Burnings, Beatings, and Bombings: Disaggregating Anti-Christian Violence in Egypt, 2013–2018

Abstract

What are the determinants of ethnic violence? Existing research has forwarded a range of often competing explanations, from political opportunism to economic competition to state incapacity. We argue that this diversity of accounts is attributable, in part, to scholars’ tendency to lump together distinct forms of ethnic violence that have different underlying determinants. We propose that scholars instead disaggregate ethnic violence and put forward a typology based on the target of the attack (properties vs. individuals) and whether assailants use arms. We demonstrate the utility of this typology by applying it to an original dataset of ethnic attacks against Christians in Egypt from 2013 to 2018. In addition to a set of shared factors, we find that unarmed attacks against property (“burnings”) are the product of political mobilization, unarmed attacks against individuals (“beatings”) are related to socioeconomic tensions, and armed attacks (“bombings”) follow the strategic logic of terrorist violence.

On August 14, 2013, followers of the Muslim Brotherhood and other Islamist political movements began marching from the main square in the Upper Egyptian city of Minya chanting against Christians, whom they accused of supporting the recent military coup against the Islamist president Mohamed Morsi. The marchers made their way to some of the city’s most prominent Christian sites, which they attacked with clubs and stones, looted, and ultimately set ablaze. Three years later, in a village located in Minya’s rural periphery, a Coptic man, Talaat Beshay Gad al-Rab, was attacked by a small group of Muslim residents. His 25-year-old son Shenouda was allegedly having a relationship with a young Muslim woman in the village and had been accused by her family of kidnapping her. Even though she denied that she had been kidnapped and a local court had acquitted the son of any wrongdoing, the woman’s family members and friends set out to avenge the supposed affront by attacking Shenouda’s father. Finally, in May 2017, on a desert road northwest of Minya, gunmen ambushed a convoy of Christian pilgrims headed to a monastery. One of the vehicles was a minibus carrying children and women, which the attackers sprayed with bullets. They then entered one of the buses, stole money and jewelry from the women, and shot 28 people. Jihadi leaflets were left strewn over the bodies, and the next day the Egyptian franchise of the Islamic State claimed credit for the attack.

The commonalities between these three events in Minya are clear: all were attacks perpetrated by nonstate actors against Christian civilians. But the attacks also have several meaningful differences. The first event was a politically motivated pogrom by Islamists looking to inflict collective punishment against the entire Christian community by targeting their places of worship. The second event was a physical attack seeking retribution against a single individual for a perceived affront to cultural mores. The third event was a preplanned operation executed by an organized and armed transnational terrorist group. Although all three events clearly fall under the category of ethnic violence, in this article we argue that the differences in their characteristics are analytically important and too often have been overlooked by scholars of ethnic politics.

Specifically, we propose that effectively explaining ethnic violence requires disaggregating the phenomenon into its distinct subtypes, which we term “burnings,” “beatings,” and “bombings.” We argue that this disaggregation has the potential to significantly advance knowledge accumulation on ethnic violence by allowing scholars to align theories and arguments to the types of violence that they are best able to explain. As we show in our literature review, the extensive scholarship on ethnic violence lacks coherence: it advances a wide range of explanations that are often in tension or difficult to reconcile. This theoretical muddle is partly attributable to the fact that scholars are analyzing ethnic violence at too high a level of aggregation, lumping together forms of violence that have different underlying determinants. A more disaggregated approach to studying ethnic violence allows us to reconcile these competing explanations and align them to specific outcomes. Indeed, conceptual disaggregation has brought important theoretical clarity to adjacent fields like civil war studies, terrorism studies, and contentious politics research. We propose that a similar shift is overdue in ethnic violence research; in fact, leading scholars in the field have been calling for precisely such a move for some time.

Our disaggregation framework is based on two criteria: the target of the attack (properties vs. individuals) and whether assailants use firearms (armed vs. unarmed). Burnings are unarmed attacks against properties, beatings are unarmed attacks against individuals, and bombings are armed attacks against either type of target. Although we do not claim that ours is the only reasonable way by which scholars might disaggregate ethnic violence, we believe our typology is effective because (1) it captures meaningful differences in the forms and motivations of violence and (2) it is straightforward to apply in the coding of quantitative event data.

We further argue that these three types of violence are likely to be associated with different underlying conditions, which align with various arguments in the scholarship. We expect burnings to be associated with political dynamics, because they often entail the mobilization of mobs or gangs by political entrepreneurs seeking to use ethnic violence for political gain. Conversely, we expect beatings to be associated with changing socioeconomic conditions, which tend to heighten local animosities between non–co-ethnic neighbors and increase the likelihood of individually targeted punitive attacks. And we expect bombings to follow the logic of terrorist violence, because these attacks require that an organized group furnish advanced firearms. Finally, because the state’s security forces might protect ethnic minorities from all manner of violence, we expect weakened coercive capacity or deliberate inaction to be associated with all three types of attacks.

To demonstrate the utility of our framework and to show that these three types of violence do indeed follow distinct logics, we analyze a novel dataset of recent anti-Christian ethnic violence in Egypt. Following the broader ethnic politics literature, we consider as “ethnic” all identities based on descent-based characteristics, including religion, language, and race (Chandra 2006; Horowitz 1985). ¹ Egyptian Copts have long been persecuted both by the state and by members of Egypt’s Muslim community (Brownlee 2013; Mahmood 2015). Our dataset comprises 413 anti-Christian attacks in Egypt from July 2013 to December 2018 sourced from local Arabic- and English-language websites, newspapers, human rights reports, and social media posts. Further, to operationalize major theories of ethnic violence, we pair this dataset with a range of time-varying and subnational variables.

In the first empirical section of this article, we provide descriptive and qualitative evidence that ethnic violence in Egypt over the five-year period from 2013 to 2018 emerged from multiple distinct processes, including political mobilization, socioeconomic tensions, terrorist organizations, and state inaction. In the second empirical section we demonstrate statistically that these explanations are also differentially associated with our three violence types, largely in the ways we predict. We find that burnings occur more frequently in districts where Islamists are highly mobilized and follow episodes of major government repression. Beatings, however, occur during periods of economic strain and are patterned by the religious calendar. Further, both burnings and beatings tend to cluster in districts where police officers had recently been on strike, suggesting that the state’s coercive capacity shapes these types of violence. Finally, bombings occur more often in symbolically significant city centers (which are often the targets of terrorist attacks) and in areas close to the home base of major terrorist groups; their incidence also decreases on Muslim holidays.

Although Egypt represents only one empirical “test case” for our typology, the findings clearly support our main claim that different types of ethnic attacks are associated with distinct conditions. As a result, generating parsimonious theoretical explanations for ethnic violence likely requires disaggregating the phenomenon into its more specific manifestations. Moreover, our analysis demonstrates the potential utility of one disaggregation approach while also welcoming the possibility that alternative typologies may accomplish similar goals.

