion 95 A Appendix: Codebook 101 B Appendix: Data Visualization 112 C Appendix: Log Count Results 118 Bibliography 123 List of Figures 2.1 Refugees as a Share of World Population since 1950 . . . . . 6 2.2 Refugees and Con ict April 2013 - May 2015 . . . . . . . . . 16 2.3 Con ict Events 2013-2015 . . . . . . . . . . . . . . . . . . . . 17 2.4 Predicted Probabilities for High vs Low Tensions . . . . . . . 24 4.1 Con ict States 1993-2009 . . . . . . . . . . . . . . . . . . . . 79 5.1 Google Keyword Search Hits . . . . . . . . . . . . . . . . . . 99 B.1 Refugee Flows and Violence 2013-2015 . . . . . . . . . . . . . 112 B.2 Violence Severity 2013-2015 . . . . . . . . . . . . . . . . . . . 113 B.3 Violence by District 2013-2015 . . . . . . . . . . . . . . . . . 114 B.4 Syrian Refugees as a Share of Population 2013-2015 . . . . . 115 B.5 Ethnoreligious Tapestry of Lebanon . . . . . . . . . . . . . . 116 B.6 Ethnoreligious Groups by District . . . . . . . . . . . . . . . . 117 viii List of Tables 2.1 Con ict Events . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.2 Random Eects and Spatial Logit MCMC . . . . . . . . . . . 27 3.1 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . 39 3.2 Ethnic Refugees and Armed Rebellion . . . . . . . . . . . . . 40 3.3 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . 44 3.4 Ethnic Refugees and One-sided Violence . . . . . . . . . . . . 46 4.1 Raw Values of Selection Equation Variables . . . . . . . . . . 81 4.2 Binary Values of Selection Equation Variables . . . . . . . . . 81 4.3 Log Count Values of Selection Equation Variables . . . . . . . 81 4.4 Untransformed Variables in Selection Equation . . . . . . . . 83 4.5 Variables for Outcome Equation . . . . . . . . . . . . . . . . 85 4.6 Organization of Model Results . . . . . . . . . . . . . . . . . 87 4.7 Marginal Eects Models . . . . . . . . . . . . . . . . . . . . . 89 4.8 Conditional Eects Models . . . . . . . . . . . . . . . . . . . 91 C.1 Organization of Model Results . . . . . . . . . . . . . . . . . 118 C.2 Marginal Eects Models with Log Counts . . . . . . . . . . . 119 C.3 Conditional Eects Models with Log Counts . . . . . . . . . . 120 C.4 Three-Stage (1) Heckit . . . . . . . . . . . . . . . . . . . . . . 121 C.5 Three-Stage (2) Heckit . . . . . . . . . . . . . . . . . . . . . . 122 ix Abbreviations EBP Ethnic Balance Power CDC Categorically Disaggregated Con ict ACD Armed Con ict Data EPR Ethnic Power Relations EAC Ethnic Armed Con ict ER Ethnic Refugees RS Resource Scarcity UNHCR United Nations High Commissioner for Refugees UNRWA United Nations Relief and Works Agency HAC Heteroscedasticity and Autocorrelation Consistent UCDP Uppsala Con ict Data Program WDI World Development Indicators GDP Gross Domestic Product IDP Internally Displaced Person DV Dependent Variable MLE Maximum Likelihood Estimation MCMC Markov Chain Monte Carlo PKK Partiya Karker^ en Kurdistan^ e ISIS Islamic State of Iraq and Syria x Symbols T l probability of targeting loyal civilians T d probability of targeting disloyal civilians C costs of targeting civilians B benets of targeting civilians V e level of violence against civilians during ethnic civil war V n level of violence against civilians during nonethnic civil war P a perceived probability of victimization in civilian's potential destination of asylum P o perceived probability of victimization in civilian's country of origin R the number of refugees eeing civil war I the number of IDPs eeing civil war F e share of forced migrants that are refugees eeing ethnic con ict F n share of forced migrants that are refugees eeing nonethnic con ict D it forced migration Z it vector of explanatory variables i;t country, time u error f it ratio of refugee ows to total migrant ows in country i in time t xi Symbols xii vector of unknown parameters cumulative distribution function of the standard normal distribution vector of known coecients for explanatory variables corr. of unobserved migrants & unobserved ratio of ref ows to migrant ows inverse Mill's ratio standard deviation In dedication to my parents for nursing me with love and unconditional support in my pursuit of success.... xiii Chapter 1 Introduction Refugees in History People have been eeing war for as long as societies have waged it. It has been said that the Egyptian King Assyrophernes erected a massive mon- ument in honor of his slain son and decreed that those who found shelter beneath it would be protected (Phillipson 1911, 348). A similar right to asylum is related in the story of Assyrian King Ninus who erects a statue of his father that bestows protections to those seeking refuge from violence. Yet, it was the Israelites who rst institutionalized the concept of a physical space of sanctuary, while the Saxon king of Kent, Ethelbert, was the rst to formally codify it in law in the 6 th century AD (Rabben 2011, 59). However, it was not until the late 15 th century, during early modern Europe's emerging system of nation-states, that political authorities began to devote tremendous resources to population control. In contrast to the monarchies and religious authorities of previous eras, these leaders relied on their exer- cise of national sovereignty to legitimize their rule. The exercise of national sovereignty not only entailed establishing physical borders separating one 1 Chapter 1. Introduction 2 nation-state from another, it also involved the delineation of sociocultural boundaries that ascribe and allocate national identity on the basis of various cultural markers, such as shared language and religious aliation (Mar eet 2007). Central to this process was dening who the nation was and, as im- portantly, who it was not. Those whose national identities did not align with national fortunes of the state were either labeled as threats to the nation or were expelled -the \others". The exodus of Muslims and Jews from the Iberian peninsula during the Inquisition, the ight of Puritans to the New World during the 16 th and 17 th centuries, and the migration of Calvinists of France and Germany to Protestant lands in the 17 th and 18 th centuries are prominent examples of these \others". In fact, it was the Huguenots, eeing to England from a hos- tile Catholic France, who were rst labeled r efugi es. It has been argued that the English monarchs, who guaranteed their security, viewed them as \both an economic resource and an ideological asset" (Mar eet 2007, 140). As a result, these refugees became entangled in the process of statecraft. The French monarchs would remove Huguenots from the French body politic only for the English Kings to snub their counterparts by accepting them into theirs. This was hardly a feature unique to the early modern world. The Sassanians sheltered the Nestorian Christians, which outraged their Byzantine foes, and even the Greek city-states were said to have provided refuge for those defecting from their rivals. Tilly (1975) argued that states make war and wars make states. If he was right, then states also make refugees and refugees also make states. That is, states contribute to condi- tions conducive to the rise of refugees and the movement of refugees from one state to another contributes to the consolidation of the borders between them because giving sanctuary to one group also entails restricting it from others. The purpose of this dissertation is to study forced migration as a cause and consequence of war-making and state-making in the modern era. I rely on Chapter 1. Introduction 3 statistical methods as well analyses of current events to show that states and rebels devote tremendous resources to the management of their populations -protecting some while driving others out. In the following section I preview the methods and key ndings of each chapter. Organization Chapter 2 builds o recent quantitative work on civil war that has iden- tied a link between refugee movements and the spread of con ict across borders. One commonly proposed mechanisms that accounts for this nd- ing identies refugee ows as a form of population pressure, which increases violence between host populations and incoming refugees. Another com- monly proposed mechanism suggests refugees with ethnic ties to groups within the host population increase violence when their presence disrupts the ethnic balance of power between rival groups. Spatial regression results using a novel geo-coded dataset of substate violence in Lebanon between March 2013 - April 2015 reveal support for the latter mechanism. Chapter 3 explores the ethnic balance of power theory in the context of two dierent forms of con ict -ethnic armed rebellion and one-sided violence -using global panel data on intrastate violence at the group and country lev- els. Although, the analysis fails to identify a signicant association between the presence of coethnic refugees and ethnic armed con ict, coethnic refugees do signicantly increase the probability of rebel led one-sided violence. I ac- count for this by introducing a complimentary theory to EBP, \the logic of population control", which contends that rebels have an incentive to target civilian supporters of rival groups as a means of undercutting the support these rivals enjoy. However, targeting civilians is costly if rebels cannot properly discriminate between their own supporters and civilians loyal to the government or opposing groups. In ethnically polarized states, rebels Chapter 1. Introduction 4 can take advantage of the salience of ethnicity by mobilizing along ethnic lines and using ethnic markers to aid in the identication of potentially dis- loyal civilians. This logic, also applies to coethnic refugees. Rebels view refugees that share ethnic ties with the civilian populations represented by rival armed groups (government or otherwise) as a threat and respond to this threat with violence directed towards these unarmed civilians. There- fore, I conclude that while coethnic refugees do not increase the likelihood of con ict onset, they do increase the chances that civilians will be targeted, refugee or otherwise. Chapter 4 examines why some civil war torn countries produce more refugees relative to their internally displaced population and others displace more of their population internally than across borders. Surprisingly, the relationship between internally displaced persons and con ict has been woe- fully underexplored. The aim of this chapter is to ll this gap in the litera- ture. Using a panel dataset of civil con icts by country-year from 1993-2010 and a two-step Heckman selection model, I show that civil wars fought along ethnic lines produce greater refugee ows relative to IDP ows than non- ethnic civil wars. I account for this nding by relying on insights drawn from the previous chapters. Specically, I argue that in con icts where combat- ants are recruited along ethnic lines, ethnic markers allow for less costly and more discriminate targeting of rival civilian populations, which in turn in- creases the share of forced migrants who seek refugee across borders relative to those displaced internally. Chapter 5 concludes with a 1) review of this dissertation's key contribu- tions, 2) a discussion of how these ndings are situated within the larger literature, and 3) discusses the implications of these ndings for issues re- lated to immigration and asylum policy. Chapter 2 Refugees and the Ethnic Balance of Power Abstract Recent quantitative work on civil wars has identied a link between refugee movements and the spread of con ict across borders. One commonly pro- posed mechanisms that accounts for this nding identies refugee ows as a form of population pressure, which increases violence between host pop- ulations and incoming refugees. Another commonly proposed mechanism suggests refugees with ethnic ties to groups within the host population in- crease violence when their presence disrupts the ethnic balance of power between rival groups. Spatial regression results using a novel geo-coded dataset of substate violence in Lebanon between March 2013 - April 2015 reveal support for the latter mechanism. 5 Chapter 2. Refugees and the Ethnic Balance of Power 6 2.1 Introduction One of the most pressing humanitarian challenges facing the international community today is the plight of millions of refugees, asylum-seekers, and internally displaced peoples. In fact, the refugee crises is worse today than in any point in history. Figure 2.1 shows the astonishing rise in the number of refugees and asylum-seekers as a proportion of world population since 1950. As of 2014, there are 54,937,556 refugees or individuals in refugee- like situations across 130 countries. Since 1995, the absolute number of worldwide refugees has increased by an alarming 170%. The majority of these people are eeing homes ravaged by violent con ict. Thus, much of the scholarly literature on the topic views refugees as a consequence of con ict and seeks to identify solutions to the crises (see for example Vernant 1953; Schechtman 1964; Ferris 1993). Yet, recent work in the eld suggests refugees are a cause of war as much as they are a product of it. Figure 2.1: Refugees as a Share of World Population since 1950 Chapter 2. Refugees and the Ethnic Balance of Power 7 For example, Salehyan and Gleditsch (2006), arguing that refugees con- tribute to con ict contagion, show that the presence of refugees from neigh- boring states increases the risk of civil war. Bohnet (2012) examines the relationship between refugee camp concentration and distribution and the outbreak of violence and nds that shorter distances between refugees and refugee settlements increases con ict risk. Fisk (2014) shows that refugee camps closer to the border tend to increase the risk of con ict. However, most refugee ows do not result in con ict (Lischer 2006) and the extant mechanisms linking refugees to violence may tell us something about the variation in the geographic distribution of con ict within a state but does little to contribute to our knowledge regarding the variation in the geographic distribution of con ict between states. For example, the ongoing civil war in Syria has produced a hemorrhaging of refugees across most of its international borders. Our current models have correctly predicted that violence from the Syrian civil war is contagious yet they are unable to explain why that violence has spread in certain directions but not others. The mechanisms linking refugees to the spread of con ict remains under theorized. The extant proposed mechanisms focus on the geographic distribution of refugee settlements at the substate level, such as distance between settle- ments, distance between settlements and major land markers such as capitals or international borders, and proximity to landscape favorable to rebellion. While these geographic factors are informative in identifying the location of con ict at the micro level, they cannot explain the variation in con ict likelihood across states at the macro level. The problem is that these sorts of geographic variables tend to exhibit far more variation within a country than between countries. After all, the logical factors that dictate the loca- tion of refugee clusters within states, such as proximity to border crossings to ease the in ux of refugees, are conditions that are likely faced by most states hosting refugees eeing neighboring con ict. In contrast to geographic Chapter 2. Refugees and the Ethnic Balance of Power 8 factors, however, demographic and economic factors can potentially explain variation both within and between states. As a result, mechanisms relying on variation in economic and demographic indicators should be identiable at both micro and macro levels. The focus of this chapter is to introduce two such mechanisms and test them at the micro level of analysis. I argue that the eect of refugees on con ict is conditional on two factors. First, refugees increase con ict levels by altering the ethnoreligious balance of power between rival groups. Second, refugees exacerbate competition over exceedingly scarce resources resulting in heightened levels of con ict. The following section discusses each mechanism in tandem. 2.2 Demographic Pressure and Economic Compe- tition 2.2.1 Refugees and Resource Scarcity The notion that population changes can lead to instability is not new; in his seminal treatise on population growth, Malthus (1798) ominously predicted that human population growth would outpace agricultural production lead- ing to a catastrophic breakdown in human socioeconomic structures. While his predictions failed to materialize, his ideas linking resource scarcity to con ict remained salient among a wave of postwar era scholars. These neo- Malthusianists argued that new technologies in mass agricultural production have led to a rapid increase in the global population, whose sustenance can- not be sustained with the world's increasingly depleted resources (see for example Osborn 1948; Vogt 1948; Ehrlich 1968). Chapter 2. Refugees and the Ethnic Balance of Power 9 Much of this work has since been decried as alarmist. What Ehrlich and others writing in his era failed to incorporate in their models is the phe- nomenon known as `demographic transition', which entails a shift from high birth rates and death rates to low birth rates and death rates (without replacement) in rst world countries (Thompson 1929). As a result, the population growth of most developed countries has been outpaced by gains in productivity. While these theories failed to gain traction overtime, the idea that population shifts can lead to con ict continuous to spark scholarly attention. More recently, scholars have begun to examine how short-term and rapid population changes can induce con ict over exceedingly scarce resources. In some instances, it has been argued that population growth and resource scarcity can actually work as a catalyst for technological innovation (Boserup 1990). Urdal (2005) nds no evidence linking population growth to con ict. On the other hand, Homer-Dixon and Blitt (1998) argue that developing countries are dependent on fresh water, cropland, forests, and sheries - depletion of any of these resources they argue leads to con ict. Maxwell and Reuveny (2000) develop an agent-based model of population oscilla- tions and warfare that largely follows the insights of Homer-Dixon and Blitt (1998). Raleigh and Urdal (2007) show that population growth and density are associated with con ict but not land degradation or water scarcity. In contrast, Theisen (2008) does nd evidence linking land degradation and water scarcity with domestic con ict. The general picture that emerges from these studies is that peace and sta- bility are sensitive to population changes. Although a consensus on the mechanism underlying this relationship remains at large, the evidence does heavily weigh in the favor of resource scarcity as a catalyst. Under condi- tions of resource abundance, rapid population growth can lead to the gradual Chapter 2. Refugees and the Ethnic Balance of Power 10 depletion of these resources, which in turn can lead to con ict. Under con- ditions of resource scarcity, however, rapid population growth leads directly to con ict over the already scarce resources. Borrowing from the literature on demography and con ict, one of the mech- anisms this chapter tests is the role refugees play in bringing about de- mographic changes that lead to competition over scarce resources and ul- timately violent con ict. For example, the war in Darfur between Arab northerners and non-Arab westerners resulted in a mass exodus of over 200,000 refugees into neighboring Chad beginning in 2003. The newly ar- rived refugees, most of whom were non-Arab Sudanese, were in immediate competition with locals over access to fresh water, government subsidies, rewood, etc. 1 By 2005 tensions had erupted between the refugees and the Arab population of Chad to the point that the non-Arab refugees began agitation movements alongside other non-Arab Chadians against the Arab Chadian majority; in less than a year Chad would collapse into an all out civil war as deadly as Sudan's (Reyna 2010). Typically refugees eeing to neighboring states are drawn from poorer seg- ments in society, with little economic resources at their disposal, exposed to great dangers both environmental and man-made, and in great need of subsidized care, thus, placing them in direct competition with local popula- tions (Loescher and Milner 2005). While international aid organizations are a great resource depended upon by many refugees, the greatest nancial and political burden ultimately falls upon the host state. As a result, refugees place themselves in direct competition with other communities in the host state {coethnic or otherwise. Competition may arise over the use of arable land for crops, access to fresh water, employment opportunities, or even 1 http://tinyurl.com/j7da6w6 Chapter 2. Refugees and the Ethnic Balance of Power 11 access to government services From this I derive my rst three hypotheses: H 1 : As the number of refugees a region hosts increases, so does its level of con ict. H 2 : Resource poor regions will experience more con ict than regions rich in resources H 3 : Refugees increase con icts more in resource poor regions than regions rich in resources. 2.2.2 Refugees and the Ethnoreligious Balance of Power In contrast to the resource scarcity theory (RS), the `ethnoreligious balance of power' (EBP) theory is couched in realist assumptions of con ict and co- operation. While realist theory of war originated in the eld of international relations, many realist insights have been adopted in the study of domestic con ict (see for example Posen 1993). Peace between groups prevails when a certain balance of power between them emerges. When that balance of power is altered in one groups favor, however, the newly empowered group may attempt to change the way in which state rent is distributed, in order to align that distribution more closely with its own interests, which can result in con ict. The sensitivity of the peace equilibrium to outside shocks is most pronounced under conditions of ethnic competition (ibid). According to Posen's eth- nic security dilemma, when states collapse and ethnic salience is a dening feature of political competition, then con ict is likely to erupt due to the increasing inability to assure credible commitments and guarantees of peace between rival ethnic groups. In many circumstances, however, state collapse itself is endogenous to ethnic con ict. Nonetheless, Posen is right to suggest that exogenous shocks to an ethnically volatile system do in fact contribute Chapter 2. Refugees and the Ethnic Balance of Power 12 to higher levels of con ict. Indeed state collapse does seem to induce ethnic con ict where ethnic tensions were previously kept in check. Similarly, it seems plausible to suggest that other exogenous shocks that introduce un- certainty or alter the balance of power between groups may also contribute to the onset of con ict, such as sponsorship and/or intervention by third parties or shifts in population numbers. Refugee ows are an example of the latter. For instance, the Syrian civil war that erupted in 2011 lead to a hemor- rhaging of Sunni refugees into Lebanon, which is threatening undermined the fragile peace between Sunnis and Shias in the country that had emerged in recent years. Sunnis, whose numbers and political power along with Lebanese Christians have continued to dwindle relative to their Shia coun- terparts, have been emboldened by the presence of Sunni refugees fresh from the battleelds in Syria, leading to frequent clashes between Assad support- ing Lebanese Shia (e.g. Hezbollah and Amal) and anti-Assad Sunnis. With a pool of ghters to draw from, the delicate ethnic balance between Sunnis and Shias in Lebanon has been jeopardized as Sunnis attempt to seize this rare opportunity to push back gains made by Hezbollah and their allies in Lebanon in recent years. As illustrated in the case above, refugee ows can alter the ethnoreligious balance of power by providing ethnoreligious groups with a greater pop- ulation base from which to extract resources and human capital (e.g. a greater potential pool of young ghters) and these ows can increase the population concentration of those groups, which allows for more ecient mobilization. These ows modify the ethnic composition of the state when the refugees are composed of groups that share an ethnic identity with one or more constituent ethnic groups in the host state. Changes in the ethnic tapestry consequently lead to violence, particularly when relations between Chapter 2. Refugees and the Ethnic Balance of Power 13 the competing ethnic groups are polarized. From this I derive my nal two hypotheses: H 4 : Regions with pre-existing ethnoreligious tensions are more likely to ex- perience con ict. H 5 : The eect of refugees on con ict levels is magnied by the presence of pre-existing ethnoreligious tensions in a region. 2.3 Research Design, Data, and Methods 2.3.1 Research Design All extant micro level statistical studies of refugees and con ict have limited their empirical analysis to sub-Saharan Africa, which leaves us wondering whether the refugee con ict nexus is unique to sub-Saharan Africa or is generalizable to other regions of the world. This study extends the empir- ical scope of analysis to a previously unexamined region of the world, the Middle East. In particular, a substate analysis of the Syrian refugee cri- sis in Lebanon will be undertaken with the aim of testing the two theories discussed above. Case selection is motivated by both theoretical and practical concerns. Lebanon is facing the most dire refugee crisis of all of Syria's neighbors -more than one in four residents in Lebanon is currently a registered Syrian refugee. Moreover, Lebanon and Syria share a similar composition of ethnoreligious groups, which is a necessary factor when testing for the conditional eects of sectarian tensions on refugee induced con ict spillover. Finally, Lebanon is the only state in the region to prohibit the establishment of formal UNHCR camps for its Syrian refugee population. With the notable exception of Fisk Chapter 2. Refugees and the Ethnic Balance of Power 14 (2014), all previous micro level studies of refugees and con ict have lim- ited their analysis to states with established camps. The eects of refugees dispersed among the host population remains largely unexamined. Using Lebanon's twenty-six second-order administrative districts as units of analysis, this study examines the spatial and temporal variation in the incidence of con ict for each month from March 2013 to April 2015. This results in a sample of 650 observations across 25 months (N = 26 districts x 25 months). 2.3.2 Measurement and Data Con ict I begin with a discussion of my dependent variable. A novel con- ict event dataset for Lebanon geo-referenced to the city level has been collected for this analysis. Over 2,500 con icts have been recorded based on information gathered from two news wire services. For the purposes of this study, the data has been subset to include only the following event types: 2 Political executions Arrests and raids by military and internal security targeting militant groups Bombings (including suicide) Assaults and clashes involving militants and/or government (targeting civilians, government forces, or militant groups) as well as assaults and clashes arising from sectarian tensions (e.g. Lebanese civilian attacks on Syrian refugees) Protests 2 Please refer to Appendix A for the dataset's codebook, which includes a more thorough discussion of the data collection procedure. Chapter 2. Refugees and the Ethnic Balance of Power 15 Individual shootings linked to militants or linked to sectarian tensions (e.g. shootings between Alawi and Sunni families in Tripoli) Abductions by militant groups This result is a dataset of about 1,300 violent con ict events across 25 months across the entire country of Lebanon. Figure 2.2 shows the number of violent con ict events over the entire sample period for each district in red. The number of violent con ict events has steadily increased for most districts and sharply increased for a few (Akkar and Baalbek) since data collection eorts began in 2013. Figure 2.3 shows the spatial distribution of con ict over time. Con ict appears to cluster in space with concentrations in the northeast and central districts of the country. Syrian Refugees Data on Syrian refugees in Lebanon is obtained from the UNHCR Syrian Refugee Response Team. 3 Each month, the UNHCR releases a map, in PDF format, of the total number of registered Syrian refugees for each district (not including Palestinians). Relying on this data, I have constructed a panel dataset with month-district as the unit of ob- servation. Data at the district level was rst made available in March of 2013, thus the temporal scope of this analysis begins in March of 2013 and extends to April of 2015. Data on refugees for the month of August 2014 is missing; therefore, the missing data was generated using exponential inter- polation. A time-series of refugee ows by district can be found in Figure 2.2. Following convention, I use the log of the number of refugees. It must be noted that these gures exclude Palestinian refugees eeing Syria, who for the most part are relocated to existing Palestinian camps throughout the country. I expect higher refugee numbers to result in greater levels of con ict. 3 http://data.unhcr.org/syrianrefugees/regional.php Chapter 2. Refugees and the Ethnic Balance of Power 16 Figure 2.2: Refugees and Con ict April 2013 - May 2015 Palestinian Camps In addition to the massive in ux of Syrian refugees, Lebanon also hosts a long-standing minority of Palestinian refugees who are located in twelve camps across eight dierent districts. Data on their geographic distribution is obtained from the UNRWA for Palestinians in Chapter 2. Refugees and the Ethnic Balance of Power 17 Figure 2.3: Con ict Events 2013-2015 the Near East. 4 This is a static and binary variable that does not cap- ture changes to the number of Palestinian refugees over the sample period. I expect districts that host Palestinian refugee camps to experience more con ict than districts without camps. 