Ending Civil Wars A Case for Rebel Victory? Monica Duffy Toft Since the end of World War II, policymakers have shown a marked preference for settling civil wars through negotiated settlements. The core recommendation of this policy is to employ third-party resources—primarily in the form of economic incentives and good offices—to halt the violence and preserve the combatants. Scholars of civil wars, for their part, have devoted the bulk of their analyses to exploring how best to achieve negotiated settlements. In recent years, however, other scholars have introduced a counterargument. Supporters of this "give war a chance" option advocate allowing belligerents to continue fighting until one side achieves a military victory.1 A survey of the literature on civil war termination makes clear that, of the two groups, the negotiated settlements camp is far more pervasive and influential than the "give war a chance" camp.2 The logic of both arguments is compelling, but flawed. The negotiated settlements camp takes it as axiomatic that the sooner violence is halted, the greater the collective benefit. The "give war a chance" camp goes too far in the other direction, assuming that it is axiomatic that violence unfettered is the best path to lasting peace. Which argument is stronger and, more importantly, under what circumstances? Although since 1990 the preferred means for ending civil wars has been negotiated settlements, these have proven largely ineffective: civil wars ended by negotiated settlement are more likely to recur than those ending in victory by one side. Although the "give war a chance" argument appears more Monica Duffy Toft is Associate Professor of Public Policy at Harvard University's John F. Kennedy School of Government. She is the author of The Geography of Ethnic Violence: Identity, Interests, and the Indivisibility of Territory (Princeton University Press, 2003) and Securing the Peace: The Durable Settlement of Civil Wars (Princeton University Press, 2010). The author would like to thank Ivan Arregum-Toft, Page Fortna, Roy Licklider, Nicholas Sambanis, Stephen Saideman, the anonymous reviewers, and the participants at seminars at the University of Chicago, Columbia, Harvard, Notre Dame, and Yale for comments and helpful suggestions on earlier drafts. Outstanding research assistance was provided by Jeffrey Friedman and Janet Lewis. 1. See Edward N. Luttwak, "Give War a Chance," Foreign Affairs, Vol. 78, No. 4 (July/August 1999), pp. 36-44. Hereafter, when I use the term "victory," I am referring to military victory, the defeat of a rival in war. 2. This claim is supported by a survey of articles published in seven leading international relations and comparative politics journals—American journal of Political Science, American Political Science Review, International Organization, International Security, journal of Conflict Resolution, Security Studies, and World Politics—from 1990 through 2005. International Security, Vol. 34, No. 4 (Spring 2010), pp. 7-36 © 2010 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology. 7 International Security 34:4 8 strongly supported by a tally of historical outcomes, a closer look reveals that the proposed causal mechanism does not operate as expected. For example, proponents of this argument have no explanation for why civil wars ending in rebel victory produce the most durable settlements. In this article, I argue that success in civil war termination—that is, ending the violence and establishing the political space for enduring peace—requires a balance of carrots and sticks. Especially since the end of the Cold War, negotiated settlements have emphasized the benefits of cooperation while downplaying the threat of punishment should either or both parties fail to live up to their obligations under the agreement.3 At the same time, the combination of the proliferation of weak states, refinements in insurgency strategy, and the wide distribution of small arms has made it relatively more difficult for even well-supplied and well-led combatants to achieve victory. Meanwhile, the damage from civil wars has become more difficult to contain, as guns and fighters flow across borders and disrupt trade and the domestic politics of neighboring states. The subject of effective civil war termination is important for three reasons. First, with regard to theory, the "give war a chance" argument forces scholars and policymakers to confront how they should think about the costs and consequences of war. If one measures the collective good in terms of a lasting peace, a systematic and general reduction in the destructiveness of war, and robust development, then, all else held equal, the "give war a chance" argument must be taken seriously.4 Second, civil wars are highly destructive. Yet they have traditionally been less subject to regulation and limitation by treaties such as the Hague and Geneva Conventions than have interstate wars. Until 1977, when the Additional Protocols to the Geneva Conventions of 1949 protecting "national liberation movements" came into force5—governments were not restricted in the amount or nature of force they could use to defeat rebels.6 Moreover, many 3. For an analysis of implementing negotiated settlements following three civil wars whose main objective was control over the central government (i.e., not secession), see Dorina Akosua Oduraa Bekoe, "After the Peace Agreement: Lessons for Implementation from Mozambique, Angola, and Liberia," Ph.D. dissertation, Harvard University, 2002. 4. Richard Falk, "Hard Choices and Tragic Dilemmas," Nation, December 20, 1993, p. 755. 5. "Protocol Additional to the Geneva Conventions of 12 August 1949, and Relating to the Protection of Victims of Noninternational Armed Conflicts (Protocol II), 8 June 1977," International Committee of the Red Cross homepage, http://www.icrc.org/IHL.NSF/FULL/47570penDocument. 6. See G.I.A.D. Draper, "Wars of National Liberation and War Criminality," in Michael Howard, ed., Restraints on War: Studies in the Limitation of Armed Conflict (Oxford: Oxford University Press, 1979), pp. 135-162. Ending Civil Wars 9 civil wars escalate to interstate wars, either by spilling across state borders or by provoking external intervention.7 Third, policymakers exert considerable effort to finding ways to advance democratic institutions and rehabilitate the economy once a war has ended.8 Therefore, knowing which postwar environments are most likely to flourish as democratic polities with liberal market conditions and which are more likely to succumb to authoritarianism, corruption, or the resumption of war is crucial. In the following section, I review the literature on civil war termination. Next, I introduce an empirical survey of the relationship between the type of effort applied to end a civil war and long-term outcomes. I then lay out my explanation and the policy implications that follow from it. Literature on Civil War Termination The idea that how a civil war ends may determine the postwar outcome is not new. In slightly different ways, both Edward Luttwak and Robert Wagner have argued that allowing wars to "reach their natural conclusion" enhances the likelihood of a durable peace and effective postwar reconstruction.9 In "Give War a Chance," Luttwak argues that "the transformative effects of both decisive victory and exhaustion are blocked by outside intervention."10 Aside from a few problematic historical cases, however, Luttwak does not explain what he means by "transformative effects" or "exhaustion." In the "Causes of Peace," Wagner claims that victories are more stable than negotiated settlements because the loser's capacity to reignite the war should be low.11 Although this is a sound hypothesis, Wagner does not test it. In a later article, Roy Licklider, who did subject the hypothesis to a statistical test, found some support for it.12 Despite the importance of Licklider's finding, security studies scholars have not tested its robustness or offered a general explanation that supports it. The 7. Stephen Ryan, Ethnic Conflict and International Relations (Aldershot, U.K.: Gower, 1990). 8. See, for example, Peter Harris and Ben Reilly, eds., Democracy and Deep-Rooted Conflict: Options for Negotiators (Stockholm: International IDEA, 1998). 9. Luttwak, "Give War a Chance"; and Robert Harrison Wagner, "The Causes of Peace," in Roy Licklider, ed., Stopping the Killing: How Civil Wars End (New York: New York University Press, 1993), pp. 235-268. 10. Luttwak, "Give War a Chance," p. 44. 11. Wagner, "The Causes of Peace." 12. Roy Licklider, "The Consequences of Negotiated Settlements in Civil Wars, 1945-1993," American Political Science Review, Vol. 89, No. 3 (September 1995), pp. 681-690. International Security 34:4 10 title of Licklider's 1995 article itself suggests why: although victory is the result in the majority of cases and it produces some interesting consequences, only negotiated settlements and their ramifications are mentioned. Victory is reduced to an analytically unimportant role. The scant literature on how different termination types influence postwar outcomes has yet to capture the strong empirical relationship between rebel victory and postwar stability that I establish here. Research on the outbreak of civil wars and on patterns of violence has increasingly centered on the behavior and activities of rebel movements: it has probed the conditions that favor the success of rebel movements;13 the resources and recruitment involved in rebel movements;14 and the decisionmaking process that rebels use when deciding whether to fight or to quit.15 Few scholars, however, have considered the possibility that victorious rebel movements display relatively common strategies in the aftermath of civil war, with important ramifications for postwar stability.16 Scholars and practitioners generally believe that if former combatants are given a voice in their political, economic, and social destiny, renewed violence can be averted. This factor alone may explain the literature's focus on negotiated settlements, because allowing former combatants a say in the postwar configuration of the government increases the likelihood that a democratic process will lead to the creation and strengthening of democratic institutions.17 13. T. David Mason, Joseph P. Weingarten Jr., and Patrick J. Fett, "Win, Lose, or Draw: Predicting the Outcome of Civil Wars," Political Research Quarterly, Vol. 52, No. 2 (June 1999), pp. 239-268; and James D. Fearon and David D. Laitin, "Ethnicity, Insurgency, and Civil War," American Political Science Review, Vol. 97, No. 1 (February 2003), pp. 75-90. 