Pages

3 August 2017

Do Civilian Casualties Cause Counterinsurgents to Fail?


“The thing about counterinsurgency is that it doesn’t really work,” the film’s narrator says. “We tried it in Vietnam. That went well. The British and the French gave it a shot, trying to hang on to their crumbling empires. It just hasn’t worked. To me, it would seem kind of simple why. You can’t win the trust of a country by invading it. You can’t build a nation at gunpoint.”

The film suggests a simple logic to back this message. A counterinsurgent must win over the “hearts and minds” of the civilian population in order to win the war.

However, a counterinsurgent that kills civilians in the course of defeating insurgents can never win “hearts and minds.” Thus, because defeating insurgents hiding among civilians almost always results in civilian casualties, counterinsurgency is impossible

We could brush this assertion off as “just Hollywood.” However, one of the most critical influences on counterinsurgency doctrine, U.S. Army Counterinsurgency Field Manual 3-24, holds a similar view. With one crucial caveat, of course.

FM 3-24 argues that excessive civilian casualties will cripple counterinsurgency operations, possibly to the point of failure. This is especially the case when the counterinsurgent doesn’t seek popular support by implementing public works projects and rendering other forms of aid, according to the manual.

Still, FM 3-24 is not clear on what exactly constitutes “excessive civilian casualties.” The manual’s authors would likely deem civilian casualties as excessive if a ground unit had the option of using more discriminate firepower to kill the enemy but chose otherwise.

For instance, let’s say a unit takes a burst of small-arms fire from insurgents hiding in a densely-populated village. Civilian casualties would likely be deemed excessive if the unit chooses to saturate the village with artillery rounds when it had the option of dropping a laser-guided BLU-126/B, Low Collateral Damage Bomb from an aircraft. True to its name, the bomb is designed to reduce unintended deaths.

The American public’s perception of counterinsurgency seems to align with the idea that excessive casualties lead to failure in counterinsurgency wars. National security experts have made similar statements for over a decade in op-eds, and popular books.

As impactful as these sources are, almost none include systematic research to back up the idea that excessive civilian casualties increase the probability of failure. This includes FM 3-24, which was written prior to the publication of systematic academic work on the subject.

Instead, the theoretical foundation of the manual is mainly derived from books written by British and French counterinsurgency practitioners that fought colonial rebellions in the 1950s and ’60s.

Practitioners’ viewpoints are valuable. But their conclusions are mostly based on anecdotes from first-hand experience. Most academic studies, on the other hand, base their conclusions on systematic evidence collected from interviews with a range of counterinsurgency practitioners as well as declassified documents and press reports.

Finally, war is not abstract for most of the academics discussed in this article. Most have taken great personal risk to interview combatants and civilians in war zones around the world.

Armored vehicle from the Ugandan African Union contingent, Mogadishu, Nov. 25, 2007. David Axe photo
People’s war

Whether it’s FM 3-24 or popular media, the conclusion that excessive civilian casualties can lead to failure rests on one fundamental assumption. That is, a civilian population caught in the middle of a war zone has the ability to choose which armed actor to support.

Many academic studies share this assumption. Although not all agree that excessive casualties lead to failure.

To be clear, the “civilian agency” assumption implies that the leaders of villages, towns, or cities and even insurgent groups are either held to the will of the population, or are cut out of the decision to support or reject armed actors.

Such “civilian agency” in war may seem problematic. But studies founded on this assumption rely on seemingly intuitive logic. Academics agree that one of the most difficult parts of counterinsurgency is called the “identification problem.” Since insurgents hide in the civilian population and don’t wear uniforms, counterinsurgents need information from the population to successfully clear an area of insurgents.

In addition to information, both counterinsurgents and insurgents need recruits and material support from the population to conduct military operations.

Academic studies that use the “civilian agency” model fundamentally agree that civilians can choose which side to fight for, and which side to provide information to. However, these studies disagree on how “excessive civilian casualties” will affect the civilian population’s decision about which side to help.