Conflicting Explanations in Ethnic Violence Studies

There is a well-developed scholarship that seeks to make sense of why violence occurs between different ethnic groups. It generally sees ethnic violence as a particular type of conflict defined by two characteristics. First, differences in ethnic identity are a primary motivation for the violent action (e.g., Brubaker and Laitin 1998: 428). Second, the violence is one-sided in that it targets everyday civilians who are not organized into an armed or political group. These criteria distinguish ethnic violence from other forms of conflict, including civil wars, rebellions, insurgencies, or criminal violence. ² These two criteria further point scholars toward a fundamental problem that motivates much of the literature: What are the conditions under which tensions between members of rival ethnic groups spill over into overt violence? To answer this question, the literature has forwarded a range of explanations that can be broadly grouped into three categories: those focused on political competition, those stressing social and economic relations, and those centered on the state and its coercive apparatus.

A first cluster of explanations proposes that ethnic violence has a political calculus. Violence may be an effective means to demobilize opponents, rally supporters, and elevate the salience of ethnic differences ahead of elections (Dhattiwala and Biggs 2012; Toha 2017; Wilkinson 2006). It may also occur when ethnic groups begin to develop autonomous political power through superior organizational or mobilizational skills (Dancygier 2010). Often these explanations focus on the mobilizational role of political entrepreneurs who use ethnically charged speech to heighten animosities between groups, provide explicit or tacit encouragement to assailants, or even directly organize followers to participate in attacks (Hardin 1997; Karapin 2002; Wilkinson 2006). These accounts often find that ethnic violence occurs around important political events, such as elections or government changes. It may also occur as “backlash” (Francisco 2004) following major acts of state repression, whereby political actors target outgroups aligned with the state whom they scapegoat for the crackdown. These backlashes can also occur when groups have been excluded from democratization processes, which they attribute to the complicity of an ethnic minority (Brooke and Nugent 2020).

A second cluster of explanations examines society-level relations between ethnic groups. An influential variant of this argument emphasizes the importance of deeply embedded identity dynamics within episodes of ethnic violence (Horowitz 1985, 2001; Petersen 2002); a related strand has looked more at the structure of institutions or relations in society that might serve to ameliorate these types of tensions (Fearon and Laitin 1996; Kopstein and Wittenberg 2011). For example, Varshney’s (2003) study of anti-Muslim riots in India shows that cross-ethnic associational ties tend to mitigate violence (see also Jha 2013). If these studies emphasize the importance of economic institutions for reducing tensions, other scholars have pointed to the centrality of economic competition between ethnic groups (Dancygier 2010; Esteban and Ray 2011; Olzak 1992; Sniderman et al. 2002).

The third cluster of arguments focuses on the institution supposedly responsible for restraining or curtailing ethnic violence: the state. State coercive actors, like the police, are entrusted with protecting ethnic minorities from predation by majority groups. But in contexts where state capacity is weak, coercive agents may lack the means to effectively restrain attackers (Kalyvas 2006; Petersen 2002). In other cases, coercive agents may lack the will to enforce order, either because they have a low level of commitment to their jobs, they harbor ethnic animosities of their own, or they see disorder as a means to stoke political polarization (Karapin 2002; Wilkinson 2006).

Of course, it is not surprising that research on a complex political phenomenon like ethnic violence has identified no single explanatory variable. Still, the range and diversity of arguments in the literature are striking. Some of these explanations generate predictions that are directly contradictory. For example, some studies find that strong ethnic leaders discourage violence to preserve intercommunal harmony (Fearon and Laitin 1996), whereas others argue that these same leaders encourage violence to achieve political gain (Wilkinson 2006). Although differences in findings could stem from analysis of disparate cases, in some instances rival explanations emerge from examination of the same case and occasionally even the very same dataset. The most famous example is the debate between Wilkinson (2006) and Varshney (2003), who came to starkly different conclusions after analyzing the same data on ethnic riots in India—the former privileging political and state-centric factors, the latter preferring a societal explanation.

We believe that such diverse and contradictory accounts are indicative of a conceptual problem in this literature. Following Sartori (1970) who cautioned against conceptual lumping and stretching, we propose that scholars are studying ethnic violence at too high a level of aggregation, grouping together different types of violence that in fact have distinct underlying processes, even if the targets share a common ethnic hue. In other words, even though narrowing the scope of analysis to specifically ethnic violence has allowed scholars to focus on a common problem—the conditions under which ethnic tensions escalate into conflict—there is still too much diversity within this category of “ethnic violence” to yield coherent and parsimonious theory.

A brief survey of the different forms of action that scholars have analyzed under the heading of ethnic violence confirms this intuition. Many studies use the “ethnic riot” as the unit of analysis, typically defined as an attack by one ethnic group against another (Horowitz 2001; Toha 2017; Varshney 2003; Wilkinson 2006). Others consider the phenomenon of pogroms (Dhatiwalla and Biggs 2012; Kopstein and Wittenberg 2011), presumably more deliberate or organized campaigns to inflict mass casualties on a minority group (though the distinction between riots and pogroms is not clear). Some scholars lump together different forms of violence in the same study. For example, Olzak (1992, 6) studies “ethnic collective action,” which she defines as “a public action of two or more persons that articulates a distinctly ethnic (or racial) claim, expresses a grievance, or attacks members of another ethic group (or property).” Here, then, we have action that is nonviolent (e.g., a protest) and violent (e.g., a riot), targeted against both individuals and properties, all bundled together. Following Olzak, Varshney (2003) and Wilkinson (2006) define an ethnic riot as an event involving violence in which two or more communally identified groups confront each other—a definition that encompasses large- and small-scale attacks against both individuals and properties from the opposing group, using various levels of arms (see Wilkinson 2006: 255–56). Dancygier (2010) similarly groups together various forms of immigrant-related violence, including relatively isolated racist attacks by one or several individuals and large-scale riots aimed at physical property destruction. Given the diversity in these forms of ethnic violence—from isolated attacks against a single victim, to mass pogroms, to well-planned armed attacks—it is no wonder that the literature has come up with so many competing arguments.

