Loot, Sponsorship, and Violence

Scholars of social movements and opposition groups have long focused on how organizations mobilize resources and support for their activities.Footnote ¹² Collier and Hoeffler brought the focus on resources and financing to the fore in the quantitative analysis of civil war by arguing that, regardless of underlying grievances, rebellions cannot survive without adequate funding.Footnote ¹³ During the Cold War opposition groups could regularly rely on one or the other superpower by adopting a Marxist or prodemocracy platform. The end of the Cold War had severe consequences for many insurgencies.Footnote ¹⁴ Although many conflicts fizzled out or terminated as Cold War sponsors cut support to their clients, other groups turned to alternative funding sources to maintain their campaign against the regime. Groups emerging since the end of the Cold War have relied heavily on illicit trade in gems and drug smuggling to fund their insurgency.Footnote ¹⁵ The Revolutionary United Front in Sierra Leone, for example, controlled diamond-mining operations and the Taliban in Afghanistan is known for its control over the opium trade. When sold on the black market, these resources provide fungible sources of revenue that sustain the organization through the purchase of arms, food, and payments to fighters. Other organizations are not “blessed” with easily extractable commodities. The Sandinistas in Nicaragua and leftist insurgents in Guatemala, for instance, relied almost exclusively on popular support to maintain the insurgency.

Weinstein draws on resource mobilization theories in arguing that rebel groups differ in their initial endowments.Footnote ¹⁶ The presence of lootable resources at the early stages of a conflict influences the organizational structure of the movement as well as the quality of recruits that the group attracts. All else equal, those rebel organizations with easy access to funding—particularly natural resources—often attract opportunistic fighters who have little commitment to the cause. By contrast, rebel organizations with poor resource endowments will appeal to individuals who are most committed to the organization's political goals. These groups must foster closer ties with their constituents because they depend on their goodwill for shelter, food, and funding. The Tigrayan People's Liberation Front (TPLF) in Ethiopia, for example, garnered support from local populations though providing selective benefits, such as land reform, parallel systems of governance, and security from government forces in the areas in which it exercised control.Footnote ¹⁷ In Mexico, the Zapatistas held rallies and public relations campaigns designed to persuade rural, indigenous populations—and international audiences—of the justness of their cause.Footnote ¹⁸ Recent research, as well as a wealth of anecdotal evidence, suggests that reliance on the population for support tends to foster a more benevolent relationship between insurgents and civilians. Although these groups may still at times target civilians, we expect that the level of such abuse is likely to be relatively low compared with groups with alternative sources of funding.

Whereas the capture of gemstones, narcotics, and other natural resources has gained much attention in recent years, foreign governments remain an equally (if not more) important source of financing for rebel organizations. Funding from the United States and Soviet Union was relatively easy to obtain during the Cold War, but other countries (for example, Iran, Libya, Sudan, Thailand) have also provided extensive material support to rebel organizations. Data collected by Cunningham, Gleditsch, and Salehyan reveal that a majority of all rebel groups active since 1945 either have the explicit support of a foreign power, or are alleged to have foreign ties.Footnote ¹⁹ Foreign financing can significantly and quickly augment an organization's capabilities. It may also reduce incentives to forge close bonds with civilian populations or potentially erode existing relations. After all, a strategy of ideological persuasion and service provision is costly, and rebels with state patronage can devote time and effort to other pursuits.Footnote ²⁰

Violence against civilians is unfortunately a common feature of civil war. Conflict environments offer numerous opportunities and incentives for rebels to exploit or harm civilians.Footnote ²¹ Fighters may employ violence strategically, targeting suspected enemy collaborators or sympathetic populations. They may also be tempted to engage in extortionate behavior that produces short-term gains, even if it is counterproductive in the long run. Conflicts create incentives and opportunities for sexual exploitation and violence as well as the redress of personal grievances and acts of revenge.Footnote ²² Furthermore, undisciplined rebel fighters may provoke disputes with local populations and engage in violence.Footnote ²³ Research in social psychology demonstrates that when otherwise decent people are placed in stressful environments with positions of power and opportunities for coercive control, the potential for abuse is high.Footnote ²⁴