1 Scots Battlegroup trooper, Shaiba, Iraq, Dec. 15, 2007. David Axe photo
Insurgent math

Let’s start with the studies that argue civilian casualties are harmful for counterinsurgents.

Matthew Kocher and colleagues analyzed every air strike in every hamlet in South Vietnam between July and December 1969 to understand how indiscriminate bombing affects territorial control by insurgents.

The data may sound too good to be true. But it’s not. American military forces recorded every airstrike in South Vietnam between 1965 and 1970. The records even include coordinates.

In addition, the authors were able to gauge the level of insurgent control by analyzing records from the Hamlet Evaluation System. The HES program required that American troops conduct monthly or quarterly surveys of every hamlet in South Vietnam in order to evaluate the pacification strategy.

Using GIS mapping to plot the airstrikes, and a series of statistical tests, Kocher and colleagues find that bombing actually decreased American and South Vietnamese control of hamlets. In addition, “bombed hamlets that were previously under insurgent control were more likely than unbombed hamlets to remain under insurgent control six months later,” according to the authors.

Kocher and colleagues argue that the population will collectively deem supporting the insurgency as increasing their odds of survival when civilians are killed indiscriminately.

In contrast, “selective targeting” of civilians would have likely avoided this outcome, according to the authors. By “selective targeting,” they mean killing individuals who are guilty of supporting the insurgency.

When “guilty” civilians are killed, according to this logic, the surrounding population understands that withholding support from the insurgency will ensure its safety. Indiscriminate killing, on the other hand, causes the population to seek the relative safety of insurgents.

However, the “selective” versus “indiscriminate” targeting conclusion remains theoretical. The authors do find that bombing in Vietnam was counterproductive. But due to a lack of available data, are not able to directly assess whether civilians actually helped insurgents maintain control of bombed hamlets.

Further, the “selective targeting” theory receives mixed anecdotal support in a previous study, and no others that I’m aware of have tested the theory with systematic evidence.

Luke Condra and Jacob Shapiro provide another study of American counterinsurgency, but this time in Iraq. The authors try to understand how unintended civilian casualties by all sides of the conflict affected the number of insurgent attacks between 2004 and 2009.

Data for “collateral damage” deaths comes from the Iraq Body Count, a non-profit organization that has tracked violent deaths in Iraq since 2003. Data for insurgent attacks is derived from the coalition’s Significant Activity Reports database.

This database features more than 193,000 “executed enemy attacks targeted against coalition, Iraqi Security Forces, civilians, Iraqi infrastructure and government or organizations.”

Condra and Shaprio plot the insurgent attacks and civilian deaths on a map covering each of Iraq’s 104 districts between 2004 and 2009. They find that as collateral damage deaths caused by the coalition and Iraqi Security Forces increase, so do the number of attacks by insurgents. Interestingly, collateral deaths caused by insurgents actually decreases the number of insurgent attacks against the coalition and ISF.

The authors argue that this outcome is explained by the fact that collateral deaths cause the population to intentionally punish the perpetrator by reducing outflows of information on the enemy.

However, there is one major caveat to their findings. The study shows that the effect of collateral damage on insurgent attacks is strongest in urban areas, and areas featuring a mixed population of Sunni and Shia.

The authors argue that this outcome is explained by the fact that mixed districts hold diverse “political preferences and so there are more people whose behavior can be swayed by civilian casualties.” And in “urban areas, there are more noncombatants around to observe insurgents’ activities and it is harder for insurgents to wield a credible threat of retribution against informers,” according to the authors.

Condra and Shapiro’s study also tests how the coalition’s “hearts and minds” campaign effected these outcomes. To do so they measured the amount of money coalition forces spent on public works projects in each district.

Overall, statistical tests show that the money had no effect on reducing insurgent attacks. This finding contradicts the assertion in FM 3-24 that the provision of public works helps to overcome a communities’ decision to withhold information in reaction to collateral damage.

To summarize, Condra and Shapiro argue that “hearts and minds” operations won’t help a counterinsurgent gain information from the community. Instead, information increases as the number of unintentional deaths decrease.

