Win,lose,or draw in the fog of civil war

Abstract

Civil war outcome studies have used expected utility logic to identify factors that affect actors’ estimates of the probability of victory, the payoffs from victory vs defeat, and the accumulated costs of fighting until victory is achieved. Tests have used static measures of national attributes and war characteristics, measured prior to the war or at its end. We use UCDP Georeferenced Event Data from 73 civil conflicts in Africa to estimate how changes in government and rebel tactical choices on where and when to fight battles affect expected utility estimates and, therefore, civil war outcomes.

Keywords

African conflicts battle data civil wars civil war outcomes conflict termination

Previous studies on the outcome of civil wars (i.e. whether they end in government victory, rebel victory, or negotiated settlement) have employed expected utility logic to identify a set of factors that should affect the choice by governments and rebels between continuing to prosecute the war or quitting, either by conceding defeat or seeking a negotiated settlement. This choice, iterated throughout the course of the conflict, is based on each party’s subjective estimates of (1) the probability of victory, (2) the payoffs from victory vs defeat, and (3) the accumulated costs of fighting from the beginning of the war through the present until that time in the future when the actor anticipates s/he will be able to achieve victory. Tests based on this logic, however, have employed as predictors of civil war outcomes static measures of national attributes (usually measured at the start of the war) and characteristics of the war itself (especially its duration and its deadliness, measured after the war has ended; for a review of this literature, see Shelton et al., 2013).

Pre-war estimates of the relative capabilities and resolve of the protagonists may not always be reliable predictors of the outcome of a war. Bargaining theories of war argue that, given the costliness of war, if all protagonists had complete and perfect information about each other’s capabilities and resolve (and, therefore, the likely outcome of the war) prior to the onset of war, they could find a negotiated agreement that would leave them all better off than they could expect to be even with a favorable outcome to war (Fearon, 1995; Walter, 2009). The outbreak of war, then, is depicted as a function of information failures. Actors over-estimate their own capabilities relative to those of their rival and under-estimate both their rival’s willingness to fight and their ability to sustain their war effort once conflict is underway. Civil wars are marked by information asymmetries as well: the rebels have far more information on the size and capability of the government’s forces than the government has on the rebel forces. War, then, becomes an information-revealing process whereby protagonists update their estimates of their rivals’ capability and resolve and, accordingly, their own expected utility from continuing to fight vs conceding defeat or seeking a negotiated settlement.

Even as war reveals information about each other’s capabilities and resolve, those estimates may not enable actors to estimate accurately the likely outcome of the war. In the fog of war, actors make tactical choices that can be disastrous, bringing on battlefield defeat despite the fact that they may have begun the war with a decided advantage in military capabilities. The French forces were defeated by Viet Minh fighters at Dien Bien Phu despite having superior numbers and fire power. Within months the French government signed the Geneva Accords that called for the withdrawal of French forces from all of Indochina (Karnow, 1983: 189–205). Thus, the outcome of a civil war may not be reliably predicted ex ante because it depends, ultimately, on tactical choices regarding when and where to deploy how much of one’s forces in the conduct of a sequence of battles over the course of the war.

The frequency, destructiveness, outcome, and changing geographic distribution of battles over the course of the war compel rivals to adjust their estimates of the payoffs from continuing to fight, conceding defeat, or seeking a negotiated settlement. In this paper we use UCDP Georeferenced Event Data (GED) from 61 civil conflicts dyads in Africa to identify the locations of battles, their proximity to key strategic locales (including major cities and the capital), and changes in their geographic dispersion over the course of the war to estimate how changing battlefield dynamics alter the protagonists’ estimates of the payoffs from continuing to fight vs conceding defeat or seeking a negotiated settlement. Adding this dynamic element to the modeling of civil war outcomes should enable us to estimate how changes in government and rebel tactical choices regarding where and when to engage their enemy affect their competing prospects for victory and/or their willingness to agree to peace talks at different points in the conflict. Developing a better understanding of how the battlefield dynamics of civil conflict affect the prospects for victory, defeat, or peace agreement should give us a better understanding of how those battlefield dynamics affect the “ripeness” of conflicts for resolution. Greig (2015) identifies certain patterns of spatial acceleration of battles toward or away from certain key locations in a civil war nation that are associated with a greater or lesser willingness on the part of protagonist to accept offers of third party mediation. Variation in the timing of armed intervention on one side or the other in a civil war has also been shown to affect the likelihood that such interventions will succeed in bringing the conflict to an earlier and less destructive conclusion (e.g. Regan, 2002). Thyne (2015) has shown that the incidence of military coups during civil wars affects the duration of civil wars. Coups represent a shock event in the civil war process that can shorten the duration of the war (depending on whether they occur early or late in the conflict) compared with what it would have been in the absence of a coup. Our examination of battlefield dynamics invokes a similar logic, but with the cumulative effect of battles representing a sequence of shock events that should affect the duration and outcome of the conflict. Refining our analysis of battlefield dynamics could give scholars and policy makers alike a more precise understanding of when in the course of a civil war and amid what battle dynamics offers to mediate a peace agreement are more or less likely to succeed.

In the next section we present a theoretical framework of the logic by which actors involved in a civil war choose between continuing to fight, conceding defeat, or seeking a peace agreement. This framework is based on the expected utility logic that has been used in previous studies to specify statistical tests of factors that explain whether an armed conflict ends in a military victory for one side or the other or in a negotiated peace agreement (Brandt et al., 2008; DeRouen and Sobek, 2004). We use that framework to derive a set of hypotheses concerning how changes in the geographic dispersion of battles over the course of the war affect the decision calculus of civil war protagonists and, therefore, the competing risk of the conflict ending in a rebel-favorable or government-favorable outcome. We test those hypotheses using UCDP Georeferenced Event Data from 61 civil war dyads in Africa to identify the locations of battles and changes in their geographic dispersion over the course of the war.

How civil wars end

Most studies of civil war duration and outcome are based on a theoretical framework that depicts the conflict as a contest between two actors, government (G) and rebels (R) (e.g. Brandt et al., 2008; Mason and Fett, 1996; Mason et al., 1999). In about one-third of civil wars, there is more than one rebel group actively battling the government at the same time (Cunningham, 2011: 13). Each government–rebel dyad (G-R) can be represented as a two-person contest in which each actor repeatedly updates its own estimates of the payoffs from continuing to fight, conceding defeat, or seeking a negotiated settlement to end the conflict. If both sides choose to continue fighting, the conflict endures. If R continues to fight while G quits, R wins, and the incumbent government is overthrown. If G continues to fight and R quits, G wins, and the incumbent regime survives. If both agree to stop fighting simultaneously, the cease fire provides an opportunity for them to negotiate a permanent peace agreement that ends the war. Under most circumstances, continued fighting is the dominant strategy for both sides (Stam, 1996: 353).

Rational choice models of civil war duration and outcome have depicted the choice between continuing to fight or quitting (by conceding defeat or seeking a peace agreement) as a function of each actor’s estimate of (1) the payoff from victory (U_v), (2) the probability of victory vs defeat (P_v vs 1-P_v), (3) the rate at which that actor is absorbing the costs of fighting (C_ti), and (4) the amount of time s/he estimates s/he will have to fight from the present (t₀) until that time in the future when s/he anticipates achieving victory (t_v). For an actor to prefer a negotiated settlement over continuing to fight, it must be the case that the expected utility from a negotiated settlement now (EU_s) is greater than the expected payoff from victory (EU_v). We can assume that the payoffs from a negotiated settlement (U_s) are less than the payoffs from victory (U_v). However, by ending the conflict in a peace agreement now, both actors avoid the costs of fighting that would accrue from the present until that time in the future when each actor anticipates achieving victory ( $\sum_{ti = 0}^{t_{v}} C_{ti}$ ). This decision calculus implies that any factor that (1) decreases an actor’s estimate of the probability of victory (P_v), (2) decreases that actor’s estimate of the payoffs from victory (U_v), (3) increases the rate at which the actor absorbs costs of conflict (C_ti), or (4) extends the actor’s estimate of the time required to achieve victory (t_v) should make that actor more inclined to seek a negotiated settlement rather than continue fighting in search of victory (see Brandt et al., 2008; Mason and Fett, 1996; Mason et al., 1999).

Existing works that use this expected utility logic to model how civil wars end have employed static measures of characteristics of the conflict itself, the protagonists’ capabilities, and the civil war nation as proxies for the parameters in the expected utility model. Most if not all of these measures are taken either at the beginning of the conflict or after the conflict has ended. For instance, the rate at which protagonists absorb the costs of war is usually approximated with some measure of total casualties divided by the duration of the conflict, both of which are measured at the end of the conflict, not as annual or monthly rates that vary over the course of the conflict. The probability of victory is typically measured with some estimate of the balance of forces between government and rebels, measured at the beginning of the war (Brandt et al., 2008; DeRouen and Sobek, 2004).

The availability of new data on battles—including where and when they occur—allows us to add a dynamic element to the modeling of this decision calculus. These data allow us to consider how the outcome and duration of the war might vary as a function of tactical choices made during the course of the war, choices that could alter an actor’s estimate of the probability of victory in ways that the static measures used in existing models do not capture. Greig (2015) introduces this dynamic element into analyses of the “ripeness” of conflicts for third-party mediation. He argues that the location and movements of battles provide both governments and rebels with information about their rival’s capabilities and resolve. That information allows them to update their estimates of the probability of victory, the time required to achieve victory, and the rate at which they will have to absorb costs in order to achieve victory. The payoffs from victory are relatively constant, although the war itself causes destruction of assets, infrastructure, and human capital, thereby diminishing the value of the prize of victory as a function of the duration and destructiveness of the conflict. For the purposes of this analysis, we will concentrate on the effects that the shifting tide of battle have on the decision parameters that do vary over the course of the war: the probability of victory, the costs of achieving victory, and the time required to achieve victory.

At the onset of civil war, both government and rebels lack complete information about their rival’s capabilities and resolve. Rebels have more information on the government’s military capabilities than the government has on the rebels’ ability to mount and sustain military operations. There is publicly available information on the government’s military budget, the number of troops in uniform, and where those troops are deployed. In contrast, information on how many combatants the rebels can mobilize for battle is less well known, even to rebel leaders themselves. On the other hand, in the early stages of a civil war, rebels are at their most vulnerable: they have to build an army, usually from scratch, in the shadows of a government that already has an organized, equipped, and trained fighting force of long standing. Indeed, most studies have found that where government victories occur, they usually occur early in the conflict. If rebels can avoid defeat in the first few years of the war, the probability of government victory begins to decline until at some point the probability of an outcome favorable to the rebels (either a rebel victory or a negotiated settlement) comes to exceed the probability of a government victory (Brandt et al., 2008). Therefore, rebels have a powerful incentive to signal the government that they, the rebels, are a formidable force that cannot be defeated easily or quickly.

Battles serve this function. Battles provide rebels with a means to demonstrate to the government (and to potential civilian supporters) their ability to mount and sustain military operations. Early in the conflict the rebels’ ability to choose where and when to fight can demonstrate to the government that the rebels cannot be defeated quickly. Guerrilla tactics of choosing where and when to attack may not induce the government to lower its estimate of its probability of victory (P_v), but that strategy can compel the government to increase its estimate of the amount of time it will take to defeat the insurgency (t_v) and, therefore, the cumulative costs of victory. Avoiding an early defeat can also enhance the rebels’ ability to recruit fighters and civilian supporters. Civilians who may prefer the rebels to the government but are deterred from actively supporting the rebels for fear of retribution will decrease their estimate of those risks as they witness the rebels engaging in (and even initiating) more battles with government forces. The rebels’ demonstrated capacity to sustain military operations will generate more civilian support, which will further enhance the rebels’ capacity to mount and sustain military operations. In this respect, battles become a means for rebels to persuade the government to reduce its estimate of its own probability of victory (P_v) and to increase its estimate of the time required to achieve victory (t_v) and, therefore, its estimate of the cumulative costs of victory ( $\sum_{ti = 0}^{t_{v}} C_{ti}$ ). By altering the government’s estimate of these three parameters of its choice between continuing to fight, conceding defeat, or seeking a negotiated settlement, battles can enhance the rebels’ prospects both for victory and for compelling the government to negotiate a peace agreement, both of which are outcomes that are favorable to the rebels, compared with either the status quo ante or rebel defeat.

Rebels, to a greater extent than governments, must prove themselves on the battlefield, not only to win, but also to induce the government to negotiate. In addition to its military advantage early in the conflict, the government has reputational incentives to avoid negotiating with rebels. By agreeing to negotiate, a government confers a degree of legitimacy on the rebels, elevating their status to something approaching equivalence with the sovereign state itself (Svensson, 2007; Zartman, 1995). This patina of legitimacy can enhance the rebels’ ability to recruit supporters among the population and to attract material support from external patrons. Therefore, we would expect governments to eschew invitations to negotiate early in the conflict (Sisk, 2009). Instead, governments will exploit the military advantage they enjoy early in the conflict and use early battles to impose costs on a nascent rebel organization in hopes of preempting its evolution into a sustainable insurgency. In El Salvador in 1932, an uprising by several thousand peasants led by Agustin Farabundo Marti was crushed by the government’s security forces, resulting in 10,000–20,000 deaths. There was not another incidence of organized violence against the regime until the 1970s (Stanley, 1996: 41–43).

There are two aspects of battle dynamics that can affect the decision calculus of both rebels and governments and, therefore, the duration and outcome of the conflict. One is the location and intensity of individual battles. Battles that occur in locations with larger populations, higher concentrations of valued assets, and higher levels of commercial activity impose higher costs on the government. In contrast, if battles are confined to peripheral regions of the nation (i.e. where there is less population and a lower concentration of economic assets), then the rebels cannot inflict as much cost on the government, and such battles will have less effect on the government’s willingness to negotiate or concede defeat.

The second dimension of battle dynamics concerns the dispersion of battles across the nation’s territory. When battles become more dispersed across a broader swath of the nation’s territory, government forces will be spread more thinly, diminishing their ability to defeat the rebels and end the war. This spreads rebel forces more thinly as well. However, unlike the government, the imperative to seize and hold territory is less urgent for the rebels. If battles are concentrated within a smaller portion of the nation’s territory, the government is more able to contain and even defeat the rebels.

Battle location

Rebels use battles to impose costs on the government. The amount of costs any given battle imposes on the government is in part a function of where the battle occurs. If rebels can only fight in peripheral regions where there are fewer people and fewer assets of value to the nation’s economy, the government can more easily absorb costs of those battles. Consequently, battles in peripheral regions will have less effect on the government’s willingness and ability to continue fighting.

If rebels can fight in or around major cities or other locales where valued economic assets are located (e.g. ports, regions with mines, oil wells or other mineral extraction assets), the government will suffer costs at a higher rate (C_ti) than it does when battles are fought in sparsely populated regions with few economic assets. Battles near major cities may or may not substantially alter the government’s estimate of its probability of victory. However, battles in such locations are likely to produce more casualties, more destruction of economic assets, and more disruption of economic activity, thereby increasing the government’s estimate of the rate at which it will have to absorb costs (C_ti) to achieve victory.

Battles that are fought in more densely populated areas also provide citizens with visible information about the capabilities of the rebels and the risk to civilians of being caught in the crossfire between government and rebel fighters. Estimates of their own personal risks of suffering the cost of conflict increase. As a consequence, citizens may choose to flee the city in order to avoid being caught in the crossfire. We have witnessed this pattern in Syria and Iraq where ISIS has fought for control of and, in some cases, seized control of cities (Thibos, 2014). Alternatively, if rebels can seize and hold key cities, the cities’ population is likely to shift their overt support to the rebels out of fear of retribution if they do not. This gives rebels access to resources that further enhance their battlefield capabilities. Should government forces retake the city, the local population can be expected to shift overt support back to the government for fear of retribution if they do not, thus denying rebels access to the human and material resources concentrated in the city (see Kalyvas, 2006; Wood, 2014; Wood and Kathman, 2013).

For these reasons such battles should increase the probability of rebel victory (relative to the probability of government victory). They should increase the probability of a negotiated settlement as well: the payoff from a peace agreement begins to approach or surpass the payoff from continued fighting for the government. For the rebels, the probability of defeat declines, and the prospects increase for a peace agreement that would provide the rebels with more net benefits than what they can expect from continued fighting. The cumulative effect of battles near major population centers should be to make the government more likely to seek a negotiated settlement and less likely to achieve a victory.

H1: as civil war battles occur closer to major cities, negotiated settlement becomes more likely.

H1a: as civil war battles occur closer to major cities, government victory becomes less likely.

Greig (2015) draws a distinction between battles fought near cities with economic value and battles fought near the nation’s capital city. While attacks on localities with economic value degrade the government’s ability to sustain its military operations, battles near the capital city pose a direct threat to the continued existence of the government. They present the government with an increased risk of defeat. Battles near the capital are also likely to affect the likelihood of a negotiated settlement. The government’s bargaining leverage is degraded by fighting near the capital. Rebels are less inclined to negotiate if the outcome of a battle near the capital leads them to conclude that victory is at hand. If the rebels can attack in and around the capital, they are less inclined to seek a peace agreement because they estimate their probability of victory to have increased. Why accept concessions from the government when your chances of achieving total victory have increased?

While battles near the capital represent an immediate threat of government defeat, governments do tend to concentrate their military forces in and around the capital. The government is more capable of inflicting casualties on the rebels in battles near the capital than in battles elsewhere. Consequently, when rebels fight near the capital, they are likely to experience an increase in the rate at which they suffer costs of conflict (C_ti). A government that sees its existence as threatened by such battles is likely to dig in its heels and fight in order to avoid a military defeat that would mean the political annihilation of the incumbent regime. Therefore, the government has little to lose by escalating is own war efforts to reverse the course of the war rather than seek a negotiated settlement.

If the outcome of battles near the capital is not a decisive victory for either side, such battles may increase both sides’ incentives to seek a peace agreement. Both sides suffer heavy costs. Neither side has reason to believe it has a better chance of decisive victory. Therefore, negotiated settlement becomes more attractive than continuing to fight. The FMLN’s “final offensive” in El Salvador in 1989 did not result in the victory they had hoped for, but while the Salvadoran military was able to push the FMLN out of San Salvador, the governing ARENA party concluded that a decisive defeat of the FMLN was not likely either. Soon thereafter, both sides agreed to peace negotiations (Stanley, 1996: 245–253). Thus, battles near the capital are likely to shorten the duration of the war, but their effect on the outcome of the war depends on the outcome of those battles.

The effect of battles near the capital should be different, however, for secessionist rebels. The goals of secessionist rebels do not include capturing the capital and seizing power over the entire nation. Instead, they seek autonomy or independence for their region, which rarely if ever includes the nation’s capital. To the extent that they can choose battle locations, we would expect them to fight to secure their homeland; attacking the capital should not be central to their battle strategy to the same extent that it would be for revolutionary rebels. When secessionist rebels choose to fight near the capital, their goal is simply to impose more costs on the government in order to induce the government to concede autonomy. Otherwise, government forces could continue to fight a war of attrition against rebel forces who have relatively secure control of their homeland but cannot impose enough costs on the government to compel it to concede autonomy and cease its war of attrition. Biafran rebels in Nigeria secured their homeland but were eventually defeated by the government because they could not impose costs on the government outside of their own homeland so that they eventually succumbed to a war of attrition. On the other hand, extending their forces into territory that is heavily defended by government forces and not vital to the rebel’s secessionist goals carries the risk of extending their forces and leaving themselves vulnerable to losses even in areas that are vital to their secessionist goals. By this logic, we would expect battles near the capital by secessionist rebels to have effects similar to those for battles near cities that are not the capital.

H2: As civil war battles occur near the capital in revolutionary (i.e. non-secessionist) conflicts, government victory becomes less likely.

H3: As civil war battles occur near the capital in secessionist conflicts, the probability of negotiated settlement and rebel victory should increase.

Battle dispersion

The dispersion of battles across the territory of the nation should also affect the protagonists’ choice between continuing to fight, conceding defeat, or seeking a negotiated settlement. Where rebels are able to mount battles across a wide swath of the nation’s territory, they are signaling their capabilities and resolve in ways that should compel the government to lower its estimate of its probability of victory and to increase its estimate of the amount of time it would require to defeat the rebels. This should make the government more willing to seek a negotiated settlement to end the war. This effect should become stronger the longer the war lasts. Rebels who can operate over a wide range of territory are less vulnerable to defeat as a result of a tactical error in one major battle. Their forces are less concentrated in the area of that one battle and, therefore, they are more able to sustain operations elsewhere. When rebels can contest government forces in locations dispersed over a large area, government forces become spread more thinly, reducing their prospects of prevailing in any given battle. Likewise, the government’s prospects for bringing the war to a conclusion with a decisive battlefield victory are lower when the rebels are able to operate over a larger territorial domain. This should make the government less likely to win and more willing to accept a negotiated settlement. By itself, territorial dispersion of battles might not enhance the prospects of rebel victory to the same extent that it increases the likelihood of negotiated settlement.

H4: As civil war battles become more dispersed across the territory of the nation, the probability of a negotiated settlement increases.

Research design

Our sample is drawn from the conflicts identified in the UCDP GED, version 1.1 (Sundberg and Melander, 2013). This dataset identifies the occurrence, geographic location, and intensity of African civil war battles from 1989 to 2010. Data limitations confine our analysis to the 1989–2006 period. Using the conflict start and end dates identified in the UCDP Conflict Termination dataset (Kreutz, 2010), we construct a database with the civil war dyad-month as the unit of analysis. Each civil war dyad consists of one government and one rebel group. Ongoing conflicts involving multiple rebel groups are treated as separate dyads. Our sample, after excluding cases with missing data in our control variables, covers 61 civil war dyads distributed across 24 African countries between 1989 and 2006. After excluding cases with missing data, this yields a sample of 1976 civil war-months for analysis. Table 1 presents summary statistics for the variables in our analysis.

Table 1.

Summary of descriptive statistics

Variable	Mean	SD	Minimum	Maximum
Major city distance	167.0629	151.221	0.0003233	654.2921
Capital distance	258.2324	208.8725	0.0002657	967.2356
Secessionist conflict × Capital distance	0.2095532	0.4070943	0	1
Secessionist conflict	49.08645	107.7469	0	691.4442
Battle dispersion	2.344253	4.223093	0	40.91055
Capital velocity	−0.172318	81.18448	−685.5727	549.5262
City velocity	1.391634	63.59042	−393.1334	462.4084
Battle dispersion velocity	0.0345171	4.183706	−34.83057	34.83057
Number of ongoing battles in month	3.472522	5.35681	0	70
Polity	7.017976	3.695388	2	18
Government military personnel	10.86337	1.697138	0	12.97154
Rebel military personnel	8.68603	1.268815	5.70711	11.15627
Ongoing conflicts	1.460709	0.7081032	1	3
Prior diplomatic efforts	0.9851053	3.353801	0	28
Total battle-deaths	7.091711	1.913093	0	10.61818

Because our focus centers upon the ways in which the violence that occurs during the course of civil wars shapes their outcomes, we adopt a competing risks approach for our analysis. The civil wars in our analysis can terminate in one of two ways: rebel-favorable and government-favorable outcomes. This convention for categorizing outcomes is used in Fortna (2015) and Gurses (2015). We code those civil wars identified by Kreutz (2010) with outcomes of “peace agreement”, “ceasefire with conflict regulation”, “ceasefire” and “rebel victory” as terminating with rebel-favorable outcomes. Conflicts coded as having ended with a government victory and those ending with no peace agreement and the same government still in power as before the conflict started (Kreutz’s “low activity” outcome) are coded as government-favorable outcomes. In our sample, we observe 21 conflicts that end with rebel-favorable outcomes and 32 that terminate with government-favorable outcomes. Two of the conflicts in our sample continue beyond the 2006 end point of our dataset. Because these cases are right-censored, we use survival analysis to deal appropriately with this censoring issue using Cox proportional hazard models.¹ In this respect, our competing risks analysis is conducted by estimating separate Cox models for each type of termination.

In order to identify the effect of battle locations and the geographic distribution of fighting on the outcomes of civil wars, we use data from the UCDP GED (Sundberg and Melander, 2013) to construct variables describing these effects. We create a variable that describes the weighted historical distance in miles between a civil war’s battles and the country’s nearest major city. To create this variable, we define major cities as the five most populous non-capital cities within a civil war state. In each month in which battles are ongoing during a civil war, we calculate the mean distance in miles between all battles in that civil war dyad and the nearest major city. We then create a weighted historical major city battle distance variable by calculating the weighted average of all of the monthly battle distances across the life of the civil war. Because we expect that the locations of more recent fighting will exert a stronger effect upon the outcomes of civil wars than battles further in the past, we construct our weighted measure so that the effect of older battles diminishes over time as a decay function. Over time, the cumulative costs of war increase, so that a battle with a given set of characteristics (e.g. proximity to strategic cities) should have a greater effect on each side’s estimate of its chances of victory than would a similar battle that occurred much earlier in the war. Our weighted historical major city distance variable is calculated as:

major city distance = \frac{\sum_{n = 0}^{\infty} \frac{1}{2_{n}} * d_{t - n}}{\sum_{n = 0}^{\infty} \frac{1}{2_{n}}}

where d is the mean distance of battles to the nearest major city in a month in which civil war battles are ongoing, and t is the number of previous months with battles. Following a similar process, we construct a second variable that measures the weighted historical distance in miles between a civil war’s battles and the country’s capital. We lag each of these variables by one month.

Besides the proximity of civil war fighting to a civil war state’s key cities, we also expect that the geographic dispersion of battles across the nation’s territory will shape the outcomes as well. To measure the dispersion of battles across a country, we use data from the UCDP GED to compute the mean distance in miles between all battles during the previous three months of conflict between a rebel group and the government. Because the influence of the proximity of battles to a nation’s cities and the impact of the dispersion of battles is affected by the size of a civil war state, we identify the geographic area of the civil war state using data from the CIA World Factbook (Central Intelligence Agency, 2009). We then calculate the measure of battle dispersion by dividing the mean three-month dyadic battle dispersion by the natural logarithm of the civil war state’s geographic area.

Our theoretical argument suggests that, just as the current locations and dispersion of battles influence the outcomes of civil wars, so do the changes in the locations and dispersion of fighting. To capture these effects, we incorporate three variables in our models. To measure the velocity at which battles move toward a country’s major cities, we include separate variables identifying the one-month change in the proximity of the nearest battles to a country’s major non-capital cities and the capital city. Following a similar logic, we also include a variable measuring the month-to-month change in the battle dispersion measure described above.

Although we expect that the geography of civil war battles will condition the ways in which civil wars end, we also recognize that other characteristics of conflicts will shape the way that they end. We use UCDP GED data to create a variable describing the total number of battle-deaths, lagged by one month, produced to-date by the conflict.² We also use UCDP GED data to construct a variable describing the number of ongoing battles within each civil war dyad during the month. Just as an increasing casualty count can place pressure on a government to end a conflict, so too can the pressure experienced by a government facing multiple rebel challenges at the same time. To account for this effect, we use data from the UCDP Conflict Termination dataset (Kreutz, 2010) to construct a variable describing the total number of ongoing government-rebel dyads a civil war state is facing during each civil war-month.

Beyond battlefield conditions, we recognize that other factors influence the extent to which a civil conflict is amenable to a peaceful settlement. Secessionist conflicts, for example, represent a unique challenge for efforts to manage civil wars. Accordingly, we use data from Cunningham et al.’s (2009) Non-State Actor dataset to identify if a rebel group fighting a civil war has a secessionist goal, coding conflicts involving a secessionist group as “1” and all other conflicts as “0”. Not only does a secessionist aim influence the prospects for negotiated agreement, but it also influences the impact of where civil war battles are fought. For a rebel group seeking to overthrow a government, taking control of the capital is a vital step toward the achievement of their goals, giving them critical access to the levers of power in the state. For a secessionist rebel group, control of the capital is less important. Instead, the ability of the rebel group to impose costs on the regime and control substantial portions of contested territory is more important to the achievement of rebel goals than control over the capital. Consequently, we include an interaction term in our model that interacts the secessionist variable with the weighted historical capital distance variable.

To control for the effect that the characteristics of the civil war state have upon the outcomes of civil wars, we look to three variables. We control for the military capacity of the civil war state with a variable that describes the number of military personnel of the state. We control for rebel military capacity by including a variable drawn from data in the Non-State Actor Dataset, v. 3.4 (Cunningham et al., 2009) that describes the number of military forces of the rebel group. In order to account for the skewness inherent to both the government and rebel force size data, we use the natural logarithm of each variable in our models.³ We also include a variable for each state’s polity score using data from Polity IV (Marshall and Jaggers, 2002). To eliminate negative values, we rescale this variable by adding 10 to the original Polity2 score reported by Polity IV.⁴ Finally, to control for the amount of third-party diplomatic attention directed to each conflict, we use data from the Civil Wars Mediation dataset (DeRouen and Bercovitch, 2011) to create a variable describing the number of previous mediation efforts, to date, applied to the civil war. This variable is lagged by one month.

Results

The results of our analysis underscore the important effects that the geography of civil war battles exerts upon the outcomes of civil wars. While the characteristics of civil wars and third-party efforts to manage them shape the outcomes of civil wars, so too does the geographic dispersion of civil war battles and their proximity to key cities. These results are summarized in Table 2. Model 1 includes only our theorized variables without controls. Model 2 includes only the theorized the battle location variables with control variables, while Model 3 includes all of our theorized battle variables with controls. The results are generally consistent across the models. Except where otherwise stated, we use results from Model 3 when discussing our results below.

Table 2.

Cox regression competing risks for civil war termination

	(1)	(1)	(2)	(2)	(3)	(3)
	Rebel favorable	Government favorable	Rebel favorable	Government favorable	Rebel favorable	Government favorable
Major city distance	0.996	1.002*	0.995	1.000	0.995	0.999
	(0.00305)	(0.000992)	(0.00307)	(0.00184)	(0.00304)	(0.00157)
Capital distance	1.001	1.001	0.998	1.005*	0.997†	1.006**
	(0.00129)	(0.00105)	(0.00159)	(0.00196)	(0.00155)	(0.00192)
Secessionist conflict × Capital distance	1.007†	1.002	1.009*	0.997	1.011**	0.998
Secessionist conflict × Capital distance	(0.00370)	(0.00218)	(0.00380)	(0.00269)	(0.00335)	(0.00307)
Secessionist conflict	0.0634*	1.592	0.0121**	1.808	0.00674**	1.331
	(0.0827)	(1.196)	(0.0168)	(2.152)	(0.00925)	(1.565)
Battle dispersion	1.119**	0.758*	1.071†	0.747	1.127**	0.636**
	(0.0408)	(0.0947)	(0.0394)	(0.134)	(0.0497)	(0.0935)
Capital velocity	1.004	0.993**			1.006	0.998
	(0.00547)	(0.00273)			(0.00415)	(0.00328)
City velocity	0.999	1.003			0.995	1.000
	(0.00678)	(0.00278)			(0.00403)	(0.00283)
Battle dispersion velocity	0.959	0.969			0.914**	1.257†
Battle dispersion velocity	(0.0258)	(0.0645)			(0.0242)	(0.153)
Number of ongoing battles in month			1.022	0.835*	1.006	0.852*
Number of ongoing battles in month			(0.0494)	(0.0684)	(0.0579)	(0.0658)
Polity			1.329**	1.136	1.345**	1.121
			(0.0977)	(0.0888)	(0.0890)	(0.0809)
Government military personnel			0.835†	1.114	0.847	0.600
Government military personnel			(0.0908)	(0.216)	(0.0918)	(0.213)
Rebel military personnel			1.016	0.516**	1.100	0.287**
Rebel military personnel			(0.351)	(0.0681)	(0.382)	(0.137)
Number of ongoing conflicts in state			3.400**	0.881	4.013**	0.917
Number of ongoing conflicts in state			(1.455)	(0.255)	(1.707)	(0.268)
Prior diplomatic efforts			1.157**	0.771*	1.142*	0.774†
Prior diplomatic efforts			(0.0603)	(0.0981)	(0.0702)	(0.109)
Total battle-deaths to date			0.846	0.598**	0.828	0.603**
Total battle-deaths to date			(0.115)	(0.105)	(0.111)	(0.102)
Time × Government military personnel						1.244†
						(0.140)
Time × Rebel military personnel						1.210
						(0.181)
Cases	76	76	61	61	61	61
Failures	26	39	21	32	21	32
Wald chi-squared	35.87	34.48	50.52	185.51	70.04	266.07
Observations	2,144	2,144	1,976	1,976	1,947	1,947

Hazard ratios reported. Robust standard errors in parentheses; **p < 0.01, *p < 0.05, †p<0.1.

Of all the battle geography factors in our analysis, the dispersion of civil war battles throughout a country exerts the strongest and most consistent effect on civil war outcomes. As expected, as the geographic dispersion of the civil war battles over the last three months increases, the likelihood of a rebel-favorable outcome increases. This effect is consistent both in the model only containing our theoretical variables and in the models that include controls. An average civil war state with a battle dispersion of 50 miles is nearly 21% more likely to see a rebel-favorable outcome at any point in time than a state with a mean-level battle dispersion of 29 miles. The likelihood of a rebel-favorable outcome grows even more substantially as the rebels challenge the government over large swaths of territory. At the point that the three-month battle dispersion reaches 200 miles, the likelihood of a rebel-favorable outcome at any point in time increases nearly five-fold relative to the mean dispersion of battles.

Figure 1 puts the effect of battle dispersion on the likelihood of rebel favorable outcomes into perspective. When civil war battles are concentrated around only a 10 mile distance, there is less than a 7% chance that the conflict will end with rebel-favorable outcome such as rebel victory or negotiated settlement within 24 months and only slightly more than a 27% chance of such an outcome within 48 months. By contrast, at a battle dispersion level of 200 miles, there is nearly a 32% chance of a rebel favorable outcome within 24 months and more than an 83% chance of a rebel-favorable outcome within 48 months. In short, as rebel groups are able to challenge and impose costs upon the government over larger swaths of territory, the prospects of the rebels forcing a negotiated settlement or achieving outright victory increase significantly.

Figure 1.

Survival until rebel-favorable outcome—battle dispersion. Figure is reproduced in color in online version.

Not surprisingly, the opposite pattern holds for the government. As the dispersion of civil war battles increases, the likelihood of a government-favorable outcome decreases. When the three-month battle dispersion within an average civil war state is at 50 miles, the likelihood of a government-favorable outcome drops by 51% relative to a conflict with a mean dispersion of battles, 29 miles. At the 200 mile battle dispersion level, the hazard of a government-favorable outcome drops by more than 99%, effectively eliminating virtually any chance of a government victory. Figure 2 illustrates this effect: when rebel fighting is confined to a 10 mile battle dispersion, there is a 12% chance of a government-favorable outcome within 24 months and a 29% chance of a positive result for the government at 48 months. This changes markedly once the range of rebel fighting expands to 200 miles. At this point, there is less than a 0.05% chance of a government victory within 48 months.

Figure 2.

Survival until government-favorable outcome—battle dispersion. Figure is reproduced in color in online version.

The two effects of increasing civil war battle dispersion on civil war outcomes make sense given the implications that increased battle dispersion carries for both rebel and government capacity. The demonstrated capacity of a rebel group to operate and fight across large sections of a country provides information that should compel the government to reduce its estimate of its own chances of victory and to increase its estimate of both the amount of time required to achieve victory and the accumulated costs the government will have to absorb in order to achieve victory. Consequently, those governments that find themselves challenged by a rebel force on several fronts across great distances should be more likely to seek a peace agreement than those that can confine the rebels to a more restricted region of conflict. Failing that, those governments are more likely to be defeated outright than are governments that can contain the geographic expanse of the rebels’ combat operations. These findings suggest that key for the government to undermine the prospects for rebel groups forcing concessions through negotiations or winning on the battlefield is containing the geographic scope of rebel activity.

Interestingly, our results in Model 3 point toward an important weakness that rebels face when strategizing their challenge to the government. While rebels are more likely to force a negotiated settlement or win a conflict outright when they fight across large chunks of territory, the prospects for a favorable outcome decrease for the rebels and increase for the government as the range of fighting is in the process of expanding. For an average civil conflict in our sample, the dispersion of battles increases by 0.24 miles over a three month period. When fighting spreads more quickly than this, however, the likelihood of a rebel-favorable outcome drops significantly. For example, when fighting expands by 10 miles over a three-month period, the likelihood of a rebel-favorable outcome drops by 6.5% and the risk of a government-favorable outcome increases by more than 18%. Increasing the three-month rate of battle expansion to 25 miles reduces the chance of a rebel-favorable outcome by more than 15% and boosts the prospects for a government-favorable outcome at any point in time by nearly 54%. This effect, however, is not significant in Model 1, which lacks control variables. This is not surprising as Model 1 does not include other dynamic forces such as cumulative battle deaths that bear upon civil war terminations.

Figures 3 and 4 illustrate the effect that the expansion of fighting has upon the time until a rebel- or government-favorable outcome. At a battle dispersion expansion level of 0.24 miles, there is about an 8% chance of a rebel-favorable outcome within 24 months and a 32% chance of a rebel-favorable outcome within 48 months. When the three-month dispersion of battles expands by 25 miles, the chances of a rebel-favorable outcome decline to 6.8 and 7.8% within 24 and 48 months, respectively. Analogously, the likelihood of a government-favorable outcome within 48 months increases from about 15% to more than 23% when battle dispersion expansion increases from 0.24 to 25 miles. While these results do not describe enormous changes in the prospects for favorable outcomes that rebels and governments face in the wake of an expanding geographic battle space, they do point to an increased risk that rebels face while they are in the process of trying to expand the scope of fighting across the country. These results suggest that rebels that try to rapidly expand the range of fighting run a risk of over-expansion that can undermine their ability to force an agreement or win on the battlefield. Instead, rapid expansion can provide an opening for the government to defeat them.

Figure 3.

Survival until rebel-favorable outcome—battle dispersion velocity. Figure is reproduced in color in online version.

Figure 4.

Survival until government-favorable outcome—battle dispersion velocity. Figure is reproduced in color in online version.

