Sustaining the peace after ethnic civil wars

Article Sustaining the peace after ethnic civil wars Conflict Management and Peace Science 30(5) 469 491 Ó The Author(s) 2013 Reprints and permissions: sagepub.co.uk/journalspermissions.nav DOI: 10.1177/0738894213499667 cmp.sagepub.com Mehmet Gurses Florida Atlantic University, USA Nicolas Rost Office of the United Nations Special Coordinator for the Middle East Peace Process (UNSCO), Israel Abstract We explore factors that influence the duration of peace after an ethnic civil war. Results from a series of survival models indicate that political and economic discrimination against the ethnic group that was involved in the war on the opposition side diminishes chances for peace. This finding is robust across different model specifications. Group size, group concentration, war duration and war outcome also influence the risk of war recurrence, although these results are not as robust as those for ethnic discrimination. The intensity of the war and its humanitarian consequences deaths, displacement, and genocide do not seem significantly to influence the duration of post-war peace. Taken together, our findings show that the way ethnic groups interact with each other after a war is a more important factor than the level of violence during the war. Peace and ethnic co-existence are possible in the aftermath of ethnic conflicts. Keywords Discrimination, ethnic civil war, peace duration Introduction A large number of theoretical and empirical studies have focused on the attributes that put countries at risk of an initial onset of civil war (e.g. Fearon and Laitin, 2003; Hegre and Sambanis, 2006; Horowitz, 1985). Yet, as Quinn et al. (2007: 168) observe, the 108 civil wars listed by the Correlates of War dataset for the period between 1944 and 1997 occurred in just 54 countries. Likewise, the 124 wars that Doyle and Sambanis (2000) report for the same period took place in 69 states. As a consequence, many studies have shifted beyond examining the factors that put a country initially at risk of civil war toward understanding the characteristics in a post-civil war environment that influence the chances of sustaining peace vs Corresponding author: Mehmet Gurses, Department of Political Science, Florida Atlantic University, 777 Glades Road, Social Science 384, Boca Raton, FL 33431-0991, USA. Email: gurses@fau.edu

470 Conflict Management and Peace Science 30(5) the risk of renewed war. As researchers have shifted focus, so have policy-makers increased their attention to post-civil war situations in recent years. In his report In Larger Freedom, then United Nations (UN) Secretary General Kofi Annan (2003: 31) asserted that roughly half of all countries that emerge from war lapse back into violence within five years. In response, the UN in 2005 set up a Peacebuilding Commission and Peacebuilding Fund to support reconstruction and peacebuilding in post-war situations. There is a widespread perception that it is particularly difficult to rebuild the peace following an ethnic war. Some have suggested that partition is the best, or even only, way to achieve peace and stability in the wake of ethnic violence. Yet many examples show that reconciliation between ethnic groups is possible within the same country, and that ethnic cooperation is more common than conflict (Fearon and Laitin, 1996). What are the factors that facilitate such reconciliation? In this study we examine the factors that increase the chances for peace and those that sharpen the risk of new violence in the post-war environment after an ethnic civil war has ended. 1 We argue that the government plays the central role in making peace possible. If the government respects the political and economic rights of an ethnic group that was involved on the opposition side in a war, and if it creates institutions such as federalism or proportional representation electoral systems that guarantee a minimum of political participation for such a group, chances increase for the post-war peace to last and the risk of war recurrence is lowered. If, on the contrary, the state discriminates against the ethnic group politically and economically, the risk of war recurrence rises. In the next section we review the literature on post-war situations to identify factors that have been found to influence whether a country will relapse into war. In the third section we discuss the political relations between ethnic groups following a civil war. From this we derive our key hypothesis ethnic discrimination increases the risk of war recurrence which we test empirically in the fourth section, using survival analysis on a sample of all ethnic wars between 1950 and 2006. We find that political and economic discrimination against the ethnic group that was involved in the war on the opposition side diminishes chances for peace. This finding is robust across different model specifications, including a model with the change in discrimination to address potential issues of endogeneity. Reducing ethnic discrimination and, to a lesser extent, creating institutional guarantees for political inclusion, on the contrary, sharply reduce the risk of war recurrence and increase the expected duration of peace after an ethnic civil war. Group size, group concentration, war outcome and war duration also influence the risks for war recurrence, although these results are not as robust across different model specifications as those for political discrimination. The level of violence during the civil war measured by the number of deaths, number of displaced people and occurrence of genocide does not seem to have a discernible impact on post-war peace duration. The level of economic development, population size, regime type (anocracy) and the presence of a peacekeeping mission also do not seem to influence our outcome variable. Literature Among scholars of post-war environments, one group has emphasized features of the previous war such as war duration, war outcome and casualties as important factors for the post-war peacebuilding process (Doyle and Sambanis, 2000, 2006; Fortna, 2004a, b, 2008; Licklider, 1995; Quinn et al., 2007; Toft, 2010). Others explain the recurrence of war by