A handful of studies do note differences in the forms of ethnic violence. For example, Karapin’s (2002) article on antiminority violence in Germany distinguishes between “small-scale violence” and larger “antiminority riots,” interrogating why the former sometimes telescope into the latter. Paul Brass (1997) asks a similar question in India: Why do small parochial incidents (like “the theft of an idol”) sometimes escalate into large-scale riots or pogroms? But these studies do not develop full typologies of violence, nor do they systematize these differences by applying them to empirical data. Nor have they prompted the scholarship more broadly to embrace disaggregation in their studies. We find this puzzling, because prominent voices in the ethnic violence field have been calling for disaggregation for quite some time. For example, Brubaker and Laitin (1998, 446) cautioned against defining a “research program around an aggregated notion of ethnic violence.” In their “plea for disaggregation,” they argue that ethnic violence is “composite and causally heterogeneous” and “involve[s] sharply opposed mechanisms and dynamics” (446). As such, we should forego attempts at elaborating universal theories of such violence and instead “analyze, and explain the heterogenous processes and mechanisms involved in generating the varied instances of what we all too casually lump together—given our prevailing ethnicizing interpretive frames—as ‘ethnic violence’” (447). Horowitz (2001, 27–28) has made a similar point: it will “not be possible to know which conditions are common and which are distinctive if we begin with a dependent variable based on merged phenomena.”

Further, we note that there has been considerable theoretical payoff from disaggregation in adjacent literatures. Recognizing the limitations of country-level analyses of civil wars, scholars in this field have embraced disaggregation at the level of ethnic groups, spatial units of analysis, and (most analogous to our study) characteristics of conflict (Cederman and Gleditsch 2009). For example, Kalyvas (2006) argues that in civil wars discriminate and indiscriminate violence follow different logics and require different theories: the latter is a function of power asymmetries and information/resource scarcity, whereas the former occurs through a process of denunciation and counter-denunciation. Similarly, Balcells (2017) argues that indirect violence against civilians in civil wars follows a military logic, whereas direct violence occurs in collaboration with local civilians who seek to consolidate political control. Disaggregation has also been embraced in the contentious politics field (e.g., McAdam et al. 2001). For example, Bishara (2021) argues that certain forms of contention are more capable of deepening and broadening movements than others.

With these successful examples in mind, we propose it is high time to answer the calls made by Brubaker, Laitin, Horowitz, and others. Even though the scholarship on ethnic violence has made significant headway in unearthing the factors that spur this form of conflict, the coherence of its contributions has been muddied by analysis of the phenomenon at too high a level of aggregation. We therefore propose moving one rung down Sartori’s proverbial ladder— essentially, to paraphrase Horowitz, “un-merging” the dependent variable—which we believe will allow for better alignment of theory with outcomes. In the next section we advance one approach by which scholars might go about such a disaggregation.

Disaggregating Ethnic Violence

In this section we propose a typology for disaggregating ethnic violence, which, following the existing literature, we understand as one-sided violence against civilians in which ethnic differences are the main motivation for action. The typology is based on two criteria: the target of the violence and the level of arms used (figure 1). We select these criteria because they are both substantively meaningful—they capture important differences in the forms and functions of violence, including the aims of that violence and its level of organization, and empirically tractable: they are easy to apply based on even cursory or limited information about an event.

Figure 1 A typology of ethnic violence

We begin by differentiating attacks by their target. Those against properties owned by members of another ethnic community (like arson or looting) we term “burnings,” and those against individuals from another ethnic group (like assaults or kidnappings) we term “beatings.” The target of an attack says something important about the motivations of the attackers. An attack against a specific individual or group of individuals likely has a retributive rationale; it entails that ethnic hostilities are narrow and focused and that the intention of the violence is to punish an individual or group of individuals for some supposed wrongdoing. In contrast, an attack against a physical property associated with an ethnic group is more likely to be motivated by a desire to inflict harm on the entire community or population, rather than some specific individuals. Such physical properties typically have symbolic significance, particularly if they are cultural sites or places of worship, and their desecration or destruction is often undertaken with the aim of collective punishment. In this sense, by distinguishing between the target of ethnic attacks we capture meaningful differences in whom they are meant to punish: the collective or the individual.

The second axis of disaggregation is the level of violence used. Drawing on Tilly’s (2003) work on collective violence, we distinguish between armed violence, which inflicts damage on an ethnic group using firearms or explosives, and unarmed violence, which inflicts harm using improvised weapons (e.g., clubs, knives, or fists). We refer to armed attacks as “bombings,” which captures both attacks using bombs and those using guns, such as shootings or assassinations. ³ We believe this distinction is important because it captures important differences in the perpetrator’s level of organization. Indeed, Kadivar and Ketchley (2018) point out that armed and unarmed attacks are generally undertaken by markedly different types of actors. In the context of ethnic violence, we expect that armed attacks require some sponsoring organization to furnish the weapons and bombs, which are not readily available to most civilians. In contrast, unarmed attacks will, by and large, be perpetrated by less organized actors, who draw on improvised weapons like sticks, clubs, stones, knives, and fists. Further, in contrast to armed violence, which requires some level of prior planning to procure and distribute the weapons, unarmed violence can emerge rapidly and spontaneously, because assailants rely on whatever is readily at hand.

Importantly, for reasons of parsimony, we choose not to subdivide armed violence according to the target of the attack (as we do with unarmed violence). We believe that the main relevant factor here is whether arms are used or not, which implies a certain level of organizational backing. Although in some cases armed attacks against individuals (e.g., assassinations) versus against buildings (e.g., explosions) could operate according to different logics—perhaps mirroring the difference between discriminate and indiscriminate violence in the civil wars or terrorism scholarship—here we choose to maintain a simpler three-part typology. ⁴

There are other ways that one might disaggregate ethnic violence. For example, per the previous point, armed attacks could be subdivided according to target, creating a four-part typology. Another obvious move would be to differentiate according to the perpetrator of the violence: a political party, a rebel group, unorganized civilians, and so on. Or, one could imagine operationalizing the scale of violence with a measure of the number of participants involved. The problem with the last two criteria is that they can be difficult to implement in the actual coding of cases. It is often unclear whether a particular group or sponsor is behind an attack, because they may want to keep this information hidden. ⁵ We also found in our data that the number of participants in an attack is often omitted or unreliable in the reporting of the incident. Furthermore, we believe that our categorization scheme captures some of the most important differences in perpetrator and scale. We have already discussed how the level of violence (armed vs. unarmed) tells us something about the level of organization behind the attack, because the arms must be furnished by someone. As for scale, it is fair to assume that attacks against a single person or a small number of individuals will involve fewer assailants, whereas unarmed attacks against properties require the mobilization of large mobs or crowds, who can surround a building and overwhelm security forces.

Because our categorization scheme captures meaningful differences in the motivations for and the organization of violence, we believe that these different types of attacks will be generated by distinct processes. Specifically, we argue that each of the different theories of ethnic violence reviewed here will be better able to explain some forms of violence than others.

