No state border is frozen : the effect of transnational ethnic groups and neighbouring state interventions on civil outcome

Master’s Program Political Science: International Relations

No State Border is Frozen: The effect of Transnational Ethnic Groups and

Neighbouring State Interventions on Civil War Outcome

Master’s Thesis

Zachary Torok 12271381 Thesis Supervisor: Dr. Mike Medeiros Second Reader: Dr. Ursula Daxecker

Acknowledgements

Thank you to everyone who helped me get to where I am now. To my Mom and Dad for supporting me and giving me all I needed. To my sister Debra for not getting sick of all the questions I asked her about marginal effects. To my friends back home and my friends here for keeping me sane throughout this process. Special shout out to all the Civil Conflict students. To the Toronto Raptors, who made this submission month a lot more bearable. And finally, to my Zadie, who told everyone to use their voice and never stop singing. He never did and neither will I.

List of Abbreviations ... 2

Abstract ... 3

Chapter I: Introduction ... 3

Chapter II: Literature Review ... 5

Ethnic Conflict ... 5

Intervention into Ethnic and non-ethnic Conflicts ... 7

Transnational Ethnic Linkages ... 9

Conflict Outcome ... 10

Rational v Non-Rational ... 12

Chapter III: Theoretical Framework ... 15

Causal Pathway ... 15

TEK Hypotheses ... 18

Intervention and TEK Hypotheses ... 19

Africa and Regions Hypothesis ... 20

Ethnic Group Size Hypothesis ... 21

Chapter IV: Research Methodology ... 21

Data Sources ... 22

Independent Variable ... 24

Dependent Variable ... 25

Control Variables ... 26

Data Strategy ... 28

Chapter V: Results and Analyses ... 29

Effect of Neighbouring Ethnic Kin ... 29

Effect of Contiguous Interventions ... 34

Africa and other Regions ... 39

Ethnic Group Size ... 41

Chapter VI: Discussion ... 44

Hypothesis 1 ... 44

Hypothesis 2 and Sub Hypotheses ... 44

Hypothesis 3, 4 and Sub Hypotheses ... 45

Hypothesis 5 ... 47

Hypothesis 6 ... 48

Chapter VII: Conclusion ... 49

Appendix ... 51

List of Abbreviations

TEK Transborder Ethnic Kin

TEK_EGIP Transborder Ethnic Kin in Power (in the neighbouring state)

TEK_MEG Transborder Ethnic Kin not in Power (in the neighbouring state)

EGIP Ethnic group in Power MEG Ethnic group marginalized govt Government

support_govt Intervention in support of the government contig_int Contiguous Intervention

Abstract

Previous research on Ethnic Civil War Outcome is limited to measuring it within the larger context of civil wars, or when measured more specifically, with the presence of a Transborder Ethnic Kin. This thesis goes above and beyond these previous works by measuring the impacts the presence of a transborder ethnic kin, the power the transborder ethnic kin holds in the neighbouring state, and the impact of neighbouring state intervention has on ethnic civil war outcome. I argue that transborder ethnic linkages can lead to irrational decision making on the part of ethnic leaders, but that the irrationality of ethnic ties can be limited by the power dynamics in the two states. In measuring 84 civil wars between 1944-2004, I found this to be true; the existing power dynamics at the start of ethnic civil conflicts is a strong predictor of conflict outcome, as are other ethnic group

characteristics.

Chapter I: Introduction

The prevalence of ethnic conflict throughout history has made it imperative for society to understand both how these conflicts begin, and just as importantly, how these conflicts end. According to one study, there have been over 600 ethnic civil wars since the end of World War II (Wucherpfennig, et al., 2012). These conflicts become more difficult to solve when the ethnic groups in conflict have a transborder ethnic kin (TEK) in the

neighbouring state. To clarify, these are ethnic groups who share transnational, and transborder, “…ethnic connections and whose settlement area is split by an international border” (Vogt, et al., 2015). Measured by ETH Zurich, there are over 600 ethnic groups who share these ethnic connections across international borders.

In conflicts where the ethnic group does have ethnic kin in the neighbouring state, there is both a higher risk of conflict onset (Cederman, et al., 2009), and also, depending on the power dynamics in both states, a higher rate of state level military intervention (Austvoll, 2005). Though there is work involving ethnic conflict outcome when there is the presence of a TEK (Gurses, 2015), there is none when involving both the power dynamics of both ethnic groups in the conflict dyad, and in the neighbouring dyad, and that measure the effect

neighbouring biased interventions have on conflict outcome.

This is what I aspire to do with my work here: examining how ethnic civil wars1 _end,

in instances where at least one ethnic group is contiguous2 _{across state borders. In these}

1 _{I will be using the terms conflict and wars interchangeably throughout this thesis.}

cases, I will be measuring conflict outcome with the presence and power of TEKs as my independent variable, and then another with neighbouring non-biased and biased

intervention. In researching this, what are the determinants of conflict outcome, and how can TEKs effect conflict outcome both in instances where the neighbouring state intervenes, and when it does not?

I intend to show that the power dynamics within the conflict state at the moment of conflict onset are the strongest predictors of conflict outcomes, but also just as importantly is the power the ethnic kin holds in the neighbouring state. After measuring this, I will measure neighbouring biased intervention and its effect on conflict outcome, both with and without also measuring TEK power.

I will be measuring ethnic conflict and their outcomes from the work of DeRouen Jr. and Sobek, and the ethnic power status from ETH Zurich, which gives me 84 observations. When reviewing the interventions in the IMI database, I will have 50 contiguous intervention observations. The conflict outcomes go from 1944 through 2004, and at the time of the last year there were ten conflicts that were still ongoing3_.

My findings both confirm and disconfirm what I expected to find. Though the interaction effects between ethnic groups in the two states did not have the significance I proposed it would, it was rather the preconditions on ethnic size in the conflict state, its geographical composition, and the power the ethnic group held in the conflict state that mattered, no matter the power its kin had in the neighbouring state. These results held true even when examining the impact of contiguous biased interventions.

These results go beyond what the existing literature is on ethnic conflict outcome. Previous research looked at conflict outcome and the impact a TEK can have on conflict outcome, but here I am interacting the power dynamics in the conflict state and neighbouring state. Further, there is little literature measuring specifically neighbouring state interventions impact on ethnic conflict outcome, something that I am measuring here.

This thesis will be structured as follows. First, I will discuss the current literature on the various themes that underline my theory and methodology. Second, I will discuss my theoretical framework, and the casual pathway linking my theory to my hypotheses. Third, I will discuss the methodological framework and my data strategy. Fourth, I will go through each of my hypotheses individually, discussing my results. Fifth and finally, I will discuss my results and extrapolate my findings, and discuss any limitations.

Chapter II: Literature Review

This review is divided into the following five subjects: 1) Ethnic Conflict, 2) Intervention into Civil Wars both Ethnic and not, 3) Transborder Ethnic Linkages, 4) Conflict Outcome, and 5) the rational or non-rational makeup of nationalism and ethnicity.

