Article

Journal of Conflict Resolution

2020, Vol. 64(10) 1916-1942
ª The Author(s) 2020
How Terrorism Spreads:
Emulation and the Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/0022002720930811
Diffusion of Ethnic journals.sagepub.com/home/jcr

and Ethnoreligious
Terrorism
Sara M. T. Polo1

Abstract
Previous research on the causes of domestic terrorism has tended to focus on
domestic determinants. Although this approach can be helpful to understand many
causes of terrorism, it implicitly disregards how the tactical choices made by similar
nonstate actors elsewhere influence a group’s decision to resort to terrorist tactics.
This study argues that the adoption of terrorism among ethnic and ethnoreligious
groups results from a process of conditional emulation. Groups are more likely to
emulate the terrorist choice of others with whom they are connected by shared
political grievances and spatial networks. The theory is tested on a new and original
group-level data set of ethnic and ethnoreligious terrorism (1970 to 2009) using
geospatial analysis and spatial econometric models. The results provide strong
support for the hypothesized mechanism leading to the diffusion of terrorism and
suggest that emulation—more than domestic and contextual factors—substantially
influences dissidents’ tactic choice.

Keywords
terrorism, tactics, diffusion, emulation, spatial econometrics

1
Department of Government, University of Essex, Colchester, United Kingdom

Corresponding Author:
Sara M. T. Polo, Department of Government, University of Essex, Wivenhoe Park, Colchester CO4 3SQ,
United Kingdom.
Email: [email protected]
Polo 1917

Recent terrorist attacks in Paris (2015) and Brussels (2016), carried out by returning
fighters from the war in Syria, have sparked fears of a spread of terrorism. Those
attacks showed how disenfranchised individuals could get inspired by the terrorist
activities of similar groups elsewhere, travel to conflict zones, receive training in
terrorist tactics, and conduct major attacks upon their return home (Schmidt 2016).
Indeed, terrorism has become an increasingly popular tactic among dissident groups.
Terrorist attacks have surged, globally, from an average of 1,523 attacks in the 1970s
to over 13,000 attacks in 2016 (Global Terrorism Database [GTD]). Moreover,
ethnic and ethnoreligious terrorism has become the most common: in 2016, four
of the five most active terrorist organizations were linked to ethnic and ethnoreli-
gious groups (i.e., Pashtun, Iraqi Sunni, Zaydi, and Kurds) killing an estimated 7,283
civilians. Despite the growing popularity of terrorism, however, not all groups
engage in this tactic. The tactical choices of the above groups stand in contrast with
those of other ethnic groups fighting conflicts in Chad, Myanmar, Liberia, and
Ethiopia, who largely avoided resorting to terrorism. Why has terrorism become a
dominant tactic among some groups and not others? To what extent are the Paris and
Brussels attacks illustrative of a broader phenomenon, whereby a group’s choice to
adopt (or reject) terrorism is influenced by the tactical choices of similar groups
elsewhere?
Despite a wealth of research on the causes of terrorism, existing work has not yet
adequately addressed the choice of terrorism by specific subnational groups and
organizations. Quantitative studies of domestic terrorism1 typically focus on the
attributes of states where terrorism emerges. Yet, this approach does not explain
why countries that are similar in many aspects (e.g., regime type, institutions, eco-
nomic wealth, instability) experience very different levels of terrorism by ethnic and
ethnoreligious groups, or why such terrorism has become so widespread. While it
may be tempting to explain domestic terrorism based solely on attributes of the
domestic context, this approach is incomplete as it disregards strategic interdepen-
dence in nonstate actors’ decision-making; that is, how the tactical choices by one
group affect the choices of other groups elsewhere. To date, only a handful of studies
have examined the diffusion of terrorism, but focusing only on transnational terror-
ism and on the country-level incidence of attacks (e.g., Braithwaite and Li 2007;
Neumayer and Plümper 2010). Domestic terrorism, however, is not a countrywide
phenomenon but typically emerges from specific subnational communities (Nemeth,
Mauslein, and Stapley 2014). Thus, we still know little about why terrorism is
chosen by some substate groups and not others and under what conditions domestic
groups respond to external incentives and “copy” the terrorist choice of others.
This article aims to fill this gap. I present a novel argument focusing on domestic
terrorism as the product of a diffusion process whereby groups observe and emulate
the tactical choice of others whom they perceive as similar to them and as an
example for their own behavior. The tactical choices and experiences of similar
ethnic groups are perceived to (and in fact may) contain relevant information on
the appropriateness of a tactical innovation in a specific political context (Simmons
1918 Journal of Conflict Resolution 64(10)

