Is the Sum Greater than the Parts? The Terms of Civil War Peace Agreements and the Commitment Problem Revisited

In a seminal work on negotiated settlements in civil wars, Barbara Walter (1997: 335) described a long‐standing historical pattern (going back a century): “Unlike interstate wars, civil wars rarely end in negotiated settlements.” Walter argued that this reflected fundamental differences in the nature of warfare within nations (i.e., civil war) when compared with warfare between nations. In civil wars, she found the prisoners’ dilemma to be inescapable. The incentives to cheat one's rival if given the chance are just too high: “civil war negotiations rarely end in successful peace settlements because credible guarantees on the terms of the settlement are almost impossible to arrange by the combatants themselves” (1997: 335). According to her theory of credible commitment, negotiations between civil war opponents produce successful agreements only when the terms are enforced by a third party.¹

Credible commitment theory has, arguably, become the dominant theoretical approach to negotiations in civil wars in the field of political science, and the empirical literature has largely developed around using various devices to solve problems of enforcement and vulnerability. Researchers have focused on three particular types of enforcement devices that civil war actors have negotiated:

• third‐party enforcement and/or peacekeeping operations (Walter 1997; Doyle and Sambanis 2000, 2006; Fortna 2004; Quinn, Mason, and Gurses 2007; Mattes and Savun 2010; Joshi 2013);

• power‐sharing provisions that divide political or military power to give the actors influence in the post‐accord period (Hartzell and Hoddie 2003; Jarstad and Nilsson 2008; DeRouen, Lea, and Wallensteen 2009); and

• monitoring and verification mechanisms that signal commitment to abide by the agreement and create an environment of accountability and compliance (Mattes and Savun 2010).

Remarkably, the long‐standing historical trends regarding how wars end that we noted above have reversed over the last two decades. Today, the majority of civil wars end in negotiated peace agreements, two thirds of which successfully end armed conflict for at least five years, and international relations scholars are now working to explain “the demise of peace treaties in interstate war” between nation‐states (Fazal 2013). Because of the magnitude of changes that have taken place in the international system over the last twenty‐five years, we believe the time is ripe to revisit the commitment problem in civil war negotiations. We suggest that issues of enforcement and vulnerability may not be as critical as they once were and that negotiators and mediators who help design contemporary agreements to end civil wars should emphasize crafting an agreement that stakeholders will perceive has the potential to accomplish large‐scale group objectives. For strategic and practical reasons that we discuss in this article, we believe that agreements with a greater quantity of reforms that fall across different policy sectors are better able to demonstrate that they have this potential.

Theories of bargaining in the civil war context have sought to explain both the scarcity of negotiated agreements and their high rates of failure — a historical pattern that has fundamentally changed in recent decades. In civil wars fought between 1940 and 1990, for example, Walter (1997: 346) found only eight negotiated settlements that successfully ended violence between the signatories for a period of five years or more. On the other hand, a 2012 survey of nearly two hundred civil war peace agreements negotiated from 1975 to 2011 found that 60 percent of those agreements ended armed conflict between the signatories for at least five years (Högbladh 2012). In fact, peacebuilding efforts seem to be ending conflicts more quickly than new ones are emerging, with a consequent decline in the incidence of ongoing civil wars over the past decade (see Goldstein 2011; Themnér and Wallensteen 2012). There has been no meaningful decrease in the number of new conflicts emerging each year, but their duration is diminishing, so the effectiveness of negotiation efforts is a plausible explanation for the overall decline.

Why has the incidence of civil war peace agreements increased, and why are they more successful than in the past? In a study of more than three hundred civil war peace agreements signed between 1990 and 2005, Christine Bell (2006: 373–378) attributed the “rise of the peace agreement” in civil war settings to recent changes in “international attention devoted to such conflict.” In particular, she cited increased “mediation” and “negotiation” efforts; increased acceptance of “standard settlement designs”; increased acceptance of agreements as “constitution making”; “the emergence of legal standards” addressing peace agreements and their enforcement; and the increasingly accepted view among domestic and international actors that peace agreements are “legal documents.”

