4. Quantitative Tests

THIS BOOK ARGUES that civil wars are more likely to end peacefully if a third party enforces or verifies demobilization and if a peace treaty offers each combatant specific political, military, and/or territorial guarantees in the first postwar government. This chapter tests this theory against every civil war that began between 1940 and 1992. I am particularly interested in answering three questions. First, is it true that third-party security guarantees and power-sharing pacts substantially increase the likelihood that a war will end in a successfully implemented settlement? Second, do combatants factor the presence of third-party security guarantees and power-sharing pacts into their decision to initiate negotiations, sign bargains, and implement any resulting terms? And third, are the conditions that encourage combatants to initiate negotiations and sign agreements the same as the conditions that then bring peace?

The chapter proceeds in five parts. The first two are, respectively, a brief discussion of the methods used to analyze the data and a summary of the general findings. Each of the next three parts focuses on one of the three stages in the process of resolving civil wars. The first of these parts finds that the costs of war and a military balance of power are particularly important in getting combatants to the negotiating table. The next reveals that third-party security guarantees, certain types of power-sharing pacts, a military stalemate, territorial goals, and mediation all have a strong impact on combatants’ decision to reach a bargain once negotiations have commenced. The final part shows that in the end only third-party security guarantees and power-sharing pacts encourage combatants to implement the terms of a bargain once it has been signed.

Analytic Methods

The primary method used to analyze the data is an ordered logit model with “Peace Process” as the dependent variable. (A nested logit model would also have been appropriate to use had a greater number of cases existed.) The peace process was divided into four categories to reflect the consecutive hurdles combatants must clear in order to successfully cooperate: (1) no negotiation, (2) negotiation, (3) signed bargain, and (4) successful implementation (see chapter 3). This method of analysis was chosen for three reasons. First, by looking at the resolution of civil wars as a multistage process, we can assess which of the factors plays a role in advancing the combatants toward peace and not just which factors play a dominant role in the implementation stage. This ensures that variables critical in the early stages of a peace process are not ignored, and that variables coming into play in the final stages are not overemphasized. It also allows us to determine at what stage in the peace process expectations about compliance become a concern. Second, ordered logit analysis allows us to take into account the conditional nature of the data (i.e., you cannot influence the bargaining process unless groups first decide to negotiate). Third, ordered logit analysis is most appropriate when the dependent variable is an ordinally ranked, categorical scale in which the outcomes are fairly evenly distributed, as is the case with this three-stage peace process.¹

A secondary method was used to analyze the conditions under which combatants will initiate negotiations. Here I used a logit analysis with “Negotiation” as the dichotomous dependent variable. I chose to run this second regression for two reasons. First, some of the hypotheses presented in chapters 1 and 2 include causal factors that materialize only after negotiations have begun and are almost certainly highly correlated with the onset of talks (i.e., mediation, third-party security guarantees, and power-sharing pacts). Including these factors in a model intended to analyze the first stage of the peace process might make it appear as if mediation, third-party guarantees, and power-sharing pacts caused combatants to initiate negotiations, when in fact the direction of causation could be the reverse. Including these factors is also likely to diminish the significance of other potentially important causal variables at this first stage and produce spurious results.² Second, the decision to initiate negotiations does not depend on any prior decisions, as is the case with the decision to sign and implement a settlement once negotiations have begun. It is reasonable, therefore, to analyze this first stage separately from the full resolution process.³

Summary of General Findings

The quantitative analysis confirms that a third-party security guarantee and power-sharing pacts significantly increase the likelihood that a civil war will end in a successfully implemented settlement, but the analysis also reveals that other factors play an important role prior to the implementation of a peace treaty. Table 4.1 summarizes the results of the statistical analyses discussed in greater detail below, showing which factors are influential at each stage.⁴

As table 4.1 shows, the costs of a war (measured by its duration and by war-related deaths) and a military stalemate affect the decision to initiate negotiations. But once negotiations begin, factors associated with post-treaty security play a much larger role in the decision to sign and then implement a peace treaty. Mediation, a military stalemate, and nonterritorial goals help combatants obtain a signed bargain, but by far the two biggest factors in getting combatants to sign and implement peace settlements are third-party security guarantees and power-sharing pacts.⁵ In short, beliefs about implementation do color decisions made in the second and third phases of the peace process.

In what follows, I investigate the substantive effect of the independent variables at stage 1, stage 2, and stage 3 of the peace process. For the negotiation phase, I present data from a simple logit model that was included to ensure that the results were not unfairly weighted in favor of variables that appear only after negotiations have commenced. For stage 2 and stage 3 (the bargaining and implementation phases) I present probabilities derived from the ordered logit analysis displayed in table 4.3.

