TY - JOUR
AU - Prorok, Alyssa, K
AB - Abstract When do rebel leaders “sell out” their constituents in the terms of peace by signing agreements that benefit group elites over the rebel constituency, and when do they instead “stand firm,” pushing for settlement terms that benefit the public they claim to represent? This article examines variation in the design of civil war settlement agreements. It argues that constituents, fighters, and rebel elites have different preferences over the terms of peace, and that rebel leaders will push for settlements that reflect the preferences of whichever audience they are most reliant on and accountable to. In particular, leaders of groups that are more civilian-reliant for their military and political power are more likely to sign agreements that favor broad benefits for civilian constituents, while leaders who do not depend on civilian support for their political and military power will sign agreements with fewer public benefits. We test this argument using original data on the design of all final peace agreements reached between 1989 and 2009, and several proxies for the group's level of reliance on civilian supporters. Using a variety of statistical tests and accounting for nonrandom selection into peace agreements, we find strong support for our hypothesis. When rebel groups launch a rebellion, they typically frame their fights as struggles to achieve broad reforms or public benefits for the whole country or a large segment of the civilian population that they claim to represent. Framing the struggle in the broadest possible terms is strategically valuable from a mobilization and recruitment perspective (McAdam 2010; Thomas 2012). Leaders, trying to build a viable armed group that can pose a realistic threat to the state, benefit from broad buy-in, both locally and internationally. Locally, they benefit from appealing to large numbers of people who they can mobilize as fighters or supporters who will provide food, shelter, and information to the rebel group (Kalyvas 2006; Weinstein 2007).1 Internationally, leaders benefit from selling their rebellion as legitimate to potential backers, and a public-benefits message can help in this regard (Jo 2015). This tendency to voice broad, public demands plays out empirically: the vast majority of rebel demands are over policies that would affect the public, such as governmental changes, political autonomy, elections, and self-determination (Thomas 2012, 146). For example, the Freedom Charter, adopted by the African National Congress (ANC) in 1955, proposed an inclusive, democratic political system and included goals relating to land reform, women's rights, education, human rights, and economic reform (Mabe 2018). Even groups notorious for their poor relations with civilians present their demands in these terms. The Revolutionary United Front (RUF) of Sierra Leone, for example, called for a return to multiparty democracy, equitable distribution of the country's resource wealth, and education and economic reforms in its founding manifesto (Abdullah 1998). Yet even though rebel demands are nearly universally articulated in terms of benefits to the public, peace agreements designed to end civil conflict, ostensibly by addressing rebel demands, are much more varied in terms of who benefits from peace. While some initiate broad reforms that reflect the public nature of rebel demands, others are packed with concessions to rebel elites or leadership. For example, consider the peace agreements reached by the ANC and RUF. The 1993 Interim Constitution Agreement between South Africa and the ANC included a variety of concessions that benefited the ANC's constituency, including economic, electoral, judicial, military, political, police, and territorial reforms. The 1999 Lome Peace Agreement between Sierra Leone and the RUF, however, proved much more favorable to RUF elites. It included economic payoffs to the RUF leadership, a guaranteed political position in the national government for the RUF leader, and an amnesty provision that included RUF elites, many of whom could have otherwise faced punishment for atrocities committed during war. What explains this variation in peace agreement design, and the disconnect in some cases between rebel demands and the terms of peace? In other words, when do rebel leaders “sell out” their constituents in the terms of peace by signing agreements that benefit group members over the rebel constituency, and when do they instead “stand firm,” pushing for settlement terms that directly benefit their public? We develop an argument that focuses on the incentives of rebel leaders to push for private payoffs versus public benefits during peace negotiations.2 We argue that peace agreements are likely to include a greater proportion of benefits for the rebel group's civilian constituency when rebel leaders are more reliant on civilian support for their military and political power. On the other hand, rebel leaders are likely to sign peace agreements that provide more group-level payoffs when the rebel organization depends less on close ties to the civilian population. We test this argument using original data on the design of civil war peace agreements. For all final agreements signed between 1989 and 2009, our original data record the direct, primary beneficiaries (constituents, rank-and-file, elites, and/or leader) of each provision in the agreement. We generate a unique measure of the proportion of public (i.e., constituent) benefits in each agreement and test the effect of civilian reliance on this original measure of settlement design. Using fractional response models and accounting for nonrandom selection into peace agreements, we find that rebel leaders who are more militarily and politically reliant on civilian constituencies tend to “stand firm,” signing agreements with more public benefits, while less civilian-reliant rebels tend to “sell out,” signing agreements with a lower proportion of public benefits. These results hold when accounting for the influence of governments and third-party mediators. This article makes several contributions to scholarship. First, it is one of the first studies to directly problematize peace agreement design as an outcome of interest. Most existing research treats settlement design as exogenous, instead focusing on how the terms of peace affect other post-conflict outcomes. It thus fails to account for how the nonrandom design of settlements affects outcomes such as conflict recurrence, implementation, post-conflict democratization, and the emergence of spoilers. This project contributes to research in each of these areas by demonstrating that the rebel-civilian relationship is a key determinant of the terms of peace, which should be accounted for when examining other post-conflict outcomes. Second, by demonstrating that agreement design is a function of rebel-civilian relations, we contribute to emerging research on civilians in wartime. Recent research shows that civilians affect rebel group behavior and the conflict environment (Weinstein 2007; Kaplan 2013; Arjona 2014). We extend this research by demonstrating that rebel-civilian relations also affect the design of peace agreements and thus the post-war environment. Finally, this project introduces a novel approach to conceptualizing and measuring agreement design. Previous studies that measure settlement design do so in relatively blunt ways, such as a count of power-sharing provisions (Hartzell and Hoddie 2007). While these studies provide valuable insights into the determinants of sustainable peace, the current project moves beyond them by thinking not just about the presence/absence of particular provisions, but also about who benefits from those terms. By relaxing the unitary actor assumption that dominates research in this field, we demonstrate that there is significant variation across audiences (i.e., members versus constituents) in terms of who benefits from peace. Such variation, furthermore, is likely to affect popular support for an agreement, its implementation, conflict recurrence, democratization, and the emergence of new rebel groups. The novel data we have collected will be particularly useful in future studies exploring these outcomes. Background Existing research on civil war settlement highlights the importance of agreement design for ensuring lasting peace. For example, settlements designed to include third-party security guarantees and extensive power-sharing provisions are critical to establishing long-term peace (Walter 2002; Hartzell and Hoddie 2007; Fortna 2008). These provisions affect sustainable settlement by alleviating fears of cheating (Mattes and Savun 2009) and overcoming information asymmetries that threaten to derail implementation (Mattes and Savun 2010). More recently, research has begun to look beyond security guarantees and power sharing, to how other aspects of settlement design affect peace and democratization. Matanock (2017), for example, finds that electoral participation improves the chances that peace will endure, while Bogaards (2013) argues that settlements with more inclusive postwar institutions improve prospects for peace. Similarly, Badran (2014) argues that agreements including more structural and procedural mechanisms to overcome impediments to cooperation lengthen the duration of peace. Joshi and Quinn (2015) also focus on sums of provisions, showing that agreements with a greater number of reforms signal greater government commitment and thus reduce the likelihood of renewed violence. Finally, recent research has begun to examine peace agreements in terms of social versus security reforms, showing that formal peace processes still largely focus on security in both design and implementation (Lee, MacGinty, and Joshi 2016). These studies provide important insights regarding how the terms of settlement affect sustainable peace. However, existing research is limited because it focuses overwhelmingly on how agreement design affects implementation or conflict recurrence, rather than problematizing settlement design as an outcome of interest. This is an important oversight, given that the terms of settlement are unlikely to be randomly assigned, with implications for what we know about how settlement design affects the durability of peace, democratization, and post-conflict recovery. Without accounting for the endogenous nature of settlement design, existing research on its effects may draw inaccurate or incomplete conclusions. A second limitation in existing research is its tendency to treat civil war combatants as unitary actors. As a result, it implicitly treats leaders, elites, rank-and-file, and constituents as having similar preferences over, and benefiting equally from, the terms of peace. This, arguably, is not true. Consider two political concessions commonly included in peace agreements: the integration of rebels into government; and electoral or party reforms. Both of these involve changes to the political system, and both are categorized as political power-sharing provisions (Hartzell and Hoddie 2007). Yet agreements vary considerably regarding which type of power sharing they include: in our data, thirty peace processes include electoral/party reform without integration, twelve include integration without electoral/party reform, and twenty-one include both. This variation is consequential because the integration of rebels into government is a group-level payoff that primarily benefits rebel elites, while electoral reforms directly benefit the public through more inclusive voter registration rules or the introduction of proportional representation. This project addresses both limitations. By relaxing the unitary actor assumption, we recognize that rebel group members and constituents likely have different preferences over the terms of peace, and that rebel leaders, who are ultimately responsible for the design and signing of peace agreements, must aggregate those preferences in ways that are most beneficial for themselves. Acknowledging this allows us to problematize settlement design as an outcome of interest in interesting ways. As we argue below, the distribution of power across a rebel leader's audience members critically affects settlement design. It is important to note, before continuing, that while our theoretical argument focuses on how rebels shape settlement design, governments and third-party mediators also critically influence the terms of peace. These actors’ contributions, while important, are not the central focus of this article. However, to the extent that they shape rebel demands or terms of settlement, we