Because the rise in civil war peace agreements has occurred concurrently with “the demise of peace treaties in interstate war” (Fazal 2013: 695, emphasis added), it seems reasonable to ask whether the underlying causes are related. Tanisha Fazal (2013: 696, 701) has attributed the demise of peace agreements in wars between nations to a “new era of accountability,” characterized by the “codification of norms” and “the emergence of a strong and growing canon” of international law regulating the conduct of actors involved in war and the termination of wars. Fazal has argued that this heightened level of “accountability” creates “disincentives for states to conclude formal peace treaties” (696) because they know that they will be “unequivocally obliged to pay the costs of compliance and to bear the consequences of noncompliance” (699). Her argument for the decline of treaties between nations is the mirror image to the arguments put forth by Bell (2006) who seeks to explain the proliferation of peace agreements in wars within nations.

These arguments by Fazal and Bell strongly suggest that changes in international norms and laws, as well as the increased prevalence and competence of peacebuilding professionals, now require conflict actors to have a greater commitment to negotiate as well as a greater willingness to comply with and enforce treaties. While actors in interstate wars seek to avoid accountability and compliance, civil war actors seem to embrace the opportunities that these new dynamics create.

Bell (2006: 378) argued that the modern “structure and language of peace agreements suggest that the parties mutually view them as legal documents.” If conflict actors view agreements as legally binding and expect high levels of compliance, then we would expect that stakeholders would seek to ensure that the terms being negotiated will achieve their intended purpose. Hence, any accord reached would be judged primarily on its potential to accomplish its objectives as perceived by the stakeholders. When the terms of an agreement are deemed insufficient, we would also expect factions or spoiler groups to emerge to oppose the agreement and cultivate opposition to it (Stedman 1997, 2001). This can occur on both sides. If stakeholders perceive an agreement to be insufficient, we would expect them to respond by calling for further negotiations in an effort to add to the current agreement or, alternatively, that there would be an impasse, which would be followed by renewed armed conflict to force additional bargaining.

How do influential stakeholders with the capacity to affect the course of a peace process — for better or for worse — evaluate the potential of an agreement to accomplish its larger objectives? We suggest that stakeholders will evaluate the potential of an accord by strategically evaluating its terms, in particular, their policy implications. The two main stakeholder groups in such conflicts are typically those influential members of the opposition who can steer opinion and mobilization efforts toward or away from the accord and those individuals within the government or the government's core constituency who lose benefits and power as a result of the reforms included in the accord. The two stakeholder groups have little or no overlap because civil wars, almost by default, are fought between groups currently in government power and groups currently excluded from government power. Peace agreements, and the policy changes they include, typically make deliberate and purposeful attempts to alter existing power structures and government expenditures. Thus, civil war agreements can be broadly understood as instruments that leaders of excluded groups use to divert some power and resources away from the government's constituency and toward their own.

Although the stakeholders to an accord include government and opposition constituencies, in this article we model negotiations primarily through the eyes of the opposition because most provisions found in peace agreements deal with reforming the government. Civil war negotiations determine which reforms the government must agree to in order to persuade the rebel group to disarm, demobilize, and pursue their interests through nonviolent means. Although the opposition typically also engages in concession‐making behavior, the nature of that engagement primarily involves setting the acceptable parameters of the government's concession‐making behavior. An agreement with few provisions will leave most government sectors unaffected and is unlikely to convince the broader rebel constituency that its benefits outweigh the costs and risks associated with disarming and demobilizing.

Thus, the potential benefit to the opposition of a particular item in the accord is the inverse of its expense to the affected government stakeholder group — an agreed‐upon term involves a transfer that generates a gain for one side and loss to the other. How can we compare the relative costs of terms for government stakeholders in one agreement versus another? Because governments measure costs primarily in terms of lost sovereignty, that is, the loss of power and authority over a certain jurisdictional space — and this applies to all levels, branches, units, subunits, departments, and offices — no government unit will voluntarily give up its power nor reduce its jurisdiction. Thus, many typical challenges to sovereignty are customarily resisted by governments.

Because the empowerment of individuals or groups via the codification of “rights” serves as a limit to government power and its application, governments have historically resisted the granting of such rights, which usually must be “won” after a struggle. Because governing elites typically use institutions to sustain the political allegiance of the constituency groups that keep them in power, they typically also resist institutional reform efforts.