14. Jeremy M. Weinstein, "Resources and the Information Problem in Rebel Recruitment," Journal of Conflict Resolution, Vol. 49, No. 4 (August 2005), pp. 598-624. 15. Paul Collier, Anke Hoeffler, and Mans Soderbom, "On the Duration of Civil War," Journal of Peace Research, Vol. 41, No. 3 (May 2004), pp. 253-273; and Karl R. de Rouen Jr. and David Sobek, "The Dynamics of Civil War Duration and Outcome," Journal of Peace Research, Vol. 41, No. 3 (May 2004), pp. 393-320. 16. For a review of the theoretical literature on civil war outcomes, see Monica Duffy Toft, Securing the Peace: The Durable Settlement of Civil Wars (Princeton, N.J.: Princeton University Press, 2010), chaps. 1, 2. For a review of some of the main statistical findings, see T. David Mason, "Sustaining the Peace after Civil War" (Carlisle, Pa.: Strategic Studies Institute, U.S. Army War College, December 2007). 17. Matthew Hoddie and Caroline Hartzell, for example, conclude that out of sixteen peace agreements signed between 1980 and 1996, those that completely implemented provisions for military power sharing among former combatants held the greatest prospects for maintaining the peace. See Hoddie and Hartzell, "Civil War Settlements and the Implementation of Military Power-Sharing Arrangements," Journal of Peace Research, Vol. 40, No. 3 (May 2003), pp. 303-320. In a later work, they further show that implementation of institutional reform in political, economic, and territorial power sharing is a crucial component of a lasting peace. See Caroline A. Hartzell and Matthew Hoddie, Crafting Peace: Power-Sharing Institutions and the Negotiated Settlement of Civil Wars (University Park: Pennsylvania State University Press, 2007). Ending Civil Wars 11 But is this the case? In the next section, I provide an empirical overview of the links between civil war termination type and long-term outcomes. A Statistical Assessment of Civil War Termination In this section, I present a statistical test of civil war outcomes. The independent variable is type of civil war outcome (negotiated settlement, cease-fire/ stalemate, or victory), and the dependent variable is war recurrence. The analysis presents a series of descriptive statistics about the character of civil war termination from 1940 to 2000,18 correlations among key variables, and a rare-events logit model. The statistical section concludes with an analysis of levels of democracy/authoritarianism and economic growth following civil war in relation to termination types. INDEPENDENT VARIABLE: TERMINATION TYPE By "negotiated settlement," I mean an ideal-type war termination in which neither side admits defeat and the combatants agree to end the violence and accept common terms on how to govern a postwar state. An external party may help to halt the violence and arrange a settlement—for example, another state, a regional or an international organization such as the United Nations or the Organization of African Unity, or a nongovernmental organization such as the International Committee of the Red Cross. For an ending to be considered a negotiated settlement, however, a third party is not required. A "cease-fire/stalemate" is similar to a negotiated settlement in that the parties agree to halt the violence. Unlike negotiated settlements, however, a cease-fire/stalemate does not attempt to achieve agreement on postwar power sharing. It simply ends the violence. By "victory," I mean an ideal-type war termination in which one side explicitly acknowledges defeat and surrenders. The terms of surrender need not be unconditional, although the closer they are to unconditionality, the closer to 18. The data set includes all cases of civil war through 2007. Analyses that include assessments by decades (e.g., 1940s, 1950s) consider civil wars that ended by December 31,1999, while other analyses include all civil wars that ended by December 31, 2002. For the complete data set, summaries of the cases, and codebook, see http://belfercenter.ksg.harvard.edu/files/Toftcwdata.xlsx. See Nicholas Sambanis, "Partition as a Solution to Ethnic War: An Empirical Critique of the Theoretical Literature," World Politics, Vol. 52, No. 4 (July 2000), pp. 437-483; Michael W. Doyle and Nicholas Sambanis, "International Peacebuilding: A Theoretical and Quantitative Analysis," American Political Science Review, Vol. 94, No. 4 (December 2000), pp. 779-801; and Fearon and Laitin, "Ethnicity, Insurgency, and Civil War." For a critical review of negotiated settlements, see Alexander B. Downes, "The Problem with Negotiated Settlements to Ethnic Civil Wars," Security Studies, Vol. 13, No. 4 (Summer 2004), pp. 230-279. International Security 34:4 12 the ideal type the victory will be. The Sri Lankan government's 2009 victory over the Tamil Tiger insurgency is a recent example. I began by combining the relevant information into a data set including all civil wars fought from 1940 to 2007.19 To be included in the data set, a civil war had to meet six criteria: (1) the focus of the war was control over which group would govern the political unit; (2) there were at least two groups of organized combatants; (3) one of the combatants was an internationally recognized state; (4) there were at least 1,000 battle deaths per year on average; (5) the ratio of total deaths had to be at least 95 percent to 5 percent, meaning the stronger side had to have suffered at least 5 percent of the casualties; and (6) the war had to have begun within the boundaries of an internationally recognized state. The first criterion embraces the notion of sovereignty and governance. In a civil war as defined here, the primary political objective of the combatants is governing, with each side rejecting the legitimacy of the other to rule over part or all of the territory of a state. The second criterion excludes spontaneous mob actions or riots, as in the Albanian pyramid crisis in 1997. The third criterion eliminates communal conflicts, which involve warring ethnic groups. The fourth distinguishes between civil wars and other types of violence such as riots and smaller-scale insurgencies. As a result, cases such as Northern Ireland are excluded. The fifth criterion captures the idea of a minimal capability of each side to conduct its military operations by inflicting casualties on the other side. This ratio criterion excludes massacres and genocides. The sixth criterion excludes wars between two sovereign states.20 The total number of wars that qualified for inclusion based on these criteria was 137. Of these, 118 experienced no violence for at least five years and are considered ended.21 19. Only civil wars that ended by 2002 are included in the logit analysis. 20. These six criteria are an amalgamation of criteria used by other scholars to define civil wars over the past two decades. For example, Licklider and Doyle and Sambanis use a death threshold of 1,000 total, as opposed to a yearly average. Furthermore, this data set includes wars that involved colonial struggles. Other data sets are inconsistent in this regard. For consistency wars that are commonly thought of as colonial wars are included in this data set. There are a total of ten such wars included here. Additionally other data sets use country years as their units of analysis (e.g., Doyle and Sambanis; and Fearon and Laitin) and differ on when to disaggregate or aggregate different wars. Doyle and Sambanis, for instance, consider Afghanistan to have suffered three separate wars. For a more comprehensive comparative analysis of this data set with others, see http:// belf ercenter.ksg.harvard .edu/ files / MTcodebook2010 .pdf. 21. Five years is a standard period to delineate the final, stable end of a war, because it typically allows for at least one election cycle. See Licklider, "The Consequences of Negotiated Settlements in Civil War"; Barbara F. Walter, "The Critical Barrier to Civil War Settlement," International Organization, Vol. 51, No. 3 (Summer 1997), pp. 335-364; and Caroline A. Hartzell, "Explaining the Stability of Negotiated Settlements to Intrastate Wars," Journal of Conflict Resolution, Vol. 43, No. 1 (February 1999), pp. 3-22. Ending Civil Wars 13 Table 1. Frequency of Civil Wars, 1940-2000 Decade Number of Civil Wars* Percentage of All Civil Wars Cumulative Percentage 1940 21 16 16 1950 23 18 34 1960 23 18 52 1970 25 19 71 1980 17 13 84 1990 20 16 100 total 129 100 * Because of rounding, the number of civil wars does not add up to 100. Table 2. Frequency of Civil Wars Ended Per Decade, 1940-2000 Decade Number of Civil Wars Percentage of All Civil Wars* Cumulative Percentage* 1940 13 12 11 1950 20 18 30 1960 14 12 42 1970 21 19 61 1980 8 7 68 1990 37 33 101 total 113 101 NOTE: Ongoing wars are not included. * Because of rounding, the percentage of all civil wars and the cumulative percentage do not add up to 100. No clear pattern of a decrease or an increase in the number of civil wars that began in each decade from 1940 to 2000 emerges from the data. As table 1 shows, the average number of civil wars that started in each decade from 1940 to 2000 is 22, with a high of 26 new wars in the 1970s and a low of 17 in the 1980s. The findings in table 2 show that the number of civil wars ending in the 1990s surpassed the numbers in previous decades. In the 1990s, 37 civil wars ended, or one-third of all such wars that began during the 1940-2000 period. During this period, most civil wars (79 wars, or 70 percent) ended in victory, followed by negotiated settlements (22 wars, or 19 percent), then cease-fires/ stalemates (12 wars, or 11 percent). Victories occurred more than four times as often as negotiated settlements and seven times as often as cease-fires/ stalemates. Figure 1 shows that the ways in which civil wars end have changed dramatically since the 1940s. Through the 1980s, victory by rebels or governments was the dominant termination