incorporate this information below. Constituent and Group Member Preferences During peace negotiations, leaders must balance the divergent interests of group members versus constituents. The constituency is composed of individuals that the group claims to represent, civilians who support the rebel cause passively, and active supporters who provide funding, food, shelter, or information to the rebels (Lichbach 1995; Kalyvas 2006; Weinstein 2007). Group members include both rank-and-file fighters and rebel elites. Due to the risks and costs group members face at conflict's end (e.g., prosecution, loss of income), we assume that rebel fighters and elites will prefer group-level benefits that accrue specifically to members of the organization. For example, provisions that call for amnesty or military integration would be preferred by group members, as they provide members with explicit benefits including protection from prosecution and future employment.3 Conversely, we assume that constituents prefer settlements that provide public benefits extending directly to the rebel constituency. Why would rebel constituents care about receiving public benefits, when, arguably, they could be just as well off if rebel leaders/elites receive private payoffs and use those payoffs to aid their constituency? For example, if rebel elites receive government positions, they may be able to influence policy to makes constituents better off. We expect rebel constituents to prefer direct public benefits because second-hand, or “trickle-down,” benefits are not guaranteed.4 In practice, second-hand benefits frequently go unrealized. For example, the 1999 Congo Agreement provided several elite-level payoffs to Ninja rebels. Mampouya and Matsanga received ministerial and special advisor posts, respectively. These payoffs, however, did not result in benefits for the rebel constituency. As Themner (2011, 52) notes, “these channels never resulted in ex-Ninjas having any real political influence. Despite being a member of the cabinet, Mampouya was not able to exert his influence on government policies.” Thus, elite payoffs that one might expect to result in secondary benefits for the public may not, in practice, deliver such benefits. Importantly, audiences care about this distinction between direct benefits and potential secondary benefits, likely because they are aware that trickle-down benefits may not be forthcoming. This was true in Ivory Coast, for example, where Patriotic Movement of Côte d'Ivoire (MPCI) supporters and rank-and-file became “increasingly suspicious that [rebel leader Guillaume Soro was] more interested in his own future political career than the grievances of the movement's adherents” (Melville 2003) when Soro stalled the peace process over ministerial posts for rebel elites. Constituents were angered, and criticized Soro, even though the ministerial posts in question could arguably have generated second-hand benefits for civilians. Given rebel members’ preferences for group-level benefits and rebel constituents’ preferences for public benefits, rebel leaders face difficult tradeoffs during peace talks, having to prioritize the preferences of one audience over the other. This is because government leaders are unlikely to give in to all rebel demands, as they are also strategic actors facing institutional constraints that affect what they are willing to concede. Importantly, what governments are willing to concede thus affects what rebel leaders can expect to achieve through negotiations.5 Given these constraints, rebel leaders must weigh the preferences of their constituents versus group members, producing agreements that may make one audience happier than another when the agreement is signed. Where leaders derive their military and political power from, we argue, affects how they make these trade-offs. Civilian Reliance and Settlement Design Our central argument is that rebel leaders whose military and political power depends more on constituent populations will “stand firm,” signing peace agreements that provide a greater proportion of public benefits, whereas rebel leaders who do not depend on civilian constituents will be more willing to “sell out,” signing agreements with fewer public benefits for the civilians they claim to represent.6 This follows for three reasons: (1) the need to maintain military effectiveness; (2) civilian influence on rebel governance; and (3) leaders’ desire to maintain political power. Maintaining Military Effectiveness First, leaders whose military power depends on civilians will need to maintain civilian support to continue the war effort effectively. Because combatants generally hash out the terms of peace while conflict is still ongoing, maintaining a strong fighting force will be important to rebel leaders, despite impending settlement. This is true not only because leaders will want to maintain military effectiveness in case negotiations fail, but also because maintaining a strong military presence allows the group to solidify or even improve its bargaining position in negotiations that take place in the shadow of conflict (Wagner 2000; Smith and Stam 2004). Therefore, maintaining military strength during negotiations is important to rebel leaders, even though settlement generally involves demobilization.7 Rebel groups vary, however, regarding the extent to which fighting capacity is derived from ties to civilians and the funding, shelter, and information they provide (Kalyvas 2006; Weinstein 2007; Thomas 2012). Some rebel groups must develop cooperative relationships with civilians in order to finance their wars, as they are relatively resource poor (Weinstein 2007). Resource-rich groups, on the other hand, have access to alternative funding sources, such as lootable resources, that render civilian support less critical. Thus, civilian-reliant rebel leaders will seek to gain concessions that benefit their constituent support base, as satisfying that base is critical to maintaining military power in the final stages of conflict. Non-civilian-reliant rebel leaders, however, will be less likely to fight for public benefits, as buy-in from the constituency is not critical to their ability to maintain an effective fighting force. Instead, they will seek concessions that benefit rank-and-file, rebel elites, or the leader herself. Joint Governance Second, close, cooperative links with the civilian population facilitate constituent influence on rebel decision-making through the development of joint civilian-rebel governance arrangements (Mampilly 2011; Arjona, Kasfir, and Mampilly 2015). In particular, rebel groups that rely on civilians to finance the war effort, supply their troops, and avoid detection by government forces have incentives to develop: (1) power-sharing governance structures that provide civilians a voice in shaping rebel behavior; and (2) inclusive governance structures that give a wide proportion of the population the ability to participate in the group's political deliberations (Weinstein 2007). Thus, these groups will create institutions that establish joint governance, including both rebel officials and civilians in decisions about rule-making, resource management, and the provision of public services. Joint governance structures are not developed by all rebel groups, however, as there are costs associated with this strategy. Rebels that include civilians in governance sacrifice some control over key decisions. Rebels are most likely to govern jointly when they are resource poor, depending on social endowments to build the movement, and when their civilian support base is homogenous, with close communal ties and shared belief systems (Weinstein 2007). Resource poverty forces rebels to develop more cooperative relationships with the civilian population to build an effective fighting force, while close social ties with a homogenous civilian base generate social pressures that allow the rebels to credibly commit to joint governance. When rebels enjoy strong economic endowments, they can forego joint governance and instead use more predatory tactics to finance the war (Weinstein 2007). This suggests that when a rebel group must rely on civilian support to mount an effective fight, especially if the support base is homogenous with close social ties, it builds reciprocal governance structures that make the leader more accountable to civilians when negotiating the terms of peace. Leaders in these circumstances will negotiate for more public benefits, as civilian input is likely to shape demands made at the negotiating table. When groups can finance their conflicts without civilian support, their leaders will be less accountable to constituents when designing agreements and will push for more group-level payoffs. Political Survival Finally, assuming that a leader's primary goal is to maintain power and avoid punishment during the transition from conflict (e.g., Goemans 2000; Croco 2011; Prorok 2016a), leaders will consider their own political prospects while negotiating. Rebel leaders may be particularly vulnerable to punishment during settlement because shifts in relative power and institutional changes during the transition generate high uncertainty (Stedman 1997; Hartzell and Hoddie 2003). Anticipating this vulnerability, leaders will seek terms of peace that maximize their chances of maintaining power now and into the post-conflict period. Maintaining political favor involves seeking benefits for the leader's most powerful audience. For civilian-reliant groups, this means rebel leaders have incentives to fight for an agreement that provides public benefits, as the constituency is an influential partner in rebel governance and thus likely influences rebel leader selection and dismissal. Leaders will be less likely to push for public benefits when they are not reliant on civilian support. In these cases, leaders will instead push for provisions that benefit the rank-and-file or rebel elites, as these are the audiences to whom the leader is accountable. Furthermore, maintaining the support of an existing constituency is easier than building a new support base as leaders look toward the transition into politics. Thus, leaders will want to please current supporters, anticipating that they will continue to make up the core of their support base. In fact, we expect leaders to negotiate for terms of peace that ensure the political relevance of their key audience into the future. For civilian-reliant leaders, this means fighting for public benefits that empower their supporters politically. Such concessions will not only curry favor with the constituency by providing a public benefit, but also ensure the rebel leader's continued political relevance as the leader of a constituency with increased political influence. For rebel leaders that do not already enjoy close ties with civilians, public benefits are unlikely to garner them significant new support, as civilians could easily attribute new rights to a benevolent government or other rebel groups in multi-actor conflicts. Without preexisting loyalties, therefore, we expect leaders to fight for group-level benefits to please members and political/security guarantees for themselves and rebel elites. Illustrative Examples The 1990s peace process between South Africa and the ANC illustrates how high civilian reliance incentivizes rebel leadership to push for public benefits. The ANC's power base was strongly civilian: it relied on a “people's power” strategy, and was involved in local governance, facilitating the development of, and working in tandem with, alternative governing structures to those of the state (Huang 2016). Reliance on broad civilian support helped ensure that ANC leaders stood firm on demands for political and economic reforms during the peace process. For example, ANC leaders insisted the government lift a state of emergency that was hugely unpopular with their constituents as a precondition for negotiations. Holding firm on this demand was clearly politically motivated; an ANC executive committee member acknowledged that the group “would lose support if we go into negotiations while our people are being slaughtered on the streets” (Chimbano 1990). Similarly, ANC spokesperson Patrick Lekota stated: “While holding grimly to our views, we shall do everything short of selling out to help the government remove the obstacles to a political settlement” (Renfrew 1990, emphasis added). These statements show that ANC leaders were influenced by constituent preferences. They were wary of appearing to sell out, out of fear that this would undermine their popular support. Ultimately, the ANC was