Governments are not unitary actors but Janus‐faced organizations with competing units and agencies controlled by multiple actors. Consequently, engaging in reforms across multiple policy areas is more costly to the government than implementing numerous reforms in just one policy sector because nearly every reform that could be offered in negotiations will challenge the interests of a particular group who will band together to oppose it. The nature of civil war, for example, the fact that civil wars typically arise following years if not decades of aggrievement on the part of groups who believe themselves to be disenfranchised or persecuted all but guarantees that those government sectors targeted for the most significant reforms are the sectors likely to be most hostile to those reforms. For example, when a peace agreement calls for police reform and the creation of integrated coethnic units, the opposition seeks reforms that are likely to be fiercely opposed.

Thus, to accommodate the opposition's reform demands, government negotiators must sometimes engage in negotiations at a second level in order to gain compliance from the actors who yield power in those policy jurisdictions, and government negotiators may find it necessary to offset the costs borne by those actors. Thus, we argue that enacting a greater quantity of reforms ranging across a greater number of multiple policy areas within government (i.e., legislative reform, judicial reform, civil administration reform, and military reform) requires more significant government commitment to the peace process and to the implementation of the terms of the agreement.

As a measure of an agreement's potential, we believe that this conceptual approach is highly generalizable because whatever the objectives of the opposition group or agreement, they are more likely to be achieved if pursued through numerous policy sectors that deal with some dimension of the dispute. Going into negotiations, the opposition knows that the sectors affected by the reforms they seek will resist implementing those reforms. As Caroline Hartzell and Matthew Hoddie (2003: 321) note, “signatories to an agreement are likely to recognize that in the often difficult and contentious process of transitioning from war to peace there is the potential that some provisions of a peace agreement will not be implemented.”

Negotiators representing the opposition therefore will seek out ways to prevent hostile factions from blocking reforms by engaging in hedging techniques. By spreading the objectives of the agreements across multiple areas of government, hedging techniques are designed to minimize the risks that a failure to implement some provisions of an agreement will undermine the overall objectives of the agreement. An opposition group seeking greater autonomy, for example, may pursue that goal through numerous channels, including constitutional reform, boundary demarcations, decentralization, civilian administration reform, police reform, and educational reform, all of which transfer power in multiple and diffuse ways.

What happens when the terms of a peace agreement are viewed as insufficient by the opposition? Such dissatisfaction can lead to the fragmentation of the opposition and create a need for more complicated bilateral negotiations with multiple groups; it can also lead to renewed violence as the opposition seeks greater leverage for future rounds of negotiations. If all negotiable opportunities have been exploited — and neither side is willing to make additional concessions — the parties will reach a “turning point” in the negotiations when “both greater conflict and enhanced cooperation are possible options” (Sisk 2009: 45). As Timothy Sisk (2009: 46) has noted, “violence is the main type of punitive tactic, or sanction, in a bargaining relationship in internal conflicts” and renewed fighting may encourage either or both parties to resume negotiations. Although this argument — the cycle of violence and negotiation and negotiation and violence — is routinely made in the bargaining literature (i.e., Fearon 1995; Wagner 2000; Filson and Werner 2002; Sisk 2009), it has gone largely unexamined in the empirical literature on civil war negotiations.

The terms negotiated in civil war negotiations present the most measurable data that actors can use to determine whether the accord will accomplish their group objectives. The literature on “spoilers” points specifically to the terms of agreement as the main source of variation that explains opposition to particular agreements. Because “parties in civil wars differ in their goals and commitment” (Stedman 1997: 10), an accord must do more than just co‐opt rebel elites to avoid falling apart once spoilers and veto players oppose it (Cunningham 2006). We argue that the number of reforms that fall under multiple policy domains within a peace agreement can serve as a rough measure of the scope of the government's commitment to the negotiations and the likelihood that the accord will accomplish the opposition's objectives. Enacting reforms across multiple different policy sectors protects the broader objectives of the agreement against hostile factions within the government that will seek to block implementation. According to this reasoning, the parties to an agreement are less likely to seek to negotiate a further agreement if the current agreement contains reforms across a greater number of different policy areas. But if the terms in an agreement are widely viewed as insufficient by influential sectors of the opposition, they may push to resume negotiations to find more acceptable terms in a subsequent agreement, and this carries the risk of renewed violence. If the government refuses to concede, renewed violence becomes likely. We thus propose the following two hypotheses:

We tested our hypotheses using two existing data sources that include information about the terms or provisions contained in peace agreements: the Peace Agreement Dataset created by the Uppsala Conflict Data Program (UCDP) at the University of Uppsala in Sweden (Högbladh 2012) and data from the Peace Accords Matrix Project created at the Kroc Institute for International Peace Studies at the University of Notre Dame in Indiana (Joshi and Darby 2013).²

The term “peace agreement” can refer to many different types of agreements:

• ceasefire agreements that regulate hostilities (often temporarily);

• process agreements that outline how negotiations will unfold but can include what issues are negotiable;

• partial agreements that deal with some substantive conflict issues but are generally intended to be followed by future accords; and

• full or comprehensive agreements that attempt to permanently end the conflict.

For this study, we used a sample of 196 peace agreements (Högbladh 2012) that includes process, partial, and full peace agreements negotiated between 1975 and 2011 for intrastate conflicts that produced at least twenty‐five battle‐related deaths per year.

In Hypothesis One, we wanted to find out whether enacting a greater number of reforms across different policy areas decreases the likelihood that the same parties will negotiate an additional accord in the future. Our first dependent variable, subsequent peace accord, refers to whether a peace agreement was followed by another peace agreement (within the same conflict ID) before 2011, the last year of observation (coded “1” if followed by subsequent accord, otherwise “0”). Among the 196 accords we examined, 146 were followed by future negotiations that produced a subsequent agreement.

The second dependent variable, recurrence of conflict, refers to whether an agreement was followed by the recurrence of at least minor armed conflict between the signatories to the accord within five years. Out of 196 peace agreements examined, the same parties returned to at least minor armed conflict seventy‐eight times (40 percent of the time) within five years.

Our primary independent variable is the quantity of multiple types of provisions contained in each agreement. Numerous studies on peace accords and their content (Hartzell and Hoddie 2003, 2007; Mattes and Savun 2010) use the word “provision” to refer to discrete topics or issues in peace agreements. The term “provision” is commonly used in legal instruments to stipulate that something be provided for or performed by one of the signatories, or to stipulate an amount of something to be set aside as a liability. We consider a provision to be a goal‐oriented reform or stipulation that something be provided, that is costly to one or more actors, and that falls under a relatively discrete policy domain (e.g., executive branch reform, demobilization, minority rights, police reform). We have examined twenty‐seven different kinds of provisions contained in the UCDP dataset:

• four provisions fell under the heading of conflict behavior (e.g., ceasefire regulation, integration within the army, and troop withdrawal);

• six provisions fell under the category of regulating a governmental incompatibility (e.g., political party, integration in civil service, elections, and power sharing in government);

• nine provisions involved regulation of a territorial incompatibility (e.g., autonomy, federalism, independence, referendum, and local government) and

• eight others fell under various miscellaneous categories (e.g., amnesty, prisoners release, and refugees).

The variable number of provisions (UCDP) ranged from a minimum of zero to a maximum of twenty‐one provisions in a single accord. For the shared subset of accords that were part of both the UCDP and PAM datasets, we substituted the quantity of PAM provisions. The number of provisions (PAM) ranged from a minimum of zero to a maximum of forty‐three different provisions in an agreement. The PAM project coded fifty‐one different types of provisions under five main categories:

• thirteen provisions fell under the category of institutions (e.g., transitional power‐sharing government, legislative branch reform, executive branch reform, and decentralization);

• eight provisions fell under the header of security (e.g., ceasefire, military reform, demobilization, reintegration, and police reform);

• fifteen provisions involved rights (e.g., citizenship, refugees, indigenous minority rights, self‐determination, women's rights, and media reform);

• seven provisions involved external arrangements (e.g., arms embargo and peacekeeping force); and

• eight provisions fell under the miscellaneous other category (e.g., ratification mechanism and refugees).