First, we expect burnings to be associated with political factors. Assailants target the buildings of an ethnic community when they want to inflict collective punishment on the whole group. Such acts of collective victimization are likely to be particularly effective in heightening ethnic cleavages and driving a wedge between entire ethnic communities, which the literature tells us is often the goal of political entrepreneurs. Moreover, burnings usually require a degree of collective action, with mobs or groups of assailants engaging in looting or arson. They therefore require someone to mobilize these mobs and spur them into action, and political entrepreneurs can provide this type of mobilizational capacity. We therefore believe that burnings will be more likely to occur where political entrepreneurs are more powerful and when they have a strong motivation to incite ethnic attacks. For example, they might use violence to exacerbate ethnic cleavages and gain a political edge in places where political control is especially contested. Or, following the literature on backlashes (Francisco 2004), they might incite attacks in response to a major state crackdown, particularly if non-co-ethnics are perceived to be in alliance with the state. We propose the following hypotheses:

• • H1a: Burnings are more likely where political entrepreneurs and their affiliated organizations are stronger.

• • H1b: Burnings are more likely in places with more political competition.

• • H1c: Burnings are more likely following major acts of state repression.

Next, we expect beatings to be more closely associated with socioeconomic factors. Because beatings entail unarmed attacks against specific individuals, they usually have a punitive rationale; they are intended to redress a particular grievance against an individual or group of individuals for some perceived affront. In the literature, social and economic factors, such as economic competition, are expected to generate violence precisely because they increase intercommunal tensions and make these types of perceived affronts more likely. Residents will scapegoat and blame their non-co-ethnic neighbors when material hardships and competition increase over scarce resources. Further, such intercommunal grievances and tensions are likely to be heightened during religious holidays, when cultural sensitivities are more acute. Finally, following Varshney (2003), we would expect these kinds of social and cultural tensions to become explosive in places where intercommunal relations and networks between ethnic groups are particularly weak. We therefore derive the following three hypotheses:

• • H2a: Beatings are more likely during periods of economic competition.

• • H2b: Beatings are more likely during religious holidays.

• • H2c: Beatings are more likely where cross-ethnic intercommunal relations are weak.

Bombings require that an organized group acquire and furnish the firearms and explosives used. Though we might envision various organized groups—such as state actors, rebel groups, and so on—providing these arms, one obvious candidate would be terrorist groups, particularly those organized around identity or religion. Terrorism has often been analyzed separately from ethnic violence, but many terrorist groups do target ethnic minorities and are connected to long-term dynamics of interethnic hostilities (Sidel 2006). As the literature on terrorism has shown, these groups tend to use violence in a highly strategic way (Byman 1998; Crenshaw 1981; McCormick 2003). They tend to attack symbolically important sites with spectacular acts of violence intended to maximize the number of civilian casualties; these acts serve simultaneously to incite fear in minority communities, increase the exposure of the terrorist group, and radicalize and mobilize potential supporters. Moreover, scholars have found that Islamist terrorists reduce their attacks during Muslim holidays due to elevated levels of societal disapproval of violence (Reese, Ruby, and Pape 2017). We expect bombings to be patterned by similar dynamics. Specifically, we derive the following hypotheses:

• • H3a: Bombings are more likely where terrorist groups have an organizational presence.

• • H3b: Bombings are more likely in highly visible and symbolically significant locations.

• • H3c: Bombings are less likely during Muslim religious holidays.

Finally, because the state might plausibly protect ethnic groups from any type of attack, we expect that when state actors are either unable or unwilling to do their jobs it may facilitate all three types of violence (Wilkinson 2006). This allows us to derive our final hypothesis:

• • H4: Burnings, beatings, and bombings are more likely where the state’s coercive capacity is weak.

In the following sections we evaluate these hypotheses and probe the extent to which our categorization scheme captures meaningful differences in ethnic violence in an important case: Egypt from 2013 to 2018.

Ethnic Violence in Egypt, 2013–2018

As a first step toward evaluating our argument about disaggregation, we begin by descriptively and qualitatively analyzing recent trends in anti-Christian violence in Egypt. Egypt is a country with a politicized ethnic cleavage, an imbalance in ethnic power, and a history of ethnic conflict, making it fairly emblematic of ethnic politics in many postcolonial states of the Global South (Clarke 2017). Egypt’s Muslim majority is dominant politically: according to the Ethnic Power Relations dataset Muslims are a monopoly power and Christians are “powerless,” an ethnic power dynamic that characterizes 44 other states in the dataset (Cederman, Wimmer, and Min 2010). ⁶ And like many of these other states (e.g., Guatemala, Rwanda, Armenia, and Malaysia), for much of its modern history Egypt has seen recurring incidents of nonstate attacks against its Christian minority community.

We analyze anti-Christian violence in Egypt from 2013 to 2018, the period directly following Abdel Fattah al-Sisi’s counterrevolutionary coup and reimposition of authoritarianism after the 2011 revolution. As we explain in detail, this period captures both an extraordinary period of violence directly following the coup and a more “normal” period from 2014 to 2018, which resembles levels of violence observed before the revolution. We consider the inclusion of both periods in our analysis a strength, because it allows us to show more clearly that ethnic violence follows different logics in different times and takes different forms. ⁷

We analyze descriptive trends in ethnic violence over this period using our catalog of anti-Christian events, which we introduce more fully in the data and methods section. We find initial evidence that all four of the major theories laid out earlier—political, socioeconomic, terrorist, and state-driven—were operative in Egypt over this period. We further find some suggestive evidence that these theories better explain certain types of violence, though we leave the more systematic examination of this question to the next section.

Political Violence and State Inaction following the Rabaa Massacre

Christians were active participants in Egypt’s 2011 revolution, which overthrew the decades-old military autocracy headed by Hosni Mubarak (Abdelrahman 2014; Clarke 2014; El-Ghobashy 2011). Indeed, during the revolution Tahrir Square was heralded as a space of striking cross-confessional solidarity. But this sense of national unity quickly faded away after Mubarak stepped down, as Egyptians set about the messy business of constructing a new political regime. In the first months of the transition, anti-Christian attacks occurred in Helwan, Qena, and Imbaba (Brownlee 2013; Tadros 2013). Egypt’s Coptic community became increasingly concerned by the growing power of domestic Islamist forces, especially the Muslim Brotherhood. As this group swept elections and then set about drafting a new constitution, Christians came to fear that they were witnessing the emergence of a permanent Islamist hegemony in Egypt. Coptic elites were therefore some of the most active participants in the counterrevolutionary movement that paved the way for Abdel Fattah al-Sisi’s July 2013 coup, which ousted Morsi and restored military rule (Clarke 2020; Holmes 2019).