type, ending between 75 and 100 percent of wars per decade. In the 1990s, however, victory ended only 41 percent of civil wars. International Security 34:4 14 Figure 1. Percentage of Civil Wars Ended, by Termination Type, 1940-2000 100 90 80 70 60 50 40 30 20 10 0 ad 1940s 1950s 1960s 1970s 1980s 1990s I I victory | | negotiated settlement [J cease-fire/stalemate Moreover, negotiated settlements ended only a handful of civil wars between 1940 and 1989. But by the 1990s, they ended 41 percent of all civil wars. Further, of all the civil wars concluded through negotiated settlements, two-thirds ended in the 1990s. Cease-fires/stalemates ended an additional one-fifth during the 1990s, but only a handful in previous decades. These findings raise two main questions. First, what explains the increase in civil wars ended by negotiated settlement in the 1990s? Second, should this form of war termination be the preferred policy in the future? Space constraints preclude a comprehensive explanation of the sudden outbreak of "peace" in the 1990s; the most likely explanation, however, is twofold.22 First, the end of the Cold War deprived the United States and the Soviet Union of the incentive to provide arms to combatants in proxy wars, even for combatants who wanted to go on fighting. In some conflicts, this sudden drop in the availability of arms would have caused a lull, until new sources of revenue to pur- 22. For a fuller treatment, see Monica Duffy Toft, "End of Victory? Civil War Termination in Historical Perspective," paper presented at the annual meeting of the International Studies Association, Honolulu, Hawaii, March 1-6, 2005. See also Virginia Page Fortna, "Where Have All the Victories Gone? Peacekeeping and War Outcomes," paper presented at the annual meeting of the American Political Science Association, Toronto, Canada, September 3-6, 2009. Ending Civil Wars 15 chase weapons could be found (e.g., Angola and Sierra Leone). A second explanation is that the United States, as the sole remaining superpower, came under increasing pressure to take moral responsibility for the world's ongoing civil wars, many of which it appeared to possess the diplomatic, economic, or military capacity to halt outright. This consistent pressure encouraged the United States (and some of its allies) to intervene in some civil wars, especially those whose destructiveness threatened U.S. national interests, such as European stability (i.e., during the wars in former Yugoslavia in the early 1990s) or those that were calculated to be cheap and easy (e.g., Somalia in 1993). There are a number of arguments for ending civil wars through negotiated settlement, the most powerful of which is that negotiated settlements reduce the number of deaths compared to victories.23 Assuming that a civil war's "costs" are measured narrowly in terms of life, negotiations to halt the violence would spare lives, thus reducing the war's costs. The logic of this argument is that without a settlement the combatants would go on killing each other, perhaps even escalating the level of violence. Yet there are at least two problems with this logic. First, even if negotiated settlements may save lives, it is also true that combatants have strong incentives to avoid sharing power in a new government. Moreover, combatants are just as likely to use an armistice as an opportunity to recover and rearm in preparation for a future fight. Therefore, negotiated settlements may have an increased likelihood of saving lives in the short term, but an equally increased likelihood of costing even more lives in the long run. Second, caution should be used when measuring costs solely in terms of lives lost: human life is a crucial cost—perhaps the most crucial cost. But what if well-intended efforts to save lives condemn the "saved" to years of desultory existence, with little hope of political liberty or economic prosperity? Some scholars argue that defeat makes rearming by the losing side highly improbable—not because the loser will lack access to weapons or because its numbers have been so diminished that even when armed they are incapable of imposing costs on winners. Rather, there is something powerful and legitimating about the recognition of defeat and victory that precludes, at least for some time, attempts to continue to resist victors through physical violence (i.e., might makes right). In addition, if combatants nearing defeat cannot assume that there will be a third party available to intervene and guarantee their security, they may give up sooner, thus sparing lives.24 23. Licklider, "The Consequences of Negotiated Settlements in Civil Wars." 24. One might assume in such cases that the rebels are most likely to flee after surrendering, because traditionally (i.e., especially prior to 1977) rebels have never enjoyed the full protections of International Security 34:4 16 DEPENDENT VARIABLE: CIVIL WAR RECURRENCE The data show that wars ending in victory were nearly twice as likely to remain settled than those concluded through negotiated settlement or a cease-fire/stalemate.25 The recurrence of wars from 1940 to 2000 ending in military victories was 12 percent (10 of 81); 22 percent (5 of 23) in negotiated settlements; and 31 percent (4 of 13) in cease-fires/stalemates. Thus, wars ended through negotiated settlement were twice as likely to reig-nite as those ending in victory. In addition, rebel victories were more stable than government victories. Therefore, not only does it matter that the war ended with a victory but also who achieved that victory. Whereas 17 percent of wars (8 of 48) ending in a government victory recurred (fewer than negotiated settlements and cease-fires/stalemates), only 6 percent of wars (2 of 33) won by rebels did so. Understanding the relationship between civil war outcomes and the duration and quality of the subsequent peace requires consideration not only of key variables but also of control variables.26 The literature on civil wars cites a number of factors affecting civil war outcomes. Four factors serve as control variables for this analysis: (1) whether the war was identity based;27 (2) whether it involved a fight over territory;28 (3) the death toll; and (4) the casualty rate. The first two factors speak to the quality of the war and the issues over which the combatants were fighting. As some research has indicated, international humanitarian law. Victorious incumbent governments have tended to deal extremely harshly with surviving rebels, who most often face exile or imprisonment, at best, and torture or murder, at worst. 25. Doyle and Sambanis discuss a related, but contradictory, finding: settlements enhanced peacebuilding between the combatants. Their finding supports a different issue, which deals with the quality of peace rather than the absence or duration of peace. See Doyle and Sambanis, "International Peacebuilding." See also J. Michael Quinn, T. David Mason, and Mehmet Gurses, "Sustaining the Peace: Determinants of Civil War Recurrence," International Interactions, Vol. 33, No. 2 (April 2007), pp. 167-193. 26. All of the models were tested using additional controls. In none did regime type, duration, or the incidence of third-party military intervention achieve significance; nor did any of the controls affect the coefficients and significance of other variables. When economic development was measured by the natural log of a country's per capita GDP in the year prior to a civil war, it did affect the significance of termination type: victory in model 1 moves to two-star significance, while both negotiated settlement and rebel victory drop below the one-star threshold. But in no model is the economic indicator significant itself, and its adverse effect on other variables probably stems from the fact that per capita GDP data are missing for more than half of the observations in the data set. On the relationship between GDP per capita and civil war onset, see Fearon and Laitin, "Ethnicity, Insurgency, and Civil War"; and Marie L. Besancpn, "Relative Resources: Inequality in Ethnic Wars, Revolutions, and Genocides," Journal of Peace Research, Vol. 42, No. 4 (July 2005), pp. 393-415. 27. This variable was coded "1" if the civil war was rooted in ethnic or religious identity, as in former Yugoslavia with Bosnian Muslims fighting Orthodox Christian Serbs and Catholic Croats. If the war was not rooted in ethnic or religious identity, then it was coded "0." 28. This variable was coded "1" if the fight was over control of territory, as in former Yugoslavia, and "0" if the fight did not center on the control of a piece of a state's territory, as in Tajikistan. Ending Civil Wars 17 identity wars are seen as more intractable. In particular, security dilemmas may emerge among ethnic groups in anarchic settings, such as exist in states experiencing civil war, undermining the groups' ability to coexist after the war.29 In civil wars where territory is central to the fight, the territory may be seen as indivisible and therefore less amenable to stable settlement.30 The remaining two factors concern the costs and intensity of the fight. Some scholars have shown that a stable settlement is more likely when the parties have reached a hurting stalemate as the total number of casualties mount. According to William Zartman, hurting stalemates should make combatants more willing to negotiate; thus, owing to exhaustion, high casualty counts and long wars may be less associated with war recurrence.31 At the same time, civil wars that are more intense, in terms of the rate of casualties over time, most likely exacerbate hostilities and fear. To account for costs, I included a variable for the log of total war-related deaths and for intensity, the natural log of war-related deaths per month.32 The best tool to assess the relationships among these variables is a logit model.33 Table 3 presents the impact of the five types of war termination: victory, negotiated settlement, cease-fire/stalemate, government victory, and re- 29. Licklider, "The Consequences of Negotiated Settlements in Civil Wars," p. 684; Barry R. Posen, "The Security Dilemma and Ethnic Conflict," Survival, Vol. 35, No. 1 (Spring 1993), pp. 27-47; Chaim Kaufmann, "Possible and Impossible Solutions to Ethnic Civil Wars," International Security, Vol. 20, No. 4 (Spring 1996), pp. 136-175; and Chaim D. Kaufmann, "Intervention Ethnic and Ideological Civil Wars: Why One Can Be Done and the Other Can't," Security Studies, Vol. 6, No. 1 (Autumn 1996), pp. 62-101. In "Partition as a Solution to Ethnic War," Sambanis found that partition is more likely to follow ethnic wars. 30. See Monica Duffy Toft, The Geography of Ethnic Violence: Identity, Interests, and the Indivisibility of Territory (Princeton, N.J.: Princeton University Press, 2003); Ron E. Hassner, "To Halve and to Hold: Conflicts over Sacred Space and the Problem of Indivisibility," Security Studies, Vol. 12, No. 4 (Summer 2003), pp. 1-33; Ron E. Hassner, War on Sacred Grounds (Ithaca, N.Y.: Cornell University Press, 2009); Stacie E. Goddard, "Uncommon Ground: Indivisible Territory and the Politics of Legitimacy," International Organization, Vol. 60, No. 1 (Winter 2006), pp. 35-68; Stacie E. Goddard, Indivisible Territory and the Politics of Legitimacy: Jerusalem and Northern Ireland (Cambridge: Cambridge University Press, 2009); and Barbara F. Walter, Reputation and Civil War: Why Separatist Conflicts Are So Violent (Cambridge: Cambridge University Press, 2009). 