4 http://www.unrwa.org/palestinerefugees Chapter 2. Refugees and the Ethnic Balance of Power 18 Population Density Districts facing greater levels of resource scarcity are expected to be more susceptible to con ict. One proxy for con ict used in the current analysis is population density. More densely populated districts impose greater strains on municipal governments, can drive down wages, and increase competition over access to government services. Therefore, I expect more densely populated districts to experience more con ict. More- over, I test the conditional relationship of population density and refugees on con ict with the expectation that the positive impact of refugees on con- ict levels is intensied in districts with higher levels of population density. I include a measure of population density (population/km2) based o of Lebanese population records at the district level from 2007. District popu- lation varies from 21,650 people/km2 (Beirut) to 62 people/km2 (Jezzine), with a mean of about 1631 people/km2. Income Low income per capita has been shown to increase the likelihood of con ict at the country level (Fearon and Laitin 2003). However, the re- lationship between income levels and con ict at the micro level is less clear. On the one hand, districts with lower income levels may face increased com- petition over access to government resources and employment opportunities, which can increase levels of con ict. On the other hand, con ict may corre- late more with higher income districts than poorer ones because wealth can provide opportunities for looting (Hegre, stby and Raleigh 2009). Thus, I include a measure of income that captures the percent of households in each district whose income is below the poverty threshold. Data on income levels is drawn from a national survey of households administered by Lebanon's Central Administration of Statistics. 5 Ethnoreligious Tensions According to the EBP theory introduced in this chapter, districts with a sensitive ethnoreligious balance of power are ex- pected to suer higher levels of con ict. Moreover, because I expect refugee 5 http://www.cas.gov.lb Chapter 2. Refugees and the Ethnic Balance of Power 19 ows to alter the delicate ethnoreligious balance of power between rival groups when refugees share a similar ethnoreligious make-up with the host population, I suspect the eect of refugees on con ict to be particularly pronounced in districts with sensitive ethnoreligious tensions. At its core, EBP theory involves two key factors, one related to characteristics of the host population and the other related to the characteristics of the incoming refugees. Not only must host populations contain rival ethnoreligious groups characterized by a delicate balance of power, the incoming refugees must in- clude at least one group involved in this rivalrous relationship. Without both these factors in play, refugee ows would not alter the existing balance of power between these groups. Unfortunately, data on the precise ethnore- ligious composition of incoming Syrian refugees is inaccessible. Moreover, Lebanon has not conducted a census since 1922, thus our knowledge of the distribution of ethnoreligious groups across Lebanon is also incomplete. Thus, in order to address these challenges, I make two assumptions. First, past studies have shown that refugees tend to ee to states where their co- ethnic kin reside (Moore and Shellman 2007; R uegger and Bohnet 2015). In addition, the three largest ethnoreligious groups in Syria -the Alawi, Shia, and Sunnis -all have counterparts among the Lebanese population. Thus, I assume that among the refugees owing from Syria into each of Lebanon's twenty-six districts, a substantial majority are either Sunni, Alawi, or Shia. Second, because the current primary political chasm in the country revolves around the divide between Shias and Alawis, on the one side, and Sunnis, on the other, I assume that the ethnoreligious balance of power in any dis- trict with substantial Sunnis and Shias/Alawis is at risk of shock due to the in ux of Syrian refugees from across the border. Based on these assumptions I operationalize pre-existing ethnoreligious ten- sions in two ways. First, any district with a mixed presence of Sunnis and Shia or Alawi is coded as having a sensitive ethnoreligious balance of power. Chapter 2. Refugees and the Ethnic Balance of Power 20 Second, any two ethnoreligiously homogeneous districts that border one an- other are also coded as having a sensitive ethnoreligious balance of power if and only if the populations of the two districts in question are from rival groups (Shia/Alawi vs. Sunni). The idea here is that although each district is homogeneous, their borders represent regions of the country where op- posing groups are in close geographic contact with one another. Therefore, refugee in uxes into these districts are also expected to increase levels of violence. As mentioned earlier, precise data on the distribution of ethnoreli- gious groups across has yet to be collected, therefore, I rely on approximate estimates based o of Lebanon's unique confessional voting system. Each ethnoreligious community has an allotted number of seats to represent their district in Parliament, which is suppose to be proportionate to each group's relative populations in that district. The threshold for representation per district is 5%. I use these seat ratios and population levels drawn from the year 2009 to reverse engineer the number of individuals from each ethnoreli- gious group for each district. 6 Then, I drop any ethnoreligious group whose numbers fall below 10% of the total district's population. Finally, using the remaining ethnoreligious groups for each district, I follow the two coding procedures laid out above. In total, eight of Lebanon's twenty-six districts are coded as having pre-existing ethnoreligious tensions. These districts are identied by light blue highlights in Figure 2.2. Distance to Capital Previous work has demonstrated that con icts tend to cluster in regions far from the capital (Buhaug 2010). As the distance from the capital increases, the projection of the central government's power diminishes and this aects the geography of con ict in two ways. First, armed groups opposing the central government are more likely to confront government forces in regions where the discrepancy between their capabil- ities is at its least. Second, the government's diminished power in regions far from the capital also hinder its ability to pacify intercommunal tensions. 6 This data is obtained from the Lebanese Elections Data Analysis (LEDA) project. Chapter 2. Refugees and the Ethnic Balance of Power 21 As a result, I expect higher levels of violence in districts further from the capital. Neighboring Con ict As Figure 2.3 shows, con ict seems to cluster in space. Thus, to account for this spatial clustering, I include a control for the levels of violence in all neighboring districts. I dene neighbors as two districts that share at least one point in common. I expect levels of con ict in one district to be positively correlated with levels of con ict in neighboring districts. I have developed a complimentary website and interactive web app using the Shiny platform. 7 The web app allows the user to visualize data on con ict, population density, Syrian refugees, and the distribution of ethnoreligious groups in Lebanon using a series of choropleth maps, time series graphs, and bar charts. A series of hypervariate data visualizations displaying each of these variables can also be found in Appendix B. 2.3.3 Method The research design and data discussed above result in four methodological challenges that must be addressed. First, the observations in the dataset likely exhibit temporal dependence. In other words, I expect con ict in one temporal period to increase the likelihood of con ict in subsequent temporal periods. Following Weidmann and Ward (2010), I address this problem in two ways. I employ robust standard errors (HAC) that control for temporal autocorrelation. Alternatively, I include in my model two temporal lags of the dependent variable. Second, because con ict events cluster in space, observations in my model are likely not independent of one another. Thus, in order to control for the eect of the presence of con ict events in neighboring 7 Note that the website takes some time to load: https://cmohamma101.shinyapps.io/lebapp/ Chapter 2. Refugees and the Ethnic Balance of Power 22 districts I follow two dierent approaches. First, I cluster my standard errors by region to capture neighborhood eects. Alternatively, I include a rst- order spatial lag of the dependent variable. A signicant coecient for the spatial lag suggests the level of con ict in one district is aected by the levels of con ict in adjacent districts. Third, my research design faces potential problems of endogeneity. I address these concern by lagging my time variant explanatory variables by one time period to ensure the causal arrow points from refugee ows to con ict and not the other way around. Finally, my dependent variable is a count measure that suers from an overdispersion of zeroes. To account for this, I employ a negative binomial regression. 2.4 Results 2.4.1 Negative Binomial Regression In this section I present the results of a series of regression models. I begin with an examination of my unconditional hypotheses. Model 1-3 employ robust standard errors clustered by administrative region (each of Lebanon's twenty-six second-order administrative districts are a member of one of six rst-order administrative regions). Model 1 in Table Table 1 conrms H 1 and H 2 , and H 4 . Controlling for the presence of Palestinian refugee camps and proximity to the capital, Syrian refugees ows, high population density, and pre-existing ethnoreligious tensions all increase levels of violent con ict in Lebanon. While the presence of Palestinian refugee camps increases con ict levels as expected, the current analysis counterintuitively nds that con ict levels are actually higher closer to the capital. Next I turn to the conditional hypotheses presented earlier. The analysis in Model 2 does not support the conditional hypotheses related to population Chapter 2. Refugees and the Ethnic Balance of Power 23 Table 2.1: Con ict Events (1) (2) (3) (4) (5) Palest. Camp 0.55 0.73 0.65 0.34 0.38 (0.12) (0.14) (0.12) (0.13) (0.12) Syrian Ref. 0.67 0.46 0.60 0.48 0.40 (0.07) (0.09) (0.08) (0.08) (0.08) Income 0.01 0.01 0.01 0.01 0.01 (0.01) (0.01) (0.01) (0.01) (0.01) Ethn. Tension 0.51 9.50 0.57 0.42 4.87 (0.15) (1.91) (0.15) (0.17) (1.94) Dist2cap. 0.01 0.01 0.01 0.01 0.01 (0.00) (0.00) (0.00) (0.00) (0.00) Pop.Dens. 0.0001 0.0001 0.0004 0.0003 0.0001 (0.00) (0.00) (0.00) (0.00) (0.00) Ref.XTension 0.92 0.49 (0.17) (0.17) Con ict.t1 0.06 0.05 (0.01) (0.01) Con ict.t2 0.04 0.03 (0.01) (0.01) Con ict.s1 0.01 0.02 (0.01) (0.01) Ref.XPop.Den. 0.00004 0.00004 (0.00) (0.00) Constant 7.64 5.55 6.96 5.59 4.85 (0.87) (0.98) (0.92) (0.95) (0.92) Observations 650 650 650 598 598 Log Likelihood 996.31 902.36 993.89 876.55 873.98 AIC 2,006.63 1,820.73 2,003.79 1,775.10 1,769.96 Note: p<0.5; p<0.01; p<0.001 Chapter 2. Refugees and the Ethnic Balance of Power 24 density H 3 . Although population density increases con ict levels, its inter- action with refugees is not signicant at conventional levels and its eect size is near zero. Model 3 tests the eect of refugees on con ict conditional on pre-existing ethnoreligious tensions. Controlling for all other factors, the eect of refugees on levels of violence is most acute in districts with pre-existing ethnoreligious tensions, which supports H 5 . This suggests EBP theory explains one mechanism by which refugee ows result in con ict. Figure 2.4: Predicted Probabilities for High vs Low Tensions Models 4 and 5 introduce the temporal and spatial lags discussed earlier. 8 The results from the previous models remain robust to the inclusion of spa- tial and temporal controls. The coecient for both temporal lags is signi- cant, suggesting that levels of con ict in one time period are in uenced by levels of con ict in the two previous time periods. Moreover, the signicant coecient in front of the spatial lag shows that higher con ict levels in one district contributes to higher con ict levels in neighboring districts. 8 Robust standard errors are kept but are not clustering by region. Also note that including rst and second-order temporal lags in the model drops the model'sN to 598. Chapter 2. Refugees and the Ethnic Balance of Power 25 Figure 2.4 shows the predicted probabilities (based on Model 5) along with condence intervals for the eect of refugees on con ict for districts with and without pre-existing sectarian tensions. The dierence between the eect of districts with and without ethnoreligious tensions on con ict clearly manifests itself at higher levels of refugee ows. In the following section, I re- estimate my model using random xed-eects and a novel spatial regression estimation technique that accounts for potential simultaneity bias in the research design. 2.4.2 Robustness Check In this section, I implement two checks of robustness. First I rerun the model using random eects. Random eects are useful when data is orga- nized hierarchically and observations are not independent of one another, typically this is the case with cross-sectional panel data. Observations are repeated at dierent time periods on the same units of analysis. Second, recent work suggests that adding spatial and temporal lags to conventional regression models results in simultaneity bias (Franzese, Hays and Schaer 2010). Moreover, this endogeneity overestimates the coecients of contagion variables at the expense of non-spatial variables. I employ a novel estimation technique using a Markov Chain Monte Carlo (MCMC) approach developed by Chi and Zhu (2008) and implemented in the computing language by Weidmann and Ward (2010). However, the implementation of this estima- tion technique has yet to be generalized to spatial count models. Thus, I run the analysis using a spatial autologistic model with a binary dependent variable. I code any district-month with more than two incidents of con ict as a 1 and all remaining district-months as a 0. Random Eects The rst two models in Table 2 shows the results of the RE estimation. Model 1 includes the rst-order temporal lag while model 2 Chapter 2. Refugees and the Ethnic Balance of Power 26 does not. 9 Temporal lags have been dropped from the second model because temporal lags tend to correlate with the random intercept in RE models, which biases the coecient of the lag variable (usually too large) and the coecient of the explanatory variables (usually too small) (Ousey, Wilcox and Fisher 2011). As expected, the eect size on the conditional variable is reduced and drops in signicance to the 0.10 threshold. However, in Model 2, where the temporal lag has been removed, the conditional variable's eect size increases and achieves signicance at the conventional 0.05 level. The marginal eect of refugees on con ict in districts with no pre-existing eth- noreligious tensions also remains positive and signicant, suggesting other mechanisms are also at work that tie refugees to the outbreak of violence. Spatial Logit w/ MCMC The nal model in Table 2 shows the results of the spatial logit model using the MCMC estimation techniques identied above. Once again, the conditional variable (Refugees X Ethnoreligious Tension) remains signicant despite the transformation of the dependent variable from a count to a binary measure and the use of MCMC estimation techniques instead of Maximum Likelihood. The results of Model 3 suggest that the interaction of pre-existing ethnoreligious tensions and refugee ows increases the odds of con ict by a factor of 3.4. 2.5 Conclusion Past studies have shown that refugees from neighboring countries increase con ict risk. In this study, I have investigated the mechanisms underlying that nding. Specically, I introduced a model that explains transborder con ict contagion as a function of pre-existing ethnoreligious tensions that are exasperated by the arrival refugees of similar ethnoreligious stock. I 9 The second-order lag is dropped from the model because its coecient was insignicant and its respective model produces a higher AIC score suggesting that the inclusion of the second lag reduces overall model t. Chapter 2. Refugees and the Ethnic Balance of Power 27 Table 2.2: Random Eects and Spatial Logit MCMC (1) (2) (3) Palestinian Camps 0.555 0.612 0.909 (0.355) (0.368) (0.334) Syrian Refugees 0.384 0.377 0.576 (0.139) (0.0204) (0.192) Income 0.0238 0.0248 0.008 (0.0197) (0.010) (0.021) Ethnoreligious Tension 3.94 5.59 12.721 (2.81) (2.72) (4.516) Distance to Capital 0.0158 0.0167 0.014 (0.00993) (0.0103) (0.010) Population Density 0.0000964 0.000104 0.0001 (0.0000404) (0.0000419) (0.0004) CONFLICT.t1 0.0237 0.778 (0.0116) (0.302) CONFLICT.t2 0.713 (0.297) CONFLICT.s1 0.0415 0.0441 0.038 (0.00799) (0.0079) (0.016) Refugees X Tensions 0.436 0.595 1.225 (0.261) (0.252) (0.425) Constant 4.62 6.967 8.769 (1.57) (0.923) (8.769) Observations 624 650 598 Log Likelihood 845.137 847.777 1.075295 AIC 1716.3 1717.6 1173.855 Note: p<0.5; p<0.01; p<0.001 Chapter 2. Refugees and the Ethnic Balance of Power 28 posit that countries with polarized ethnoreligious congurations are more susceptible to con ict contagion. When transborder refugees share similar ethnoreligious congurations with the host population they can alter the balance of power between rivals groups, resulting in higher levels of con ict. In addition, I argue that this mechanism should be identiable at macro and micro levels because substantial variation between units is expect at both levels. I test this theory against an alternative that identies refugees as a form of demographic pressure that aicts regions suering from resource scarcity most acutely. The results of my analysis conrm past studies that have identied a sig- nicant and positive relationship between refugees and con ict. Moreover, I nd support for the notion the eect of refugees on con ict is conditioned by the ethnoreligious congurations shared by the refugees and host popula- tion. Regions characterized by delicate ethnoreligious congurations are at particular risk for con ict contagion. These ndings are robust to a number of dierent specications and estimation techniques. Mechanisms linking the interaction of refugees and resource scarcity to con- ict, however, are unsupported by the data. While the marginal eect of population density on con ict is positive and signicant, low income is in- signicant as is the conditional eect of refugees and population density on con ict levels. These results also corroborate the ndings of scholars exam- ining the economic impact of refugees on host populations. For example, in a recent working paper, Del Del Carpio and Wagner (2015) show that the recent Syrian refugee in ux into Turkey's informal labor sector, while displacing natives, actually increased formal sector labor opportunities for locals. Foged and Peri (2015), examining refugee ows into Denmark be- tween 1991 and 2008, similarly nd that natives displaced by refugee arrivals tend to enjoy increased employment opportunities in less manual-intensive sectors that pay higher salaries. Chapter 2. Refugees and the Ethnic Balance of Power 29 What remains unclear is what type of violence the presence of refugees is most associated with. Do refugees increase violence among militants? Between government forces and rebels? Or does the presence of refugees increase the chances these groups target noncombatants? In the following chapter, I test the ethnic balance of power theory in the context of two dierent forms of violence. First, I examine coethnic refugees and ethnic rebellion using a global dataset of disempowered ethnic groups from 1975- 2009. Second, I examine the relationship between coethnic refugees and one-sided violence committed by government and rebel groups using a global country-year dataset. The results suggest that coethnic refugees have no signicant eect on the likelihood of ethnic rebellion but do increase the chances that rebels will target noncombatants when they do ght. I account for this nding by introducing a complimentary theory to EBP, the logic of population control, which argues warring parties face incentives to control populations in contested territory by protecting loyal civilians and targeting those sympathetic to rivals. Chapter 3 When Ethnic Groups Rebel: Refugees and Transborder Kin Abstract In this chapter, I explore the ethnic balance of power theory in the con- text of two dierent forms of con ict -ethnic armed rebellion and one-sided violence -using global panel data on intrastate violence at the group and country levels. Although, the analysis fails to identify a signicant associ- ation between the presence of coethnic refugees and ethnic armed con ict, coethnic refugees do signicantly increase the probability of rebel led one- sided violence. I account for this by introducing a complimentary theory to EBP, \the logic of population control", which contends that rebels have an incentive to target civilian supporters of rival groups as a means of undercut- ting the support these rivals enjoy. However, targeting civilians is costly if 30 Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 31 rebels cannot properly discriminate between their own supporters and civil- ians loyal to the government or opposing groups. In ethnically polarized states, rebels can take advantage of the salience of ethnicity by mobilizing along ethnic lines and using ethnic markers to aid in the identication of potentially disloyal civilians. This logic, also applies to coethnic refugees. Rebels view refugees that share ethnic ties with the civilian populations represented by rival armed groups (government or otherwise) as a threat and respond to this threat with violence directed towards these unarmed civilians. Therefore, I conclude that while coethnic refugees do not increase the likelihood of con ict onset, they do increase the chances that civilians will be targeted, refugee or otherwise. 3.1 Introduction Recent literature in the eld has identied a link between refugees and the spread of con ict across borders. One causal mechanism that I have oered to account for the ndings of previous work is the role that shared ethnic ties play as a catalyst for con ict. I argue when refugees share ethnic ties with politically active groups in their host state, their presence can alter the existing ethnic balance of power (EBP) between competing groups. If tensions are hostile between these rivals, then even minor shifts in demo- graphics that advantage one group over another can increase the number of violent events. In the previous chapter, I tested this proposition at the substate level in Lebanon and found evidence to support it. In this chapter, I test this EBP theory globally at the country and group levels in the context of two dierent forms of con ict -ethnic armed rebellion and one-sided violence. Although, the analysis fails to identify a signicant association between the presence of coethnic refugees and ethnic armed con ict, coethnic refugees do increase Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 32 the probability rebels will target unarmed civilians. I account for this nd- ing by introducing a complimentary theory to EBP, \the logic of population control", which contends that governments have an incentive to target civil- ian supporters of rival groups as a means of undercutting the support their rivals enjoy. However, targeting civilians is costly if governments cannot eas- ily discriminate between their own supporters and the civilians loyal to the opposition. The conditions are dierent when rival groups mobilize along ethnic lines because ethnic markers aid in the identication of potentially disloyal civilians. This logic, also applies to refugees who share ethnic ties with these same groups. 3.2 Ethnic Armed Rebellion and One-sided Vio- lence While the empirical approach of the previous chapter leveraged its disag- gregated unit of analysis to uncover interesting links between demographic shifts and outbreaks of violence, the approach adopted suers from three data limitations that can be overcome by moving the unit of analysis either to the country or ethnic group level. First, due to data limitations, the analysis of the previous chapter took for granted the ethnic background of the Syrian refugees owing into Lebanon. The data on Syrian refugees at the substate level in Lebanon that the UN- HCR has released does not include demographic data, such as religious or ethnic aliation. As a result, I have approached the empirical analysis with the assumption that Syrian refugees tend to migrate to regions of Lebanon inhabited by Lebanese whom they share ethnic and/or religious ties with. If this assumption holds true, however, I face another challenge. If ethnic refugees are the only form of refugees in a particular region then I cannot know if it is their shared ethnic status that is responsible for the increased Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 33 chances of con ict or if it is simply because they are forced migrants who bring with them weapons and dangerous ideologies from the battles elds they ed. Recent released data on the ethnoreligious background of refugees (ER dataset) can help ameliorate this problem (Wimmer, Cederman and Min 2009). Because this data is not territorially disaggregated, a novel country or group year analysis must be undertaken. Second, the logic behind the theory of EBP posits that coethnic refugees are a potential source of recruitment for combatant groups. Therefore, we should expect groups who enjoy the benets of recruiting from larger pools of coethnic refugees to be more likely to rebel against the state. Although the substate data collected for the empirical investigation of the previous chapter includes demographic information on the victim and perpetrator of each con ict event, records for all observations are incomplete and analogous demographic information pertaining to refugees at the substate level does not exist. The ACD2EPR dataset includes group level data on the outbreak of ethnic civil war (Vogt et al. 2015), which combined with the ER dataset (R uegger and Bohnet 2015), can form the basic design of a group-year analy- sis that examines the relationship between the presence of coethnic refugees and the likelihood of rebellion for each country's ethnic groups. As mentioned above, while the data from the previous chapter also includes information on the identity of victims and the perpetrators of violence, that sort of information is simply unavailable for many of the observations in the dataset. Therefore, while the analysis of the previous chapter revealed a statistically signicant link between refugees and violence, it is unable to identity what type of violence the presence of refugees is increasing. This is problematic for our understanding of the mechanics behind the refugee- con ict nexus. Without this information, we cannot know, for example, if the presence of refugees is destabilizing because these refugees are actively involved with ghting or if the increase in violence is a result of hostile groups targeting civilian refugees. Moving the analysis to the country level Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 34 allows me to leverage data on one-sided violence released by the Uppsala Con ict Data Program (UCDP) to test this proposition (Eck and Hultman 2007). Finally, due to the diculty in obtaining accurate geocoded data at the substate level, the analysis of the previous chapter was conned to the state of Lebanon. Using data at the group and country level allows me to expand the geographic scope of this analysis beyond the Levant. Thus, the purpose of this chapter is to provide a further test of EBP theory at two new levels of analysis, across new regions of the world, and with data that allows me to test all parameters of my theory at once. To that end, in this chapter I will introduce a group-year analysis that examines the impact of refugees on the likelihood of rebellion for their coethnic brethren in their host states. I will also present the results of a separate country-year analysis that looks at the impact of coethnic refugees on one-sided violence in host states. Ultimately, what I nd is that refugees exert no statistically signicant im- pact on the likelihood that their coethnic groups in the host state will rebel, but they do increase the probability of rebel led one-sided violence. On the one hand, both the substate and the country-year analyses revealed a substantial link between refugees and con ict conditioned by the presence of volatile ethnic tensions. On the other hand, the group-year analysis fails to uncover a signicant relationship between ethnic con ict onset and the presence of ethnic refugees. I account for this discrepancy by introducing a complimentary theory to EBP, which I label the \logic of population con- trol". I argue that the presence of coethnic refugees can alter the ethnic balance of power, but changes in the ethnic balance of power between rival groups contributes more to how con ict is fought than to the specic like- lihood of its occurrence. In other words, changes in relative power between rival ethnic groups impact the tactics of violence adopted by warring parties. The presence of refugees with ethnic ties to one group, I argue, increases Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 35 the incentives rival groups have to target refugees and their local coethnic civilians. The logic of population control contends that refugees are a valuable source of human capital in the form of recruits, public solidarity, and even eco- nomic activity. Rival groups are keenly aware of this and, thus, rightly or wrongly, view these refugees as a threat that needs to be controlled. Thus, coethnic refugee ows to states with volatile ethnic relations increases the likelihood of civilian targeted one-sided violence as rebels (or governments) target refugees with the aim of undercutting their support for their coethnic brethren. The remainder of this chapter is divided into two sections. In the rst section, I review the design and results for the analysis of the eect of refugees on the probability of ethnic group rebellion. In this second section, I introduce the design and present the results of the country-year analysis that reveals a link between the spread of coethnic refugees across borders and one-sided violence. I conclude with a discussion of a number of important observable implications of this theory, which I proceed to test in the following chapter. 