Gurses and Rost 471 pointing to the inability or unwillingness of the state to address economic and political grievances (Walter, 2004). Commitment problems constitute a third obstacle to post-war peace. Walter (2002) finds that a lack of credible commitments from outside parties increases the risk of war recurrence. A fourth group of studies has looked at whether the combatants fight along ethnic or religious lines to explain the success or failure of peacebuilding after the end of a civil war (Kaufmann, 1996; Licklider, 1995; Sambanis, 2000). Of these factors, the role of ethnicity in increasing the risk of war starting and recurring has been the subject of much controversy. Some scholars have not found ethnicity to play an important role in the onset of war. They instead emphasize the role of opportunity structures (Fearon and Laitin, 2003) or struggle for resources (Collier and Hoeffler, 2004; Ross, 2006) as the main causes of civil conflict. In a thorough sensitivity analysis, Hegre and Sambanis (2006) find ethnic fractionalization to influence low-level internal conflict but not civil war. Other researchers, however, have identified ethnicity ethnic fractionalization, polarization or concentration as an element not only in the outbreak and duration of civil war but also with regard to sustaining the peace in the aftermath of conflict. Ethnic civil wars are shown to be harder to resolve and end in a negotiated settlement than ideological wars (Cederman and Girardin, 2007; Horowitz, 1985; Kaufmann, 1996; Licklider, 1995; Sambanis, 2001). According to Horowitz (1985: 12), in deeply divided societies, strong ethnic allegiances permeate organizations, activities, and roles to which they are formally unrelated and shape both political and economic activities. Political leaders may exploit ethnic lines or manipulate and even create ethnic affiliations to further their goals. Those who argue that political grievances lie at the core of ethnic conflicts often find that prospects for peace in the aftermath of war are closely tied to how ethnic grievances are addressed (Cederman et al., 2010, 2011; Gurr, 1993; Horowitz, 1985; Sambanis, 2001). One possibility for addressing grievances and providing opportunities for participation is democratization. A democratic political system addresses ethnic minority demands by building institutions capable of channeling these demands into nonviolent forms of participation and competition. A domestic corollary of the democratic peace theory holds that democracy lowers the risk of civil war because democratic institutions and culture provide non-violent means for change (e.g. Hegre et al., 2001). Nevertheless, findings on this proposed link between democracy and conflict are, at best, mixed. Empirical studies on the impact of democracy on civil war onset and recurrence have produced an insignificant (Cederman et al., 2009; Fearon and Laitin, 2003) or a weak and nonlinear relationship (Hegre et al., 2001; Walter, 2004). Institutional democracy, then, is not enough. Empirical studies that include institutional democracy as a factor, often using the Polity measure, do not capture the full picture. Formally democratic systems may still discriminate against certain ethnic groups whose political rights are not respected. The process of democratization, especially if it stalls halfway, may even lead to an increase in instability and violence (Hegre et al., 2001). Political systems that provide opportunities for ethnic groups to participate and make their voices heard may be more important in addressing ethnic grievances. The behavior and attitude of the government toward ethnic groups matter. Regimes that discriminate against the political rights and economic interests of an ethnic group create grievances. Particularly in the volatile post-war situation, ethnic groups may perceive a return to arms as the only way to address their grievances. In the next section, we elaborate on the link between discrimination and post-war peace before subjecting the argument to an empirical test.

472 Conflict Management and Peace Science 30(5) Security dilemma, state and ethnic civil war Arguments for territorial partition as a solution to ethno-nationalist civil wars rely on the logic of the security dilemma to explain why partition is preferable to other solutions. This concept is borrowed from the Realist theory of International Relations and builds upon the absence of a sovereign to explain conflict: in the absence of a sovereign, each ethnic group has an incentive to strengthen its own position, in turn paving the way for preemptive war. The outbreak of conflict among ethnic groups exacerbates this situation, feeds distrust and fear, and generates the perception of a first-strike advantage (Kaufmann, 1996; Posen, 1993). Therefore, according to proponents of partition, security dilemmas created in ethnic war make restoring civil politics in multi-ethnic states shattered by war. impossible. destroy the possibilities for ethnic cooperation. intensify violence, motivate ethnic cleansing, and prevent de-escalation unless the groups are separated (Kaufmann, 1996: 137). Building upon this line of argument, Chapman and Roeder (2007: 679) argue that incompatible identities are the driving force behind nationalist conflicts in countries where the disputed population rejects membership in the larger nation-state, and can better be resolved and managed if ethnic groups are partitioned. Critics of the partition argument, however, have pointed to the difficulties and unrealistic assumptions of creating defensible ethnic enclaves. The complications involved the creation of economically unsustainable small states, the possible transformation of an intrastate into an interstate conflict, disputes among the successor states, disagreements over the borders have led some scholars to conclude that partition should not become a preferred solution to ethnically divided societies. [But it] may be the best option when ethnic groups are so distrustful of one another that they are unable to live together (Rothchild and Roeder, 2005, 12; also see Horowitz, 1985; Sambanis, 2000; Sambanis and Schulhofer-Wohl, 2009; Walter, 2004). Partition has been rare since the Second World War, as most countries have stayed intact after civil wars. Thus we seek to examine factors that increase the chances for peace to last after a war. Atrocities committed during a war seem to make peaceful coexistence impossible, even grotesque, yet we know that it happens all the time (Licklider, 1995: 681). It is possible, we argue, when discrimination is avoided and grievances are addressed. Grievance-based arguments ascribe the root causes of rebellion to a sense of deprivation and injustice, for example when an ethnic group feels disadvantaged or when the state systematically restricts an ethnic group s access to political and economic power (Cederman et al., 2010; Gurr, 1993; Horowitz, 1985). Thus, to the extent that ethnic fears are addressed, the prospects for peaceful co-existence should increase. Similar to Cederman et al. (2010), we see the state play the main role in addressing ethnic grievances. The state has three options in how to treat ethnic groups, in particular the group whose members fought against the government. First, it can actively discriminate against the ethnic group. Second, it can respect the political rights and economic interests of the group. Third, in addition to respecting the rights and interests of a group, it can reach out to the members of the group and create opportunities for political participation, such as the additional layers of governance in federalist systems or proportional representation electoral systems, which offer minority parties institutional guarantees for some level of influence. 2 Posen (1993), in his application of the security dilemma argument, points to disappearing sovereigns as the main factor behind the rise of an environment in which every group has to fend for itself. In some cases, where the central government is too weak to manage