Consequently, research on civilian targeting should examine not only the opportunities and incentives for such behavior but also the potential constraints that impede it. Some military organizations work to suppress bad behavior, instill discipline, and punish transgressions; however, others are either indifferent to abuse or incapable of constraining it. In the context of a counterinsurgency campaign, US and North Atlantic Treaty Organization (NATO) forces in Afghanistan placed a high premium on avoiding civilian casualties through error and “collateral damage.”Footnote ²⁵ Rebel organizations with strong incentives to foster close collaboration with civilians similarly place greater emphasis on avoiding egregious behavior, seek to recruit capable, committed soldiers, and work to punish those actors who abuse civilians. All else equal, those groups with access to resources provided by foreign sponsors are more likely to attract opportunistic thugs who are inclined to prey on civilians.Footnote ²⁶ Easy access to resources encourages the rapid expansion of the group, which can complicate leadership's ability to screen recruits, and can result in newer, more mercenary members crowding out activists.Footnote ²⁷

We take these arguments as a starting point, but sponsorship is not as simple as a state providing resources to a foreign rebellion. Rather, sponsors expect returns on their “investment,” and they often attempt to influence the actions of their clients. Governments can, of course, attack foreign rivals with their own military forces. However, direct military confrontation is potentially costly. The state must pay the direct expense generated by the conflict, and leaders often suffer audience costs at home and abroad stemming from unpopular wars.Footnote ²⁸ In order to spare themselves these costs, state leaders may delegate the conflict to agents by providing military and monetary support to insurgent groups challenging the target state.Footnote ²⁹ But states do not give funds without conditions attached. Rather, they expect to gain some agenda control over the movement they sponsor, shaping their strategies and goals. Some rebel groups are almost entirely the creations of foreign powers. For example, the Resistência Nacional Moçambicana (RENAMO) in Mozambique came into being largely as a result of Rhodesian and South African aid as these states sought to unseat the leftist government.Footnote ³⁰ Other rebel movements, however, may originate from a grassroots political movement but become dependent upon foreign financing over time. Hezbollah in Lebanon, for example, largely originated in the grass roots Shi'a political and militant movements of the late 1970s. However, it was the direct sponsorship and influence of Iran's Revolutionary Guard that molded the nascent movement into a unified guerilla organization.Footnote ³¹

The relationship between foreign sponsors and rebel groups can be understood in principal-agent terms. Governments contract with rebels to provide foreign policy services. Rather than fight their rivals directly, states issue funds and arms to insurgent groups as an indirect means of warfare.Footnote ³² Rebels receive increased resources but lose some autonomy over their operations because continued funding is contingent on carrying out objectives that are at least partly determined by the interest of the sponsor. For example, Pakistan closely monitors and directs the activities of anti-Indian groups, including the selection of targets.Footnote ³³ Regardless of ethnic or ideological affinities, sponsors rarely provide resources without strings attached. Foreign sponsorship implies some degree of influence over the group and becomes a means of coercive bargaining between states.Footnote ³⁴

The potential for agency loss or moral hazard is a significant concern for sponsors.Footnote ³⁵ Rebels may carry out actions that are unwelcome or embarrassing to their sponsor states. First, they may shirk responsibilities and devote suboptimal effort to the war while privately consuming the sponsor's resources. Second, rebels can prove incompetent and fail to effectively challenge the target state. For instance, Sudan and Zaire largely created the Allied Democratic Forces and the West Nile Bank Front with the intentions of toppling Museveni's regime in Uganda. In both cases, however, the movements largely engaged in terrorist attacks on civilians and were never effective at weakening the regime's control over the state.Footnote ³⁶ Finally, the rebel agents may engage in egregious behaviors that are either contrary to the goals and strategic interests of the principal or that generate domestic and international backlash against it. Contra violence against peasants and aid workers in Nicaragua publically embarrassed the Reagan administration and led to widespread condemnation from both domestic and international human rights organizations. At the extreme, rebels can turn on their former patrons, such as when Laurent Kabila in the Democratic Republic of the Congo abandoned his former ally, Rwanda, leading to the Second Congo War.Footnote ³⁷