Ethnic Conflict

The breakup of Yugoslavia and the U.S.S.R in the early 1990s provoked fears of rising ethnic tensions and conflict. The subsequent wars in Yugoslavia, the genocide of Tutsis at the hands of Hutus in Rwanda in the 1990s gave rise to primordialist sentiments amongst scholars; the conflicts in the Balkans were simply a form of ancient hatreds (Cohen, 1997). In an attempt to attach an international relations school of thought to ethnic conflict, the security dilemma was discussed as an option to better understand how ethnic conflicts can begin.

In the aftermath of the dissolution of the U.S.S.R. there was a worry over a potential conflict in the Ukraine over their possession of Soviet nuclear weapons. There was a fear that this one issue could provoke conflict between Ukrainians and Russians. However, in discussing the potentiality for ethnic conflict, authors presumed the same ethnic conditions hold across all adversarial ethnic groups (Posen, 1993).

The fear over ethnic conflict breaking out anywhere was high, while at the same moment a unipolar world system was beginning to take shape, leaving some international relations experts unsure of the future of global security and the effect of ethnicity on it. Ethnic dyads across neighbouring state boundaries influence foreign policy choices and behaviour (Davis & Moore, 1997), while others discussed how transborder ethnic linkages can affect the onset of separatist violence (Ayers & Saideman, 2000).

The 2000s saw a shift in ethnic civil war thought, with a rise of ideas linking ethnic conflicts to economic opportunity and the lure of natural resources. This method of examining ethnic conflicts is the so-called greed school of thought, instead of political and economic objective grievances. The former type of thought, the greed school, was propagated by authors such as Collier, Hoeffler, Fearon, and Laitin. They presumed the primacy of economic opportunities and natural resources as better predictors of conflict than the makeup of a state’s ethnic or religious grievances (Fearon & Laitin, 2003) (Collier & Hoeffler, 2004).

These new arguments were bolstered by work centered around micro cleavages within existing macro cleavages of civil wars. These works showed how the micro cleavages in communities could be just as dangerous if not more so than the macro cleavages of ethnicity, religion or ideology. The conflict cleavage in a Spanish town during the Spanish Civil War did not centre around Republicans or Nationalists, but rather around two doctors caught up in a

rivalry (Kalyvas, 2003). In another example, during the War in Bosnia in the 1990s, a Bosniak Muslim leader named Fikret Abdić set up his autonomous region by co-opting Muslims in Bosnia. Fighting broke out between Muslim forces loyal to Abdić, who received support from the Serbian government, and those loyal to the main Bosniak army (Christia, 2008). Economic opportunity in this case cut across the main cleavage of ethnic conflict.

Ethnic conflicts though do not begin with these micro cleavages cutting across already existing cleavages. Discrimination against a group provides leaders and group members with grievances that can be activated. Disaggregating ethnicity and conflict to a centre-periphery divide then shows when powerful ethnic groups are not given political power and access, there is an increased probability of ethnic conflict onset. This probability increases when the excluded ethnic group is located further from the capital and/or the terrain separating the capital from the ethnicity is mountainous or rough (Buhaug, et al., 2008).

There is further evidence against the totality of the greed model propagated by these authors. Ethnic group level inequality, absolutely and relatively to other ethnic groups in the state, increase the probability of conflict onset. Though the finding is restricted to the post-Cold War era, the absence and even presence of oil has no impact on the probability of conflict, directly contradicting a crucial finding from previous work (Fearon & Laitin, 2003). Horizontal inequality at a political or economic level, advantaged or disadvantaged, is indeed a crucial variable in Civil War onset (Cederman, et al., 2011).

Three conditions do have to be in place for opportunity, or greed, to take advantage of already existing grievances. 1) a strong a base of support has to exist, 2) a large enough pool of material whether that be money or resources has to be in place, and 3) a weak central government (Denny & Walter, 2014). These three conditions do not matter if ethnic grievances do not already exist within the state, and even then, other conditions matter. The population size of the aggrieved ethnic group must be large enough, and the group should be geographically concentrated far from the capital, or close to state borders. As Denny and Walter write, divisions along ethnic lines will be persistent across time; one can change your ideology or religion, but ethnicity tends to stick with individuals, “…ethnic identity is also one of the best predictors of how individuals are likely to vote, now and in the future” (Denny & Walter, 2014).

More than likely, violence in ethnic conflicts is a combination of both of these phenomena; ethnic violence does exist within both interstate and intrastate conflicts. Legitimate widespread grievances can exist and be the opportunity that secessionist or extreme leaders can use to incite conflict and violence. Political and economic discrimination can

reduce separatism onset (Ayers & Saideman, 2000), yet in Northern Ireland persistent and real grievances held by Irish Catholics fueled violence for decades. Clearly, though findings can have strong external validity, they can also fail to properly capture specific cases.

Intervention into Ethnic and non-ethnic Conflicts

Before I begin this section, I am only looking at unilateral interventions on behalf of one side during a conflict. There is no need to discuss multilateral interventions by a supranational force whether they be neutral or whether they are biased4_.

External intervention into ethnic, ideological, or religious civil conflicts can be motivated by a number of different factors, but is posited that most unilateral third-party interventions are motivated by a desire to end the conflict rather than prolong it (Regan, 2000). The intervening state does so to ensure political stability in their neighbourhood and the larger region. Rarely according to Regan do external unilateral interventions attempt to solve the underlying issue that causes violence. In intervening, Regan posits that external actors change the cost-benefit calculation of each party to ensure the cost of continued fighting be so high as to be not worth it. However, if ethnic ties exist across state boundaries, this changes the utility model for ethnic leaders (Regan, 1996).

External interventions are also rare when the casualty rate is high (Elbadawi & Sambanis, 2000), a finding supported by others (Regan, 1996), and the former find that interventions into primarily ethnic conflicts is rarer than other types of conflicts. Interventions into ethnic or religious conflicts have a higher success rate than ideological conflicts, and when the strategy of intervention is both military and economic, this success rate jumps further (Regan, 1996).

Ethnic fractionalization is also not associated with external intervention, meaning that if two ethnic groups are in conflict with each other and are of similar size, external intervention is rare. This does not take into account whether an ethnic group has a TEK, but it would make sense for the TEK in the secondary dyad to not intervene if its kin in the primary conflict dyad is not in need of external support. Second, increased duration of conflicts is associated with external interventions. Finally, interventions are less likely in regions where there are high levels of democracy (Elbadawi & Sambanis, 2000), furthering the Democratic Peace hypothesis, in which democracies do not get into conflict with each other (Babst, 1964).

4 _{Though I do not control for U.N. intervention in most of my hypotheses, when measuring conflict outcome in}

different regions, I do measure the effect of U.N. intervention in H5, which I will elaborate on when discussing my methodology and data strategy.

Neither of the above articles or writers discuss intervention into solely ethnic conflicts however, so it is worth discussing in more detail the current literature on interventions into ethnic conflicts. Framing intervention from a purely rational actor standpoint, the decision and strategy of intervention depends on what the intervening actor believes its payoff from intervention to be. Regan hypothesizes that interventions into ethnic conflict will have a higher chance of success, writing that in ethnic conflicts, there is also the prospect of counter-interventions, thus making each side more amenable to peaceful outcomes (Regan, 2000).