Gurses and Rost 473 interethnic relations (e.g. Lebanon) and the state collapses (Somalia, Afghanistan) or disintegrates (former Yugoslavia), the security dilemma logic may apply, thereby undermining the chances for peaceful coexistence. The government s capacity to manage and mediate ethnic relations varies across post-war situations, yet countries that have experienced ethnic rebellions have rarely collapsed. The state, often as an instrument in the hands of the dominant ethnic group, can substantially alter the dynamics that could instigate an ethnic security dilemma. The state can alleviate fears and uncertainties about the future and thereby prevent the rise of a security dilemma, or raise them by oppressing members of an ethnic group. By engaging in discrimination, the government also signals that it may not be fully committed to maintaining peace with the ethnic group involved in an earlier war. While there may be a role for peacekeepers or other outside forces to contribute to post-war peace (Fortna, 2004a, b; Walter, 2002), it is the state that can greatly shape the chances for the post-war peace by incorporating ethnic demands and redressing the factors that led to the war outbreak. The state can use its power to regulate and mediate among warring ethnic groups, and serve as a guarantor for disadvantaged ethnic groups. Deliberative and carefully crafted polices to create possibilities and institutions for political participation can help overcome fears and grievances including the fear of extinction, a pivotal factor behind ethnic conflicts (Horowitz, 1985) and lay the groundwork for interethnic cooperation. Ethnic discrimination, on the contrary, may lead to renewed ethnic conflict. The establishment of formal power-sharing institutions may create incentives for cooperation and eliminate the fear of betrayal and cheating among different ethnic factions (Hartzell and Hoddie, 2003, 2007; Horowitz, 2008). 3 These institutions encourage the previously excluded group to work within the existing political system. In this way, the aggrieved ethnic group is brought back into the fold and accommodation may lead to peaceful co-existence. Proportional representation voting systems lead to less antagonistic outcomes than winnertakes-all majoritarian systems and allow groups to be politically represented even if they do not constitute a national or regional majority. Proportional representation, particularly in ethnically divided societies, can help alleviate fears of political dominance by the larger ethnic group over the minority (Lijphart, 1977; Reynal-Quorel, 2002). Similarly, federalism allows ethnic groups a certain degree of self-governance and alleviates fears of a system that is dominated from the center (Horowitz, 2008). Especially for geographically concentrated ethnic groups those that are a national minority but a regional majority federalism allows for political participation and even a degree of self-rule. The use or avoidance of ethnic discrimination in the aftermath of a civil war may be an indicator of an underlying government strategy to confront and exclude an ethnic group, or to accommodate it to avoid further conflict. Yet, by focusing on political and economic discrimination against ethnic groups and change in ethnic discrimination, while also examining political institutions such as federalism and the electoral system, our argument should capture most (and the most important) aspects of such a policy. Another concern is endogeneity: what if discrimination is a consequence of ethnic tensions and a result of civil war, rather a factor that contributes to war recurrence? In robustness tests, described in detail below, we include the change in discrimination from before the civil war to the post-war period, rather than just post-war discrimination. Thus, we can better discern whether ethnic discrimination is merely part of a protracted situation of ethnic tensions, or whether changes in government policy increasing or softening ethnic discrimination after a civil war help to keep the peace. Some may also argue that the use or avoidance of ethnic discrimination is merely a sign of the existence or absence of underlying tensions between ethnic groups. Yet,

