Violent Competition and Terrorist Restraint

Abstract

A large literature has argued that domestic competition increases a militant organization's use and severity of terrorism to differentiate their “brand” and “outbid” other organizations. However, most empirical analyses infer such competition from the quantity of groups present in a geographic area. This approach neglects specific group relationships, such as cooperation, rhetorical or violent rivalry, or peaceful coexistence. We introduce a behavioral measure of group competition and argue that variation in the quality, rather than the quantity, of competition affects the violence profile of militant groups in unexpected ways. Violent competition, where militants attack one another, imposes significant constraints on group resources and increases groups’ dependence on civilian support, which exacerbates the costs of a popular backlash against brutality. Moreover, violent competition effectively substitutes for crowding out rivals via outbidding. As competition becomes extreme, we posit that groups increasingly opt for a strategy of terrorist restraint and reduce the share of high-profile attacks on soft civilian targets. We test this argument at the macro and micro levels with cross-national data on 290 organizations in civil war (1970–2018) and granular data on the subnational targeting strategy of the Islamic State in Syria (2013–2018). Both analyses provide robust support for our argument. The findings shed light on the strategic limitations of outbidding and provide important insights for research and policy.

Armed conflicts around the world are characterized by increasing levels of intergroup competition and infighting. Fatalities in conflicts between nonstate actors have risen steadily since 2013 and today remain at a historically high level.¹ Many of the armed groups engaged in conflict with one another also resort to terrorism as a tactic. Indeed, in recent years, terrorist infighting has reached unprecedented intensity.²

While a growing body of work studies the causes of cooperation and conflict between armed groups, including rivalry and competition,³ the effects of such competition on the dynamics of violence and terrorism remain debated.⁴ The dominant explanation of the effect of intergroup competition on the behavior of militant groups focuses on a logic of outbidding,⁵ whereby increases in competition produce an escalation of terrorist violence, especially in its most brutal forms, as groups attempt to brand their violence and drum up popular support vis-à-vis rivals. While the outbidding mechanism is intuitively appealing, its empirical record appears mixed, with only limited support for the theory.⁶ This suggests that the ways groups actually respond to competition may be more complex.⁷

To illustrate this puzzle, consider the gruesome terrorist attack at the Kabul airport in August 2021, where a suicide bomber detonated an explosive belt at one of the airport gates during the evacuation from Afghanistan, killing at least 170 Afghan civilians and 13 members of the US military. This attack was claimed by the Islamic State–Khorasan Province (IS-KP), the Taliban's main rival in Afghanistan, and such rivalry has been invoked to explain the escalation of violence.⁸ While this is a plausible explanation in line with outbidding theory, a closer look at the conflict between IS-KP and the Taliban reveals some surprising patterns. Data in the Global Terrorism Database (GTD) suggest that when the Taliban faced the most intense violent competition with IS-KP, its propensity to target civilians actually declined. In other words, for the Taliban, increases in intergroup competition are associated with a decrease in the group's overall brutality. This reaction is puzzling and cannot be explained by the theory of outbidding, which would have predicted an escalation of violence, especially against civilians. What explains this behavior?

The question of how violent intergroup competition affects the use of terrorism has received incomplete answers from the literature for two main reasons. The first is how competition is conceptualized and measured. Nearly all studies focus on the number of groups active in a given country or conflict, or the change in that number.⁹ While these measures are good approximations of how crowded a conflict marketplace is, they tend to overlook the specific relationships among those groups, which could include cooperation and even alliances, as well as peaceful coexistence.¹⁰ Moreover, where competition occurs, it could be merely rhetorical or ideological.¹¹ In other words, the focus has been on the quantity of (potential) competition rather than its quality.

Second, with respect to outcomes, research has either examined the effect of competition on the number of terrorist attacks (the quantity of terrorism) or considered qualitative aspects of terrorism in isolation. Therefore, we do not know how competition affects the overall violence profile of groups, including substitution of brutality with restraint and vice versa. Aggregating all terrorist attacks into a single number overlooks variation in target selection and degrees of brutality, whereas isolating specific forms of violence obscures the overall violent strategy of a group; a group that targets civilians 10 percent of the time is different from a group that targets civilians exclusively. Only a holistic focus on violence profiles can tell us whether civilians, governments, or other targets are more or less at risk as a result of violent competition, a question of great relevance not only for scholars but also for policymakers.

This article aims to fill these gaps. We develop a theory of how violent intergroup competition affects the violence profiles and terrorist strategy of militant groups simultaneously engaged in civil war. We argue that violent intergroup competition shifts the composition of terrorist violence, reducing attacks on civilians relative to government and hard targets. In other words, the quality, rather than the quantity, of competition shapes a group's violence profile and generates incentives for terrorist restraint versus escalation. Our theory emphasizes two aspects that have been downplayed in previous studies of outbidding: the costs of terrorist brutality and its decreasing marginal returns.

Terrorist brutality entails significant legitimacy and reputation costs for armed groups.¹² This is especially true when groups are simultaneously engaged in a civil war against the government, a context where civilian support is paramount for a group's viability and survival.¹³ The concomitant engagement in violent competition with other nonstate actors increases resource pressures and, in turn, a group's dependence on civilians. Because civilians provide key resources to militant groups, attacks that alienate the civilian population are particularly counterproductive. Thus, when intergroup competition turns violent, more competition incentivizes terrorist restraint rather than escalation.