When evaluating whether an agreement was followed by an additional agreement and the effectiveness of the agreement at eliminating future armed conflict, we controlled for the number of dyads — previous research has shown that a greater number of dyads can make negotiated solutions more difficult to achieve (Cunningham 2006). We also controlled for whether the central incompatibility of the conflict was over territory or government policies using the UCDP data (Themnér and Wallensteen 2012) and for the degree of country‐level ethnic fractionalization (Alesina et al. 2003), which has been shown to increase the risk of civil war recurrence (Mason et al. 2011). (Ethnic fractionalization is a measure of ethnic diversity; it reflects the probability that two randomly selected individuals from a population belong to different ethnic groups.)

Similarly, the risk of renewed war is also affected by the duration and intensity of the civil war (Zartman 1989; Doyle and Sambanis 2000). For measures of conflict duration and conflict intensity we have used data provided by Lotta Themnér and Peter Wallensteen (2012). We coded conflict duration in days (logged), and coded conflict intensity as a “1” if the conflict reached the level of “war” at any time.

In addition, because democracies are less likely to experience conflict recurrence (Mason et al. 2011), we controlled for whether the country was a democracy in the year prior to the signing of the accord using data from the Polity IV project (Marshall and Jaggers 2012). The Polity IV measure reflects the country's level of electoral competitiveness and openness, the nature of general political participation, and the level of constraints on executive authority. We also controlled for state capacity because of the strong connections between state weakness and the risk of armed conflict (i.e., Collier 1999) and between state capacity and peace agreement implementation (DeRouen et al. 2010). We included several proxies for state capacity (all lagged one year): infant mortality rate and the natural log of population and gross domestic product (GDP) per capita (World Bank 2012).

Our independent variable (the quantity of different types of provisions) might be influenced by several confounding factors. Because the content or number of provisions in a peace agreement can be related to past agreements and the presence of mediation, we also believe it is important to control for these factors. The variable prior peace agreement is coded “1” if the agreement in our sample was preceded by any prior agreements and “0” otherwise. Previous studies (i.e., Gartner and Bercovitch 2006; Bercovitch and Gartner 2007) show the presence of third‐party mediation to have path‐dependent effects. Therefore, we controlled for mediation in civil wars by using the Civil War Mediation data used by Karl DeRouen, Jacob Bercovitch, and Paulina Pospieszna (2011). The mediation variable is coded “1” if mediation preceded the accord and otherwise “0.” Conflicts that generate more refugees and internally displaced persons create regional security concerns and more external involvement (see Salehyan 2009), which might lead to more concessions being made because of pressure from these kinds of actors. These two variables come from the World Bank (2012).

We have performed our statistical analysis using a set of logit models because our dependent variables are binary; the use of such models is well established in the conflict recurrence literature. We used robust standard errors clustered on the specific conflict. The results are presented in Table Two.

We find strong support for our first hypothesis that when agreements contain provisions across a greater number of policy areas, the conflict actors are less likely to pursue another agreement. Models One through Four in Table Two contain the variable number of provisions (UCDP) and offer substantial evidence in support of our argument. Models Five through Seven use number of provisions (PAM) as the main independent variable and the results show, again, agreements that featured a greater number of different types of provisions were less likely to be followed by subsequent agreements.

Of the control variables, we found that the longer the civil war had lasted, the more likely the current accord would be followed by a subsequent accord. This suggests that as a civil war becomes more protracted, each side sees its chances of victory declining. We also found that both measures of state capacity had significant effects, specifically, we found higher levels of infant mortality and lower GDP per capita to be associated with an increased likelihood of a subsequent agreement — this suggests that stronger states are more likely to resist attempts to negotiate subsequent agreements. As would be expected, process agreements were significantly more likely to be followed by future agreements. We found no consistent support that the remaining controls had any significant effects.

We not only argued that agreement insufficiency would lead to the resumption of negotiation activities to draft an additional agreement, but also that insufficiency could lead to renewed violence as part of the bargaining process. Table Three above presents our analysis on the question of whether having a greater range of provisions across policy areas reduces the risk of renewed armed conflict between the signatories to the agreement. Across all the models presented in Table Three, the number of different types of provisions has a highly significant effect.