However, like Kocher and colleagues, Condra and Shapiro can’t directly observe whether information from the population actually accounted for variation in insurgent attacks. The intelligence is still classified.

Moving to the current war in Afghanistan, Jason Lyall and colleagues take a unique approach to understanding how collateral damage effects civilian support of all combatants.

Public opinion polling would likely be the best way to gauge a counterinsurgent’s popularity. Unfortunately, reliable polling in a war zone doesn’t exist. A government loyalist probably won’t express his derision for the government to a Taliban pollster. And vice versa.

So instead of directly asking Afghans their opinions of International Security Assistance Force and the Taliban directly, Lyall and colleagues devised what’s called an “endorsement experiment.”

The experiment, conducted in 2011, involved randomly selected male residents of Pashtun areas being “asked to express their opinion toward a policy endorsed by a specific actor whose support level [the authors] wishes to measure (here, ISAF and the Taliban).” The endorsed policies include prison reform, anti-corruption initiatives, and electoral reform. All policies were selected to avoid “don’t know” answers.

The authors then “contrasted [responses to policies with endorsements from ISAF or the Taliban] with those from a control group of respondents that answered an identical question without the endorsement. Higher levels of enthusiasm for a policy with an endorsement relative to those without it are viewed as evidence of support for the endorsing actor.”

Afghanistan in 2011 posed some difficulties for the researchers.

It was not possible to simply call Afghans in Pashtun tribal areas. And three American academics from the Ivy Leagues surely weren’t going to trapes around Taliban country conducting public surveys. Instead, locals were hired through a private Afghan polling company to conduct the surveys.

The survey is immense, and includes “2,754 male respondents in 204 villages within 21 districts of five Pashtun-dominated provinces.” This includes Helmand province, where men conducting the survey had to brave improvised explosive devises, fire-fights, check-points and highway robbery. In fact, one of the men hired to take the survey was injured by and IED.

Without even accounting for the effect of collateral damage, the survey shows that neither the ISAF’s nor the Taliban’s policies were popular in 2011. In a result reminiscent of “favorability” numbers for Donald Trump and Hillary Clinton throughout 2016, respondents overwhelmingly viewed both the ISAF and the Taliban negatively. But the survey does show that respondents generally disliked the Taliban far less than they disliked ISAF.

The measure of “collateral damage” comes from self-reports by respondents. Survey-takers asked whether the respondent experienced physical harm. Respondents were also asked whether his family were injured or killed, and whether his property was damaged and by whom.

One might think self-reports in Pashtun areas would favor the Taliban. However, the survey shows that “37 percent of respondents indicated that they had personally experienced victimization by ISAF, while a similar 33 percent reported Taliban victimization.” A surprisingly balanced result.

Finally, respondents were asked if the Taliban or ISAF attempted to redress physical harm or damage to property. ISAF victims usually received a $2,500 payment, while Taliban victims receive monthly stipend, funeral oration, “or basic staples such as Kerosene or foodstuffs.” The survey found that 16 percent of ISAF victims were approached, while the Taliban approached 60 percent of their victims.

This evidence allows Lyall and colleagues to gauge the effects of “hearts and minds” operations. They also measure the amount of development aid ISAF put into the areas where the surveys were taken.

Similar to Condra and Shapiro, Lyall and colleagues find that civilian harm caused by the ISAF increases support for the Taliban, and decreases support for the ISAF. Crucially, Lyall and colleagues part ways with Condra and Shapiro in their conclusion that “Taliban-inflicted harm does not translate into greater support for ISAF and has only a marginally negative effect on Taliban support.”

Interestingly, payments directly to victims were successful in garnering support for both sides. However, Lyall and colleagues warn that the ISAF’s sample is too small to make broad conclusions. In addition, they note that the ISAF selects payees that will most likely be swayed by the money.

Finally, economic assistance — outside of death payments — had no effect on these outcomes. This finding directly contradicts FM 3-24’s “hearts and minds” thesis.