31. I. William Zartman, ed., Elusive Peace: Negotiating an End to Civil Wars (Washington, D.C.: Brookings Institution Press, 1995). 32. The logic of this argument can be found in Kaufmann, "Possible and Impossible Solutions to Ethnic Civil Wars"; and David A. Lake and Donald Rothchild, "Containing Fear: The Origins and Management of Ethnic Conflict," International Security, Vol. 21, No. 2 (Fall 1996), pp. 41-75. 33. The main findings of the logit model were confirmed using a Cox proportional hazard model. A negotiated settlement increased the likelihood of war recurrence in each year by almost 400 percent, whereas military victories reduced the likelihood by roughly 75 percent (both findings were significant at the p = 0.01 [two-tailed tests]). In addition, rebel victories reduced the likelihood of recurrence by more than 20 percentage points over government victories (—68 percent to —44 percent). This model used the same controls as those used in the logit model presented here. See author's website for logfiles of a more comprehensive statistical analysis at http://belfercenter.ksg .harvard.edu/files/MTcodebook2010.pdf. Table 3. Logit Results of Civil War Termination Type on Recurrence Model 1 Model 2 Model 3 Model 4 Model 5 First Differences Victory -1.935*** (0.607) — — — — -0.238*** Negotiated settlement — 1.850*** (0.853) — — — 0.273*** Cease-fire/stalemate — — 1.248* (0.770) — — 0.200* Government victory — — — -0.578 (0.585) — -0.055 Rebel victory — — — — -1. 466** (0.879) -0.111** Identity 2.348** (0.819) 2.236*** (0.843) 2.273*** (0.794) 2.361 *** (0.901) 2.030*** (0.742) 0.181 *** Territorial war -0.997* (0.684) -0.416 (0.707) -1.077* (0.747) -0.723 (0.638) -0.828* (0.616) -0.037 LN average deaths per month 0.739*** (0.204) 0.762*** (0.209) 0.497*** (0.176) 0.564*** (0.190) 0.528*** (0.165) 0.283*** LN total deaths multl Constant -0.395** (0.171) -2.813** (1.439) -0.422*** ( 0.176) -4.537*** (1.812) -0.233 (0.184) -4.040*** (1.526) -0.302** (0.171) -3.641 *** (1.479) -0.211 (0.181) -3.919*** (1.676) -0.1 52*** Pseudo R 0.242 0.209 0.182 0.168 0.197 N = 11 6 in all models. One-tailed t-tests. ***p < 0.01, **p < 0.05, *p < 0.10. Huber/White robust standard errors appear in parentheses. Clustered on state. First differences for nontermination-type controls are given for model 2. First differences for casualty variables are taken for one standard deviation below the mean versus one standard deviation above the mean. Ending Civil Wars 19 bel victory on civil war recurrence. A "0" indicates no recurrence, and "1" indicates recurrence. This serves as the dependent variable. Given that the coefficients of a logit model are not directly interpretable as unit changes on the dependent variable (as they would be in a linear regression), the table includes "first differences"—factors that demonstrate the effect of war termination type on the probability of war recurrence while holding the alternate independent and control variables at their mean.34 The data presented in table 3 reveal that victory reduces the likelihood of civil war recurrence by 24 percent, relative to all other types of civil war termination. Conversely, negotiated settlements increase the chances of recurrence by 27 percent, relative to all other types. Both findings are statistically significant. As regards its substantive impact on war recurrence, war termination type is shown not only to be almost as important as a two standard deviation increase in the intensity of the violence,35 but also more important than issues of identity and territory, which have already received a good deal of attention and analysis in the literature.36 The statistical analysis thus confirms the hypothesis that civil wars ending in victory are less likely to recur than those ending with a negotiated settlement. Moreover, models 4 and 5 in table 3 show that rebel victory seems to be driving this relationship. Although the data show that recurrence of civil war following a negotiated settlement is more likely, are fewer people likely to die in the longer term than if the combatants fight until one side achieves victory? To assess whether negotiated settlements in civil wars end in fewer deaths, I analyzed four measures—total deaths, total deaths per capita, total battle deaths, and total battle deaths per capita—against the two options of negotiated settlement and victory for significance using a two-sample t-test. I also tested war duration (as measured in months) regarding its significance to negotiated settlements and victory.37 The findings in table 4 show that civil wars ended through negoti- 34. This analysis measured average effect of each termination type relative to a control group comprising all other cases. A pair-wise comparison of victories to negotiated settlements alone strengthens the results (the first difference becomes —0.28). 35. In the logscale, the difference between 4.8 and 8.4 translates into an increase of 124 casualties per month to 4,532 (a thirty-seven-fold increase). 36. In a separate analysis, Quinn, Mason, and Gurses found that ethnic revolutions and secessionist wars were not more likely to recur. The difference in findings (especially regarding ethnic revolutions, which seem to capture identity-based civil wars in this analysis, and are consistently shown to be significant in relation to recurrence) results from differences over which cases are included and their coding. Quinn, Mason, and Gurses use the Doyle and Sambanis data set, which excludes a number of wars of independence (e.g., identity based) and failed negotiated settlements. Given these discrepancies, I am more confident that this data analysis more accurately captures the dynamics. Quinn, Mason, and Gurses, "Sustaining the Peace." 37. To conduct the duration tests, I added the values of duration for each country's civil wars together. If, for example, in the original data set there were two separate observations for Angola International Security 34:4 20 ated settlements resulted in significantly more deaths (as measured by total deaths per capita) and that this relationship is statistically significant. The other measures, besides duration (itself a cost), did not reach statistical significance. Among recurring civil wars, there were more total deaths on average, but this finding did not reach statistical significance (see table 5). The total number of battle deaths per capita was not much greater between wars that recurred and those that did not. Wars that recurred were significantly longer than those that did not recur. Thus, war recurrence appears to be costly. Therefore, the results in tables 3, 4, and 5 show not only that civil wars ending in negotiated settlements are more likely to begin anew and to last longer than wars ended by other means, but that the wars following these failed settlements are significantly more deadly. The five civil wars that ended with negotiated settlements but later reignited (Angola, Iraq, Lebanon, the Philippines, and Sudan) led, on average, to 0.015 total deaths per capita compared with 0.011 for all recurring civil wars; and an average of 0.005 battle deaths per capita compared to a mean of 0.003 for all recurrences. By both measures, recurring civil wars following negotiated settlements were roughly 50 percent more deadly.38 Thus, if war recurrence—with its corresponding increase in deaths, sacrifices in opportunity costs, destruction of infrastructure, and possible escalation of violence—represents a high cost, then negotiated settlements appear to be more costly than allowing the combatants to fight until one side emerges victorious. The empirical evidence, therefore, does not support the normative argument that negotiated settlements save lives. Could there be other compelling reasons to support negotiated settlements? For example, what if, despite the higher likelihood of war recurrence following a negotiated settlement, the postwar environment promotes greater democratization or economic growth? In such cases, the quantitative loss of life might be redeemed by an enhanced quality of life that allows for political freedom or prosperity. Again, the results do not favor negotiated settlements, but rather called "Angola Ha" (Angola Civil War) and "Angola lib" (UNITA Warfare), the duration in months for each module was added to measure the total hardship a country endured as a result of both the initial war and its resumption. Where a recurred civil war is ongoing, it is truncated in 2006. These counts come from a number of sources, including J. David Singer and Melvin Small's Correlates of War data set, ver. 3-0, http://www.correlatesofwar.org; and the Oslo International Peace Research Institute data set, Data on Armed Conflict, http://www.prio.no/CSCW/Datasets/ Armed-Conflict/. The data and supporting documentation used for tables 4 and 5 are available on the author's website at http://belfercenter.ksg.harvard.edu/files/MTcodebook2010.pdf. 38. This finding holds even when excluding the deadliest of these civil wars—Sudan—as a