Scholars of the micro-dynamics of armed conflict largely agree that collateral damage—the unintentional killing or harming of civilian noncombatants—is not only morally reprehensible but often strategically costly for combatants in war. The logic behind this is straightforward: civilian communities can join or aid multiple different parties in armed conflicts, and killing or harming them alienates them from the perpetrator and pushes them toward its rivals or opponents. This idea is now buttressed by a variety of detailed empirical studies from different conflicts, including Vietnam, the Palestinian territories, Iraq, and Afghanistan.Footnote 1 As summarized in one such effort, the empirical record demonstrates that “both sides pay a cost for causing collateral damage.”Footnote 2
However, this dynamic is of little surprise to most combatants, who already understand it well and undertake a number of actions to mitigate these costs. Chief among these is the practice of giving “condolence payments”—that is, providing material compensation to civilians for damage inflicted. This tactic is widespread. In the case of the US, it dates to World War I and has been used in every major intervention since then.Footnote 3 Today, the scale of these efforts is substantial: a report by the Government Accountability Office (GAO) shows that the US military distributed more than USD 20 million in these payments in Iraq in 2005 alone.Footnote 4 Meanwhile, a number of other countries such as Iraq, Afghanistan, Pakistan, Colombia, and Israel have also provided compensation to wartime victims,Footnote 5 as have prominent militant groups such as the Afghan Taliban and even branches of Al Qaeda. The use of material compensation to mitigate the impact of collateral damage is a common feature of modern war.
Do these payments work? This question has significant implications not only for international security policy, but also our theoretical understanding of how civilians react to material incentives after they experience harm. I examine these dynamics during the heart of the Iraq War, which pitted the US-led Coalition and the Iraqi national government against Sunni Arab insurgents (and, at times, Shi'a Arab militias) seeking territorial and political control. Specifically, I use micro-level conflict event data available via the Empirical Study of Conflict Iraq War database (ESOC-I).Footnote 6 This allows me to combine detailed data on 193,264 Iraqi insurgent attacks with 19,961 incidents of collateral damage and 4,046 Coalition condolence payments in Iraq from 2004 to 2008. Using these data, I build on previous studies testing the impact of collateral damage on insurgent attacksFootnote 7 and analyze how post-harm mitigation shapes wartime violence.
This analysis reveals that Coalition condolence payments do reduce local insurgent violence, and their impact is substantial. Calculations based on the model suggest that a USD 1,000 to 2,000 boost in condolence payment spending in a given district yields one fewer insurgent attack in that district over a six-month period. Moreover, this effect is consistent across different types of compensation, with both sustainable “in-kind” assistance programs and pure cash transfers diminishing insurgent attacks (with no statistically significant difference between them). These patterns are consistent with rationalist models of armed conflict in which civilians accurately recognize and respond to conflict events in ways that maximize their odds of survival.Footnote 8 In this framework, post-harm compensation acts as a “costly signal” of selective violence and unintentional harm, leading civilians to rationally update their beliefs about the relevant event. In so doing, it shifts their judgments about the perpetrator's “type,”Footnote 9 disrupting the process of alienation that would otherwise facilitate more insurgent violence.
These results have two primary sets of implications. From a policy perspective, they provide evidence that there is not only a normative justification for combatants to compensate victims but a strategic one as well. The study reveals that such compensation is among the most successful tactics in the commander's tool kit, and that it should be prioritized after any collateral damage event. In this sense, it meshes nicely with a recent paper that finds similar results in Afghanistan.Footnote 10 Together, these studies provide firm evidence across conflicts and programs that post-harm compensation is effective at mitigating violence after collateral damage. From a theoretical perspective, the findings add another layer to the growing debate over civilian populations in war zones. Specifically, they indicate that—even with the strong grievances that arise after civilian casualties—relatively subtle signals of intent by out-group actors elicit greater civilian collaboration. In other words, the results suggest that civilians are not blinded to countervailing signals by biases or grievances, and that—while their initial or baseline attitudes are not unimportant—they can and do update in the face of new and relevant pieces of information about combatant behavior.