3.3 Coethnic Refugees and Armed Ethnic Rebel- lion In the previous chapter, I examined how Syrian refugees contributed to in- creases in the levels of violence at the substate level in Lebanon. I found that refugees do increase con ict but I also discovered that the impact was even stronger for districts with previously volatile ethnic relations. Unfor- tunately, due to the limitations of the data collection process, I was unable Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 36 to identify precisely what kind of violence this entailed. Thus, it is un- clear whether this increased con ict takes the form of violence between rival armed groups emboldened by the arrival of pools of new recruits or if it is in the form of violence targeted specically at refugees and other unarmed civilians. Framed more simply, are refugees actively causing this violence as much of the media and even some academic work has framed the topic or are they victims of violence in the very places they have sought refugee? I argue that the empirical analysis of this chapter lends support to this latter claim. In this section, I have designed a ethnic group-year analysis that leverages newly released data on the ethnoreligious background of all refugees between 1975-2009, which I use to test the proposition that the presence of refugees emboldens their coethnies in the host state to rebel. 3.3.1 Hypotheses Given the ndings of the previous chapter, I argue that refugees increase the probability of con ict by altering the demographic balance between rival groups. Because they are a potential source of recruitment for groups they share ethnic ties with, rivals of these groups (or rivals in government) view these refugees as a potential threat. This destabilization is what EPB theory argues contributes to con ict in the rst place. The presence of refugees either emboldens their coethnic brethren in the host state, agitates rivals of these groups, or both; the result in each scenario is more con ict. From this, I derive my primary hypothesis: H 1 : The presence of coethnic refugees emboldens ethnic rivals excluded from sharing power with the government to rebel. Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 37 3.3.2 Methodological Approach and Data The methodological approach adopted in this section is designed to examine whether the presence of refugees with ties to host ethnic groups increases the likelihood these groups rebel against the government. Thus, I use a logistic regression model with country-group-year as my unit of analysis. That is, each observation is a record of each politically active ethnic group in each country in each year between 1975-2009. The dependent variable in this model is binary and indicates the onset of rebellion by a specic ethnic group in a particular country. I employ robust standard errors clustered by country-group (some ethnic groups exist in more than one country so I cluster using within country ethnic groups) to account for the nested nature of the panel data and I introduce region-xed eects to control for unob- served heterogeneity from one region of the world to another. I also include a count for the number of years of peace since the last rebellion as well as its squared and cubed term to account for temporal autocorrelation a la Carter and Signorino (2010). The data used for the dependent variable, armed rebellion by an ethnic group, is a recoding of UCDP/PRIO's Armed Con ict Data (ACD) by the coders of the Ethnic Power Relations (EPR) dataset at the University of California, Los Angeles (Wimmer, Cederman and Min 2009). The result is a dataset (the Ethnic Armed Con ict (EAC) dataset) of all politically active ethnic groups across the world categorized according to the \degree of access to central state power by those who claimed to represent them" (ibid, 326). It also includes the dependent variable used in this study {a binary indicator of the onset of armed ethnic rebellion. Armed ethnic rebellion is dened as armed con ict where at least 25 combatants have been killed annually and \at least one party is the government of a state" (ibid, 326). What makes it \ethnic" is that the rival party must be an armed group that makes \ethnic aims" (such as achieving ethno-national autonomy) and the armed rebels Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 38 must be mobilized along ethnic lines (i.e. recruits are drawn primarily from the same ethnic group and alliances are forged along ethnic ties). Once an ethnic group rebels, years of ongoing con ict are dropped from the analysis because this is a model that is testing the onset of ethnic con ict and not its duration. Furthermore, only ethnic groups excluded from power are included in the analysis. Therefore, if the ruling government is composed of parties representing one or more ethnic groups, those ethnic groups are not included in the analysis because by denition they cannot rebel against themselves. It must be noted that requiring one party to the con ict to be a government of a state, omits a wide range of forms of violence, such as violence against noncombatants, violence between rebels, terrorism, and even genocide. The primary independent variable of this study is the presence of coethnic refugees. A dataset of the ethnic background of all refugee ows from 1975- 2009 has recently been released, known as the Ethnicity of Refugees (ER) dataset (R uegger and Bohnet 2015). This dataset takes the dyadic data on refugee ows between countries released by the UNHCR and identies up to three dierent ethnic groups per ow {a majority group and up two minority groups using information drawn from experts as well as primary sources. Each group is allotted a particular share of the total number of refugees of each year and aggregate numbers are then reverse engineered from these proportions. I log the total value. I include a host of group and country level variables in the analysis. In terms of group level variables, I include, the log of a group's population, the log of a group's population as a share of total country population, the history of con ict between the ethnic group and the state (operationalized as the total number of ethnic con icts between the group and the government of that state), and whether the ethnic group had lost a share of power in the ruling coalition in the previous two years. In terms of country level controls, I include whether the state is a democracy, which I obtain from the Polity IV dataset (Marshall and Gurr 2014) and I use the log of GDP/capita drawn Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 39 from the World Bank's World Development Indicators (WDI). I also control for unobserved region-to-region heterogeneity using region xed-eects. All variables are lagged by one year. Table 3.1 shows the descriptive statistics for each of these variables. Next, I turn my attention to the results. Table 3.1: Descriptive Statistics Statistic N Mean St. Dev. Min Max Ethnic Con ict 251,580 0.005 0.067 0 1 Democracy 247,937 0.202 0.402 0 1 Previous Con ict 251,580 0.076 0.297 0 3 log(Share of Population) 251,580 1.008 1.098 0.002 4.595 log(Ethnic Refugees) 251,580 0.074 0.886 0.000 14.950 log(GDP/capita) 250,978 8.266 1.108 5.179 15.010 log(Population) 250,978 11.715 1.863 6.272 14.094 Peace Years 251,506 38.721 15.493 0 63 3.3.3 Results Table 3.2 shows the results of the logistic regression model with robust stan- dard errors in parentheses. All three peace year terms are signicant, indi- cating a clear time trend in the outbreak of rebellion over the sample time period. Of the remaining controls, previous con ict history and recent down- grade in power sharing status increase the chances of ethnic group rebellion, while armed groups representing ethnic groups with greater populations are actually less likely to rebel against the government. Most importantly, H 1 is unsupported by the data, which suggests there is not enough evidence to support the claim that refugees increase the chances of armed rebellion by their transborder kin. This says nothing, however, of levels of one-sided violence that may emerge in response to the presence of refugees. In the next section I examine whether the presence of refugees with shared ethnic ties can contribute to increased chances of civilian targeted violence. Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 40 Table 3.2: Ethnic Refugees and Armed Rebellion Dependent variable: Onset of Ethnic Con ict Intercept 0:33 (2:43) Democracy 0:18 (0:39) log(Ethnic Refugees) 0:75 (1:01) log(Share of Pop) 0:23 (0:12) Previous Con ict 1:18 (0:18) log(GDP/capita) 0:46 (0:17) log(Population) 0:24 (0:08) Peace years 0:17 (0:05) Peace years 2 0:01 (0:00) Peace years 3 0:00 (0:00) Downgraded Status 2:03 (0:37) Asia 1:25 (1:13) Eastern Europe 1:33 (1:10) Sub-Saharan Africa 0:66 (1:16) North Africa/Middle East 1:43 (1:12) Latin America 0:01 (1:20) Observations 247,865 AIC 12803 2 1,694.988 (df = 15) Note: p<0.05; p<0.01 Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 41 3.4 Coethnic Refugees and One-sided Violence The previous section failed to uncover evidence linking the presence of co- ethnic refugees with more ethnic armed con ict. In this section, I examine whether refugees alter the ethnic balance of power in ways that increase one-sided violence against civilians. 3.4.1 Hypotheses Although the previous section did not uncover a statistically signicant relationship between ethnic armed rebellion and the presence of coethnic refugees, coethnic refugees may impact other forms of con ict, such as vi- olence aimed at unarmed civilians. In fact, I posit that the presence of refugees with ethnic ties to the host population does alter the ethnic bal- ance of power between groups but that these changes do not make armed con ict more likely. Instead, they change the way in which armed con ict is waged. Specically, I argue that refugees are a valuable source of human capital in the form of recruits, public solidarity, and even economic activity. Rival groups are keenly aware of this and, thus, rightly or wrongly, view these refugees as threats to be controlled. Thus, the presence of coethnic refugees increases the likelihood of one-sided violence as rebels (or govern- ment forces) attempt to control the population by targeting refugees with the aim of undercutting the support they provide to their coethnic brethren in the host state. I label this the logic of population control because warring parties have an interest in controlling populations sympathetic to their own cause while undercutting the support of populations their rivals enjoy. Because coethnic refugees are identiable by the same ethnic markers that place their local coethnic brethren at risk of targeting, they too are at risk of persecution. Governments as well as warring rebels have an interest in targeting the Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 42 civilian support base of their rival groups. However, targeting civilians is costly if con icting parties cannot distinguish between sympathizers and those sympathetic to their rivals. If these con icting parties can identify supporters based on ethnic markers, this task is eased. Refugees, like the local civilian population, then are at risk of persecution if they share ethnic traits with these locals. From this, I derive my remaining two hypotheses. H 2 : Rebels in states hosting coethnic refugees are more likely to target civil- ians during con ict than rebels in states that do not host coethnic refugees. H 3 : Governments of states hosting coethnic refugees are more likely to tar- get civilians during con ict than government's of states without coethnic refugees. 3.4.2 Methodological Approach and Data To test each of these hypotheses, I rely on logistic regression. I include region-xed eects to account for unobserved heterogeneity from region to region and I employ robust standard errors clustered by country to obtain unbiased estimates from the panel data. To account for temporal autocor- relation, I follow Carter and Signorino (2010) and include the count of years since the last occurrence of violence along with its squared and cubed terms. The dependent variable for one-sided violence (1975-2013) is derived from the UCDP one-sided violence dataset (Eck and Hultman 2007), which I convert from a count measure to a binary one, where 1 indicates at least one violent event in that given year and 0 none. The primary independent variable is the logged number of coethnic refugees a state hosts derived from the ER dataset (R uegger and Bohnet 2015). As a check of robustness, I also include a binary measure of this variable where 1 indicates the presence Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 43 of coethnic refugees and 0 indicates their absence. 1 In addition to coethnic refugees, I include a measure of the logged number of none coethnic refugees in each country to determine whether refugees with no ethnic ties to the country have a comparable eect on violence. 2 I include the same set of controls for both models. I expect that rough and mountainous terrain hinders eorts to target civilians, particularly for government led forces. To control for this, I include a logged measure of the share of mountainous terrain. Furthermore, past studies have shown that various forms of internal con ict have become more prevalent and more deadly since the Cold War (Kalyvas 2001), so I include a dummy for the end of the Cold War. It is reasonable to suspect that one-sided targeting of civilians is more likely in the midst of civil war, so I include a dummy indicator for ongoing civil wars drawn from the ACD dataset (Wimmer, Cederman and Min 2009). I also include a number of social, economic, and political indicators, including the log of GDP/capita, the log of the infant mortality rate, and a dummy indicating whether a country experienced less than 2% growth in the previous scal year all drawn from the World Bank WDI dataset. Moreover, I control for whether a country is a democracy using the Polity IV coding of countries above the threshold of 6 (Marshall and Gurr 2014). Similarly, I control for regime stability by including the log of the number of years since a change of more than 2 points in a country's Polity IV score. I also suspect that one-sided violence is least likely to occur in highly ethni- cally fractionalized societies and highly homogenized ones. 3 In the former, 1 The results are robust to both measures so I present the results of the count variable only. 2 I also restrict this measure to only include refugees from neighboring states. The results remain the same, thus, I report ndings only for all refugees, irrespective of origin. 3 I rely on a measure of \ethnic fractionalization" as opposed to \ethnic polarization" because past research has shown that polarization aects the level of con ict intensity. In contrast, fractionalization has been show to increase the chances of con ict but not its severity. For more on this topic see Esteban and Ray (2008). Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 44 if ethnic groups are too numerous, mobilization usually takes place along alternative lines (Esteban and Ray 2008), which may reduce the ability of warring parties to identify loyal and disloyal pools of civilians (Caselli and Coleman 2013), including refugees (see next chapter). Likewise, the salience of ethnicity is obviously low in homogeneous societies and for similar rea- sons I suspect homogeneous societies to also see less one-sided violence. To control for this curvilinear relationship where very low and very high levels of fractionalization are more peaceful than moderately fractionalized soci- eties, I include both the commonly used measure of ethnic fractionalization derived from the Herndahl index and its squared term. Finally, all the vari- ables in this analysis are lagged by one year. Table 3.3 shows the descriptive statistics for each of these variables. Table 3.3: Descriptive Statistics Statistic N Mean St. Dev. Min Max One-Sided Viol. Rebel 5,185 0.105 0.306 0 1 One-Sided Viol. Gov. 4,994 0.059 0.235 0 1 Civil War 5,065 0.23 0.42 0 1 Post Cold War 4,963 0.59 0.49 0 1 Democracy 4,994 0.42 0.49 0 1 Slow Growth 4,600 0.31 0.46 0 1 Peace Years 5,026 12.54 10.01 0 35 Regime Durability 4,963 2.50 1.33 0.00 5.30 log(Refugees) 5,185 7.70 4.24 0.00 13.45 log(Ethnic Refugees) 4,745 2.24 4.39 0.00 14.95 Ethnic Frac. 5,095 0.42 0.29 0.001 0.93 log(GDP/capita) 5,065 1.30 1.35 1.83 8.10 log(Population) 5,065 9.40 1.47 6.10 14.09 log(Infant Mortality) 4,958 0.14 1.04 2.71 1.81 log(Mount. Terrain) 5,095 2.15 1.39 0.00 4.42 Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 45 3.4.3 Results I test two separate models -one for one-sided violence committed by gov- ernment forces and the other for one-sided violence committed by rebels, the results are shown in Table 3.4. 4 What I uncover is that the presence of refugees with ethnic ties to the host population signicantly increases the chances of civilian targeting by rebel forces but not by government forces. The rst model fails to accept H 3 , which anticipates increased chances of civilian targeting by government forces as a response to the presence of coethnic refugees. This may suggest that government forces do not view refugees supporting their transborder kin as threats or, if they do, they may face other constraints in employing violence, such as international obliga- tions to safeguard refugees. More likely, however, a selection eect in the destination choices of refugees accounts for this insignicant nding. A civilian eeing con ict faces three choices when caught in the midst of crossre; 1) she must choose whether to stay with her family or to relocate them to a safer place 2) if she stays, she may take up arms or continue bearing the brunt of the war as a noncombatant but if she ees, she faces the additional choice to either migrate to safer regions of her own country as an internally displaced person (IDP) or to seek asylum across international borders as a refugee 3) if she makes the choice to lead her family to safety across international borders, she must nally decide where the safest place to ee is among her available choices. That choice is partly constrained by geography; mountains, deserts, and bodies of water are harder and more dangerous to traverse. But that choice is also shaped by social and political dynamics. For example, there is ev- idence suggesting refugees, ceteris paribus, are more likely to ee to more 4 Region-xed eects are removed from the government led model because of conver- gence issues caused by perfect separation. This was not an issue for the other model, which retains the region-xed eects. Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 46 Table 3.4: Ethnic Refugees and One-sided Violence (Government) (Rebel) (Intercept) 3:82 4:72 (0:88) (1:00) Regime Durability 0:05 0:07 (0:09) (0:10) log(Refugees) 0:02 0:07 (0:03) (0:04) log(Ethnic Refugees) 0:01 0:05 (0:02) (0:02) Ethnic Fractionalization 2 2:67 4:50 (1:64) (1:83) Ethnic Fractionalization 2:43 3:55 (1:38) (1:66) log(GDP/capita) 0:23 0:44 (0:16) (0:21) Democracy 0:60 0:29 (0:24) (0:27) log(Population) 0:24 0:26 (0:09) (0:11) log(Infant Mortality) 0:23 0:01 (0:28) (0:28) Slow Growth 0:66 0:29 (0:20) (0:23) log(Mountainous Terrain) 0:02 0:05 (0:08) (0:08) Civil War 1:31 2:20 (0:26) (0:27) Post Cold War 0:85 0:71 (0:32) (0:37) Peace years 0:73 0:97 (0:11) (0:09) Peace years 2 0:05 0:07 (0:01) (0:01) Peace years 3 0:00 0:00 (0:00) (0:00) Num. obs. 4398 4275 AIC 1193.8 1204 2 1193.8 1207.6 Fixed-Eects None Region Note: p<0.05; p<0.01 Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 47 democratic and less volatile neighbors than to autocratic and unstable ones. Schmeidl (2001); Moore and Shellman (2004, 2007); R uegger and Bohnet (2015) recently provided empirical conrmation of a long standing suspicion that refugees tend to ee to countries where their transborder kin reside. Moreover, leaders of neighboring states may even limit entry of asylum seek- ers due to concerns over their own domestic stability (Neumayer 2005). As a result of a con uence of these various factors, refugees tend to migrate to countries whose governments are most supportive of their presence. This may include countries where the transborder kin of these refugees reside, such as Albania, which hosts ethnic Albanian refugees from Kosovo. It may also include countries who share political fortunes with certain groups among the refugees, such as Muammar Qadda who actively opened his borders to refugees eeing the Chadian civil war or Saddam Hussein who gave shelter and arms to the Iranian Mujaheddin-e-Khalq during the Iran-Iraq war. If refugees are more likely to ee to countries ruled by governments that are more tolerant of and receptive to their presence then the association between the presence of transborder refugees and rebellion would be understated. If refugees do migrate to countries where they enjoy shared ethnic and politi- cal ties with those in power, however, then opposition groups may view these incoming refugees as liabilities to be dealt with. If the incoming refugees not only share political ties with groups in power, but also ethnic ones, then rebels can more easily identify which refugees are potential recruits for their war eort and which are potential recruits for the war eort of their rivals. The second model tests this by regressing the use of one-sided vio- lence by rebels on the logged number of coethnic refugees along with a host of control variables. These ndings conrm H 2 ; rebels respond to the in- creased presence of transborder refugees by targeting noncombatants. There is substantial anecdotal evidence to back up this claim as well. For exam- ple, the news media has highlighted the persecution Syrian refugees face in Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 48 Lebanon at the hands of ISIS sympathizers, 5 how Boko Haram specically targets IDP camps in Nigeria with suicide bombings and organized attacks, 6 and Congolese rebels who target camps in Nord-Kivuamong 7 among others, which suggests refugees, like local populations, contribute to shaping the contours of con ict. However, not all refugees are targeted in equal fashion. In fact, the model of rebel led one-sided targeting suggests none coethnic refugees have no sta- tistically signicant impact on the likelihood of one-sided violence. I have argued that warring parties have an incentive to \control the population" by undercutting the support base of their rivals. Doing so involves target- ing civilians sympathetic to the cause of their rivals, an aim that is easier to achieve when society's divisions fall along ethnic lines. Refugees form an important support base for rival parties. Rebels know this so they tar- get refugees, like civilians, as a means of undermining their rivals' capacity. Therefore, it is entirely expected that none coethnic refugees complicate ef- forts to distinguish friend from foe. That is not to say that the presence of none coethnic refugees is not destabilizing in other ways. Indeed, the literature I discussed earlier has typically focused on the relationship be- tween con ict and refugees irrespective of ethnicity. For example, refugees, it has been argued, may contribute to economic pressures that increase the likelihood of con ict in the host state irrespective of their ethnic ties, 8 but in the context of one-sided violence, it appears they exert no statistically signicant impact. 5 See for example, http://english.al-akhbar.com/node/21557 6 See for example, http://www.theguardian.com/world/2016/feb/10/nigerian-refugee- camp-hit-by-double-suicide-bombing-boko-haram 7 See for example, http://www.pbs.org/newshour/updates/africa-july-dec08- congo 1031/ 8 See for example, the Regional Center for Strategic Study's work on this topic at http://www.rcssmideast.org/en/Article/144/Syrias-refugees-burden-neighboring- countries-economies-#.VxdCvdCNvt8 Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 49 Of the control variables in the model of rebel led violence, the log of popula- tion, ongoing civil war, and the post Cold War era all increase the probability of rebel-led one-sided violence. Moreover, the coecient of ethnic fraction- alization is negative and signicant, while its squared term is positive and signicant, which suggests only moderately fractionalized states are at risk of one-sided con ict (this is true of both government led and rebel led one- sided violence). Finally, the positive and signicant coecients for North Africa/Middle East and for sub-Saharan Africa identify these two regions as the most dangerous for civilians relative to the referent group of Western Europe, the US, and Canada. It must be noted that the empirical observation that greater levels of coeth- nic refugees are associated with greater levels of civilian targeted violence does not by itself establish that refugees are necessarily the targets of such violence. It is perfectly reasonable, for example, to suggest that the asso- ciation between coethnic refugees obtains because they are either actively involved in the violence or because they embolden their transborder kin to rebel. However, if we are to believe that refugees tend to ee to countries that are more inviting of their presence, and we have every reason to believe they do, then the association between coethnic refugees and government led, not rebel led, violence should be signicant, which it is not. In other words, if refugees ee to countries with more accepting governments and if there existed a refugee warrior phenomena that was fueling one-sided violence, then we would observe government allied forces targeting more civilians, as opposed to rebel forces targeting them. In the concluding section, I highlight a number of observable implications of these ndings and discuss approaches to testing them, which I carry out in the next chapter. Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 50 3.5 Conclusion In the previous chapter, I showed that the presence of refugees increases incidents of con ict in Lebanon, particularly in districts with volatile ethnic relations. I accounted for this nding with the suggestion that refugees alter the ethnic balance of power between rival ethnic groups, which increases opportunities and incentives to engage in con ict. In this chapter, I further explored this proposition using a global dataset of all ethnic groups. The results failed to identify a link between coethnic refugees and the onset of ethnic armed rebellion, which can only be reconciled with the ndings of the previous chapter if at least one of the following three are true; one, the association between the refugees and con ict is understated because refugees ee to countries where their transborder kin are in government and not opposition (selection eect), two, coethnic refugees contribute to con ict severity but not to the onset of con ict itself, or, three, coethnic refugees contribute to other forms of con ict all together, such as one-sided violence. I tested this latter possibility using a global country-year dataset of one- sided violence. The results of this latter analysis point to a link between the presence of coethnic refugees and rebel led one-sided violence. That is, coethnic refugees are associated with violence not because they are perpet- uating it, but because they are victims of rebels groups who target them. This is a major nding that frames the link between refugees and violence in ways yet to be explored by the eld. For example, of the six major works in the eld that quantitatively assesses the relationship between refugees and con ict, not a single one attempts to theoretically or empirically link increases in violence to attacks on civilians, including refugees themselves, despite the abundant anecdotal evidence reported in mainstream and alter- natives media sources. 