With this, Regan is stating that while it is hard to shake ethnic identity, prospective defeat for one ethnic community should lead it to act for its own survival and perhaps drive for peace. This is perhaps a different read on the ethnic hardening hypothesis (Kalyvas, 2008), as well as ethnic outbidding which I will discuss later on. However, Regan does not examine ethnic kin ties across state borders. I will discuss the importance of ethnic ties across state borders later on.

Interventions across state borders is also dependent on a willingness to take advantage to improve relations with states, or to ensure regional and neighbouring security. Contiguous borders increase the threats faced by states that does not extend to states that are far removed from the conflict. Further, when the conflict is coded as ethnic, there is a higher chance of neighbour states intervening, though this is not disaggregated to transborder kin ties. Like other authors though, its noted that ethnic kin ties do effect state decisions (Kathman, 2010).

Austvoll finds that interventions into ethnic conflict increases when one’s ethnic kin is fighting. Without ethnic ties, the probability of a third party intervening on either side decreases. Interestingly, Austvoll finds the only significant relationship of intervention exists when an ethnic group in power (EGIP) intervenes on behalf of its marginalized ethnic kin (MEG) in the conflict dyad. If both ethnic groups are in power, or both are marginalized, then the secondary ethnic dyad will intervene on behalf of the government (Austvoll, 2005). This will be discussed later on.

In cases where states have multiple ethnic kin, yet one is dominant politically and demographically, the state is more likely to intervene on behalf of their ethnic kin, especially so when their neighbouring kin is geographically concentrated. Even still, the rate of intervention is high when the kin are not geographically contiguous, and not concentrated geographically (Huibregtse, 2010).

Ethnic kin intervention into conflict then must be balanced with the ties of kinship and weighing the rational benefits and costs of such action. Ethnic leaders will face pressure from their constituents to intervene in conflicts and act aggressively, while also facing the fear of

political ethnic outbidding and outmaneuvering. Huibregtse does state however that, “if a kin state considers only ethnic ties, it will intervene to help their brethren at any cost and regardless of the other factors” (Huibregtse, 2010). Taking this further, ethnic kin can be led down a path of committing valuable resources to ensure at least a favourable outcome for their kin regardless of the costs and consequences.

Here, it is clear there is a separation between the rational choice, utility model that scholars use to determine ethnic kin actions, and the ethnonational bonds that bind people together as outlined by Walker Connor. This will be discussed in further detail later.

Transnational Ethnic Linkages

Transnational ethnic linkages can not only effect actions taken by ethnic kin, but also effect state behaviour and rhetoric. Davis & Moore find that advantaged ethnic minorities can increase conflict onset between states if their kin in the other state is disadvantaged (Davis & Moore, 1997). This relationship is also found if the ethnic groups share a border. This is found by others as well; kin protests and rebellion in neighbouring states and in the same region increase the probability of separatist movements and violence (Ayers & Saideman, 2000). Similarly, the presence of transborder ethnic groups increases the risk of civil war onset (Gleditsch, 2007).

Martin Austvoll measures a similar finding, with kin dyads in both states there is an increased chance of conflict and intervention. Disaggregated further, Austvoll finds that if the ethnic group in power, EGIP, and the target ethnic kin in the conflict state is marginalized, MEG, then there is an increased probability of intervention (Austvoll, 2005). As mentioned before, this relationship is not found in any other scenario with ethnic groups being in power or not.

This is not absolute finding however; Cederman et al. do not find a significant relationship if the intervening TEK is in power. They do however find an increased probability of conflict onset if the excluded ethnic group in the conflict state is large in proportion to the rest of the state population (Cederman, et al., 2009). This is also found by other authors in separate works (Buhaug, et al., 2008) (Gleditsch, 2007). Further, conflict onset increases when the ethnic dyad is located far from the capital or when the terrain is mountainous (Cederman, et al., 2009) (Buhaug, et al., 2008).

Though there is a relationship between ethnic group’s size, its access to power, and neighbouring ethnic kin presence or intervening on its behalf, its impact is limited. Ethnic kin in the primary conflict dyad can use their TEK to access a greater pool of resources, increasing

their own bargaining position (Gleditsch, 2007). Ethnic kin in the secondary dyad can also be more confrontational and more likely to support aggressive actions. This is also something that is found in diaspora communities. Walker Connor agrees with this; minority ethnic groups in a different state than their homeland will strive to remain as one with their own ethnicity and maintain high levels of nationalism, rather than loyalty to the state. This is a contrast between nationalist loyalty with patriotism, which is loyalty to the state. In this case, Connor is at least rhetorically resistant to the idea of multi-national states (Connor, 1993).

As Gleditsch and Salehyan note separately, the advantage of having a TEK can be nullified if the two states have a strong relationship (Gleditsch, 2007) (Salehyan, 2007). Austvoll finds this when examining the intervention risk of a MEG on behalf of its MEG; in that instance the MEG in the secondary neighbouring dyad is more likely to intervene on behalf of the government (Austvoll, 2005). Rival states are more likely to see conflict continuation, as the secondary dyad may provide support for transnational rebels. States face a high cost of crossing state borders to intervene in neighbouring conflicts, but this also works in the reverse; there is a high cost to states in choosing to intervene in the secondary dyad to attack transnational rebels (Salehyan, 2007).

Regardless, the closed polity approach, of states existing as one and in isolation from each other does not hold even if there are no transborder ethnic kin. States are affected by violence in neighbouring states, especially so if there is a TEK in conflict. Transborder ethnic kin are also related to secession and irredentist movements (Lake & Rothchild, 1998), a finding replicated by others (Ayers & Saideman, 2000). As one article on ethnonational kin quoted, “…too many theories…assume…irrelevancy as far as the international environment is concerned and assume also that internal political development or decay occur without regard to external factors.” (Cederman, et al., 2009) Though geopolitics can limit overt military aid from one ethnic group to another, there are other methods of support. But limiting ethnicity to a back role is to not take it seriously enough, a thought echoed by Walker Connor which will be discussed in more detail later.

Conflict Outcome

There are four ways intrastate conflict can cease: 1) government victory, 2) rebel victory, 3) truce, or 4) treaty. Depending on the type of conflict waged by rebel forces, a simple goal of rebels is to keep the rate of new recruits above the rate of rebel deaths (Sànchez-Cuenca, 2007). Further, rebels, especially so in democracies and pseudo democracies, must keep popular support high enough in order to wage effective campaigns of violence. Though ethnic

outbidding and competition can resort to more indiscriminate acts of violence (McCauley & Moskalenko, 2008), for example the IRA in Northern Ireland and the ETA in the Basque Country had almost near monopolies on the use of rebel violence. Instead, both groups had to contend with public perception within their own religion and ethnic base (Sànchez-Cuenca, 2007).