Another reason for restraint concerns the diminishing returns of severe attacks on civilians, especially when groups simultaneously engage in violent competition. Escalation may be beneficial in the initial stages of competition, as more groups enter the conflict marketplace.¹⁴ However, brutality can be increased only up to a point, and when other groups jump on the bandwagon it ceases to be an innovation and becomes routine. Moreover, the very logic of using escalation to gain credibility and shift constituency support away from rivals becomes moot once a group is set on physically eliminating its direct competitors. When militant groups attack each other, the battlefield becomes the arena where groups build their reputation and demonstrate their superiority. This is because the battlefield can reveal otherwise private information about a group's capabilities and resolve not only to combatants but also to their civilian audience. Under violent competition, escalation against civilians is unhelpful but also unnecessary: a group can reap greater benefits from a strategy of restraint.

We test the observable implications of our theory at the macro and micro levels. At the macro level, we conduct a cross-national analysis on 290 militant organizations engaged in terrorism and civil war between 1970 and 2018. For this analysis, we introduce a new measure of violent intergroup competition. At the micro level, we take the case of the Islamic State of Iraq and al-Sham (ISIS) in Syria between 2013 and 2018. While Syria is one of the most competitive contemporary conflict environments, ISIS is a hard case for our theory, since the group is widely recognized as particularly brutal toward civilians. The availability of fine-grained, subnational data on ISIS's terrorist targeting strategy under episodes of violent competition allows us to examine when and where competition occurs and the specific spatiotemporal dynamics of ISIS responses, as well as to probe our causal mechanisms. Both analyses provide robust support for our theory.

This study makes several contributions. First, it provides a behavioral conceptualization and measurement of intergroup competition focusing on quality rather than quantity. It empirically demonstrates how violent competition shifts the composition of terrorist violence and creates incentives for restraint rather than escalation. In other words, the quality and quantity of competition produce substantially different effects on groups’ violent behavior.

Second, the argument provides a theoretical counterpart to outbidding theory by explaining why and when intergroup competition leads militant groups to de-escalate violence against civilians. In doing so we offer a novel explanation of terrorist behavior that accounts for the empirical puzzle of many terrorist groups’ behaving in ways that are inconsistent with the expectations of outbidding theory.

Finally, this research emphasizes the strategic limitations of outbidding and the possible shortcomings of the frequent analogy between terrorist organizations and firms in competitive markets. As long as groups compete nonviolently for the support of the same constituency, the firm analogy is helpful to understand groups’ strategies to differentiate themselves and outbid competitors. However, unlike firms, militant groups often engage in violence against one another. The requirements and costs of violent conflict, as well as civilians’ reaction to it, can fundamentally alter the expected benefits and costs of terrorist escalation and brutality.

Previous Research

Terrorism is the deliberate creation and exploitation of fear through violence or threat of violence by nonstate actors against noncombatant targets in the pursuit of political change.¹⁵ It is a tactic of indirect targeting which operates through the intimidation of a larger audience beyond the immediate victims or physical targets. Among often-cited determinants of terrorism, scholars have focused on competition between groups as a driving force for the use of wide-scale, attention-grabbing attacks.¹⁶ The logic rests on the notion of outbidding: when multiple armed groups compete for support from the same constituency, their violent acts will intensify in scope and number. By engaging in particularly remarkable forms of violence, a group can stand out from its rivals and signal resolve, creativity, and credibility. Such signaling is meant to indicate to the public that this group has the best odds of revising the status quo.

While the idea of outbidding is important, this theoretical focus tends to downplay the reputational costs of terrorist escalation. While terrorism can impose costs on the government and signal resolve, it also risks alienating civilians and forfeiting public support.¹⁷ Even in studies that acknowledge the risks of terrorism in response to competition, there is an implicit assumption that the benefits of increased violence and brutality outweigh the possible costs.¹⁸ This assumption is not always warranted. Indiscriminate attacks on civilians are a minority of all the attacks reported in the Global Terrorism Database, which suggests a general preference of perpetrators for more selective forms of terrorism. Moreover, recent research on competition between armed groups suggests that escalation is not always beneficial¹⁹ and that groups often respond to competitive pressures by increasing tactical diversity.²⁰ This increase in diversity does not necessarily mean escalating brutality. In fact, sometimes groups diversify using nonviolent tactics.²¹

Beyond the costs of extreme forms of violence, scholarship has not been sufficiently clear on when competition is expected, what form it takes, or how variation in forms of competition affects terrorist behavior. Most studies have measured competition by the number of actors in the conflict marketplace²² or the number of groups sharing ideological ties.²³ The focus on the quantity of potential competitors disregards the quality of interactions between them and limits our understanding of competition between groups, both conceptually and empirically. A focus on the number of actors or ideological ties is potentially misleading. The number of actors could indicate opportunities for collaboration or alliance formation, as opposed to competition;²⁴ and if competition does occur, it could be be rhetorical rather than violent.²⁵ As a result, it remains unclear how violent intergroup competition affects terrorist strategies, including incentives for escalation versus restraint. Understanding these dynamics is important not only because violent intergroup conflict is spiraling around the world, but also because violent competition imposes different challenges and constraints on groups, compared to nonviolent or rhetorical competition.

In examining the effect of competition, most studies examine the quantity of terrorism and civilian targeting or qualitative aspects of it such as attack severity or the use of suicide bombings.²⁶ While examining the quality of terrorism resulting from competition is an important step in the right direction, existing analyses only tell us whether or not a certain form of terrorist violence occurs. What is missing is an understanding of the effect of competition on the violence profile of terrorist organizations—that is, the organization's relative preference for specific tactics and targets. Little is known about whether increased competition makes brutality the defining feature of an organization's violence or whether it simply makes the group's tactics more diverse but not necessarily more brutal or extreme.