and Elkins 2004). Terrorist choices are therefore, in some sense, interdependent. But
terrorism does not spread everywhere and observing other groups’ terrorism does not
automatically lead to adoption. Political similarity between groups, especially
shared political marginalization, facilitates mutual identification and creates a feel-
ing of common grievances. This makes a group more willing to emulate the terrorist
choice of similar groups elsewhere. Yet, for successful emulation to occur, groups
also need to be capable to adopt terrorism. Geographic or network proximity to
groups who already engage in terrorism lowers the resource and skills constraints for
the adoption of terrorist tactics, provides logistical advantages for their use, and
increases a group’s overall perception of (terrorist) efficacy. Together, shared grie-
vances and direct ties between potential adopters and transmitters determine the
optimal conditions for the emulation of terrorism.
Group-level analyses of terrorism and diffusion have so far been hampered by the
lack of available data. In this article, I introduce a new and original data set of ethnic
and ethnoreligious terrorism, linking organizations in the GTD to ethnic groups in
the Ethnic Power Relations (EPR) database (Cederman, Min, and Wimmer 2010).
This is the first actor-based global data set on the use of terrorist tactics which
identifies the specific subnational ethnic communities from which terrorist organi-
zations have emerged. Using geospatial data on the geographic distance between
ethnic group settlements and nonspatial data on groups’ political status, I generate
measures of connectivity between all politically relevant ethnic groups between
1970 and 2009 based on their degree of political similarity and spatial proximity.
I then examine the effect of previous adoption of terrorism by an ethnic group on the
likelihood that other connected ethnic groups in the same and other countries resort
to terrorism and find a strong positive effect. Results from a Bayesian spatial probit
further suggest that indirect effects, or spatial feedbacks, reflecting tactical diffusion
are often more important than the direct effect of domestic attributes for explaining
the adoption of terrorist tactics. Moreover, while spatial proximity alone is insuffi-
cient to account for diffusion and operates in conjunction with identification via
shared grievances, some degree of proximity to “transmitters” gives organizations a
competitive advantage for the adoption of terrorist tactics relative to purely nonrela-
tional mechanisms based on media effects. The findings are robust across different
model specifications and estimations, when taking into account alternative diffusion
mechanisms such as spillover or competition, and while controlling for a number of
exogenous external conditions and spatially clustered group-level factors as well as
common shocks, which increase confidence that the results are not driven by a
simple common exposure.
This study contributes to research on terrorism, diffusion, and ethnic conflicts in
several important ways. By conducting the first systematic group-level analysis of
the diffusion of domestic terrorism, it shows that this very common type of terrorism
is often a product of strategic emulation between groups. This contributes an alter-
native explanation of domestic terrorism focused on the interdependence of groups’
tactical choices.
Polo 1919

Second, moving beyond the conventional focus on states as relevant spatial

units in conflict diffusion, I provide a novel, unified framework for understanding
the diffusion of violent tactics between nonstate actors, both within and across
national boundaries. This framework identifies specific sender and receiver groups
involved in diffusion dynamics and explains why some ethnic groups engage in
terrorism while others, even those in the same state or with similar domestic
circumstances, do not.
Third, the new data set of terrorist tactics of ethnic and ethnoreligious groups,
combined with advanced spatial econometric techniques, allows testing the mechan-
isms of terrorism diffusion in ways not possible before, and to identify key channels
through which intergroup diffusion occurs. This approach further improves upon
existing research on conflict diffusion as it demonstrates that nonrelational and
relational mechanisms of diffusion, based, respectively, on ideas and direct ties, are
not necessarily mutually exclusive but complement to each other in very specific
ways. Finally, this research has important implications for policy-making. Identify-
ing the external causes of domestic terrorism, the groups more likely to choose
terrorism, and the mechanisms through which terrorist tactics diffuse can inform
more effective policies for countering and preventing terrorism.