Installing a base of support then from the population can strengthen bargaining positions for rebels and increase the probability of truces or treaties between the government and rebels. Even still, government victory is the most likely outcome in almost all types of intrastate conflicts. The probability of government victory is at its highest in the earliest months of the conflict (DeRouen Jr. & Sobek, 2004). As the conflict progresses in duration, the probability of rebel victory increases, as does the probability of truces or treaties. This outcome pattern is replicated in other studies (Fett, et al., 1999). However, other scholars find that while duration does increase the probability of rebel victory, truce, and treaties, they also find that government victory probability increases rapidly once conflict duration goes past 15 years (Brandt, et al., 2008).

As a proxy for state strength, one study finds that a strong bureaucracy severely restricts the possibility for rebel victory, though not effecting government victory or any other outcome. Government army size increases the probability of each possible outcome (DeRouen Jr. & Sobek, 2004), a finding not shared by other authors which find that army size increases the probability of government victory while decreasing other possible outcomes (Fett, et al., 1999). This is confirmed in other studies; that an army’s size in proportion to its population increases the probability of government victory and reduces the duration of intrastate violence (Brandt, et al., 2008).

Though not looking at ethnic conflict outcome directly, they find that ethnic fractionalization actually increases the probability of all outcomes (Brandt, et al., 2008), but over time, decreases all possible outcomes. As they put it, ethnic fractionalization simply increases the duration of intrastate conflicts, with the probability of truce or treaty being the most probable outcome (Brandt, et al., 2008). DeRouen and Sobek find something similar in that ethnic fractionalization increases the probability of treaties, but not of truces, while decreasing the probability of government victory (DeRouen Jr. & Sobek, 2004).

However, if the intrastate conflict is ethnic, a sharp decrease in rebel victory, while increasing the probability of truce as an outcome (DeRouen Jr. & Sobek, 2004). Another study finds something similar; ethnic intrastate conflicts increases the probability of government victory while decreasing the probability of rebel victories (Fett, et al., 1999). “Rebel forces

enjoy a greater probability of victory in separatist wars than in revolutions, and governments enjoy a higher probability of victory in ethnic wars…” (Fett, et al., 1999). Interestingly as well, more borders can increase the probability of rebel victory; perhaps this gives rebels a greater chance to flee violence from the government and establish bases outside the state (DeRouen Jr. & Sobek, 2004).

Finally, intervention into intrastate conflicts has a large effect on conflict outcome. UN intervention can dramatically increase conflict outcomes of truce and treaties, though they do not measure intervention of other kinds (DeRouen Jr. & Sobek, 2004). In other works, authors find that neutral interventions increase the probability of government victory while reducing settlements. Biased interventions increase the probability of victory for whatever side it intervenes for, while reducing the probability of settlements according that same study (Fett, et al., 1999). Though they do find that increased duration combined with neutral or biased interventions increases the probability of settlements. This is also found by Brandt et. al, where intervention leads to a negotiated settlement as the most likely outcome (Brandt, et al., 2008). Disaggregated further, interventions into ethnic conflict reduces the chance of successful outcomes, but military intervention in support of the government leads to higher success. Though Regan does not discuss interventions from neighbouring states, it is still of note that interventions on behalf of the government are most likely to be successful in government victory, and that ethnic conflicts are tricky to intervene in (Regan, 2000).

Disaggregated even further, ethnic conflicts with TEKs are more probable to end in settlements than in rebel victories. However, Gurses does not measure intervention by TEK into conflicts, rather he only uses the presence of a TEK as sufficient evidence. Regardless, his work is one of few that attempts to capture the effect neighbouring states specifically can have on ethnic conflict outcome (Gurses, 2015).

As a whole, intrastate conflicts are more probable to end in government victory especially in the outset. Neutral interventions increase the probability of truce and treaty as outcomes, while biased interventions increase the probability of victory for the side receiving assistance. Further, ethnic conflicts are more likely to end in government victory, and even if rebels have a TEK, there is only a significant probable increase in settlement as an outcome, rather than rebel victory.

Rational v Non-Rational

Across intrastate conflict literature, and ethnic intrastate conflicts as well, authors presume actors act rationally. Gurses examines transnational ethnic conflicts through an

expected utility model, as does Regan in his intervention works. Collier and Hoeffler, and Fearon and Laitin wrote that intrastate conflicts are driven more by greed and opportunity rather than ethnic grievances. Buhaug, Cederman, and Rød examine conflicts through a different rational choice phenomenon, writing that certain conditions allow for ethno-nationalist civil wars (Buhaug, et al., 2008). Cederman et. al show how a combination of demographic size and power effects ethnic conflict onset (Cederman, et al., 2009). Salehyan also writes that having access to extraterritorial bases in neighbouring states can increase onset and duration of conflicts (Salehyan, 2007). Finally, DeRouen and Sobek examine civil war conflict outcome through an expected utility model (DeRouen Jr. & Sobek, 2004).

In examining language vitality and conflict intensity, a similar rational choice theoretical perspective is followed. In doing so, an inverted U curve hypothesis is formed where there is a middle range of language vitality and conflict intensity (Medeiros, 2017). As language can be seen as part of one’s ethnicity, it is worth inclusion here. An inverted U curve is present in other literature as well; ethnically polarized states leads to prolonged civil wars (Elbadawi & Sambanis, 2000) (DeRouen & Sobek, 2004). Interventions are also hypothesized to be impacted by the power ethnic groups hold. State actions are also impacted by the power access ethnic groups have (Davis & Moore, 1997) (Austvoll, 2005). Austvoll specifically walks through EGIP and MEG actions through a rational choice framework. Finally, the poorest and wealthiest ethnic groups are more likely to experience ethnic conflicts, another U curve phenomenon (Cederman, et al., 2011).

While that is not necessarily incorrect, it is worth noting that the rational choice expected utility model of ethnicity is not the only model. Using this model to examine ethnicity, ethnic conflicts, interventions, and especially ethnic interventions is short-sighted. Walker Connor is critical of scholars’ over-reliance on rationality to explain ethnic actions. Though rational choice theory does have its place, it is not the be all end all of examining ethnic group action. In ethnic heterogeneous states, it is not objective criteria that determines whether a group can become a nation. A “…nation is a psychological bond that joins a people and differentiates it…from all non-members…” (Connor, 1993). In multinational states then, ethnonational groups will have greater ties to ones own state. Connor believes this to be even more so with TEKs, as he uses the dissolution of the U.S.S.R and Yugoslavia to show. Ethnic ties to ones own state assisted in the creation of new independent states of Latvia, Ukraine, Albania, Croatia etc. Nationalism at its core appeals to emotions, our hearts; “people do not voluntarily die for things that are rational.” (Connor, 1993)

Without quoting all of Walker Connor, he argues that economic grievances are not a precondition for conflict, but rather a catalyst or as an exacerbator of potential ethnic strife. In Switzerland, he gives the example of a plurality of citizens voting to deport guest or immigrant workers despite the government publicly stating how much of a negative effect that would have on the Swiss economy (Connor, 1994). The Baltic states during U.S.S.R. rule chose to resist economic investment believing it would increase the migration of ethnic Russians into that area, effecting the demographic makeup of their own ethnic homelands (Connor, 1994).