The coup set in motion a series of political events that culminated in a surge of ethnic violence against Egypt’s Christians in August 2013. We can see this spike clearly in figure 2, which plots the number of ethnic attacks per month from July 2013 to December 2018. When Morsi was ousted, the Muslim Brotherhood responded with massive street mobilization; hundreds of thousands of their followers converged on a square in front of Rabaa al-Adwiyya mosque in Cairo and swore not to leave until Morsi had been restored (Biagini and Ardovini 2022; Ketchley 2017). On August 14 the military-led government reacted with a bloody crackdown, killing approximately 1,150 people (Human Rights Watch 2014). In response, local Muslim Brotherhood branches and allied Islamist groups went on the attack. One of their targets was Christian communities, whose members they blamed for backing the coup and enabling the massacre of their supporters. Event data indicate that individuals associated with these groups were implicated in 129 attacks after the Rabaa Massacre, many of which targeted Christian churches. Moreover, though these attacks were triggered by political events in the capital, the violence was national in scale, occurring in 16 of Egypt’s 27 governorates, with a concentration in the more heavily Christian governorates of Upper Egypt (e.g., Minya, Asyut, and Beni Souef).

Figure 2 Anti-Christian attacks, July 2013–December 2018

A report published in the aftermath of these attacks by the Egyptian Initiative for Personal Rights (EIPR) provides rich qualitative insight into the central role of political entrepreneurs from the Muslim Brotherhood and other Islamist groups in orchestrating this violence. The report summarized the events as follows: “Most of the churches were set on fire by marchers supporting the deposed president, and some of the marches set out from sit-ins, as in cities in Minya, Assyout, and Sohag. In other cases, such as in cities and districts in Fayyoum, Minya, and Giza, appeals went out to attack churches from mosques known to be dominated by Islamist currents” (EIPR 2014, 95).

The EIPR report also includes firsthand accounts from individuals who witnessed the attacks. For example, the priest of the Mar Girgis Church in Minya described an assault on a Christian orphanage: “What happened is that on Wednesday, 12 August, the day they cleared the Nahda and Rabaa sit-ins, a huge group of terrorists ⁸ came and attacked the place here. As you can see, the building here was burned to the ground with everything inside it, after they took whatever they could carry. It was an organized operation, first looting and theft then burn it down” (EIPR 2014, 97). In addition to churches and other communal sites, the report also discusses the deliberate targeting of Coptic-owned homes and businesses by similar crowds of Morsi supporters.

A second important source of insight on the attacks comes from an unpublished report written by the Egypt-based researcher Steve Negus, based on field research conducted nine months after the episode (Negus 2014). It focuses on the Upper Egyptian governorate of Minya, where the three episodes discussed in this article’s introduction took place. Minya has long been an epicenter of anti-Christian violence in Egypt due to an explosive mix of demographic conditions: it is both an Islamist stronghold and yet hosts the largest proportion of Christians of any governorate in Egypt: Christians make up one-third of its population (3–5). In our dataset it is the governorate with the largest number of violent events, both during the Rabaa week and otherwise. Negus’s reporting confirms that the anti-Christian attacks in the governorate after Rabaa were instigated by supporters of Mohamed Morsi and mosque preachers aligned with the Brotherhood. Moreover, witnesses he interviewed confirmed that “the attackers were more interested in burning and in looting computers, wiring, and other valuables from the churches than in attacking people” (19). He also found that, by and large, these attacks did not involve weapons: “while some people might have brought arms to the attacks, these were crowd assaults, not armed assaults” (20).

Finally, Negus found considerable evidence that the lack of state coercive capacity, specifically police inaction, allowed the violence to spiral out of control. Local police forces in the south had been cowed by the experience of the 2011 revolution. Their disillusionment only increased when Morsi was elected president, because many police officers have long-standing animosities toward the Muslim Brotherhood. During the Rabaa violence, Negus (2014, 19) reported that many police forces lacked the confidence and resolve to confront the Islamists, holing up in their stations or abandoning them entirely. The EIPR report (2014, 105) also notes that the performance of the security forces during the attacks was “characterized by sluggishness and inaction.” The report further documents incidents in which police were not only slow to respond but also actively facilitated attacks against Christians (93).

Low-Scale Social and Cultural Attacks

In the months after the post-Rabaa attacks, sectarian violence in Egypt receded; as can be seen in figure 2 and in our online appendix, these levels resembled those during the months and years preceding the 2011 revolution. September 2013 saw only 9 attacks, October saw 12, and November and December saw 2 each. The average rate of monthly attacks over the four years following Rabaa was 6, and no month contained more than 20 attacks. There was also a meaningful qualitative shift in the types of violence that mark this period. In general, the violence was far more localized and smaller in scale, with issues like church construction, intermarriage, and religious conversion catalyzing attacks that usually targeted one or several individuals and rarely resulted in serious injuries or deaths. ⁹ For example, in January 2014, in Asyut, a brawl ensued when several Muslim boys were called out for harassing a Christian girl. In the spring of the following year, the village of al-Galaa in Samalut, Minya, saw a series of clashes and attacks against Coptic individuals in response to plans to build a new church. Another common form of attack was the abduction of Christians for ransom. For example, in September and October 2013, a number of Christian doctors and medical professionals were kidnapped and held for ransom in Minya province.

Anti-Christian Terrorism

In addition to politically motivated riots and small-scale parochial attacks, the trends in our data point to a third wave of ethnic violence that was operative in Egypt during this period. Figure 3 plots the number of monthly deaths from sectarian violence from July 2013 to December 2018. In contrast to figure 2, which showed a relatively steady rate of incidents from 2014 to 2018, this figure shows a striking increase in the intensity of violence beginning in late 2016, with spikes in deaths occurring in December 2016, April 2017, and May 2017.

Figure 3 Christian deaths from ethnic violence, July 2013–December 2018

The trend can be attributed to a wave of terrorist violence staged by the Islamist militant group Ansar Beit al-Maqdis (ABM), which was formed in the Sinai Peninsula immediately after the 2011 revolution. Militants took advantage of the collapse of the security state to essentially take control of large parts of the Sinai, and they soon began staging attacks against security forces, infrastructure, and civilians. These attacks increased in the aftermath of the 2013 coup, as the organization began to recruit more followers from among the disillusioned supporters of the Morsi government (Awad and Hashem 2015). Then in November 2014 ABM publicly declared allegiance to the Islamic State and began staging increasingly audacious attacks against Egypt’s security forces, which the group considered its primary enemy (Awad and Hashem 2015; Jumet and Gulmohamad 2020). The organization also attacked Coptic communities, bombing Coptic churches and staging armed assaults on groups of Copts.

Overall, then, this analysis suggests that, from 2013 to 2018, Egypt was racked by multiple distinct types of ethnic violence. First, after the massacre of Islamists in Rabaa Square on August 14, 2013, there was an outburst of politically motivated and state-facilitated rioting in which Morsi supporters targeted sites of communal importance to Christians. Second was the persistent targeting of Christians in low-scale attacks and kidnappings prompted often by perceived affronts to communal and social norms. Third, there was a surge in militant-sponsored attacks on Christian religious sites and communities intended to maximize casualties and incite a sectarian war. In other words, we see that all four of the major explanations for ethnic violence seem to have been operative, to some degree, in Egypt during this period. We also have some early evidence that the factors highlighted in these explanations were associated with different types of attacks (e.g., many burnings occurred during the politically motivated post-Rabaa wave of violence). To examine these relationships more systematically we turn next to a statistical analysis of these same violence patterns.