able to deliver on its constituency's public demands, and ANC leaders maintained power into and beyond the post-conflict transition. In contrast, the agreement signed between António Bento Bembe—leader of the Front for the Liberation of the Enclave of Cabinda (FLEC)/the Cabindan Forum for Dialogue (FCD)—and the Angolan government in 2006 reflects the risks leaders face for failure to deliver on their key constituency's demands. Bembe represented FCD, an umbrella group including the rebel group FLEC, the Catholic Church, and Cabindan civic groups in peace talks (Africa Confidential 2006). FLEC enjoyed widespread civilian support in Cabinda and was supported by Cabinda's hugely influential Catholic Church and civil society groups (Pearce 2005, 147). Further, because FLEC/FCD did not profit from Cabinda's rich natural resources (Middleton and Miller 2008), Bembe was highly dependent on local civilian and church support for military/political power. The Memorandum of Understanding (MOU) signed in 2006 failed to live up to constituent expectations. Bembe did not deliver independence or even autonomy for Cabindans. Instead, the region was offered only “special administrative status,” while Bembe and top FLEC members were granted numerous payoffs, including a blanket amnesty, ministerial posts, and positions within Angola's state-owned oil company (Africa Confidential 2006). In reaction, civilian members of the FCD rejected the deal, calling it a “sham” (Africa Confidential 2006). Raul Danda, a member of Cabindan civic association Mpalabanda, said Bembe “was sold to the Angolan government” (The New Humanitarian 2010), and church representative Reverend Jose Marcos Kapa stated: “We, the church, do not consider these agreements valid because they have not included us, the civil society or politicians” (Siona 2006). Displeasure with the terms of peace translated into serious repercussions for Bembe. He was sidelined politically, with FCD representatives claiming he signed the MOU without their authority (Siona 2006; The New Humanitarian 2010). FLEC, which had reunified in 2004 with Bembe at the helm, quickly fractionalized again. As a result, Bembe was effectively removed from both opposition and rebel leadership. He no longer wielded any significant power within the Cabindan independence movement, and while he remained leader in name of his rump FLEC-R faction, the locus of power within Cabindan politics shifted decidedly away from Bembe. Finally, unlike ANC and FLEC leaders, Renamo leader Aphonso Dhlakama's power was not dependent on civilian support. Renamo relied on violence and coercion to control locals (Weinstein 2007, 230). In Renamo-controlled areas, “governance was not joint and did not involve power sharing” (Weinstein 2007, 183). Instead, Dhlakama's power derived from maintaining the loyalty of rebel elites, which he secured by providing contraband-funded payoffs to commanders and senior combatants (Weinstein 2007, 271). Given his lack of civilian reliance, Dhlakama entered peace talks in 1992 with little incentive to push for public benefits and strong incentives to secure payoffs for senior commanders. During negotiations, Dhlakama willingly backed down from public demands, such as his call for a “discussion on the return of Mozambican refugees from neighboring countries” (Reuters News 1991). Similarly, while he paid lip service to political reforms, in practice Dhlakama used stall tactics to delay them. Instead, political positions for rebel elites and guarantees that he and other rebels would not be arrested were Dhlakama's top priorities during negotiations. He insisted on “guarantees that no one will go to jail” (Reuters News 1992) and the inclusion of Renamo members “in a national army and police force, and its leaders given a say in government” (Esipisu 1992). Dhlakama's willingness to give in on public demands had no negative consequences for the rebel leader. He remained Renamo leader until his death from ill health in 2018, running for president on the Renamo party ticket in every postwar election beginning in 1994 (Cowell 2018). Testable Implication The above discussion suggests that rebel leaders who depend on civilians for their military/political power will be more likely to fight for public benefits in the terms of settlement than rebel leaders whose power does not derive from civilian support. H: Civilian-reliant rebel leaders will reach peace agreements with a greater proportion of public benefits than rebel leaders who are not civilian-reliant. Research Design We test our argument using original data on the design of civil war peace agreements. Our list of peace agreements comes from the Implementation of Peace Agreements Dataset (IPAD) (Cil 2016). IPAD identifies eighty peace processes, i.e., unique peace agreement-dyad pairs, across sixty-five civil conflict dyads and sixty-eight final agreements included in the UCDP Peace Agreement Dataset (Högbladh 2011).8 All peace agreements in the dataset are final settlements signed between one state government and one rebel group in which the agreement lasts for at least one month and partially or fully addresses the incompatibility in the conflict.9 While other datasets on agreement provisions exist, we prefer IPAD because it: (1) records peace processes at the dyad-level, which allows us to capture variation in benefits and civilian reliance in multi-party agreements; and (2) records whether a given provision includes concessions from government to rebels. Using IPAD as our starting point, we collect original data on peace agreement design. The resulting dataset uses the agreement-dyad as the unit of analysis. Agreement Design For each agreement in IPAD, we use the agreement text to determine who benefits from each provision.10 We coded each provision as benefiting any or all of the following actors: constituents, rank-and-file, rebel elites, and rebel leaders. The key distinction, in this article, is between public benefits (i.e., benefits to rebel constituents) and private, or member, benefits (i.e., benefits to rank-and-file, elites, and/or leaders). A provision is coded as providing public benefits if the rebel constituency is a direct and primary beneficiary of the provision. Direct beneficiaries are actors who have first-hand access to the benefit. That is, the actor is guaranteed access to the provision's benefits if that provision is implemented. Direct beneficiaries are distinct from second-hand, or indirect, beneficiaries, as the latter are not guaranteed to benefit, even if a given provision is implemented. Second, the rebel constituency must be a primary beneficiary of a provision for it to be coded as a public benefit. Primary beneficiaries must be explicitly referenced in the provision, or barring explicit reference, must be the actor who the provision is most obviously intended to benefit.11 Economic reforms, for example, are coded as public benefits, as they directly benefit the constituency and the public is the primary beneficiary of these reforms. Reintegration, on the other hand, primarily benefits rebel group members rather than the constituency, and is therefore generally coded as a group-level benefit. This is because reintegration provisions provide payoffs, job training, etc. to ex-combatants: civilians are excluded from access to the specific benefit (e.g., job training) provided in reintegration provisions. While civilians may sometimes benefit indirectly from fighter reintegration, this is a second-hand benefit and is therefore excluded as a public benefit. It is also important to note that each provision in each agreement is coded based on how it applies in that specific agreement, meaning that the same type of provision can benefit different actors across different agreements. Additionally, it is possible for any given provision to benefit multiple actors, such that some provisions provide both private and public benefits. For example, two reintegration provisions in our data include reintegration benefits for orphans and the wounded in addition to ex-combatants. In this instance, reintegration is coded as both a public and a group-level benefit, as both constituents and group members are direct, primary beneficiaries. The above coding rules are relatively strict. To address concerns that this measure is overly restrictive, we develop an alternative coding using more inclusive criteria. Specifically, the alternative coding includes direct, non-primary beneficiaries, in addition to direct, primary beneficiaries. For example, while electoral reforms that introduce a multiparty system primarily benefit the public, these reforms also directly benefit rebel leaders/elites, as they will likely lead to a new political party that forms as a result of the reforms, even if this is not explicitly stated in the provision. In the inclusive coding, therefore, these reforms are coded as both public and group-level benefits. Similarly, demobilization of government forces primarily benefits group members, as they are on the front lines of the conflict. However, rebel constituents also directly benefit from government troop withdrawal, if troops withdraw from a region where constituents live (e.g., Northern Ireland). This provision is therefore sometimes coded as both a public and a private benefit in the lenient version.12 Based on these coding rules, we have coded the beneficiaries of 661 provisions across all peace agreements in the dataset. Figure 1 shows the distribution of benefits across provision types for both measures. Coding details are included in Appendix C in the supplementary files. Figure 1. Open in new tabDownload slide Distribution of group and public benefits by provision type Figure 1. Open in new tabDownload slide Distribution of group and public benefits by provision type We aggregate the provision-level data to the agreement level. Public-benefits proportion captures the proportion of provisions that benefit the rebel constituency. To create this measure, we sum the number of provisions for which public benefits equals 1 and divide by the total number of provisions that include benefits for at least one actor within the rebel group. Thus, the measure is a proportion that captures the total number of public benefits relative to the total number of provisions with benefits to any rebel actor.13 The resulting variable ranges from 0 to 1, where higher values indicate that a larger proportion of beneficial provisions includes public benefits. The mean value is 0.448. The alternative coding, public-benefits proportion (inclusive) also ranges from 0 to 1, with a mean of 0.558. To illustrate this measure more concretely, consider the following agreements. First, the 1999 Agreement on Ending Hostilities in the Republic of Congo included provisions calling for the return of all former armed forces to their former ranks (rank-and-file/elite benefit), the reintegration of former combatants to civilian life (rank-and-file benefit), and the release of political and military prisoners detained during the war (rank-and-file/elite benefit). However, it did not include any broader political, economic, or military reforms that directly, primarily benefited the public. Therefore, the value of public-benefits proportion for this agreement is zero. The Cairo Agreement signed between the Government of Sudan and the National Democratic Alliance (NDA), on the other hand, receives the highest value of public-benefits proportion. All substantive provisions in the agreement tackled broader reforms to the political and territorial system, as well as the national armed forces of Sudan. These provisions include the introduction of a pluralist democratic system, the establishment of a constitutional review commission to improve respect for human rights and fundamental freedoms, the establishment of a decentralized federal system enabling the people of Sudan to rule themselves, and the professionalization of national armed forces, all of which directly, primarily benefit the public. One potential concern about our dependent variable (DV) is that some provisions are arguably “worth more” than others. If the more important provisions are systematically different from less important provisions in terms of who they benefit, our results may be picking up this variation rather than the effect of civilian reliance. We do not think this is a serious concern, as our data already exclude regulatory/administrative provisions. However, to allay any concerns, we generate an alternative DV that includes only power-sharing provisions based on Hartzell and Hoddie's (2007) operationalization, as these are likely the “most