474 Conflict Management and Peace Science 30(5) given that all the ethnic groups we examine in our analysis have just fought a civil war against the government, there are by definition strong ethnic tensions in all cases. Still, the government has a choice whether to employ ethnic discrimination in the aftermath of a war or to avoid it. Two examples may help to explain this choice that governments face in the aftermath of an ethnic war. Even in tense situations characterized by ethnic hostilities, governments can choose to lessen discrimination. At the same time, they can decide to discriminate against ethnic groups in situations that might otherwise seem amenable to post-war reconciliation. First, the Ibos in Nigeria an ethnic minority that primarily lives in the northeast became subject to harsh political and economic discrimination following Nigeria s independence in 1960. Members of this ethnic group had faced serious collective violence even before independence, and after two massacres in 1966 the Ibo embarked on a secessionist war against Nigeria (Horowitz, 1981: 179). The bloody civil war that started in 1967 lasted four years and ended in the government s military victory. According to the Minorities at Risk (MAR) project, the Ibos faced a high level of political and economic discrimination before and during the war (the MAR project assigns a 4, the highest score, for both). Despite initial discrimination, the bloody civil war that ensued and a government military victory over this secessionist rebellion, the war has not recurred. The status of the Ibos began to improve following the end of the war and the ethnic discrimination they faced before and during the war began to decline, down to 3 (social exclusion/neutral policy) in 1971, until reaching 0 (no discrimination) in 1999. Thus, the Ibos gradually faced lower ethnic discrimination after the war. On the contrary, the Kurds in Turkey rebelled against discriminatory and repressive state policies in 1984 and were defeated militarily in 1999. Following the defeat of the Kurdish insurgent group (the Kurdistan Workers Party, known by its Kurdish initials, PKK) on the battlefield, peace lasted only six years. Despite some improvements in the status of the Kurdish minority in Turkey in the early 2000s, overall discrimination against the members of this ethnic group remained high in the post-war years between 1999 and 2005. The political and economic discrimination against the Kurds before and during the war did not see any change following the end of the armed conflict in 1999. 4 The war broke out again in 2005 and it has been intensifying since then. Although in both cases, rebellions were started with the goal of seceding from the state, no partition occurred. In Nigeria, discrimination against the Ibos was ended and the war did not recur; in Turkey discrimination continued and the war recurred. If states have a choice of whether to use discrimination in the aftermath of a civil war, why would they ever choose to discriminate? Trying to answer this question is beyond the scope of this study. However, studies on why governments choose to repress (e.g. Pierskalla, 2010) find that states use repression to exclude groups from political and economic power. Walter (2006a) finds that states choose not to accommodate secessionist ethnic groups also in order to build a reputation among other ethnic groups. Governments must see at least some chance of their policies succeeding in solidifying their grip on power rather than leading to a new civil war, otherwise they would not choose repression. While the same may be true for ethnic discrimination, in this study we cannot answer the question of why and under which circumstances governments discriminate. To summarize, we argue that the state can lay the groundwork for a peaceful coexistence by using careful policies and by creating institutions to bring the warring ethnic group back to the political arena. The absence of ethnic discrimination in post-war situations can help

Gurses and Rost 475 avoid a recurrence of the fighting, while providing opportunities for political participation can lay the groundwork for a sustainable peace. Therefore, the hypothesis that we propose links the chances for a lasting peace to the post-civil war level of discrimination. Specification and data Our sample includes all ethno-nationalist civil wars that began and ended between 1950 and 2006. To make our results comparable to other studies on ethnic civil wars (e.g. Chapman and Roeder, 2007; Sambanis, 2000, 2001; Sambanis and Schulhofer-Wohl, 2009) we primarily rely upon the war list from Sambanis (2004). We focus on the dyadic relationship between an ethnic group and the state; hence our sample includes cases that involve an ethnic group engaging in armed conflict with its own government. To do so, we consulted Chapman and Roeder (2007), 5 although we exclude cases that did not have a clear ethnic component. 6 Further, we update the list by coding war resumptions for two ethnic conflicts that recurred after 2004. 7 The dependent variable is the risk of war recurrence during each year after the end of an ethnic civil war. The lower the yearly risk of war recurrence, the longer the expected duration of post-war peace. There are 70 ethnic civil wars 8 in the dataset, 12 of which ended in partition. 9 We do not make any normative claims for or against partition and do not question that under specific circumstances it can be a solution to ethnic conflict. We exclude those cases that ended in partition for two main reasons. First, these cases do not show any variation in terms of peace failure (our dependent variable) and post-war discrimination (our primary explanatory variable), thus making it difficult to obtain any meaningful and robust relationship between these two variables. Second, and perhaps more importantly, we focus on the effect of post-war discrimination on the risk of war recurring. Thus our analysis focuses on those cases where the ethnic groups continued living within the same territorial boundaries, and our results only apply to countries that did not partition after an ethnic war. In fact, the underlying process leading to partition might show qualitative differences from the cases in which state borders remained intact after the end of the conflict. The argument and evidence presented in this study do not apply to those cases that resulted in physical partitioning of warring groups or cases that may be viable candidates for partition. We believe that this is an equally important question that requires additional attention and research. We were able to include 56 cases where countries stayed intact. The duration of post-war peace in these cases varies widely, begging the question as to what explains this variation. Of these 56 cases, 25 failed (i.e. war recurred) during the timeframe under analysis. The peace duration ranges from 1 year to 47 years with a mean of 10.88 and a standard deviation of 9.52 years. Explanatory variables The level of political and economic discrimination in the aftermath of the war is the key explanatory variable. Several studies have used the Polity composite democracy index, which gages institutional aspects of democracy on a scale from 210 (least democratic) to þ 10 (most democratic), to examine the effect of democracy on civil war. While the Polity measure provides a good picture of the overall level of institutional democracy, it suffers from an ad hoc aggregation rule (Treier and Jackman, 2008: 202) and non-continuous scores, which can be arrived at through different combinations (Gleditsch and Ward, 1997; Vreeland,