Empirical Context
I examine the impact of post-harm civilian compensation during the Iraq War from 2004 to 2008. The conflict, emerging soon after the 2003 invasion, pitted the US-led Coalition forces and Iraqi national government against Sunni Arab insurgents seeking control of the country, with Shi'a Arab and Kurdish militias also struggling for power. The Iraqi case is useful for us because of the unique availability of detailed, disaggregated event data on 4,046 condolence payments by one of the key combatants—the Coalition—across space and time, plus data on collateral damage and insurgent violence with which they can be combined.
Our dependent variable is the number of insurgent attacks as captured by the military's “significant activity” (SIGACT) database. SIGACT includes information about the location, date, time, and type of attack for 193,264 insurgent attacks against Coalition forces, the Iraqi national government, Iraqi Security Forces, and civilians from February 2004 through December 2009. The data do not contain Coalition raids and operations during which no insurgents returned fire, so they measure insurgent-initiated attacks or firefights with insurgent forces. Moreover, they most likely undercount attacks against civilians and other kinds of “soft targets” when Coalition forces are absent.Footnote 11 However, since our main interest is in insurgent activity directed against Coalition troops, these are second-order concerns.
The primary independent variable is Coalition condolence payments. Data for these are from the US Army's Iraq Reconstruction Management System. The data contain the start and end dates, project type, funding organization, and cost of a wide range of reconstruction projects from March 2003 through December 2008 in Iraq by the Coalition. Reconstruction was first dominated by the USD 18.4 billion Iraq Relief and Reconstruction Fund, an all-purpose fund focusing on large-scale infrastructure and security projects early in the conflict. As these funds diminished, however, they were increasingly replaced by small community-level programs run by the State Department or the Department of Defense (DoD). Chief among these are the United States Agency for International Development's (USAID) Economic Support Fund, which focuses on economic and social welfare initiatives, and DoD's Commander's Emergency Response Program (CERP), which allows local military officers to implement a wide range of small-scale initiatives within their areas, from building schools or roads to paying local militias or civilians.
Crucially for our analysis, the data contain information on the two key types of post-harm compensation that the Coalition dispensed. First, they include data on 2,945 condolence payments, or compensation for injury, death, or property damage by Coalition forces. These payments were disbursed by brigade-level military officers as part of CERP and are the main type of post-harm compensation by the military in Iraq.Footnote 12 Second, the data contain information on 1,101 Marla Ruzicka Iraq War Victims Fund payments. Dispensed by USAID rather than DoD, Ruzicka payments provide civilians “in-kind” vocational training and livelihood assistance rather than cash after suffering harm by Coalition forces. This follows a line of thinking that such tailored assistance programs help civilians more than simple cash transfers to victims or their families.Footnote 13
While the DoD condolence payments and USAID Ruzicka payments were not the only avenues to compensate civilians after harm in Iraq, they were the most important. A 2007 GAO report on post-harm compensation examined all of the programs used for such purposes in Iraq.Footnote 14 CERP condolence payments were the most prominent, with roughly USD 21.5 million spent on such payments in fiscal year (FY) 2005 alone and another USD 7.3 million in FY2006. Meanwhile, over USD 17.8 million was spent on the Ruzicka payments from 2005 through mid 2007, suggesting a broadly similar yearly outlay. Two other programs were used in Iraq. One (the State Department's claims program) was modest, with outlays in the thousands per year.Footnote 15 The other (the Foreign Claims Act or FCA) was substantial, but became overshadowed by the more flexible CERP and Ruzicka programs as it was legally circumscribed to events that occurred outside “combat”—which has been interpreted quite broadly in Iraq and Afghanistan.Footnote 16 In sum, we can be reasonably confident that the data used cover the two most central types of post-harm compensation used in the war.Footnote 17