In the next section, we shift the focus from the quantity to the quality of competition and develop a theory of how violent competition affects the overall terrorist profile of militant groups and shifts the composition of their violence.

The Quality of Competition and Terrorist Restraint

Our argument is rooted in a rationalist approach whereby terrorism is a means for militant groups to achieve their goals. Groups have not only long-term or maximalist goals, such as regime change or secession, but also proximate goals. These include attracting and retaining members and supporters, drawing attention to the group's cause, demonstrating strength and resolve to government opponents, and outcompeting rival groups. The theoretical framework we develop in this section is based on the link between terrorist violence and proximate goals.

Proximate goals are key to the survival of militant groups in the context of a civil war. Mobilizing and maintaining popular support is critical here, since civilians provide material and human resources that are necessary for the war effort, as well as information and security, by allowing militants to hide among them.²⁷ Moreover, civil wars are complex environments where threats to a group's survival can originate not only from the government (and its allies) but also from other nonstate actors. In fact, over 40 percent of all intrastate conflicts involve multiple nonstate actors, frequently in competition with one another. Conflict among militant groups can occur for multiple reasons. Groups may disagree on specific actions, responses, or tactics, but also on more fundamental issues such as goals and ideologies. Even when groups nominally fight on the same side and share the same overarching goals, conflict can emerge over the support of the population as groups attempt to establish primacy as leader of the rebellion or bulwark of the interests of the constituency they claim to represent.²⁸

Competition between militant groups can take various forms. Watts distinguishes between escalating competition and destructive competition. Escalating competition takes place when “terror groups attempt to outpace each other through expansive competition and occurs when competing groups separate geographically and the perpetration of successful attacks leads to gains in notoriety and subsequent increases in resources.” Destructive competition occurs when “terror groups attack each other and arises predominantly ... when terrorist factions occupy the same terrain,” where they compete for fighters and money.²⁹

The idea of escalating competition clearly reflects a logic of outbidding. But because it takes place among groups that operate in different geographical spaces, it tends to be primarily rhetorical or discursive. In other words, groups can engage in escalating competition, and increase their level of violence and civilian targeting, even without coming into direct contact with one another. The transnational outbidding among Salafi jihadist groups in different parts of the world is a good illustration.³⁰ On a domestic level, Palestinian groups are an often-cited example of escalating competition among proximate groups, although empirical evidence beyond this particular case appears rather limited.³¹ For example, studies of competition between al-Qaida and the Islamic State in Syria and Iraq have failed to uncover similar outbidding dynamics.³² Beside their occurrence in the same or different geographical spaces, a fundamental difference between escalating and destructive competition is that the latter focuses on directly attacking the rival with the aim of physically eliminating it.

Destructive competition between nonstate actors does not take place in a vacuum. Our theoretical argument focuses on groups in civil war, a context in which militants compete violently with the state while also competing with each other. At the same time, there are important differences between these two types of competition. For groups engaged in civil war, violent competition with the state is an inherent characteristic of the conflict. This stems from the very definition of civil war, which is based on a threshold of violence between government and armed nonstate groups’ forces (twenty-five battle deaths per year for a lower-intensity intrastate conflict, and a thousand for a full-scale civil war, according to the Uppsala Conflict Data Program).³³ Violence against state forces is also a necessity if militant groups want to prevail militarily or obtain favorable terms in a peace agreement. Put differently, violent competition between state and nonstate forces happens in all civil wars. This is not the case with violent competition between militant, nonstate groups. Competition with other nonstate actors sometimes does not occur at all; sometimes it is nonviolent or rhetorical; and sometimes it includes physical violence. As we will argue in the next section, this violent dimension of competition between groups fundamentally changes the expected costs and benefits of terrorist escalation, with implications for the violent strategies groups adopt also in their conflict with the state.

Why Violent Competition Makes Escalation Counterproductive

For militant groups fighting a government opponent, engaging in destructive, violent competition with other groups effectively entails fighting on two fronts simultaneously. This situation imposes major demands on group resources and fighting capabilities, beyond those incurred in the conflict with the state. Capability constraints and the group's survival require prioritizing among targets and focusing attacks on the group's direct enemies rather than the civilian population. Moreover, since civilians provide crucial resources for the war effort, preserving civilian support becomes paramount. Violent strategies that undermine civilian compliance are therefore counterproductive. Specifically, terrorist attacks that victimize the civilian population become even more costly for a group if civilians withdraw support in response. Furthermore, in the presence of violent intergroup conflict, in-group civilians are likely to be caught in the middle, either as collateral damage or due to their perceived allegiance to rival groups. This reduces civilians’ tolerance for violent strategies that place them further at risk, directly or indirectly (such as when attacks on civilians trigger government retaliation against militants’ constituency).

Fighting on multiple fronts also exposes a group to a greater risk of losses and a need to promptly replenish its ranks to ensure its viability. Terrorist attacks that victimize civilian fence-sitters thus can become highly counterproductive because they may incentivize those civilians to seek protection from, and support, the government. And when attacks are directed at government supporters, the internecine conflict on the militants’ side undermines their ability to capitalize on the potentially mobilizing effects of government retaliation.³⁴ Put simply, violent intergroup competition poses challenges and strain which are very different from simply operating in a crowded conflict marketplace.

Even when competition occurs alongside in-group/out-group cleavages, violent competition often takes place in contested areas, where groups do not have complete control, rather than in militant groups’ strongholds. Such areas are characterized by a severe information problem.³⁵ Each group is uncertain about civilian loyalties, the identity of enemy supporters, and more generally, the extent to which the civilian population supports it or its rival. In this scenario, targeting of civilians can backfire by inducing them to either withhold information and refuse to cooperate with the group or to declare support for the rival altogether.