The explanation Lyall and colleagues present for these outcomes is fairly simple. When the Pashtun-dominated Taliban inflicts harm, Pashtun civilians believe the combatants were forced to do so by the foreign invader. But when the ISAF inflicts harm on Pashtuns, the preconceived notion that foreign invaders are seeking to do harm is confirmed.

This conclusion helps us to rethink the “civilian agency” assumption. Again, the mainstream understanding of this assumption is that civilians will be swayed to the side of the combatant that kills fewer civilians and can provide the most funding for public works.

However, the findings by Lyall and colleagues show that killing civilians will not necessarily impede a combatant’s success on the battlefield as long as the killers and victims have the same ethnicity.

Lyall’s 2010 study on the Russian counterinsurgency war in Chechnya in the early 2000’s provides similar conclusions, but with a wholly different logic.

Russia’s fight against Chechen insurgents was absolutely brutal.

Rape, looting, disappearances, torture and massacres were widespread and common. Much of these abuses were carried out during Russia’s sweeping operations. These operations involved hundreds of soldiers laying siege to a village for several days while also searching for insurgents, collaborators and weapons caches.

By 2003, pro-Russian militias made up of Chechens were conducting sweeps separate from Russian forces.

Lyall’s study attempts to determine how effective these sweeps were in reducing insurgent attacks. In particular, Lyall’s study considers whether Chechen militias were more effective in reducing attacks than their Russian counterparts.

To do so, Lyall mapped hundreds of sweeping operations, mainly using newspaper reports. He was able to distinguish between sweeps conducted by Chechen militia, and those conducted by Russian-only forces. To gauge the effect of the sweeps, Lyall compared the number of attacks during the 90-day period before the sweeps to the 90-day period after the sweeps.

Lyall found that the period following Russian-only operations saw a seve per cent increase in insurgent attacks. Remarkably, Chechen-only sweeps reduced attacks by 33 percent.

What’s even more astounding about this result is that civilian killings and other abuses occurred at almost identical levels during both Russian and Chechen operations, according to Lyall.

What explains this? Lyall argues that Chechen militia are better able to identify insurgents and civilian collaborators than Russians. Unlike their Russian counterparts, the Chechen militia often knew the identities of people in the village. Most Russian forces also do not speak Chechen.

However, the Chechen militias’ familiarity with certain areas also meant that they are better able to protect civilians that provide information on insurgents.

In short, Lyall argues that co-ethnic militia are better able to protect, threaten, disappear or kill both insurgents and civilians guilty of supporting insurgents.

To back this conclusion, Lyall notes that a central tactic of the Chechen militia during the sweeps involved kidnapping the families of insurgents. The Chechens were able to use their knowledge of the community to communicate to an insurgent that surrender would keep his family from being tortured and killed. Apparently, the Russians were bad at this practice because they could not identify insurgents’ families as easily.

Unfortunately, this is not the only study that shows civilian targeting can actually be a productive strategy in counterinsurgency. At least under certain conditions.

David Axe photo
When killing works

Jason Lyall’s 2009 study of indiscriminate artillery shelling by Russia during the same war in Chechnya provides a similarly macabre conclusion. Using similar methods to his 2010 study, Lyall shows that villages pummeled with artillery saw a reduction in “the mean number of insurgent attacks relative to nonshelled villages.”

Lyall provides two explanations for this outcome. First, the artillery strikes may have displaced the population, cutting insurgents from informants and potential recruits. Second, the population may have blamed the insurgents for the carnage and punished insurgents by withholding information needed to attack the Russians.

However, like all of the other studies, Lyall doesn’t have the data to explain exactly what caused the reduction in attacks.

Trying to understand variation in insurgent attacks, territorial control and popularity are all crucial measures for evaluating any counterinsurgency strategy. However, none of these studies discussed thus far address how civilian casualties affect a combatant’s ability to recruit fighters.

Luckily, Stathis Kalyvas asks this very question in his study of German counterinsurgency during the occupation of Greece during World War II.

A good test case for this question would be a war in which the counterinsurgent accidentally killed civilians. But an even better test case would involve a counterinsurgent that intentionally massacres civilians. This is because recruitment should be impossible. At least according to FM 3-24, and popular media.