The Spatial Dimension of Domestic Terrorism

Terrorism can be defined as the threat or use of violence by substate actors in order
to obtain a political or social goal through the intimidation of a wider audience
beyond the immediate victims (Enders and Sandler 2012). It is clear from mapping
the distribution of domestic terrorism that incidents of terrorism cluster in certain
areas and are relatively rare in others (see Online Appendix C). The existence of
such spatiotemporal clusters raises questions about the relative effect of domestic
versus external factors as determinants of terrorism (Galton 1889). To date, how-
ever, it remains unclear whether, and under what conditions, strategic interdepen-
dence and diffusion can lead to the adoption of domestic (as opposed to
transnational) terrorism.
Existing studies on the causes of domestic terrorism tend to focus on individual-
level or state-level factors and their influence on willingness and/or opportunities to
engage in terrorism (e.g., Krueger 2007; Asal, Brown, and Schulzke 2015; Choi and
Piazza 2014; Findley and Young 2011). While very insightful, this approach is
incomplete for two main reasons. First, it generally assumes that the motivating
factors for terrorism are a function of the domestic environment. However, if we
look specifically at subnational groups, such as ethnic groups, many of these have
not resorted to terrorism even when the environment was conducive to collective
mobilization. In South America, for instance, there are many politically and eco-
nomically marginalized ethnic groups (e.g., indigenous groups in Colombia, Vene-
zuela, Ecuador, Peru, and Chile), but despite the numerous civil wars and high levels
of terrorism in the region, no organization acting on behalf of these groups has
1920 Journal of Conflict Resolution 64(10)

resorted to terrorist tactics. In contrast, marginalized ethnic groups elsewhere have

resorted to terrorism, as in the case of the Kurds, Balochs, African Americans,
Bretons, Corsicans, Moro, and Acehnese.
Second, grievances and opportunities correlate not only with terrorism but with
many different forms of violent (and nonviolent) collective action (Cederman, Gle-
ditsch, and Buhaug 2013). Put differently, terrorism is only one option from a
broader menu of dissent tactics and strategies. Yet, so far, there has been little
attention to how decisions about specific tactics are made in the context of alterna-
tives. Why do groups’ tactical choices converge on terrorism? The domestic (or
closed-polity) approach de facto disregards how other actors’ choice of terrorism
informs decisions about appropriate tactics through learning and emulation effects.
It also fails to address the timing of adoption. Why have some groups—such as the
French Basques, Corsicans, Baloch, and Kurds—started to use terrorism only at a
specific point in time?2 To answer these questions, it is necessary to explicitly take
into account the interdependence of groups’ tactical choices and diffusion effects.
Unfortunately, there is a daunting gap in research on the actor-level diffusion of
domestic terrorism. While the civil war literature has long recognized diffusion
dynamics, little systematic evidence exists regarding the diffusion of terrorism
(Braithwaite and Chu 2018). The few existing studies of terrorism diffusion do not
examine domestic terrorism3 but focus primarily on less common international
terrorist events (Bove and Böhmelt 2016; Braithwaite and Li 2007; Midlarsky,
Crenshaw, and Yoshida 1980; Neumayer and Plümper 2010). Moreover, although
these studies have pioneered research on the topic, their dominant focus on country
linkages as possible channels of diffusion4 masks the fact that only some groups
within a country choose terrorism and others, usually the majority, do not. Put
differently, there are specific sender and receiver groups in diffusion processes.
Hence, analyzing the spread of terrorism across countries does little to clarify who
are the specific groups that use terrorism and why or what makes some groups
ostensibly unresponsive to demonstration effects.
Existing studies have not gone far enough to unpack the subnational and transna-
tional dynamics of terrorism diffusion partly because of a lack of actor-level data.
My actor-based data set allows addressing this gap in ways not possible before.
Moreover, by modeling linkages between substate groups, it is possible to analyze
specific mechanisms though which actor-level diffusion occurs. Several mechan-
isms have been suggested, such as emulation and competition, but these are largely
left untested. As a consequence, we still lack a theoretical and empirical framework
for understanding intergroup diffusion and how the terrorist choices by one group (or
lack thereof) affect choices of other groups, in the same country and elsewhere.