important” provisions. This version of the DV includes provisions regarding changes to the electoral system, and/or seats in the legislative and executive branches of government (political), integration into state security forces (military), the creation of new autonomous areas and/or decentralization of local governance (territorial), and the equitable distribution of state revenues and/or the creation of economic policies to improve land tenure, post-conflict recovery, and the development of disenfranchised areas (economic). This variable ranges from 0 to 1, with a mean of 0.50. Our results using this DV remain unchanged (see Table A4 in the supplementary files). Civilian Reliance We use three main variables to proxy civilian reliance, with additional measures included as robustness checks. First, we record whether the rebel group relies on an ethnic group for support. We use the ACD2EPR dataset from the Ethnic Power Relations project for this information. ACD2EPR provides a list of ethnic groups that are linked to each UCPD Conflict Actor via: (1) rebel claims to represent the ethnic group; (2) rebel recruitment from the ethnic group; and (3) the rebel group receiving support from at least 50 percent of the members of the ethnic group (Wucherpfennig et al. 2012). We create an aggregate measure of ethnic support that is coded 1 if the rebel group claims to represent, recruits from, and receives support from an ethnic group during conflict, and zero otherwise. Ethnic support is a useful proxy for civilian reliance because close ties via claims, recruitment, and support indicate that the rebels are reliant on the ethnic group for their military power. It also suggests that the rebel leader has a built-in constituency that he can turn to for political power in the post-conflict period. Finally, support from a homogenous constituency such as an ethnic group indicates that close social ties are likely present, making joint governance more likely. We therefore expect ethnic support during the conflict to increase the proportion of public benefits in the settlement. This variable is coded 1 for thirty observations (37.5 percent). Second, we record whether a rebel group had access to external sanctuary, or territory allowing them to establish bases outside of the conflict country, during conflict. We expect that a rebel group's civilian reliance will be weak when it has external bases. Many critical tasks required to build up and maintain fighting capacity, such as housing and training troops and extracting supplies to maintain the rebel army, are undertaken elsewhere, with the support of the host country, rather than via close ties to local communities in the conflict country.14 As a result, the leaders of these groups will be less reliant on the civilian population in their origin country. We therefore expect external sanctuary to decrease the proportion of public benefits in the terms of settlement. To code external sanctuary, we use information from the UCDP External Support Dataset (Högbladh, Pettersson, and Themnér 2011). We use the Access to Territory variable, which includes instances when “an external actor allows a warring party to set bases on the territory it controls, permits sanctuary or cross-border military action for the supported waring party.” External sanctuary is coded 1 when a rebel group had access to territory in external state(s) during conflict, and zero otherwise, with thirty-seven observations coded 1 (46.25 percent). Third, we record whether rebel groups used contraband funding during conflict. We expect rebel groups that are resource poor to develop close ties with the civilian population in order to build their military forces, feed and shelter troops, and finance their rebellion (Weinstein 2007). Groups that have access to contraband funding, however, will be less likely to develop close ties to the civilian population, and will be less reliant on civilian support to build and maintain their forces. Thus, we expect contraband funding to decrease the proportion of public benefits in settlements. Our measure of contraband funding comes from the Rebel Contraband Dataset (RCD), version 2 (Walsh et al. 2018). RCD measures if and how rebels earn income from the exploitation of both natural resources and criminal activities. We use these data to create a single indicator of whether the rebel group engaged in resource or criminal extortion during the conflict. We limit our measure to extortion, rather than including other types of funding, because extortion most closely captures exploitative, coercive behavior by the rebels that would likely preclude close ties with the population. Other common funding types, such as smuggling, may involve positive, cooperative relationships with civilians (e.g., rebels protect smugglers), so are excluded from our measure. The variable is coded 1 if extortion was used during conflict, and zero otherwise. Contraband funding is coded 1 in thirty-eight cases (48 percent).15 In addition to our three main proxies for civilian reliance, we run additional analyses using variables from Huang's (2016) Rebel Governance Dataset (RGD). These data provide detailed information on the relationship between rebels and civilians, so are ideally suited to measuring civilian reliance. We cannot use these data in our main analyses because they are based on a different civil war list (Doyle and Sambanis 2006) and time period (1950–2006), so we lose several observations when using RGD. Still, we present a robustness check using civilian dependence, an ordinal measure coded 1 if rebels did not depend systematically on civilian aid, 2 if rebels depended on a mix of civilian and non-civilian aid, and 3 if rebels depended exclusively on civilian aid (Huang 2016). In a second robustness check, we use Huang's (2016) measure crime, which is coded 1 if rebels engaged in widespread criminal activity targeting civilians in order to finance their rebellion. This variable is useful as an alternative to the contraband funding measure, to get at coercive, exploitative relationships with civilians. Results presented in Table A3 in the supplementary files show, as expected, that civilian dependence significantly increases and crime significantly decreases the proportion of public benefits in an agreement. Control Variables Control variables included in the analyses focus on confounders, or factors likely to impact both the extent of civilian reliance and the proportion of public benefits. Descriptive statistics for all controls are provided in Table A1 in the supplementary files. First, we control for whether the conflict is ethnic, secessionist, or over autonomy. Conflict type is likely to affect agreement design by shaping the types of demands that rebels make, and thus what is on the negotiating table. Further, conflict type should also affect the capacity of rebel groups to draw support from the civilian population, as rebels fighting ethnic and secessionist conflicts are likely to enjoy the support of co-ethnics and proximate communities. Controlling for conflict type ensures that our measures of civilian reliance are not simply capturing the effects of ethnic war or secession. Finally, controlling for conflict type allows us to empirically account for variation in rebels’ goals. Rebels fighting ethnic or secessionist conflicts are likely to have broader population goals than other rebel groups, so accounting for this ensures that our measures of civilian reliance are not simply capturing groups with broader goals, instead of groups accountable to the public.16Conflict type is measured as a dummy variable, coded 1 if the conflict is ethnic, secessionist, or over autonomy, and zero otherwise.17 Data for this variable come from the Non-State Actor (NSA) dataset (Cunningham, Gleditsch, and Salehyan 2013). Second, we control for rebels’ relative strength. Stronger rebel groups are likely to gain greater concessions from the state, as relatively weaker governments will feel more pressure to give in to rebel demands. This makes costlier public concessions more likely. Additionally, stronger rebel groups are likely better able to attract and retain civilian supporters, as they can credibly promise protection in exchange for support (Kalyvas 2006). We therefore include a measure of rebel strength, which is coded 1 for groups that are weaker than the state, 2 for groups at parity with the state, and 3 for groups that are stronger than the state. Data for this variable come from the NSA dataset (Cunningham, Gleditsch, and Salehyan 2013). Third, we control for conflict intensity as a proxy for the costs of conflict. Government officials may be more likely to make costly, public concessions to end a highly costly conflict. Furthermore, high conflict intensity suggests that the rebels were able to amass enough troops to engage the government in more intense fighting, and an important reason for this may be that the rebels had a large civilian support base to draw resources and manpower from. Conflict intensity is coded 1 if the conflict reached 1,000 battle deaths in at least one year during the war, zero otherwise. Data come from the UCDP Dyadic Dataset (Harbom, Melander, and Wallensteen 2008). Fourth, we control for technology of rebellion (Kalyvas and Balcells 2010). Research shows that popular support is a central feature of guerrilla/irregular war, whereas rebels who fight conventional conflicts do not require the same level of civilian collaboration (Kalyvas 2006; Balcells and Kalyvas 2014). At the same time, technology of rebellion also affects conflict outcomes (Balcells and Kalyvas 2014). We therefore include a dummy variable, conventional, which equals 1 if the war is fought conventionally, zero otherwise.18 Data for this variable come from Balcells and Kalyvas (2014). Finally, we control for civilian victimization by the state. The rebels’ ability to mobilize civilian support is likely to be higher if the state engages in violence against civilians, particularly indiscriminate violence (Kalyvas 2006). Further, the government's culpability for human rights abuses during the conflict is likely to reduce its willingness to make public concessions that open the political system, reform the judiciary, etc., as culpable government leaders will fear punishment for their bad behavior (Nalepa 2010). We therefore control for whether the government has committed civilian victimization using a dummy variable, government OSV (one-sided violence), coded 1 if the government engaged in deliberate targeting of civilians during the conflict, and zero otherwise. These data come from UCDP One-Sided Violence (OSV) dataset (Eck and Hultman 2007).19 In addition to the above factors, we expect characteristics of the government and third-party mediators to influence agreement design. We do not add these controls to our main analysis due to our limited sample size. However, in secondary analyses presented in Table 3, we include state and mediator variables to ensure that our key findings remain robust. Just like rebel leaders, government leaders are strategic actors who weigh the costs/benefits of concessions. They also face institutional constraints that affect what they are willing to offer in the terms of settlement. Specifically, a government leader whose power depends on a small number of elites will be less likely to offer public concessions that threaten regime insiders’ private benefits. Instead, this leader will pay off rebel elites, as the political and economic consequences would be far worse for regime insiders if the terms of peace included public reforms (e.g., democratization, economic liberalization). Conversely, democratic leaders, whose political power already derives from a broad constituency, will find it less costly to provide public benefits (Bueno De Mesquita et al. 2003). This suggests that the size of the government leader's support base (i.e., winning coalition) affects the extent to which he/she will be willing to offer public benefits in the terms of settlement. Building on Mattes and Rodriguez (2014) regarding autocracies’ coalition sizes, we expect personalist leaders to be least likely to offer public benefits, democracies to be most likely to do so, and other autocracies (e.g., military and single-party regimes) to fall in the middle. To measure regime type, we use information from the Polity IV project (Marshall and Jaggers 2010) and the Autocratic Regimes Dataset (ARD) (Geddes, Wright, and Frantz 2014). States are coded as democracies (baseline category) if they received a 6 or higher on the polity scale at conflict's end, as personalist autocracies