TABLE 4.1

Factors That Encourage Combatants to Negotiate, Sign, and Implement a Peace Plan, by Stage of Process

Stage 1: Initiating Negotiations	Stage 2: Reaching a Bargain	Stage 3: Implementing a Peace Plan
The duration of war	A third-party security guarantee	A third-party security guarantee
The number of war-related deaths	A territorial pact	A territorial pact
A military stalemate	A political pact	A political pact
	A mediator
	A military stalemate Nonterritorial goals

The Conditions under Which Combatants Negotiate

The decision to begin negotiations is the essential first step toward a peace settlement. No settlement can ever be reached unless the parties first agree to meet at the bargaining table. Table 4.2 presents the results of the logit regression with “Negotiation” as the dependent variable and shows that the factors that bring governments and rebels to the negotiating table are very different from the factors that allow them to reach and sign settlements. Only two factors significantly affect whether combatants initiate peace negotiations (p < .05): the duration of war and the number of war-related deaths. One additional variable (military stalemate) appears to have a positive effect on the onset of negotiations, although it fails to achieve standard levels of statistical significance (p < .10).⁶

TABLE 4.2

Logit Analysis of Factors Affecting the Decision to Negotiate

Independent Variable	Coefficient	Standard Error
Duration of war	.02*	.01
War-related deaths	.003*	.001
Military stalemate	1.88∧	1.01
Executive constraints	.26	.18
Ethnic division	.51	.84
Total goals	−.31	.76
Territorial goals	−1.73	1.18
Difficulty of division	.36	.47
Constant	−2.67∧	1.59
Pseudo R2	.36
χ2	23.53**

Note: N = 72. Heteroscedasticity-consistent standard errors, clustered by country, were employed.

^∧p < .10 *p < .05 **p < .01

None of the other variables (executive constraints, regime type,7 ethnic division, difficulty of division, territorial or total goals) had a significant effect on the decision to negotiate. Ethnic civil wars are no less likely to experience negotiations than nonethnic wars. Combatants with very different racial and tribal characteristics are just as likely to initiate negotiations as are combatants from the same ethnic group. Civil wars in Sudan, Nigeria, Yugoslavia, and Zimbabwe were driven by ethnic issues, yet the combatants in each initiated numerous formal peace negotiations. In contrast, the combatants in Costa Rica, Cuba, Paraguay, and Bolivia faced no ethnic divisions yet consistently refused to talk.

Combatants who fight in wars where the stakes are easy to divide, as was the case in Indonesia, Pakistan, and Sudan, are also no more likely to begin peace negotiations than combatants who fight for stakes that were harder to divide, as was the case in Zimbabwe, El Salvador, and China. Similarly, wars in which the rebels pursue limited goals are also no more likely to experience negotiations than wars in which the rebels have more absolute aims. Combatants appear more concerned with the immediate cost of war when they decide to initiate negotiations, and not on issues that might make bargaining easier once negotiations commence.

Finally, more democratic governments, or governments operating under a heavier burden of executive constraints, are also no more likely to negotiate with rebels than less democratic governments. Civil wars fought in Costa Rica, Peru, and the Philippines were no more likely to experience negotiations than civil wars fought in far less democratic countries such as Cuba, Burma, or Iran.

Table 4.2 illuminates which variables presented in chapters 1 and 2 influence the decision to initiate negotiations. Logit results, however, are notoriously difficult to interpret. In order to make the results easier to understand, the King, Tomz, and Wittenberg simulation procedure was used to convert the logit coefficients to expected probabilities. The strong relationship between each of these variables and the decision to initiate negotiations is more readily apparent using this procedure, the results of which are shown in table 4.3.⁸

The Costs of War

Consistent with theories that emphasize the costs of war, both the duration of war and the number of war-related deaths have a sizable and significant effect on combatants’ decision to initiate negotiations. As wars get longer, the probability of having active negotiations increases.⁹ Holding other variables at their mean values, combatants in the longest wars are 70 percent more likely to pursue a negotiated settlement than combatants in the shortest wars. Similarly, combatants fighting the most deadly civil wars are 39 percent more likely to pursue negotiations than combatants in the least deadly wars. As war-related deaths increase, so too does the likelihood of negotiations.¹⁰

These findings offer strong support for rationalist accounts of war termination that view combatants, whether civil or interstate, as driven by rationally motivated cost-benefit calculations. Incumbent governments and insurgents appear to be highly cost sensitive and will initiate negotiations as the cost of war rise. This finding calls into question arguments that view civil wars as uniquely emotional or irrational conflicts that are insensitive to material considerations. Civil war combatants may wish to portray themselves as more dedicated to their ideals and less willing to compromise, but their decision to pursue peace is clearly influenced by the price they pay to fight.

TABLE 4.3

Predicted Probability That Combatants Will Initiate Negotiations

Variable	Probability Estimate	Probability Difference
Duration of war
Less than 1 month	.27
396 months	.97
		.70*
War deaths per year
1,000	.55
1,275,000	.92
		.37*
Military stalemate
No	.51
Yes	.84
		.33∧

Note: Probabilities are derived from logit analysis presented in table 4.2.

^∧p < .10 *p < .05

A Military Balance of Power

Military stalemates also appear to have a positive effect on combatants’ decision to initiate negotiations. Holding all else constant, combatants who fight each other to a standstill are 33 percent more likely to pursue peace negotiations than combatants who are unable to stop their enemy from making military advances. Stalemates that occurred in Sudan, El Salvador, and Mozambique, for example, did appear to have a significant positive influence on combatants’ decision to pursue peace. In each case, negotiations commenced after a military stalemate set in, indicating that the decision to prove peace was also influenced by the combatants’ belief in their own ability to defeat their opponent.