The Diffusion of Ethnic and Ethnoreligious Terrorism

Before turning to the analysis of terrorism diffusion, it is important to clarify what
constitutes ethnic and ethnoreligious terrorism and why this type of terrorism
Polo 1921

Figure 1. Ethnic terrorism 1970 to 2009.

warrants specific attention. Organizations that adopt terrorism as a tactic to advance

their political objectives have often emerged from specific subnational communities.
Ethnic groups are cultural communities based on the belief in a common ancestry and
shared culture (Cederman, Gleditsch, and Buhaug 2013, 23). Some ethnic groups,
such as Sunni and Shia groups in Iraq and Lebanon, define their ethnic identity also in
relation to a particular religion; I define these groups as ethnoreligious.5 Historically,
the most common type of terrorism has been ethnic and ethnoreligious terrorism, that
is, terrorism perpetrated by organizations which make claims on behalf of a specific
ethnic or ethnoreligious group. Notable examples of such terrorist organizations
include, among others, the Irish Republican Army, the Corsican National Liberation
Front, Basque Fatherland and Freedom, the Kurdistan Workers Party, Hezbollah,
Hamas, al-Aqsa Martyrs Brigades, Lashkar-e-Jhangvi, al-Qaida in Iraq/Islamic State
of Iraq, Ansar al Sunna, and the Taliban. Figure 1 illustrates the adoption of terrorism
by ethnic and ethnoreligious groups between 1970 and 2009. The blue areas are spatial
polygons representing the geographic settlements of the groups that have adopted
terrorist tactics based on the Geo-EPR data (Wucherpfennig et al. 2011). Data on the
terrorist activities of organizations claiming to represent specific ethnic and ethnor-
eligious groups come from a new data set described in the Data section. It is clear from
Figure 1 that the adoption of terrorism is not random but is concentrated in some
geographic areas while absent in others. In the next sections, I develop an argument for
why and how groups choose terrorist tactics which accounts for these patterns.6

How Decisions about Tactics and Strategies Are Made

I begin by assuming that ethnic groups, and their leaders, operate in the context of
uncertainty about courses of action and of constraints on both rationality and
resources for mobilization (Checkel 2013; Lake and Rothchild 1998; Simmons,
1922 Journal of Conflict Resolution 64(10)

Dobbin, and Garrett 2008). Specifically, when deciding whether and how to mobi-
lize collectively, groups often face uncertainty regarding which tactics to use and
which tactics are most appropriate and effective. Even in the presence of political or
economic grievances, ethnic groups can choose, at least in principle, among several
courses of actions. For instance, they can keep the status quo, resort to nonviolent
resistance, or adopt violent tactics. In addition, actors are constrained in their ability
to process information about specific tactics (Tversky and Kahneman 1974). Gath-
ering, elaborating, and evaluating such information can be daunting and groups may
still not be able to determine the appropriateness of a given tactic to their specific
circumstances.
To overcome such uncertainty and constraints on information processing, ethnic
groups often look for examples (Simmons and Elkins 2004). Relying on the expe-
rience of others constitutes a useful cognitive heuristic to resolve the dilemma of
tactic choice. Ethnic group leaders can infer from the prevalence of a particular
tactic within the repertoire of other groups, or by the sheer number of adopters, that
this may be the best thing to do. This process applies also to the choice of terrorist
tactics. When terrorist tactics appear to be gaining attention and attracting partici-
pants, groups may have an incentive to copy that tactic and capitalize on its per-
ceived success (see also Cunningham, Dahl, and Frugé 2018, 594). From a bounded
rationality perspective, emulation of others’ behavior is thus a very simple but useful
focal point for choosing dissent tactics.7
Yet, observing other groups’ adoption of terrorism is a necessary but not suffi-
cient condition for actual emulation. Increasing the availability of information about,
and the visibility of, terrorist tactics only increases the potential audience for such
tactics; it does not necessarily elicit a response from that audience (Hill, Rothchild,
and Cameron 1998). The spread of terrorism can stall in the face of inattention, if
others’ actions do not resonate with the group members, or if a group perceives too
high a risk. The diffusion of terrorism through emulation is not automatic and differs
from the simple spread of ideas. Terrorism is a conflict tactic, subject to specific
incentives and constraints. In particular, not all groups that employ terrorism are
regarded as relevant or compelling examples to emulate. And not always does the
behavior of other groups positively influences a group’s own calculations of benefits
to be gained and opportunities for effective terrorist action. To understand the
conditions under which terrorist emulation actually occurs, it is necessary to con-
sider specific linkages between sender and (potential) receiver groups and how the
senders’ behavior influences recipient groups’ willingness and capability to initiate
and sustain terrorist activities.