if ARD coded them as pure or hybrid-personalist at conflict's end, and as nonpersonalist autocracies for all remaining states. We also include a measure of third-party mediator involvement. Research suggests that biased mediation produces agreements with more elaborated institutional arrangements (Svensson 2009), while the type of leverage a mediator has influences whether the terms of settlement reflect the parties’ underlying grievances (Menninga and Reid 2017). These insights suggest that the presence and identity of mediators influence the extent to which peace agreements reflect the public's interest. Agreements are more likely to include public benefits when mediators are present, and certain mediator types are more likely to push for public benefits. Specifically, non-governmental organizations (NGOs) and other grassroots organizations are particularly likely to promote settlements that provide public benefits because they have context-specific leverage (see Menninga and Reid 2017): they have a presence on the ground, knowledge of the local context, and direct relationships with civilians affected by the war. International organization (IO) mediators generally have less context-specific leverage, though they may have mandates that include the promotion of public reforms. Finally, state mediators are least likely to have context-specific leverage or any public-reform mandate, making them least likely to push for public benefits. To test these expectations, we include a categorical variable, third-party mediators, coded 3 if NGOs or other grassroots or faith-based organizations were involved, 2 if IOs were involved, 1 if only state mediators were involved, and 0 if no mediators were present during negotiations leading to final settlement. Data come from IPAD, which records third-party presence during negotiations based on information from the Managing Intrastate Conflict (MIC) in Africa dataset (Croicu et al. 2013). Results and Discussion The main results of our analysis are presented in Table 1. We present fractional response models (fractional logit regressions) with robust standard errors, as our DV is measured as a proportion that ranges between 0 and 1.20 Fractional logit regression estimates the conditional mean of y, E(y|x), when y is continuous within the range [0, 1] (Papke and Wooldridge 1996). Alternative models such as logistic regression, which is suitable for binary DVs, or linear regression, which can produce predicted values that are outside the bounds of 0 and 1, are not suitable for fractional DVs. We also present Heckman selection models to address nonrandom selection into settlement agreements (see Table 2, Models 1–3). In these models, stage one predicts whether a peace agreement was reached and stage two predicts the design of the agreement, allowing us to account for the possibility that our key variables influence not only settlement design but also the likelihood of settlement.21 Table 1. Fractional logit results . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.478** 0.592*** (0.204) (0.223) External sanctuary −0.465** −0.622*** (0.204) (0.227) Contraband funding 0.0239 0.100 (0.223) (0.251) Conflict type 0.593** 0.584** 0.694** 0.560** 0.515* 0.710** (0.250) (0.252) (0.269) (0.281) (0.276) (0.309) Rebel strength 0.00472 0.121 0.120 0.0516 0.204 0.205 (0.174) (0.168) (0.176) (0.183) (0.165) (0.185) Conflict intensity 0.533** 0.490** 0.507** 0.491** 0.432** 0.452* (0.212) (0.199) (0.220) (0.231) (0.216) (0.240) Government OSV 0.109 0.272 0.218 0.737*** 0.951*** 0.869*** (0.213) (0.208) (0.212) (0.231) (0.232) (0.231) Conventional −0.498** −0.590*** −0.517** −0.493** −0.626*** −0.517** (0.217) (0.216) (0.229) (0.243) (0.240) (0.253) Constant −0.679** −0.495 −0.754** −0.636* −0.387 −0.788** (0.306) (0.318) (0.351) (0.343) (0.356) (0.393) Observations 75 75 75 75 75 75 . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.478** 0.592*** (0.204) (0.223) External sanctuary −0.465** −0.622*** (0.204) (0.227) Contraband funding 0.0239 0.100 (0.223) (0.251) Conflict type 0.593** 0.584** 0.694** 0.560** 0.515* 0.710** (0.250) (0.252) (0.269) (0.281) (0.276) (0.309) Rebel strength 0.00472 0.121 0.120 0.0516 0.204 0.205 (0.174) (0.168) (0.176) (0.183) (0.165) (0.185) Conflict intensity 0.533** 0.490** 0.507** 0.491** 0.432** 0.452* (0.212) (0.199) (0.220) (0.231) (0.216) (0.240) Government OSV 0.109 0.272 0.218 0.737*** 0.951*** 0.869*** (0.213) (0.208) (0.212) (0.231) (0.232) (0.231) Conventional −0.498** −0.590*** −0.517** −0.493** −0.626*** −0.517** (0.217) (0.216) (0.229) (0.243) (0.240) (0.253) Constant −0.679** −0.495 −0.754** −0.636* −0.387 −0.788** (0.306) (0.318) (0.351) (0.343) (0.356) (0.393) Observations 75 75 75 75 75 75 Notes: Robust standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01. Open in new tab Table 1. Fractional logit results . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.478** 0.592*** (0.204) (0.223) External sanctuary −0.465** −0.622*** (0.204) (0.227) Contraband funding 0.0239 0.100 (0.223) (0.251) Conflict type 0.593** 0.584** 0.694** 0.560** 0.515* 0.710** (0.250) (0.252) (0.269) (0.281) (0.276) (0.309) Rebel strength 0.00472 0.121 0.120 0.0516 0.204 0.205 (0.174) (0.168) (0.176) (0.183) (0.165) (0.185) Conflict intensity 0.533** 0.490** 0.507** 0.491** 0.432** 0.452* (0.212) (0.199) (0.220) (0.231) (0.216) (0.240) Government OSV 0.109 0.272 0.218 0.737*** 0.951*** 0.869*** (0.213) (0.208) (0.212) (0.231) (0.232) (0.231) Conventional −0.498** −0.590*** −0.517** −0.493** −0.626*** −0.517** (0.217) (0.216) (0.229) (0.243) (0.240) (0.253) Constant −0.679** −0.495 −0.754** −0.636* −0.387 −0.788** (0.306) (0.318) (0.351) (0.343) (0.356) (0.393) Observations 75 75 75 75 75 75 . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.478** 0.592*** (0.204) (0.223) External sanctuary −0.465** −0.622*** (0.204) (0.227) Contraband funding 0.0239 0.100 (0.223) (0.251) Conflict type 0.593** 0.584** 0.694** 0.560** 0.515* 0.710** (0.250) (0.252) (0.269) (0.281) (0.276) (0.309) Rebel strength 0.00472 0.121 0.120 0.0516 0.204 0.205 (0.174) (0.168) (0.176) (0.183) (0.165) (0.185) Conflict intensity 0.533** 0.490** 0.507** 0.491** 0.432** 0.452* (0.212) (0.199) (0.220) (0.231) (0.216) (0.240) Government OSV 0.109 0.272 0.218 0.737*** 0.951*** 0.869*** (0.213) (0.208) (0.212) (0.231) (0.232) (0.231) Conventional −0.498** −0.590*** −0.517** −0.493** −0.626*** −0.517** (0.217) (0.216) (0.229) (0.243) (0.240) (0.253) Constant −0.679** −0.495 −0.754** −0.636* −0.387 −0.788** (0.306) (0.318) (0.351) (0.343) (0.356) (0.393) Observations 75 75 75 75 75 75 Notes: Robust standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01. Open in new tab Table 2. Heckman and matching results . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.148** 0.395* (0.0579) (0.219) External sanctuary −0.101* −0.477** (0.0527) (0.200) Contraband funding 0.0126 −0.016 (0.0555) (0.214) Conflict type 0.112 0.144* 0.137 0.677*** 0.561** 0.473* (0.0756) (0.0846) (0.0891) (0.258) (0.252) (0.262) Rebel strength 0.0480 0.0532 0.0601 −0.097 0.064 0.135 (0.0579) (0.0513) (0.0583) (0.169) (0.168) (0.188) Conflict intensity 0.0990* 0.108** 0.0990* 0.593** 0.427** 0.571*** (0.0511) (0.0509) (0.0552) (0.238) (0.186) (0.219) Government OSV −0.0114 0.0456 0.0404 0.260 0.202 0.090 (0.0442) (0.0440) (0.0458) (0.238) (0.208) (0.223) Conventional −0.126** −0.142*** −0.135** −0.326 −0.696*** −0.639*** (0.0505) (0.0516) (0.0553) (0.203) (0.226) (0.247) Constant 0.318* 0.431* 0.214 −0.640* −0.305 −0.618* (0.183) (0.221) (0.224) (0.387) (0.279) (0.324) Peace agreement Ethnic support −0.158 (0.156) External sanctuary −0.0622 (0.151) Contraband funding −0.153 (0.202) Battle deaths (ln) −0.0847* −0.0858* −0.0868* (0.0473) (0.0472) (0.0495) Multiple rebel groups 0.233* 0.234* 0.212 (0.139) (0.139) (0.144) Conflict type −0.0352 −0.0961 −0.152 (0.163) (0.149) (0.155) Total PKO (ln) 0.0955*** 0.0937*** 0.0777** (0.0254) (0.0233) (0.0314) Government troop support 0.163 0.140 0.336 (0.232) (0.243) (0.269) Rebel troop support −0.392 −0.404 −0.484* (0.286) (0.266) (0.283) Nonpersonalist autocracy 0.00762 0.0344 −0.0856 (0.253) (0.269) (0.305) Personalist autocracy 0.327 0.315 0.245 (0.215) (0.206) (0.248) GDP per capita (ln) −0.218*** −0.223*** −0.251*** (0.0692) (0.0724) (0.0773) Constant 0.0464 0.0957 0.434 (0.630) (0.664) (0.739) athrho 0.00178 −0.189 0.270 (0.569) (0.603) (0.672) lnsigma −1.633*** −1.596*** −1.560*** (0.0743) (0.0895) (0.170) Observations 1,016 1,016 948 . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.148** 0.395* (0.0579) (0.219) External sanctuary −0.101* −0.477** (0.0527) (0.200) Contraband funding 0.0126 −0.016 (0.0555) (0.214) Conflict type 0.112 0.144* 0.137 0.677*** 0.561** 0.473* (0.0756) (0.0846) (0.0891) (0.258) (0.252) (0.262) Rebel strength 0.0480 0.0532 0.0601 −0.097 0.064 0.135 (0.0579) (0.0513) (0.0583) (0.169) (0.168) (0.188) Conflict intensity 0.0990* 0.108** 0.0990* 0.593** 0.427** 0.571*** (0.0511) (0.0509) (0.0552) (0.238) (0.186) (0.219) Government OSV −0.0114 0.0456 0.0404 0.260 0.202 0.090 (0.0442) (0.0440) (0.0458) (0.238) (0.208) (0.223) Conventional −0.126** −0.142*** −0.135** −0.326 −0.696*** −0.639*** (0.0505) (0.0516) (0.0553) (0.203) (0.226) (0.247) Constant 0.318* 0.431* 0.214 −0.640* −0.305 −0.618* (0.183) (0.221) (0.224) (0.387) (0.279) (0.324) Peace agreement Ethnic support −0.158 (0.156) External sanctuary −0.0622 (0.151) Contraband funding −0.153 (0.202) Battle deaths (ln) −0.0847* −0.0858* −0.0868* (0.0473) (0.0472) (0.0495) Multiple rebel groups 0.233* 0.234* 0.212 (0.139) (0.139) (0.144) Conflict type −0.0352 −0.0961 −0.152 (0.163) (0.149) (0.155) Total PKO (ln) 0.0955*** 0.0937*** 0.0777** (0.0254) (0.0233) (0.0314) Government troop support 0.163 0.140 0.336 (0.232) (0.243) (0.269) Rebel troop support −0.392 −0.404 −0.484* (0.286) (0.266) (0.283) Nonpersonalist autocracy 0.00762 0.0344 −0.0856 (0.253) (0.269) (0.305) Personalist autocracy 0.327 0.315 0.245 (0.215) (0.206) (0.248) GDP per capita (ln) −0.218*** −0.223*** −0.251*** (0.0692) (0.0724) (0.0773) Constant 0.0464 0.0957 0.434 (0.630) (0.664) (0.739) athrho 0.00178 −0.189 0.270 (0.569) (0.603) (0.672) lnsigma −1.633*** −1.596*** −1.560*** (0.0743) (0.0895) (0.170) Observations 1,016 1,016 948 Notes: Standard errors clustered on dyad ID. *p < 0.10, **p < 0.05, ***p < 0.01. Cubic polynomials of time omitted from output for Models 1–3. Open in new tab Table 2. Heckman and matching results . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.148** 0.395* (0.0579) (0.219) External sanctuary −0.101* −0.477** (0.0527) (0.200) Contraband funding 0.0126 −0.016 (0.0555) (0.214) Conflict type 0.112 0.144* 0.137 0.677*** 0.561** 0.473* (0.0756) (0.0846) (0.0891) (0.258) (0.252) (0.262) Rebel strength 0.0480 0.0532 0.0601 −0.097 0.064 0.135 (0.0579) (0.0513) (0.0583) (0.169) (0.168) (0.188) Conflict intensity 0.0990* 0.108** 0.0990* 0.593** 0.427** 0.571*** (0.0511) (0.0509) (0.0552) (0.238) (0.186) (0.219) Government OSV −0.0114 0.0456 0.0404 0.260 0.202 0.090 (0.0442) (0.0440) (0.0458) (0.238) (0.208) (0.223) Conventional −0.126** −0.142*** −0.135** −0.326 −0.696*** −0.639*** (0.0505) (0.0516) (0.0553) (0.203) (0.226) (0.247) Constant 0.318* 0.431* 0.214 −0.640* −0.305 −0.618* (0.183) (0.221) (0.224) (0.387) (0.279) (0.324) Peace agreement Ethnic support −0.158 (0.156) External sanctuary −0.0622 (0.151) Contraband funding −0.153 (0.202) Battle deaths (ln) −0.0847* −0.0858* −0.0868* (0.0473) (0.0472) (0.0495) Multiple rebel groups 0.233* 0.234* 0.212 (0.139) (0.139) (0.144) Conflict type −0.0352 −0.0961 −0.152 (0.163) (0.149) (0.155) Total PKO (ln) 0.0955*** 0.0937*** 0.0777** (0.0254) (0.0233) (0.0314) Government troop support 0.163 0.140 0.336 (0.232) (0.243) (0.269) Rebel troop support −0.392 −0.404 −0.484* (0.286) (0.266) (0.283) Nonpersonalist autocracy 0.00762 0.0344 −0.0856 (0.253) (0.269) (0.305) Personalist autocracy 0.327 0.315 0.245 (0.215) (0.206) (0.248) GDP per capita (ln) −0.218*** −0.223*** −0.251*** (0.0692) (0.0724) (0.0773) Constant 0.0464 0.0957 0.434 (0.630) (0.664) (0.739) athrho 0.00178 −0.189 0.270 (0.569) (0.603) (0.672) lnsigma −1.633*** −1.596*** −1.560*** (0.0743) (0.0895) (0.170) Observations 1,016 1,016 948 . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.148** 0.395* (0.0579) (0.219) External sanctuary −0.101* −0.477** (0.0527) (0.200) Contraband funding 0.0126 −0.016 (0.0555) (0.214) Conflict type 0.112 0.144* 0.137 0.677*** 0.561** 0.473* (0.0756) (0.0846) (0.0891) (0.258) (0.252) (0.262) Rebel strength 0.0480 0.0532 0.0601 −0.097 0.064 0.135 (0.0579) (0.0513) (0.0583) (0.169) (0.168) (0.188) Conflict intensity 0.0990* 0.108** 0.0990* 0.593** 0.427** 0.571*** (0.0511) (0.0509) (0.0552) (0.238) (0.186) (0.219) Government OSV −0.0114 0.0456 0.0404 0.260 0.202 0.090 (0.0442) (0.0440) (0.0458) (0.238) (0.208) (0.223) Conventional −0.126** −0.142*** −0.135** −0.326 −0.696*** −0.639*** (0.0505) (0.0516) (0.0553) (0.203) (0.226) (0.247) Constant 0.318* 0.431* 0.214 −0.640* −0.305 −0.618* (0.183) (0.221) (0.224) (0.387) (0.279) (0.324) Peace agreement Ethnic support −0.158 (0.156) External sanctuary −0.0622 (0.151) Contraband funding −0.153 (0.202) Battle deaths (ln) −0.0847* −0.0858* −0.0868* (0.0473) (0.0472) (0.0495) Multiple rebel groups 0.233* 0.234* 0.212 (0.139) (0.139) (0.144) Conflict type −0.0352 −0.0961 −0.152 (0.163) (0.149) (0.155) Total PKO (ln) 0.0955*** 0.0937*** 0.0777** (0.0254) (0.0233) (0.0314) Government troop support 0.163 0.140 0.336 (0.232) (0.243) (0.269) Rebel troop support −0.392 −0.404 −0.484* (0.286) (0.266) (0.283) Nonpersonalist autocracy 0.00762 0.0344 −0.0856 (0.253) (0.269) (0.305) Personalist autocracy 0.327 0.315 0.245 (0.215) (0.206) (0.248) GDP per capita (ln) −0.218*** −0.223*** −0.251*** (0.0692) (0.0724) (0.0773) Constant 0.0464 0.0957 0.434 (0.630) (0.664) (0.739) athrho 0.00178 −0.189 0.270 (0.569) (0.603) (0.672) lnsigma −1.633*** −1.596*** −1.560*** (0.0743) (0.0895) (0.170) Observations 1,016 1,016 948 Notes: Standard errors clustered on dyad ID. *p < 0.10, **p < 0.05, ***p < 0.01. Cubic polynomials of time omitted from output for Models 1–3. Open in new tab We also present models with weights generated using Covariate Balancing Propensity Score (CBPS) matching (Table 2). If the distributions of the confounders in our main analyses are not similar across the values of our main independent variables, it may still be difficult to properly control for them (Gelman and Hill 2006, 200). Specifically, the bias in our estimate of the “treatment” effect depends on whether we specify the correct functional form of the confounders. By making the distributions of the confounding variables more similar across treatment categories, we reduce problems that may result from misspecification of the functional form of included variables. As a result, the estimated effect of our main independent variables should be less biased post matching. We therefore balance the sample by matching on the control variables described above, using one of the three proxies for civilian reliance as the “treatment” in each model (Models 4–6) in Table 2. In Table 1, a positive coefficient estimate indicates that a given variable leads to an increase in the proportion of public benefits in the agreement. Models 1–3 use each of our three main proxies for civilian reliance and the main, or strict, coding of benefits for our DV (i.e., direct, primary public benefits). The results in these models generally support our hypothesis. Rebel leaders that receive ethnic support produce agreements with a significantly higher proportion of public benefits than rebel leaders who do not rely on ethnic support. Post-estimation results shown in Figure 2, furthermore, show that the predicted proportion of public benefits increases from 38.9 percent to 50.2 percent when increasing ethnic support from 0 to 1. This 29 percent increase is significant at the p < 0.05 level. Figure 2. Open in new tabDownload slide Predicted proportion of public benefits across different measures of civilian reliance Figure 2. Open in new tabDownload slide Predicted proportion of public benefits across different measures of civilian reliance Results for external sanctuary, our second proxy for civilian reliance, also support the Hypothesis. Model 2 shows that rebel groups that had access to extraterritorial sanctuary during conflict sign agreements that are significantly less favorable to the general public than rebel groups without external bases. Figure 2 shows that the predicted public-benefits proportion falls from 48.2 percent to 37.3 percent when external sanctuary increases from 0 to 1. This 23 percent decrease is significant at the p < 0.05 level. Finally, the results for contraband funding in Model 3 provide less support for our theoretical expectations. We expected that groups that fund their rebellions through criminal or natural resource extortion would be less reliant on civilian support and would therefore be less likely to sign agreements that favor the public. Instead, we find that contraband funding has no significant effect, and there is virtually no difference in predicted public-benefits proportion with or without contraband funding (Figure 2). This insignificant result could be because contraband funding's effect is primarily on leader and elite-level benefits, rather than rank-and-file or public benefits. Groups that rely on contraband funding may have leaders/elites who profit from conflict, due to resource extraction and other elicit activities. If this is the case, governments may need to provide payoffs specifically to leaders and elites (not rank-and-file) to compensate them for the financial losses that will occur when conflict ends and their resource/criminal extortion comes to an end. Our measure of public benefits, therefore, may simply not be nuanced enough to capture this dynamic. Models 4–6 in Table 1 present the results using our alternative coding of the DV, which uses the more inclusive rules to determine who benefits. The effects of our three civilian reliance proxies remain the same. Turning to the controls, we find that conflict type and conflict intensity significantly increase the proportion of public benefits in an agreement across all models, as expected. Ethnic/secessionist conflicts result in more public-benefits-oriented agreements, and governments make more public concessions when faced with higher conflict costs. As expected, Conventional conflicts consistently result in significantly lower public benefits in the terms of settlement. Finally, rebel strength and government OSV do not consistently affect the proportion of public benefits in an agreement. Rebel strength is insignificant across models, though consistently positive as expected, while government OSV is positive, rather than negative, across models. The central findings in Table 1 are confirmed in Table 2, which presents the results of Heckman selection models (1–3) and after CPBS matching (4–6).22 The findings across these models remain consistent with the main results. Ethnic support has a significant positive effect, while external sanctuary has a negative, significant effect on the proportion of public benefits in the agreement. The findings are also confirmed in Table 3, which includes additional controls for government regime type and third-party mediation. Our findings for civilian reliance hold when controlling for these alternative explanations. The results provide only partial support for expectations regarding government and mediator variables. Personalist autocracy is negative, as expected, but its effect is not significantly different from the baseline category, democracy. Nonpersonalist autocracy is also insignificant across models. This may be because regime type affects not only the government leader's costs of public concessions, but also the need to make public concessions: democracies have less need to make public concessions due to existing institutional openness. The effects of costs and need, therefore, likely cancel out. Third-party mediator variables are more in line with expectations. NGO mediation significantly increases the proportion of public benefits in five out of six models in Table 3. Table 3. Additional controls results . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.476** 0.547** (0.241) (0.262) External sanctuary −0.438** −0.626*** (0.214) (0.230) Contraband funding −0.175 −0.139 (0.228) (0.256) Conflict type 0.516** 0.570** 0.495* 0.455* 0.509* 0.464 (0.252) (0.270) (0.278) (0.276) (0.290) (0.301) Rebel strength −0.119 −0.0370 −0.00359 −0.129 −0.0380 0.0114 (0.161) (0.149) (0.151) (0.174) (0.150) (0.160) Conflict intensity 0.322 0.292 0.269 0.171 0.156 0.127 (0.196) (0.187) (0.196) (0.219) (0.209) (0.223) Government OSV 0.381* 0.452** 0.498** 1.135*** 1.210*** 1.251*** (0.226) (0.216) (0.225) (0.279) (0.274) (0.285) Conventional −0.370* −0.482** −0.369* −0.304 −0.470** −0.311 (0.216) (0.214) (0.222) (0.224) (0.211) (0.228) Nonpersonalist autocracy 0.384 0.477 0.250 0.384 0.564 0.239 (0.383) (0.394) (0.370) (0.438) (0.479) (0.434) Personalist autocracy −0.397 −0.153 −0.561 −0.594 −0.210 −0.745 (0.385) (0.421) (0.407) (0.476) (0.532) (0.499) State mediator 0.199 0.166 0.279 0.430 0.373 0.511 (0.320) (0.346) (0.335) (0.349) (0.374) (0.369) IO mediator 0.209 0.349 0.369 0.584 0.742** 0.760** (0.317) (0.298) (0.300) (0.368) (0.361) (0.361) NGO/grassroots mediator 0.480 0.520* 0.672** 0.777** 0.786** 0.945** (0.330) (0.304) (0.326) (0.377) (0.364) (0.379) Constant −0.796** −0.744* −0.738* −0.871** −0.795 −0.861* (0.338) (0.419) (0.441) (0.401) (0.503) (0.511) Observations 75 75 75 75 75 75 . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.476** 0.547** (0.241) (0.262) External sanctuary −0.438** −0.626*** (0.214) (0.230) Contraband funding −0.175 −0.139 (0.228) (0.256) Conflict type 0.516** 0.570** 0.495* 0.455* 0.509* 0.464 (0.252) (0.270) (0.278) (0.276) (0.290) (0.301) Rebel strength −0.119 −0.0370 −0.00359 −0.129 −0.0380 0.0114 (0.161) (0.149) (0.151) (0.174) (0.150) (0.160) Conflict intensity 0.322 0.292 0.269 0.171 0.156 0.127 (0.196) (0.187) (0.196) (0.219) (0.209) (0.223) Government OSV 0.381* 0.452** 0.498** 1.135*** 1.210*** 1.251*** (0.226) (0.216) (0.225) (0.279) (0.274) (0.285) Conventional −0.370* −0.482** −0.369* −0.304 −0.470** −0.311 (0.216) (0.214) (0.222) (0.224) (0.211) (0.228) Nonpersonalist autocracy 0.384 0.477 0.250 0.384 0.564 0.239 (0.383) (0.394) (0.370) (0.438) (0.479) (0.434) Personalist autocracy −0.397 −0.153 −0.561 −0.594 −0.210 −0.745 (0.385) (0.421) (0.407) (0.476) (0.532) (0.499) State mediator 0.199 0.166 0.279 0.430 0.373 0.511 (0.320) (0.346) (0.335) (0.349) (0.374) (0.369) IO mediator 0.209 0.349 0.369 0.584 0.742** 0.760** (0.317) (0.298) (0.300) (0.368) (0.361) (0.361) NGO/grassroots mediator 0.480 0.520* 0.672** 0.777** 0.786** 0.945** (0.330) (0.304) (0.326) (0.377) (0.364) (0.379) Constant −0.796** −0.744* −0.738* −0.871** −0.795 −0.861* (0.338) (0.419) (0.441) (0.401) (0.503) (0.511) Observations 75 75 75 75 75 75 Notes: Robust standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01. Open in new tab Table 3. Additional controls results . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.476** 0.547** (0.241) (0.262) External sanctuary −0.438** −0.626*** (0.214) (0.230) Contraband funding −0.175 −0.139 (0.228) (0.256) Conflict type 0.516** 0.570** 0.495* 0.455* 0.509* 0.464 (0.252) (0.270) (0.278) (0.276) (0.290) (0.301) Rebel strength −0.119 −0.0370 −0.00359 −0.129 −0.0380 0.0114 (0.161) (0.149) (0.151) (0.174) (0.150) (0.160) Conflict intensity 0.322 0.292 0.269 0.171 0.156 0.127 (0.196) (0.187) (0.196) (0.219) (0.209) (0.223) Government OSV 0.381* 0.452** 0.498** 1.135*** 1.210*** 1.251*** (0.226) (0.216) (0.225) (0.279) (0.274) (0.285) Conventional −0.370* −0.482** −0.369* −0.304 −0.470** −0.311 (0.216) (0.214) (0.222) (0.224) (0.211) (0.228) Nonpersonalist autocracy 0.384 0.477 0.250 0.384 0.564 0.239 (0.383) (0.394) (0.370) (0.438) (0.479) (0.434) Personalist autocracy −0.397 −0.153 −0.561 −0.594 −0.210 −0.745 (0.385) (0.421) (0.407) (0.476) (0.532) (0.499) State mediator 0.199 0.166 0.279 0.430 0.373 0.511 (0.320) (0.346) (0.335) (0.349) (0.374) (0.369) IO mediator 0.209 0.349 0.369 0.584 0.742** 0.760** (0.317) (0.298) (0.300) (0.368) (0.361) (0.361) NGO/grassroots mediator 0.480 0.520* 0.672** 0.777** 0.786** 0.945** (0.330) (0.304) (0.326) (0.377) (0.364) (0.379) Constant −0.796** −0.744* −0.738* −0.871** −0.795 −0.861* (0.338) (0.419) (0.441) (0.401) (0.503) (0.511) Observations 75 75 75 75 75 75 . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Model 6 . Ethnic support 0.476** 0.547** (0.241) (0.262) External sanctuary −0.438** −0.626*** (0.214) (0.230) Contraband funding −0.175 −0.139 (0.228) (0.256) Conflict type 0.516** 0.570** 0.495* 0.455* 0.509* 0.464 (0.252) (0.270) (0.278) (0.276) (0.290) (0.301) Rebel strength −0.119 −0.0370 −0.00359 −0.129 −0.0380 0.0114 (0.161) (0.149) (0.151) (0.174) (0.150) (0.160) Conflict intensity 0.322 0.292 0.269 0.171 0.156 0.127 (0.196) (0.187) (0.196) (0.219) (0.209) (0.223) Government OSV 0.381* 0.452** 0.498** 1.135*** 1.210*** 1.251*** (0.226) (0.216) (0.225) (0.279) (0.274) (0.285) Conventional −0.370* −0.482** −0.369* −0.304 −0.470** −0.311 (0.216) (0.214) (0.222) (0.224) (0.211) (0.228) Nonpersonalist autocracy 0.384 0.477 0.250 0.384 0.564 0.239 (0.383) (0.394) (0.370) (0.438) (0.479) (0.434) Personalist autocracy −0.397 −0.153 −0.561 −0.594 −0.210 −0.745 (0.385) (0.421) (0.407) (0.476) (0.532) (0.499) State mediator 0.199 0.166 0.279 0.430 0.373 0.511 (0.320) (0.346) (0.335) (0.349) (0.374) (0.369) IO mediator 0.209 0.349 0.369 0.584 0.742** 0.760** (0.317) (0.298) (0.300) (0.368) (0.361) (0.361) NGO/grassroots mediator 0.480 0.520* 0.672** 0.777** 0.786** 0.945** (0.330) (0.304) (0.326) (0.377) (0.364) (0.379) Constant −0.796** −0.744* −0.738* −0.871** −0.795 −0.861* (0.338) (0.419) (0.441) (0.401) (0.503) (0.511) Observations 75 75 75 75 75 75 Notes: Robust standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01. Open in new tab Additional analyses are presented in the supplementary files. These include models using alternative measures of the DVs and key independent variables, alternative specifications of controls, models with additional controls, and tests to determine if civilian reliance's effect is conditional on civilian costs. We refer the reader to the supplementary files for these tests but note that all findings remain robust in these analyses. Conclusion This article has argued that the design of peace agreements is strategic. Rather than simply reflecting the public demands espoused by rebel leaders at conflict's start, peace agreements vary in terms of their favorability to rebel constituents. Empirical tests using original data on who benefits from the terms of peace demonstrate that rebel groups with strong ethnic support bases sign agreements that are significantly more populist in nature, while rebel groups with access to external sanctuaries sign agreements that are less beneficial to their constituencies. As one of the first studies to directly examine the determinants of peace agreement design, this project addresses important gaps in existing research. It suggests that research focusing on the effects of agreement design are missing part of the story by often treating the terms of settlement as exogenous. This study highlights the importance of accounting for settlement design, as endogenous design may have implications for the state of knowledge on implementation, conflict recurrence, post-conflict democratization, public support for peace, and the emergence of spoilers. Further, by loosening the unitary actor assumption, this study presents a novel way of conceptualizing settlement design and answers an interesting puzzle: why the terms of peace often do not reflect the type of demands rebel groups make at conflict's start. By focusing on who, within a rebel group/support base, benefits from the terms of peace, this project identifies novel variation in settlement design that previous studies have overlooked. In particular, our approach has implications for research on power sharing. Our data show that there is substantial variation across power-sharing provisions in terms of who primarily benefits. Sometimes power sharing involves broad-based reforms such as electoral proportional representation or the transfer of resources and powers to an autonomous region, but at other times, it means integrating rebel elites into preexisting governing structures by providing them with guaranteed positions. This is important because different power-sharing strategies (i.e., elite-focused versus public-focused) could have significant consequences for the long-term viability of the peace, a question that should be addressed in future research. Additionally, by demonstrating that agreement design is a function of rebel-civilian relationships, we contribute to emerging research on how rebel governance affects subsequent group behavior (Weinstein 2007) and how rebel leaders affect conflict dynamics (Prorok 2016a, 2016b; Uzonyi and Wells 2016). Finally, our theory and novel data are pertinent for understanding additional questions about peace agreement design, as well as how design affects implementation and conflict recurrence. For example, the implementation of public benefits may depend on private benefits for rebel leaders/elites, suggesting that settlements with certain combinations of private and public benefits may improve the likelihood of implementation. Understanding agreement design, therefore, likely has implications for durable peace. Future research should take advantage of our data and build on our theoretical argument to explore how “who benefits” from peace affects longer-term outcomes, such as implementation success, the risk of conflict recurrence, spoiling behavior, and democratization. Supplementary Information Supplementary information is available at the International Studies Quarterly data archive. Footnotes 1 This does not mean rebels will appeal to all civilians, but that they will appeal to some civilian constituency (e.g., co-ethnics), rather than framing their goals as exclusively benefiting group members. 2 Peace agreement provisions involve public benefits for the rebel constituency if civilian supporters have direct, first-hand access to the provision's benefits and if they are the primary beneficiary of that provision. Examples and details are provided below. We use the term public benefits to refer to benefits for the rebel constituency. We use the terms private, group-level, and member benefits/payoffs interchangeably to refer to benefits for rebel group members. 3 Group members’ preferences may vary depending on whether they are high or low-commitment types (Weinstein 2007). On average, however, we expect group members to prefer group-level benefits. 4 Populations that are more negatively affected by war may not require as many public benefits to continue supporting the rebels, because the value of peace is so high to them: the peace that group-level payoffs produce may be enough. We test this possibility in Table A6 in the supplementary files by interacting civilian reliance with an indicator for violence against civilians. 5 Thus, settlement terms result from a dynamic bargaining process, wherein finding a mutually acceptable deal requires overlap between government and rebel preferences. We control for government preferences in our analysis to help isolate the effect of rebels’ constituencies. 6 Our argument relates to selectorate theory in several ways (Bueno De Mesquita et al. 2003): rebel leaders are forward-thinking, providing benefits to groups that are most likely to keep them in power (i.e., the equivalent of their winning coalition). Also, like selectorate theory, when a leader's key constituency is larger, he will provide them with more public rather than private benefits. We hesitate to draw too close a connection to selectorate theory, however, because the presence of alternative local authorities (i.e., government and possibly other rebel groups) means that exit may be a more viable option for constituents than selectorate theory suggests, with implications for how loyalty relates to provision of benefits. Further, rebel leader selection is less institutionalized than state leader selection. Unlike selectorate theory, rebel leaders can be highly civilian-reliant even if civilians are not directly involved in their selection through institutionalized processes such as elections. 7 This mechanism requires civilians to have information about the terms of negotiations while they are ongoing. Arguably, this may not always be the case, as some negotiations occur in secret. However, in some cases, citizens have real-time information about the terms of negotiations, particularly if civil society groups are involved in the talks (e.g., Cabindan Forum for Dialogue, discussed below) or if one side decides to leak information for strategic reasons. Importantly, even if civilians do not know the terms of settlement during talks or only have limited information, civilian-reliant leaders still have incentives to pursue benefits for civilians because of anticipated future costs, once the terms of peace are revealed during implementation (see the political survival argument below). 8 Some final agreements include multiple rebel groups, while in other cases a dyad signs more than one final agreement. Each of these observations is recorded as a separate peace process, resulting in eighty dyadic peace processes. Due to missing data, our analysis includes seventy-five observations. 9 See Appendix B in the supplementary files for more detailed information. 10 The data exclude regulatory/administrative provisions, focusing instead on substantive provisions. 11 See Appendix C in the supplementary files for more information on coding rules and examples. 12 We prefer the strict measure, as we think it better captures how audiences view the terms of peace. However, all results are consistent across both measures. 13 A proportion is preferred to a count of public benefits because it ensures we are not simply capturing more detailed agreements in our results. 14 Sanctuary is unique in this respect. Other types of support, such as funding, do not clearly preclude close ties between rebels and civilians. However, adding external funding to our analysis does not affect our results (see Table A5 in the supplementary files). 15 Correlations among the three civilian reliance proxies range from 0.06 to 0.16. Including all three in one model does not affect the results (Table A3). 16 Unfortunately, we do not have data on rebel demands. Information on the number of rebel demands does exist for Africa, but this measure is less ideal than conflict type because it tells us nothing about the content of demands. 17 Results hold if we drop ethnic war from this control (see Table A3 in the supplementary files). 18 The base category combines irregular and symmetric nonconventional (SNC) conflicts. Results are consistent if we alter the base category. 19 This variable is limited because OSV data before 1989 are not available. However, our findings remain consistent if we drop this variable. 20 Results hold if we cluster standard errors by dyad or peace agreement. 21 The second stage of the Heckman models use ordinary least squares (OLS). Selection-equation variables are discussed in the supplementary files. The selection model allows us to indirectly account for the fact that settlement is a dynamic bargaining process in which combatants’ bargaining positions are constrained by their key constituencies, and that selection into settlement is potentially a nonrandom result of that process. 22 Figure A1 in the supplementary files shows the balance of the confounders across our three main independent variables before and after CBPS matching. Authors listed alphabetically. Equal authorship implied. Deniz Cil is a postdoctoral scholar at the Center for International Development and Conflict Management (CIDCM) at the University of Maryland. Her research focuses on civil war dynamics, conflict resolution, peace agreement implementation, and peacekeeping. Alyssa K. Prorok is an assistant professor of political science at the University of Illinois at Urbana-Champaign and currently a Linowes Fellow at the Cline Center for Advanced Social Research. Her research focuses on civil conflict dynamics and resolution and international law. Notes Author's note: We thank Jacob Aronson, Kyle Beardsley, Michael Findley, Nicholas Grossman, Caroline Hartzell, Paul Huth, Carla Martinez Machain, Molly Melin, Elizabeth Menninga, Lindsay Reid, Jakana Thomas, Gary Uzonyi, and participants in the 2018 Folke Bernadotte Academy Research Workshop in Bogota, Colombia for valuable comments and feedback. Part of the this work was supported by the National Science Foundation (grant no 1424033) and the Smith Richardson Foundation (grant no 20140298). The views and findings expressed in this paper are those of the authors alone. References Abdullah Ibrahim . 1998 . “Bush Path to Destruction: The Origin and Character of the Revolutionary United Front/Sierra Leone.” The Journal of Modern African Studies 36 ( 2 ): 203 – 35 . Google Scholar Crossref Search ADS WorldCat Africa Confidential . 2006 . “Cabinda Dreaming: The Fight for Cabindan Independence Is Being Put to Bed By the Luanda Regime.” Africa Confidential , August 25. https://www.africa-confidential.com/article-preview/id/1851/Cabinda_dreaming . OpenURL Placeholder Text WorldCat Arjona Ana . 2014 . “Wartime Institutions: A Research Agenda.” Journal of Conflict Resolution 58 ( 8 ): 1360 – 89 . Google Scholar Crossref Search ADS WorldCat Arjona Ana , Kasfir Nelson , Mampilly Zachariah . 2015 . Rebel Governance in Civil War . Cambridge : Cambridge University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Badran Ramzi . 