Of course, to investigate the impact of post-harm compensation on insurgent attacks, we also have to include information on the harm that precipitated it to begin with. To that end, I use data on collateral damage incidents from the Iraq Body Count (IBC), an Iraqi NGO dedicated to tracking Iraqi civilian casualties during the war through international and local media reporting plus hospital records, morgue figures, and other sources of data.Footnote 18 These data include the date, location, actors, and tactics of 19,961 incidents of collateral damage, accounting for 59,245 total civilian casualties from March 2003 through June 2009. These data were cleaned by Condra and Shapiro and attributed using event descriptions to Coalition, insurgent, sectarian, or unknown forces.Footnote 19 Like the insurgent violence and civilian compensation data, the incidents are geo-located by district. Validity checks suggest that unknown killings and imprecise locations (about 10% of the sample) are not merely a function of reporting biases driven by violence levels.Footnote 20
Descriptive Analysis
Figure 1 plots the weekly numbers of insurgent attacks (via SIGACT), collateral damage incidents by Coalition forces, condolence payments by Coalition forces, and Marla Ruzicka payments by Coalition forces across Iraq from 2004 through 2008. The level of insurgent attacks trended steadily upward as the conflict grew from 2004 through 2007, spiking after the bombing of the Golden Mosque in Samarra in early 2006 by Al Qaeda in Iraq (AQI). Finally, violence began to fall off in mid-2007 as the mix of the “surge” (the deployment of 30,000 more Coalition forces and change in their use) and the “Anbar Awakening” (the realignment of Sunni Arab tribes and nationalist insurgents against AQI) kicked in and started to pacify the situation.Footnote 21 The figure thus effectively highlights the broad ebbs and flows of the conflict during this period. Meanwhile, the top-right quadrant presents the weekly plot of collateral damage incidents by Coalition troops from 2004 to 2008. The amount of Coalition collateral damage does not mirror the overall trend in insurgent violence. Instead, it is relatively stable over time, with slight peaks in late 2004 (during the battles for Fallujah) and early 2008 (during the middle of the surge).
Figure 1. Insurgent violence, coalition collateral damage, and post-harm compensation over time in Iraq, 2004–08
Finally, the bottom two quadrants show the amount spent per week (in thousands of dollars) on condolence payments and Ruzicka payments by the Coalition to Iraqi civilians over time. Looking first at condolence payments, we can see that there was little compensation in the first year while the program was still in its infancy,Footnote 22 followed by a large spike in early 2005 (after the Second Battle of Fallujah) after which the series smooths out and becomes reasonably constant. Turning to the Ruzicka payments, the series is somewhat lumpier, with sharp peaks in early 2005, 2006, and especially 2007 after which point it flattens out substantially. In fact, the plot ceases to fluctuate in mid-2007 and remains flat for over a year.Footnote 23 Overall, these two plots indicate that the USAID-funded Ruzicka payments and DoD-funded condolence payments were quite independent, drawing from different organizational budgets and constraints. Both series show little apparent relationship with the macro-level trends in insurgent violence.
Empirical Strategy
I now turn toward the task of assessing how compensation affects insurgent violence. The study runs from the start of 2004 through the end of 2008, covering all 104 Iraqi districts. The unit of analysis is the district-half year, which yields 1,040 (10*104) total observations.Footnote 24 Using half-years affords ample time to capture the full claims process and the subsequent effects of compensation, while also retaining a substantial number of observations for our analysis. Both insurgent violence and compensation are divided by district population (or a transformation thereof) to create SIGACTs per 1,000 residents and spending per capita.Footnote 25 The logic of this is that the population of each district is likely linked to the violence perpetrated (and compensation received) within it in ways unrelated to the analysis, such as due to sheer scale.Footnote 26
The Data-Generating Process
To estimate the causal impact of condolence payments on violence, we need a plausibly exogenous source of variation in payments. Perhaps the primary threat to such exogeneity would be a “strategic selection story” in which post-harm compensation was a product of local Coalition troop quality. In this story, wiser Coalition units (and leaders) would do a better job of providing compensation to civilians they harmed in their operations, and of winning over the local civilian populace and prosecuting the war in other ways. This would make it difficult to tell what was causing better outcomes in higher-compensation areas: the compensation or the forces that gave it.