So far, we have argued that violent intergroup competition makes a terrorist escalation counterproductive by increasing dependence on the civilian population and lowering civilian tolerance for the adverse impacts of escalation. Specifically, violent competition between militant groups increases the legitimacy and reputational costs of terrorist targeting of civilians beyond what the militants were already facing in their conflict with the state. In this regard, the civil war literature has highlighted several factors that shape the expected benefits and costs of terrorism in civil war, such as the balance of capabilities, sources of support, and government behavior.³⁶ Violent intergroup competition introduces an additional cost that affects militants’ strategic calculus of violence, for example, by exacerbating existing constraints or introducing a new source of vulnerability, and thus pushes militants toward greater moderation in their treatment of civilians.³⁷

A strategy of terrorist restraint can be a more effective response to violent competition for several reasons. While some militant groups may engage in restraint for principled reasons, most explanations focus on strategic nonstate actors and on the utilitarian nature of restraint in armed conflict.³⁸ Restraint toward civilians can be costly and limit the tactics and strategic choices available to militant groups. However, it can be useful to the extent that it buys domestic and international legitimacy and support. Most militant groups in civil war depend on some level of domestic support to be successful. “Good behavior” toward civilians, especially at a time of increased resource pressure, is more likely to be rewarded with greater cooperation. Thus restraint helps groups win the hearts and minds of the population and foster civilian support, which is paramount for survival. Moreover, terrorist restraint can reassure government opponents of the trustworthiness of the group and even increase the chances of negotiations. The Tuareg rebels in Mali adopted this strategy to differentiate themselves from AQIM and Ansar Dine. Terrorist attacks by Tuareg rebels dropped substantially after the Bamako peace agreement in 2015. When government opponents are abusive, restraint can further help a nonstate actor differentiate itself and bolster claims of the incumbent's illegitimacy.³⁹ Restraint can also attract support from external actors. Militant groups often seek to appeal to international audiences. Many of these audiences share a preference for civilian protection.⁴⁰ Restraint signals to these audiences that a group is worthy of support and widens the overall pool of possible external sponsors.⁴¹

It is plausible that militant groups less reliant on local support may experience lower costs from civilian targeting. However, in the short term, all groups in civil war will suffer the increased resource pressures stemming from the opening of additional military fronts. Even groups with external sources of support are subject to these constraints, and unless they can persuade the external sponsor to step up efforts, they are likely to experience a resource shortfall. This, in turn, increases the relevance of local support and strategic restraint even for these groups, at least in the short term.⁴²

Why Violent Competition Makes Escalation Unnecessary

Escalating and destructive forms of competition reflect different strategies toward rival groups. In fact, they can be regarded as substitutes. Both strategies aim at undermining a rival, but through different mechanisms. In one case violence is used indirectly to shift civilian support away from a rival, while in the other case, violence is employed directly against the rival.⁴³

Escalating intergroup competition aptly captures a logic of outbidding. Groups that operate in a crowded marketplace need to brand themselves as the most radical to “gain credibility and win the public relation competition,”⁴⁴ especially to convince the base of the group's superior commitment and viability. This creates incentives for an escalation in the number but also the severity of terrorist attacks. However, when competition with other groups takes a violent form, and militants directly attack one another, such outbidding becomes unnecessary because the goal is now to physically eliminate the adversaries. Put differently, directly targeting the other group is effectively a substitute for crowding them out via outbidding. Attacking civilians when groups simultaneously fight one another is unhelpful for tipping the balance and may even suggest group weakness. Indeed, the violent conflict between militants reveals otherwise private information about group capabilities and resolve.⁴⁵ A group that targets civilians while losing the fight against a violent rival may not appear as strong, capable, or resolved as in the absence of such violent competition. While reputational gains are the basis for outbidding, the battlefield is another place to build reputation and arguably a more informative one. All in all, the strategy of outbidding may be effective as long as competition remains primarily nonviolent or rhetorical. But once a group decides to directly target its rival/s, the struggle for primacy plays out more on the battlefield than in the court of public opinion.

Escalating and destructive competition could be temporally related and adopted in sequence. But even when competition begins nonviolently, terrorist escalation yields decreasing marginal returns and is unlikely to persist when groups transition to attacking one other. On an operational level, a group cannot increase the severity of its attacks indefinitely due to organizational constraints: killing a hundred people may be possible, but killing a thousand is much harder. As a result, the severity of violence will likely settle on a plateau that is a function of the group's capabilities. On a tactical level, the novelty of a tactic necessarily decreases with its frequency, and this in turn progressively reduces its social impact. As more groups jump on the bandwagon and emulate extreme tactics, the benefits of brutality for establishing a group's brand, notoriety, and superior resolve will diminish.⁴⁶ Moreover, when competition between groups becomes extreme and militants transition to destructive competition, greater civilian victimization is unlikely to significantly boost a group's profile. Not only does the battlefield become the place to build a reputation in the eyes of civilian audiences, but civilians themselves may become less tolerant of the human costs of violence once they are caught in the crossfire of intergroup conflict. Ultimately, the benefits of outbidding are not guaranteed, and there are limits to how much support a group can muster with this strategy when fighting its rivals at the same time.⁴⁷ In this context, moderation in civilian targeting might prove more effective for the group's survival and access to critical resources and support.

These considerations lead to the following observable implications.

H1: When a group is engaged in violent competition against other groups, the proportion of its terrorist attacks that are directed at soft civilian targets decreases.

H2: As the number of attacks against rival groups increases, the proportion of its terrorist attacks that are directed at soft civilian targets decreases.