Identification through Shared Grievances

As previously argued, exposure to other groups’ terrorist behavior by itself does not
lead to emulation. Information about other groups’ tactical choices fosters emulation
only when the other group is perceived as a reference, hence as a model for a group’s
Polo 1923

own behavior. In other words, emulation is dependent upon identification with an

actor (Simmons and Elkins 2004; Simmons, Dobbin, and Garrett 2008). The policy
diffusion literature has emphasized the importance of similarity between potential
adopters and transmitters (Shipan and Volden 2008). Among ethnic groups, the
sharing of a similar political status is a key feature that fosters mutual identifica-
tion. A group’s political status is determined by the level of group access to state
power, that is, the degree to which a group is politically included or excluded (e.g.,
Cederman, Gleditsch, and Buhaug 2013). By defining the position of an ethnic
group within the polity, political status creates visible identification points
between groups who share similar political circumstances and status, even when
they live in different states.
At the same time, not all forms of shared political status are equally conducive to
emulation. Groups excluded from political power, rather than groups in power, are
more likely to respond positively to the mobilization of other groups facing similar
circumstances (see also Metternich, Minhas, and Ward 2017). Observing the terror-
ist choice of other politically excluded groups increases the domestic salience of
existing structural inequalities and contributes to transforming these into actual
grievances. Following the mobilization of similar groups elsewhere, politically
excluded groups become more aware of their own disadvantage in relation to the
groups in power, realize the injustice of such a condition, and are spurred to change
the status quo based on the example of others (e.g., Kuran 1998). An increase in the
use of terrorism by excluded groups elsewhere has a two-fold effect. First, it makes
group-level grievances more salient in the domestic discourse. Second, and more
importantly, it activates those grievances by providing a model for a specific tactic
and mobilization trajectory to redress them. Put differently, as terrorism becomes
prevalent within the repertoire of marginalized reference groups, members of a
politically excluded ethnic group become more likely to infer that such a tactic
might be the most appropriate to adopt. Ethnic entrepreneurs play an important role
in this process. As Tilly (2003, 34-35) notes, ethnic entrepreneurs specialize in
activating group identities, connecting distinct groups and networks, and coordinat-
ing collective action. As a result, they can leverage the terrorist momentum produced
by marginalized ethnic groups elsewhere to activate emotions and gut responses
within their own group. These gut responses are often more effective at spurring
support for terrorism than complex calculations of costs and benefits (see also
Tversky and Kahneman 1974).
I have argued that the process of mutual identification and perception of shared
grievances, which is necessary for the emulation of terrorism, is stronger when both
sender and receiver groups are disadvantaged and excluded from access to political
power. This does not mean that politically advantaged groups do not have grievances
or cannot identify with others. However, these groups’ high stakes in the status quo
and the availability for them of alternative, legal means to address potential grie-
vances make these groups less responsive to terrorism cascades elsewhere. This
leads to heterogeneous responses among (potential) recipients of diffusion effects.
1924 Journal of Conflict Resolution 64(10)

To illustrate, consider the case of ethnic groups in Iran. While Iran is located in a so-
called terrorism hot spot, only the Kurds, Baluchis, and Arabs have resorted to
terrorism while the Persians and the Azeri have not. Similarly, in France, only the
Corsicans, Bretons, and Basques have adopted terrorism following the example of
other marginalized groups in Spain, Italy, and United Kingdom, while no organiza-
tion claiming to represent the French has resorted to terrorism.