2014 . “Intrastate Peace Agreements and the Durability of Peace.” Conflict Management and Peace Science 31 ( 2 ): 193 – 217 . Google Scholar Crossref Search ADS WorldCat Balcells Laia , Kalyvas Stathis N. . 2014 . “Does Warfare Matter? Severity, Duration, and Outcomes of Civil Wars.” Journal of Conflict Resolution 58 ( 8 ): 1390 – 418 . Google Scholar Crossref Search ADS WorldCat Bogaards Matthijs . 2013 . “The Choice for Proportional Representation: Electoral System Design in Peace Agreements.” Civil Wars 15 ( suppl 1 ): 71 – 87 . Google Scholar Crossref Search ADS WorldCat Bueno De Mesquita Bruce , Smith Alastair , Morrow James D. , Siverson Randolph M. . 2003 . The Logic of Political Survival . Cambridge, MA : MIT Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Chimbano Chola . 1990 . “ANC Says South Africa Must Lift State of Emergency Before Talks.” Reuters News, February 11 . OpenURL Placeholder Text WorldCat Cil Deniz . 2016 . “The Implementation of Peace Agreements Following Civil Wars and Post-Conflict Outcomes.” PhD dissertation, University of Maryland, College Park, MD . OpenURL Placeholder Text WorldCat Cowell Alan . 2018 . “Afonso Dhlakama, Mozambique's Opposition Leader, Dies At 65.” The New York Times, May 6. https://www.nytimes.com/2018/05/06/obituaries/afonso-dhlakama-dead-mozambique.html . OpenURL Placeholder Text WorldCat Croco Sarah E. 2011 . “The Decider's Dilemma: Leader Culpability, War Outcomes, and Domestic Punishment.” American Political Science Review 105 ( 3 ): 457 – 77 . Google Scholar Crossref Search ADS WorldCat Croicu Mihai , Melander Erik , Nilsson Marcus , Wallensteen Peter . 2013 . “Mediation and Violence: Searching for Third Party Intervention That Matters.” Paper presented at the Annual Meeting of the International Studies Association, San Francisco, CA . OpenURL Placeholder Text WorldCat Cunningham David , Gleditsch Kristian Skrede , Salehyan Idean . 2013 . “Non-State Actors in Civil Wars: A New Dataset.” Conflict Management and Peace Science 30 ( 5 ): 516 – 31 . Google Scholar Crossref Search ADS WorldCat Doyle Michael W. , Sambanis Nicholas . 2006 . Making War and Building Peace: United Nations Peace Operations . Princeton, NJ : Princeton University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Eck Kristine , Hultman Lisa . 2007 . “One-Sided Violence Against Civilians in War: Insights from New Fatality Data.” Journal of Peace Research 44 ( 2 ): 233 – 46 . Google Scholar Crossref Search ADS WorldCat Esipisu Manoah . 1992 . “Mozambique Rebel Leader Says Chissano Starving Millions.” Reuters News, August 9 . OpenURL Placeholder Text WorldCat Fortna Virginia Page . 2008 . Does Peacekeeping Work? Shaping Belligerents’ Choices after Civil War . Princeton, NJ : Princeton University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Geddes Barbara , Wright Joseph , Frantz Erica . 2014 . “Autocratic Breakdown and Regime Transitions: A New Data Set.” Perspectives on Politics 12 ( 2 ): 313 – 31 . Google Scholar Crossref Search ADS WorldCat Gelman Andrew , Hill Jennifer . 2006 . Data Analysis Using Regression and Multilevel/Hierarchical Models . 1st edition. New York : Cambridge University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Goemans H.E. 2000 . War and Punishment: The Causes of War Termination and the First World War . Princeton, NJ : Princeton University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Harbom Lotta , Melander Erik , Wallensteen Peter . 2008 . “Dyadic Dimensions of Armed Conflict, 1946—2007.” Journal of Peace Research 45 ( 5 ): 697 – 710 . Google Scholar Crossref Search ADS WorldCat Hartzell Caroline , Hoddie Matthew . 2003 . “Institutionalizing Peace: Power Sharing and Post-Civil War Conflict Management.” American Journal of Political Science 47 ( 2 ): 318 – 32 . Google Scholar Crossref Search ADS WorldCat Hartzell Caroline , Hoddie Matthew . 2007 . Crafting Peace: Power-Sharing Institutions and the Negotiated Settlement of Civil Wars . University Park, PA : Pennsylvania State University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Högbladh Stina . 2011 . “Peace Agreements 1975-2011: Updating the UCDP Peace Agreement Dataset.” In States in Armed Conflict 2011, 2012 , edited by Therése Petterson , Themnér Lotta , 39 – 56 . Uppsala : Uppsala University . OpenURL Placeholder Text WorldCat Högbladh Stina , Pettersson Therése , Themnér Lotta . 2011 . “External Support in Armed Conflict 1975–2009: Presenting New Data.” Paper presented at the Annual International Studies Association Convention, Montreal, Canada, March . OpenURL Placeholder Text WorldCat Huang Reyko . 2016 . The Wartime Origins of Democratization: Civil War, Rebel Governance, and Political Regimes . Cambridge : Cambridge University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Jo Hyeran . 2015 . Compliant Rebels . New York : Cambridge University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Joshi Madhav , Quinn J. Michael . 2015 . “Is the Sum Greater than the Parts? The Terms of Civil War Peace Agreements and the Commitment Problem Revisited.” Negotiation Journal 31 ( 1 ): 7 – 30 . Google Scholar Crossref Search ADS WorldCat Kalyvas Stathis N. 2006 . The Logic of Violence in Civil War . Cambridge : Cambridge University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Kalyvas Stathis , Balcells Laia . 2010 . “International System and Technologies of Rebellion: How the End of the Cold War Shaped Internal Conflict.” American Political Science Review 104 ( 3 ): 415 – 29 . Google Scholar Crossref Search ADS WorldCat Kaplan Oliver . 2013 . “Protecting Civilians in Civil War: The Institution of the ATCC in Colombia.” Journal of Peace Research 50 ( 3 ): 351 – 67 . Google Scholar Crossref Search ADS WorldCat Lee SungYong , MacGinty Roger , Joshi Madhav . 2016 . “Social Peace Vs. Security Peace.” Global Governance 22 ( 4 ): 491 – 512 . Google Scholar Crossref Search ADS WorldCat Lichbach M.I. 1995 . The Rebel's Dilemma . Ann Arbor, MI : University of Michigan Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Mabe Pule . 2018 . “Freedom Charter Remains a Call to Action - ANC - POLITICS | Politicsweb.” Politicsweb, June 26. http://www.politicsweb.co.za/politics/freedom-charter-remains-a-call-to-action–anc . OpenURL Placeholder Text WorldCat Mampilly Zachariah Cherian . 2011 . Rebel Rulers: Insurgent Governance and Civilian Life during War . Ithaca, NY : Cornell University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Marshall M.G. , Jaggers K. . 2010 . “Polity IV Project: Political Regime Characteristics and Transitions, 1800–2010.” University of Maryland. http://www.cidcm.umd.edu/Inscr/Polity . OpenURL Placeholder Text WorldCat Matanock Aila M. 2017 . Electing Peace: From Civil Conflict to Political Participation . Cambridge : Cambridge University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Mattes Michaela , Rodríguez Mariana . 2014 . “Autocracies and International Cooperation.” International Studies Quarterly 58 ( 3 ): 527 – 38 . Google Scholar Crossref Search ADS WorldCat Mattes Michaela , Savun Burcu . 2009 . “Fostering Peace After Civil War: Commitment Problems and Agreement Design.” International Studies Quarterly 53 ( 3 ): 737 – 59 . Google Scholar Crossref Search ADS WorldCat Mattes Michaela , Savun Burcu . 2010 . “Information, Agreement Design, and the Durability of Civil War Settlements.” American Journal of Political Science 54 ( 2 ): 511 – 24 . Google Scholar Crossref Search ADS WorldCat McAdam Doug . 2010 . Political Process and the Development of Black Insurgency, 1930–1970 . Chicago : University of Chicago Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Melville Chris . 2003 . “Ivorian Rebels Agree to Give Up Key Cabinet Portfolios.” IHS Global Insight, March 7 . OpenURL Placeholder Text WorldCat Menninga Elizabeth J. , Reid Lindsay . 2017 . “Mediator Leverage and Context-Specific Peace Agreements.” Paper presented at the International Studies Association Annual Convention, Baltimore . OpenURL Placeholder Text WorldCat Middleton John , Miller Joseph C. , eds. 2008 . “Cabinda, Angola.” In New Encyclopedia of Africa , 2nd ed., vol. 1 , 293 – 4 . New York : Charles Scribner's Sons . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Nalepa Monika . 2010 . Skeletons in the Closet: Transitional Justice in Post-Communist Europe . Cambridge : Cambridge University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Papke Leslie E. , Wooldridge Jeffrey M. . 1996 . “Econometric Methods for Fractional Response Variables with an Application to 401 (k) Plan Participation Rates.” Journal of Applied Econometrics 11 ( 6 ): 619 – 32 . Google Scholar Crossref Search ADS WorldCat Pearce Justin . 2005 . An Outbreak of Peace: Angola's Situation of Confusion . Claremont, South Africa : New Africa Books . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Prorok Alyssa K. 2016a . “Leader Incentives and Civil War Outcomes.” American Journal of Political Science 60 ( 1 ): 70 – 84 . Google Scholar Crossref Search ADS WorldCat Prorok Alyssa K. . 2016b . “Led Astray: Leaders and the Duration of Civil War.” Journal of Conflict Resolution , December. http://dx.doi.org/10.1177/0022002716680265 . OpenURL Placeholder Text WorldCat Renfrew Barry . 1990 . “ANC, Government Negotiators Express Optimism About Talks.” The Associated Press, May 3 . OpenURL Placeholder Text WorldCat Reuters News . 1991 . “Mozambique Peace Talks Restarted After Three-Week Stall.” Reuters News , May 28 . OpenURL Placeholder Text WorldCat Reuters News . 1992 . “Mozambican Rebels Ready to Stop War.” Reuters News , July 4 . OpenURL Placeholder Text WorldCat Siona Casimiro . 2006 . “Cabinda Church and Civil Representatives Reject Reported Deal with Angola Government.” Associated Press International, July 5 . OpenURL Placeholder Text WorldCat Smith Alastair , Stam Allan C. . 2004 . “Bargaining and the Nature of War.” Journal of Conflict Resolution 48 ( 6 ): 783 – 813 . Google Scholar Crossref Search ADS WorldCat Stedman Stephen John . 1997 . “Spoiler Problems in Peace Processes.” International Security 22 ( 2 ): 5 – 53 . Google Scholar Crossref Search ADS WorldCat Svensson Isak . 2009 . “Who Brings Which Peace? Neutral Versus Biased Mediation and Institutional Peace Arrangements in Civil Wars.” Journal of Conflict Resolution 53 ( 3 ): 446 – 69 . Google Scholar Crossref Search ADS WorldCat The New Humanitarian . 2010 . “The Death of One Man Does Not End a War.” The New Humanitarian , April 26. http://www.thenewhumanitarian.org/news/2010/04/26/death-one-man-does-not-end-war . OpenURL Placeholder Text WorldCat Themnér Anders . 2011 . Violence in Post-Conflict Societies: Remarginalization, Remobilizers and Relationships . New York : Routledge . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Thomas Jakana L. 2012 . “Don't Stop Till You Get Enough: Rebel Group War Aims and Conflict Resolution in Civil Wars.” PhD dissertation, Pennsylvania State University. https://etda.libraries.psu.edu/catalog/15203 . OpenURL Placeholder Text WorldCat Uzonyi Gary , Wells Matthew . 2016 . “Domestic Institutions, Leader Tenure and the Duration of Civil War.” Conflict Management and Peace Science 33 ( 3 ): 294 – 310 . Google Scholar Crossref Search ADS WorldCat Wagner R. Harrison . 2000 . “Bargaining and War.” American Journal of Political Science 44 ( 3 ): 469 – 84 . Google Scholar Crossref Search ADS WorldCat Walsh James Igoe , Conrad Justin M. , Whitaker Beth Elise , Hudak Katelin M. . 2018 . “Funding Rebellion: The Rebel Contraband Dataset.” Journal of Peace Research 55 ( 5 ): 699 – 707 . Google Scholar Crossref Search ADS WorldCat Walter Barbara F. 2002 . Committing to Peace: The Successful Settlement of Civil Wars . Princeton, NJ : Princeton University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Weinstein Jeremy M. 2007 . Inside Rebellion: The Politics of Insurgent Violence . Cambridge : Cambridge University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Wucherpfennig Julian , Metternich Nils W. , Cederman Lars-Erik , Gleditsch Kristian Skrede . 2012 . “Ethnicity, the State, and the Duration of Civil War.” World Politics 64 ( 1 ): 79 – 115 . Google Scholar Crossref Search ADS WorldCat © The Author(s) (2020). Published by Oxford University Press on behalf of the International Studies Association. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Selling Out or Standing Firm? Explaining the Design of Civil War Peace Agreements
JF - International Studies Quarterly
DO - 10.1093/isq/sqaa010
DA - 2020-06-01
UR - https://www.deepdyve.com/lp/oxford-university-press/selling-out-or-standing-firm-explaining-the-design-of-civil-war-peace-0ranqN1p8Q
DP - DeepDyve
ER -