9 In fact, theories tying refugees to con ict in each 9 See for example, Salehyan and Gleditsch (2006); Salehyan (2008); Forsberg (2009); Bohnet (2012); Fisk (2014); Shaver and Zhou (2015). In fact, Zolberg, Suhrke and Aguayo Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 51 of these articles focus on how refugees produce dividends in the form of re- cruits for armed rebels. Much of the work in this dissertation relies on this same theoretical understanding but argues that this is precisely what makes refugees valuable targets for repression. Thus, the ndings of this chapter provide a framework for thinking about refugees and their role in con ict that departs from conventional views of refugees as \warriors" in con ict (Zolberg, Suhrke and Aguayo 1992; Adelman 1998). If refugees are a source of power in domestic politics that can advantage one group over its rivals, then rebels have an incentive to control the popula- tion of their country as a means of bolstering their own support base and undermining the support of their rivals. This logic of population control entails targeting civilians and refugees who support their rivals' claims to power. Doing so, however, can be costly if rebels cannot distinguish between noncombatants who sympathize with them and those sympathetic to their rivals. But if society is mobilized along ethnic lines, then discriminating be- tween friend and foe is easier. In the context of refugee ows, this suggests that ethnic refugees should increase the chances of one-sided violence but non-ethnic refugees should not. This much, at least, has been conrmed by the empirical analysis of the current chapter but there are other observable implications of this theory as well. Some of these observable implications have been identied and examined by other scholars. For example, if rebels have an incentive to augment their public support while undercutting that of their rivals and if doing so is easier when society is divided along eth- nic lines, then we should expect civilian targeting to be more severe during ethnic con ict than nonethnic con ict. Indeed, there is much work to back up this claim (Valentino, Huth and Balch-Lindsay 2004; Eck 2009; Caselli and Coleman 2013) but there are (1992) coined the term \refugee warrior" decades ago to describe forced migrants who take up arms in their destinations of refuge. Chapter 3. When Ethnic Groups Rebel: Refugees and Transborder Kin 52 other observable implications that have yet to be examined. In the follow- ing chapter I look one of these. I suggest that if rebels have an incentive to target noncombatants and if doing so is easier under conditions of ethnic con ict, then we should observe a statistically signicant dierence between the patterns of forced migration resulting from ethnic civil wars and those re- sulting from nonethnic con ict. I hypothesize that ethnic civil wars produce a greater share of refugees relative to internally displaced persons (IDPs) than nonethnic civil wars and I nd empirical evidence in support of this claim. The theory of the logic of population control can account for this nding. Civilians eeing violence have the choice to ee to safer regions of their own countries or to seek asylum in foreign states. If all the civilians fear is the crossre of con ict, they will be more inclined, I argue, to stay as close to their homes as possible without placing themselves and their families at risk. Typically, this means relocating to other regions of the same country. However, if the combatants have an incentive to specically target civilians, then civilians will nd it in their interest to relocate to places warring par- ties cannot follow, which in the context of an actual functioning Westphalian system of states means seeking asylum across international borders as reg- istered refugees. The next chapter will explore these propositions in more detail. These ndings also have important policy implications. If the prevailing view in academic and popular circles is that refugees actively contribute to con icts with and among populations that host them, then policy aimed at addressing the refugee-con ict nexus may do more harm than good. My results suggest that policies aimed at securing the safety of refugees them- selves are one of the best ways to reduce the violence associated with their presence. Chapter 4 The Logic of Population Control Abstract Why do some civil war torn countries produce more refugees relative to their internally displaced population and others displace more of their population internally than across borders? Surprisingly, the relationship between inter- nally displaced persons and con ict has been woefully underexplored. The aim of this chapter is to ll this gap in the literature. Using a panel dataset of civil con icts by country-year from 1993-2010 and a two-step Heckman selection model, I show that civil wars fought along ethnic lines produce greater refugee ows relative to IDP ows than non-ethnic civil wars. I ac- count for this nding by relying on insights drawn from the previous chap- ters. Specically, I argue that in con icts where combatants are recruited along ethnic lines, ethnic markers allow for less costly and more discriminate targeting of rival civilian populations, which in turn increases the share of 53 Chapter 4. The Logic of Population Control 54 forced migrants who seek refugee across borders relative to those displaced internally. 4.1 Introduction Much of the literature on refugees and con ict contagion has focused on the conditions under which refugees are more or less likely to spread violence across borders. There is some evidence, for example, that the eects of refugees on the spread of violence is conditioned by a number of factors, such as shared ethnic ties with the host population, resource scarcity, and political instability in the host state. Specically, this dissertation has argued that refugees contribute to violence in two primary ways. First, refugees alter the existing balance of power between competing ethnoreligious groups at the substate level, which can result in increased levels of violence in the host state if such relations are polarized to begin with. Second, coethnic refugees also increase the likelihood of one-sided violence, especially at the hands of non- state actors, suggesting that their presence is a potential threat to certain rebel groups. The logic in both these circumstances is uniform; refugees represent both a threat and a boon to rebel groups. As much as coethnic refugees are an attractive target for recruitment and resource extraction for one rebel group, their presence is viewed as a threat by competing rebel groups. What is less clear in the literature, however, are the dynamics that con- tribute to the emergence of refugees in the rst place. In many ways, we've put the proverbial cart before the horse. We know how con ict refugees contribute to the spread of violence yet we do not know which con icts are most likely to result in a large exodus of refugees in the rst place. From the perspective of the con ict contagion literature, we know which countries Chapter 4. The Logic of Population Control 55 are at greatest risk of contagion but we do not know which countries are the most contagious. While the forced movement of people is a feature endemic to most con ict- ridden states, 1 the proportion of civil war-torn states with active refugee ows represent only 51% of all con ict years between 1993-2011. 2 The range in the number of forced migrants eeing con icts is also non- trivial. For example, the 1994 civil war and genocide in Rwanda resulted in the ight of a little over 1.8 million refugees. Similarly, in 2006, the intercommunal violence and civil war that plagued war-torn Iraq resulted in nearly 1.2 million people eeing their country. Compare those annual gures, with the number of refugees that ed Tajikistan's civil war in the 1990s {only 5,000 refugees were displaced over the course of ve years. Of course not all victims of forced movement in con ict become refugees {a good portion of them are displaced internally. In fact, in many cases the forced displacement of victims of war within boarders exceeds the displacement of victims of war across borders. Take for example Pakistan's civil war against the Taliban insurgency, which resulted in the internal displacement of over 1.7 million people in the year 2009 alone. This stands in great contrast to the 2,279 individuals who sought and found refuge outside Pakistan's borders that same year. Why do some civil war torn countries produce more refugees relative to their internally displaced population and others displace more of their population internally than across borders? Surprisingly, the relationship between inter- nally displaced persons and con ict has been woefully underexplored. The aim of this chapter is to ll this gap in the literature. Using a panel dataset 1 For example, between 1993 and 2011, only one country enduring con ict featured no refugees living outside its borders, Papua New Guinea between 1993-1994. 2 Active refugee ows refer to the number of refugees forced to ee in a given year from a given country, as opposed to refugee stock, which is a count of the cumulative total of refugees. Chapter 4. The Logic of Population Control 56 of civil con icts by country-year from 1993-2010 and a two-step Heckman selection model, I show that civil wars fought along ethnic lines produce greater refugee ows relative to IDP ows than non-ethnic civil wars. I ac- count for this nding by relying on insights drawn from the previous chap- ters. Specically, I argue that in con icts where combatants are recruited along ethnic lines, ethnic markers allow for less costly and more discriminate targeting of rival civilian populations, which in turn increases the share of forced migrants who seek refugee across borders relative to those displaced internally. In what remains of this chapter, I introduce a theory, which I label the \the logic of population control", that accounts for variation in levels and pat- terns of forced migration resulting from civil wars. This logic of population control is grounded in theoretical contributions from three disparate research programs {ethnic con ict, civilian targeting, and forced migration. In the sections that follow, I brie y review the literature in each of these research agendas in that order, I provide a stylized narrative of the theory, and I present my methodological approach to testing the observable implications of the theory. Finally, in the last two sections I provide an interpretation of the results and conclude with a discussion that ties these ndings to the con ict contagion research agenda. 4.2 Literature Review 4.2.1 Ethnicity and Ethnic Civil Wars Although the literature disaggregates civil wars in a number of dierent ways, such as \old vs new wars" (Kalyvas 2001), \territorial vs governmen- tal" (Gleditsch et al. 2002), \symmetric,non-symmetric, conventional, and irregular" (Kalyvas 2005), etc., perhaps the most widely employed typology Chapter 4. The Logic of Population Control 57 of civil war contrasts ethnic with non-ethnic violence. This study embraces this latter approach. In general, justications for analytically separating ethnic and nonethnic con ict focus on motivations, structural conditions, and opportunities that dierentiate one form from the other. 3 Justications based on motivation typically highlight the unique set of eth- nic grievances that result from dierential treatment of groups. For example, Sambanis (2001) nds evidence that ethnic war is waged in response to po- litical grievances, while nonethnic war is associated with lack of economic opportunity. In other words, ethnic discrimination is a uniquely motivating factor in rebellion. Others argue that ethnic and religious discrimination produce grievances that result in con icts that are more dicult to manage. For instance, Hassner (2009) suggests that religious grievances that result in territorial con ict are more intractable and dicult, if not impossible, to resolve due to the motivating beliefs of those involved over the sacredness of the territory being fought over. Similarly, Fearon (1995, 1998, 2004) argues that ethnic wars produce commitment problems that result in extended du- ration of warfare relative to nonethnic con icts. Furthermore, Kaufmann (1996) contrasts \ethnic civil wars" with \ideological civil wars". The key dierence between these two forms of war he suggests is the level of exibility of individual loyalties. Where loyalty in ideological con icts is quite uid, in ethnic con icts it is far more rigid. According to Kaufmann, the rigid- ity of loyalty is what makes ethnic con icts particularly dicult to resolve peacefully. 3 This distinction has not been accepted wholesale in the eld. See for example (Mueller 2000, 215) who argues ethnic con icts resemble other forms of violence in so far as they are waged by small groups of combatants purported to \ght and kill in the name of some larger entity." Chapter 4. The Logic of Population Control 58 In contrast to motivation-based explanations, justications based on the dif- ferent structural conditions of ethnic and nonethnic warfare rely on the eth- nic composition of con ict prone societies. Researchers who proer these jus- tications identify variation in levels of ethnic polarization and/or fractional- ization as factors that condition warfare in unique ways that require tailored approaches (Reynal-Querol 2002; Garcia-Montalvo and Reynal-Querol 2004; Bhavnani and Miodownik 2009). Finally, opportunity justications typically focus on the conditions unique to mobilization when ethnicity is salient and highly politicized. These sorts of explanations look to the ways that ethnic markers shape the ability of groups to mobilize for warfare. Caselli and Coleman (2013) argue ethnic markers help enforce group membership by reducing \free-riding". In homogenous societies members of the losing group can easily pass themselves o as mem- bers of the winning coalition but in ethnically heterogeneous societies ethnic markers make this task far more dicult. The fact that leaders in homoge- nous societies understand this ex post dilemma reduces their incentives to mobilize ex ante in none-ethnically salient states. These ethnic markers not only reduce the free-riding problem, they also allow groups to easily identify and target loyal populations for their recruitment eorts. Thus, ethnicity plays a role in reducing the coordination costs associated with mobilization. These reduced barriers to mobilization, of course, result in greater risks of con ict escalation Eck (2009). While ethnic markers provide groups with more ecient means to mobilize, they also provide combatants with more eective means of targeting pools of populations loyal to rivals. In eect, the opportunity structures of ethnic civil wars increase incentives to target civilians (Valentino, Huth and Balch-Lindsay 2004). The distinction between ethnic and nonethnic con ict made in this study is motivated by this very phenomenon. Thus, I embrace the opportunity- based justication for analytically separating ethnic from nonethnic con ict. If ethnic markers provide combatants with better opportunities to target the Chapter 4. The Logic of Population Control 59 loyal population base of rival groups, then it stands to reason that civilians are at a greater risk of victimization in ethnic civil wars. The resulting atmosphere of violence should increase the incentives of forced migrants to seek refuge across international borders (as opposed to hiding among the general population). In the following section, I discuss the extant literature on the targeting of civilians during combat and tie it to the literature on forced migration and ethnic con ict. 4.2.2 Civilian Victimization Civil war literature has identied a number of factors that contribute to the victimization of civilians during con ict, including autocratic regime types (Engelhardt 1992; Har 2003; Valentino, Huth and Balch-Lindsay 2004), use of guerrilla tactics by rebels (Valentino, Huth and Balch-Lindsay 2004), and desperation to win (Downes 2006). Scholars studying this topic have also embraced the ethnic-nonethnic distinction in examining civil wars and their analyses suggest ethnic con ict increases the chances of civilian target- ing relative to nonethnic con ict (Downes 2006; Fjelde and Hultman 2010). Valentino, Huth and Balch-Lindsay (2004, 381) argue that because \it is more dicult for individuals to disguise their ethnicity than their political aliation" combatants are better able to discriminate between friendly and hostile civilian populations. In the absence of ethnic cleavages, combatants nd it dicult to distinguish friend from foe; as a consequence, nonethnic civil wars actually result in less civilian targeting than ethnic ones (ibid). This has important implications for the patterns of forced migration that re- sult from domestic warfare. The following section identies these observable implications after brie y reviewing recent quantitative literature on forced migration. Chapter 4. The Logic of Population Control 60 4.2.3 Forced Migration A number of signicant factors have been shown to increase the risk of forced migration in a given a country. Typically, these fall into one of two categories {push or pull factors. Push factors refer to characteristics and conditions of countries that force populations to ee their homes to safer des- tinations abroad. Examples of push factors include natural disasters (Drabo and Mbaye 2011) and various forms of violence (Schmeidl 2001; Moore and Shellman 2004; Davenport, Moore and Poe 2003). Pull factors refer to at- tractive neighborhood characteristics that make movement across borders less costly than seeking refuge within borders (or staying put all together). These include the regime type of destination countries (Moore and Shellman 2007), shared ethnic aliation in destination countries (R uegger and Bohnet 2015), and hospitable neighboring geography (Moore and Shellman 2007). An analysis of the factors that contribute to one form of forced migration (international refugees) or another form (internally displaced persons) must incorporate both of these elements. Surprisingly, the relationship between external and internal displacement has only received scant attention in the eld thus far. The only work to date that I am aware of that quantitatively compares refugee ight and IDP movement to one another is Moore and Shellman (2007). They employ a two-step Heckman model on a global panel analysis of country-years between 1976-1995. Their analysis suggests that levels of violence in neighboring states increases the proportion of IDPs ows relative to refugee ows. In their formulae, victims of displacement weigh the dan- gers they perceive at home against those they see in their potential points of destination. When conditions in their potential points of destination are more favorable than the conditions they face at home, they are more likely to migrate (and vice versa). The current analysis embraces both the methodol- ogy and theoretical foundations of Moore and Shellman's article. However, Chapter 4. The Logic of Population Control 61 it departs from their approach in two ways. First, where Moore and Shell- man's narrative emphasizes the agency migrants enjoy in determining their own choice of destination, the present analysis examines the role that com- batants play in limiting that agency and in uencing that choice. Second, where Moore and Shellman examine \characteristics of countries" that aect patterns of forced migration, the present analysis also examines the charac- teristics of con icts that aect that phenomenon. This latter point is not trivial and, in fact, represents a major departure from the methodological approach of Moore and Shellman and much of the existing literature on forced migration. Instead of examining patterns of forced migration for all countries as others have done, I am specically interested in the patterns of forced migration un- der the strategic environment that victims and combatants of civil war nd themselves. Therefore, I restrict my analysis to country-years experiencing at least one civil war. I argue this exclusion criteria is justied because the choice to stay, ee to other regions, or to seek refugee in other countries is inherently dierent for those in an environment of organized warfare than those eeing economic hardships, natural disasters, or other forms of polit- ical violence. 4 That dierence, I contend, arises from the unique incentives combatants (i.e. rebels and governments) have to control the ow of pop- ulation within and between their territories and regions controlled by their rivals. Furthermore, I suggest the type of war being waged shapes these very incentives, perhaps to a great deal. The next section explains why. In what follows, I introduce the \logic of population control", I establish the 4 This exclusion criterion can potentially introduce bias into the model's estimates if civil war-torn and peaceful states dier in unobservable ways either related to the like- lihood of forced migration or to the proportions of forced migrants that are refugees (or IDPs). Please see the Data and Methods section for a more in-depth discussion of the selection problem and the tools I use to manage the issue. Chapter 4. The Logic of Population Control 62 assumptions of the theory, I introduce the relevant actors and their inter- ests, I formalize the opportunity structures that govern their behaviors, and I provide a stylized account of the theory at work. 4.3 The Logic of Population Control The logic of population control is based on the premise that combatants have an incentive to control populations loyal to them and to undermine the control their rivals enjoy over their own loyal populations. In the previous chapters, I examined the role of ethnicity in fomenting con ict in regions of Lebanon heavily populated by Syrian refugees. Specically, I argued that refugees alter the balance of power between rival ethnic groups, which can result in increased mobilization eorts towards organized violence. My anal- ysis showed that 1) higher refugee numbers increase con ict and 2) ethnically polarized regions are particularly susceptible to this threat. Refugees, like the local population, are a vital resource for the groups that they share ethnic aliations with. Therefore, rival groups view refugees of rival ethnic groups as a threat. In Lebanon, this manifested itself in one-sided attacks on refugees, organized attacks between rival groups, and increased intercom- munal violence amongst the local population. The analyses from the previous chapters suggest that control over popula- tions is an important goal for combatants. Asserting control over loyal popu- lations and undercutting the support their rivals enjoy provides combatants with a comparative advantage in mobilization eorts and in resource extrac- tion. In the same way that groups have an interest in targeting refugees of rival ethnic groups, combatants also have an incentive to target the local populations of rival ethnic groups. Thus, for the very same reason that the presence of coethnic refugees is more destabilizing in a host country than the presence of non-coethnic refugees, civil wars fought along ethnoreligious Chapter 4. The Logic of Population Control 63 divisions are also more destabilizing than those fought along other societal divisions (i.e. class, ideology, etc.). They are destabilizing in ways that are particularly destructive towards civilians. In the previous section, I dis- cussed the literature on civilian targeting that identied ethnic civil wars as the most violent form of civil war that civilians endure. The theory in- troduced here examines how this logic of population control in uences the patterns of forced migration that result from ethnic and nonethnic con ict alike. Next, I turn my attention to the assumptions of the theory. 4.3.1 Assumptions I begin with a number of important assumptions. First, although civil- ians make a `decision' to ee (either within or across borders), ultimately \forced migration" invariably entails a level of coercion that really places more agency in the hands of perpetrators of the violence (combatants) than in the hands of its victims (refugees and IDPs). Therefore, I assume that combatants maintain some degree of in uence over the decision of civilians to stay, to ee to other regions of the country, or to seek refugee in other states. My second assumption is that population control is an important aim of combatants in civil war. Control over populations allows parties to a con- ict to extract human and material resources in the form of soldiers, eld doctors/nurses, taxes, and general economic production. Population control also allows combatants to homogenize their populations in support of their war aims by expelling, detaining, or killing sympathizers of the opposing group. For the very same reason that a controlled but robust and supportive population is critical for success in con ict, warring parties have an interest in undercutting the population support their rivals enjoy. One common way of doing so in the midst of warfare is directly targeting civilians. Chapter 4. The Logic of Population Control 64 My next assumption rests on the dierence between war-torn states where ethnicity plays a salient role in the society's divisions and where it does not. The costs of and barriers to population control in ethnically salient con icts are, as already discussed, less than in nonethnic con icts because ethnic markers allow groups to better discriminate between supporters and those sympathetic to opposing groups. Therefore, I contend that dierent types of civil wars produce dierent logics of population control. Furthermore, many civil wars increase the threat of bodily harm to civil- ians and they react to this threat by eeing their homes. They can either ee to other regions of their own country or attempt to seek refugee across the border in neighboring states. I assume, all else equal, movement across international borders is more costly than internal displacement. Therefore, civilians should prefer to relocate as close to their original location of resi- dency as possible without exposing themselves to high risks of victimization. Ultimately, the decision to relocate internally as opposed to seeking asylum elsewhere, wrests on whether the civilian thinks his/her probability of vic- timization is higher outside the country than inside. That calculation, I argue, is in uenced by the likelihood that combatants will target civilians, which is itself in uenced by the type of civil war waged. My nal assumption is that rightly or wrongly, civilians are viewed by rebels (or governments) representing rival groups as a potentially threatening re- source at the disposal of their enemies. Under conditions of non-ethnic con- ict the tools of targeted repression become blunted because governments (or rebel groups) nd it dicult to distinguish between loyal and disloyal pools of civilians. Because mass repression can undercut support among previously loyal populations, rival victims of war are better able to seek refugee undetected within the borders of the state by hiding among popula- tions that warring groups are hesitant to target or nd it dicult to target eectively. Chapter 4. The Logic of Population Control 65 In contrast, the conditions of ethnic con ict produce a dierent logic of repression and ight. Ethnic markers provide warring groups with the ability to not only identify supportive populations but hostile ones as well, a point (Kaufmann 1996, 21) makes when he suggests that combatants \can treat all members of the other ethnic group as enemies without risk of losing a recruit." In eect, conditions of ethnic con ict limit the domestic destination options available to eeing victims of war. 4.3.2 Actors and Interests The notion that combatants have incentives to control the type of forced migration that results from their participation in ongoing violence I label the theory of population control. The narrative of this theory focuses on three actors {rebels, governments, and civilians. Each of these groups has their own interests as well. Rebels want to limit the resources of the government and one way to do so is to target populations loyal to them. Similarly, governments want to undercut the support base of opposing rebel groups, so they too benet from targeting civilian populations loyal to their rivals. At the same time, both governments and rebels have an incentive to safeguard loyal populations. Therefore, overzealous targeting of civilian groups that puts their own potential supporters at risk is costly. Finally, civilians too are self-interested actors whose primary goal is to reduce their personal likelihood of persecution at the hand of combatants. LetT l denote the probability of targeting loyal civilians and letT d denote the probability of targeting disloyal civilians. Line 4.1 identies the incentives versus the constraints combatants face under conditions of nonethnic civil war, T l =T d (4.1) Chapter 4. The Logic of Population Control 66 That is, when groups are mobilized along nonethnic lines, the probability of targeting loyal civilians should be roughly equal to the probability of target- ing populations loyal to rivals because distinguishing between friend and foe is more dicult without observable markers that aid in dierentiating one from the other. In contrast, under conditions of ethnic con ict, the likeli- hood of targeting disloyal civilians is greater than the likelihood of targeting one's own population base, T l >T d (4.2) The balance between T d and T l is in uenced by the ability of combatants to target rivals' civilian support discriminately. When groups are mobilized along ethnic lines, the dividends, from attacks on civilians is higher than the risks associated with civilian victimization because there are fewer costs to targeting civilians discriminately. However, when ghting takes place along the lines of non-ascriptive identities, such as class, fewer observable markers exists that reveal the loyalties of the civilian population. Under these conditions, combatants nd the chances of targeting loyal civilians, T l and disloyal ones, T d , roughly the same. Let C denote the costs associated with targeting civilians and B the benets. When T d >T l then, C >B (4.3) In other words, when the probability of targeting disloyal civilians (T d ) is higher than the probability of targeting loyal ones (T l ) then the costs (C) of targeting civilians is less than the benets (B) accrued from the tactic. If, V e , represents the level of violence against civilians during ethnic civil war and V n the level of violence during nonethnic civil war, then, Chapter 4. The Logic of Population Control 67 V e >V n (4.4) Line 4.4 indicates that under conditions of ethnic con ict violence against civilians is greater than under conditions of nonethnic con ict. Civilians also make a cost-benet analysis. P o refers to the probability of perceived victimization in a civilian's country of origin andP a refers to the probability of perceived victimization in a civilian's potential destination of asylum. When P o are equal P a , P o =P a (4.5) civilians will choose to relocate to domestic destinations. That is, when the perceived chances of being targeted is the same at home that it is abroad (or in the process of traveling abroad) then civilians will elect to stay within the borders of their country. Likewise, if the perceived probability of persecution abroad is higher than at home, P o <P a (4.6) then civilians will seek refugee in regions of the country they nd safer than the battle grounds from which they escaped. On the other hand, if the perceived probability of persecution at home is higher than the perceived likelihood of victimization abroad (and in the process of traveling abroad), P o >P a (4.7) then civilians will seek refugee across international borders. LetR represents the number of refugees eeing civil war andI represent the number of IDPs Chapter 4. The Logic of Population Control 68 eeing civil war. F e and F n represent the share of forced migrants that are refugees for ethnic and nonethnic civil wars respectively. F e = R R +I (4.8) F n = R R +I (4.9) Holding all else equal, combatants nd C <B, when T d >T l is true. And whenT d >T l is true, civilians will calculateP o >P a . It is the contention of this chapter thatT d >T l is true during ethnic con icts more so than during nonethnic con icts. If P o >P a then, all else equal, F e >F n (4.10) In other words, during conditions of nonethnic con ict the share of forced migrant ows that are composed of refugees is fewer than during conditions of ethnic con ict. Thus compared to nonethnic civil wars, civil wars charac- terized by ethnic cleavages are more likely to produce con ict environments that push civilians to seek asylum among foreign populations than to nd shelter among their own. 4.3.3 Stylized Narrative The logic of population control, thus, identies two patterns of ight. When the costs of targeting loyal populations are higher than the benets accrued from employing this tactic, C > B, civilians become bystanders in danger of crossre. As such, their pattern of ight can be characterized as one that Chapter 4. The Logic of Population Control 69 avoids regions of high intensity warfare between rival groups. IfC <B, how- ever, civilians are not mere bystanders in war. In eect, they become targets of war and their ight patterns will re ect their intent to not only avoid re- gions of high intensity con ict but also peaceful places either controlled by rival adversaries or at risk of control (or targeting) by such groups. From these two sets of equivalencies we are left with a stylized narrative that describes the conditions under which civil wars produce more or less refugees as a share of total forced migrants. Imagine for a moment a head of a household caught in the crossre between rebel and government forces. She faces the choice to stay with her family, ee with them to more peaceful regions of the country, or make the potentially dangerous trek across inter- national borders to safety. Therefore, she makes two choices. First, whether to stay or ee and, second, once on the move, whether to relocate to other regions or to seek refugees across the border. Her rst decision is simple; she will relocate with her family in attempt to evade the threat of crossre. Her second choice, however, depends on the perceived likelihood that the violence will follow her and her family to their choice of destination. The perception of this likelihood is in uenced by the deliberate actions of the warring parties because at the very moment she is deciding whether to simply ee or seek actual asylum, group decision makers are faced with a choice whether to target her and her family as they ee. On the one hand, if ethnic markers reveal the direction of her loyalty then the choice to target is simpler to make. If the civilian's ethnic markers reveal her to be a of a rival ethnic group then the risks associated with targeting her are lower and decision makers will likely make the choice to target. If the civilian markers identify her as supportive, then rival groups may target her instead. On the other hand, if ethnic markers do not coincide with the cleavages of warfare, then choosing to target such a civilian is costly {you may or may not have just targeted a supporter. Thus, under conditions of nonethnic Chapter 4. The Logic of Population Control 70 con ict the propensity to target civilians is reduced and civilians like that head of the household use this information when deciding where to ee. If they experience deliberate targeting by combatants, they will take this as a cue that the violence will follow them to where they ee. Therefore, civilians will be more likely to ee to regions outside the reach of warring parties (i.e. outside the country). But if they are not targeted and, as such, view the threat they face in the con ict as incidental, then they will be more inclined to seek refugee in places they consider safe from crossre. The following section posits a set of hypotheses derived from the discussion above. 