Data and Research Design

We test our expectations using a data set of organizations engaged in terrorism in civil war. The universe is drawn from the UCDP/PRIO Armed Conflict Dataset and the GTD. We hand-match organizations across both data sets to determine which organizations in the GTD are parties to an intrastate armed conflict. We refrain from matching terrorist organizations to umbrella rebel organizations to avoid inflating or over-attributing terrorist attacks to specific groups.⁴⁸ Our matching process considers the specific years a group is involved in a civil conflict that reaches the twenty-five-battle-deaths threshold, as well as those that fall below the threshold, as long as the group is active. The unit of analysis is the organization-year. The data set contains 290 unique organizations from 1970 to 2018.

We focus on militant groups that engage in terrorism for several reasons. First, the largest body of work we wish to speak to in this article focuses on outbidding, which looks primarily at violence by terrorism users. Examining these actors allows us to draw more meaningful comparisons. Second, our argument links violent competition with terrorist target choice, which can be examined for only groups that engage in some form of terrorism. Third, a focus on groups that use terrorism may be regarded as a hard case for our theory of restraint, since these groups are willing and able to escalate attacks on civilians, whereas it would be unsurprising to see restraint from groups that never engaged in terrorism.

The dependent variable is the proportion of attacks against soft targets. Looking at the quality of violence by examining how often an organization attacks soft civilian targets in comparison to hard targets allows a more precise investigation into profiles of terrorist violence and composition shifts. Moreover, to empirically assess restraint, we need to consider groups’ relative preference for civilians versus other types of targets, and this cannot be done by simply counting the absolute number of attacks on civilians. We operationalize the proportion dependent variable with the sum of attacks on hard and soft targets as the denominator. Soft civilian targets include private citizens, civilian businesses, religious figures, civil society actors, the media, and educational institutions.⁴⁹ This is in comparison to hard, or official, targets, which include governmental assets and personnel, the police, and local infrastructure.⁵⁰ As Polo writes, “The logic of this classification reflects the degree to which specific attacks are likely to harm innocent civilians and therefore generate a popular backlash against the group.”⁵¹ Of the groups in our sample, al-Shabaab maintains a consistent preference for soft targets, while the Revolutionary Armed Forces of Colombia and the Taliban in Afghanistan demonstrate considerable restraint.

The main explanatory variable is intergroup infighting. We introduce a new, behavioral measure of violent competition. Using GTD target indicators, we extract information on the occurrence and intensity of violent competition by tracing attacks by the organization in question on other organizations in the same conflict environment.⁵² For the occurrence of competition, we create a binary variable measuring the instance of violent competition. For the intensity of competition, we calculate the number of attacks by this organization on any other organization in a given year. This variable is logged in our empirical analyses. Figure 1 displays the number of attacks on other organizations for a selection of groups in our sample.⁵³

FIGURE 1. Infighting attacks

Our measures of infighting focus on violent interactions between militant groups. We define infighting, or violent competition, as physical acts of violence between two nonstate actors. The measures are behavioral and differ from country- or group-level measures of competition which capture the number of nonstate actors active in the conflict or the change in that number.⁵⁴ Our measure also differs from recent departures from this operationalization, such as Blair and colleagues' and Conrad and colleagues' explicit coding of rivalries between militant organizations.⁵⁵ While useful, and more illustrative of competition, both measures offer only an indication of the presence of competition, as opposed to its intensity. Moreover, the coding of competition is quite broad. Blair and coauthors classify rivalries as nonviolent noncooperation between militant groups and code competition as an instance of two organizations with conflicting goals working to defeat one another. This coding structure does not explicitly capture violent interactions. It is also limited in its focus on organizational goals, which do not always mean competitive environments. In our data, 36 percent of organizations launched at least one attack on another organization. The Islamic State and the Taliban are among the organizations launching the most violent attacks on other groups. The Taliban, for example, engages in violent competition with local militants, such as the National Uprising Movement, and IS-KP. In contrast, organizations such as Abu Sayyaf in the Philippines and al-Shabaab in Somalia launch relatively few attacks on other groups, despite several other groups being present in both countries and conflicts.

As control variables, we include a number of group-, conflict-, and country-level factors. We outline the sources for these variables and their operationalization in the online supplement. At the group level, we control for ideology, group size, number of terrorist attacks, foreign sponsorship, territorial control, and the proportion of an organization's attacks on soft targets in the previous year. At the conflict and country level, we control for regime type, common conflict-level indicators of competition (such as the number of groups in the country-year and the change in that number from the previous year), government repression, and ethnic diversity, as well as the natural log of GDP and population size in the previous year. Table 1 provides summary statistics for the core variables in our analysis.

TABLE 1. Summary statistics

Our empirical strategy has two stages, examining the proposed relationship at the macro and micro levels. At the macro level, we conduct a cross-national examination from 1970 to 2018. The estimation relies on a series of fractional logistic regressions, examining how the occurrence and intensity of violent competition affect civilian targeting. At the micro level, we take the case of the Islamic State in Syria from 2013 to 2018, using a series of linear models with temporal and district-level fixed effects to test our expectation.

Macro-Level Results

Before turning to the empirical analyses, we present descriptive statistics on the relationship between violent competition and soft terrorism. Our data reveal important trends in violent competition. Figure 2 graphs infighting attacks and nonstate conflict fatalities from 1989 to 2018. In recent years, terrorist infighting has reached unprecedented intensity. The number of violent attacks between organizations, and the number of fatalities from these conflict episodes, is large. There is a noticeable increase in violent competition from 2011 onward. It declines after 2015, but there remain a very high number of reported attacks between terrorist organizations.