Thus, Germany’s intentional brutality toward civilians during the occupation of Greece makes it a “hard” test case for an alternative theory.

Kalyvas spent years in his home country of Greece interviewing former militia fighters, insurgents, and civilians. He also found a treasure trove of archival material in Greece that allowed him to plot civilian killings at a level of detail never seen before.

But he was only able to collect such data for the Argolid region of Greece. While limited in scope, such fine-grained data allows for a better understand of what actually occurred on the ground.

Through statistical tests Kalyvas found that civilian killing had no real effect on the German military’s ability to recruit local militia fighters. Instead, Kalyvas found that recruitment was mainly dependent on the amount of territory the Germans controlled. In fact, the greater the territorial control, the more recruits joined the German’s militia.

Ethnicity didn’t matter, and neither did political ideology. There were two main Greek political parties, one that supported the Germans and one that did not. One would think that maybe swaying anti-occupation parties to the side of the Germans would be impossible. But it wasn’t. Territorial control determined recruitment above all else.

Elders at shura, Marzak, Afghanistan, Jan. 24, 2012. David Axe photo
Leaders matter more

Early on in the movie War Machine, Brad Pitt’s character brings an Afghan army captain to a meeting with top American diplomats. The captain explains that the elections are not seen as necessary by most Afghans.

“Why are we having an election?” the captain asks. “People vote for whoever the local leaders tell them to vote for. Yeah. Because they don’t want to have their heads chopped off.”

This quote captures the underlying logic found in studies that don’t buy into the idea that civilians can choose which side to support in a war zone. These studies provide evidence that insurgent leaders make these decisions. These studies find that the most successful counterinsurgents coerce, co-opt or kill leaders of insurgent organizations.

Christopher Day and William Reno pose the most direct challenge the civilian-centric approach found in FM 3-24, and other sources. The authors note that most counterinsurgency strategy by African militaries fundamentally contradict the “hearts and minds” thesis found in FM 3-24. These states simply do not have the resources to carry out development projects, and massacres of civilians are commonplace, according to the authors.

Yet, African counterinsurgents are fairly successful. Day and Reno find that “Only about 10 percent [of rebels in Africa] have been unambiguously victorious.” Further, they note that 30 percent of Africa’s 150 civil wars ended with rebels being completely destroyed

Day and Reno’s study seeks to answer why African counterinsurgents are so successful even though they don’t implement a “hearts and minds” strategy. To answer this question, they carefully trace the history of Uganda’s counterinsurgency wars with various rebels since the late 1980s.

They find that, like other African counterinsurgents, Uganda was most fearful of insurgent movements that were headed by men that had formerly been members of the state’s political apparatus.

This type of insurgent group was often the best at recruiting soldiers and funding military operations, according to Day and Reno. But instead of trying to win over potential civilian recruits, Uganda was often able to successfully bribe the leaders into peace with power-sharing deals.

On the other hand, Uganda felt little threat from insurgent groups headed by leaders with little to no political connections to the state. For these groups, the Ugandan military simply crushed the insurgents militarily.

Civilians mattered little in either case, according to Day and Reno. At various points, both the Ugandans and insurgents brutalized the population. At other times, the armed actors were relatively restrained.

Ultimately, Uganda’s success came from the provision of patronage or “from the barrel of a gun,” to paraphrase War Machine.

Lee J.M. Seymour’s study of the civil wars in Sudan in the early 2000s also finds that patronage is a vital component in counterinsurgency. However, he is not directly testing why a counterinsurgent won or lost.

Instead, Seymour tries to understand why leaders of factions within insurgent groups often switch alliances to the counterinsurgent.

This is an important question for counterinsurgency practitioners. As the Russia found in Chechnya, and the American military found during the “Awakening” in Iraq in 2007, allying with former insurgents makes victory a whole lot easier.

To answer this question Seymour put together a dataset comprised of 92 insurgent factions in Sudan. Fascinatingly, he found that “41 of 92 factions switched sides at some point (with 23 of them switching two or more times).” Seymour also spoke with rebels, government troops and aid workers during extensive field work in Sudan.