Yet, the key with this selection story (and most others) is that the provision of compensation hinges on the intentional decision making of local Coalition troops (and leaders), and not on forces outside of their control. In fact, identifying the impact of condolence payments is greatly aided by the fact that the mechanisms by which they were made—and the amounts provided—were shaped by a variety of idiosyncratic financial, bureaucratic, and geographic constraints largely beyond the control of the soldiers on the streets and their local commanders.
First, the provision of post-harm compensation was deeply shaped by local fiscal pressures. For example, the chance of condolence payments was shaped by not only the allocation of CERP funds across districts but also whether those funds happened to be available when a claim is made. CERP funds can be spent in many areas, from building hospitals to hiring security guards. Moreover, a survey of “reconstruction leaders” in Iraq found that while condolence payments were seen as an effective use of CERP, others like agriculture, water and sanitation, rule of law, and contract security were prioritized more.Footnote 27 This means the odds of compensation at any given time were shaped by the degree to which funds were “left over” from other higher priorities and their (idiosyncratic) unit-level payment schedules. For example, one ex-claims officer in Iraq recalls an especially deserving victim who “did not receive compensation after his daughter died from a cluster munition because funds for condolence payments were unavailable when he visited the [Baghdad] convention center.”Footnote 28 Choices about whether to pay thus often “turn[ed] on little more than the availability of funds” Footnote 29 when a claimant walked in the door, rather the strategic thinking or understanding of local forces.
Second, the provision of post-harm compensation was shaped by bureaucratic factors like the inclination of the claims officer attached to the unit when a claim was filed.Footnote 30 CERP-based condolence payments were typically under the authority of a brigade-level (or higher) military commander, but in practice were approved by a Judge Advocate (JAG)—a military lawyer—with vast discretion to accept or reject them. A review of hundreds of claims obtained via an ACLU Freedom of Information Act (FOIA) request found that the process was plagued with inconsistencies, discrepancies, and ambiguities that “invest massive discretionary authority in US claims personnel.”Footnote 31 In other words, differences between JAGs were a key factor in the chances of a claim. In the words of Jonathon Tracy, an ex-JAG and claims officer himself: “I know plenty of lawyers who did not pay any condolence payments at all … There was no reason for it. It was clearly not combat, and the victim was clearly innocent, all the facts are there, witness statements, but they wouldn't pay them.”Footnote 32 While this opens up the prospect that political or strategic factors may shape these personnel, they will not affect them all equally—some (like Tracy) may base their choices on the merits of the case, while others may make strategic calculations or award nothing at all. In this way, variation in the preferences of the bureaucratic personnel in charge of the case—as distinct from the local commander or the quality of the area's “troops”—play a key role.Footnote 33
Third, compensation also hinged on the proximity of the incident to the nearest Coalition center that processes civilian claims. In fact, most claims were initiated by civilians who brought them to the attention of local Coalition forces at Civil Military Operation Centers (CMOCs), but these could be miles away depending on where the incident occurred. For example, in Fallujah, the major CMOC after the invasion was a center run by the Marines known as the Fallujah Liaison Team (FLT). Civilians seeking compensation in Fallujah had to visit the FLT to submit claims, regardless of whether the incident occurred ten blocks or ten miles away. This is key because studies have shown that the proximity of local institutions like police stations shapes public willingness to come forward, report abuses, and seek assistance.Footnote 34 This is likely exacerbated in contexts like wartime Iraq, in which travel via road is notoriously risky and dangerous.Footnote 35 Thus, the provision of compensation is likely to be shaped by where an incident happened to occur relative to a claims center, independent of the local troops’ or leaders’ thinking. In sum, qualitative information on the data-generating process suggests that the provision of compensation is strongly shaped by several factors quite divorced from strategic decision making or at odds with it. This helps ease concerns about the most problematic types of selection bias driving any results.