Model One, which is a basic model containing only a control for prior peace accord, indicates that, as the number of different kinds of provisions in an agreement increases, the risk that the same parties that negotiated that document will return to war within five years is significantly reduced. Model Two expands upon this basic test by adding important controls for the number of dyads, territorial incompatibility, conflict duration, conflict severity, and regime characteristics. We see in the results that the effect of the number of different provisions in the accord is highly significant and that having prior agreements is associated with a reduced risk of violence. This suggests a complex and important relationship between the terms in a current agreement and how that agreement relates both to the past and to the future. The fact that a current agreement was preceded by a past agreement suggests that further negotiations were needed to address issues that were not sufficiently addressed in the previous agreement and that doing this reduced the incentive for further violence. This finding remains consistently significant across all remaining model specifications in Table Two.

In Model Three, the number of provisions in the accord remains highly significant even after we controlled for such potentially relevant and confounding factors as prior agreement, the presence of prior mediation, ethnic fractionalization, population size, and GDP per capita. Among the controls, only the presence of prior mediation showed a significant effect and was positively correlated with renewed fighting. This is consistent with the findings of prior studies that suggest that mediators are drawn to more difficult, protracted conflicts and have professional incentives to produce an agreement, regardless of quality (see Gartner and Bercovitch 2006; DeRouen and Möller 2013). In Model Four, we replaced GDP per capita with infant mortality rate (arguably a better indicator of poverty) and found that higher infant mortality significantly increased the risk of renewed violence. This is consistent with previous results that weak state capacity increases the risk of civil war recurrence (Quinn, Mason, and Gurses 2007). Earlier we found that weaker states were significantly more susceptible to demands for further negotiations toward another accord. This current finding suggests that the opposition group may use violence to force negotiations if the state is reluctant to negotiate any further. The control variables for the number of dyads, territorial incompatibility, conflict duration, regime type, ethnic fractionalization, and population size had no significant effect on recurrence. Models Five, Six, and Seven in Table Three replicate Models Two, Three, and Four using an alternative measure of the number of different types of agreement provisions taken from the Peace Accords Matrix project. As seen in Model Five, number of provisions (PAM) is highly significant and negatively correlated with post‐accord conflict recurrence. We ran several robustness tests and the results can be seen in Table Four below.

We must consider the possibility that our main independent variable may be masking some important internal dynamics that might offer an alternative theoretical account for our findings. The results suggest that peace agreements that include a greater range of provisions are less likely to be followed by either further agreements or renewed violence. It could be the case, however, that the relationship is being driven by the presence of particularly important provisions within the composite measure — perhaps it is not the number and range of provisions that matter, but the inclusion of certain particular provisions, and agreements that have a greater number of provisions are most likely to include the really important ones. The results might also reflect the fact that certain agreements are designed to permanently end the conflict and agreements intended to be final could be expected to contain a higher number of provisions and should be followed by fewer agreements and less violence.

These questions could be explored in several different ways. For example, we could include the most prominent provisions from prior research in a model as individual regressors. Another option is to change the composition of the sample so as to drop those accords most likely to contain these prominent provisions. In Models Eight and Nine, we included peacekeeping operations, local power sharing, political power sharing, and military power sharing as independent variables in the model as a way of gauging the relative importance of these provisions compared with the totality of the terms in the accord. As shown in Table Four, we found the coefficient for the number of different types of provisions in the agreement to be highly significant and negatively associated with civil war recurrence. The provisions for peacekeeping troops and local, political, and military power sharing do not reach statistical significance when included in the same model with the total number of provisions in the accord. The findings from Models Eight and Nine combined with the descriptive findings shown in Table One suggest that these provisions are not the primary mechanisms that conflict actors use to overcome commitment problems in negotiations because most agreements lack these provisions (as shown in Table One).

In Model Ten, we dropped every peace agreement categorized as a “full” agreement in the UCDP Peace Agreement Dataset (Högbladh 2012). Full agreements, as a subsample, are those agreements where one or more dyad agrees to settle the whole incompatibility. After we dropped full agreements from the analysis (which reduced the sample size by 30 percent), we still found significant support for the impact of the quantity and range of provisions in each agreement. This strongly suggests that the strength of our findings is not a sampling artifact arising from the fact that full agreements have more substantive content. Stated differently, our results are not being driven by a low‐risk subset of cases that contain a higher than average number of provisions.