Although battle dispersion exerted a consistent effect upon the ways in which civil wars terminate, we observed more mixed effects with respect to the proximity of civil war battles to a country’s key cities. Contrary to our expectations, we saw little impact of the proximity of battles to major, non-capital cities on the likelihood of a rebel- or government-favorable outcome. While proximity to non-capital cities had a significant effect upon government-favorable outcomes in Model 1, this effect became insignificant once control variables were included in the model. In contrast, the proximity of battles to the capital did have a significant effect upon the outcomes of civil wars. The effect of battle proximity, however, differs in non-secessionist and secessionist conflicts. These predicted survival curves for capital battle proximity in non-secessionist conflicts are plotted in Figures 5 and 6.

Figure 5.

Survival until rebel-favorable outcome—battle proximity to capital (non-secessionist). Figure is reproduced in color in online version.

Figure 6.

Survival until government-favorable outcome—battle proximity to capital (non-secessionist). Figure is reproduced in color in online version.

In non-secessionist conflicts, the likelihood of a rebel-favorable outcome increases and a government-favorable outcome decreases when battles occur closer to the capital. When the closest fighting occurs 370 miles from the capital (the 75th percentile in our data), both the rebels and the government have a nearly identical chance of achieving a favorable result within 48 months, about 25%. When the nearest battles occur within 1 mile of the capital in a non-secessionist conflict, we see a sharp divergence in the prospects of a favorable outcome for the rebels and government. At this proximity to the capital, the rebels have a 16% chance of achieving a favorable outcome within 24 months and a 55% chance of such an outcome within 48 months. When fighting is near to the capital, the prospects for an outcome favorable to the government are bleaker, with less than a 5% chance of a positive outcome within 48 months.

While fighting near the capital is beneficial for rebels in non-secessionist conflicts, our results suggest that battle proximity to the capital in secessionist conflicts actually undermines the chances for a rebel-favorable outcome. Changing the proximity of the nearest battle to the capital in a secessionist conflict from 370 miles to within 1 mile reduces the likelihood of a rebel-favorable outcome within 48 months from 11 to 0.05%. Battles close to the capital in secessionist conflicts, however, reduce the likelihood of a government-favorable outcome as well. At a minimum capital distance of 370 miles, there is a 17% of a government-favorable outcome within 48 months but only a 7% chance of such an outcome when the nearest battle occurs within 1 mile of the capital. This suggests that fighting near the capital in secessionist conflicts points to a stalemated conflict, rather than one in which an end is near.

While this finding first seems counterintuitive, one potential explanation for this result is that targeting the capital represents a strategic error for rebel groups seeking to obtain independence for their homeland and signals their inability to inflict sufficient costs on the government through fighting within the disputed territory. While targeting the capital might be seen as a way for a secessionist group to impose high costs on the regime in the hope that doing so will compel the government to yield to an agreement, threatening the capital may actually be counter-productive for secessionist rebels. Challenging the capital risks overextending a secessionist group’s forces, exposing them to a higher risk of battlefield defeat and thereby potentially weakening their capacity to capture and defend their own homeland further down the road. At the same time, challenging the regime in its capital city may more deeply entrench a belief within the regime that the rebel group represents an existential threat to the incumbent regime, potentially forcing state leaders to conclude that the only means of dealing with a rebel group that has threatened the government in its seat of power is by using any means necessary to defeat the rebels rather than attempt to negotiate a settlement.

While the proximity of battles to the capital significantly influenced civil war outcomes, our results showed no evidence that neither the proximity of fighting to non-capital cities nor the speed by which battles moved toward such cities carried any effect on outcomes. This suggests that, to the extent to which rebels are able to challenge the government across large sections of the country, it is the broad dispersion of battles that most directly imposes the costs and pain on governments that lead to rebel-favorable outcomes, rather than the pressure rebels apply by fighting near non-capital cities.

While our results demonstrate that the geography of civil war battles plays an important role in influencing how civil wars end, our results also show that many of the conventional characteristics of the conflict and its participants are still linked to civil war outcomes. For example, mounting battle deaths over time significantly reduce the likelihood of a government-favorable outcome without significantly impacting the prospects for a positive outcome for the rebels. This suggests that an increasing civil war death toll, while demonstrating that the rebels are able to fight hard enough to avoid a government victory, also appears to sour the belligerents on the value of a peaceful settlement without increasing the potential for a rebel victory. The presence of other threats and the capacity of the government to respond to them also influence the outcomes of civil wars. The chances of a rebel-favorable outcome sharply increase as the number of other ongoing civil war dyads increases, but decrease significantly as government military capability grows.

Conclusion

Existing models of civil war termination have relied on expected utility logic to identify factors that affect protagonists’ choice between continuing to fight in the quest for victory or stopping in defeat or in hopes of negotiating a peace agreement with their rivals. Tests of those models have relied largely on static measures of national attributes and characteristics of the war itself, all measured either at the start of the conflict or after its termination. We know from history that tactical choices during the war can and do affect the outcome in ways that predictions based on static ex ante measures cannot anticipate. We have made an effort in this paper to incorporate some of those battlefield dynamics into the analysis of civil war outcomes. The findings add further support to the expected utility logic that underlies most models of how civil wars end. But the inclusion of changing battlefield parameters provides us with some insights into the conflict process that previous models cannot. And some of these new insights have potentially valuable implications for conflict management efforts.

First, evidence on the impact of battle dispersion suggests that where rebels have demonstrated the capacity to conduct combat operations over a wide swath of the nation’s territory, the prospects for government-favorable outcome are rather remote and, therefore, the government’s best option at that point would be to seek a negotiated settlement. Even if that government chose to continue fighting, it would have to count on absorbing the costs of conflict for a longer period of time; in effect the government would have to reverse the trend in battle dispersion before the prospects of ever achieving victory improved. By the same token, rebels who cannot sustain operations over a wide range of the nation’s territory are unlikely to achieve victory. Their best option is to demonstrate that they can sustain military operations, even if confined to a relatively limited portion of the nation’s territory, and hope that the prospect of absorbing those costs of conflict for a protracted period of time will induce the government to seek a peace agreement. For third parties with an interest in conflict management, these findings suggest that the willingness of the protagonists to accept conflict management offers is probably greater where battles are widely dispersed. Although we did not test this directly, we suspect that, under conditions of both widely dispersed battles and narrowly confined battle dispersion, the prospects for successful conflict management increase with time. Where battles are widely dispersed, the longer the rebels conduct such operations without achieving victory, the more likely they will be to settle for a peace agreement. Likewise, a government that has managed to confine the battle to a relatively limited portion of the nation’s territory should become more willing to negotiate the longer it fights a rebel group without defeating them.