FIGURE 2. Nonstate conflict fatalities and terrorist attacks on violent nonstate actors in civil war

Conflicts between groups in Afghanistan, India, Syria, and Yemen are among the most violent. In India, for example, CPI–Maoist has hostile relations with several rival organizations, including the Tritiya Prastuti Committee and Salwa Judum, an anti-Maoist vigilante group. While these organizations do target civilians, there is a general decline in civilian targeting when they face violent competition. During periods of intense infighting, CPI–Maoist reduced its share of attacks on civilians from 50 percent to 44 percent. More generally, and across all organizations in our data, the average share of attacks on civilians is approximately 62 percent in organization-years without infighting, and 51 percent in organization-years with infighting. This decline provides descriptive support for our hypotheses. We now turn to the statistical analyses to examine the hypothesized trends systematically.

The empirical models estimate the proportion of attacks against soft civilian targets (Table 2). The standard errors are clustered on the organization to account for the nonindependence of observations within each group over time. We present identical model specifications for both measures of intergroup infighting.⁵⁶ Models 1 and 4 present the bivariate specification. Models 2 and 5 present the main specification. Models 3 and 6 include additional control variables.

TABLE 2. Macro-level results (fractional logit, GLM)

Notes: Standard errors in parentheses, clustered on organization. The “count” variable is log(count + 1). *p < .05; **p < .01; ***p < .001.

Consistent with our expectation, when organizations face violent competition from other militant groups, they are less likely to attack soft civilian targets. Figure 3 illustrates the predicted proportion of soft civilian targets against both measures of violent competition. When an organization does not face violent competition, the expected proportion of attacks on soft civilian targets is approximately 63 percent. But with violent competition this proportion shrinks by a quarter, to about 48 percent. We observe a similar trend in the intensity of competition. An organization is less likely to attack soft civilian targets when faced with intense violent competition. This corroborates our second expectation.

FIGURE 3. Predicted Proportion of attacks on soft targets

Micro-Level Results: Islamic State in Syria, 2013–2018

The macro-level analysis provides initial support for our theoretical expectation. However, it mainly captures aggregate differences in targeting strategies between groups. It does not allow an examination of a specific organization at various points in time and space. To examine terrorist restraint under episodes of violent competition, we turn to a micro-level analysis on the Islamic State of Iraq and al-Sham (ISIS) in Syria from 2013 to 2018.

Emerging from the remnants of al-Qaeda in Iraq, ISIS launched its operation in Syria under the leadership of Abu Bakr al-Baghdadi. ISIS in Syria is an appropriate case to examine for several reasons. In the period under examination, the organization navigated an ambitious governance project across Syria.⁵⁷ It sought to control “all aspects of life within its territory” and justified violence through a rigid interpretation of sharī’ah law.⁵⁸ At the height of its insurgency, ISIS controlled territory equivalent to the size of Portugal and “governed the lives of eight million civilians.”⁵⁹ Attacks on soft civilian targets accounted for approximately 70 percent of the organization's violence.

Beyond its relationship with civilians, ISIS faced competition from other militant groups in Syria, such as Jabhat al-Nustra, Ahrar al-Sham, Hayat Tahrir al-Sham, Jaysh al-Islam, and Hurras ad-Deen. As a conflict environment, Syria was particularly prodigious, with many organizations and intergroup interactions, including both violent rivalries and cooperation.⁶⁰ ISIS in Syria therefore presents an extreme case. Given high values of the dependent variable, the underlying causal mechanisms, even when considered alone, should strongly determine a particular outcome as a paradigmatic example of the process.⁶¹ It is also, arguably, a hard case. ISIS's virulent bellicosity is well documented.⁶² It seems counterintuitive to expect the organization to demonstrate restraint and decrease violence toward civilians at any given time.

Figure 4 illustrates the timing of attacks by ISIS across fourteen governorates (sixty-five districts) in Syria. The proportion of soft civilian targets is indicated on the left, and the number of infighting attacks on the right. ISIS faced competition in seventeen districts in Syria, though its extent varied considerably across districts.⁶³ The most intense periods of infighting were in Mount Simeon, Deir Ez-Zor, and Raqqa. ISIS launched relatively few attacks on other organizations in Damascus and Homs, despite these districts seeing a lot of violence. No infighting occurred in districts such as As-Suwayda and Mayadin. Based on our expectations, ISIS should decrease its proportion of attacks on soft civilian targets where it faces violent competition, across space and time. The significant constraints on group resources and increased dependence on civilians should exacerbate the costs of backlash during periods of intense competition, leading to restraint.

FIGURE 4. Proportion of ISIS attacks on soft targets and infighting

To examine changes in ISIS's targeting preferences, we use highly disaggregated, granular data on ISIS in Syria between 2013 and 2018. The data are from the GTD. We estimate a linear model of the proportion of attacks on soft civilian targets by ISIS against the presence of violent competition.⁶⁴ The unit of analysis is the district-month.⁶⁵ We control for ISIS battle deaths, terrorist attacks by ISIS, territorial control, the provision of local services (such as tax, Islamic education, municipal services, police, courts, and welfare), and the proportion of attacks on soft targets in the previous year.⁶⁶ We specify the models with district and year fixed effects.⁶⁷ The results are displayed in Table 3. The standard errors are clustered by district.

TABLE 3. District-month analysis on ISIS in Syria (linear model)

Notes: Standard errors are clustered by district and reported in parentheses. *p < .05; **p < .01; ***p < .001.