4.4 Hypotheses In the previous two sections I introduced the logic of population control. In this section I discuss a number of observable implications of this theory and formally present a set of hypotheses, which I empirically test in subsequent sections. The most basic observable implication of the theory is that in the midst of civil war, refugees respond to direct threats to their personal safety by relocating either to other regions within their own countries or to safe regions in other countries. From this I derive my rst hypothesis: H 1 : Controlling for all other factors, one-sided violence against civilians increases the probability of forced migration among civil war torn states. Moreover, for the two-step Heckman correction to work properly at least one variable must act as an \instrument" for the eect of forced migration; that is, it must predict forced migration without aecting the composition of forced migrants. In this case, one-sided violence acts as the instrument. If one-sided violence has a signicant independent marginal eect on the composition of forced migration then its role as an instrument is compro- mised because it aects the probability of forced migration to begin with. But if, as I suspect, one-sided violence increases the probability of forced Chapter 4. The Logic of Population Control 71 migration (H 1 ) without aecting the composition of forced migration (i.e. the share of forced migrants that are refugees), then introducing an inter- action of one-sided violence and civil war type (as is suggested by (H 3 ) and ((H 4 ) below) should pose no problems for estimation even if the interaction itself is signicant. Thus I hypothesize the following null eect: H 2 : One-sided violence has no independent eect on the proportion of forced migrants that are refugees. The key proposition presented in the section discussing actors and their interests is that combatants have incentives to target civilians and when they do, refugees have a choice to ee to other regions of their home countries or to ee to safety across international borders. If the war is fought along ethnic lines, then combatants can more eectively target these civilians based on observable ethnic markets. Realizing this, civilians in the midst of ethnic con icts will nd it less attractive to seek refugee in their own countries, which brings me to my next hypothesis: H 3 Controlling for all other factors, compared to nonethnic civil wars, civil wars characterized by ethnic cleavages produce more refugees as a share of total forced migrants. As discussed earlier, ethnic markers increase the incentives combatants have to target civilians because it allows for more discriminate targeting and previous research has veried this claim. Therefore, ethnic markers act as a sort of intervening variable where the eect of one-sided violence on migrant patterns is heightened if the con ict is waged along ethnic lines. In fact, I suspect that one-sided violence against civilians increases the share of refugees relative to IDPs but only under conditions of ethnic con ict. Thus, the eect of civil war type on composition of forced migrants is conditional on the presence or absence of one-sided violence: H 4 : Controlling for all other factors, the positive eect of ethnic civil wars Chapter 4. The Logic of Population Control 72 on the share of forced migrants that are refugees is conditional on the pres- ence of one-sided violence. If the civilians being targeted nd themselves in nonethnic con ict, however, then I expect they will be more inclined to relocate to regions within their own country than to neighboring states primarily because of the diculty in traversing international borders. If it is easier to hide among the local population, which it is under conditions of nonethnic con ict, civilians will do so. From this I derive my next hypothesis. H 5 : Controlling for all other factors, one-sided violence against civilians under conditions of nonethnic con ict should decrease the share of forced migrants that are refugees. H 6 : Controlling for all other factors, civil war type should not aect the share of forced migrants that are refugees absent one-sided violence. Similarly, I do not expect ethnic con ict to increase the share of forced mi- grants that are refugees when one-sided violence is not a feature of warfare. In fact, if the only mechanism by which ethnic con ict aects patterns of forced migration is the level of one-sided violence it produces, then it stands to reason that ethnic civil wars free of one-sided violence should have no sig- nicant eect. H 6 re ects this expectation that the eect ethnicity on the share of migrants that are refugees is conditional on the presence of one-sided violence against civilians. The next section introduces the methodological approach, estimation techniques, and data. Chapter 4. The Logic of Population Control 73 4.5 Data and Methods 4.5.1 Methodological Approach The empirical strategy adopted in this chapter is primarily driven by the methodological challenge posed by selection bias in at least one stage of the analysis. The problem of sample selection is a form of omitted variable bias that arises from a nonrandom selection of data. When a subset of the data is systematically excluded due to a particular factor then exclusion of the subset can bias estimates. This problem can emerge as an artifact of the research design or when sub- jects self-select into certain groups. As an example of the latter, if a re- searcher is interested in the eect of drug use on mental illness, simply regressing mental illness on past drug use will yield bias results if individu- als use drugs to self medicate. The researcher may identify more drug use with greater mental health problems but the relationship may very well be overstated if we think the reason for that higher drug use had to do with greater levels of mental illness to begin with. This is also known as endoge- naity bias because the selection into treatment groups is endogenous to the outcome. In regards to the former, if a researcher is interested in examining the relationship between education and wage oers but only has access to wage oers of individuals currently employed then the factors that predict participation in the labor force may bias the relationship between levels of education and wage oers, specially if labor force participation and wages are related. Both these forms of bias may exist in the current study and their threats to inference are to be taken seriously. I begin with a discussion of subject- self selection (also known as endogeneity or treatment selection) referred to above in the anecdote of the study of drug use and mental illness. I have already established that ethnic con icts better facilitate combatant Chapter 4. The Logic of Population Control 74 recruitment and mobilization. Not only do ethnic markers allow leaders to better commit to the rank and le, they provide leaders with assurances that their spoils of war will not be diluted by free riders ex post. Furthermore, these ethnic markers allow combatants to more eectively target civilians loyal to rival groups. Therefore, there exists at least some nominal benet to mobilizing along ethnic lines. However, leaders may face countervailing factors that push them to mobilize along alternative cleavages instead. If leaders are capable of determining the societal ssures along which war is waged and for some unobservable reason(s) leaders that choose to mobilize their rank and le along nonethnic lines also happen to be leaders that are less willing to target civilians during warfare, then the eect of ethnic civil wars (in contrast to nonethnic wars) on the proportion of forced migrants that are refugees will be overstated. One way to address this threat to inference is to rst model the likelihood that civil war is fought along ethnic lines and then move on to examining the outcome of interest. This is may be unnecessary, however, because 1) I contend that while leaders certainly have an interest in mobilizing their rank and le according to the ways they see most benecial to their cause, their actually ability to do so in meaningful ways is very limited. In fact, the very politicization of ethnicity is a process that takes shape over many years and under the in uence of many factors and 2) the research design faces other antecedent biases in the chain of selection that are more important to address. I turn to those next. Although, I begin with a total sample of all country years between 1993- 2010 (based on data availability), I am only interested in the pressures that forced migrants face in the midst of civil war. Thus, I subset the data to country-years experiencing an ongoing civil war. An argument could be made that excluding all none civil war states will introduce selection bias in the nal estimates. But this misses the point; exclusion of observations based on a particular selection of a population of interest does not induce Chapter 4. The Logic of Population Control 75 bias on its own. In fact, Wooldridge (2010) asserts \sample selection can only be an issue once the population of interest has been carefully specied" (551). He suggests that if the researcher is interested in a subset of a larger population then the appropriate approach is to specify a model for the part of the population based on randomly selected data from that subset. In this study, I am only interested in examining the strategic environment victims and combatants of civil wars nd themselves. Therefore, selection based on my \population of interest" {civil wars {should not in uence my results. 5 Next, I need to examine whether some civil war torn states produce more refugees relative to IDPs than others and accounts for that dierence. How- ever, not all civil war-torn states experience forced migration. In fact, be- tween 1993-2010 49% did not. Therefore, I need to further subset the pop- ulation of civil war-torn states to include only those with active refugee and/or IDP ows. This stage of selection may bias estimates. Here I em- ploy the two-stage heckit correction method developed by Heckman (1979). The heckit method entails identifying a \selection equation" {a probit model that estimates the likelihood that a given civil war-torn state experiences forced migration. From the selection equation I obtain the inverse of the 5 Indeed, the canonical heckit correction method (Heckman 1979) for identifying and addressing selection eects fails to uncover a selection bias produced by exclusion of non- civil war states; the inverse Mill's ratio of the bivariate probit estimation does not attain signicance. Had selection bias been revealed at that stage as it has been in a subsequent stage, then a bivariate probit model would be used to identify both selection eects (se- lection into civil war states and, given a set of civil war-torn states, selection into those that produce forced migration) and the inverse Mill's ratio obtained from both selection equations would be included as additional regressors in the outcome equation that es- timates the eect of civil war type on pattern of forced migration. Such an approach, a multi-stage selection model, would be warranted if the inverse Mill's ratios (identied below) of both selection equations attain signicance. However, this method identies bias only in the second stage of selection, selection into civil war-torn states that produce forced migration. In other words, excluding civil war free country-years from the sample does not bias the results, while exclusion of states that do not produce forced migrants does. Because selection at the rst stage poses no threats of inference, I only present results of the standard two-stage heckit model. Please see Appendix C for the results of the multi-stage selection model. Chapter 4. The Logic of Population Control 76 Mill's ratio, which is the ratio of the probability density function to the cu- mulative distribution function of a distribution, and use it as a regressor in the outcome equation {a linear model estimated in ordinary least squares. Given the nested nature of the panel data, I also employ robust standard errors clustered by country. Moreover, I include region xes eects to control for unobserved region-to-region heterogeneity in estimating the rst stage of the equation (selection into civil war-torn states). Finally, to control for temporal eects, I include a lag of the dependent variable in both stages of estimation. What I end up with is a two-stage heckit selection model with 361 civil war-torn country-years between 1993-2011. The following section formally introduces the model and estimation technique. 4.5.2 Estimation Technique As mentioned in the previous section, my analysis involves a two-step esti- mation process. Given a population of war-torn states, I must rst estimate the likelihood that such states produce refugee and/or IDP ows. In the subsequent step, among the remaining pool of states (civil war-torn states that produce forced migrant ows), I must identify factors that in uence the share of forced migrants that are refugees. The rst stage of analysis involves a basic probit regression written as follows, P (D i t = 1jZ i t) =(Z ) (4.11) whereD it indicates forced migration (D it = 1 if countryi experienced forced migration in timet andD it = 0 otherwise),Z it = 1 is a vector of explanatory variables, is a vector of unknown parameters, and is the cumulative distribution function of the standard normal distribution. In the second stage, I include a transformation of the predicted individual probabilities Chapter 4. The Logic of Population Control 77 (inverse Mill's ratio) as an additional regressor in a model I estimate using OLS, which is notated as, f it =X it +u it (4.12) wheref it denotes the ratio of refugee ows to total migrant ows in country i in time t, which is not observed if the country does produce any forced migrants in a given year. Based on equations 4.11 and 4.12, the condi- tional expectation of the proportion of refugees to all migrant ows given the country experienced forced migration is written as follows, E[f it jX it ;D it = 1] =X it +E[u it jX it ;D it = 1] (4.13) If we assume the error terms are jointly normal (i.e. multivariate normal distribution) then we obtain the following, E[f it jX it ;D it = 1] =X it + u (Z it ) (4.14) where is the correlation between unobserved determinants of propensity to produce forced migrants and unobserved determinants of the ratio of refugee ows to total forced migrant ows u, u is the standard deviation of u and is the inverse Mill's ratio evaluated at Z it . Hall (2002) argues that the standard two-step estimator results in inconsistent standard error estimates. This can be overcome using a variety of robust methods. Alternatively, rather than estimating the equations using the standard two- step process, a maximum likelihood estimator (MLE) approach can be used. Chapter 4. The Logic of Population Control 78 6 While the two-step method controls for the eect of variables in the selection equation on the outcome equation by including the inverse Mill's ratio, the MLE approach removes the eect of the variables in the selection equation from the outcome equation altogether. 7 I present the ndings of both in the results section. A nal note in regards to the estimation technique must be made before moving on to the data. For the heckman correction to successfully remove bias three key assumptions must be met. First, the standard estimation assumptions of both the outcome and selection equation equations must not be violated. Second, the selection equation must be specied well. Third, and perhaps most dicult to meet, at least one signicant variable must aect the selection equation but have no independent signicant eect on the outcome equation. In other words, one or more variables in the selection equation must act as an instrument that aects the probability of a country experiencing forced migration in a given year but not the composition of that migration (i.e. ratio of refugees to total forced migrants). I posit that the eect of ethnic civil wars on the composition of forced migrants is conditional on the presence of one-sided violence. One-sided vi- olence during nonethnic civil wars should either have no eect on the com- position of forced migrants or should reduce the share of refugees relative to total forced migrants (i.e. increase the share of IDPs). Similarly, ethnic con ict, absent, one-sided violence, should have no eect on the composition of forced migrants. However, I also have argued that civilians respond to concerns over their personal safety by relocating to safer regions either in or out of their countries. If one-sided violence not only predicts forced migra- tion but also in uences the composition of forced migration then it cannot be used as an instrument. On the other hand, if there exists no independent 6 This analysis was conducted in 'ssampleSelection package, which uses the Newton- Raphson algorithm by default to maximize the log-likelihood function of the estimator. Alternative algorithms produced near identical results. 7 See ? for more. Chapter 4. The Logic of Population Control 79 marginal eect of one-sided violence on the composition of forced migrants then the instrument is valid. 4.5.3 Data First it must be noted that the population of cases under examination only include country-years in which at least one civil war occurs as dened by Bartusevi cius (2016) newly released dataset on Categorically Disaggregated Civil Wars (CDC). 8 Given certain data availability issues discussed below, the population of cases is restricted to between 1993-2009, which results in 365 country-year observations. Figure 4.1 is a choropleth map of the 45 states ravaged by civil war during the sample time period. Figure 4.1: Con ict States 1993-2009 The dependent variable in both the selection and outcome equations is drawn from the UNHCR's database of persons of concern. 9 This database includes complete records of the countries of origin and asylum of refugees between 1951-2013 (incomplete data for 2014-2015). It also includes data on the num- ber of internally displaced persons (and \persons in IDP-like situations") 8 https://henrikasbartusevicius.com/cdc-1-0/ 9 http://www.unhcr.org/pages/4a013eb06.html Chapter 4. The Logic of Population Control 80 between 1993-2013 (incomplete data for 2014-2015). Data for both refugees and IDPs are total stocks of migrants; however, the logic of population con- trol is concerned with active migrant ows. To obtain ows, I take the rst dierence between the stock of migrants in time t and the stock of migrants in time t 1 and then I truncate all negative values to zero. 10 For the selection equation, where I estimate the probability of forced migra- tion among civil war states, I simply I add up the total number of refugee ows, IDP ows, and ows of persons in IDP-like situations and then con- vert this total into a binary measure; 0 for states with no active forms of forced migration and 1 for states with at least some form of forced migration (active refugee and/or IDP ows). For the outcome equation I am interested in the share of migrants that are eeing refugees. To obtain this information I simply take the ratio of refugee ows to the total number of migrant ows. IfY it is the dependent variable at time t in countryi, then letR it represent the number of refugee ows at timet in countryi andI it the number of IDP ows at time t in country i, Y it = R it R it +I it (4.15) Prior to introducing the data of the regressors for each equation, a note must be made in regards to the operationalization of some of these covariates. As a check for robustness, whenever a data source is a count, such as the number of individuals displaced by natural disasters, I operationalize it in two ways, either as a natural log of the original count variable or a binary variable. To obtain binary scores, I simply choose zero as a threshold. For example, to 10 This may underestimate total migrant ows because some individuals may repatriate back to their homes, thereby reducing the total stock of forced migrants, which would reduce my measure of total migrant ows even if the total number of forced migrants eeing con ict did not decrease. In eect, this measure con ates repatriation with decreases in refugee out ows. Because my population of cases is drawn from civil war-torn states I suspect the number of forced migrants returning home is likely very minuscule. Chapter 4. The Logic of Population Control 81 operationalize the number of individuals displaced by natural disasters as a binary variable I code all country-years with at least one individual aected by natural disasters as a 1 and all remaining country-years (i.e. those with no individuals displaced) as a 0. For the logged counts, I just add one to the base and take the natural log. Table 4.1 shows the raw values for each of these variables, Table 4.2 shows their logged values, and Table 4.3 shows their binary values. Table 4.1: Raw Values of Selection Equation Variables Variable N Mean St. Dev. Min Max One-sided Viol. 361 3,184.87 37,262.76 0 501,069 Forced Migrant 365 70,796.72 248,107.60 0 2,107,111 Intercomm. Viol. 361 55.41 199.24 0 2,127 Natural Disasters 361 373,037.40 733,338.50 32 4,695,110 Table 4.2: Binary Values of Selection Equation Variables Variable N Mean St. Dev. Min Max One-sided Violence 365 0.660 0.474 0 1 Forced Migration 365 0.638 0.481 0 1 Intercommunal Violence 365 0.195 0.396 0 1 Natural Disasters 365 0.321 0.467 0 1 Table 4.3: Log Count Values of Selection Equation Variables Variable N Mean St. Dev. Min Max log(One-sided Deaths) 361 3.525 2.832 0.000 13.125 log(Intercom. Deaths) 361 0.974 2.042 0.000 7.663 log(Natural Disasters) 361 10.833 2.548 3.497 15.362 The primary independent variable of interest (i.e. the instrument) in the selection equation is one-sided violence, which I draw from UCDP (Sundberg Chapter 4. The Logic of Population Control 82 2009). One-sided violence is dened as \the use of armed force by the government of a state or by a formally organized group against civilians, which results in at least 25 deaths" in a single event (see UCDP one-sided violence codebook). I operationalize this as the natural log of the number of deaths after having added 1 to the base and as a binary variable where 0 indicates less than 25 civilians deaths and 1 more than 25 (see Table 4.2 and 4.3). I also include a set of control variables. These include the natural log of number of deaths as- sociated with intercommunal violence (and its binary operationalization, see Table 4.2 and 4.3), and the natural log of the number of individuals aected (killed and displaced) by natural disasters derived from The International Emergency Disasters Database EMDAT (and its binary operationalization, see Table 4.2 and 4.3). I also include a set of variables that need no transformation. I control for involvement in international war by including a binary indicator of interna- tional con ict drawn from the Correlates of War (COW) dataset (Sarkees and Wayman 2010). I also suspect that forced migration is a feature more common to earlier stages of con ict than later ones; thus, I control for con- ict duration operationalized as the number of years since the civil war began. I also control for civil war intensity with an indicator coded as 1 if the civil war resulted in more than 1,000 deaths in a given year and 0 if be- tween 25 to 1,000 casualties were recorded (obtained from the CDC dataset itself). Finally, I include a rst order temporal lag of the DV (forced migra- tion in time t 1). Table 4.4 shows the descriptive statics for each of these remaining variables. My primary independent variable of interest in this study, the type of civil war fought, appears in my outcome equation. Specically, I am interested in whether countries experiencing ethnic civil wars produce dierent pat- terns of forced migration than countries experiencing nonethnic civil wars. Chapter 4. The Logic of Population Control 83 Table 4.4: Untransformed Variables in Selection Equation Variable N Mean St. Dev. Min Max Interstate War 365 0.055 0.228 0 1 Con ict Duration 365 5.638 4.640 0 17 Con ict Intensity 365 0.159 0.366 0 1 I rely on Bartusevi cius (2016) newly released CDC dataset. I opt for the CDC data over the Ethnic Armed Con ict (EAC) dataset and the Armed Con ict Data to Ethnic Power Relations (ACD2EPR) dataset because the EAC and ACD2EPR denitions of ethnic con ict are more restrictive than the theory tested in this study necessitates. EAC and ACD2EPR code con- icts as ethnic if two conditions are met; one, combatants \explicitly pursue ethno-nationalist aims, motivations and interests" and, two, the combat- ants \recruit ghters and forge alliances on the basis of ethnic aliations" (Cederman, Wimmer and Min 2010). The CDC, in contrast, codes a con ict as ethnic if and only if its participat- ing groups recruit members along ethnic lines. Because we cannot observe the aims and goals of ethnic groups (at best we can take their public an- nouncements at face value) and because the logic of population control is concerned with observable markers of ethnicity and not with war aims, the CDC dataset is better suited to this study. If combatants recruit based on ethnic aliation then it makes more sense that they will also target civilians based on ethnic aliation. This is a simple binary indicator, 1 if a country- year experiences ethnic con ict and 0 if it experiences nonethnic con ict. However, some states experience both ethnic and nonethnic con icts in the same year so to manage this overlap I code any country-year with at least one ethnic con ict as a 1 (i.e. as ethnic con ict). An additional observable implication of the theory of the logic of popula- tion control is one-sided violence is an intervening variable between civil war Chapter 4. The Logic of Population Control 84 type and composition of forced migration. Therefore, an interaction of eth- nic con ict with one-sided violence should show an increase in the share of forced migrants that are refugees relative to the interplay between nonethnic con ict and one-sided violence. In other words, not only do I suspect that one-sided violence against civilians increases the likelihood of forced migra- tion (stage 1, selection equation), I also suspect it increases the share of refugees relative to IDPs but only under conditions of ethnic con ict. I op- erationalize this conditional eect as an interaction term between one-sided violence and ethnic con ict in the outcome equation. Specically, I interact the binary variable for ethnic con ict with a binary variable indicating the presence of at least one event of civilian targeting. 