potential outlier. When Angola (another particularly deadly conflict) is excluded, the number of per capita battle deaths is similar to the global mean, but per capita deaths remain higher for recurred negotiated settlements. To exclude both Sudan and Angola simultaneously—thus removing the two deadliest of five total failed negotiated settlements—makes little sense. Table 4. The Difference in Means of Death Counts for Negotiated Settlements Victory Mean (A/) Negotiated Settlement Mean (A/) Pr ITI > |t| t-test Total deaths Total deaths per capita Total battle deaths Total battle deaths per capita Duration in months 118,113 (80) 0.0089 (63) 71,226 (73) 0.0042 (59) 65.7 (81) 145,048 (23) 0.0174 (22) 54,348 (22) 0.0063 (21) 172.9 (23) 0.7077 0.1001 0.7108 0.3409 0.000 0.3760 1.6628 -0.3719 0.9582 4.5576 Table 5. The Difference in Means of Death Counts for Recurred Civil Wars No Recurrence Mean (A/) Recurrence Mean (N) Pr ITI > |t| t-test Total deaths Total deaths per capita Total battle deaths Total battle deaths per capita Duration in months 135,897 (88) 0.01007 (70) 85,512 (80) 0.00543 (65) 97 (89) 258,377 (14) 0.01120 (14) 70,517 (14) 0.00276 (14) 180 (14) 0.2034 0.8485 0.8042 0.3330 0.0184 -1.2802 -0.1916 0.2486 0.9743 -2.3970 International Security 34:4 22 appear to favor rebel victories. I explore the significance of this finding in the next section. POST-CIVIL WAR POLITICS The purpose of this article is not only to understand complex phenomena but also to provide policymakers with guidance. It is therefore useful to consider what impact the type of war termination might have on the nature of postwar political institutions. Policymakers who are deciding whether to intervene in another state's civil war must consider a variety of factors, including possible interruption in global commerce, threats to strategic resources, quality of life in a postwar environment, and ties to allies. Although the data set does not permit testing of all of these factors, it does allow exploration of links between the type of war termination and the degree of political liberty that survivors are apt to enjoy in the years following the cessation of a civil war. To track regime types by differences in war termination type, I estimated changes in the level of democratization in a state that had experienced civil war, based on the polity variable from the Polity IV project data.39 The polity variable ranges from —10 to 10; a score of —10 corresponds to authoritarian regimes (with no democratic qualities), and a score of 10 corresponds to the most democratic regimes (with no authoritarian qualities). In general, civil wars begin in states with few democratic qualities. An assessment of the average polity scores before and at the outset of civil war indicates a high degree of authoritarianism in these states, as shown in table 6. For example, five years prior to the start of civil war, their mean score was —3.63, with a slight decrease in authoritarianism as war approached.40 These scores are significantly lower than for all states in the international system. For the same period (1940-2002), the mean polity score for all states was —0.33. Thus, civil wars tend to occur within more authoritarian states.41 39. Polity IV Project, Polity IV Dataset (College Park: Center for International Development and Conflict Management, University of Maryland, 2000), ver. p4v2000. 40. The results in table 6 indicate a shift toward less repression as the war approaches, although repression increases just slightly during the year the war breaks out. This decrease in authoritarianism might be indicative of these states' efforts at placating a discontented society. This "loosening up" argument goes only so far, however. As the war approaches, the level of authoritarianism does increase ever so slightly. Such a dynamic lends credence to the idea that "liberalization" of the system is precarious because it might put too much stress on that system, leading to more calls for even more liberalization. If these calls are not met, frustration sets in, repression increases, and violence ensues. 41. Similarly, Fearon and Laitin, as well as Havard Hegre, Tanja Ellingsen, Scott Gates, and Nils Petter Gleditsch, show that anocracy makes conditions ripe for insurgents. Authoritarianism does not allow insurgents to emerge, whereas democratic states seem to allow grievances to be handled without resort to violence. See Fearon and Laitin, "Ethnicity Insurgency and Civil War"; and Hegre, Ellingsen, Gates, and Gleditsch, "Toward a Democratic Civil Peace? Democracy, Political Ending Civil Wars 23 Table 6. Mean Polity Scores for All States Involved in Civil War, Five Years Prior to Start of War, 1 940-2002 Five years prior One year prior Year war started Mean Polity Score -3.63 (a? = 74) -2.53 (n = 81) -2.63 (n = 83) Table 7. Polity Scores by Termination Type before and at the Start of the Civil War Average Polity Score War Termination Type Five Years Before One Year Before Year War Began Victory -4.26* -3.44** -3.31 ** Negotiated settlement -3.40 -2.47 -3.33 Cease-fire/stalemate 4.00 2.89*** 2.89*** Government victory -4.16 -3.09 -2.43 Rebel victory -4.41 -4.05 -5.11** ***p < 0.01, **p < 0.05, *p < 0.10. The table presents t-tests (one-tailed) for the significance of difference in means across the categories. For example, the upper left cell indicates whether the average polity score for cases of victory is statistically different from those cases that did not end in victory. See the appendix for supporting data at http://belfercenter.ksg.harvard.edu/files/MTcodebook2010.pdf. So, are authoritarian states that experience civil war likely to pursue the path toward greater democratization after the war ends? Does the type of settlement matter? Table 7 presents the mean polity scores of states five years before civil war broke out, one year before, and the year the war began. The results in table 7 indicate that whether a state signs a negotiated settlement is unrelated to its polity score. Statistically, democracies and authoritarian states are equally likely to end their civil wars through negotiated settlement. This is not true, however, of cease-fires/stalemates, which were associated with more democratic states (e.g., India).42 The mean polity score for cases of victory was substantially different (i.e., more authoritarian) than for those cases not ending in victory. A disaggregation of the data shows that civil wars ending in rebel victory occurred within more authoritarian states. Change, and Civil War, 1816-1992," American Political Science Review, Vol. 95, No. 1 (March 2001), pp. 33-48. For the first full exploration of this thesis and supporting evidence, see Jack Snyder, From Voting to Violence: Democratization and Nationalist Conflict (New York: W.W. Norton, 2000). 42. Because there are so few cases, interpreting the findings related to cease-fires/stalemates must be done with care. International Security 34:4 24 Figure 2. Change in Polity Score before and after Civil War, by Termination Type, 1946- 5 years start 5 years 10 years 15 years 20 years before year after after after after *See the appendix for supporting data at http://belfercenter.ksg.harvard.edu/files/ MTcodebook2010.pdf. Does the level of democratization increase or decrease within post-civil war states?43 The average change in polity scores after a civil war can help to answer this question. Figure 2 illustrates the difference between a state's average polity score at the start of a civil war and the average change in polity scores five to twenty years after the war has ended. As figure 2 shows, negotiated settlements are associated with greater levels of authoritarianism as the postwar period is extended. Although the level of postwar democratization increases initially (nearly five points in the average polity score), this change is short lived, eventually giving way to rising levels of authoritarianism. After five years, the trend is toward authoritarianism, which is statistically significant. The pattern is particularly marked after twenty years, when the average change in polity score is close to —6. This finding reinforces the observations presented above about civil war recurrence. Incumbent governments faced 43. Moreover, they do not address whether democratization affects civil war recurrence. I included this factor in an earlier model; it had no effect. Ending Civil Wars 25 with what appears to the likelihood of renewed war seem to sink precipitously into authoritarianism as they attempt to avert another round of fighting. Cease-fires/stalemates do not appear to have an impact on the level of autocracy or democracy. A state's level of repression at the start of the war is a good indication of its level after the war. Because there are only 13 cases against which to run tests, caution must be used in weighting these findings too heavily. Although, in general, victory does not seem to have much impact on regime type, significant patterns emerge when victories are disaggregated into government and rebel victories. As figure 2 demonstrates, when governments win, repression increases by one to two polity points over the following decades. This makes sense, considering that the government—the same government that just suffered a civil war—is wary of opening up the political arena. Even when governments win, however, they do not seem more apt to crack down than they had been before the war. In contrast, levels of autocracy decreased after rebel victories. Within ten years of the end of civil war, the level of autocracy decreased by more than one point; and by twenty years, that level more than doubled. Although still within the authoritarian range, the point change is dramatic enough to indicate that repression eases for a good portion of the citizenry. In fact, on average, states with rebel victories perform best on the democratization front (i.e., statistically significant at the twenty-year mark). Given this finding, the common image of civil war pitting a corrupt, tyrannical government against freedom-loving rebels may not be entirely unfounded. In sum, the relationship between polity scores and war termination types provides further evidence that negotiated settlements may not be the best way to