Finally, kin protest and separatist movements positively effect the probability of ethnic kin separatism in the observed state. Yet, if the protests and separatism are conducted by other ethnic groups, this effect is lost (Ayers & Saideman, 2000). In a rational choice, expected utility model this should not be the case. Successful and even just active separatism should effect the probability of separatism no matter the group in an expected utility model. Ayers and Saidemen thus are able to show how much ethnic ties matter beyond rationality models.

Rational choice is also not the sole driving factor for the making of ethnonational communities. Rational choice can not understand loud and exuberant patriotism, nor, “…bitter hatred which appear to supply the emotional fuel to these conflicts…” (Wimmer, 2002). Rational choice can not explain why ethnonational bonds are used in some instances but not in others. To be sure, this does not mean the opposite is true. As Wimmer writes, neo-romantic models of ethnonational communities can not fully explain these ties. It can not explain why some ethnic communities do not form strong ethnonational bonds while others do (Wimmer, 2002).

This is not to say that ethnonationalist ties do not lead individuals or groups to make solely irrational or ill-advised decisions. It is to state that that a rational choice model is not the most effective or only way to examine ethnic conflict outcome that invoke diaspora communities and TEKs. I believe it is a combination of irrational ethnonational bonds that push TEKs to intervene and to pursue maximal gains up to a point. Ethnic groups will want to see their ethnic kin survive by intervening and postponing defeat, yet not push back so much to invite greater international attention, all things being equal. In a simpler manner, ethnic ties across borders lead to greater probably of conflict onset, which leads to higher chance of intervention, but not prolonged conflicts, or even conflicts that lead to their ethnic side claiming victory. At that stage, rationality is perhaps the best explanation for conflict outcome. If transborder ethnic kin have the ability to intervene on behalf of their kin in conflict they will actively pursue an outcome favourable for their kin. If restricted by existing power dynamics, this effect may be lost and thus favourable outcomes to the kin as well.

Chapter III: Theoretical Framework

The fundamental dissonance in civil war conflict onset is between the greed and grievance debate. Yet this debate only captures the onset of civil wars, not its outcome. While the two terms, greed and grievance, are painted in narrow black and white stripes it would be unfair to state that greed would be translated solely to actors taking advantage of resources for their own self-interests, whether those interests be political or monetary. The greed model, one adapted by Fearon and Laitin and by Collier and Hoeffler, can be interpreted as actors having the economic tools and resources to achieve certain ends.

In tweaking this argument to adopt conflict outcome, authors adopt rational choice and expected utility models to arrive at similar hypotheses. It is correct, as DeRouen and Sobek do, to hypothesize that government capability is a predictor of government victory, or that rough terrain is a predictor of outcomes favourable to rebel forces. But in examining conflict outcome without disaggregating to the ethnic level, DeRouen and Sobek and other authors do not capture in nuance the factors that contribute to their findings.

If rational choice, greed, and opportunity are the most correct methods of understanding ethnic civil conflict outcome, then I believe Ayers and Saideman would have found that non-kin protests and separatist movements in adjacent states would have an effect on active separatism in the observed state. Crucially, they do not find this, and in one instance show the power and influence kin ties have even across state borders, and how much ethnicity matters in relation to rationality.

Simply plugging in ethnic kin as a third variable in an expected utility equation as Mehmet Gurses does is also not enough to fully capture TEK influence. It does not capture the nuance of nationalism and ethnic ties fully. In exploring my topic and theory, I realized that scholars like Gurses, DeRouen and Sobek, Mason et. al, missed in not exploring the combined effects of TEKs neighbouring state intervention and expected rational choices.

Causal Pathway

From Regan, we know that interventions into ethnic civil conflicts are as common as interventions into ideological civil conflicts (which were increased by the Cold War international system) (Regan, 2000) (Balcells & Kalyvas, 2010). We know that ethnic conflict onset probability increases when transborder ethnic kin ties are present, and that these ties can influence foreign policy decisions at the federal level. Neighbouring states with ethnic kin ties increase state cooperation, yet can also increase the risk of conflict onset (Davis & Moore, 1997). Ethnic conflicts are more likely to end in truces which suggest a return to conflict in the future and we can link ethnic intrastate conflicts with longer conflicts, which increases the

chance of external intervention (DeRouen Jr. & Sobek, 2004) (Elbadawi & Sambanis, 2000). Through this path, we can see that ethnicity matters when examining conflict duration and outcome, without having even examined TEKs.

TEKs, and other ethnic kin in states outside the region can be opted, or they can co-opt elites into acting more aggressively and in irrational manners. Being far away from the potential conflict dyad means they do not face the consequences of conflict and can believe themselves to be relatively safe. This is something that Walker Connor and others hint at. By being away from the conflict zone, external ethnic masses and elites have the freedom to act more dangerously. Franjo Tudjman, President of Croatia in 1991, received funding and support from external elites with extreme views towards the Serb minority in Croatia (Fearon, 1998). The Free Aceh Movement was able to survive in part due to external Acehnese diaspora elites providing huge quantities of resources. At the same time, the Acehnese believed that the diaspora would always provide support for their movement (Missbach, 2013).

Those are just extremist actions taken by diaspora elites; neighbouring ethnic groups provide support in different capacities as well (Salehyan, 2007). Basque separatists, ETA, used the neighbouring French border to escape Spanish forces, while the PLO used base camps in Jordan to train and escape Israeli forces. This opportunity is compounded when the neighbour state is of the same ethnicity. For the ETA, there was a sizeable Basque population in the neighbouring regions of France.

For Connor, ethnic ties are stronger together then ethnic ties to the multiethnic state. And in instances of neighbouring ethnic groups these ties are exacerbated. Though not the most apt example in this case, Connor illustrates that in World War I, Socialists and Marxists were surprised when the working classes of France and Germany sided with their own nationalities rather than with their own class across state lines (Connor, 1994). Though obviously an interstate conflict, it is an example of ethnic ties trumping other international non-ethnic ties.

What is known as ethnic outbidding can occur in both the conflict dyad and in the neighbouring dyad. In competing for the same base of support, groups can act in more extreme methods than other groups to show commitment to a cause. In Croatia, Tudjman won the Presidency but was supported by extremists in his own party and could not afford to give out concessions to the Serbs in fear of losing their support and in fear of being politically attacked by members of the opposition party (Fearon, 1998).

If members of an ethnic kin outside the homeland dyad are more aggressive and extreme in their actions and beliefs, then they are more likely to provide support for their kin fighting in the conflict dyad. This support can be rhetorical, or it can be more substantial. As

ethnic groups invest in their kin and provide more support and intervene with their own military forces, the dynamics of the conflict change to a situation more favourable to this ethnic group. There are confounding factors to this that will be outlined when I discuss my control variables, but the principle of mission creep should remain constant (Hoffmann & Schneckener, 2011).