Factors That Affect the Decision to Sign a Bargain

The question remains, however, whether the conditions that encourage combatants to initiate negotiations at a specific point in the war are sufficient to convince them to sign and implement settlements. A good portion of the literature on civil wars has assumed that once the military, political, and economic conditions on the ground favor negotiations, the biggest barrier to a negotiated settlement has been surmounted. To a degree this is true. Combatants have no chance to end their war in a negotiated settlement unless they are willing to meet their enemy at the bargaining table, and during the period analyzed in this study, more than half of all civil war combatants were unwilling to proceed this far. The fact that it is so difficult to convince combatants to talk to each other, however, does not mean that the conditions that prompt this important first step will enable opponents to settle their war. In fact, history shows that roughly two-thirds of all negotiations (twenty-four out of thirty-seven cases) broke down into renewed war. This high failure rate strongly suggests that other factors come into play once combatants attempt to reach, sign, and implement mutually agreeable deals.

Once combatants decide to negotiate, they face the very different and challenging task of resolving underlying conflicts of interest, something that often requires concessions and cooperation. The factors that allow combatants to navigate this phase of the resolution process, therefore, should be quite different from the conditions that bring them to the table in the first place.

Table 4.4 presents the results of the ordered logit with “Peace Process” as the dependent variable.

Although the logit coefficients show which factors significantly increase the likelihood of peaceful settlement, they do not reveal which of the factors matter in each of the three stages. A military stalemate, for example may play a large role in convincing combatants to initiate negotiations but only a minor role in the decision to implement any subsequent peace plan. The logit coefficients also do not give us much help in interpreting how important each variable is. It is not clear, for example, what a value of −2.91 for territorial goals means, given that the dependent variable is an ordered categorical variable. Territorial goals may be statistically significant; but it may also turn out to have almost no substantive affect on the probability that combatants will sign and implement a settlement. In order to rectify these two problems, I once again convert the results from table 4.4 into probability estimates, shown in table 4.5. This allows me to calculate the degree to which each of the variables influences decisions at each stage in the process of resolving civil wars. Table 4.5 focuses specifically on the second stage of the resolution process—whether combatants reach and sign settlements.

TABLE 4.4

Ordered Logit Analysis of Factors Affecting the Outcome of Civil Wars

Independent Variable	Coefficient	Standard Error
Third-party security guarantee	4.68**	1.62
Territorial pact	3.48*	1.80
Political pact	3.66∧	2.16
Military pact	−1.69	1.86
Duration of war	.01	.01
War-related deaths	−.0005	.001
Military stalemate	1.81∧	.94
Executive constraints	.11	.16
Ethnic division	.66	.78
Total goals	−.40	.81
Territorial goals	<2.91*	1.38
Difficulty of division	.23	.38
Mediator	3.62***	1.02
Constant 1	2.62	1.29
Constant 2	6.33	1.58
Constant 3	9.50	1.81
Pseudo R2	.60
χ2	67.10***

Note: N = 72. Heteroscedasticity-consistent standard errors, clustered by country, were employed.

^∧p < .10 *p < .05 **p < .01 ***p < .001

Predicted probabilities presented in this table confirm that combatants choose to sign bargains for reasons that are quite different from those that first bring them to the negotiating table. Once combatants initiated negotiations, five completely different factors affect their decision to reach and sign a peace agreement. Only third-party security guarantees, territorial and political pacts, mediation, and nonterritorial goals have a sizable and significant effect on the decision to sign a peace agreement; the duration of war and the number of war-related deaths no longer exert any significant impact on combatants’ ability to move forward. One additional factor, a military stalemate, continues to play a substantively important role in the decision to sign a peace treaty, but is not quite statistically significant (p < .10). Each of these relationships is discussed in more detail below.

TABLE 4.5

Predicted Probability That Combatants Will Sign a Bargain

Variable	Probability Estimate	Probability Difference
Third-party security guarantee
No	.05
Yes	.55
		.50**
Territorial pact
No	.09
Yes	.49	.40*
Political pact
No	.07
Yes	.45
		.38∧
Military stalemate
No	.09
Yes	.32
		.23∧
Territorial goals
No	.23
Yes	.03
		−.20*
Mediator
No	.04
Yes	.43
		.39**

Note: Probabilities are derived from the ordered logit model presented in table 4.4. Please note that the King, Tomz, and Wittenberg simulation procedure allows tests of significance only to the 99 percent level.

^∧p < .10 *p < .05 **p < .01

Third-Party Security Guarantees

As table 4.5 shows, third-party security guarantees have the greatest impact on the willingness of combatants to sign peace settlements. Holding all other causal variables constant at their mean values, governments and rebels are 50 percent more likely to accept a compromise bargain if a third party offers to enforce or verify the transitional terms and then follows through with its promise. If a third party offers to verify or enforce demobilization, as the British did in Zimbabwe in 1979 and the United Nations did in Bosnia in 1994, combatants are significantly more likely to sign a peace treaty than if no such offers are made. Indeed, if a third party did not step forward to enforce or verify, combatants have only a 5 percent probability of signing a settlement.