Adoption Capacity and Expected Efficacy

The emulation of terrorism is not just influenced by identification between groups
based on shared political exclusion. Marginalized groups are not able to plan and
execute terrorist attack at will even when inspired to do so by similar groups else-
where (Braithwaite and Chu 2018). The choice of terrorism is constrained by access
to resources such as weapons, personnel, training, and experience; yet, these
resources are crucial for orchestrating effective terrorist campaigns. If group leaders
want to impose significant pressure on the government and attract attention, they
need to maintain a certain level of activity through organized campaigns. Failure to
recruit enough members to the terrorist organization and to carry out successful
attacks signals group weakness and generates high audience costs for the leadership.
In what follows, I argue that the location of reference groups who already engage in
terrorism is a very important factor that allows a group to overcome the organiza-
tional and resource constraints surrounding the adoption of terrorism and increases
expected efficacy. Thus, for successful terrorist emulation to occur, it matters not
only who the other groups are but also where they are.
Unlike peaceful protests and demonstrations, terrorism is a more costly and risky
tactic which requires specific skills and resources. The start-up costs of initiating a
terrorist campaign in complete isolation can be extremely high. Recently, individual
incompetence and lack of adequate training have led to the failure of several
attempted terrorist attacks in Europe and the Unite States (see Burke, The Guardian,
September 15, 2017). Although such terrorist incompetence saved many lives, it
represented a major setback for those who planned the attacks. In contrast, some of
the most severe and damaging attacks recently occurred in Europe, such as the Paris
and Brussels attacks, were characterized by very high levels of sophistication and
required the work of munition specialists who trained in conflict zones, especially
Syria (see Callimachi, The New York Times, March 29, 2016). The terrorists them-
selves had previously traveled to Syria, where they joined the Islamic State and
received training in terrorist tactics.
The above examples illustrate the importance of resources, capabilities, and
training for conducting successful attacks. Establishing direct ties with more expe-
rienced groups that already engage in terrorism lowers the barriers to initiating and
sustaining terrorist attacks and increases a group’s expectation of efficacy. Creating,
or joining, local networks of terrorist organizations facilitates knowledge transfer
and learning; allows for joint training and planning of attacks; and increases the
Polo 1925

availability, supply, and exchange of weapons, personnel, and other terrorism-

specific capital; it even allows groups to forge alliances, make common cause, and
launch coordinated campaigns against the state (see also Buhaug and Gleditsch
2008). Terror networks therefore operate as a crucial force multiplier by bolstering
groups’ adoption potential, capabilities, efficacy, and lethality (Asal and Rethe-
meyer 2008). Geographic proximity is a critical component of such networks that
tend to be spatially delimited (Asal et al. 2016; Perliger and Milton 2016; Sageman
2004). These networks, and the consequent opportunities for successful terrorist
actions, are much more difficult to create and maintain among groups that live very
far away, and the material advantages of these networks are less likely to accrue to
groups that are relatively isolated from the other members.
Moreover, groups’ tactical choice-set is usually limited to those tactics that are
immediately accessible to them (Tilly 1978). This is especially true for nonstate
actors, who do not have the same capabilities of governments and face more severe
constraints on the adoption of tactical innovations (Horowitz 2010). Some degree of
geographic proximity to politically similar groups who already engage in terrorism
makes the latter a more accessible tactic, by mitigating resource- and skills-
constraints and providing means for effective terrorist emulation.8
It may be argued that groups should be less likely to emulate terrorism if this does
not lead them to achieve their ultimate goals. The success of terrorism, however, is
difficult to define in practice. On the one hand, governments rarely make major
concessions to terrorists due to reputational concerns (Pape 2003). Terrorists, on the
other hand, tend to push alternative narratives which portray them as unyielding
fighters and true believers that cannot be defeated or intimidated by a display of
government military power. Moreover, groups not only have long-term goals but
also proximate goals which include attracting public attention, recruiting members,
demonstrating mobilization potential, and imposing political costs on the govern-
ment. These goals are a necessary precondition for the achievement of most other
goals, and studies have shown that terrorism can be effective at achieving such
proximate goals (Bapat and Zeigler 2016).
The expectation of government repression is also unlikely to deter emulation.
Groups often use terrorism precisely to provoke repression (Bueno de Mesquita and
Dickson 2007). Moreover, ethnic entrepreneurs can leverage repression to exacer-
bate out-group antagonism and persuade members of their constituency to support
extreme terrorist actions. Hence, there are benefits to be gained from terrorism
provided that groups are inspired by the example of others they identify with and
able to organize effective campaigns. But effective action depends also on the
availability of resources, skills, and membership which are facilitated by proximity
to terrorist groups and local networks.
Figure 2 summarizes the general mechanism of diffusion discussed so far.
Although exposure to other groups’ terrorism is a necessary condition, it is insuffi-
cient, on its own, to generate emulation. The diffusion of terrorism through emula-
tion requires, first, a process of mutual identification between groups based on the
1926 Journal of Conflict Resolution 64(10)