Consistent with our expectation, ISIS is less likely to attack soft civilian targets when faced with violent competition (Figure 5). In district-months where ISIS does not face violent competition, the expected proportion of attacks on soft civilian targets is approximately 60 percent. With violent competition, the proportion shrinks by three-quarters, to approximately 15 percent. During episodes of infighting, therefore, ISIS maintains a strategy of restraint and reduces its attacks on soft civilian targets.

FIGURE 5. Predicted proportion of ISIS attacks on soft targets

A strategy of restraint is advantageous in times of competition because the need to guarantee resources and support outweighs the possible benefits of escalation. Moreover, the risk of backlash is more intense in periods of violent competition. This is evident in ISIS's terrorist strategy across Syria. To briefly illustrate the causal mechanisms at play, we draw on primary and secondary source materials on the activities of ISIS and other conflict actors in Syria.⁶⁸

In 2013, ISIS routinely executed civilians in public for straying from Sunni Islam and breaking its stringent interpretation of sharī’ah law. Zelin, for example, details the execution of three Alawites in the main square of Deir al-Zour and the killing of two young boys from Zahra and Nubl in Aleppo governorate for alleged heresy.⁶⁹ Alongside its egregious violence, the organization attempted to push a social program. The initial determination was, however, met with popular backlash, including large anti-ISIS demonstrations in Raqqa and Aleppo.⁷⁰ The organization became particularly wary of civilian backlash during episodes of rhetorical and violent infighting with other organizations. ISIS's excessive use of force was criticized by rival organizations, such as Jabhat al-Nusra, and other key jihadi ideologues.⁷¹ During intense periods of competition with the Supreme Military Council, ISIS spokesperson Abu Muhammed al-Adnani warned, “Everyone who belongs to [the Council] is a legitimate target for us,” while leader al-Baghdadi extended an arm of recognition to civilians, declaring violence was only to be used against “official” Alawis and Shi'a targets.⁷²

Moreover, in the face of violent battles with the Syrian Revolutionaries Front, Jaysh al-Mujahidin, and the Islamic Front in 2014, ISIS demonstrated considerable restraint to guarantee civilian support and promoted “good governance” to decrease the risk of a coordinated backlash.⁷³ Across its strongholds in Raqqa and Aleppo, the organization focused heavily on da'wah (invitation to embrace Islam) outreach initiatives, opening complaint offices and establishing Islamic Services Committees to oversee electricity supply, healthcare, education, street cleaning, and food provision.⁷⁴ Thus, during episodes of intergroup infighting, ISIS sought to mark clear delineations between legitimate combatants and illegitimate soft civilian targets, concentrating on maintaining popular support. Aside from this focus on civilian backlash and preserving support, ISIS also sought to secure resources and received a number of pledges from local tribes, such as the al-‘Umur and al-Mawali tribes in Badiyya.⁷⁵

Overall, the case of ISIS in Syria nicely illustrates the implementation of a strategy of restraint during periods of violent competition. In places and times where the organization faces competition, it is less likely to attack soft civilian targets. Taken together, the results support our argument. The results speak to existing theories on terrorist strategies in civil war,⁷⁶ but offer a significant departure from outbidding dynamics. As opposed to the escalation of violence, we argue and show that organizations display restraint and reduce attacks on soft civilian targets when faced with violent competition.

Robustness

The results are robust to several alternative empirical specifications and estimations. These tests and their results are outlined in the online supplement.

For the macro-level analysis, we check the results’ robustness to an ordinal measure of infighting (Table A6), a linear model specification with temporal, group, and country-level fixed effects (Table A7), alternative dependent variables focusing on the incidence and number of attacks on soft civilian targets (Tables A8–10), the exclusion of military targets (Table A11), and a reduced sample without Syria (Table A12). We also check the results’ robustness to alternative measures of violent competition independent of terrorism, using the MGAR data set (Table A13)⁷⁷ and models accounting for the brutality of competitors and ideological partners (Tables A14–15), ethnic and religious claims (Table A16), foreign support (Tables A19–24), previous violence (Table A25), battlefield dynamics (Table A28), nonviolent mobilization (Table A29), and organization strength (Tables A30–31).

For the micro-level analysis, we run the main models with a GEE and GLM estimation strategy (Tables A32–33), a count of infighting (Table A34), the exclusion of military targets (Table A35), monthly temporal fixed effects (Table A36), and an alternative dependent variable focusing on counts of attacks on soft civilian targets (Table A37). We also explore whether the brutality of ISIS's rivals matters (Tables A40–42). All results are consistent with those reported earlier. To probe whether audiences matter, we use placebo tests distinguishing between domestic and foreign civilian targets. Consistent with our argument, violent competition only decreases violence against domestic civilian targets (Table A38).

Extensions

In addition to testing the robustness of our results, our data permit a variety of additional tests which further illustrate the effects of violent competition on terrorism.

In-Group and Out-Group Dynamics

Scholarly work often associates in-group and out-group dynamics with rebel incentives to target civilians.⁷⁸ However, the incentive to target civilians may be mitigated by the simultaneous occurrence of violent inter-rebel competition. Our argument suggests that such a mitigating effect should be stronger where an organization shares the same audience as their rival. If organizations can direct attacks toward out-group civilians, the legitimacy and reputation costs of violence may be lower. We explore this possibility in Table A39 in the online supplement. Using the case of ISIS in Syria, we identify the likely ethnic identity of targets based on the ethnic composition of districts in disaggregated geo-spatial data.⁷⁹ The results, displayed in Figure 6, indicate that in locations inhabited by the ISIS in-group (Sunni Arabs), violent competition is associated with a marginally lower proportion of attacks on soft targets relative to out-group and ethnically mixed locations.