476 Conflict Management and Peace Science 30(5) 2008). More importantly, this measure s emphasis on the institutional aspects of a polity fails to identify an ethnic group s grievances. Any notable progress or deterioration in an ethnic group s position within the system without observable institutional changes would remain hidden. For instance, the Kurds relationship with the Iraqi regime has fluctuated from being peaceful to being violent and repressive, while the political regime has remained authoritarian. The war in the late 1960s and the negotiations that ensued between the Iraqi government and the Kurds generated a relatively free and peaceful environment in the following years, while the Polity score for Iraq sank from 25 in1967to27 in 1968 and remained so until 1979. Thus the Polity index fails to capture the dynamic relationship between different groups (Gurses, 2010). Since we try to capture such dyadic relationships between an ethnic group and the state, we opt to use the data provided by the Minorities at Risk Project (2009). Ethnic discrimination, for the purpose of this study, is defined as restricting a person s political rights or curtailing their economic interests on the basis of that person s ethnic affiliation. The MAR project monitors and assesses the overall levels of political and economic discrimination against ethnic groups on two separate scales, each coded on a yearly basis and ranging from 0 (no discrimination) over 1 (neglect/remedial policies), 2 (neglect/no remedial policies), 3 (social exclusion/neutral policy) to 4 (exclusion/repressive policy). 10 We use data on the political and economic discrimination an ethnic group faces following the end of the war to assess the extent to which ethnic grievances are addressed in the aftermath of the war. 11 As a combined measure of overall post-war discrimination and to avoid problems of multi-collinearity we use the time-varying mean of political and economic discrimination. 12 The combined post-war discrimination variable ranges from 0 to 4 with a mean of 2.19 and a standard deviation of 1.31. The mean of this variable for those cases that failed during the time period under analysis is 2.94, whereas it is 1.27 for post-war periods that lasted, indicating some initial evidence for our hypothesis (Pr T. t =\0.0001). We take a first look at the bivariate relationship between our discrimination variable and the recurrence of war in Table 1 with the figures rounded to integers. The table includes all 885 post-civil war years (fewer are included in the multivariate analyses that follow owing to missing data). A higher level of discrimination is associated with a higher risk of war recurrence; with every one-point increase on the discrimination variable, the risk of war increases substantially, increasing by a factor of 21 from situations with no discrimination to situations with high discrimination. We include two variables to test whether political institutions that provide ethnic groups with a possibility of political participation are linked to a reduced risk of war recurrence. Norris s (2009) Democracy Time-Series Dataset provides data on federalism, based on Watts (1998). Golder (2005) provides data on electoral systems from which we construct a binary variable for countries with at least some elements of proportional representation electoral systems. We have updated both variables, as they did not cover the full time period we study, and have transformed them into dichotomous variables. Control variables Discrimination may be related to other factors, such as war outcome or economic development. To avoid risks of possible endogeneity, we employ a number of control variables on characteristics of the war, country and group that have been found to be associated with peace building after civil war.

Gurses and Rost 477 Table 1. Ethnic discrimination and war recurrence Ethnic discrimination Peace continues War recurs Total Low 0 277 1 278 99.6% 0.4% 1 131 2 133 98.5% 1.5% 2 139 3 142 97.9% 2.1% 3 163 8 171 95.3% 4.7% High 4 146 13 159 91.8% 8.2% Total 856 27 885 post-civil war years 96.9% 3.1% War outcome. Scholars have pointed to the effect of war outcome on post-war peace as it often leads to a redistribution of political and economic resources among adversaries (Doyle and Sambanis, 2000, 2006; Fortna, 2004a, b; Hartzell et al., 2001; Hartzell and Hoddie, 2003; Licklider, 1995; Quinn et al., 2007; Toft, 2010) and influences post-war democracy (Gurses and Mason, 2008; Joshi, 2010; Wantchekon, 2004). Following the literature, we define three outcomes: rebel victory, government victory and settlement, using data from Doyle and Sambanis (2000) and Sambanis (2004). Of the 54 cases that we were able to include in our full model (model 2 in Table 3), three ended in rebel victory, 25 ended in government victory and 26 resulted in a settlement. War costs. While some have argued that war duration leads to war weariness, thereby helping to sustain the peace (Sambanis, 2000; Walter, 2004), the ethnic security dilemma and partition arguments would expect ethnic positions to harden during long wars, making war recurrence more likely. We use duration in years to put this argument to a test. To measure the intensity of the war we employ a set of variables that are used in other studies. Since we use Sambanis s (2004) definition of civil war, we primarily rely on Doyle and Sambanis s (2000) and Sambanis s (2004) coding notes to collect data on logged total deaths. Where these data are not available we consult the Uppsala Conflict Data Project (2010). We use the logged total number of refugees and internally displaced people to measure forced displacement. Refugees and internally displaced persons may have a destabilizing effect on post-war peace, as their return can create enormous social, economic and political strains. Data on displacement are drawn from Moore and Shellman (2004), and where their data was missing, from Doyle and Sambanis (2000). The genocide variable is a dichotomous variable and is set equal to 1 for those wars during which a genocide occurred, 0 otherwise (Harff, 2003; Political Instability Task Force, 2010). Country-level control variables. Studies have shown that civil wars are less likely in highly democratic and highly autocratic regimes, making regimes that fall in between more susceptible to war onset (Hegre et al., 2001; Reynal-Quorel, 2002; Sambanis, 2001). Thus, following Wimmer et al. (2009), we define a variable using Polity IV data (Marshall and Jaggers, 2009)