This principal-agent relationship may also provide incentives to kill innocents. In some cases, rebels target civilians in an effort to demonstrate their commitment to the sponsor and maintain their resource flows. Such violence may either simply serve as a signal that an otherwise ineffectual group is “doing something” to carry out its sponsor's wishes or, more egregiously, targeting civilians may represent a strategy explicitly dictated by the sponsor. With respect to the former, Hovil and Werker find that the Allied Democratic Forces in Uganda terrorized civilians to demonstrate their commitment to their external sponsors, Congo and Sudan, because they were unable to directly challenge the state.Footnote ³⁸ As for targeting civilians, the war crimes tribunal in Sierra Leone war recently convicted Liberian President Charles Taylor for deliberately encouraging civilian abuse by militant factions as a strategy to destabilize the country and plunder its resources. Many sponsored militant groups—Hezbollah (Iran/Syria), Lashkar-e-Taiba (Pakistan), and RENAMO (South Africa/Rhodesia)—are notorious for targeting noncombatants, and such abuses are often tolerated (and sometimes encouraged) by external funders. To reiterate, we argue that foreign sponsorship creates incentives for violence against civilians. This occurs because foreign support stands as a substitute for local support, reducing the group's reliance on the population. Moreover, in some cases, the sponsors may tolerate attacks on civilians, and a few may either directly or indirectly encourage attacks on civilians.Footnote ³⁹ This leads to our first hypothesis:

Our proposed mechanism suggests a corollary argument about resources and support. If lucrative natural resources and foreign support have similar effects—namely, reducing reliance on civilian populations, and attracting poorly disciplined fighters—then, empirically, they should be indistinguishable from one another. Foreign funding is simply another resource. However, if foreign support comes with additional incentives to injure civilians as a signal of commitment, then the effect of foreign support on civilian killings should be greater and more robust than that of lootable resources. Note that we do not argue that reducing reliance on civilians and attracting thuggish fighters are unimportant. Instead, our principal-agent logic implies an even greater likelihood of abuse when foreign funding is available.

For some foreign sponsors, however, mass killings by their rebel agents may be met with disapproval and even punitive action. Foreign sponsors have different tolerances for human rights violations and war crimes by their rebel agents. Authoritarian regimes may not be as concerned with cultivating an image of respect for human rights and may not care about the fate of civilians in the war zone. They often do not face strong domestic human rights pressures and pay mere lip service to global human rights norms.Footnote ⁴⁰ Democracies, on the other hand, are generally more sensitive to human rights activism and have a deeper commitment to international norms regarding war crimes. Although democratic regimes have certainly backed rebel movements with questionable human rights standards,Footnote ⁴¹ they are generally more likely to select less abusive rebel agents and to sanction rebel groups that engage in bad behavior when compared with authoritarian sponsors.

Human rights have frequently been linked to US foreign military assistance programs,Footnote ⁴² and the US Congress has often forced the issue through legislation. For instance, the “Leahy Law,” introduced to the Foreign Operations Appropriations Act by Vermont Senator Patrick Leahy, forbids military assistance to foreign governments that violate human rights. Congress has also acted to prohibit arms transfers to rebel organizations believed to engage in civilian abuse. For instance, an amendment to the US Arms Control Act of 1976 sponsored by Senator Dick Clark of Iowa explicitly prohibited military aid to UNITA forces in Angola, even though this was later circumvented by the Reagan administration. Similarly, activists placed significant public pressure on Congress and President Reagan to stop funding human rights violators in Central America during the wars in El Salvador, Guatemala, and Nicaragua. With respect to Nicaragua in particular, the Boland Amendment in 1982—named after House Representative Edward Boland—specifically forbade funding the Contras in their effort to topple the Sandinista regime. Again, while the Reagan administration ultimately undermined this provision, leading to the Iran-Contra scandal, the very fact that scrutiny over assistance to militant organizations exists provides incentives to avoid publically embarrassing human rights violations.

In more general terms, principals seek to avoid moral hazard through both screening and sanctioning mechanisms.Footnote ⁴³ For democratic states, agency loss includes human rights violations by rebels. Principals use screening mechanisms to identify agents with preferences most similar to their own. They may host meetings or establish low-level contacts in order to gather further information on the organization. For example, the United Kingdom has scrutinized the behaviors of the Syria opposition in order to assess the group's “commitment to human rights and an inclusive government.”Footnote ⁴⁴ Sanctioning involves punishment (and withdrawal of support) if agents commit actions that are contrary to the principal's interests. With respect to democratic sponsors funding rebel groups, this implies that the causal relationship likely runs in both directions. Rebels who are less abusive are more likely to attract the support of democracies, and those supported by democracies are more likely to be punished for egregious acts.