We also had to decide whether to include in our analysis thirty‐eight process agreements. DeRouen and his colleagues (2009), who used an earlier version of the same data, included these agreements (2009: 371); however, two earlier studies (Jarstad and Nilsson 2008; Nilsson 2008) excluded them. Looking at these thirty‐eight process agreements in more detail reveals some interesting characteristics about this subset. As might be expected, process agreements, which chart a process for moving forward, include few provisions. Process agreements included a mean of only 3.7 provisions (standard deviation = 2.87), and they were more likely to break down into renewed conflict than other agreements. Out of the thirty‐eight process agreements, twenty‐three of them (61 percent) failed within five years compared with 40 percent for the entire sample. Because process agreements contain disproportionately fewer provisions and are more conflict prone, and are designed to be followed by further negotiations, including them in the sample clearly boosts the analysis in the direction that supports our theory.

We thus dropped all process agreements from the sample and replicated the analysis, which yielded a sample of 157 accords. As shown in Models Eleven and Twelve in Table Four (which replicate Models Four and Seven in Table Three), the correlations between the quantity of provisions and our two dependent variables remain statistically significant, which supports the argument that our results were not being driven by a high‐risk subset of agreements that contained a lower‐than‐average number of provisions. Overall, all models support our contention that evaluating peace accords by the quantity and range of different types of provisions they contain is a practical method of predicting which accords are most and least likely to be followed by renewed conflict or renewed negotiation.

To compare the effects of the variables in a consistent manner, we calculated the marginal effects for the individual variables in Models Four and Seven at their minimum and maximum values while holding all other variables at their means. Table Five reports these marginal effects as percentage changes. As the number of UCDP provisions in a peace accord increases from zero to twenty‐one (Model Four), the probability of the same parties negotiating another accord declined by 60 percent. Including at least three provisions in an accord reduced the chances that a future accord would be negotiated by roughly 5 percent, while increasing that number to six reduced it by 13 percent. When the number of PAM provisions in an agreement increased from seven to forty‐three (Model Seven, Table Three), the chances that the parties would negotiate another accord declined by 75 percent.

We see a similar pattern for the recurrence of armed conflict. When the number of UCDP provisions in a peace accord increased from zero to twenty‐one (Model Four in Table Three), the probability that violence would resume declined by 98 percent. As the number of PAM provisions increased from zero to forty‐three (Model Seven), the probability of armed conflict recurrence declined by 99 percent.

In this article, we sought to contribute to the literature on civil war negotiations and peace agreements by focusing on concession‐making behavior and how the terms of peace agreements reflect the potential of those agreements to resolve the conflict. Our findings suggest that the peacemaking potential of a negotiated agreement between civil war adversaries is greatly enhanced when reforms are pursued across many different policy areas.

These findings have important implications for negotiation and mediation strategies in the civil war context and for the design of peace agreements. They suggest that conflict actors should view accord content holistically and not just focus on key provisions and that negotiators should propose reforms that will address the underlying incompatibility across as many different policy areas and institutions as possible. Although enforcement devices are important in the implementation setting because combatants are put at risk by their participation in the implementation process, it seems that such enforcement mechanisms as power sharing and peacekeeping forces alone are insufficient to ensure success.

Our findings about the relationship between current agreements and past and future agreements suggests that the myriad of social, political, economic, and cultural conflicts that give rise to civil wars demands a commensurate level of reforms in the policy areas that perpetuate the conflict. The results suggest that there are few shortcuts to negotiating lasting settlements in civil wars. Why are agreements containing a greater number of provisions across multiple areas likely to be more successful? We argue that the efforts of spoilers to undermine such agreements will be less successful because these agreements will better satisfy a greater number of popular concerns and reflect a greater level of commitment on the part of the government, and subsequent agreements and renewed violence will not be seen as necessary if the terms of the current agreement allow stakeholders to promote it as a historical milestone for the group.