478 Conflict Management and Peace Science 30(5) to distinguish those regimes with some democratic elements but that are not full democracies (i.e. anocracies; those that fall between 26 and þ 6 on the Polity scale). Population and income have been shown to have powerful effects on initial war onset and peacebuilding measures (Fearon and Laitin, 2003; Hegre and Sambanis, 2006). Therefore, we include logged real GDP per capita and logged population size as controls, with the data taken from the Penn World Tables. Group characteristics. Group size and group concentration influence the decision to go to war by lowering the opportunity cost for rebellion (Cederman et al., 2010; Walter, 2006b). We use MAR data for the proportion of the ethnic group in question out of the total population and the degree of geographical concentration to control for this argument. The latter is measured on a scale of 0 3, with higher numbers indicating higher levels of concentration. Peacekeeping. Several studies have argued that the presence of peacekeepers helps sustain the peace in post-war countries (Doyle and Sambanis, 2000, 2006; Fortna, 2004a, b; 2008; Quinn et al., 2007; Sambanis, 2008; Walter, 2002). Relying on Mullenbach s (2005) dataset, we include a dichotomous variable, coded 1 for years during which any type of peacekeeping mission was present. The descriptive statistics are presented in Table 2. Analysis Given the outcome variable, peace duration in years, we estimate a series of survival models to test our hypothesis. The unit of analysis is the post-ethnic war country year and the models assess the risk of the peace failing in a given year. We start by employing a Cox regression model. 13 While some variables, such as war duration or outcome, are static over the postwar period, the values for the explanatory variable and several of the control variables (e.g. population and GDP per capita) vary from year to year. Parametric models, such as Weibull regression, better control for such variation regardless of whether other failures occur in the same interval (Cleves et al., 2002: 182). Therefore, to control for the possibility that the baseline hazard may follow a monotonic (Weibull) or a non-monotonic (log normal or log logistic) shape, as a robustness check, we analyze the data using parametric survival models (Box- Steffensmeier and Jones, 2004; Cleves et al., 2002). 14 We opted for using Weibull regression, as one of the most widely used survival models in the literature, for robustness checks. The results are presented in Table 3. The results indicate a strong support for our hypothesis: high levels of discrimination endanger peace in the aftermath of ethnic civil wars. This finding is robust across all models. Model 1 shows the results from our basic model with only some confounding factors. Model 2 displays the results from the full model, using Cox regression. Adding controls for war, country and group characteristics does not lead to a substantial change in the key finding on the association between discrimination and post-war peace duration. Every unit increase in the level of discrimination increases the risk for peace failure by 152% in model 1 and by 179% in model 2. Results from a Weibull regression are substantially the same (model 3). An increase in the post-war level of discrimination increases the risk for war resumption by 191%. This effect is visualized in Figure 1, which is based on model 3 and where the postwar discrimination variable varies from 0 to 4, while holding other variables at their

Gurses and Rost 479 Table 2. Summary statistics for the independent variables Variable N Minimum Maximum Mean Standard deviation Post-war discrimination 600 0 4 2.20 1.32 Change in post-war discrimination 609 23 4 0.67 1.25 Government victory (dummy) 600 0 1 0.60 0.49 Rebel victory (dummy) 600 0 1 0.06 0.23 Settlement (dummy) 600 0 1 0.35 0.48 GDP per capita (logged) 600 5.87 10.27 7.73 0.83 Population (logged) 600 6.4 14.09 10.55 1.84 Anocracy (dummy) 600 0 1 0.29 0.45 War duration 600 1 36 6.12 6.07 Casualty (logged) 600 6.7 14.51 10.15 1.95 Forced displacement (logged) 600 0 15.53 7.87 6.22 Genocide (dummy) 600 0 1 0.47 0.49 Peacekeeping (dummy) 600 0 1 0.097 0.30 Group concentration 600 0 3 2.27 1.09 Group proportion 600 0.003 0.83 0.17 0.18 Federalism (dummy) 600 0 1 0.27 0.45 Proportional representation (dummy) 600 0 1 0.09 0.29 Post-War Discrimination and Peace Duration.1.2.3 Hazard function 4 3 0 2 1 0 0 10 20 30 40 50 analysis time postwar discrimination=0 postwar discrimination=2 postwar discrimination=4 postwar discrimination=1 postwar discrimination=3 Figure 1. Post-war discrimination and peace duration. mean values. The hazard, as displayed in Figure 1, sharply increases as discrimination intensifies. To investigate the role of political institutions, we add variables for federalism and proportional electoral systems to our model. As shown in model 5, federalism does not seem to have a statistically significant impact. Proportional representation voting systems help to