11 All control variables are lagged one year. In addition to the primary explanatory variables of con ict type and one- sided violence, I introduce a host of controls in the outcome equation as well, which can be classied into one of two groups { neighborhood and do- mestic factors. Neighborhood factors, on the one hand, refer to attributes of a country's region that render seeking refuge across international bor- ders more or less attractive. I expect civilians we be disinclined to seek refugee across international borders if they suspect the likelihood of being targeted is as high or higher in neighboring states as it is in his/her own country. Thus, I account for characteristics of neighboring states, which in- clude controls for whether any neighbors of a country are experiencing civil wars (CDC dataset), one-sided violence (one-sided con ict UCDP dataset), and intercommunal con ict (non-state con ict UCDP dataset, see Sundberg, Eck and Kreutz (2012)). I also control for borders. I expect island countries to produce fewer refugees on average so I control for whether a country is an island or not. I also suspect fewer borders increases the burden neigh- bors face in accommodating eeing refugees and, therefore, should decrease 11 I also test this with a continuous operationalization of one-sided violence as discussed earlier. Chapter 4. The Logic of Population Control 85 their willingness to accept large in uxes of refugees. Thus, I control for the number of international borders a country shares with its neighbors. Domestic factors, on the other hand, refer to characteristics of a country that make internal displacement more or less attractive for forced migrants. For example, I control for population density in case higher density countries make internal relocation more dicult and external migration more feasible. I also include a control for regime type using the Polity IV data (Marshall and Gurr 2014). All control variables are lagged one year. In addition to these controls, I also include temporal controls in both the selection and outcome equations. In both cases, I employ a rst order temporal lag of the dependent variable to control for any autocorrelation in the errors. Table 4.5 shows the descriptive statistics for each of these variables. The next section presents the ndings. Table 4.5: Variables for Outcome Equation Variable N Mean St. Dev. Min Max Density 361 114.32 115.30 5.38 408.38 Island 365 0.09 0.28 0 1 Neighb. Civil War 365 0.60 0.49 0 1 Neighb. Intercom Viol. 365 0.41 0.49 0 1 Neighb. One-sided Viol. 365 0.70 0.47 0 1 Number of Borders 365 4.98 2.89 0 14 Refugee:Total 37 0.49 0.49 0.00 1.00 Ethnic Civil War 365 0.65 0.48 0 1 Chapter 4. The Logic of Population Control 86 4.6 Analysis 4.6.1 Organization of Results I have estimated a number of Heckman style models using two dierent estimation techniques and two dierent operationalizations of the control variables; the results of only some of these models are presented in this sec- tion. One set of models is estimated using the traditional two-step method and the other set is estimated using MLE (Newton-Raphson algorithm). Furthermore, one set of models includes binary operationalizations of the control variables and the second uses the natural log of raw counts. What I end up with is 1) a model with dummy controls estimated using the two-step method, 2) model with dummy controls estimated using MLE, 3) a model with log counts estimated using the two-step method, and 4) another model with log counts estimated using MLE. Furthermore, each of these models contains two stages of estimation {a selection and outcome equation. I also estimate models with and without the interaction terms. Including the two separate equations for each model, a total of sixteen equa- tions are estimated. Instead of reporting the results for all sixteen models, I present the ndings for the binary response variable models only because the results are robust across the two dierent operationalizations. The re- sults for the model using log count explanatory variables can be found in Appendix C. The results are also largely robust to the two dierent esti- mation techniques; however, the ndings of the two-step method and MLE did diverge in a couple of noticeable ways. Although the literature on selec- tion models suggests MLE estimates are more consistent and robust (?), I present results for both. Table 4.6 lays out the components of each model and identies whether its results are located in the results section or the Appendix. The models shown in that table include the full two-stage model (i.e. not its selection Chapter 4. The Logic of Population Control 87 Table 4.6: Organization of Model Results Model Variable Estimation Eects Section Model 1 Binary 2step Marginal Results Model 2 Binary 2step Conditional Results Model 3 Binary MLE Marginal Results Model 4 Binary MLE Conditional Results Model 5 logCount 2step Marginal Appendix Model 6 logCount 2step Conditional Appendix Model 7 logCount MLE Marginal Appendix Model 8 logCount MLE Conditional Appendix Three-stage Binary 2step Marginal Appendix and outcome components). Counting the outcome and selection models for each model results in 19 separate equations (the three-stage model has three equations). Next I turn my attention to the results of the eight equations of interests {selection and outcome stage models with and without interaction terms estimated using MLE and 2step estimation and a set a binary control variables (the rst four models in bold found in Table 4.6). 4.6.2 Results I begin by examining the results of the marginal eects selection model (i.e. without interaction eects) in the rst stage of estimation. In this stage of estimation, I have regressed the probability of forced migration on the presence of international con ict, intercommunal con ict, one-sided violence, natural disasters, and civil war intensity. Model 1 MLE and Model 2 2step of Table 4.7 shows the selection results of the MLE and 2step estimation respectively for the model without interaction eects in its outcome stage. The positive sign and statistical signicance of the coecients of one-sided violence suggest that civilians respond to the risk of persecution by eeing, either as refugees or IDPs, which conrms H 1 . Episodes of forced migration Chapter 4. The Logic of Population Control 88 in the previous year also increase the chances of forced migration, suggesting that 1) similar factors likely persist over the course of a con ict that push civilians to ee and 2) civilians learn from the past to inform their decisions in the future (if they see others persecuted and eeing in time t 1 they will be more likely to follow suit and ee in time t). Forced migration is no more or less likely to occur earlier in con icts, while occurrence of natural disasters, intercommunal violence, international con ict, and war intensity also all fail to reach statistical signicance. It appears as though the number one factor that in uences forced migration is the deliberate and organized targeting of civilians. Now I turn to the outcome equation. and in Model 1 are both signi- cant, suggesting that selection bias is present absent a well-specied selection equation in the rst stage. In other words, Model 1's outcome equation con- rms that using a two-stage Heckman like selection model is the appropriate approach {a nding that is also conrmed by Model 2's outcome equation, whose inverse Mill's ratio is also signicant. A number of factors aect the composition of forced migrants. For example, island countries produce signicantly more IDPs as a share of total forced migrants than countries with territorial borders. More importantly, the na- ture of con ict divisions -whether a civil war is fought along ethnic lines or not -also aects the ratio of refugees to IDPs. The positive sign and statis- tical signicance of the coecients of the ethnic civil war variable suggests that states embroiled in ethnic civil war produce more refugees relative to total forced migrants than states embroiled in nonethnic civil war, which conrms H 3 . This is true of the models estimated using both techniques (MLE and 2step). H 2 posits that although one-sided violence increases the chances of forced migration (H 1 ), it has no independent marginal eect on the composition of forced migrants (i.e. the share that is composed of either IDPs or refugees). Both models fail to uncover any statistically signicant relationship between one-sided violence and the dependent variable of the Chapter 4. The Logic of Population Control 89 Table 4.7: Marginal Eects Models Model 1 MLE Model 2 2step Selection Outcome Selection Outcome Civil War Intensity 0:10 0:10 (0:19) (0:20) Forced Migration t-1 0:55 0:54 (0:15) (0:15) Interstate War 0:24 0:23 (0:29) (0:29) Con ict Duration 0:02 0:02 (0:02) (0:02) Intercommunal Violence 0:03 0:05 (0:17) (0:18) Natural Disasters 0:19 0:09 (0:15) (0:15) Civilian Targeting 0:68 0:12 0:69 0:12 (0:15) (0:07) (0:15) (0:07) Ethnic War 0:18 0:15 (0:07) (0:07) Population Density 0:00 0:00 (0:00) (0:00) Island 0:39 0:37 (0:13) (0:14) Civil War Neighborhood 0:04 0:04 (0:07) (0:08) Intcom.Viol Neighborhood 0:05 0:05 (0:06) (0:07) Civil.Targ Neighborhood 0:05 0:04 (0:08) (0:08) Borders 0:02 0:02 (0:01) (0:01) Ref:Forced Migrants t-1 0:35 0:35 (0:08) (0:08) (Intercept) 0:29 0:67 0:22 0:02 (0:16) (0:12) (0:16) (0:35) 0:39 0:47 invMillsRatio 0:88 Adj. R 2 0.17 0.14 Num. obs. 227 365 227 365 Note: p<0.05; p<0.01 Chapter 4. The Logic of Population Control 90 outcome equation, which conrms H 2 . This is an extremely important nd- ing because had such an eect existed, one-sided violence could not be used as an instrument in the selection equation, while the estimates of the out- come equation would remain biased (i.e. the selection bias would persist). Surprisingly, there is no evidence to suggest any of the remaining neighbor- hood eects in uence the share of forced migrants that are refugees. The number of international borders, the occurrence of civil war, intercommunal violence, or one-sided violence in neighboring states has no statistically sig- nicant eect on the composition of forced migrants. Although surprising, this may be accounted for by the fact civilians likely have incomplete infor- mation in regards to the likelihood of persecution abroad. Their impression of where the safest place to ee is likely formed by information they have about conditions at home and not those abroad. Therefore, we would expect civilians react to domestic factors rather than neighborhood ones when infor- mation is scarce -a typical feature of warfare. Regime type and population density also fail to reach signicance. I now turn to the results of the conditional eects model. Because the interaction eects are not introduced until the second stage, the results of the conditional eects selection equation are nearly identical to the results of the marginal eects selection equation. Here, I am only interested in the conditional eects of ethnic con ict and one-sided violence on the share of forced migrants that are refugees. To do this, I can examine the interaction between civil war type and one-sided violence. Table 4.8 shows the results of the MLE and 2step models, both of which identify a statistically signicant relationship between the share of forced migrants that are refugees and the interaction between civil war type and one-sided violence, which conrms H 4 and H 5 . Moreover, the marginal eect of ethnic civil war when one-sided violence does not occur is insignicant, which also conrms H 6 . The one dierence Chapter 4. The Logic of Population Control 91 Table 4.8: Conditional Eects Models Model 3 MLE Model 4 2step Selection Outcome Selection Outcome Civil War Intensity 0:10 0:10 (0:19) (0:20) Forced Migration t-1 0:53 0:41 (0:15) (0:15) Interstate War 0:24 0:15 (0:29) (0:30) Con ict Duration 0:02 0:02 (0:15) (0:02) Intercom. Viol. 0:03 0:05 (0:18) (0:18) Natural Disasters 0:20 0:09 (0:15) (0:15) Civilian Targeting 0:68 0:27 0:69 0:04 (0:15) (0:09) (0:15) (0:18) Civi.Targ. X Eth.War 0:26 0:25 (0:12) (0:08) Ethnic War 0:18 0:03 (0:07) (0:05) Population Density 0:00 0:00 (0:00) (0:00) Island 0:41 0:39 (0:13) (0:14) Civil War Neighborhood 0:05 0:04 (0:07) (0:14) Intcom.Viol Neighborhood 0:05 0:04 (0:06) (0:06) Civil.Targ Neighborhood 0:07 0:06 (0:08) (0:08) Borders 0:02 0:02 (0:01) (0:02) Ref:Forced Migrants 0:25 0:34 (0:05) (0:08) (Intercept) 0:29 0:79 0:22 0:15 (0:16) (0:13) (0:16) (0:35) 0:39 0:50 invMillsRatio 0:89 Adj. R 2 0.17 0.14 Num. obs. 227 365 227 365 Note: p<0.05; p<0.01 Chapter 4. The Logic of Population Control 92 that arises between the two estimation techniques, is that the marginal eect of one-sided violence on forced migrant composition is insignicant for nonethnic civil wars when estimated using the 2step method but signicant and negative when estimated using the MLE approach. The MLE results suggest one-sided violence during nonethnic civil war increases the share of IDPs rather than the share of refugees. Although not shown in the coecient tables, the marginal eect of one-sided violence on the composition of forced migrants during ethnic civil wars is negative and statistically signicant. These ndings suggest that civilians eeing ethnic civil wars have more incentives to seek refuge across borders than in \safer" regions of their home countries if the violence they are eeing is specically directed towards them. This has important implications for the spread of con ict, which I turn to in my concluding section. 4.7 Conclusion In this chapter I have introduced a theory, the logic of population control, which accounts for some of the variation in patterns of forced migration resulting from on going civil wars. The logic of population control draws on the ndings of the previous chapters as well as the extant literature on ethnic con ict and civilian targeting. In the previous chapters, I argued that coethnic refugees contribute to in- creased levels of violence because they can alter the delicate ethnic balance of power in host states with volatile ethnic relations. Because refugees are a potential source of recruitment for coethnic groups in the host state, ri- val groups, rightly or wrongly, view them as a threat. It is for this very same reason that combatants also view local civilians perceived as loyal or sympathetic to their rivals as a threat. However, whether combatants tar- get civilians loyal to their rivals largely depends on if they can determine Chapter 4. The Logic of Population Control 93 whether the civilian is a supporter or not. The literature on civilian target- ing suggests ethnic markers aid in this process of identication. If ethnic markers increase violence against civilians, then civilians eeing ethnic con ict are more constrained in their choice of destination. Simply relocating to regions away from the crossre may not be enough; combat- ants may target the regions they ee to as well. Therefore, I have hypothe- sized that states suering ethnic civil wars on average should produce more refugees relative to IDPs than states aicted with nonethnic civil war. To test this proposition I applied a two-step Heckit style selection model to a population of all country-years experiencing civil war between 1993-2010. The results of my analysis are consistent my expectations that a statistically signicant dierence exists between patterns of forced migration resulting from ethnic and nonethnic civil wars. The nding that ethnic con icts contribute to refugee ight across borders has serious implications for the study of con ict contagion more generally. Previous literature has examined the conditions under which states produce more or less refugees, but these studies, with one or two notable excep- tions, have failed to appreciate the counter factual to refugee ight, which is internal displacement. Civilians face two primary choices during con ict that can impact their chances of survival; 1) to stay put (either as a civilian or to take up arms) or to ee, and 2) whether to ee to other regions of their own countries or to seek asylum in foreign countries. The current literature on forced migra- tion either aggregates refugees and IDPs into a single category or focuses exclusively on refugee ows. Aggregating all forms of forced migration into a single category allows us to identify the factors that contribute to civilian ight, the rst choice civilians in con ict make, but it cannot speak to the second choice they face -their destination of refuge. Moreover, examining only refugee ows leads researchers to under appreciate how IDP ows can Chapter 4. The Logic of Population Control 94 also contribute to con ict contagion. For example, opposition groups in neighboring states may nd safe zones for IDPs to be useful as safe havens for their own activities. 12 Thus, examining refugees and IDPs together but as analytically distinct elements can help match dierent mechanisms of con ict contagion with dierent patterns of forced migration. It must be noted, however, that the quantitative approach used in this chapter provides evidence that is at best consistent with the theory of the \logic of population control" but cannot rule out other mechanisms at work. For example, it may be the case that ethnic con icts are more likely to occur in states that are predisposed to produce a higher refugee to IDP ratio than other states with otherwise similar proles for unknown reasons (if these reasons were identiable, quantiable, and available they would be included in the extant model). Nonetheless, the model of population control proposed in this chapter is not only consistent with the available data on refugees, IDPs, and civil war, it is a logical consequence of the ndings of a number of other studies on civilian targeting and ethnic con ict (see section 4.2.2). 12 The Kurdish PKK in Turkey has found support in the safe zones along Syria's Turkish borders where many Kurdish IDPs have amassed. For more in- formation on how the PKK has used these regions in Syria see http://www.al- monitor.com/pulse/security/2016/04/turkey-pkk-clashes-last-stronghold.html. For more on IDPs and the spread of con ict see Bohnet (2012). Chapter 5 Conclusion Key Contributions Forced migration is both a consequence and a cause of the spread of con ict across borders. States have a vested interest in controlling the populations they rule and rebels have incentives to undermine this control. Under con- ditions of ethnic con ict, population control is facilitated by ethnic markers that aid combatants in identifying loyal populations and those sympathetic to rivals. The targeting of noncombatants, civilians and refugees alike, is a natural consequence of this friction. The aim of this study has been to examine this friction under a number of dierent conditions. This study makes ve main contributions to the literature on forced mi- gration and con ict. First, it has identied a refugee-con ict nexus at the substate region in Lebanon. There exist no substate analyses of refugees and con ict outside of sub-Saharan Africa. Thus, this represents the rst test of that association in a previously unexplored region of the world. Sec- ond, it identies ethnic tensions as a conditioning factor between violence and the presence of refugees -regions with pre-existing ethnic tensions that 95 Chapter 5. Conclusion 96 host refugees are more likely to see violence than ethnically stable regions that host similar refugees. I argue this is the result of changes in the ethnic balance of power between competing groups. I also tested a second com- monly proposed mechanism linking refugees and con ict, competition over exceedingly scarce resources. Using population density as a proxy, my anal- ysis nds no support for the hypothesis that resource scarcity drives con ict between refugees and the host population or among the host population itself. Third, this study has shown that ethnically mobilized civil wars are more likely to produce greater refugees relative to IDPs than nonethnic civil wars, which has important implications for our understanding of con ict conta- gion. Consistent with numerous previous studies and supported by the ndings uncovered in chapters 1 and 2, the movement of refugees across borders has been shown to be associated with the spread of con ict from one country to another. Thus, if knowing which countries are more prone to producing refugees and why can allow us to formulate better predictive models of con ict contagion, which in turn help us make more condant pol- icy recommendations in order to more eectively coordinate the handling of refugee crises in the future. Fourth, this study has focused on how governments and rival rebel groups have incentives to control populations, particularly during times of con ict. In fact, I argue that states not only have an interest in protecting civilians loyal to their cause but to punish civilians sympathetic to their rivals. This means that states have strong incentives to allow entry to refugees that those in power share ethnic ties with. Previous literature has already established that refugees are more likely to ee to countries they share ethnic ties with, however, my research suggests not all ethnic ties are equal drivers of mi- gration. Future iterations of this project will benet from testing whether refugees are more likely to ee to countries ruled by individuals whom they share ethnic ties with than countries where they share ties with opposition Chapter 5. Conclusion 97 groups or no groups at all. In other words, are states more likely to admit refugees who share ties with those in power or, conversely, are they less likely to admit refugees who share ethnic ties with rebel groups? Such an analysis could also be extended to the realm of migration studies, if such a mechanism is true of the relationship between rulers and incoming economic migrants as well. Do rulers of countries with fragile ethnic tensions prefer to host migrants they share ethnic ties with? Similarly, are rulers less likely to admit immigrants who share ties with rival rebel groups? And are such policies eective in consolidating power, particularly when the ethnic balance of power between rival groups is relatively evenly distributed (or when those in power are part of the demographic minority, such as the Sunni in Bahrain or Alawi in Syria)? Finally, the empirical analysis has shown that refugees do not actively con- tribute to con ict onset but, instead, shape the way con ict is waged. Exist- ing studies on con ict and refugees typically highlight the role refugees play as active combatants -in the form of recruits or public advocates. However, this research has shown that the association between refugees and violence is largely conned to a particular form of violence, one-sided violence, which suggests refugees are victims of the spread of violence as much if not more than they are active contributors to it. This has a number of signicant policy implications that are discussed in the following section. Policy Implications The conventional wisdom in the literature on forced migration and con ict suggests refugees pose a danger to the host state, typically in the form of recruits and added pressure on the local population. This is most succinctly captured in the notion of the \refugee warrior". Yet analyses of disaggre- gated features of con ict suggest refugees, far from being active participants Chapter 5. Conclusion 98 in con ict, are typically its victims -even in the regions they have sought refugee. The evidence presented in this study suggest the refugee-con ict nexus is best viewed as a security issue for both the refugees and the host population (as opposed to the host population only), if it is to be most eectively addressed. Moreover, the nding that the signicant association between refugees and con ict is mediated by ethnic ties between forced migrants and their hosts, has important policy implications. These ndings may point to alternative means of organizing refugees in the host population. Currently, the UNHCR advocates for the establishment of camps to house refugees. However, if refugees are susceptible to target- ing by rebel groups, these camps render refugees as vulnerable as sitting ducks. Perhaps a more eective way to organize the distribution of refugees in a host state is to diuse their presence across dierent regions and to limit the number of refugees to regions with volatile ethnic tensions. This not only minimizes the added pressures local populations face, it also limits the ability of rival groups to target specic camps as a means of undermin- ing the support of their rivals or bolstering their own recruitment eorts. There is evidence to suggest that at least some decision-makers have come to this same conclusion. For example, leaders in Lebanon, a state who has hosted numerous Palestinian refugee camps scattered across its landscape for decades, has refrained from establishing centralized camps for Syrian refugees and, instead, has opted for a policy of general dispersion. Despite the low-intensity violence that is currently rocking Lebanon (which has been well documented in this study), the country has managed to avoid the type of high-intensity violence that has become commonplace in Iraq and Turkey in recent years. The ndings of this analysis also suggest that growing concerns over the in ux of refugees to Western states may be overblown. Figure 5.1 shows Chapter 5. Conclusion 99 Figure 5.1: Google Keyword Search Hits Google searches for the terms \Refugees", \Syrian Civil War", and \Ter- rorism" since 2014 along with the actual incidents of con ict. The y-axis in the graph represents Google's normalized search metric for each search term 1 . The x-axis represents the time range from 2014-2016, along with the locations of major terrorist incidents in the Western world during that same time period. Figure 5.1 shows that searches for refugees, the Syrian civil war, and ter- rorism co-vary to a considerable degree, suggesting popular perceptions of refugees, war, and terrorism are tied. While it is no surprise that people associate refugees with war, the fact that terrorist incidents are closely fol- lowed by commensurate changes in popular search habits, represents a more general fear of refugees held by the public. The red vertical line identies 1 According to Google Trends each data point is divided by the total searches of the geography and time range it represents. The resulting numbers are then scaled to a range of 0 to 100. See the following link for more information https://support.google.com/trends/answer/4365533?hl=en Chapter 5. Conclusion 100 the US Congress's vote on bill H.R. 158 that successfully revoked visa waiver stipulations for European dual citizens of Sudan, Iran, Iraq, and Syria, in the wake of the San Bernardino attacks. Whether politicians who passed H.R. 158 were genuine in their concern for the link between refugees and terrorism or whether they were capitalizing on fears of the public, the pop- ular perception of refugees as vectors of con ict transmission has some clear eects on public policy. Based on the ndings of this study, I argue that we can do more to limit violence associated with refugees by doing more to guarantee their security. For as long societies wage war, people will be eeing it. Appendix A Appendix: Codebook Event data on substate con ict in Lebanon was manually coded from two local news wire services -Naharanet and National News Agency. Naharnet is Lebanon's rst independent Digital Media providing real-time news in English. 1 National News Agency is Lebanon's ocial state news and in- formation service run by the Ministry of Information. 2 The data collection eorts cover the period from March 2013 to April 2015. Each observation is coded for seventeen variables. Only events that are related to political con ict or sectarian con ict are coded for the dataset. It must be noted, for the purposes of this study only certain categories of con ict are included in the sample. For example, events related to illicit be- havior (e.g. drug tracking and human smuggling undertaken by militants) that do not involve violence are dropped from the sample. Also dropped are all events related to cross-border con ict with other countries, such as Syr- ian and Israeli military raids into Lebanese territory as well as cross-border shelling of Lebanese towns and villages from the Syrian side of the border. 1 http://www.naharnet.com 2 http://nna-leb.gov.lb/en 101 Appendix A Codebook - Chapter 2 102 Actors, Denitions, and Variables The variable name used in the dataset is found in brackets. 1. The date the event occurred [date] Description: Day, Month, and Year in which the event took place. Notes: If the event took place over more than one day. Each day was coded as a separate event. 