Data and Method

Our principal data source for the statistical analysis is the Eshhad catalog of anti-Christian attacks in Egypt over 2013–18 (the same dataset used to create the plots in figures 2 and 3). The catalog draws on more than 50 Arabic- and English-language news sources. ¹⁰ We cleaned these data and assigned each event to its census district. Given the potential for “news holes” (Oliver and Maney 2000) during periods of upheaval, we expanded the dataset for the intense period of violence after the Rabaa massacre using human rights reports compiled by the Egyptian Initiative for Personal Rights and Human Rights Watch. These sources yielded 54 additional events. We also removed some events from the Eshhad dataset, either because they were difficult to verify (e.g., entries sourced from Facebook posts) or because they did not meet the criteria for an ethnic attack. Ultimately, we ended up with a dataset of 413 anti-Christian attacks, which we reproduce in table C1 in the online appendix.

For each event the dataset recorded the date of its occurrence, the location (governorate and district), the number killed and injured, and the main narrative details. We used these narrative details, as well as the underlying source text, to code each event into our three ethnic violence categories. The following definitions guided our coding: burnings were defined as events in which improvised weapons were used to inflict damage on physical property, ¹¹ beatings as events in which improvised weapons were used to inflict harm on individuals, and bombings as events in which firearms or explosives were used to inflict harm on individuals or physical property. Typically, events could easily be assigned to one of these three categories. For example, the type of target that had been attacked was usually clear; a narrative would explicitly mention that a building had been mobbed, burned, or looted or that an individual person had been abducted or beaten. For those events where individuals were injured during an attack on a building, the event was still coded as a “burning” because the main target of the attack was a property. The only borderline cases (which were rare) occurred when an individual was hurt in front of or in the vicinity of a building, and it was unclear whether the building or the person was the main target of the attack. Determining whether weapons were used was even more straightforward; whenever a narrative mentioned that a firearm or bomb was used, we coded the event as a “bombing.” The only borderline cases (which, again, were rare) were those in which assailants carried weapons but did not use them or they fired their weapons into the air. As an indication of the ease with which our typology could be applied, three coders independently reached consensus on 94% of the events. For the 29 events about which coders disagreed, mostly due to ambiguities like those just noted, we used the categorization agreed on by two of the three coders.

Dependent Variable

To assess how different geographic and temporal predictors explain the three categories of ethnic violence that we laid out, let y be the dependent variable with J nominal categories coded as “2” for a district-day that witnessed a burning, “3” a district-day that witnessed a beating, and “4” for one that witnessed a bombing. The base category (set as “1”) records district-days on which no event took place. This is naturally analyzed using multinomial logistic regression, where the occurrence of sectarian violence in district (i) located in a governorate (g) on a given day (t) is predicted as

$$ \Pr \left({\mathrm{y}}_{\mathrm{igt}}\hskip0.35em =\hskip0.35em \mathrm{m}|\;{\mathrm{x}}_{\mathrm{igt}}\right)\hskip0.35em =\hskip0.35em \frac{\exp \left({\mathrm{x}}_{\mathrm{kigt}}{\unicode{x03B2}}_{\mathrm{m}}\right)}{1+{\sum}_{\mathrm{j}\hskip0.35em =\hskip0.35em 2}^{\mathrm{J}}\exp \left({\mathrm{x}}_{\mathrm{kigt}}{\unicode{x03B2}}_{\mathrm{m}}\right)}\hskip0.33em \mathrm{for}\hskip0.5em \mathrm{m}>1 $$

where m is the specific event type to be predicted, and x_k is a vector of time-varying and time-invariant independent variables measured at the district and governorate levels. Our data structure is thus a hierarchical panel, and so two-way standard errors are clustered on the governorate and district. As an early indication that our categories were capturing meaningful differences in types of violence, a Wald test confirmed that these outcome categories are statistically distinct from one another (p < 0.001) as a function of x_k.

Independent Variables

In the section, “Disaggregating Ethnic Violence,” we argued that different theories of ethnic violence may be better suited to explaining some types of attacks than others. To operationalize these various explanations, we drew on an array of data, some of them original and hand coded.

First, we used several measures to operationalize political variables, which we argued are more likely to be associated with burnings. Following H1a, we used two measures to capture the organizational and mobilizational strength of the Muslim Brotherhood in a district. After the July 2013 coup, the Brotherhood rallied its members and supporters to occupy squares and major thoroughfares around the country, where they remained until the Rabaa massacre in mid-August 2013. To capture the local mobilizational strength of the Brotherhood, we used a binary variable for whether an occupation was taking place in a district, sourced from Ketchley (2017). As a second way of capturing the Brotherhood’s organizational strength, we drew on data from the first round of the 2012 presidential election in Egypt, using the percentage of the vote won by the Brotherhood’s candidate Mohamed Morsi.

Another political explanation for ethnic violence argues that entrepreneurs use violence to politicize ethnic cleavages for electoral gain. This type of violence is expected to emerge in places where political control is more competitive (H2b). We measured political competition in a district with the percentage of the vote in the 2012 elections won by Ahmed Shafik, the candidate favored by the majority of Egypt’s Coptic community. Following Dhattiwala and Biggs (2012) we used a squared version of this measure, which captures non-monotonic effects. If violence was more likely in more competitive districts, we would expect it to emerge at intermediate levels of Shafik support, which implies a district evenly split between Shafik supporters and supporters of his main rival Mohamed Morsi. H1c predicts that state repression can trigger a backlash against ethnic groups that are perceived to be allied with the regime. In our case, the Rabaa massacre is just such an act of state repression, and so we included a dummy variable for the seven days following the massacre. We might also expect that people living in areas where the relatives of those who died at Rabaa lived would be more likely to lash out at proximate Christians. To account for this, we included the square rooted count of individuals killed at Rabaa by their home district, sourced from Ketchley and Biggs (2015).

We proposed that beatings would be more closely associated with socioeconomic variables. H2a argued, for example, that beatings are more likely during periods of economic competition. We tested this with a measure of percentage change in the Consumer Price Index (CPI), taken from the Egyptian Central Bank, which captures changes in the cost of living. H2b leads us to expect that beatings are more likely during religious holidays. To measure this, we included dummy variables for Muslim and Christian holidays. We also included a dummy for whether the day in question was a Friday, the day of prayer for Muslims. Next, per H2c, we captured the strength of co-ethnic communal relations with the square rooted count of the number of co-ethnic protest events that occurred in a district in the pre-coup period, sourced from Ketchley (2017).