Political and Territorial Pacts

Combatants are also significantly more likely to sign a peace treaty that includes a territorial or political pact than if their agreement does not. Holding other variables constant at their mean values, if a peace treaty includes a provision for territorial autonomy, governments and rebels are 40 percent more likely to sign than if a peace treaty includes no such terms.¹¹ Rival factions do appear to demand some administrative control over previously occupied regions in order to preserve a political base, should things go badly. This supports the notion that combatants in civil wars wish to prevent the full concentration of power in a single administration, at least in the first postwar government.

Similarly, the combatants are 38 percent more likely to sign a bargain if it includes guaranteed positions in the new government, although this relationship is less statistically significant (p < .10). Rival factions appear concerned with the postwar distribution of power and do seem to demand guaranteed representation as the price for peace. In contrast, the presence of a military pact does not appear to have any significant effect on combatants’ willingness to sign peace settlements, a finding that suggests that rebels are more concerned with wresting political control from the hands of an entrenched and privileged elite than with obtaining military representation. This is understandable, since rebels are more likely to have the means to recruit new insurgents to their cause than to obtain legitimate administrative control over territory and political institutions.

Skeptics may argue that combatants should be more likely to sign treaties that include generous guarantees of territorial and political power. As I will show in chapter 5, however, the decision to extend political or territorial guarantees does not automatically lead to a signed settlement, and the decision to withhold these guarantees does not always prevent this result. In fact, some 30 percent of the treaties were signed without a territorial or a political pact. Moreover, only 53 percent of the treaties with a territorial or political pact eventually ended with a successfully negotiated settlement, a detail that enhances confidence in the finding. In short, it appears that specific political or territorial guarantees do increase combatants’ confidence in an agreement and play an important role in shifting their preferences from war to peace.

These results offer strong support for the credible commitment theory of civil war resolution. Incumbent governments and rebels clearly look down the road and consider problems of enforcement and post-treaty security when deciding whether to sign a peace treaty or return to war, and they clearly hesitate to sign unless a third party exists to solve this problem for them.

Military Stalemate

A military balance of power continues to have a positive effect on resolution of civil wars despite not being quite statistically significant (p < .10). Holding all else constant, combatants who fight to a standstill are 39 percent more likely to sign a peace treaty than combatants who are unable to stop their enemy from making military advances. There are at least two reasons why civil war combatants may initiate negotiations and sign a bargain in the presence of fairly balanced military capabilities. First, military stalemates impart important information to combatants regarding their relative war-fighting abilities and help eliminate disagreements and informational asymmetries that might otherwise prevent agreement. It is easier to reach a bargain when each party knows the extent of its rival’s military strength. Second, stalemates tend to bestow relatively equal bargaining power on each of the adversaries and thus promise a more equal distribution of power in any new government. If it is true, as the credible commitment theory claims, that combatants only implement those settlements that extend substantial power-sharing pacts, then only those combatants who are fairly equally matched on the battlefield will have the necessary leverage to obtain these terms.¹²

Territorial Goals

The most striking finding shown in table 4.5 is the effect of territorial goals on the likelihood of signed bargains. Although combatants involved in territorial wars are no more likely to initiate negotiations than combatants in nonterritorial confrontations, they are 20 percent less likely to reach a mutually acceptable settlement once they do begin to talk.¹³ Combatants who fought for territorial control, as the Sudanese did in the 1960s and 1970s and the Croatians did in the early 1990s, were significantly less inclined to reach and sign settlements than combatants who competed for nonterritorial goals such as control of government.

This result is quite startling. If anything, one would expect to find a positive relationship between territorial wars and combatants’ willingness to sign a bargain, since it was hypothesized that the stakes of war should be easier to divide in territorial conflicts. In addition, the credible commitment theory implies that combatants fighting for territorial separation have an easier time cooperating since they need not worry about exploitation after a treaty. It is possible that territorial disputes represent situations where combatants have already failed to work together and were pursuing separation as a result. In this case, it is the combatants’ failure to negotiate that causes rebels to shift their goals in favor of territorial separation, not the other way around. This might explain why territorial conflicts have such a low rate of negotiation.

Mediation

The final factor that has a sizable and significant effect on the willingness of civil war combatants to reach and sign bargains is outside mediation.¹⁴ As table 4.5 shows, governments and rebels are 39 percent more likely to bargain successfully with the help of a mediator than on their own (p < .01). Richard Holbrook, George C. Marshall, and Lord Owen and Cyrus Vance probably deserve credit for peace agreements reached under their guidance in Bosnia, China, and Croatia respectively. Moreover, the fact that this relationship is so strong despite the rough measure of mediation (the presence or absence of an outside mediator, rather than a more refined measure of the mediator’s ability) offers particularly clear support for mediation theories of resolution. An intermediary’s ability to press one or both parties to accept a proposal, to manipulate the agenda, control information, and sequence offers in ways that push combatants toward agreement does appear to “grease the wheels” of the bargaining process and make settlement more likely. This finding lends support to bargaining theorists who argue that informational problems often stand in the way of negotiated settlements. Mediators do appear to help solve this problem, perhaps by teasing out private information, or simply opening up better channels of communication.