2. The actor's identity [actor] Notes: The dierence between actor and target is that actor refers to the participant who instigates the con ict event. Cases in which the identity of the initial instigator is unknown then the rst participant named in the primary sourced is coded as the actor. Description: Actor identity refers to the specic identity of an actor involved. For example, if a Lebanese civilian attacks a Syrian refugee then the actor identity is coded as a Lebanese civilian. If Lebanese police arrest a militant then the actor's identity is Lebanese Internal Security Force. Values: The following are a list of all coded actors: { lebauth: Unspecied Lebanese authority { lebmilint: Lebanese Military Intelligence { lebarmy: Lebanese Armed Forces { lebisf: Lebanese Internal Security Forces { lebmun: Lebanese Municipal Authorities (includes local po- lice, mayor's oce, etc.) { palisf: Palestinian Internal Security Forces Appendix A Codebook - Chapter 2 103 { isrlmil: Israeli Military { syrmil: Syrian Military { syrcg: Syrian Coast Guard { lebcivil: Lebanese civilian { isrlcivil: Israeli civilian { syrref: Syrian refugee or Syrian national not identied specif- ically as a rebel or with a rebel group { palref: Palestinian refugee or Palestinian national not iden- tied specically as a rebel or with a rebel group { hezb: Hezbollah militant group (shia) { amal: Amal militant group (shia) { rebpal: Palestinian rebel groups such as Popular Front Gen- eral Command, Fatah, Hamas, etc. { rebfsa: Rebel group Free Syrian Army (sunni) { rebisis: Rebel group Islamic State , includes oshoots such as Abdullah Azzam Brigades (sunni) { rebanf: Rebel group Al Nusra Front (sunni) { rebskh: Sheikh Khaled Hablas rebel group (sunni) { rebbsaa: Sheikh Ahmad Asir rebel group (sunni) { rebunid: Unidentied rebel group { unil: United Nations Interim Force in Lebanon { unrwa: United Nations Relief and Works Agency { other: None of the above (i.e. national not from Lebanon or Syria) 3. The actor's group [actor group] Description: Actor Group refers to the religious or institutional background of an actor. For example, a Sunni Lebanese civilian Appendix A Codebook - Chapter 2 104 would be coded as a Sunni as would a Sunni militant. The insti- tution of an actor is coded only if the actor is acting on behalf of a government. For example, a Lebanese Army force and a Lebanese Police would both be coded as Lebanese government. Values: The following are a list of all coded groups: { sunni: Sunni Muslim { shia: Shia Muslim { druze: Druze { alawi: Alawi { jewish: Jewish { christian: Maronite, Catholic, Protestant, Armenian Catholic, Greek Orthodox { palgov: Palestenian Authority { syrgov: Syrian Government { lebgov: Lebanese Government { isrlgov: Israeli Government { intlorg: International Organization (such as UNRWA or UNHCR) { other: None of the above 4. The target's identity [target] Description: The dierence between actor and target is that actor refers to the participant who instigates the con ict event. Cases in which the identity of the initial instigator is unknown then the rst participant named in the primary sourced is coded as the actor. Values: See actor identity 5. The target's group [target group] Appendix A Codebook - Chapter 2 105 Description: The dierence between actor and target is that actor refers to the participant who instigates the con ict event. Cases in which the identity of the initial instigator is unknown then the rst participant named in the primary sourced is coded as the actor. Values: See actor group 6. The number killed [dead] Description: The number of individuals, civilians and non-participants included, who were killed during the course of the event or as a result of it) The coding for number killed is not always perfectly re ected in the source URL for two reasons 1) the numbers are updated as new news articles are reported on an event (even though those other URLs are not coded in the source column) and 2) because some events were coded using additional URLs that are not re- ported, 7. The number injured [injury] Description: The number of individuals, civilians and non-participants included, who were injured during the course of the event. 8. The number arrested [arrest] Description: The number of individuals who who were arrested during the course of the event. Arrests only refer to arrests maid by Lebanese government. 9. The category of the event [category] Appendix A Codebook - Chapter 2 106 Description: This variable categorizes each event according to the type of con ict event, ranging from clashes involving numer- ous people on both sides to assassinations, bombings, and even hostage taking. Values: Many dierent categorizations { exect: Event involves execution outside of battles unrelated to basic shooting (such as execution of hostages) { arrestraid: Event involves raid or arrest by military or in- ternal security forces { bombard: Event involves bombarment with weapons such as surface-to-air grade missiles outside the context of a clash (dened below) { assault: Event involves use of violence b/w 1-3 individu- als (not including `shooting'), involves weapons other than projectiles and guns. { shooting: Event involves shooting arising from dispute b/w 1- 3 individuals (such as individual murder) { protest: Event involves protest or other collective actions such as solidarity marches and strikes but also includes acts such as rioting and forced closures of streets and shops { abduction: Event involves use of kidnapping, abduction, ransom, or hostage-taking { clashes: Event involves clashes bw armed groups (4+ indi- viduals involved in total) with guns, artillery, grenades, rock- ets, missiles, and/or mortars that wholly takes place within Lebanon (includes siege) { suicbatt: Event involves suicide bombing { bombatt: Event involves explosive device besides suicide attack within Lebanon (IED, bomb, intentional mine, not Appendix A Codebook - Chapter 2 107 hand grenades unless placed as a detonated bomb -so not used with `hands' as a projectile weapon such as a missile launcher, etc.). "Non-events" are not coded. For example, if a bomb is found that has not gone o, that is not coded. is NOT coded. Only events that transpire are coded { curfew: Event involves imposition of curfew on refugees by authorities at any level of government { crossclash: Event involves land re (gunre, mortars, mis- siles but not bombs from planes) from across Syrian or Israeli border, or it involves territorial violation by Israeli or Syrian governments { aerbom: Event involves aerial bombing (cross-border in- cluded) { airspace: Event involves nonviolent illegal entry of Lebanese airspace (Syria and Israel are the usual actors) { maritime: Event involves nonviolent illegal entry of Lebanese territorial waters { territ: Event involves nonviolent illegal entry of Lebanese land territory 10. The classication of the event [classication] Description: This variable classies events according to the na- ture of the underlying tension between each actor-target. Values: Seven classications { perscon: Event associated with con ict over individual or peronal act of violence/illicit activity (i.e. non { political/economic/social such as family dispute) { relcon: Event associated with con ict bw religious groups and/or social groups (includes arrest of Syrian refugees for lack of refugee papers) Appendix A Codebook - Chapter 2 108 { polcon: Event associated with con ict over policy between competing political factions or citizens and government (all incidents of terrorism or related to terrorist groups coded as polcon including arrests for possession of military style weapons) { illcon: Event associated with con ict over illicit group activ- ities associated with militants (such tracking, drugs, etc.) { syrcrosscon: Event associated with cross border con ict from Syria (includes illegal entry into the country by refugees, ghters, etc.). If a single cross-border assault (such as aerial bombing or mortar/gunre) hits multiple villages within Lebanon, each location that is struck is coded as a separate event { isrcrosscon: Event associated with cross border con ict from Israel { econcon: Event associated with con ict over economic is- sues (for example protest over food prices, these are very rare in the data simply because news reports rarely report any information that a coder can infer to suggest the event involves economic related issues/ demands/tensions). 11. The town the event took place [town] Notes: Closest town to the event. 12. The region the event took place [region] Description: This refers to informal regional names. 13. The district the event took place [district] 14. The governate the event took place [governate] 15. The latitude of the event [latitude] Notes: Closest geo-coded point of reference to the event Appendix A Codebook - Chapter 2 109 16. The longitude of the event [longitude] Notes: Closest geo-coded point of reference to the event 17. The URL of the primary sourced used to code the event [url] Notes: When multiple sources identify the same event only one source is listed. General Notes 1. If a single cross-border assault (such as aerial bombing or mortar/gun- re) hits multiple villages within Lebanon, each location that is struck is coded as a separate event 2. Arrests only refer to arrests maid by the Lebanese government (and not for example by Syrian military forces who enter into Lebanon to detain militants. That would be coded as a 'syrcrosscon' (illegal Syrian entry of Lebanese territory). 3. Only the primary two agents are coded. For example if a clash breaks out between Sunni and Shia groups and as a direct result of those clashes the army arrests one or more of the actors/targets involved then only the two actors/targets that were initially involved in event are coded even though the event includes an arrest by an army ocer. This is only the case if the arrest happens immediately within the time frame of the event (same day for example, anything past a day gets coded as a separate event) 4. Events cannot last more than one day -for events related to one another in a larger battle, each day is coded as an independent event. 5. If an event involves action by more than one participant (i.e. a reac- tion) the event is coded according to the attributes of the 1st action. Appendix A Codebook - Chapter 2 110 For example, if an event involves an abduction or hostage taking by actor A and the victim is target A but the news wire also states that the hostage was freed on the same day after the army arrested actor A then only the abduction as coded for the variable `category' (and not the subsequent arrest). Similarly, the actor is coded as the rst actor and the target is coded as that rst actor's target (even though eventually our actor A in the example above moves from being an actor to a target when he is arrested by the army). Keep in mind, however, because the event ultimately resulted in the arrest of actor A the variable `arrest' is still coded as 1 (or if he was killed in the raid then 1 for dead). 6. `Non-incidents' are not included such as `Lebanese Amry Intelligence discovered an detonated bomb in a parked car' 7. Explosions caused by land mines from previous con ict are not in- cluded (unfortunately more common than one would think) 8. If two dierent event classications occur at once (for example clashes occur between two families and security forces get involved and arrest those involved then the very rst action to perpetuate the con ict is coded. In this case if arrests were made immediately at the time of the clashes the event would be coded as `clashes' and not `arrestraid' despite the fact that an arrest was made and the variable for arrest would still be coded as 1. Conversely if an arrest was being made and clashes erupted or shooting erupted immediately (within the frame of action such as during the time the arrest was being made or im- mediately following it within a couple hour block) as a result of the attempted or successful arrest then the event would be coded as an `arrestraid' and not as `clashes' or `shooting'. 9. It is not always clear who the actor and who the target are. Usually the actor is the agent who initiates the confrontational event but if Appendix A Codebook - Chapter 2 111 the news report does not include that info or if no clear initiator even existed then the default is to always use civilians as the target. In cases where civilians are not one of the agents then it reverts to the government. For example, if an an unidentied rebel group (rebunid) is in a clash with Lebanese army (lebarmy) and its unclear who the target and who the actor is then the actor by default gets set as the government. In cases in which neither civilians nor Lebanese govern- ment actors are involved then the group that sustains most injury or death gets coded as the actor and if its still unclear then the rst group in the news wire to be mentioned is coded as the actor. 10. Incidental victims do not get coded as targets though if they are in- jured or killed a death and/or injury will be coded. For example, if a clash between the army and some unidentied rebel group takes place and a Palestinian refugee is killed in the cross re then the target and actors would be either Lebanese army or the rebel group (depending on who initiated) but it would not be the Palestinian refugee. 11. Only drug tracking arrests are reported, so no events involving drug possession or drug transport are coded unless violence is involved in which case it is reported even it is just possession or transport 12. Only violent events are (with the exception of nonviolent arrests of militants and nonviolent illegal entry of militants into the country) that are related to political con ict are included Appendix B Appendix: Data Visualization Figure B.1: Refugee Flows and Violence 2013-2015 0 50 100 150 Jun 13 Sep 13 Dec 13 Mar 14 Jun 14 Sep 14 Dec 14 Mar 15 Jun 15 Refugees in Thousands Violence Refugees 112 Appendix B. Data Visualization - Chapter 2 113 Figure B.2: Violence Severity 2013-2015 10 20 30 40 Jun 13 Sep 13 Dec 13 Mar 14 Jun 14 Sep 14 Dec 14 Mar 15 Jun 15 Violent Events 0 25 50 75 Jun 13 Sep 13 Dec 13 Mar 14 Jun 14 Sep 14 Dec 14 Mar 15 Jun 15 Deaths 0 200 400 600 800 Jun 13 Sep 13 Dec 13 Mar 14 Jun 14 Sep 14 Dec 14 Mar 15 Jun 15 Injuries 0 100 200 300 Jun 13 Sep 13 Dec 13 Mar 14 Jun 14 Sep 14 Dec 14 Mar 15 Jun 15 Arrests Appendix B. Data Visualization - Chapter 2 114 Figure B.3: Violence by District 2013-2015 0 50 100 150 Baalbek Akkar Baabda Beirut El Hermel El Meten Aley Chouf El Batroun El Koura Violent Events 0 100 200 Baalbek Baabda Akkar El Hermel Beirut Aley Chouf El Meten El Batroun El Koura Deaths 0 200 400 600 Baabda Baalbek Akkar El Hermel Beirut Aley Chouf El Batroun Bent Jbeil El Koura Injuries 0 200 400 600 Baalbek Akkar Beirut Baabda El Meten El Koura Chouf El Hermel Aley El Batroun Arrests Appendix B. Data Visualization - Chapter 2 115 Figure B.4: Syrian Refugees as a Share of Population 2013-2015 Apr−2013 May−2013 Jun−2013 Jul−2013 Aug−2013 Sep−2013 Oct−2013 Nov−2013 Dec−2013 Jan−2014 Feb−2014 Mar−2014 Apr−2014 May−2014 Jun−2014 Jul−2014 Aug−2014 Sep−2014 Oct−2014 Nov−2014 Dec−2014 Jan−2015 Feb−2015 Mar−2015 Apr−2015 Refugees/Population 0.250.500.751.001.25 Appendix B. Data Visualization - Chapter 2 116 Figure B.5: Ethnoreligious Tapestry of Lebanon ## OGR data source with driver: ESRI Shapefile ## Source: "/Users/cyrusmohammadian/Desktop/diss.data/LBN_adm-2/", layer: "LBN_adm2" ## with 26 features ## It has 18 fields Greek.Orthodox Greek.Catholic Maronite Armenian.Orthodox Armenian.Catholic Christian.total Alawaite Shia Shia.Alawi.total Sunni Druze 0 25 50 75 Group % Appendix B. Data Visualization - Chapter 2 117 Figure B.6: Ethnoreligious Groups by District 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 25 50 75 0/100 Akkar Aley Baabda Baalbek Batroun Bcharr.. Beirut Bekaa−West Bent Jbayl Chouf Hasbaya Hermel Jbayl Jezzine Kesrouan Koura Marjayoun Matn Minie−Danniyeh Nabatiyeh Rachaya Saida Sour Tripoli Zahl.. Zgharta Sect Armenian Catholic Armenian Orthodox Greek Orthodox Greek Catholic Maronite Druze Shia Sunni Alawi Appendix C Appendix: Log Count Results Table C.1: Organization of Model Results Model Variable Estimation Eects Section Model 1 Binary 2step Marginal Results Model 2 Binary 2step Conditional Results Model 3 Binary MLE Marginal Results Model 4 Binary MLE Conditional Results Model 5 logCount 2step Marginal Appendix Model 6 logCount 2step Conditional Appendix Model 7 logCount MLE Marginal Appendix Model 8 logCount MLE Conditional Appendix Three-stage Binary 2step Marginal Appendix 118 Appendix C Log Count Results - Chapter 4 119 Table C.2: Marginal Eects Models with Log Counts Model 5 MLE Model 6 2step Selection Outcome Selection Outcome Civil War Intensity 0:24 0:11 (0:20) (0:21) Forced Migration t-1 0:50 0:35 (0:15) (0:15) Interstate War 0:19 0:23 (0:29) (0:32) Con ict Duration 0:02 0:02 (0:01) (0:02) log(Intercom Death) 0:00 0:01 (0:03) (0:04) log(Nat.Dis. Victim) 0:03 0:04 (0:04) (0:03) log(Civilian Deaths) 0:11 0:11 (0:03) (0:03) Ethnic War 0:17 0:17 (0:06) (0:06) Population Density 0:00 0:00 (0:00) (0:00) Island 0:34 0:39 (0:13) (0:13) Civil War Neighborhood 0:05 0:05 (0:07) (0:07) Intcom.Viol Neighborhood 0:07 0:05 (0:06) (0:07) Civil.Targ Neighborhood 0:04 0:04 (0:08) (0:08) Borders 0:01 0:02 (0:01) (0:01) Ref:Forced Migrants t-1 0:25 0:30 (0:05) (0:08) 0:41 0:60 invMillsRatio 0:50 Adj. R 2 0.14 0.13 Num. obs. 227 365 227 365 Note: p<0.05; p<0.01 Appendix C Log Count Results - Chapter 4 120 Table C.3: Conditional Eects Models with Log Counts Model 7 MLE Model 8 2step Selection Outcome Selection Outcome Civil War Intensity 0:19 0:11 (0:21) (0:21) Forced Migration t-1 0:45 0:35 (0:15) (0:15) Interstate War 0:21 0:23 (0:31) (0:32) Con ict Duration 0:02 0:02 (0:02) (0:02) log(Intercom Death) 0:00 0:01 (0:04) (0:04) log(Nat.Dis. Victim) 0:03 0:04 (0:04) (0:03) log(Civilian Deaths) 0:10 0:05 0:11 0:05 (0:03) (0:02) (0:03) (0:02) Civi.Death X Eth.War 0:18 0:19 (0:08) (0:07) Ethnic War 0:03 0:17 (0:10) (0:06) Population Density 0:00 0:00 (0:00) (0:00) Island 0:41 0:41 (0:13) (0:13) Civil War Neighborhood 0:05 0:05 (0:07) (0:07) Intcom.Viol Neighborhood 0:04 0:05 (0:06) (0:06) Civil.Targ Neighborhood 0:06 0:04 (0:08) (0:08) Borders 0:02 0:02 (0:01) (0:01) Ref:Forced Migrants t-1 0:04 0:30 (0:01) (0:08) 0:37 0:40 invMillsRatio 0:17 Adj. R 2 0.14 0.15 Num. obs. 227 365 227 365 Note: p<0.05; p<0.01 Appendix C Log Count Results - Chapter 4 121 Table C.4: Three-Stage (1) Heckit Selection Stage 1 (Intercept) 0:75 (0:70) log(Population) 0:16 (0:07) Polity 0:00 (0:02) Polity 2 0:00 (0:00) Regime Stability 0:20 (0:23) Ethnic Fractionalization 0:38 (0:37) log(Mountainous Terrain) 0:03 (0:07) Infant Mortality 0:33 (0:14) Civil War Neighborhood 0:13 (0:08) Peaceyears 2:57 (0:24) Peaceyears2 0:57 (0:07) Peaceyears3 0:04 (0:01) Slow Growth 0:18 (0:19) AIC 266.86 Log Likelihood -120.43 Num. obs. 2,720 p< 0:01, p< 0:05 Appendix C Log Count Results - Chapter 4 122 Table C.5: Three-Stage (2) Heckit Selection Stage 2 Outcome Civil War Intensity 0:10 (0:20) Forced Migration t-1 0:41 (0:15) Interstate War 0:15 (0:30) Con ict Duration 0:03 (0:02) log(Intercommunal Deaths) 0:05 (0:18) log(Natural Disaster Victims) 0:09 (0:15) log(Civilian Deaths) 0:70 (0:03) Ethnic War 0:17 (0:06) Population Density 0:00 (0:00) Island 0:40 (0:13) Civil War Neighborhood 0:03 (0:07) Intcom.Viol Neighborhood 0:04 (0:06) Civil.Targ Neighborhood 0:05 (0:07) Borders 0:02 (0:01) Ref:Forced Migrants t-1 0:30 (0:06) invMillsRatio 0:03 0:56 Adj. R 2 0.14 Num. obs. 365 227 Note: p<0.05; p<0.01 Bibliography Adelman, Howard. 1998. \Why refugee warriors are threats." Journal of Con ict Studies 18(1). Bartusevi cius, Henrikas. 2016. \Introducing the Categorically Disaggre- gated Con ict (CDC) dataset." Con ict Management and Peace Science 33(1):89{110. Bhavnani, Ravi and Dan Miodownik. 2009. \Ethnic polarization, Ethnic Salience, and Civil War." Journal of Con ict Resolution 53(1):30{49. Bohnet, Heidrun. 2012. Large Refugee Populations, Resource Scarcity and Con ict. In Annual Conference of the European Political Science Associ- ation. p. 21{23. Bohnet, Heidrun, Fabien Cottier and Simon Hug. 2013. \Con ict-induced IDPs and the Spread of Con ict." Presentation at the European Political Science Association . Boserup, Ester. 1990. \Economic and Demographic Relationships in Devel- opment." Population and Development Review 16(4):775{779. Buhaug, Halvard. 2010. \Dude, where's my con ict? LSG, relative strength, and the location of civil war." Con ict Management and Peace Science . Carter, David B and Curtis S Signorino. 2010. \Back to the future: Modeling time dependence in binary data." Political Analysis 18(3):271{292. 123 Bibliography 124 Caselli, Francesco and Wilbur John Coleman. 2013. \On the theory of ethnic con ict." Journal of the European Economic Association 11(s1):161{192. Cederman, Lars-Erik, Andreas Wimmer and Brian Min. 2010. \Why Do Ethnic Groups Rebel? New Data and Analysis." World Politics 62(01):87{119. Chi, Guangqing and Jun Zhu. 2008. \Spatial Regression Models for Demo- graphic Analysis." Population Research and Policy Review 27(1):17{42. Davenport, Christina, Will Moore and Steven Poe. 2003. \Sometimes you just have to leave: Domestic threats and forced migration, 1964-1989." International Interactions 29(1):27{55. Del Carpio, Ximena V and Mathis C Wagner. 2015. \The impact of Syri- ans refugees on the Turkish labor market." World Bank Policy Research Working Paper (7402). Downes, Alexander B. 2006. \Desperate Times, Desperate Measures: The Causes of Civilian Victimization in War." International Security 30(4):152{195. Drabo, Alassane and Lingu ere Mbaye. 2011. \Climate change, Natural dis- asters and Migration: An Empirical Analysis in Developing Countries.". Eck, Kristine. 2009. \From Armed Con ict to War: Ethnic Mobilization and Con ict Intensication*." International Studies Quarterly 53(2):369{388. Eck, Kristine and Lisa Hultman. 2007. \One-Sided Violence Against Civil- ians in War Insights from New Fatality Data." Journal of Peace Research 44(2):233{246. Ehrlich, Paul, R. 1968. \The Population Bomb." Ballantine, New York . Engelhardt, Michael J. 1992. \Democracies, Dictatorships and Counterin- surgency: Does Regime Type Really Matter?" Journal of Con ict Studies 12(3). Bibliography 125 Esteban, Joan and Debraj Ray. 2008. \Polarization, fractionalization and con ict." Journal of peace Research 45(2):163{182. Fearon, James D. 1995. \Rationalist explanations for war." International organization 49(03):379{414. Fearon, James D. 1998. \Commitment Problems and the Spread of Ethnic Con ict. The International Spread of Ethnic Con ict." Princeton, NJ, Princeton University Press 52:269{305. Fearon, James D. 2004. \Ethnic Mobilization and Ethnic Violence." Barry R. Weingast and . Fearon, James D and David D Laitin. 2003. \Ethnicity, Insurgency, and Civil war." American Political Science Review 97(01):75{90. Ferris, Elizabeth G. 1993. Beyond Borders: Refugees, Migrants and Human Rights in the Post-Cold War Era. WCC publications Geneva. Fisk, Kerstin. 2014. \Refugee Geography and the Diusion of Armed Con- ict in Africa." Civil Wars 16(3):255{275. Fjelde, Hanne and Lisa Hultman. 2010. Weakening Your Enemy: Con- stituencies and the Location of Violence against Civilians in Africa, 1989- 2006. In SGIR 7th Pan-European International Relations Conference. Foged, Mette and Giovanni Peri. 2015. \Immigrants' Eect on Native Work- ers: New Analysis on Longitudinal Data.". Forsberg, Erika. 2009. \Refugees and Intrastate Armed Con ict: A Conta- gion Process Approach.". Franzese, Robert J, Jude C Hays and Lena M Schaer. 2010. Spatial, Tem- poral, and Spatiotemporal Autoregressive Probit Models of Binary Out- comes: Estimation, Interpretation, and Presentation. In Annual Confer- ence of the American Political Science Association. Bibliography 126 Garcia-Montalvo, Jose and Marta Reynal-Querol. 2004. \Ethnic Polariza- tion, Potential Con ict, and Wivil Wars." Potential Con ict, and Civil Wars . Gleditsch, Nils Petter et al. 2002. \Armed con ict 1946-2001: A new dataset." Journal of Peace Research 39(5):615{637. Hall, Bronwyn H. 2002. \Notes on Sample Selection Models.". URL: http://eml.berkeley.edu// bhhall/e244/sampsel.pdf Har, Barbara. 2003. \No Lessons Learned from the Holocaust? Assess- ing Risks of Genocide and Political Mass Murder Since 1955." American Political Science Review 97(01):57{73. Hassner, Ron Eduard. 2009. War on Sacred Grounds. Cornell University Press. Heckman, James J. 1979. \Sample Selection Bias as a Specication Error." Econometrica 47(1):153{161. Hegre, H avard, Gudrun stby and Clionadh Raleigh. 2009. \Poverty and Civil War Events: A Disaggregated Study of Liberia." Journal of Con ict Resolution 53(4):598{623. Homer-Dixon, Thomas and Jessica Blitt. 1998. Ecoviolence: Links among Environment, Population, and Security. Rowman & Littleeld Publishers. Kalyvas, Stathis. 2001. \New and old civil wars." World politics 54(1):99{ 118. Kalyvas, Stathis N. 2005. \Warfare in civil wars." Rethinking the Nature of War pp. 88{108. Kaufmann, Chaim. 1996. \Possible and Impossible Solutions to Ethnic Civil Wars." International Security 20(4):136{175. Bibliography 127 Lischer, Sarah Kenyon. 2006. Dangerous Sanctuaries: Refugee Camps, Civil war, and the Dilemmas of Humanitarian Aid. Cornell University Press. Loescher, Gil and James Milner. 2005. \Security Implications of Protracted Refugee Situations." The Adelphi Papers 45(375):23{34. Malthus, Thomas Robert. 1798. An Essay on the Principle of Population, as it aects the Future Improvement of Society: with remarks on the Spec- ulations of Mr. Godwin, M. Condorset, and other writers. J. Johnson. Mar eet, Philip. 2007. \Refugees and history: why we must address the past." Refugee Survey Quarterly 26(3):136{148. Marshall, Monty G. and Robert T. Gurr. 2014. \Polity IV Project: Political Regime Characteristics and Transitions, 1800-2013.". URL: http://www.systemicpeace.org/polity/polity4.htm Maxwell, John W and Rafael Reuveny. 2000. \Resource Scarcity and Con- ict in Developing Countries." Journal of Peace Research 37(3):301{322. Moore, Will H and Stephen M Shellman. 2004. \Fear of Persecution Forced Migration, 1952-1995." Journal of Con ict Resolution 48(5):723{745. Moore, Will H and Stephen M Shellman. 2007. \Whither Will They Go? A Global Study of Refugees' Destinations, 1965-1995." International Studies Quarterly 51(4):811{834. Mueller, John. 2000. \The banality of \ethnic war"." International Security 25(1):42{70. Neumayer, Eric. 2005. \Asylum Recognition Rates in Western Europe Their Determinants, Variation, and Lack of Convergence." Journal of Con ict Resolution 49(1):43{66. Osborn, Faireld. 1948. \Our Plundered Planet.". Bibliography 128 Ousey, Graham C, Pamela Wilcox and Bonnie S Fisher. 2011. \Some- thing old, Something New: Revisiting Competing Hypotheses of the Victimization-oending Relationship among Adolescents." Journal of Quantitative Criminology 27(1):53{84. Phillipson, Coleman. 1911. The international law and custom of ancient Greece and Rome. Vol. 1 Macmillan and Company, limited. Posen, Barry R. 1993. \The Security Dilemma and Ethnic Con ict." Sur- vival 35(1):27{47. Rabben, Linda. 2011. Give refuge to the stranger: the past, present, and future of sanctuary. Left Coast Press. Raleigh, Clionadh and Henrik Urdal. 2007. \Climate Change, Environmen- tal Degradation and Armed Con ict." Political Geography 26(6):674{694. Reyna, Stephen P. 2010. \The Disasters of War in Darfur, 1950{2004." Third World Quarterly 31(8):1297{1320. Reynal-Querol, Marta. 2002. \Ethnicity, Political Systems, and Civil Wars." Journal of Con ict Resolution 46(1):29{54. R uegger, Seraina and Heidrun Bohnet. 2015. \The Ethnicity of Refugees (ER): A new dataset for understanding ight patterns." Con ict Manage- ment and Peace Science . Salehyan, Idean. 2008. \The externalities of civil strife: Refugees as a source of international con ict." American Journal of Political Science 52(4):787{801. Salehyan, Idean and Kristian Skrede Gleditsch. 2006. \Refugees and the spread of civil war." International Organization 60(02):335{366. Sambanis, Nicholas. 2001. \Do Ethnic and Nonethnic Civil Wars Have the Same Causes? A Theoretical and Empirical Inquiry (Part 1)." Journal of Con ict Resolution 45(3):259{282. Bibliography 129 Sarkees, Meredith Reid and Frank Wayman. 2010. \Resort to War: 1816 - 2007." Con ict Management and Peace Science . Schechtman, Joseph B. 1964. The Refugees in the World: Displacement and Integration. Barnes. Schmeidl, Susanne. 2001. \Con ict and forced migration: a quantitative review, 1964-1995." Global migrants, global refugees: Problems and solu- tions pp. 62{94. Shaver, Andrew and Yang-Yang Zhou. 2015. \Questioning Refugee Camps as Sources of Con ict.". Sundberg, Ralph. 2009. Revisiting One-sided Violence{A Global and Re- gional Analysis. In States in Armed Con ict, ed. Lotta Harbom and Ralph Sundberg. Department of Peace and Con ict Research. Sundberg, Ralph, Kristine Eck and Joakim Kreutz. 2012. \Introducing the UCDP Non-state Con ict Dataset." Journal of Peace Research 49(2):351{ 362. Theisen, Ole Magnus. 2008. \Blood and Soil? Resource Scarcity and Internal Armed Con ict Revisited." Journal of Peace Research 45(6):801{818. Thompson, W.S. 1929. \Population." American Journal of Sociology. (34). Tilly, Charles. 1975. \The Formation of National States in Western Europe." Princeton UniversityPress . Urdal, Henrik. 2005. \People vs. Malthus: Population Pressure, Environ- mental Degradation, and Armed Con ict Revisited." Journal of Peace Research 42(4):417{434. Valentino, Benjamin, Paul Huth and Dylan Balch-Lindsay. 2004. \\Draining the sea": Mass killing and guerrilla warfare." International Organization 58(02):375{407. Bibliography 130 Vernant, Jacques. 1953. The Refugee in the Post-War World. Yale Univer- sity Press. Vogt, Manuel et al. 2015. \Integrating Data on Ethnicity, Geography, and Con ict The Ethnic Power Relations Data Set Family." Journal of Con- ict Resolution . Vogt, William. 1948. Road to Survival. New York: William Sloane Asso- ciates, Inc. Weidmann, Nils B and Michael D Ward. 2010. \Predicting Con ict in Space and Time." Journal of Con ict Resolution 54(6):883{901. Wimmer, Andreas, Lars-Erik Cederman and Brian Min. 2009. \Ethnic pol- itics and armed con ict: a congurational analysis of a new global data set." American Sociological Review 74(2):316{337. Wooldridge, Jerey M. 2010. Econometric Analysis of Cross Section and Panel Data. MIT press. Zolberg, Aristide R, Astri Suhrke and Sergio Aguayo. 1992. Escape from violence: Con ict and the refugee crisis in the developing world. Oxford University Press on Demand.