Although separable in principle, both processes—screening and sanctioning—occur in any principal-agent relationship. The worse the initial screening, the more sanctioning may be required, but practically speaking, principals engage in both types of activity. Empirically, we can find examples of both screening and sanctioning. For instance, before the NATO intervention in Kosovo, there was a robust debate in Western countries about whether or not to arm the Kosovo Liberation Army, and such debates hinged upon the trustworthiness of the opposition (screening).Footnote ⁴⁵ Later, when abuses by the KLA came to light, Kosovar leaders came under indictment by the UN tribunal in the Hague on war crimes charges (sanctioning).Footnote ⁴⁶ Similarly, while India had been supporting the Tamil Tigers (LTTE) in Sri Lanka for many years, mounting reports of unreliability and abuse by the organization—among other reasons—lead to a withdrawal of support.

The recent NATO intervention in Libya provides an additional example. During the 2011 uprising in Libya, international actors expressed serious concern over the rebels' “true” identity and interests before offering international recognition and military assistance. Many world leaders reserved judgment on this point until well into the insurgency, and only after meeting with opposition forces. Even as rebel forces were advancing toward the capital, Western media outlets and human rights organizations paid close attention to reports of massacres and reprisal killings by the Libyan opposition.Footnote ⁴⁷ As the Muammar Gaddafi regime faltered, NATO members gradually offered recognition and assistance to the Transitional National Council (TNC), the main opposition group. In recognizing the TNC as the legitimate representative of the Libyan people, British Foreign Secretary William Hague emphasized its commitment to a more “inclusive political process,” respectful of the wishes and rights of the Libyan people.Footnote ⁴⁸ Following the ouster of the old regime, foreign governments and nongovernmental organizations (NGOs) have pressed the new Libyan regime to respect human rights, disband militias responsible for reprisal killings, and many have expressed concern over the extrajudicial execution of Gaddafi.Footnote ⁴⁹

There are at least three reasons why democratic states are more likely to seek reliable rebel movements and sanction bad behavior. First, democratic states may be genuinely more committed to human rights norms and standards; at the very least, they place strong rhetorical emphasis on respect for physical integrity rights. Violating these standards can be costly to a state's reputation, both domestically and internationally. Second, executives in democracies—who are tasked with carrying out foreign policy—are more constrained by domestic legislatures, courts, and other institutions that scrutinize their behavior and can interject human rights principles. Third, because of their open political systems, democracies are more likely to have strong human rights lobbies and face popular pressure. Activists can “name and shame” governments for bad behavior, including support for human rights violating rebel organizations.Footnote ⁵⁰ Such shaming efforts coupled with a strong organizational presence can affect state behavior.Footnote ⁵¹

We believe that this last mechanism—opportunity for human rights activists to scrutinize state foreign policy—is critical. Human rights organizations (HROs) work to bring human rights abuses to light and pressure states to adhere to international standards. The power of the transnational human rights movement lies in its ability to put pressure on multiple governments at once, both to improve human rights standards domestically and to pressure foreign governments to do the same.Footnote ⁵² Although they do not have the coercive resources of states or the financial resources of firms, they influence politics through persuasion and through their ability to name and publically shame human rights abusers.Footnote ⁵³ Naming and shaming can be particularly effective when based on international treaties and institutions—such as the Geneva Conventions, UN Convention on Genocide, and the International Criminal Court—which explicitly delineate acceptable standards of behavior, thereby providing clear benchmarks for assessment. Moreover, in democracies, these principled actors can be especially important in persuading the public (that is, voters) and putting moral concerns on the policy agenda. Because of freedom of association and speech, human rights organizations are also likely to have greater influence in democratic polities. As a result, human rights activists and other advocacy groups have secured some notable successes, including making the prosecution of war crimes part of the international agenda.Footnote ⁵⁴ We will soon test for the independent effects of democratic institutions and the strength of the human rights lobby. To summarize this discussion, democracies are subject to significantly greater human rights pressures than autocratic regimes. Autocratic principals may tolerate or even encourage civilian killings by their agents to foster instability in the target. This leads to our second set of hypotheses:

Principal-agent theory suggests that agents with multiple principals are less constrained in their behavior than those with a single principal.Footnote ⁵⁵ A single principal is able to give clear directives and the agent reports to a sole actor. In cases where the rebel organization has multiple principals, however, rebel groups can play these actors off against one another to extract a better bargain on more favorable terms. Not being as reliant on a single funding source, such rebels are less sensitive to sanctioning and threats from any particular principal. In order to sanction the bad behavior of a rebel organization with multiple principals, these actors must coordinate their action with one another, and overcome the collective action problem when the interests of the principals diverge. Acting collectively increases the costs of punishing rebels. With potentially contradictory directives and the ability to find alternative sources of funds, rebels backed by multiple states thus have fewer incentives to protect civilians.Footnote ⁵⁶

However, the sponsors' regime types should condition this effect. As a single autocratic sponsor is unlikely to be concerned with human rights violations, we do not expect multiple autocratic sponsors to be more or less sensitive to rebel organizations' civilian killings. We do expect the effect of democratic principals to be diminished when multiple foreign actors back the group. If a rebel agent finds constraints imposed by democratic sponsors too limiting, the organization can turn to other principals that have a higher tolerance for war crimes. Therefore, in addition to our simple hypothesis about multiple principals, we also posit a conditional effect depending on regime type.

This discussion yields five distinct hypotheses, which we test after briefly introducing the data and measurement of the key conceptual constructs.Footnote ⁵⁷

Data

Our unit of analysis is the individual rebel organization, with one observation per organization-year. The expanded Non-State Actors (NSA) in Conflict dyadic data, recently constructed by Cunningham, Gleditsch, and Salehyan,Footnote ⁵⁸ defines the sample of conflict groups. As opposed to highly aggregate research designs, which utilize country-years or conflict-years, our research design disaggregates wars to the level of the rebel organization. This allows us to look at the behavior and attributes of each organization separately within the same conflict and country. In multiactor conflicts we are able to avoid problems of overaggregation associated with lumping all opposition groups together as “rebels,” which ignores meaningful variation across organizations in terms of foreign funding, abusiveness, and other features of interest.

To test the relationship between foreign support and rebel violence against civilians, we analyze recent data on one-sided violence against noncombatants collected by the Uppsala Conflict Data Program (UCDP), which provides information on the count of people killed by the group during the year.Footnote ⁵⁹ The definition of one-sided violence applied by the UCDP is the intentional and direct use of violence against noncombatants, resulting in at least twenty-five deaths per annum. This definition excludes deaths by siege or infrastructure damage as well as deaths from battlefield error, negligence, or crossfire.Footnote ⁶⁰ Because of the inclusion criteria, the death rate is sometimes much lower than media reports and other counts of “civilian deaths” suggest. However, this high bar for inclusion also makes the data ideally suited to testing hypotheses on purposeful attacks on civilians because it excludes “collateral damage” and other incidental violence. Because of the temporal bounds of the one-sided violence data set, the merged data sets yield a sample of roughly 1,000 observations for more than 200 rebel groups in more than 100 conflicts for the years 1989–2009.Footnote ⁶¹

To test our first hypothesis regarding the relationship between foreign support and civilian victimization, we employ the measure of foreign support in the NSA.Footnote ⁶² This is a binary indicator for whether an insurgent organization received military, economic, or other material support from any foreign state. Although this variable is sufficient for testing the basic hypothesis on support, our theory posits that different sources of support will lead to different levels of civilian victimization. Specifically, our theory and hypotheses require information on three features of the external support: (1) the number of supporters, (2) the sponsors' domestic political institutions, and (3) the strength of human rights organizational presence in the sponsor state. We calculate the number of sponsors by summing the number of states providing material support to the insurgency in a given year. Because the NSA data provide the name of each foreign sponsor, we can determine whether each foreign sponsor is democratic by matching it with data on domestic political institutions from the Polity IV data set.Footnote ⁶³ For human rights organizational presence, we use the number of human rights organizations (HROs) with permanent offices in these states compiled by Murdie and Bhasin.Footnote ⁶⁴ We believe that permanent offices are important because they indicate a long-standing, institutional basis for activism and represent more reliable data than raw membership numbers.

We use several different permutations of our support variables. support is a binary variable accounting for the presence of any foreign support. number of supporters is the count of foreign supporters. democratic support is a binary variable indicating that at least one democratic state supported the insurgent group. We are also interested in how the composition of collective foreign support influences group behavior. We therefore construct the variable percent democratic, which captures the proportion of supporters that are democracies when multiple sponsors exist. This allows us to determine if a greater concentration of democratic supporters imposes a more severe constraint on the behavior of the sponsored group. Finally, we construct an interaction term composed of our number of supporters and percent democratic variables to examine whether increasing numbers of supporters mitigates the influence of democratic supporters.Footnote ⁶⁵

Previous research suggests that conflicts in which lootable resources play a significant role may lead to more violence against civilians.Footnote ⁶⁶ In particular, Weinstein's argument presumes that the causal mechanisms linking lootable resources and foreign support with violence should be largely analogous. Our argument, though related to Weinstein's in its emphasis on rebel resources, specifies a different set of mechanisms, which arise from the principal-agent relationship between state sponsors and rebel organizations. If Weinstein's theory is correct, variables accounting for lootable resources and foreign support should both be statistically significant and similar in effect. However, we have argued that foreign sponsorship provides additional incentives to abuse civilians as a way for groups to signal commitment to sponsors. We account for the presence of lootable resources in all of our models. The measure is a binary indictor for the presence of lootable resources (gems or drugs) within the conflict area. Geospatial data on the location of gems is from LujalaFootnote ⁶⁷ and Gilmore and colleagues,Footnote ⁶⁸ whereas drug data are from Buhaug and Lujala.Footnote ⁶⁹

We also include several control variables. Recent research has suggested that the relative balance of capabilities between insurgent and government forces influences rebel strategies of violence. Wood, for example, argues that stronger insurgents are less likely to target the population because they are better able to provide security and other benefits that encourage recruitment and foster loyalty among the population.Footnote ⁷⁰ In a similar vein, Metelits suggests that when insurgent groups face increasing competition from the government, they are more likely to turn toward violence against civilians.Footnote ⁷¹ To account for the effect of insurgent military capabilities, we include a binary measure called fighting capacity, which indicates whether or not the group enjoyed a moderate or high level of capacity to win battles against the government. The measure is from the NSA data.

Recent research suggests that territorial control is strongly related to the types of violence armed actors employ. Kalyvas demonstrates that armed groups are less likely to employ indiscriminate violence against civilians in areas where they possess strong territorial control.Footnote ⁷² Balcells similarly shows that during the Spanish Civil War insurgents were more likely to resort to using “indirect violence” such as aerial and artillery bombardment against areas in which their support and control was low.Footnote ⁷³ The absence of spatially disaggregated data on territorial control prohibits us from directly testing these arguments. However, we believe that the ability to exert effective control over territory in the aggregate should be negatively correlated with violence against civilians. territorial control is a binary indicator measuring whether the group exerted at least a moderate level of territorial control; it is taken from the NSA data set.

We also include features of the conflict environment. Previous research demonstrates that more intense conflicts create greater incentives for violence against civilians.Footnote ⁷⁴ The variable war represents conflict years in which the number of battle deaths exceeded 1,000, which is the standard definition of civil war in the literature. The measure is taken from the UCDP. Second, we control for the duration of the insurgency. duration is the count of years since the beginning of the conflict. Past research has also shown a positive relationship between insurgent and government violence.Footnote ⁷⁵ As such, we include government one-sided violence as the log-transformed value (plus 1 to account for zeros) of the count of one-sided violence perpetrated by the government in a given year. We included a lagged dependent variable to address temporal dependence and serial correlation.