Factors Affecting the Decision to Implement an Agreement

Reaching and signing a peace treaty, however, does not guarantee that negotiations will succeed. As I discussed in chapter 1, in almost half of all civil wars where combatants signed a bargain (ten of twenty-three cases), the factions reverted to violence and the civil war continued. Comprehensive peace agreements were signed in Laos, China, the Philippines, Angola, Afghanistan, Chad, Uganda, Somalia, Liberia, and Rwanda, yet all failed to bring peace. In this section, I show why the conditions that allow combatants to reach a bargain will not convince them to end a war, and I discuss what can be done to improve the chances that a bargain, once struck, will be implemented.

Once again, table 4.6 displays changes in expected probabilities for the implementation phase of the resolution process as each of the significant independent variables varies from its lowest to its highest level. Once combatants agree on the final terms of a settlement, only third-party security guarantees and power-sharing pacts are decisive in convincing them to follow through with cooperation. Three other factors are significantly related to implementation, but all have only a negligible effect on the final outcome.

Once treaties have been signed, combatants are 20 percent more likely to implement the terms if a third party offers a security guarantee, 18 percent more likely to do so if the agreement includes a territorial pact, and 16 percent more likely to implement an agreement if it includes a political pact. By contrast, a military stalemate, territorial goals, and a mediator increase the likelihood of implementation by less than five percentage points each.

Third-Party Security Guarantees

Once combatants have signed a peace agreement, all of the other factors purported to influence the outcome of civil wars fade away except third-party security guarantees and power-sharing pacts. Combatants clearly are concerned with their short and long-term survival if they do implement a peace agreement and are unlikely to proceed without strong guarantees about their future security.

TABLE 4.6

Predicted Probability That a Settlement Will Be Successfully Implemented

Variable	Probability Estimate	Probability Difference
Third-party security guarantee
No	.004
Yes	.20
		.20**
Territorial pact
No	.005
Yes	.19
		.18*
Political pact
No	.004
Yes	.16
		.16∧
Military stalemate
No	.005
Yes	.033
		.03∧
Territorial goals
No	.02
Yes	.002
		−.02*
Mediator
No	.002
Yes	.05
		.05**

Note: Probabilities are derived from the ordered logit model presented in table 4.4.

^∧p < .10 *p < .05 **p < .01

The unmistakable and powerful relationship between security guarantees and the outcome of civil wars is even more apparent when the eventual outcome is compared to the presence or absence of third-party guarantees. As table 4.7 shows, signed bargains are almost always implemented if a third party guarantees the safety of combatants as they demobilize. They are almost never implemented if a third party does not.

TABLE 4.7

Implementation of Settlements with and without Third-Party Security Guarantees

	No Implemented Settlement	Implemented Settlement
No third-party security guarantee	9	2
Third-party security guarantee	1	11

Between 1940 and 1992, combatants in twenty-three civil wars signed formal peace settlements. Third-party security guarantees were offered in twelve of these cases, and all but one of these settlements were successfully implemented. Outside powers guaranteed Lebanon’s 1958 agreement, the Dominican Republic’s Act of Dominican Resolution, Sudan’s Addis Ababa agreement, Zimbabwe’s Lancaster House agreement, the Tela agreement in Nicaragua, the Chapultepec accords that ended El Salvador’s civil war, Cambodia’s 1991 Paris peace agreement, Guatemala’s 1996 agreement, Mozambique’s Rome accord, Croatia’s Geneva peace agreement, and Bosnia’s Dayton accord, and all brought peace. The only signed settlement that was not implemented even though security guarantees were offered was Angola, and in this case Jonas Savimbi appeared to have no interest in peace. In contrast, nine of eleven signed settlements failed to end their respective wars without outside guarantees.

Two civil wars did end in negotiated settlements without the assistance of an outside security guarantee (Colombia in 1958 and Yemen in 1970). Yet a closer look at these two cases confirms the predictions of the credible commitment theory. Colombia and Yemen were the only two wars where the opposing parties had no partisan armies to demobilize once a settlement was signed. In Colombia, the war was fought by small bands of armed peasants rather than the national army, which remained relatively uninvolved in the fighting. In Yemen, the royalist rebels had no regular army and procured fighters through negotiations with powerful regional sheikhs.¹⁵ Thus, once the Conservatives and Liberals in Colombia and the Royalists and Republicans in Yemen agreed to cooperate, they simply bought the loyalty of either the powerful Colombian generals or the Yemeni tribes and in this way obtained a relatively neutral and ready-made national force. In short, these adversaries could agree to cooperate because they could bypass the vulnerable demobilization and reintegration period all other combatants faced.¹⁶

Territorial and Political Pacts

Power-sharing pacts played almost an equally critical role in the final outcome of civil war negotiations. Fifty-five percent of peace treaties that included a political, military, or territorial guarantee were successfully implemented.¹⁷ Power-sharing pacts, however, were not by themselves sufficient to ensure the successful implementation of a treaty. Combatants in nine cases (45 percent of the total) were willing to offer each other detailed power-sharing guarantees but were then unwilling to implement the terms when third-party security guarantees failed to materialize. Laos’ agreement in Geneva in 1962 promised to create a new government consisting of seven neutralist representatives and four representatives from each of the right wing parties, and equally integrate the armies of the three political factions into a single national army. Uganda’s peace accord of 1985 distributed important council seats to members of all the warring factions and called for the formation of a new national army. And Rwanda’s 1994 accord included some of the most detailed consociational arrangements of any settlement in the study. Combatants in all of these cases, however, refused to implement any terms. In short, even the most detailed power-sharing arrangements were not enough to ensure a successfully negotiated settlement in the absence of third-party guarantees.