Does a quantitative look tell the same story? To check, I estimate models with post-harm compensation as the dependent variable. This allows me to analyze what predicts how much compensation an area gets. Specifically, I run two models, one predicting condolence spending per district*half year and the other predicting Ruzicka spending per capita per district*half year. I include in these models the following variables: lagged insurgent violence (SIGACTs), civilians killed by both the Coalition and the insurgents, other competing types of reconstruction spending from the same sources (that is, noncondolence small CERP spending and non-Ruzicka USAID spending), Coalition troop strength, population density, percent urban, US PRT (Provincial Reconstruction Team, a measure of local Coalition expertise) presence in each district, and original data gathered for this study on CMOC presence in each district.Footnote 36 If compensation were driven by local counterinsurgent quality, we would expect it to be positively linked to other forms of small reconstruction spending as well as PRT presence. Meanwhile, we can also check whether it is affected by other forms of selection bias, such as a patronage logic of rewarding less violent areas (fewer previous SIGACTs). As in the analysis that follows, all models are estimated in first differences, using OLS with district-level clustering and fixed effects.Footnote 37
Table 1 shows the results of the analysis. The payments are not going to areas that spend more on CERP generally or have a PRT presence. This is not consistent with the idea that they are driven by local counterinsurgent quality, since small and targeted reconstruction spending along with local civilian expertise have both been shown to be key parts of counterinsurgent gains in Iraq.Footnote 38 Similarly, they do not go to areas with more previous insurgent attacks (either toward the Coalition or Iraqi civilians), belying the notion that they are a reward to friendly or calm areas for good behavior.
Table 1. The predictors of post-harm civilian compensation in Iraq, 2004–08
Notes: Results from first-differenced OLS regressions with clustering by district. Compensation and other spending is per capita while insurgent violence is per 1,000 residents, both over six months. Standard errors in parentheses. *p < .10; **p < .05; ***p < .01.
In contrast, we can see that some of the variables do predict compensation. Other USAID spending negatively and significantly predicts Ruzicka spending, while other small CERP spending almost negatively and significantly predicts condolence spending (p = 0.11). This lends support to the idea discussed earlier—that payment is often made only when there is funding “left over” from other higher organizational priorities, and that budgetary issues unrelated to the incentive to pay claims act as a key constraint on compensation.
Moreover, we can see that the level of condolence spending is positively and significantly predicted by CMOCs, troop strength, and percentage urban. These variables can be understood as capturing the accessibility of compensation to people—after a collateral damage event, it is easier to seek payment in areas with more Coalition forces or in urban centers with more bases (and CMOCs). This is consistent with the notion that variation in the geographic circumstances of civilians influences their ability to seek compensation. While we do not find this manifest in exactly the same way in the Ruzicka model, we do see that percentage urban and population density increase compensation, suggesting a similar dynamic at work (the lack of significant effects for troop strength and CMOCs here are unsurprising, since the payments are given by USAID and not the military). This analysis of the data-generating process provides little evidence that compensation is especially strategic in nature. Instead, it suggests that, conditional on collateral damage, compensation is shaped in key ways by factors like funding availability and geographic accessibility, further mitigating concerns about a strategic selection process confounding any results.