FIGURE 6. Predicted proportion of ISIS attacks on soft targets

We also consider the identity of ISIS competitors in each location, using data from the UCDP Non-State Conflict Dataset (Table A42). We find that ISIS reduces its attacks on civilians (measured here by civilian deaths) in locations where it shares an audience with rivals. But where audiences differ, ISIS increases its attacks on civilians. While the micro-level analysis offers a more accurate test of our audience expectation, we also consider it at the macro level. We include tests for organizations with and without ethnic constituencies and control for organizations that share the same audience as their rival (Tables A17–18). The results mirror those already presented.

Brutality of Competitors

The decision to decrease attacks on civilians is likely shaped by the behavior of rival organizations. When violence against civilians becomes routine, it is reasonable to believe that the costs of perpetration are much lower than under normal circumstances. However, the lower costs of civilian victimization do not explain how violence can benefit perpetrators for the purpose of outbidding, because at this point a group can no longer distinguish itself from its competitors. In other words, when civilian violence becomes routine, the motivation for continuing to use it is unlikely to be related to outbidding. Instead, we argue, when there are high levels of civilian victimization and violent competition, an organization may still have incentives to become more restrained to gain an edge over violent rivals.

We examine this empirically in the case of ISIS in Syria using data on the intensity of infighting and rivals’ attacks on civilians (Table A41). The results, illustrated in Figure 7, indicate that in district-months where ISIS faces intense infighting and where rivals launch large attacks on civilians, the organization decreases its attacks on civilians. This test highlights that, even in conditions where rivals victimize civilians, there is still an incentive to demonstrate restraint.

FIGURE 7. Substantive effect of interaction

Civilian Support

Violent competition with other groups adds costs associated with civilian victimization, with implications for the targeting strategies nonstate actors employ in the context of the civil war. It is plausible that groups who are less reliant on civilians may be less affected, but, at least in the short term, we argue all groups will suffer the increased resource pressures stemming from fighting on multiple fronts simultaneously. Therefore, we expect all organizations to decrease attacks on civilians when faced with violent competition, even those that do not strongly rely on local support. We probe this expectation at a macro level and run several models accounting for the existence of external support (Table A19) as well as the type of external support (Tables A20–24). We find no evidence that an organization's reliance on foreign support alters the decrease in civilian targeting when faced with violent competition.

Substitutable Tactics

While our argument focuses on substitution within violence, it is important to consider possible substitution between violence and nonviolence. For example, armed groups could divert resources away from governance activities to sustain their violent conflict with a rival without reducing attacks on civilians. Still, there are reasons to suspect that groups heavily invested in governance activities are reluctant to forgo them in the attempt to save resources and step up violence.⁸⁰ There is evidence that rebels who govern may be particularly concerned with building legitimacy and proving themselves as more capable rulers than the incumbent government.⁸¹ Renouncing governance could damage a group's perceived legitimacy and undo the organization's previous and costly efforts.

We investigate this substitution empirically.⁸² At a macro level, we do not find a statistically significant difference between governing rebels and nongoverning rebels based on whether they face any violent competition (Table A26). Only when violent competition increases in intensity do we find some evidence that governing rebels become even more restrained. At a micro level, in the case of ISIS in Syria, we can investigate this descriptively. Service provision occurs in 15 percent (7 of 48) of districts without infighting; yet it occurs in 65 percent (11 of 17) of districts with infighting. This suggests the tactics are not entirely substitutable and ISIS does not appear to reduce governance efforts when engaged in infighting. Moreover, when looking at attack patterns, infighting reduces not only the proportion but also the absolute number of attacks on soft civilian targets (Table A37). That is, ISIS becomes more restrained in both relative and absolute terms. The substitution we observe is within terrorist violence rather than between violent and nonviolent tactics.

Conclusion

We have examined the effects of violent intergroup competition on terrorist violence against soft civilian targets. We have argued that, when militant groups are engaged in violent conflict with one another, their propensity to target civilians relative to harder targets decreases. In the context of civil conflicts, where militant groups fight against the government and are especially dependent on civilian support, the opening of a second front against another nonstate actor increases a militant group's dependence on civilians for resources and incentivizes restraint toward them. Our argument highlights the limitations of outbidding strategies in this context. Once a militant group shifts toward destructive competition, aimed at physically eliminating rival groups, escalation of terrorist brutality toward civilians becomes counterproductive, but also unnecessary.

Our research contributes to a growing body of work seeking to unpack the causes and consequences of cooperation and conflict between nonstate actors. Future studies could build on this research in several ways.⁸³ First, future work could extend our focus on the quality of competition beyond violence and consider how nonviolent interactions between militant groups, such as rhetorical and ideological competition, affect their behavior, tactics, and targeting strategies. Recent data sets that disaggregate various forms of violent and nonviolent competition,⁸⁴ for example, can open up new avenues to study patterns of complementarity and substitution and enable a more explicit comparison of the effects of nonviolent/rhetorical versus violent competition. Second, and relatedly, future research could extend the data collected to investigate violent competition outside of terrorism users. While we believe focusing on groups that use terrorism provides a hard case for our theory of restraint, it is important to examine the effect of competition on groups that do not already engage in terrorism, as well as outside civil war contexts. Third, our micro-level analysis reveals important in-group and out-group dynamics in the effects of violent competition. However, we have only scratched the surface, and we welcome studies of the consequences of inter- and intra-ethnic competition more broadly. Finally, considering the multi-party nature of many contemporary civil wars, there is much work to be done to better understand how conflict (and cooperation) between rebels affect the dynamics and outcomes of civil wars.