None of the other variables presented in chapter 3 appear to have a significant effect on the willingness of combatants to implement a peace agreement.

The Costs of War

In the end, neither the duration of war nor war-related deaths were sufficient to convince combatants to sign and implement a peace settlement. A high number of deaths and protracted violence might have convinced combatants to initiate negotiations, but once these negotiations were under way, they had no significant impact on the eventual outcome of a war.

The fact that combatants become insensitive to the costs of war once negotiations begin indicates that other more important factors come into play as combatants look down the road and decide whether or not they will benefit from a peaceful solution to war. In keeping with the logic of the credible commitment theory, once combatants begin to talk about terms of settlement, the everyday costs of war should fade in comparison to the potentially greater costs of post-treaty abuse. At this point, civil war combatants should base their decision to sign and implement a peace settlement on the potential costs they expect to pay for being duped, and not on the material costs they paid prior to negotiations.

Territorial Goals

As mentioned earlier, territorial goals had an unusual effect on the peace process. Whereas combatants who pursue territorial secession or greater regional autonomy were no more likely to initiate negotiations, territorial goals had a strong negative effect on their willingness to sign a treaty, and a negative, although weaker, effect on their decision to implement the terms.¹⁸ As I explained earlier, the fact that civil wars fought over territory are less likely to reach bargains than wars fought for political control could be because these wars represent conflicts that are so intractable that exit becomes the only option.

Military Stalemate

Although a military stalemate was statistically related to the implementation of a peace agreement, its impact on this final stage was also largely irrelevant.¹⁹ Consistent with expectations arising from the credible commitment theory, a balance of power should have little effect on combatants’ willingness to implement a peace settlement since whatever defensive advantage existed disappears as soon as combatants are forced to demobilize. Combatants who are nervous about their future would obtain no added sense of security knowing they have symmetrical capabilities if these capabilities disappear as implementation proceeds. The fact that government armies and rebel armies frequently did not disband equally or simultaneously made a balance of power even less important at this stage. Second, a balance of power can quickly turn into a decisive rout if one side demobilizes while the other reneges on the agreement and launches a surprise attack.

This finding offers partial support for balance-of-power theories but clearly indicates that such a balance is not sufficient to convince groups to proceed with a risky implementation period. A balance of power may make war sufficiently costly to bring combatants to the table, and it may impart the necessary information to allow them to reach a bargain, but if a peace settlement requires combatants to reduce the number of soldiers under arms and break down partisan armies, even a robust military balance will not convince them to implement the terms.

Constraints on the Executive

Once again, the degree of executive constraints continued to have no effect on leaders’ decisions to sign and implement negotiated settlements. Leaders of democratic regimes appeared to be under no greater pressure to initiate negotiations, sign bargains, or implement peace settlements than their nondemocratic counterparts. Institutional constraints, therefore, were unable to facilitate cooperation in the risky post–civil war environment.

One can imagine a number of reasons why democratic leaders would be no more willing to implement a peace treaty than leaders who operated under fewer institutional restraints.²⁰ First, if it is true that a negotiated settlement could leave a group worse off than continued war, then citizens are unlikely to press for settlement in the absence of effective guarantees. Voters may wish to see their leader attempt to negotiate, but if a pact comes at the price of a high-risk transition, it is unlikely that they will push their leader so far. Second, citizens and subgroups will have greater difficulty influencing the outcome of negotiations as long as closed-door negotiations allow their leaders to control the amount of information they share with their public. Democratic constraints are effective only if citizens can trace the breakdown of a treaty to their leader.

The fact that executive constraints seem to play no role in the peaceful resolution of civil wars challenges an assumption found in the literature. International relations scholars have debated whether democratic institutions can facilitate cooperation. What these findings seem to show is that democratic constraints do not make leaders more likely to cooperate in situations where the costs of cheating are high. When leaders face a particularly dangerous implementation period, it is third-party verification and enforcement that matters, not regime type.

Mediation

The evidence presented in this chapter also shows that mediation has very little impact on combatants’ decision to implement peace plans even though mediation was consequential in pulling together the plan.²¹ In the end, wars in which a mediator was present during negotiations were more apt to end in a negotiated settlement because they reached a mutually acceptable settlement more frequently, but mediation did not appear to be the final variable upon which successful implementation hung. This seems to indicate that in many civil wars the failure of negotiations is not a failure of bargaining but rather the inability to solve problems of post-treaty enforcement.

Identity, Goals, and Divisibility

The remaining theories based on ethnic differences, total goals, and divisibility continued to have no predictive power in determining how civil wars end. Once again, civil wars that broke down along ethnic lines were no less likely to end in a negotiated settlement than nonethnic wars. This finding challenges popular notions that ethnic wars are more resistant to compromise solution than nonethnic wars.²² It also lends further support to the rational actor model that tends to view all combatants as driven by the same cost calculations regardless of ethnic affiliations or identity.

The actual distribution of population and resources also had no measurable impact on combatants’ decision to pursue a negotiated settlement at any stage of the peace process. Wars where it would be relatively easy to divide the population and resources were no more or less likely to end in negotiated settlements. Finally, theories that claim that total wars should be more difficult to resolve through negotiated settlements than those fought over limited goals are also not supported in these relationships. The data clearly show that wars fought for total aims such as full control over the government or social revolution were as likely to end in negotiated settlement as those fought for less total aims such as a share of governmental control or political reform.

This finding suggests that it may be wrong to draw inferences about the likelihood of compromise based on the stated goals of the belligerents. Once violence erupts, civil wars appear to face the same difficult resolution problems, regardless of the initial aims, ideology, or demands of the participants. Once fighting begins, combatants appear to get frozen into adversarial positions, and the structure of the situation appears to determine to a large extent how long they will fight, and whether they will accept and implement the terms of a settlement.

Conclusion

The empirical results presented in this chapter are striking. Various circumstances in a civil war may encourage enemies to negotiate, and certain additional factors push groups to sign a compromise settlement, but unless groups receive a guarantee from an outside country or international organization that ensures their safety during demobilization, and unless they are willing to include specific political or territorial pacts in their peace treaties, they are apt not to implement the settlement. Once negotiations begin, the success of these negotiations really is driven by combatants’ beliefs that a third party will step in to verify and enforce demobilization and political and territorial power will be shared.

This finding provides strong support for the credible commitment theory and attests to the unique role post-treaty security plays in the resolution of civil wars. The seventy-two civil wars that began between 1940 and 1992 differed from each other in many ways, yet one fact appeared interchangeable. Combatants who wished to end their war though peaceful means encountered similar implementation problems. The costs of war, the goals of the combatants, the military conditions on the battlefield, and mediation did encourage groups to pursue settlement. In the end, however, it was the structure of the postsettlement transition period that determined whether negotiations would end in war or peace.

The importance of third-party security guarantees and power-sharing pacts was not, however, the only finding to come out of this chapter. The findings presented confirm how important it is to view the termination of war as a three-stage process. Previously, academics and other observers of civil wars have failed to draw clear distinctions between different stages of the resolution process. Theorists have either emphasized the importance of the first stage (whether combatants choose to negotiate) or the second stage (whether combatants choose to reach and sign a compromise solution). My research not only demonstrates that a third stage in the process is critical to the ultimate success or failure of negotiations to end civil wars, but also reveals that very different conditions encourage cooperation at each step along the way.

Finally, one issue is not dealt with in this chapter that merits serious thought. Although the findings presented here offer strong support for the credible commitment theory of civil war resolution, they are far from a complete test of the theory. It is still possible that these findings are spurious, since outside security guarantees and power-sharing pacts may only be offered in cases that would have succeeded on their own and not in cases that were bound to fail. In the next chapter I look more critically at the interrelationships between third-party security guarantees, power-sharing pacts, and the outcome of civil wars to see if the findings presented here are truly robust.

¹ One problem with using ordered logit is the potential for selection bias that is inherent in many ordinally ranked, categorical scales. In the case of civil wars, combatants may sign a settlement but fail to implement the terms. We cannot determine from this outcome whether one of the parties would have been willing to agree to a negotiated settlement, as we observe only the final outcome. One statistical method for dealing with this issue is selection bias probit, also known as multivariate probit, which explicitly takes into account an estimate of the effects of selection bias. However, a selection bias probit model could not be used because of the relatively small number of civil wars in the data set. I have therefore attempted to address selection bias through secondary logit analyses, reported below.

² In fact, when I included these three variables in an additional logit analysis with negotiation as the dependent variable (not shown), the results were overdetermined and, as a result, produced unreliable coefficients and standard errors. I interpret this as confirmation of this suspicion.

³ Unfortunately, this makes it impossible to test whether third-party guarantees and power-sharing pacts had an impact in the first stage of the peace process. I cannot, therefore, rule out the possibility that groups are basing their decision to negotiate at least in part on whether they believe mediators and third-party guarantees will be available during the bargaining and implementation phases. The bias introduced to the logit model by these additional variables makes it impossible to specifically test this possibility.

⁴ The results reported in table 4.1 are resilient to a wide range of variations in model specification (not shown). Multicolinearity does appear to weaken the coefficients for two of the variables, total goals and military stalemate. When nonterritorial goals is removed from the equation, having total goals is positively and significantly related to the successful resolution of a civil war. Given the unintuitive nature of this finding, it is likely that the relationship that emerges is spurious. In all other cases, excluding potentially colinear variables does not produce significant changes in the remaining causal variables.

⁵ This supports findings from a comparative study of five cases by Fen Hampson. In Nurturing Peace he found that peace settlements were more durable in cases “where there was unified and sustained third-party involvement in both the negotiation and implementation of the agreement” (207).

⁶ The failure of military stalemate to achieve the .05 level of significance is most likely attributable to multicolinearity between several of the independent variables. The existence of a military stalemate is moderately correlated with a war’s duration and the number of war-related deaths in the war at .23 and .30, respectively. While these correlations are by no means overwhelming, when either war duration or the number of war-related deaths is excluded from the model, the military stalemate dummy becomes statistically significant at the .05 level. Moreover, the magnitude of the coefficient on the military stalemate dummy is only modestly affected by including or excluding these other two variables, suggesting that this latter causal factor does indeed meaningfully affect the outcomes of civil wars in the manner predicted by the theory.