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Perspectives on state capacity and the political geography of conflict

PDF

The social making of authoritarian environmentalism: protest-litigation nexus and policy changes in China

PDF

Emergency-war machine: national crisis, democratic governance, and the historical construction of the American state

PDF

The business of nationalism: how commodification sustains bilateral tensions

PDF

A theory of status-quo terrorism: democracies in conflict and their proclivity to outsource repression

PDF

When passions run high: emotions and the communication of intentions in face-to-face diplomacy

PDF

Thawing rivalries and fading friendships: a multi-method approach to rapprochement and alienation

PDF

When bullets are not enough: determinants of chemical weapons usage and abstention

PDF

Agents of war: Cambodian refugees and the containment of radical opposition

PDF

Forced march to modernity: State-imposed cultural change and regime stability in 20th-century east Asia

PDF

Towards an inter-disciplinary approach to the study of transnational relations

PDF

Shaking the foundations of violent civil conflict: institutions, disasters, and the political economies of state-rebel interaction

PDF

Powellism, Thatcherism and the Conservative Party, 1945-87: the party as the site of ideological transition

PDF

Refugee protection and empowerment joint venture: understanding how and why public and private actors cooperate to tackle the global refugee crisis

PDF

Are you as old as you feel? A quantitative and qualitative analysis of leader age and foreign policy decision-making

PDF

Developing countries in the digital era: state and business interactions for industrial upgrading

PDF

The partial Good Samaritan states: China and Japan in the international relations of autocracy and democracy

PDF

Perceiving and coping with threat: explaining East Asian perceptions toward China’s rise

PDF

Voices unheard, stories untold: Black women, police violence and political participation

PDF

The paradox of personalism: internal political constraints on nuclear posture in authoritarian regimes

Asset Metadata

Creator Mohammadian, Cyrus (author)

Core Title Forced migration as a cause and consequence of conflict

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Political Science and International Relations

Publication Date 07/27/2016

Defense Date 06/13/2016

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Civil War,conflict,Ethnicity,forced migration,IDP,Lebanon,OAI-PMH Harvest,Refugees

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor James, Patrick (committee chair), Sellers, Jefferey M. (committee member), Wilson, John P. (committee member)

Creator Email cmohamma@gmail.com,cmohamma@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-283917

Unique identifier UC11280116

Identifier etd-Mohammadia-4648.pdf (filename),usctheses-c40-283917 (legacy record id)

Legacy Identifier etd-Mohammadia-4648.pdf

Dmrecord 283917

Document Type Dissertation

Format application/pdf (imt)

Rights Mohammadian, Cyrus

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA