Why Do Some Militaries Kill Civilians During Civil War, While Others Do Not?
Domestic politics play a major role
In August 2017, residents of Tu Lar To Li fled as Burmese soldiers closed in on the remote village located in Burma’s Rakhine state, according to the testimony of 20-year-old Hasina in a recent Human Rights Watch report.
In the wake of years of ethnic cleansing between Buddhist and Muslims in Rakhine state, and the incursion of government troops, Rohingya insurgents have reemerged and begun to fight. In keeping with a counterinsurgency strategy that has defeated nearly 60 insurgencies over as many years, the Burmese military has taken to destroying entire villages.
Hasina witnessed the brutal counterinsurgency operations first hand. As the villagers were trying to flee “some were trapped on a riverbank and she saw dozens murdered on the beach,” Human Rights Watch reported. “She tried to hide her infant daughter under her shawl, but a soldier noticed the baby, snatched her away and tossed her into the fire.”
The soldiers later attempted to rape Hasina and stabbed her mother-in-law to death. The soldiers also murdered three children related to Hasina using spades. And before departing the soldiers attempted to burn Hasina and her family alive in their own home, according to her testimony in the report.
This incident is but one of many in Rakhine State. Mass killings, rape and scorched-earth tactics are clearly a part of an intentional strategy by the Burmese military, according to Human Rights Watch.
Media coverage of civil wars often focuses on such horrifying massacres, justifiably leading many to believe that most military’s fight insurgents the same way as the Burmese army. But a recent study by University of Minnesota professor Jessica Stanton found this perception to be wrong.
Looking at all civil wars between 1989 and 2010, Stanton finds that “24.5 percent of governments massacred civilians, 47.1 percent burned civilian homes and crops, 21.6 percent deliberately bombed or shelled civilian targets and 13.7 percent forcibly expelled civilians from territory. Yet, 49 percent of governments refrained from using any of these four forms of violence.”
Stanton notes that some violence against civilians occurs in all civil wars, even among the 49 percent of militaries that practice restraint. For instance, U.S. and allied forces in Afghanistan have largely abided by the law of armed conflict, according to Human Rights Watch. Yet, in 2012, one U.S. soldier systematically murdered 16 civilians in Kandahar province, Afghanistan.
However, Stanton and other researchers are interested in studying why some militaries intentionally include killing civilians as a part of their overall counterinsurgency strategy, while others do not. Focusing on how militaries act as a whole, rather than focusing on aberrant soldiers, allows researchers to offer concrete policy recommendations on how to prevent systematic civilian killing in civil war.
Since the early 1990s academics in the field of political science have produced a trove of valuable studies on violence against civilians in civil war. One of the most consistent findings in the academic literature is that battlefield dynamics have a strong bearing on whether a state decides to target civilians.
Yale University professor Stathis Kalyvas’ The Logic of Violence in Civil War is one of the most important studies on battlefield dynamics and civilian targeting. Kalyvas’ book investigates why combatants “selectively” or “indiscriminately” target civilians caught in the middle of a civil war.
A homicide is “selective” when a combatant has information that a civilian is an enemy collaborator. For example, during the Nazi occupation of Greece during World War II, “Germans raided the village [of Anifi] and used a hooded informer to arrest several men whom they later shot. This is an instance of selective violence,” according to Kalyvas.
A homicide is “indiscriminate” when the combatant possesses no such information. As an example, Kalyvas points to a story of German troops massacring all civilians outside past curfew during the occupation of Greece.
It’s true that in either case, civilians are killed. But making this distinction answers important question on why civilians are killed in civil war that would be impossible to answer otherwise.
Kalyvas argues that the level of selective or indiscriminate targeting varies within civil wars, and that this variation depends on level of territorial control enjoyed by a combatant. This explanation applies to both government and rebel forces.
At top — Iraqi troops in Mosul in 2017. Matt Cetti-Roberts photo. Above — a Rohingya refugee camp in Myanmar. Photo courtesy of Rohingya activists
Take a situation in which government troops occupy 80-percent of a village, but rebels control the remaining 20 percent. In this case, government troops are likely to mainly practice selective killings, while rebels are likely to kill indiscriminately.
Kalyvas explains this variation by arguing that higher levels of territorial control mean combatants can collect better information on the identity of enemy collaborators. Thus, combatants with more control can more accurately, or “selectively,” kill civilian collaborators, while the combatant with less control are forced to kill indiscriminately.
But when a village is fully under government control, troops are unlikely to commit any violence against civilians. This is because control also means combatants can better deter collaboration with the enemy. If violence does occur, however, it’s likely to be selective.
Within the government-dominated territory rebels practice indiscriminate violence against civilians. But they do so at a much higher level than when they maintained some control. This is due to the reality that rebels without any control of a village can’t deter collaboration with the government, and can’t collect information on collaborators. In this scenario rebels are faced with a higher number of potential collaborators, and thus more people to kill.
The theory is fairly intuitive up to this point. But there is one final territorial control scenario.
Kalyvas notes that in some rare instances combatants split control of a village more-or-less 50-50. That is, the government fully controls the village during the day, and rebels fully control the village at night. In this scenario, both combatants carry out very little violence against civilians. If any.
Under a parity of control, no civilians are willing to inform about enemy collaborators to either combatant. Civilians are aware that allies of those that are killed maintain equal ability to inform, as well.
In other words, the guarantee of reciprocity for informing deters collaboration.
Kalyvas provides compelling evidence. But only from the German occupation of Greece during World War II. And mainly in the Arigold region of Greece. His data sources include interviews with dozens of people alive during the occupation. But also rare archival data that includes court documents providing detail of who was killed and why in the Arigold region.
He illustrates his theory with highly detailed case studies, and tests a portion of his theory statistically. But the descriptive statistics are probably his most compelling evidence. He codes the level of control for 61 villages in the Arigold over the course of the civil war, and the number and location of civilians selectively or indiscriminately killed by each side.
In total, 725 civilians were killed, 366 selectively and 359 indiscriminately. The Germans and allied militia were responsible form 48.69 percent of the civilians killed, and rebels the other 51.31 percent.
The pattern of killing generally conforms to his theoretical predictions. But the rebels did tend to kill a higher number of people selectively than expected when control was 50-50. Also somewhat outside of the theory is the fact that the incumbents — German troops or allied militias — also selectively killed more civilian when they had total control.
Kalyvas explains these deviations with the reality that indigenous rebels naturally have better access to local information, and, for similar reasons, practice more credible deterrence capabilities.
One major criticism for Kalyvas’ work is that it only covers one region during one conflict, and thus can’t be generalized across cases. Although a 2009 study by Kalyvas and Adman Kocher shows the theory largely explains variations in violence in the case of Vietnam, there’s some truth to this.
But such data on selective/indiscriminate civilian targeting is nearly non-existent in most civil wars. This is especially the case for civil wars that took place before special tribunals, NGOs, and journalists kept track of civil killings.
In short, understanding how territorial control and information effects civilian killing would be nearly impossible in a study that includes all civil wars. Thus, to provide practitioners and academics fine-grain knowledge of how civilian-killing actually works during war, scholars need to provide mirco-comparative studies like Kalyvas’s.
At least three other studies have since used “selective” and “indiscriminate” targeting as the main measure for civilian targeting. And the distinction between “selective” and “indiscriminate” targeting actually helps us better understand the patterns of ISIS’s executions in Iraq.
Still, as Stanton’s data shows, 49 percent of government forces refrain from intentionally killing civilians, and thus would not fall under the “selective” or “indiscriminate” categories.
The German occupation of Athens during World War II. Photo via Wikipedia
Much of the literature on battlefield dynamics and civilian targeting focus on trying to understand why some militaries target civilians while others don’t.
One of the first and most important studies in this category is the book Final Solutions: Mass Killing and Genocide in the 20th Century by Dartmouth University professor Benjamin Valentino. A major component of the book focuses on explaining why mass killings occur in some wars, but not others.
To be considered a “mass killing” a government must kill 50,000 or more civilians over a five-year period or less. This is a much higher standard than Stanton’s, and others’ definitions of “civilian targeting.” “Mass killing” is a measure attempting to essentially understand genocide, while “civilian targeting” simply measures whether a combatant intentionally kills civilians as a part of an overall strategy.
But understanding Valentino’s argument is useful for our purposes nonetheless, since the basic mechanisms in his theory are tested in most studies by scholars investigating “civilian targeting.”
Valentino argues that states are most likely to kill mass numbers of civilians in war when insurgents rely on guerilla warfare as their main strategy. The logic for this is fairly simple. Guerilla warfare usually requires insurgents to rely on civilians for recruits, food, shelter, and even revenue through taxation. In other words, civilians are the main source of power for guerillas.
Valentino argues that militaries often resort to mass killing in order to cut off the insurgents’ power base. Using a metaphor to explain his argument, Valentino notes that Mao Zedong once described guerillas as fish and that civilians are the water in which they swim. Many militaries metaphorically drain the water in order to more easily catch the fish.
However, Valentino admits that most militaries fighting guerillas do not resort to mass killing. He says that two factors explain this variation.
First, if a counterinsurgent believes “that their adversaries are not receiving substantial support from local civilian populations, they have little incentive to target large numbers of civilians in counterguerrilla operations,” according to Valentino.
He points out that the Rhodesian military abstained from killing civilians in mass numbers during their fight against anti-colonial rebels in the 1970s largely because rebels based operations in neighboring Zambia and Mozambique, and thus had little support from civilians in Rhodesia, now Zimbabwe.
A second reason that may induce restraint from mass killing is “when insurgent military operations remain at levels that do not pose a major military threat to the government,” according to Valentino. He adds that when the insurgent threat is low, mass killing is likely to be averted even when guerillas garner heavy support from local civilians.
For example, the Irish Republican Army received significant support from Catholics in Northern Ireland. However, “the IRA and related paramilitary organizations were responsible for approximately 1,800 total deaths, including 640 civilians, in thirty years of fighting in Northern Ireland,” Valentino writes.
To put the low casualty numbers in perspective Valentino points out that the 64 people killed during the insurgency in 1984 amounted to only “about one-third the number who died in traffic accidents in Ulster that year.” The IRA violence “put political pressure on the British government but gave it little incentive to resort to systematic killing of civilians.”
Valentino’s book laid the foundation for many future studies on civilian killing in war. Yet the book doesn’t test his argument, but instead illustrates his argument using a number of historical examples.
However, Valentino went on to statistically test his theory in 2004 with the help of two colleagues in a study titled “Draining the Sea: Mass Killing and Guerrilla Warfare.” This study covers both civil wars and wars between states, but still provides important insights into the connection between battlefield dynamics and civilian targeting.
To determine whether insurgents relied primarily on guerilla warfare the authors use historical case studies of each of the 147 conflicts between 1945 to 2000 in their dataset. The severity of the guerilla threat was determined by the size of the guerrilla force and their ability to kill government troops.
The level and scope of civilian support was determined through historical accounts as well. If the war was fought guerilla-style and active civilian supporters numbered 100,000 and over, the insurgents were coded as having significant civilian support.
The authors find that guerilla warfare itself has a significant impact on the likelihood of mass killing. As does the severity of the conflict, including its duration, and to an even greater extent the level of civilian support for insurgents.
Although covering interstate wars between 1900 and 2003, two other studies find that battlefield dynamics provide a strong explanation for systematic targeting of civilians during war. These two studies, George Washington University professor Alex Downes’ “Targeting Civilians in War,” and Valentino and colleagues’ “Covenants Without the Sword,” find guerrilla warfare to be a significant predictor of civilian targeting.
Iraqi troops in Mosul in 2017. Matt Cetti-Roberts photo
But they also point to other battlefield dynamics as fairly good predictors of civilian targeting, including wars of attrition and annexation, wars with high battle deaths and wars with a long duration.
Civilian-targeting during wars of attrition “is a coercive strategy meant to sap the morale of an adversary’s population or undermine the enemy’s ability to fight,” Downes argues. And wars of annexation often necessitate the removal of “an adversary’s population to gain control over the conquered area,” he adds.
High battle deaths and long duration increases the likelihood that states will resort to civilian targeting during these wars in order to decease the costs of fighting. Essentially, states become desperate to win.
While these studies cover interstate wars only, when combined with the findings of Kalyvas and Valentino, they provide justification for the argument that battlefield dynamics have a significant bearing on the likelihood of civilian targeting in civil war.
Indeed, John Carroll University professor Jen Ziemke’s 2012 study “Turn and Burn: Loss Dynamics and Civilian Targeting in the Angolan War” presents a theory of civilian-targeting in civil war that combines the battlefield desperation arguments forwarded by Downes and Valentino with the territorial control argument forwarded by Kalyvas.
Ziemke spent two years using newspapers to geo-locate civilian killings, battles and changes in territorial control by all combatants during the four decades of the Angolan civil war.
She finds significant statistical evidence that the loss of territory was the main reason combatants from either side intentionally murdered civilians. Periods of “sustained territorial loss,” the duration of the conflict, and the shift from conventional to guerilla war were all factors that increased combatants’ propensity to kill civilians upon the loss of territory.
“During times of loss, the number of civilians considered potential enemies, traitors or defectors increases rapidly because the cost of being wrong escalates drastically,” Ziempke explains. “And because sustained territorial and battlefield losses increase the odds that civilian allies may defect to join the winning side, losers end up killing civilians of all persuasions, even those allies who had always been considered longtime supporters, in order to scare them and prevent the defection of the rest of the group.”
“In short, the crucial problem of civilian targeting should not be seen as a static phenomenon: rather, loss severely exacerbates this problem.”
Scholars also study how internal government characteristics of a state and pressure from the international community affect civilian targeting in war. Importantly, many of these studies also account for the role of battlefield dynamics.
University of Wisconsin-Madison professor Scott Straus’ “Making and Unmaking Nations” provides a compelling theory for the most destructive form of civilian targeting during civil war: Genocide.
Straus argues that the main causes of genocide during civil war are a combination of battlefield dynamics, and the “founding narratives” of the state.
Over time, leaders within a state construct a “founding narrative” that one group within the population should dominate the economy and politics of the country, and other groups should be excluded.
Take for instance the view by the Hutu-led government in Rwanda that ethnic Hutus should dominate the economy and politics of the country. The Ottoman Turks held a similar view on the subordination of the Armenians and the Tutsi-led government of Burundi had a similar view of ethnic Hutu.
But in all of these cases, these “founding narratives” did not result in genocide until the excluded group was perceived to be a military threat. “As the military crisis escalates, so do the fears that a victory for the group representing the excluded group will result in devastation for the primary group,” Straus explains.
“In military crises, moreover, state elites are more likely to associate the military enemy with the dangerous social group such that winning the war becomes synonymous with destroying the threating social group,” he adds.
However, exclusionary political narratives plus military threat does not always equal genocide. A few factors can mitigate genocide. Influential members of society can blunt exclusionary narratives by propagating a counter-narrative stressing inclusiveness. But counternarratives are really only possible in democracies, according to Straus.
Genocide also requires that the state military is competent enough to carry out the complex and often resource-intensive task of identifying, rounding up, and killing of an entire group. Especially since the target group usually has armed organizations on their side.
A Rohingya boy walks over an open sewage ditch in the Sithmagi IDP camp. Adryel Talamantes photo
Some states practice restraint from genocide because their economy would be negatively impacted by mass killings, such as agriculture and tourism-based economies. Finally, international intervention can stop a genocide from occurring through economic sanction and military intervention.
Straus illustrates his theory through comparative case studies of civil wars from sub-Saharan Africa. To some, this approach is not as rigorous as cross-national statistical testing. But like Kalyvas and Ziemke, Straus is able to more fully identify how his theory actually functions in war.
However, a 1993 study by Barbra Harff does find statistical evidence that governments practicing exclusionary politics are more likely to commit genocide during civil war than inclusive governments.
Interestingly, Harff also tests the role of the international community in genocide. She finds that governments with higher levels of trade openness were less likely to commit genocide. However, membership to regional or international organizations had no effect on the likelihood of genocide onset.
Other evidence about the role of the international community are mixed. Valentino and colleagues’ 2006 paper, discussed earlier, also found no evidence of a restraining effect for states signatory to various international legal treaties prohibiting civilian targeting during war.
However, military intervention by foreign governments finds mixed results. A 2005 study by Krain finds that foreign military intervention reduces the severity of genocides or stops the brutality all together.
A 2012 study by Wood and colleagues finds that such intervention decreases the likelihood of civilian targeting by the side the foreign military allies itself with. However, such biased intervention increases the likelihood and severity of civilian targeting by the opposing combatant.
This outcome is explained using a similar theory forwarded by Downes, and Valentino and colleagues. Biased intervention increases the desperation of the opposing combatant, according to the 2012 study.
One last international characteristic studies by civilian targeting scholars is regime type. One might think that democratic countries would be less likely to target civilians during war. Democratic leaders should be sensitive to the costs incurred when voters see their own troops murdering civilians.
Surprisingly, there is mixed evidence for this argument.
Valentino and colleagues’ 2004 paper claims that a government’s level of democracy reduces the likelihood of mass killing. However, their 2006 paper found no effect for democracy on civilian targeting. While Downes finds that democracies are actually more likely to target civilians than autocrats are.