increase the prospects for greater democratization following a civil war: negotiated settlements are associated with higher authoritarianism over time. This finding is important for at least four reasons. First, it suggests that more democratic processes do not necessarily lead to more democratic outcomes. Although the average polity score five years after the termination of civil wars by negotiated settlements increases dramatically, this change does not. In all likelihood, what is being captured here is the effect of elections, mandated by the settlement itself. Years later, however, after the first round of elections and the departure of third parties, a cycle of opposition to the government and reactive repression starts to accelerate. This is consistent with my finding that negotiated settlements are more apt to lead to the recurrence of war. Generally speaking, the state's political dynamics seem to move in a decidedly negative direction. Second, any general explanation of the relationship between war termination types, on the one hand, and durable peace and robust reconstruction, International Security 34:4 26 Figure 3. Percent Per Capita GDP Growth before and after Civil War, by Termination Type* 3 -0.5 H-1-1-1- 5 years start 5 years 10 years 20 years before year after after after *See the appendix for supporting data at http://belfercenter.ksg.harvard.edu/files/ MTcodebook2010.pdf. on the other, will have to contend with this finding, especially if democracy promotion is the reason that negotiated settlements are sought. Third, although negotiated settlements are precarious in terms of both recurrence and increased authoritarianism, these findings suggest, if only tentatively given the small number of cases, that cease-fires/stalemates may be a better, albeit impermanent, solution. Fourth, although victories in general may not increase postwar civil liberties and freedom, rebel victories are associated with lower levels of authoritarianism. POST-CIVIL WAR PROSPERITY The next question is whether civil war termination type and economic development are linked. Suppose another cost or benefit of a certain type of war settlement materially influences the likelihood that survivors will struggle in poverty or recover and prosper. Using gross domestic product (GDP) as a proxy for economic development, I measured the impact of a given war termination type on postwar economic reconstruction. Figure 3 shows the per- Ending Civil Wars 27 centage change in GDP over 5,10, and 20 years (n = years after the wars terminated, with the year prior to the outbreak of war as the benchmark); that is, the percentage change in GDP five years after the termination of the civil war is the GDP at the five-year benchmark minus GDP from the year prior to the outbreak of war. Likewise, the percentage change in GDP ten years after the war was terminated is the GDP at the ten-year benchmark minus GDP from the year prior to the outbreak of war. The results depicted in figure 3 show that economic growth or decline is unrelated to the type of civil war settlement. Most of the states that suffered civil wars followed the same trajectory, with little divergence. The highest degree of divergence occurred among states whose civil wars ended with a rebel victory. These states suffered a decline in GDP immediately following the war. Within ten years, however, they recovered, displaying the same level of economic performance as states whose civil wars ended in something other than a rebel victory. SUMMARY Overall, the statistical analyses reveal that the type of civil war settlement materially affects the prospects for enduring peace and democratization and, to a lesser extent, economic prosperity. In addition, they are able to track a settlement's impact over an extended period of time. The analyses produced three main findings. First, civil wars ending in negotiated settlements are much more likely to recur. Second, negotiated settlements are no more likely to lead to democracy than other types of settlements. Rather, rebel victories are more likely to produce this result. Third, economic growth trends do not seem to be correlated with the type of civil war termination. A Theory of Civil War Termination This section offers a general theoretical explanation for civil war termination that is consistent with the article's empirical findings. Using Uganda as an illustrative case, it accounts not only for why peace eludes negotiated settlements but for why rebels might be more apt to allow greater democratization and governments less so. As a way of simplifying an array of plausible explanations for why some civil wars end and stay ended yet others recur, I characterized approaches to civil war termination as falling into two rival camps, noting that negotiated settlements receive the greatest support in both the scholarly and policymaking communities. In negotiated settlements, the aim is to preserve both sides as political actors, who are then aided in their work toward a compromise po- International Security 34:4 28 litical solution, which, it is hoped, will result in peace and prosperity. Allowing a civil war to continue until the emergence of a victor would only result in more killing and destruction and possibly spill over into neighboring states. In victory, the outcome effectively eliminates one side or, more commonly, damages it to the point where it must abandon its political objectives. An obvious advantage to this policy is that winners are spared the costs of having to consider another actor's wishes. Moreover, insofar as many of the world's peoples consider victory strong evidence of a just cause, a great deal of immediate legitimacy attaches to winners, enhancing their position in the transition from war to peace. Yet in victory, unlike either cease-fires/stalemates or negotiated settlements, it matters which of the two ideal-type combatants—the incumbent government or the rebel group—is the victor. I argue that logically, and generally, each of these actors starts with advantages and disadvantages. Incumbent governments enjoy international recognition and legitimacy. They can print money or manipulate the economy to raise capital to purchase weapons and recruit troops. They generally have standing militaries and police forces and control the mass media. Rebels start out with none of these advantages. The implication is that in organizing a challenge to the incumbent government, rebels cannot afford to make many mistakes. They must be innovative, organized, and above all efficient in the distribution and deployment of scarce resources. In short, rebels need a high level of institutional capacity if they are to survive and win. Rebels also require something else. Often, they begin the process of rebellion by criticizing the government's shortcomings and highlighting (or at times inventing) popular grievances. In cases where incumbent governments overreact to these challenges, the result will be a sharpening of grievances and a shift in public perception that the rebels may have a just cause. Rebels who are successful beyond the goal of merely surviving will be able to use this added legitimacy and increasing popular support to amass more resources and develop their institutional capacity.44 This legitimacy and institutional capacity are the factors needed to achieve postwar political and other reforms. Thus, rebel legitimacy and capacity may be endogenous to rebel victory. Rebels' decision to fight or quit during the course of a civil war may be conditional on their capacity to avoid defeat. Rebel victory frequently occurs under conditions of revolt against authoritarianism, in which rebel groups are more likely to gain support through promises of reforms to recruits and would-be constituents. This war-fighting capacity can 44. Jeremy M. Weinstein, Inside Rebellion: The Politics of Insurgent Violence (Cambridge: Cambridge University Press, 2006); and Collier, Hoeffler, and Söderbom, "On the Duration of Civil War." Ending Civil Wars 29 provide the institutional basis for postwar governance and for protection from future threats to the state. Consider Uganda's rebel victory in 1986. UGANDA In describing Uganda's post-civil war recovery the World Bank praised it as a "major turnaround" characterized by "astonishing efficacy" in aspects of the government's behavior.45 Uganda's recovery in the late 1980s and 1990s illustrates, and in a number of ways supports, my argument for why rebel victory increases the likelihood of postwar stability.46 Although myriad factors contributed to this success, the capacity and legitimacy of the victorious National Resistance Army (NRA) rebel group and that of its political wing, the National Resistance Movement (NRM), were pivotal. To appreciate Uganda's achievements since 1986, it is critical to recall what came before: markedly repressive and violent rule under a notorious despot, Idi Amin, followed by the return of Milton Obote, during whose rule civil war raged in the country. Under Amin, most progress that had been attained since Uganda's independence was largely reversed; by some estimates, 1 million people were displaced from their homes, and because of government coercion, most skilled laborers left the country, especially after Amin expelled the Asian population in 1972 and Britain cut off diplomatic relations in 1976. Between 1970 and 1980, Uganda's GDP declined by 25 percent, and exports declined by 60 percent.47 Throughout Amin's rule, Uganda's polity score was —7, an extreme autocracy. Much of the abuse and mismanagement that became commonplace under Amin reportedly continued in the 1980s during Obote's second presidency, from 1980 to 1985, which coincided with Uganda's civil war. The war marked another devastating episode in the country's postindependence period, with estimates of war-related death tolls ranging from 100,000 to 300,000. During the war, however, Yoweri Museveni's NRM developed institutions and broad-based support that served it well following the NRM's assumption of government control after the war. The war also produced a high degree of cohesion within the NRM leadership 48 45. Ritva Reinikka and Paul Collier, eds., Uganda's Recovery: The Role of Farms, Firms, and Government (Washington, D.C.: World Bank, 2001). 