The Model
To estimate the effect of compensation on insurgent violence, I use a panel regression strategy with first differencing and fixed effects. First differences allow us to focus only on changes in insurgent activity from one half year to the next within each district, and ensures that the results are not driven by cross-sectional differences in insurgent violence or underlying political loyalties between districts. Fixed effects function as a series of district-specific time trends that allows us to account for the divergent trajectory of each district throughout the course of the war.Footnote 39 Finally, I include a number of key covariates representing conflict dynamics that vary over time. Thus, the base model is:
$$Y_{i\comma t}-Y_{i\comma t-1} = \alpha \lpar {c_{i\comma t}-c_{i\comma t-1}} \rpar + \beta \lpar {d_{i\comma t}-d_{i\comma t-1}} \rpar + \gamma \lpar {e_{i\comma t}-e_{i\comma t-1}} \rpar + G_y + H_i + I_{g\comma y} + \mu _{i\comma t}$$where Y i,t is the level of insurgent attacks in district i at time t, c i,t is the spending on condolence payments in district i at time t, and d i,t is the spending on Ruzicka payments in district i at time t. Meanwhile, e i,t is a vector of other time-varying conflict dynamics including the amount of Coalition collateral damage, insurgent collateral damage, Coalition troops, other (noncondolence) small-scale CERP spending, other (non-Ruzicka) USAID spending, CMOC presence, and PRT presence in district i at time t. Finally, G y is a set of half-year fixed effects, H i is a set of district fixed effects, and I g,y is a set of interactions between governorate-level vote shares for Sunni Arab parties in the 2005 parliamentary elections and half years meant to pick up broad sectarian shifts such as the Sunni Awakening. Table A2 in the appendix reports a correlation matrix for the variables in the base model. Models are estimated with OLS unless otherwise specified.Footnote 40
Base Results
Table 1 shows the base results. The first model includes only the post-harm compensation, the second adds Coalition and insurgent collateral damage, the third adds other small CERP and USAID spending, and the fourth adds troop strength, CMOC presence, and PRT presence. As is clear, a naive model with only the two types of post-harm compensation suggests that they have a limited or mixed impact on rates of insurgent violence: condolence spending has no impact, while Ruzicka spending does. Yet once we add Coalition and insurgent collateral damage, another picture emerges. With collateral damage included (columns 2 to 4), both types of compensation significantly reduce insurgent violence. The coefficient on condolence spending is now significant, and the Ruzicka coefficient remains significant but is much larger. The shift occurs because, as noted, post-harm compensation is dispensed after collateral damage incidents, which are known to amplify insurgent attacks.Footnote 41 When we do not control for this fact, it looks like condolence payments have a limited effect on insurgent attacks. However, once we have done so, this is no longer the case; they have a clear violence-reducing effect. Thus, to identify the effect of post-harm compensation on insurgent violence, it is essential to control for the harm that prompts it in the first place.
We can also compare the effects of the two types of post-harm compensation. As noted before, some practitioners argue that sustainable and tailored assistance programs such as the Ruzicka Fund are more effective in undoing the effects of collateral damage than simple one-off cash transfers.Footnote 42 Yet, while the coefficient on Ruzicka payments is substantially larger than the one on conventional condolence payments in the three models, a Wald test on the equality of the two shows that this difference is not statistically significant at conventional levels (for example, p = 0.37 in model 3, the preferred specification). This suggests that both types of payments have similar effects—post-harm civilian compensation is effective in depressing insurgent violence regardless of how it is provided.
Turning to the covariates, we see that Coalition collateral damage substantially increases insurgent attacks,Footnote 43 although insurgent collateral damage does not have the opposite impact here. Meanwhile, other small CERP spending does not have a significant influence on insurgent violence in our model. And finally, the effect of Coalition troop strength is nearly significant in predicting more attacks as well (p = 0.12), which may be a result of heavier troop concentrations going toward more violent areas, giving insurgents more targets to attack, or simply observing and reporting more of the incidents (especially against Iraqis) that occur nearby.
Table 2. The effect of post-harm civilian compensation on SIGACTs per half year
Notes: Results from first-differenced OLS regressions with clustering by district. Civilian compensation is per capita while insurgent violence is per 1,000 residents, both over six months. Standard errors in parentheses. *p < .10; **p < .05; ***p < 0.01.