In mission creep, an actor becomes involved in a conflict to achieve a narrow goal. If it fails to achieve that goal in quick fashion, it may become bogged down in a larger conflict with more expansive goals beyond its original intentions. This is most apparent in the U.S. invasions of Afghanistan and Iraq. The same concept can occur with TEKs as they intervene on behalf of their kin, expecting a quick and successful outcome. However, if this does not occur quickly, this can extend the civil war and create a more intense conflict. This aligns with previous research: long-lasting intrastate conflicts are associated with external interventions (Elbadawi & Sambanis, 2000). And as mentioned before, longer lasting intrastate conflicts are associated with an increase in truce or treaties (DeRouen Jr. & Sobek, 2004).

This effect of TEKs intervening with military force is not tempered by the international system (Balcells & Kalyvas, 2010), but it is tempered by the power dynamics in the conflict dyad and the secondary dyad (Austvoll, 2005). It is here that rational choice theory does matter when examining TEK ties and intervention. Though only measuring conflict onset, not outcome, ethnic intervention on behalf of their kin is found when the intervening ethnic kin dyad is in power, and their conflict kin are marginalized (Austvoll, 2005). This aligns with previous research; marginalized kin with real and perceived grievances and inequality, are more likely to be involved civil war conflicts (Cederman, et al., 2011). This finding is lost when measuring other kin power types (Austvoll, 2005).

If this intervention is lost to the power dynamics in the conflict dyad, then the outcome will change with it. Stripping one side of intervention in an ethnic conflict decreases the chance of victory for the rebels (Fett, et al., 1999). In this case, the probability of victory increases for the government (Fett, et al., 1999), or there is an increased chance of a truce (DeRouen Jr. & Sobek, 2004). There are other confounding factors to this outcome that will be brought up when I discuss my control variables, namely the ethnic group population distribution. This goes beyond what other scholars have researched; like Gurses, I will be controlling for ethnic population distribution and size, but I will be going further in my research. Gurses does not control for the power ethnic groups have in both dyads, thus failing to capture the true effect of conflict outcome. Furthermore, Gurses does not actually capture neighbouring state intervention as Austvoll does; instead he uses the presence of TEKs as a proxy for their effect on conflict outcome.

In summary, the irrational ties of ethnicity across state boundaries are termed not by the region, or by the international system, but by the rational conditions of ethnic power balances. Scholars can be correct in examining ethnic conflict outcome through the prism of rational choice utility models, but to box in ethnic ties solely into this model is too not examine ethnic ties in a more substantive manner.

TEK Hypotheses

As coded by DeRouen and Sobek, and from already existing literature, biased intervention on behalf of one actor in a conflict increases the probability of victory (Fett, et al., 1999). Therefore, in a straightforward statement, we can extrapolate this to the main hypothesis, without any considerations.

H1: Ethnic groups in conflict with a transborder ethnic kin leads to victory as the most probable outcome for this ethnic group.

In this instance, the bonds of ethnic ties supersede rational choice models, in that ethnic groups push for victory with ‘blinders’ on, and not caring for the consequences of their actions. I do not expect that I will end up accepting this hypothesis, but it is a base hypothesis from which to build off of.

H1a: Ethnic kin in conflict without a transborder ethnic kin will raise the probability of the conflict outcome ending in victory for the opposition.

From previous research, we know the importance of access to external bases of support, ethnic or not (Salehyan, 2007). Removing direct access to men and material deprives rebels of a key addition to survival; the rate of new fighters and resources being above the rate of loss (Sànchez-Cuenca, 2007).

H2: The probable effect of H1 is amplified or constrained by the power dynamics in the conflict dyad and in the neighbouring dyad.

H2i: If the ethnic group in conflict is marginalized, and its TEK is in power in the neighbouring state, the probability of an outcome favourable to the ethnic kin and above H1.

H2ii: If the ethnic group in conflict is in power, and its TEK is also in power in the neighbouring dyad, the most probable outcome is government victory and higher than H1 and H2i.

H2iii: In any other power dynamic involving the kin in the conflict dyad and the neighbouring dyad, the probability of the observed ethnic kin winning falls below H1.

In previous work on transborder ethnic kin and the effect on intervention, state ties can supersede the ties of ethnicity between groups across state borders. In instances where both

ethnic groups are marginalized, the secondary marginalized group might not have the opportunity to provide support for its ethnic kin in conflict. Furthermore, the state itself can provide support to its neighbour in conflict to fight the marginalized group to further its own interests (Austvoll, 2005) (Saideman, 2002). In a more specific example, the Turkish government does not support Kurdish fighters in Syria, while the PKK attempts to provide some support for the Kurdish forces in Syria.

In previous work on transborder ethnic groups and conflicts, the data does not fully capture the intervention of non-state actors. This is a methodological issue across many works: Austvoll captures interventions by state forces, not by the individual ethnic groups, Gurses does not capture intervention at all, rather simply the presence of TEKs as a proxy for their effect. Unfortunately, the existing and available data does not allow me to go above this and capture non-state armed actors intervening on behalf of their fellow ethnic kin. Due to this, my hypotheses on neighbouring intervention effect on conflict outcome will mirror those of Austvolls, but with focus on conflict outcome, not onset.

Intervention and TEK Hypotheses

H3: When both the ethnic groups in the conflict dyad and in the neighbouring state are in power, biased intervention will favour the government and lead to outcomes favourable to the government.

H3i: When both the ethnic groups in the conflict dyad and in the neighbouring state are marginalized, biased intervention will favour the government and lead to outcomes favourable to the government.

H4: When the ethnic group in conflict is in power and its neighbouring ethnic kin is marginalized, intervention will favour the ethnic group in conflict, and lead to government victory as the most likely outcome.

H4i: When the ethnic group in conflict is marginalized and its neighbouring ethnic kin is in power, intervention will favour the ethnic group in conflict, and lead to outcomes favourable to the marginalized ethnic kin.

The current literature does not take into account the potential for neighbouring interventions during ethnic conflicts. Unfortunately, the data that exists today does not allow me to conduct a study measuring non-state armed group contiguous interventions into ethnic kin conflicts. But we can use state interventions as a proxy, which is something that Austvoll does. In one study, there is no significant effect of biased intervention on conflict outcome, but again, that is not neighbouring intervention (Fett, et al., 1999). Interventions in support of the

government though are more likely to end in success (Regan, 1996). But none of these take into effect the ethnic power dimensions.

Though Austvoll does not study conflict outcome, he finds that the power dynamics matter most for direction of intervention, and likelihood of intervention (Austvoll, 2005). If we take that and extend his findings to the belief that biased intervention increases probability of said side winning, then we can build the hypotheses above. As stated, the power dynamics in the conflict state and the neighbouring state play an outsized role in effecting conflict outcome.

H3 and H3i have similar outcomes because if both ethnic groups are in power, not only

are their ethnic ties to maintain, but state stability as well. If both are marginalized, then states may feel the need to intervene on behalf of the government, or at least restrict the ability of ethnic rebel groups to support their ethnic kin. However, when there is a power asymmetry between the neighbouring dyad and the conflict dyad, this is where I expect to see intervention on behalf of the ethnic kin, and thus victory for this side.