We expected bombings to follow the logic of terrorist violence. Following H3a, we used a (square root) measure of the distance between a given district and the governorate of North Sinai, where the terrorist group Ansar Beit al-Maqdis was based. To test H3b—that ethnic violence sponsored by terrorist groups may occur in major urban centers where it will garner maximum attention from local and international audiences—we constructed a distance measure in kilometers (square rooted) for each district from its centroid to the centroid of the district where the governorate headquarters is located. Finally, H3c suggests that bombings are less likely during Muslim holidays, when terrorists tend to exercise restraint. This hypothesis was captured by the same dummy variables for religious holidays and Fridays mentioned earlier.

Finally, H4 suggests that all three types of violence should be patterned by state incapacity. In early 2013, local police forces in Egypt launched a national wave of strikes demanding higher wages, better equipment, and more institutional support. We drew on an original dataset of these strikes (n = 143), sourced from Clarke (2020). ¹² We used the number of strikes that occurred in a district (square rooted) to capture the resources and resolve of the local police in that location. If weak policing capacity or deliberate police inaction facilitated attacks in that district, we would expect this variable to be positive.

We also include several contextual variables to account for plausible confounders. To account for the underlying population of Christians and Christian religious infrastructure, we included two measures: the logged number of Christians in a district (from the 2006 Egyptian census) and the number of Coptic churches (square rooted). The data on Coptic churches were taken from an online list of churches compiled by Egypt’s Coptic community. ¹³ We also included various measures from the 2006 census capturing the demographic characteristics of a district—the percentage of population employed in agriculture, the percentage with a university degree, and population density—in case more rural, urban, or educated districts had a higher or lower tendency toward ethnic violence. To measure any inertial effects of violence we included a one-day-lagged binary indicator of any type of ethnic violence in a district.

Results

Our results are reported in Table 1. Model 1 begins with our contextual variables and those measures capturing political motivations. Model 2 introduces our socioeconomic variables. Model 3 adds our measures for terrorist organization and strategy, and model 4 adds the variable measuring state incapacity. Coefficients are expressed as relative risk ratios (the exponent of β) and thus represent the risk of an event type occurring relative to the base category of no event.

Table 1 Predicting the Incidence of Burnings, Beatings, and Bombings

Z-scores in parentheses

P-values (two-tailed); p*<.05, p**<.01, p***<.001

The large number of coefficients and test statistics complicates reporting the results from multinomial models. For the sake of simplicity, we concentrate on the findings from model 4, which is akin to the full model. Marginal effects for statistically significant continuous variables are reported in figure 4. A strength of a multinomial modeling approach is that we can also test whether one event type is more likely to occur relative to another. ¹⁴ In this section, we indicate when these between-outcome differences are significantly different from each other. Because these tests involve multiple comparisons, we confine attention to differences that are significant at p < 0.01. ¹⁵ In online appendix figure A1, we also visualize between-type differences as a link plot.

Figure 4 Marginal effects

Beginning with our contextual variables, we see that the number of Christians in a district is a substantive and significant predictor of burnings and beatings, but not bombings. Next, the availability of more Christian religious infrastructure in a district is a positive and significant predictor of all three violence types. In addition, we see that beatings were 1.84 times more likely following a violent sectarian event the previous day, whereas the risk of a bombing increased by 27.9 times—and these differences are themselves statistically significant. ¹⁶ Demographic measures of agricultural employment, university education, and population density are not significantly associated with any event type.

Turning to our political variables, we find support for the idea that, when the motivation for violence is political, violence will tend to target an entire community by damaging that community’s communal properties. Both variables operationalizing H1a—Brotherhood occupations and vote support for Morsi—are positively and significantly associated with burnings. Burnings were 12 times more likely to occur in districts where the Muslim Brotherhood had strong mobilizational capacity, as measured by the presence of an anti-coup sit-in (p < 0.001). Increasing electoral support for Morsi increases the risk of burnings (p < 0.001; see figure 4e). It is also associated with an increased risk of beatings, but not bombings. We find little support for H1b: that political competition heightens the risk of burnings. The squared term on vote share for Ahmad Shafiq is not robustly associated with burnings or either of the other two violence types. We do, however, find support for H1c: that burnings are more likely after major acts of state repression. In the seven days following the Rabaa massacre we see that districts were 115 times more likely to experience a burning. The crackdown also increased the incidence of beatings, although to a lesser extent: beatings were six times more likely during this period. However, the seven days following the Rabaa massacre did not see an increased risk of bombings. ¹⁷ Note also that, when comparing between event types, burnings were significantly more likely to occur in the aftermath of the Rabaa massacre than either beatings or bombings (both p < 0.01). There is no robust association between ethnic violence of any type and the home districts of those killed at Rabaa.

Next, we evaluate H2a–2c, which posited that socioeconomic and cultural variables should be more associated with beatings because they exacerbate low-level intercommunal tensions. We find support for H2a: that beatings are more likely during periods of economic competition. A 1 SD increase in CPI, which measures the cost of living, is associated with an 11.8% increase in the risk of beatings (figure 4d). We also find support for the idea that religious holidays affect the incidence of beatings (H2b). These attacks were 2.4 times more likely during Muslim holidays, when Muslims may have been more sensitive to perceived affronts. For the Christian holidays variable, our results are inconclusive, because our models suffer from separation problems. There were no burnings or beatings in our dataset that occurred on a Christian holiday (and only one bombing), and as a result these outcomes are perfectly predicted, which explains the zero coefficient. We recognize this as a potential though unavoidable problem and note that coefficients in the full model do not change substantively with the exclusion of these variables. ¹⁸ Finally, we find little support for H3c: that cross-ethnic ties decrease the likelihood of violence. None of our three violence types is associated with recent histories of co-ethnic mobilization.

We see support for the idea that terrorist activity and terrorist strategic logic are behind the incidence of bombings. H3a called attention to the areas where terrorists have an organizational presence. We see that the distance from North Sinai (where ABM is based) is positively associated with burnings and beatings, whereas bombings are more likely to occur closer to North Sinai and this latter relationship is statistically significant (p < 0.05). Moreover, when we compare across event types, we see robust differences: bombings are more likely than both beatings and burnings to occur closer to North Sinai. We also find support for H3b. Bombings are more likely in highly visible and symbolically significant locations; they are more likely to occur closer to a governorate’s administrative center (figure 4f). We find that the same relationship holds with burnings, presumably because important communal sites would also be located close to urban centers. Moreover, these between-category differences (beatings vs. burnings and beatings vs. bombings) are themselves statistically significant. Finally, separation issues on our Muslim holidays variable make it difficult to evaluate H3c statistically. However, the fact that not a single bombing occurred during a Muslim holiday (the source of the separation issue) lends support to this hypothesis.