His main conclusion is that insurgent factions chose to switch sides mainly when they were offered “weapons, ammunition, and support against local competitors.”

The ethnicity or political demands of the insurgent leader had no bearing on his decision to change alliance. Contrary to Kalyvas’ findings on the German occupation of Greece, territorial control also had no bearing on outcomes.

Civilian killing didn’t enter into the calculation to switch sides either.

For instance, Seymour notes that a leader within the insurgent group known as the Sudan People’s Liberation Army claimed that his defection was in part due to civilian abuses. However, his faction continued to practice the same level of brutality once on the government’s side.

But not all studies find that bribes are essential.

Paul Staniland’s study on insurgent groups in Kashmir and Sri Lanka finds that the actions of the counterinsurgent have only a marginal effect of the decision of the government to switch sides.

Based on field work in the two countries, Staniland finds that when infighting occurs between factions within insurgent groups, the weaker side often allies with the government. In Staniland’s view, the will to survive trumps ethnicity or political goals for insurgent leaders.

Like Seymour, and Day and Reno, Staniland finds that civilian killings — widespread among combatants in Sri Lanka and Kashmir — didn’t affect the decision to flip.

Similar to Staniland, Fotini Christia finds that power is crucial to attracting alliances.

Christia’s 2012 book Alliance Formation in Civil Wars is already a seminal work in political science. Christia conducted extensive field work in Afghanistan and Bosnia where she received unprecedented access to insurgent leaders, government officials/troops, aid workers, academics and other regional experts. She even gained a working knowledge of Farsi/Dari and Bosnian/Croatian/Serbian to conduct the field work.

Her findings boil down to this. During civil wars with more than two armed groups — think Syria — insurgent leaders decide who to ally with based on two factors. The alliance should increase the likelihood of winning. However, the alliance must also allow for the insurgent leader to gain a significant amount of political power once the war ends. If it becomes clear during the course of the war that this will not happen, the insurgent leader will switch sides.

This phenomenon holds true even if it’s clear that the alliance is about to win the war. That is, unless it becomes clear that an insurgent group has become so dominant that no alliance formation could possibly challenge it. If this occurs, insurgent groups will bandwagon to the winning side.

Insurgent leaders gauge the relative power in their coalition mainly based on battlefield performance, according to Christia. Ethnicity, religion, political goals and treatment of co-ethnic civilians, have no bearing on the decision to switch sides.

In fact, Christia shows that insurgent leaders “pick allies whose support will result in optimizing their wartime returns, and then look to their identity repertoires for characteristics shared with their allies while not shared with their foes.” This propaganda is unimportant to insurgent leaders, but “serve as important signals to the rank and file.”

The most compelling illustration of her argument occurred during the 1992-1998 Afghan intra-Mujahedeen civil war. She identifies the four main insurgent groups based on ethnic affiliation, Pashtuns, Tajiks, Uzbeks and Hazaras.

What may come as a surprise to the popular American view of ethnicity in Afghanistan, “there were seven instances of alliance shifts” during the war.

Tajiks and Pashtuns never allied during the war. However, the Uzbeks and Hazaras fought on both sides multiple times. Sometimes as allies. Sometimes as foes. In fact, between 1995 and 1996 two separate Hazara factions fought on opposing sides.

Each shift brought a new propaganda narrative explaining the new alliance. The most damning shift for the “hearts and minds” thesis came in 1998 when the Hazara switched to the side of the Pashtuns. The latter organization was known as the Taliban by that point in the war.

What’s so damning about this shift is that the Taliban massacred thousands of Hazara civilians just a few months prior to the alliance. The leader of the Hazaras, Mohammed Akbari, told Christia that he justified this alliance to his people based on historical economic ties between Pashtun landlords and Hazara peasants.

In reality, however, the Hazaras were a part of a mass bandwagoning of leaders from all ethnic insurgent groups to the side of the Taliban once it became clear that they were going to win.

No comments:

Post a Comment