Africa and Regions Hypothesis

H5: On the continent of Africa, more cross-border interventions by state actors will increase the probability of an outcome favourable to the government.

The borders of many African states were drawn arbitrarily by European colonizers with no regard for the geographical ethnic composition of these states. In doing so, ethnic groups, already brutalized by European colonizers with some groups having complete access to power over other groups, were divided unevenly across state lines. With a larger number of TEKs then, one would presume to have more cross-border interventions and assistance from secondary dyads. However, there is some evidence to suggest otherwise; that the borders of African states are actually hardened compared to other regions (Kathman, 2010). With fewer interventions on behalf of rebel actors, this decreases their chance of a favourable outcome for rebels. This can mean more assistance on behalf of governments to other governments.

However, there is evidence that civil war conflicts in Africa are more likely to end with treaties than government victories. This evidence though is constrained by the fact that many conflicts were still ongoing when the study was finished (DeRouen Jr. & Sobek, 2004). Because of this, the outcome most probable is not outright victory, but outcomes favourable to the government. In running my models, I will make Africa the reference category to which base other regions on.

Ethnic Group Size Hypothesis

The interaction effect between conflict intensity and onset with ethnic group size and political inequality leads to the U curve and inverted U curve phenomena, I am hypothesizing that this U curve is also seen when measuring conflict outcome.

H6: Smaller ethnic groups with limited access to power will see agreements as the most likely outcome, and larger ethnic groups with and without access to power will see government victory as the most likely outcome.

Previous work on ethnic conflict onset has studied the population size of the ethnic group in conflict and the size of the TEK as well. One work shows that large TEKs can have a conflict dampening effect, while those ethnic groups in the middle of the population size see a higher probability of conflict onset (Cederman, et al., 2013). As previous work on conflict intensity and conflict onset has shown, horizontal inequality is associated with a higher rate of conflict at both ends of the U curve (Cederman, et al., 2011). Those ethnic groups that are the most unequal relative to other ethnic groups in the state are more likely to see conflict, while those that are richer, or more equal than other groups, are also more likely to see conflict.

As my methodology for this hypothesis will show, I am examining the conflict outcome probability of ethnic groups by population size, and those groups that are politically unequal and those that have the most power. I expect to see both a U-Curve and an inverted U-Curve of rebel victory probability when measuring power status and population share respectively.

Chapter IV: Research Methodology

The data in my final research includes a combination of ETH Zurich datasets on ethnic conflicts, ethnic group geographical composition, their power at the regional level and federal level, and the presence of transborder ethnic kin. This data makes up some of the control variables of my data and is necessary to fully examine the power disparity between ethnic groups across state borders. From there, I merged the ETH Zurich datasets with the civil conflict dataset from DeRouen and Sobek where I can capture the civil conflict outcome as my dependent variable. Finally, I used the IMI Intervention Dataset to capture intervention from state and non-state actors as my independent variable (Pearson & Baumann, 1993) (Kisangani & Pickering, 2008). In total, my dataset captures ethnic civil conflicts from 1944 through 2004 with 84 conflict outcomes.

Data Sources

As written before, the control variables are taken from the ETH Zurich Ethnic Power Relations Dataset Family. From them, I took the datasets on Transborder Ethnic Kin, Ethnic Conflict, and Geographically Referenced Ethnic Groups. Each of these datasets is based off the Ethnic Power Relations (EPR) Core Dataset which, like the Minority at Risk Dataset (MAR), observes the political access of ethnic groups globally. Unlike the MAR dataset, the EPR data observes not just ethnic groups at risk, but all ethnic groups and their access to power at the federal and regional level (Vogt, et al., 2015).

The GROWup Dataset

ETH Zurich has public data that does go beyond the data I’ve listed below but are also from the same source. Crucially, it provided data on transborder ethnic kin ethnic power, whether it was marginalized or in power that the main TEK dataset did not. Second, it provided data on ethnic group distribution (Girardin, et al., 2015).

The EPR Core Dataset (1946-2017)

Observing over 4 000 ethnic groups globally, the dataset gives each group its own identifying number. For example, Hungarians living in Hungary are given the identifying code 31001000, but in Slovakia the code given is 31702000. In each state, the data codes a group into one of seven categories: 1)Monopoly, 2)Dominant, 3)Senior partner 4)Junior partner, 5)Powerless, 6)Discriminated, and 7)self-exclusion. Further, the dataset codes the proportional demographic size. In doing so, the dataset captures political access at the federal level, at the executive level, and at lower political levels. Finally, each group is splintered into different time periods, to better reflect the shifting power dynamics within states (Vogt, et al., 2015).

The Transborder Ethnic Kin Dataset (TEK)

Using the same identifying codes for ethnic groups, this dataset captures over 600 TEK connection across state borders, with some groups having multiple kin across multiple states. The data also does not stop at capturing TEK with contiguous ties, it observes ethnic groups living in the neighbouring states but do not have direct geographical links. The dataset the ethnic group the same TEK across multiple states to make it easier to observe each ethnic group’s geographical distribution easier and therefore to interpret.

Using this data on ethnic groups and their transborder ethnic linkages instead of the MAR dataset gives me two advantages. First, because the coding process is the same, it allows for consistency across regions and ethnic groups. This makes combining datasets easier and leaves less room for error. Second, unlike the MAR dataset, this dataset observes the political power each ethnic group has across borders as well. This allows for better observation

involving political access for ethnic groups in the conflict dyad and in the secondary dyad (Vogt, et al., 2015).

Ethnic Conflict Dataset (1946-2017)

The Ethnic Conflict Dataset I am using combines the Non-State Actor dataset with the Ethnic Power Relations dataset (ACD2EPR) instead of the UCDP/PRIO dataset. The ACD2EPR data takes non-state armed actors and connects these actors with ethnic groups in conflict. In doing so, the data does three key things. 1)This confirms the ethnic nature of the conflict, 2) The data confirms that armed groups can claim representation of their ethnic groups 3) These armed groups receive public support from the ethnic group they claim to represent (Wucherpfennig, et al., 2012)

The data allows for one ethnic group to support multiple armed actors, as well as multiple ethnic groups to support the same armed actor. Finally, like the TEK and EPR dataset, the coding system is the same, thus making combining datasets easier and smoother.

The GEO-EPR Dataset

Finally, the GEO-EPR dataset takes the original EPR data for politically relevant groups, and codes each group to a geographical location. The group ID codes are the same as the ethnic conflict data, and the transborder ethnic kin. It establishes demographic distribution of regional groups, regional and urban groups, and aggregated groups For other groups, their distribution is stopped by state borders, and migrant groups are not coded. Because it is built from the EPR dataset, it matches the geographical distribution with the power ethnic groups possess over time (Wucherpfennig, et al., 2011). This goes above and beyond the MAR data used by Gurses.