480 Conflict Management and Peace Science 30(5) Table 3. Event history analysis for the duration of peace after ethnic civil war Variable Model 1 (Cox) Model 2 (Cox) Model 3 (Weibull) Model 4 (Cox) Model 5 (Cox) Post-war discrimination 2.52*** 2.79*** 2.91*** 3.82*** [0.82] [1.05] [1.27] [1.93] Change in post-war discrimination 0.40*** [0.12] Government victory 0.13* 0.13* 0.17** 0.10** 0.05* [0.16] [0.16] [0.13] [0.11] [0.08] Settlement 0.28 0.27 0.40 0.35 0.20* [0.29] [0.25] [0.25] [0.29] [0.19] GDP per capita (logged) 0.98 1.05 0.94 0.84 1.46 [0.21] [0.22] [0.21] [0.19] [0.48] Population (logged) 0.92 0.96 0.89 0.86 0.82 [0.08] [0.15] [0.17] [0.14] [0.12] Anocracy 1.38 1.59 1.36 0.82 0.59 [0.55] [0.76] [0.64] [0.42] [0.25] War duration 0.93 0.94 0.92* 0.90* [0.05] [0.06] [0.04] [0.05] Casualty (logged) 0.92 0.94 1.18 0.87 [0.15] [0.17] [0.21] [0.17] Forced displacement 1.04 1.06 1.00 1.06 [0.06] [0.07] [0.06] [0.06] Genocide 0.72 0.50 0.93 0.76 [0.53] [0.49] [0.63] [0.58] Peacekeeping 0.57 0.48 1.26 0.82 [0.55] [0.52] [0.85] [0.82] Group concentration 1.93*** 1.89*** 1.67*** 2.45*** [0.39] [0.41] [0.35] [0.61] Group proportion 32.57** 28.48** 8.02 41.56* [53.61] [47.59] [10.75] [85.42] Proportional representation 0.02*** [0.02] Federalism 0.52 [0.30] Number of subjects 56 54 54 54 53 Observations 651 609 609 609 600 Wald x 2 (13) 43.71 37.60 50.50 64.14 160.33 Probability. x 2 \0.0001 0.0003 \0.0001 \0.0001 \0.0001 Note: Hazard ratios are reported. Robust standard errors clustered on country in parentheses. *** Significant at 0.01 level; ** significant at 0.05 level; * significant at 0.10 level (two-tailed). The estimates were obtained using Stata 10.1. increase the duration of post-war peace, but the results depend on whether we include mixed systems. Using a strict definition of proportional systems, the variable is not significant; including mixed systems, as shown here, it is. The finding on post-war discrimination variable holds. Robustness tests The key explanatory variable, post-war discrimination, measures discrimination in each year of the post-war period, but it does not take into account the pre-war discrimination. To

Gurses and Rost 481 investigate this issue further, we look at the change in the level of discrimination from the pre-war level to the post-war level. We subtract the level of discrimination in the year before war onset from the post-war level of discrimination. This variable measures whether there was an improvement or deterioration in an ethnic group s status. As shown in model 4, improvements in a group s status significantly lower the risk for war resumption. Every unit decrease in the post-war level of discrimination relative to pre-war level reduces the risk for peace failure by 60%, vindicating the findings obtained earlier. Further, we checked for the possibility of changes in the level of discrimination an ethnic group faces during (rather than before and after) the war. The overall level of discrimination an ethnic group faces tends to remain stable during the war. The mean discrimination for 39 cases (out of 70 cases in the sample) remained the same compared with the pre-war level of discrimination. Of the remaining 31 cases, 20 experienced a decline in the level of discrimination and 11 had an increase. The Baluchis in Pakistan, for instance, faced a slightly increased discrimination during the war years (1973 1977) compared with the pre-war level in 1972 (2.5 vs 2). The mean discrimination the Somalis in Kenya faced during the war (1963 1967) decreased from 3 to 2.8. To control for the change in the level of discrimination during the war, we generated a variable by looking at the change between the level of discrimination during and in each year after the war. The analysis produced very similar results to those presented in model 4. As a further robustness analysis and to check whether excluding partition cases introduces a selection bias, we analyze the data by adding those cases that ended in the physical separation of warring groups. Of the 12 cases that were coded as separated, none failed during the time period under analysis. Partition, compared with those cases that ended in rebel victories (our reference category), is statistically significant yet the finding on partition is sensitive to model specification because it is a perfect predictor of war recurrence (results not shown). We ran our models without the Polity variable for the level of democracy, to check whether the variable or its interaction with the discrimination variable influences our results, and obtained substantially similar results. Further, we re-ran models 1 5 with the discrimination variable lagged by one year to address concerns about possible endogeneity. The results were substantially similar: ethnic discrimination is associated with a high risk of war recurrence. An important number of ethnic groups have engaged in multiple civil wars against their governments. Thus, treating these observations as if they were independent could influence the estimates for variance and coefficients (Box-Steffensmeier and Jones, 2004: 155). To give an example, the Chittagong Hill Tribes engaged in one civil war against Bangladesh whereas the Hutus in Burundi rebelled three times for the period under analysis. To control for this, we generated a count variable and then stratified the estimation by this variable to allow each strata to have its own baseline hazard rate while coefficients are restricted to be the same across strata (Box-Steffensmeier and Jones, 2004: 160). The results from this stratified Cox regression (not shown) were substantially similar to those reported in in Table 3. Finally, to account for possible unobserved heterogeneity in the data we conducted robustness checks using frailty models. Unobserved heterogeneity can result owing mainly to subgroups or individual cases in the sample with distinct failure rates. In other words, some observations or groups of observations might be inherently more likely to experience failure (i.e. war recurrence) than others, hence influencing the hazard rate for remaining cases in the sample (see, for instance, Zorn, 2000). We analyzed model 3 of Table 3 using g