46. For a fuller treatment of the Uganda case, see Toft, Securing the Peace. 47. Severine Rugumamu and Osman Gbla, "Studies in Reconstruction and Capacity Building in Post-Conflict Countries in Africa: Some Lessons of Experience from Mozambique, Rwanda, and Uganda" (Harare, Zimbabwe: African Capacity Building Foundation, December 2003). 48. Dan M. Mudoola, "Institution-building: The Case of NRM and the Military in Uganda, 1986-1989," in Holger Bernt Hansen and Michael Twaddle, eds., Changing Uganda: The Dilemmas of Structural Adjustment and Revolutionary Change (London: James Currey, 1991), pp. 230-246. International Security 34:4 30 In the immediate aftermath of hostilities, the NRM drew on institutions that the NRA had developed for war making, such as the National Resistance Council and its supporting structure, including its network of political mobi-lizers, its political and diplomatic committee, and its finance and supply committee. After Museveni's ascension to the presidency in January 1986, the NRM-led government expanded these committees. Representative government also expanded; in February 1986, Museveni created a hierarchy of Resistance Councils at the village, parish, subcounty, county, and district levels. These councils served to promote the legitimacy of the new government and to foster political participation from broad social groups.49 All adults automatically became members of their village Resistance Council, which managed local affairs, including development projects. According to one scholar, despite being plagued by corruption, the councils provided crucial institutions through which the national government could hold elections for the legislature and could reach areas on the periphery of its control.50 The inclusiveness of the new government's institutions extended beyond the NRM's allies. The NRM welcomed its former rivals, offering top government posts to leaders from other political parties as a way to broaden its political base.51 Furthermore, the NRA successfully absorbed soldiers from what had been opposing armies during the war and recruited from all ethnic groups, processes that expanded the military and neutralized the risk of resumed war. This broad-based and expanding army also assumed social tasks, including growing food for the population 52 In its early years of governance, the NRM enjoyed a good deal of popular support, despite the limits of its democratic reforms. Part of this support appears to have carried over from the NRM's broad appeal during the civil war, when local communities had supplied the NRA with food, shelter, and information about the opposition 53 Moreover, during the fighting, the NRM had publicly decried state-inflicted violence. Later it was able to convince the public of its commitment to democratic reform. The NRM's relative successes in achieving political reforms were matched, and probably surpassed, by economic reforms that promoted growth. Uganda is commonly heralded by the international financial community, which has praised the NRM's openness to economic liberalization as a development suc- 49. Ibid. 50. Thomas P. Ofcansky, Uganda: Tarnished Pearl of Africa (Boulder, Colo.: Westview, 1996). 51. Rugumamu and Gbla, "Studies in Reconstruction and Capacity Building in Post-Conflict Countries in Africa." 52. Nelson Kasfir, "The Ugandan Elections of 1989: Power, Populism, and Democratization," in Hansen and Twaddle, Changing Uganda, pp. 247-278. 53. Ofcansky, Uganda. Ending Civil Wars 31 cess. The NRM made economic reconstruction a high priority, arguing that it was crucial for stabilization.54 At the end of the war, in 1986, Uganda's per capita GDP was just 58 percent of what it had been at independence, but this improved to 69 percent by 1994 and 80 percent by 1997 55 Although Uganda's post-1986 history has been remarkably stable, this stability and the country's democratic credentials have not gone unchallenged. Until only recently, the country faced an insurgency from the Lord's Resistance Army in the north, and the central government has stymied crucial democratic reforms. Museveni remains Uganda's president. Although he repealed prohibitions on political parties other than his own only in 2005—declaring Western multipartyism to be destabilizing—and he continues to block efforts to impose term limits on his presidency, Uganda has made remarkable strides toward greater democracy and prosperity. The point is not that rebel victory is a panacea. Rather, rebel victory can usher in considerable reform. And with regard to two sought-after outcomes—durable peace and political liberty—it outperforms negotiated settlement, the international community's currently preferred means of ending civil wars. These findings warrant careful consideration, particularly when the international community considers the costs and benefits of negotiated settlement. In the next section, I pull together the threads of my overall argument and supporting evidence to begin weaving a general theory of civil war termination, which requires settlements that promise former combatants both benefits from compliance and harm from defection.56 THE POLICY PROBLEM: THE TWO CAMPS AS COUNTERPRODUCTIVE EXTREMES If this analysis has so far read as a general indictment of negotiated settlement as the best way to end civil wars, then I have only partially achieved my aim. Thus far I have relied on a statistical analysis to support my claim that from 1940 to 2002, most negotiated settlements did not produce lasting peace (at least as compared to victory), but instead tended to break down. This, in turn, resulted in renewed, and at times intensified, violence that heaped costs both on belligerents and on regional and interstate actors to whom peace would have been a great boon. I have argued that at their ideal extremes, neither negotiated settlements (as 54. John Kiyaga-Nsubuga, "Uganda: The Politics of 'Consolidation' under Museveni's Regime, 1996-2003," in Taisier M. Ali and Robert O. Matthews, eds., Durable Peace: Challenges for Peacebuilding in Africa (Toronto: University of Toronto Press, 2004), pp. 86-112. 55. Reinikka and Collier, Uganda's Recovery. 56. Ultimately, readers must decide whether the Ugandan example is reflective of the larger universe of relevant cases or merely an outlier. International Security 34:4 32 typically designed and implemented) nor victories are apt to resolve civil wars in a way that (1) spares lives, property, and cultural treasures; (2) endures; (3) creates space for greater political liberty; and (4) establishes the conditions necessary for economic reconstruction, recovery, and development. Statistically, there is a clear association between rebel victory and positive outcomes on all the above dimensions, save economic development, which remains equivocal. This in itself is an important empirical finding. Rather than suggest that well-meaning third parties should allow civil war combatants to fight until victory, however, I use this idea as a foil for understanding why negotiated settlements have proven counterproductive along many of these dimensions. What is it about negotiated settlements that has produced failure, and what is it about victories that has led to success? Might it be possible to incorporate the more effective mechanisms of victory into negotiated settlements, so as to determine a new type of settlement that achieves all of the political objectives being sought? An examination of the negotiated settlements that have ended civil wars is illuminating in this regard. To begin, establishing a solid representative set of institutions must be a main objective. Most of the negotiated settlements include extensive provisions for establishing executive offices, legislatures, free and fair elections, and judiciaries, as well as demobilizing, demilitarizing, and reintegrating the armed forces. The emphasis is to make soldiers into citizens by giving them money and resources to reintegrate them into society. Yet, the means and methods to refashion and reinstitutionalize the military— security-sector reform (SSR)—is given only secondary consideration, or, if provisions regarding the police and armed forces are written into the agreements, then their implementation is allotted only meager resources. Although discussions of demobilization, demilitarization, and reintegration are common in the academic and policy literature, security-sector reform has received little consideration in academic research and has only recently become a priority among policymakers.57 Although the concept of SSR emerged in the late 1990s, it was not until the United States ousted the Taliban from Afghanistan that policymakers set up an SSR fund, finally recognizing the "vicious circle of security and reconstruction/development,"58 in which lack of 57. What little academic research exists on SSR addresses current policy practices. See, for example, Nicole Ball, "Transforming Security Sectors: The IMF and World Bank Approaches," Conflict, Security, and Development, Vol. 1, No. 1 (April 2001), pp. 45-66; Dylon Hendrickson, "A Review of the Security Sector Reform," Working Paper, No. 1 (London: King's College London, Centre for Defence Studies, 1999); and Chris Smith, "Security-Sector Reform: Development Breakthrough or Institutional Engineering?" Conflict, Security, and Development, Vol. 1, No. 1 (April 2001), pp. 5-20. 