IMI Intervention Dataset

Recorded in two different datasets, one measures the movement of state troops across state borders from 1946-1988, and the second one from 1989-2005. In this instance, the act of intervention has to be an intentional act by the foreign state, and not be taken by non-state actors such as guerilla fighters or paramilitaries. In total, 1114 cases were collected, and the dataset allows for analysts to measure specifically neighbouring interventions and the underlying strategy behind the act of intervening. Though the dataset is coded in two separate forms, the methodology remained consistent from one collection to the next collection (Kisangani & Pickering, 2008).

Independent Variable

There will be two sets of independent variables from two different sources. First, from the IMI Intervention dataset I have taken an independent variable marked “contiguous intervention”. In conjunction to that, from the ETH Zurich dataset I’m using a variable marked “transborder ethnic kin presence”. As will be discussed later, from both of these datasets I am using other variables as control variables (Vogt, et al., 2015).

In measuring transborder ethnic kin effect as an independent variable, ETH Zurich uses a dummy variable that’s coded as, (tek_presence). In total, of the 84 conflicts, 22 do not have a transborder ethnic kin, and 62 have at least one. Of those 62, 51 were in power while 11 were not. This will be my first independent variable and will be included in only the first model. To better measure ethnic group power, I will be using the following variables as my main independent variables before capturing neighbouring intervention, and in the models I do capture neighbouring intervention.

Access to Political Power (status_code) is the main indicator of ethnic access to power in the primary conflict dyad, and in the neighbouring dyad. Ample evidence suggests a link between MEGs as most likely to be embroiled in conflict, with support from a TEK that is in power (Cederman, et al., 2009). This effect may be lost with other power relationships between ethnic groups (Austvoll, 2005). This data will be from the Ethnic Power Relations data from ETH Zurich. The (status_code) variable is coded from 1-3: 1 being ethnic groups excluded from power, 2 being ethnic groups that share power, and 3 meaning that ethnic groups are the dominant political force.

I will be using this (status_code) variable when measuring only when studying the marginal effects of power status in H6. For the other measurements I am using dichotomous variables (EGIP) and (MEG) that captures the power the ethnic group has in the conflict dyad. EGIP codes the ethnic group as in power at 1 and marginalized as 0. For the variable

MEG the coding is the opposite, 1 for being marginalized, 0 for in power.

In doing so, this aligns the variable with the coding structure for transborder ethnic kin power from ETH Zurich. TEK power, (TEK_EGIP) is also a dichotomous variable, with 1 being coded as in power, and 0 is coded as marginalized, and (TEK_MEG) is coded in the opposite way. Powerless, discriminated, and self-excluded ethnic groups in both the TEK power and ethnic group variables are coded as 0. Ethnic groups who share power, dominate or have monopolized power are coded as 1 in both variables as well.

This matches both variables to make interaction effects easier to measure and

interpret. It also aligns the interaction effect with previous work on transborder ethnic effects on both conflict onset and outcome. The main effect I hope to capture is transborder kin effect on conflict outcome, and from there to have it interact with contiguous biased interventions. Thus, the latter will only be brought in in later models.

To further break down the data on contiguous intervention, I recoded the direction of the neighbouring intervention into a dummy variable, where intervention in support of the government is coded as 1, whereas 0 is intervention in support of the rebels. This allows for greater nuance to understand intervention effects, while not losing any observations. Thus, instead of using the variable (contig_int) as my main independent variable, my independent variable will be coded as (support_govt)5_{. Neighbouring interventions coded as being in}

support of governments or against rebel forces are coded as 1, whereas interventions in favour of rebel forces or against government forces are coded as 0.

Dependent Variable

My dependent variable are all ethnic civil conflict outcomes from 1944 through 2004. The conflict outcome dataset from DeRouen and Sobek captures all forms of civil conflicts, not just ethnic ones. To only explore ethnic conflicts, I first removed all conflicts that were not coded as ethnic. From there, I matched up the conflicts that were coded as such to the Ethnic Conflict dataset by country name and by year of conflict start date and end date. The authors use this dataset instead of the PRIO dataset because it codes 1 000 battle deaths in a year as a civil war, and because of its outcome coding. If violence resumes within two years of termination, it is coded as a new conflict (DeRouen Jr. & Sobek, 2004). This gives more cases for which to measure from.

In the data, conflicts are coded into four different outcomes: victory for the government, victory for the rebels, truce, or treaty. A truce is defined different from treaties as ending hostilities with no further settlement. To make this easier and to make my findings more robust, the DeRouen Jr. and Sobek folded the truce and treaty outcomes into one outcome variable, ‘agreement’. In my results then I will have four outcome possibilities: victory for the government, victory for rebel forces, agreements, and ongoing conflicts.

Defining what is a favourable outcome to either actor in a civil conflict is an interesting exercise. Obviously, victory for either side is the most preferred outcome. However, in

5 _{I am still using contig_int as an independent variable in Table 4, and in my appendix in Table 3 I am using}

previous research authors delineate favourable conflict outcomes differently. Gurses for example states that conflict outcomes favourable to rebel forces are either victory or agreements. I agree with this methodology because rebel goals can differ based on context. In some cases, rebels are only fighting for survival and thus seek only more expansive freedom and rights. In essence, not all rebels are looking to overthrow a government. This is dependent on rebel motivations, whether that is greed, grievance, or a mix of both motivators.

However, I would state that government forces have one objective: to defeat rebels at a cost that is not too high to render governments inoperable. Taking Max Weber’s definition of the state, the state has the monopoly of violence within its territorial boundaries (Weber, 1919/1958). Thus, making an agreement with rebel forces erodes the state monopoly on violence. Taking this perspective means only government victory is a favourable outcome for governments. Though it can be argued that in some cases agreements can be favourable for governments, I believe that this can be said in cases when the conflict is ideological. Ideologies can shift over time, while ones own ethnicity can be harder to shake off for some. Thus, while it is hard to defeat an idea, agreements can be reached between ideologically opposed forces. It is harder to do so during ethnic conflicts (DeRouen Jr. & Sobek, 2004)6

This also gives a different methodological frame for outcomes favourable to governments in comparison to rebel forces. This means that agreements will be neutral in their outcome, but also favourable to rebel forces.

From there, I coded an (outcome) variable from 0-3, with 0 coded as ongoing, 1 with conflicts ending in agreement, 2 with rebel victory, and 3 as government victory. Ongoing conflicts, or 0, will be my base or reference outcome.

The data also captures information that I will be using as control variables, including logged income and the Gini Coefficient.

Control Variables

I expect that the Cold War still had an impact on intervention rates (Balcells & Kalyvas, 2010). Because of this, I have split up the data into three different time periods, one of which marks conflict onset from 1989 onwards. This will be my reference category, meaning that conflict outcomes will be compared to those that begin after the cold war ends7_.

6 _{Interestingly, ethnic wars as coded by DeRouen Jr. and Sobek do not see a marked change in any outcome,}

except truces. In this case, an increase in truce shows a return to conflict within two years. This finding shows how hard it can be to shake away ethnic grievances, or greed, and thus make permanent peace hard to reach.