482 Conflict Management and Peace Science 30(5) and inverse Gaussian frailty models. These models showed no evidence of unobserved heterogeneity and produced substantially the same results as reported in Table 3. 15 Control variables Of the control variables, the findings on war outcome indicate that, compared with rebel victories, government victory lowers the risk for war resumption. 16 Some have argued, on the contrary, that government victories are often little more than a lull in the fighting (Quinn et al., 2007; also see Toft, 2010). Settlements, too, seem to help sustain the peace; yet this effect is noticeable in model 5 only. 17 Although there is no consensus on the effect of war outcomes, an important number of studies point out the effect of outcomes on the chances of sustaining the peace. A number of studies have stressed the precarious nature of negotiated settlements as the protagonists fighting capabilities remain intact (Atlas and Licklider, 1999; Licklider, 1995; Wagner, 1993). Others have argued that this effect is contingent upon the dimensions of power-sharing incorporated into the terms of agreement (Hartzell and Hoddie, 2003, 2007), or whether peacekeeping forces are available to back up the peace (Quinn et al., 2007; Walter, 2004). Mattes and Savun (2009) find that political power-sharing provisions are paramount in explaining the variation in peace duration after civil war settlements. Toft (2010) concludes that rebel victories are more likely to produce durable peace and lead to democracy. Government victory, Quinn et al. (2007; also see Mason et al., 2011) argue, compared with rebel victory, is more likely to preserve the condition of dual sovereignty (Tilly, 1978), and thus should be followed by a less durable peace. Kreutz (2010), on the contrary, finds that government victories produce more durable peace than rebel victories. Fuhrmann and Tir (2009) find that one-sided military victories, regardless of which side wins, prolong peace. Mukherjee (2006) points to an interactive relationship between war outcome and power-sharing agreements and finds no direct effect of government or rebel victory on the duration of peace after the war ends. As such, while we acknowledge that other studies have found war outcome to play an important role, owing mainly to our sample and model specification, the results may not necessarily be comparable to those found by others, making us cautious to draw clear conclusions about the influence of war outcome on the duration of peace. The causal link between war outcome and post-war discrimination deserves more attention, yet we believe this is a separate research question for future studies. We do not seek to examine post-war discrimination as a function of war outcome. Rather we seek to examine whether a decline in post-war level of discrimination improves the chances of sustaining the peace in the aftermath of civil war. The finding on war duration, although weakly significant in model 5 only, is consistent with the finding on this variable in the literature; long wars prolong the peace. Other variables that are used as measures of war cost (casualties, displacement, genocide) are not significant, indirectly indicating the possibility of a peaceful coexistence even after bloody ethnic civil wars. Population size and economic development are two measures that are useful for distinguishing between countries that generally are at a high risk of civil war and those that have a low risk. However, once a civil war has occurred, these measures are too broad to distinguish countries that are at risk of war recurrence from those countries with a good chance for lasting peace. The anocracy variable seems to have no noticeable effect on the peace duration, a

Gurses and Rost 483 finding that confirms Cederman et al. s (2009) results. In contrast to Doyle and Sambanis (2000, 2006) and Fortna (2004a, b), we did not find a statistically significant impact for peacekeeping in our sample. Yet, as we do not specifically examine the relationship between peacekeeping and war recurrence, we leave a closer examination of this non-result similar to those for war outcome and anocracy for future, more detailed research. Moreover, our sample includes ethnic civil wars only. The insignificant findings on variables that the existing literature has found to affect the risk of civil war (e.g. GDP per capita and peacekeeping) could be due to this difference in the sample. Thus, the results from this study do not necessarily invalidate the earlier findings on the link between economic development and conflict but rather point to potentially different dynamics at work for civil wars that are fought along ethnic identities. The measures used to control for group characteristics are associated with a higher risk of war recurrence. Group concentration is highly significant across all four models in which it is included; group proportion is statistically significant in three of the four models. It is easier for larger, more concentrated groups to mobilize resources and to return to war if they feel that the government does not respect their rights. To summarize, to the extent the government addresses ethnic grievances, chances for peace increase. The war s characteristics, such as casualties, duration and outcome, that others have found to be of critical importance in shaping the prospects for the post-civil war peace, cease to be the chief variables. What happens after the war seems to be of greater importance than what happened during the war and how the war ended. Conclusion We have argued that it is possible for ethnic groups to live together in one country following the end of an ethnic war. Even after particularly violent civil wars, peace between ethnic groups is possible. What seems to matter more, from our analysis, is what happens after the war ends. Political and economic discrimination, in our models, is the most robust factor in influencing whether the peace will last. Discriminatory state policies against aggrieved ethnic groups and exclusionary institutions sharply increase the risk of war to recur. Repressive state polices not only enhance the risks of initial war onset but also depress the chances for sustaining the peace afterwards. The results are in line with Cederman et al. s (2011: 492) conclusion that unless the claims of marginalized groups are taken seriously it is unlikely that such conflicts can be understood or solved. Civil conflict can deepen ethnic tensions and reinforce political ethnic identities. Yet, ethnic identity is essentially a fluid construct that is subject to social processes (Anderson, 2006; Smith, 1991). Thus, the government can alleviate the fear of ethnic groups with carefully targeted policies and institutions that lay the groundwork for peaceful coexistence. The results show that avoiding discrimination and, to a lesser extent, putting in place institutional guarantees for political participation may contribute to reducing the risk of a new war. Such polices, in the long run, can help to de-politicize ethnicity in such a way that it loses its salience as a motive for ethnic rebellions. If, on the contrary, the government discriminates against members of the ethnic group that has been involved in the civil war, chances for the peace lasting fall and the risk of war recurring increases. To conclude and illustrate the main result that governments in post-war situations should avoid ethnic discrimination and give ethnic groups a political voice we generate