We can also estimate the substantive impact of the compensation from this model. The third column of Table 1 tells us that, within a given district, another dollar of condolence spending per capita yields 0.52 fewer insurgent attacks per 1,000 residents over a six-month period. In more concrete terms, this shows that, in an average district, a USD 1,950 increase in condolence spending will stop one more insurgent attack over a six-month period.Footnote 44 The same calculations imply that one additional Coalition collateral damage incident leads to 8.3 more insurgent attacks over a six-month period. This means that the cost of offsetting the typical Coalition collateral damage incident is about USD 16,185 ($1,950*8.3) in condolence payment spending. While we would urge caution in deploying this specific point estimate without care, the results do clearly suggest that the impact of post-harm compensation is significant in substantive as well as statistical terms.Footnote 45
Robustness Checks
To boost confidence in these results, I conduct a wide range of robustness checks. First, one key test to probe for endogeneity or omitted variables in time series or panel data is to lag the dependent variable on the LHS of the equation and analyze its relationship with the primary explanatory variable(s) on the RHS. If the condolence and Ruzicka payments predict (lower) past insurgent violence, this would suggest that the relationship may be reversed or entirely spurious. Yet the results of this test (see Appendix Table A3) show that this concern is unfounded: neither type of post-harm civilian compensation predicts prior insurgent violence. Second, a related test to help guard against these issues is to flip the model and analyze whether the lead of insurgent violence predicts present civilian compensation. This would also suggest a potentially problematic relationship because condolence payments might be sent to areas expected to be cooperative and peaceful in the future—in other words, there could be an “anticipation bias.”Footnote 46 However, the results of this test (see again Appendix Table A3) also show no evidence that this is the case: future insurgent attacks do not predict present condolence payments. Together, these tests suggest that condolence payments are not driven by past or future (anticipated) insurgent violence, bolstering our confidence in the causal arrow in our relationship.
In addition to these tests, I include a rich set of additional covariates to ensure that the results are robust to other relevant factors. First, I add other major types of Coalition spending to the equation, including spending on (1) small (less than USD 50,000) non-CERP reconstruction projects, and (2) large (more than USD 50,000) non-CERP reconstruction projects. This ensures that the effect of condolence payments is not proxying for other categories of spending in a given district. Second, I include lagged measures of the conflict dynamics—that is, insurgent violence, collateral damage by both sides, and Coalition troop strength—into the model to account for any short-term trends in the fighting that might influence local patterns of violence and compensation. Third, I add the unemployment rate, population density, and urbanization percentage in each district to control for socioeconomic features of districts that could shape their propensity for violence and demand for compensation. Fourth, I include sectarian and unknown civilian casualties in the model to account for the possibility that other types of victimization boost compensation and insurgent attacks. Across all these tests, the core results that both types of compensation significantly reduce insurgent attacks are unchanged (see Appendix Table A4). Finally, I also replicate the base models without district fixed effects to ensure their inclusion is not critical to our findings. The results (see Appendix Table A5) show that my findings are substantively unchanged without district fixed effects. Overall, the robustness of our primary results to this array of additional covariates and tests helps substantially boost our confidence that post-harm compensation reduces insurgent attacks.Footnote 47
Conclusion
Despite efforts to avoid it, collateral damage is inevitable in war. Whether an errant artillery shell, a botched checkpoint stop, or a calculated choice by a commander weighing different types of risk, even the best-organized and best-trained armies will kill some civilians in any prolonged military engagement. From a moral perspective, it is widely recognized today that combatants have an obligation to provide compensation to these victims.Footnote 48 Yet, with biases toward out-group populations and alternative ways to allocate scarce wartime resources (i.e., to the fighting), they often neglect to do so. This study reveals that such neglect is strategically myopic. There is a convincing strategic rationale for combatants to distribute post-harm compensation to civilian victims in war zones because doing so significantly undercuts enemy attacks and attempts to exploit their battlefield mistakes. Specifically, I found that the compensation provided by the Coalition to its civilian victims in Iraq from 2004 to 2008 substantially reduced ensuing insurgent violence in affected areas. These results yield important policy implications, giving combatants a strong strategic incentive to compensate civilians they harm in war. In particular, they provide a firm basis for recommending the expansion of programs along the lines of CERP condolence payments and the USAID Ruzicka payments in any future population-centric conflicts that may arise.
The findings also contribute to the growing scholarly debate about what shapes the perceptions and reactions of civilians in war zones. The emerging literature on the micro-dynamics of armed conflict has produced at least two main rival schools of thought about civilian populations in war zones. In one, they are best approximated as “rational peasants” who accurately recognize and react to conflict events in ways that increase their odds of survival.Footnote 49 In another school of thought, they act more like “ethnic partisans” whose attitudes and behaviors are deeply shaped by group identities and who are heavily biased against out-group actors.Footnote 50 The debate largely hinges on the question of whether civilians rationally update in the face of new information in conflicts or interpret it in ways consistent with their existing beliefs.