58. The fund's name is LOFTA (Law and Order Trust Fund for Afghanistan) and is administered by the United Nations Department of Peacekeeping. Michael Schmunk, German diplomat, interview by author, Cambridge, Massachusetts, December 2005. Ending Civil Wars 33 progress in the reconstruction and development arenas intensifies security problems, which in turn inhibit meaningful reconstruction and development. The postwar situation in Iraq following the removal of Saddam Hussein from power initially seemed to represent a step backward, when the United States disbanded the armed forces, leaving Iraq in the throes of civil war. It remains to be seen whether security-sector reform will achieve success in Iraq and thus avert civil war. Lack of attention to SSR can have devastating consequences. Following negotiated settlements, for example, a military is left to fend for itself. What often follows is the reemergence of multiple sets of militaries/militias/rebel organizations ready to do the bidding of any political leader who can provide (or has provided) resources to sustain their vision of how to win. Colombia is a case in point.59 For decades, the government gave priority to hurting the Revolutionary Armed Forces of Colombia (known by its Spanish acronym FARC) over achieving victory. The idea was similar to that found in strategic bombing campaigns: raise the costs of resistance until the adversary agrees to come to the negotiating table. But after a number of failed negotiated settlements over the course of decades, the Colombian military became frustrated with the negotiations and what it perceived as a lack of funds necessary to defeat the rebels.60 As a result, militias have proliferated, with some including former government military personnel, leaving the government to face new rebel organizations in addition to FARC. Only in the late 1990s, with U.S. help under Plan Colombia did the Colombian government commit itself to defeating the rebels in an effort to consolidate the state's control over its territory and institutions. The logic of this argument reduces to a single hypothesis: the more unified a postwar political administration is, the less likely war will recur. Yet a monopoly over the use of force does not seem to be enough to prevent the recurrence of civil war. The ability of one side to harm the other does not explain all of this study's findings. Although it might explain why military victories and cease-fires/stalemates are more stable than negotiated settlements, it cannot explain why rebel victories are more stable and result in less autocratic outcomes than government victories. 59. For background reading on the history of Colombia and its war, see David Bushnell, The Making of Modern Colombia: A Nation in Spite of Itself (Berkeley: University of California Press, 1993); Bert Ruiz, The Colombian Civil War (Jefferson, N.C.: McFarland, 2001); and Frank Safford and Marco Palacios, Colombia: Fragmented Land, Divided Society (New York: Oxford University Press, 2001). 60. William Aviles, "Institutions, Military Policy and Human Rights in Colombia," Latin American Perspectives, Vol. 28, No. 1 (January 2001), pp. 31-55. International Security 34:4 34 A key strength of negotiated settlements is their credible promise to provide mutual benefit as a reward for continued compliance. The collective goods of (1) no more violence and (2) the opportunity to participate in an electoral process and a government that promises to represent and protect the interests of the citizens in some form are important benefits. Where negotiated settlements fall short, however, is in the credibility of their promise to inflict harm should one or both sides in a settlement fail to comply with its terms. This is one of the reasons why much of the literature on civil war termination stresses the need for third parties,61 which can threaten to harm one or both sides if a settlement's provisions are undermined. Yet third-party involvement is often limited to getting the parties to the bargaining table or agreeing to early participation in the immediate implementation stage. Furthermore, third parties are rarely accorded the right to impose the terms of the settlement by use of force, or if given that right, refuse to do so. Finally, not every war attracts enough international interest that third parties are willing to become engaged and stay engaged, especially militarily. So, even strong promises of intervention to enforce compliance are often not credible. Thus, while negotiated settlements are good at providing benefits, they are less effective in following through on their threats and are therefore not self-sustaining. In contrast, when governments achieve victory, the military is left intact, as are the other branches of government. As a result, the government retains the capacity to repress or harm the population. According to the polity data, governments that win civil wars remain about as authoritarian as they were at the start of the war and well into the future. As I have shown, however, these victories are more stable than negotiated settlements and stalemates/cease-fires but less stable than rebel victories. The reason is that when rebels win, they are in a position not only to harm (or threaten to harm) their populations but also to benefit them. In winning, a rebel military organization remains capable of containing moves against its government. But because it is a rebel organization, it has to appeal not only to a portion of its domestic audience for approval but also to an international community not predisposed to the overthrow of national governments. This is also consistent with the move toward the greatest level of democratization following rebel victories. Rebels need to buttress the legitimacy of their win: allowing greater liberalization of the political system is an effective means of doing so.62 Uganda under Museveni 61. Compare Barbara F. Walter, Committing to Peace: The Successful Settlement of Civil Wars (Princeton, N.J.: Princeton University, 2002); and Walter, "The Critical Barrier to Civil War Settlement." 62. For an impressive effort at demonstrating just how strong international pressure is on developing countries to liberalize their politics and economics, see Roland Paris, At War's End: Building Peace after Civil Conflict (New York: Cambridge University Press, 2004). Ending Civil Wars 35 exemplifies this pattern, highlighting how rebel victories provide both key factors of benefit and harm. Another way to understand this dynamic is to look at successful negotiated settlements (what was it about the terms of the settlement that stand out from other settlements?) and to think in the abstract about the terms that might make a victory more stable. Consider, for example, the negotiated settlement that ended the decades-long fight between the Farabundo Marti National Liberation Movement and the rightist government of El Salvador. This settlement lasted, but unlike most before or since, it contained robust provisions for the reconstruction of El Salvador's security forces.63 It contained both carrots and sticks.64 Therefore, to achieve success, negotiated settlements need to offer a mix of benefits and threats. Securing the peace demands (perhaps paradoxically) making these threats as credible as the benefits, and then balancing them on a case-by-case basis in negotiated settlements or, failing that, in support of victory by one side or the other. Conclusion This study of civil war termination makes at least two important scholarly contributions. First, it underscores the importance of extending the time horizon for judging success or failure of post-civil war environments beyond the five years that have limited most previous analyses. Extending the time horizon challenges the three problematic assumptions underpinning the current preference for attempting to end civil wars by negotiated settlements of a certain type: (1) the more quickly the violence can be halted, the greater the number of lives that can be saved, and "lives saved" is the only cost of consideration; (2) escape from the constant fear of death in civil war is generally sufficient to motivate combatants to make the compromises necessary to produce lasting peace; and (3) a threat of further violence is immoral or unpalatable. Although it is true that from 1940 to 2002, most negotiated settlements halted the killing in the short term, and often led to greater democratization, both peace and democratization have tended to be short lived, ending after the first or second election cycle. The longer-term analysis reveals that negotiated settlements 63. See Charles T. Call, "Assessing El Salvador's Transition from Civil War to Peace," in Stephen John Stedman, Donald Rothchild, and Elizabeth M. Cousens, eds., Ending Civil Wars: The Implementation of Peace Agreements (Boulder, Colo.: Lynne Rienner, 2002), pp. 383-420; and Toft, Securing the Peace, chap. 5. 64. In fact, although the war was largely fought over land and class issues, most of the provisions of the agreement focused on transformation of the security sector. International Security 34:4 36 have tended to lead to a greater likelihood of war and repression, whereas rebels who win civil wars seem better able to keep the peace and allow for more democratization. Second, expanding scholarly analysis of civil war termination types beyond the current default—negotiated settlements of the "all carrots and no sticks" type—to include victories provides a much larger set of cases and variables to drawn upon to enhance understanding of the conditions most likely to support long-term stability, democracy, and prosperity. Axiomatically supporting negotiated settlements that contain only provisions for benefits is bad policy because these generally do not last, and because when civil wars reignite, they may escalate the violence in either quantity or quality. If stability, democracy, and development are truly valued objectives, then it is crucial to understand what it is about victories—and, in particular, rebel victories—that lend themselves to durable outcomes. Rather than simply shift to a willingness to support rebels in achieving victory, however, I have argued that third parties pay greater attention to security-sector reform during negotiations, thereby increasing the possibility of achieving both short- and long-term gains in better-designed, implemented, and supported settlements. Such settlements, which could credibly guarantee both benefits from cooperation and harm from defection, are likely to hold out better prospects for enduring peace, liberty, and prosperity following a civil war. Failing that, support in pursuit of victory, especially victories by rebels, may be a worthy objective. The alternative is continuing to make promises of intervention that are progressively less credible to both domestic and other target audiences, and accepting civil wars that will last decades. This, in itself, is something the developed world—where civil wars are rare—can no longer afford to ignore as a matter of interest, if not moral principle.