The second set of findings concern the impact of battles near strategically valuable cities. The findings here also lend credence to expected utility logic. The findings suggest that, in non-secessionist conflicts, as long as the government can confine the fighting to regions away from the capital, the chances of a rebel victory are diminished. Secessionist rebels who stray from their homeland and approach the capital face a greater risk of defeat. Advancing on other major cities, for both revolutionary and secessionist rebels, does not necessarily increase their own prospects of a favorable outcome. Instead, it is the ability of rebels to fight across large portions of a country’s territory, rather than proximity to major cities, that improves their prospects for a favorable outcome. In this respect, the geographic scope of fighting seems to bear more directly on the outcomes of civil wars than the geographic locations of fighting.

In general, these findings suggest that governments that are confronted with fighting spread over a wide range of territory, especially for a protracted period of time, could be persuaded that their best option is to seek a negotiated settlement. Their chances of victory are remote, in terms of both the probability and the time required to achieve victory. Therefore, the costs of continuing to fight will accrue for a long period of time. Just as rationalist models of war imply that, with complete information about each other’s capabilities and resolve, government and rebels should be able to negotiate an agreement prior to war that would leave them better off than any outcome they could hope to achieve from war, so the information revealed by widely dispersed fighting should persuade a government that there should be a set of peace agreement terms that the government would deem preferable to continued fighting in a war that appears unlikely to end in a government victory in the foreseeable future.

Our findings also point to some avenues for future research to further develop our understanding of how the way in which events unfold on the battlefield shapes the opportunities for conflict management and how civil wars end. For example, our findings regarding the proximity of battles to a state’s capital depart significantly from our theoretical expectations. Our analysis, however, only focuses upon the locations of battles, not their outcomes. Research that incorporates how battle outcomes shape the subsequent behavior of civil war belligerents and the opportunities for conflict management would be a valuable addition to the literature. At the same time, while this paper has offered some insights into the ways in which where rebels fight the government influence the manner in which the government responds to the rebel threat, further research could also explore how the locations of battles and their success and failure influence the subsequent expectations of each protagonist. For example, as rebels grow more able to threaten the capital, do they remain cohesive, or do they splinter between those who seek a negotiated settlement with the government and those who demand to fight on until the government is defeated? How do governments respond to these two distinct potential rebel responses to fighting near the capital?

Footnotes

Author note

Replication materials, including the dataset and relevant do files, are available online through the Sage CMPS website. They can also be obtained from the authors by email request: Michael.Greig@unt.edu or David.Mason@unt.edu or Jesse.Hamner@unt.edu.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Notes

References

Brandt

Mason

Gurses

Petrovsky

Radin

(2008) When and how the fighting stops: Explaining the duration and outcome of civil wars. Defence and Peace Economics19(6): 415–434.

Central Intelligence Agency (2009) The World Factbook. Available from https://www.cia.gov/library/publications/the-world-factbook/index.html (accessed 8 March 2013).

Cunningham

(2011) Barriers to Peace in Civil War. New York: Cambridge University Press.

Cunningham

Gleditsch

Salehyan

(2009) It takes two: A dyadic analysis of civil war duration and outcome. Journal of Conflict Resolution53(4): 570–597.

DeRouen

Bercovitch

(2011) Introducing the Civil Wars Mediation (CWM) Dataset. Journal of Peace Research48(5): 663–672.

DeRouen

Sobek

(2004) The dynamics of civil war duration and outcome. Journal of Peace Research41(3): 303–320.

Fearon

(1995) Rationalist explanations for war. International Organization49(3): 379–414.

Fortna

(2015) Do terrorists win? Rebels’ use of terrorism and civil war outcomes. International Organization69(summer): 519–556.

Greig

(2015) Rebels at the gate: Civil war battle locations, movement, and openings for diplomacy. International Studies Quarterly59(4): 680–693.

10.

Gurses

(2015) Transnational ethnic kin and civil war outcomes. Political Research Quarterly68(1): 142–153.

11.

Kalyvas

(2006) The Logic of Violence in Civil War. New York: Cambridge University Press.

12.

Karnow

(1983) Vietnam: A History. New York: Viking Press.

13.

Kreutz

(2010) How and when armed conflicts end: Introducing the UCDP Conflict Termination Dataset. Journal of Peace Research47(2): 243–250.

14.

Marshall

Jaggers

(2002) Polity IV Project: Political Regime Characteristics and Transitions, 1800–2002. Dataset Users’ Manual. Center for International Development and Conflict Management, University of Maryland.

15.

Mason

Fett

(1996) How civil wars end: A rational choice approach. Journal of Conflict Resolution40(4): 546–568.

16.

Mason

Weingarten

Jr Fett

(1999) Win, lose, or draw: Predicting the outcome of civil wars. Political Research Quarterly52(2): 239–268.

17.

Regan

(2002) Third-party interventions and the duration of intrastate conflicts. Journal of Peace Research46(1): 55–73.

18.

Shelton

Stojek

Sullivan

(2013) What do we know about civil war outcomes?International Studies Review15: 515–538.

19.

Sisk

(2009) International Mediation in Civil Wars: Bargaining with Bullets. New York: Routledge.

20.

Stam

III (1996) Win, Lose, or Draw: Domestic Politics and the Crucible of War. Ann Arbor, MI: University of Michigan Press.

21.

Stanley

(1996) The Protection Racket State: Elite Politics, Military Extortion and Civil War in El Salvador. Philadelphia, PA: Temple University Press.

22.

Sundberg

Melander

(2013). Introducing the UCDP georeferenced event datasetJournal of Peace Research50(4): 523–532.

23.

Svensson

(2007) Bargaining, bias and peace brokers: How rebels commit to peace. Journal of Peace Research44(2): 177–194.

24.

Thibos

(2014) 35 Years of Forced Displacement in Iraq: Contextualising the ISIS Threat, Unpacking the Movements. Florence, Italy: Migration Policy Center, European University Institute.

25.

Thyne

(2015) The impact of coups d’état on civil war duration. Conflict Management and Peace Science, first published 25March2015; doi:10.1177/0738894215570431

26.

Walter

(2009) Bargaining failures and civil war. Annual Review of Political Science12(1): 243–261.

27.

Wood

(2014) From loss to looting: Battlefield costs and rebel incentives for violence. International Organization68(4): 979–999.

28.

Wood

Kathman

(2013) Too much of a bad thing? Civilian victimization and bargaining in civil war. British Journal of Political Science44: 685–706.

29.

Zartman

(1995) Elusive Peace: Negotiating an End to Civil Wars. Washington, DC: Brookings Institution Press.