While no one study can resolve this debate, in this research note I aim to contribute to the conversation. By systematically examining the strategic effects of post-harm compensation by Coalition troops in the Iraq War from 2004 to 2008, we can see whether civilians react to combatants’ signals about their intentions after battlefield mistakes in ways that affect their conflict behavior. Overall, the patterns in the data are best explained by a rationalist account of civilian reactions in which the compensation serves as a costly signal of the unintentionality of harm by the user of force, shifting civilians’ expectations about the chance of future threats to their physical security from the actor. This explanation best accounts for the fact that the compensation is effective in this case even though it is provided in the face of strong grievances (i.e., after collateral damage incidents) and comes from a foreign occupier (i.e., the “away team”). In sum, the study suggests that civilians can and do rationally update in the face of perpetrator signals like condolence payments.
These findings also dovetail nicely with a recent study in the context of Afghanistan. In particular, Lyall examines the impact of the Afghan Civilian Assistance Program (ACAP)—a USAID program to compensate Afghan civilians harmed in the war—on Taliban insurgent violence.Footnote 51 Mirroring my results, he finds that the program is effective in diminishing insurgent violence over substantial time periods at the local level. In this sense, the studies together provide a strong basis for concluding that post-harm mitigation is effective at reducing local violence in asymmetric war. For readers comparing the two, it is worth noting that Lyall's study examines ACAP, which is a tailored in-kind assistance program extremely similar in design to the Marla Ruzicka program. While some of the effect sizes in my study are larger for the Ruzicka as opposed to the conventional condolence payments, I find that both types are broadly effective. My results thus suggest that the general signal rather than the specific form it takes is what drives this pacifying effect. Future research can build on both studies and examine how this and other aspects of program design shape the consequences of post-harm mitigation.Footnote 52
There are some key limitations to the study that suggest opportunities for future research. First, future studies should explore whether civilian compensation by other actors and in other settings has similar consequences to those found here. Lyall's recent study is a good start, showing similar effects from compensation by international forces in Afghanistan as those found here in Iraq. However, scholars should also look for ways to examine whether compensation by nonstate groups—from Palestinian militants to African rebels to Latin American cartels—has a similar impact. Additionally, the effect of offering material compensation by combatants can be compared to the other ways in which they signal their intentions after mistakes, such as issuing apologies or punishing offenders.
Second, to gain empirical leverage on the question, I focused on the reactions of the civilian communities (districts in Iraq) directly targeted by the combatant signals. That said, civilians who do not learn about these events through personal experience or local networks—for example, those who live in Baghdad and observe events unfolding in Basra, Mosul, or Erbil—may interpret them very differently. If the rationalist mechanism posited here is indeed at work, and we see the observed behavior because civilians are updating their beliefs about the local chances of harm from different combatants, those who do not live in these areas would not have an incentive to alter their beliefs or behaviors. While this may be less relevant in the case of condolence payments, which are not particularly likely to be communicated to a mass audience, many conflict events like air strikes, terrorist attacks, and other forms of violence are broadcast to millions of civilians in the conflict zone (and beyond). This means that the reactions of this wider audience may differ in key ways from those living directly in the “line of fire.” Examining the reactions of these two different types of civilians to new information is a promising path for scholars who wish to deeply understand the nature of civilians’ beliefs and behaviors in conflict environments.
Data Availability Statement
Replication files for this research note may be found at <https://doi.org/10.7910/DVN/GBATE9>.
Supplementary Material
Supplementary material for this research note is available at <https://doi.org/10.1017/S0020818320000193>.
Acknowledgments
Many thanks to Christopher Gelpi, Richard Herrmann, Jan Pierskalla, Jacob Shapiro, Bradley Holland, Peter Mansoor, Daniel Kent, Kyle Larson, and two anonymous reviewers for their thoughtful feedback and advice on earlier versions of this paper.