Characteristics of the conflict state may also influence one-sided violence by rebels, so we control for two relevant state-level variables. We include a control variable indicating the institutional composition of the conflict state. Democracies may encourage violent rebel attacks on civilians by virtue of the linkages between the preferences of the citizens and the actions of the state.Footnote ⁷⁶ We measure democracy with a dummy variable indicating a score of 6 or greater on the combined Polity IV scale. We also control for the state's population size. Countries with larger populations present more opportunities for violence against citizens. population is measured using the natural log of the conflict state's total population.Footnote ⁷⁷

Model and Results

The structure of the dependent variable is a time-series cross-section of rebel groups with annual counts of one-sided violence against civilians by rebel organization. Because the dependent variable demonstrates significant overdispersion, we estimate a negative binomial model. In order to account for within-group correlation, we employ robust standard errors clustered on the insurgent-government dyad. Taken as a whole, the results from the statistical model support the hypothesized relationships discussed earlier. Model 1 in Table 1 presents the results for the most basic model, which includes only the binary measure indicating whether the rebels received material support from a foreign patron.Footnote ⁷⁸ In this model, the coefficient for the aggregate support variable is positive and achieves statistical significance at the 0.1 level. This result provides some support for the general argument that support from foreign sponsors encourages higher levels of violence against noncombatants. However, this must be taken with a grain of salt because we expect democratic and autocratic sponsorship to have very different effects.

Model 2 tests the relationship between democratic sponsorship and insurgent behavior. As before, the variable for any support is statistically significant and positive, indicating that support is associated with greater abuses against the local population. Indeed, once we parse out democratic sponsorship, this variable is more statistically significant. As expected, the coefficient for the variable accounting for the presence of support from at least one democratic sponsor is negatively signed and statistically significant, suggesting that groups with at least some support originating from democracies are less likely to target civilians than rebel groups supported only by non-democratic states. Indeed, the coefficient on democratic support nearly cancels out the effect of sponsorship in general.Footnote ⁷⁹ Model 3 shows the results for our alternative measure of democratic support, which accounts for the proportion of foreign sponsors that are democracies. The coefficient for this variable is negative and statistically significant, suggesting that as the percentage of a group's supporters that are democratic states increases, the less violence it employs against civilians. These results lend support to our second hypothesis.

Next, we examine the effect of multiple principals. In Model 4 our indicator of the number of foreign supporters is positive and statistically significant. This result holds when we include our indicator for proportion of democratic sponsors in Model 5. Consequently, we find support for H3a, which argues that multiple principals reduce the constraints on agents and encourage greater levels of violence against civilians. However, this variable does not account for regime type among the several foreign patrons. Model 6 explores the contingent influence of the number of principals on the constraining effect of democracy. The coefficients for both constituent terms are positive but statistically insignificant. However, looking at the significance of individual components in the interaction can be misleading.Footnote ⁸⁰ The interaction term composed of the percent democratic variable and the number of supporting states variable is both negative and significant with 90 percent confidence intervals. This indicates support for H3b. We return to this relationship in our discussion of marginal effects.

Although the results of our first model offer support for the hypothesis that the resources provided by foreign sponsors may encourage violence against civilians, our discussion and quantitative analysis also demonstrate that not all support is the same.Footnote ⁸¹ Democratic leaders—who are more likely to express an interest in human rights and are more likely to face domestic backlash when agents engage in high levels of civilian killing—are more likely to take steps to curb such behavior. Because of selection by democracies and constraints imposed, insurgents supported by democratic states are less likely to engage in high levels of violence than their counterparts whose support comes from nondemocratic states.

An additional finding of note surrounds the role of lootable resources in shaping insurgent behavior. If foreign sponsorship is simply another resource that does not require cooperative relationships with civilians, it should have effects similar to that of gemstones and narcotics. However, the results presented here fail to support this conjecture. The variable accounting for the presence of lootable resources is positive across the specifications but fails to achieve statistical significance. Although this certainly does not serve as a definitive test of Weinstein's theory, the results nonetheless indicate that there is no statistically significant relationship between the presence of loot in the conflict zone and the extent of civilian abuse.Footnote ⁸² Foreign sponsorship has a noticeably larger and more significant effect than natural resources alone.

Turning to the control variables, the severity of the conflict is positively correlated with the severity of abuse, and the coefficient is statistically significant.Footnote ⁸³ One might presume that severe conflicts simply produce greater collateral damage, yet the one-side violence data we have used as our dependent variable explicitly exclude such incidental and unintentional killing. Government violence against civilians measure is also positive, and it is marginally significant across the various model specifications. A group's overall ability to effectively challenge the government on the battlefield—its fighting capacity—is also related to its decisions regarding violence against civilians. Groups that enjoy at least a moderately high level of fighting capacity are more likely to engage in high levels of violence against civilians. This result is contrary to previous findings and suggests the need to further explore the relationship between rebel capabilities and violence strategies.Footnote ⁸⁴ However, the variable accounting for a group's ability to exert control over territory is negatively correlated with civilian targeting, and the coefficient is significant. This provides some support to theories suggesting that the ability to control and govern territory makes civilian abuse less likely.Footnote ⁸⁵