Lastly, our police strike variable allows us to assess H4: that coercive incapacity or deliberate inaction facilitates all three types of violence. We find, in line with this hypothesis, that both beatings and burnings were significantly more likely in districts that had recently witnessed police strikes. A 1 SD increase in police strikes raises the relative risk of beatings by 72 percent and burnings by 76 percent (figure 4c). However, there is no robust association between areas where police withdrew from the streets and bombings.

In online appendix section B, we provide several tests of the robustness of these findings, including dropping observations with extremely small and extremely large predicted probabilities to address potential concerns of hidden collinearity, increasing the duration of the variable capturing the post-Rabaa period, and varying the lag of recent anti-Christian events. We also include results from a logistic regression model using a re-merged version of the dependent variable—all ethnic violence events, regardless of type. The results from this model reinforce the importance of disaggregation: several variables that are associated with different violence types in the other models (e.g., CPI changes, Muslim holidays, distance to North Sinai) are not statistically significant in this model. In addition, variables that were associated only with one type of ethnic violence appear to predict ethnic violence in general in this model. In other words, the picture of ethnic violence revealed in these analyses is quite different when using a merged versus a disaggregated measure of the phenomenon.

Discussion

At the most general level, it is clear from these analyses that, although commonalities exist between our three violence types, there is also considerable variation in the factors that predict their occurrence. The analyses also provide support for many of the specific hypotheses laid out at the end of the section “Disaggregating Ethnic Violence.”

Against this backdrop, our findings lend support to the idea that a political logic is behind the incidence of burnings, which tend to inflict collective punishment on an entire community by targeting their communal sites. Burnings, we saw, made up a disproportionate number of the attacks after the Rabaa massacre, as organized groups of Morsi supporters sought to punish Egypt’s Coptic community for its perceived support of the military-backed coup. Moreover, underlining the importance of political entrepreneurs for this type of violence, we found that these burnings were much more likely in districts where the Muslim Brotherhood had recently organized a sit-in. This finding aligns with our qualitative analysis, which revealed that many burnings during the Rabaa week were instigated by Morsi supporters marching from sit-ins in central city squares. Further, another measure of political organization—vote support for Morsi in the 2012 elections—was also associated with this type of violence, as it was with beatings.

Beatings, we discovered, are more likely when the cost of living goes up. This finding is in line with theories of ethnic violence arguing that economic downturns raise tensions between ethnic groups due to economic competition. The frictions that emerge from economic competition over scarce resources are likely to manifest themselves at a highly local level; neighbors or co-residents from different ethnic groups may exact revenge against specific individuals whom they believe are to blame for their economic grievances. We also found that the incidence of beatings increases during Muslim holidays when Muslims are likely to be particularly sensitive to perceived cultural affronts. During these periods they may seek retribution against a specific individual whom they believe has offended them. Indeed, in our qualitative analysis we found that many beatings were tied to precisely these types of cultural disagreements, such as the building of unregistered churches or interfaith marriages.

We find support for the idea that bombings occur according to the rationales of terrorist violence. We see, for example, that bombings are more likely to take place closer to the symbolic center of a governorate, in line with the idea that terrorists stage these types of attacks to maximize their publicity and exposure before local and global audiences. (We also find that burnings are more likely near governorate centers, which aligns with what we know about the pattern of attacks during the Rabaa week, when assailants started off in central urban squares and targeted major churches.) Bombings were also more likely to occur closer to North Sinai, where the terrorist group ABM was based. Although our statistical results were inconclusive, the fact that no bombings occurred during Muslim holidays nevertheless suggests an underlying terrorist logic—that militants restrain themselves during these holy periods in anticipation of societal disapproval of violence (Reese, Ruby, and Pape 2017).

Finally, we found that in areas where the police had proven themselves either incapable or unwilling to enforce order, as measured by previous strikes, the likelihood of both beatings and burnings (but not bombings) was higher. This finding dovetails with what we learned from the qualitative data on the post-Rabaa massacre wave of violence: police forces, cowed and embittered by their experience during the revolution, failed to intervene to prevent attacks and, in certain cases, may have even facilitated this violence. Finally, the fact that bombings seem not to be affected by this variable potentially provides even more support to the idea that this type of attack is driven by a terrorist logic. Well-armed terrorist groups are likely to be less concerned by the presence or absence of police forces in a district, choosing their sites of attack according to other criteria.

Conclusions

Our analyses generated several important conclusions about the correlates of anti-Christian violence in Egypt from 2013 to 2018. Using a typology based on the nature of the target and the level of violence, we disaggregated this violence into three types: burnings, beatings, and bombings. The analyses suggest that because the motivations for these attacks and the organizational capacity of the attackers differ, the processes that generate them are also distinct: political factors shaped the occurrence of burnings, economic and religious factors seemed to affect beatings, variables capturing the strategic logic of terrorist organizations were best suited to explaining bombings, and measures of coercive incapacity were associated with burnings and beatings. Moreover, when comparing these findings to a model of ethnic violence in general, we find very different results, suggesting that considerable nuance is lost when using an aggregated measure of the phenomenon.

Of course, there are some important caveats regarding the status of these findings. First, they are based on observational analyses and do not necessarily speak to causal relationships. More research is needed to establish the causal connections between these factors and each of our violence outcomes. But our results nevertheless provide suggestive evidence that the underlying processes that generate these discrete forms of violence may be distinct. Second, the findings all emerge from analysis of a single case, and we cannot say whether these relationships are generalizable. We do believe our findings are intuitive and theoretically explicable—and that they are likely to hold in many other countries with politicized ethnic cleavages and imbalanced ethnic power relations. But we also grant that case-specific differences may exist. For example, the 2013 coup was clearly a watershed political event, which generated an extreme spike in ethnic violence. Our study does not allow us to say what other types of political events might have similarly explosive potential (e.g., elections, uprisings). Another example is the association between bombings and terrorist organization. This relationship might not exist in a context where firearms are much more readily accessible, or it might be that a different type of organized group, like a rebel organization or a branch of the state, is responsible for furnishing arms. More research is needed to determine the extent to which the findings in this paper generalize to other contexts.

Setting aside our specific findings, we hope that our typology will be helpful to scholars faced with a range of conflicting theories regarding ethnic violence. Our framework is useful because it is both tractable, in that it can be applied easily to many datasets and cases, and conceptually meaningful, in that it captures important substantive differences in various forms of ethnic violence. Although we recognize that other disaggregation strategies may also be fruitful, we do believe that the typology we have proposed could be applied productively to other cases. Ultimately, whether scholars choose to adopt our framework or not, we hope that this article has convincingly established the merit of disaggregating ethnic violence. Doing so, we believe, will allow us to develop more tailored and parsimonious explanations, ultimately advancing our collective knowledge and understanding of this important phenomenon.