Downes explains that democracies are more sensitive to the costs of losing a war than autocrats are. Democratic leaders view losing a war and practicing restraint as potentially costlier at the ballot box than targeting civilians and winning, according to Downes.
University of Minnesota professor Jessica Stanton’s book Violence and Restraint in Civil War statistically and qualitatively tests most of the theories presented in this article. Stanton argues that state leaders “use violence and restraint strategically, weighing the costs of violence based on an assessment of their own relationship with domestic and international constituents and weighing the benefits violence based on an assessment of their opponent’s relationship with its constituents.”
When the costs are disproportionate to the benefits, states intentionally practice restraint. However, when the benefits outweigh the costs, leaders choose to target civilians.
This is a fairly intuitive answer to why some governments target civilians, while others practice restraint. The difficult part for researchers is finding a way to identify “domestic costs” and “international costs” that can be identified across numerous cases.
Stanton argues that states that are democratic, feature unstable regimes or are ethnically inclusive governments all would incur heavy costs from their constituencies for targeting civilians in civil war. Democracies intentionally practice restraint because leaders fear the costs at the ballot box, explains Stanton.
Still, some democracies exclude ethnic groups from participating in government. And some autocracies “seek to appeal to a broad domestic audience, even though they might lack domal institutional mechanisms to enforce government accountability to the domestic public.”
Thus, it’s important to also distinguish between governments that are ethnically inclusive from those that are ethnically exclusive, according to Stanton. She argues that the more inclusive a government is, the more likely it will practice restraint. She measures variance in inclusiveness using the oft-cited
Finally, leaders of unstable governments fear that civilian targeting will cause constituents to pull support, resulting in the collapse of their already fragile regime. Unstable governments are measured using the Polity IV database. This database measures regime type using a score between -10 (most autocratic) to +10 (most democratic).
A government is considered “unstable” if, five years or less prior to the onset of rebellion, its regime score changed by three or more within three years. In addition, regimes that incurred a military coup attempt prior to rebellion are considered unstable.
States that are the most susceptible to international costs of targeting are those with unstable governments, and those that began fighting at the end of the Cold War. Unstable governments fear they will collapse if Western states in the international community pull their support.
The end of the Cold War ushered in an era in which governments “use international human rights and humanitarian norms as a basis for seeking international support,” according to Stanton. Thus, governments that begin fighting after 1989 should be more prone to practice restrain in order to gain support of Western states.
The benefits of ethnically-cleaning civilians should be greatest when constituents for the rebel group are concentrated in one area of the country, according to Stanton. Further, governments are likely to practice “terrorism” against civilians when rebel groups try to actively govern their own constituents. Think ISIS in Mosul prior to the Iraqi offensive.
Guerilla war, however, should not entice civilian targeting, according to Stanton. And neither should high battle deaths, government troops fighting a relatively large number of rebels or variance in territorial control. Essentially, none of the battlefield dynamic theories identified in this article should help explain a government’s decision to victimize civilians, according to Stanton.
Finally, Stanton also tests other international factors. She notes whether a state is signatory to international laws concerning treatment of civilians in war, or is party to international or regional political organizations. She also measures a state’s level of trade with other countries.
Stanton’s statistical results confirm her cost-benefit theory. Further, the battlefield dynamic theories in this article do not reach conventional levels of statistical significance in Stanton’s model. However, Kalyvas’ territorial control theory can’t easily be measures in a cross-national statistical analysis. Thus, this theory is tested in the case studies. Stanton finds that her theory works better.
Finally, none of the alternative theories for the influence of the international community explain restraint. That should come as no surprise. Studying organized violence is extremely difficult.
For instance, Stanton defines “inclusive governments” as those “in which no ethnic group has “dominant” or “monopoly” control.” Therefore, Stanton defines the Apartheid regime in South Africa as “inclusive” by Stanton because whites split power between Afrikaners and English-speakers. Stanton uses the ethnic power relations dataset to code “inclusive” and “exclusive” governments. Since this dataset also accounts for whether a group is “powerless” or “discriminated” against, there’s no reason Stanton shouldn’t have also included these two definitions in here coding of “exclusive governments.”
Further, none of the statistical studies on civilian targeting control for civil wars fought between two or more conventional armies with armor and other sophisticated weapons, or civil wars between symmetrically powerful forces using mainly light weapons. This is a serious flaw, seeing as there is robust evidence that these types of wars produce distinct battlefield dynamics and types of armed organizations.