Lastly, population size is significant and positive in most model specifications. Thus, countries with larger populations are more likely to face higher levels of violence against civilians. Other controls fail to achieve statistical significance in our models. Conflict duration is positively correlated with violence toward civilians, though it is not statistically significant. We likewise fail to find support for the proposition that conflicts in democratic states invite higher levels of violence from insurgents. The lagged dependent variable also fails to achieve statistical significance. Although this result is perhaps surprising, it suggests that violence against civilians is more a function of changes in the conflict environment than of the group's “type.”

In order to illustrate the substantive impact of various aspects of support, we also calculate the predicted effects on the level of violence insurgents employ against civilians. Based on Models 1 and 2 in Table 1, we estimated the marginal effects of support and democratic support. Figure 1 displays these effects graphically using a violin plot, which is a convenient way to display marginal effects of explanatory variables on expected changes in the dependent variable.Footnote ⁸⁶ It combines the advantages of a boxplot, which show the average effect and the uncertainty surrounding it, with a kernel density plot, which shows the entire distribution of the variable's effect. Specifically, we construct a box plot of the marginal effect under two high probability scenarios, and then add a rotated normal kernel density plot to each side of the box plot.

Figure 1. Violin plots of marginal effects for foreign support (top) and foreign democratic support (bottom) on one-sided violence (OSV)

Holding all covariates at their mean values, and binary variables at their modal values, the expected number of one-sided deaths in a given year more than doubles when rebel organizations have foreign support—supported insurgents are predicted to deliberately kill more than double the number of civilians per year. This indicates strong support for our claim that rebel organizations with foreign support are more likely to engage in one-sided violence against civilians than those without foreign support. Our second hypothesis suggested that rebel groups receiving external support from democracies are less likely to engage in one-sided violence against civilians than those groups that receive support from autocracies. Consistent with this hypothesis, the predictions suggest that democratically sponsored rebel groups engage in roughly half as much one-sided violence as their nondemocratically sponsored counterparts. Interestingly, these predictions indicate that groups supported by democratic sponsors are expected to behave similarly to those with no foreign support. These results further reinforce the argument that the sponsor type matters—when sponsors care about human rights, rebel agents are more likely to act with greater restraint toward civilians.

Figure 2 visually displays the interaction between the number of foreign supporters and the proportion of democracies in the support group along with uncertainty estimates for each prediction. On the x axis is the number of democratic sponsors for a given number of total sponsors, which we vary by panel (one through six sponsors), and on the y axis we show the predicted number of one-sided violence incidents. We find that the marginal effect of additional democracies on one-sided violence incidents decreases as the total number of sponsors increases. For an average rebel group with six sponsors, the expected number of one-sided violence incidents is approximately ten if none of the sponsors are democratic (0 percent). It drops to three if one of the sponsors is democratic (17 percent), and is just one expected incident if two of the sponsors are democratic (33 percent). Yet the marginal effect of additional democratic sponsors in the mix has a decreasing effect on the expected count of one-sided violence incidents. For a rebel group with four external sponsors, the expected number of one-sided violence incidents is approximately four if none of the sponsors are democratic (0 percent). It drops to about one if one of the sponsors is democratic (25 percent), and just 0.5 if two of the sponsors are democratic (50 percent), and then to 0.25 for three democratic sponsors (75 percent). This suggests that both multiple and democratic sponsorship are important factors in predicting the magnitude of violence against civilians, but that their effects are conditional. The marginal effect of additional democratic sponsors is proportionally less than for the first one—the slope is initially steep but tapers off as the number of democratic sponsors increases, and the slope is significantly steeper for rebel groups with multiple sponsors than for those with only a couple sponsors. In sum, this implies that having at least one democratic sponsor can significantly diminish incentives to engage in civilian victimization, especially for rebel groups with multiple sponsors, but that this “democratic effect” has diminishing returns.

Figure 2. Interaction effect of the number of sponsors and the proportion of democratic sponsors on one-sided violence