⁷ Regime type was excluded from the final model as it was highly colinear with the variable executive constraints. Additional tests that included regime type but excluded executive constraints suggest that this variable also played no significant role at any stage of the peace process.

⁸ This technique involves conducting repeated simulations of a given model to estimate expected values for each β coefficient, as well as expected probabilities derived from transforming these coefficients. The “difference” in the expected probabilities for each variable is derived by setting all of the other variables to their mean value and then separately calculating the change in the probability that combatants would reach each level of the settlement process when a given causal variable moves from its lowest to its highest value. If the 99 or 95 percent confidence intervals surrounding these predicted differences exclude the possibility of zero effect (i.e., they do not run from negative to positive), we may conclude that variations in a given causal variable do produce statistically significant differences in the probability of reaching a given stage of the settlement process.

⁹ Mason and Fett had similar findings using a somewhat different list of cases. Although their analysis did not distinguish at what phase in the peace process the duration of war was likely to have the greatest effect, they did find that the longer a war lasted, the more likely it was to end in a negotiated settlement. As we will see, however, the duration of war only had a significant effect in the first phase of the peace process, a fact that Mason and Fett’s analysis could not tease out.

¹⁰ This stands in contrast to Mason and Fett’s findings. They found some support for the possibility that higher casualty rates were less likely to lead to negotiated settlements.

¹¹ Three separate studies had similar findings using somewhat different sets of cases. See Charles King, “Devolution of Power and Negotiated Settlements in Civil Wars,” Georgetown University, April 1997; Kaufman, “Possible and Impossible Solutions”; and Gurr, Minorities at Risk.

¹² For an excellent discussion on the effects of the distribution of power on settlement see R. Harrison Wagner, “Peace, War, and the Balance of Power,” American Political Science Review 88 (1994): 593–607.

¹³ This finding is compatible with one made by Ted Gurr. In his Minorities at Risk study, Gurr found that wars of secession were longer and more violent than those that did not aim for independence. This finding is supported by data on interstate conflicts. Evan Luard and Kalevi Holsti have found that competing governments are less likely to resolve disagreements over territory than almost any other issue, and territorial issues are one of the most frequent sources of war. See Evan Luard, War in International Society (London: I. B. Tauris, 1986); and Kalevi J. Holsti, Peace and War: Armed Conflicts and International Order, 1648–1989 (New York: Cambridge University Press, 1991).

¹⁴ Two other scholarly works had similar findings for interstate wars. Jacob Bercovitch and William Dixon both found that third-party mediation was consistently effective in preventing escalation and promoting peaceful settlement in conflicts between states.

¹⁵ An excellent account of this war can be found in Robert W. Stookey, Yemen: The Politics of the Yemen Arab Republic (Boulder: Westview Press, 1978), especially 243–45 and 258.

¹⁶ This finding indicates that the credible commitment theory of civil war resolution may be applicable only in those cases where the groups control separate partisan armies, and does not apply to those situations where a negotiated peace treaty creates few security risks.

¹⁷ Similar results emerged from a related study of intrastate wars. An analysis of all civil wars between 1945 and 1997 found that settlements that included military, political, or economic institutional guarantees in the terms of a settlement were most likely to be stable. See Caroline A. Hartzell, “Explaining the Stability of Negotiated Settlements to Intrastate Wars,” Journal of Conflict Resolution 43, no. 1 (1999): 3–22.

¹⁸ This finding is supported by a study of all civil wars and separatist revolts since 1945 by Mason and Fett. They found that separatist wars were no more likely to end in negotiated settlements than revolutions. This stands in contrast to a comparative study of eighty-one international and civil conflicts in Africa, Europe, and the Middle East between 1945 and 1985 by Hugh Miall. When one looks only at conflicts in these three regions and includes interstate conflicts as well, territorial goals make peaceful settlement more likely. See Miall, The Peacemakers.

¹⁹ As previously noted, military stalemate is moderately correlated with both the duration of war and the number of war-related deaths. However, taking military stalemate out of the regression analysis does not increase the impact of either war duration or the number of war-related deaths at any stage.

²⁰ A similar finding was made in regard to economic reform. See Joel S. Hellman, “Winners Take All: The Politics of Partial Reform in Postcommunist Transitions,” World Politics 50, no. 2 (1998): 203–34.

²¹ One other scholarly work had a similar finding. In their study on the contribution of UN peace operations to peacebuilding, Michael Doyle and Nicholas Sambanis found that UN mediation had no significant effect on successful peacebuilding. They did not, however, break down the analysis into three phases and therefore could not say whether mediation had an effect earlier in the peace process. See Michael W. Doyle and Nicholas Sambanis, “International Peacebuilding: A Theoretical and Quantitative Analysis,” American Political Science Review 94 (2000): 778–801.

²² Quantitative studies by Roy Licklider, “The Consequences of Negotiated Settlement in Civil Wars, 1945–1993,” American Political Science Review 89 (1995): 681–90, and by Mason and Fett, “How Civil Wars End,” had similar findings.