Exposure to other
groups’ terrorism

sender & receiver Identification via Adoption capacity geographic proximity

political exclusion shared grievances Expected efficacy b/w sender & receiver

Emulation

Figure 2. Mechanism of diffusion.

sharing of similar grievances. As Figure 2 illustrates, this is more likely when both
sender and receiver groups are politically excluded. Second, groups need to be opti-
mistic about the availability of terrorism-specific resources and skills, which boost
their capacity to effectively carry out attacks. This condition is influenced by the
degree of geographic proximity to groups that already adopt terrorist tactics. However,
a generic proximity to terrorism (such as the presence of a terrorism hot spot in the
neighborhood) will be insufficient to spur emulation. Proximity to terrorism operates
in conjunction with identification between groups through shared political exclusion.
In the absence of the latter, groups will be less likely to collectively emulate terrorist
violence despite the available opportunities.9 To illustrate, consider the case of the
indigenous groups of South America (e.g., Quechua, Indigenous People of the Ama-
zon, Aymara, Tupi-Guarani). This region experienced several civil wars and some of
the highest levels of terrorism ever recorded, mainly inspired by leftist ideologies. Yet,
marginalized indigenous groups did not respond positively to such violence and no
organization claiming to represent these groups has resorted to terrorism. Instead,
nonviolent tactics have rapidly spread among these groups to become one of their
primary methods of resistance (Cunningham and Sawyer 2017).
Based on the above discussion, I formulate the following hypothesis:

An ethnic group is more likely to adopt terrorism when it shares a status of

political exclusion with other ethnic groups who engage in terrorism and the
latter are geographically proximate.

Data and Research Design

To test the hypothesis on the diffusion of terrorism, I compile a new data set, the
GTD2EPR, of ethnic and ethnoreligious terrorism from 1970 to 2009. This data set
links terrorist organizations in the GTD with politically relevant ethnic groups in the
Polo 1927

EPR data set (EPR-ETH; Cederman, Min, and Wimmer 2010). To link terrorist
organizations with corresponding ethnic groups, I collected information on whether
organizations claim to represent a specific ethnic group, complementing the infor-
mation in the Armed Conflict to EPR data set (Wucherpfennig et al. 2012) with
additional research on organizations not involved in a civil war.10 This data set
allows me to identify the specific ethnic and ethnoreligious communities from which
terrorist organizations have emerged.
The dependent variable is a group-level binary indicator of terrorism, based on
one or more attacks perpetrated by organizations linked to an ethnic group per
year.11 Only domestic terrorist attacks are considered, and I exclude attacks perpe-
trated by ethnic organizations in other countries.
To estimate terrorist diffusion between groups, I rely on spatial regression models
where a connectivity matrix W specifies the dependence structure between all
politically relevant ethnic groups in every given year. I generate two sets of spatial
weights matrices that reflect the two different forms of connectivity presented in the
theory, namely, similarity based on shared political grievances and degree of spatial
proximity. To operationalize shared political grievances, I consider whether two
groups share a status of political exclusion. The EPR coding of political exclusion
focuses on exclusion from access to central government. Based on the political status
of each ethnic group for every given year, I construct a binary matrix of political
similarity where elements are 1 if two groups share political exclusion and 0 other-
wise. I measure geographic proximity between ethnic groups using GeoEPR, a geo-
referenced version of the EPR-ETH data set, which encompasses information on the
specific settlement areas (polygons) of ethnic groups from 1946 to 2009. GeoEPR
includes 812 group polygons and over 700 unique ethnic groups. Group polygons are
not fixed but can change over time reflecting the emergence of new countries or
changing settlements within countries. For each year from 1970 to 2009, I calculate
the minimum distance between each group polygon and all other polygons and
generate a binary connectivity matrix for each year.12
For each year, the two matrices are multiplied element by element (i.e., Hada-
mard product) so that each element (or weight) wijt of the final matrix represents the
product of elements wijt of the original two matrices. This interaction produces a
matrix, WExclusionProximity , which reflects the specific combination of shared polit-
ical grievances and geographic proximity which is posited in the theoretical argu-
ment (nonspatial and spatial weights). The matrices are then combined into a single
NT  NT block diagonal matrix.13
To test the hypothesis that ethnic groups are more likely to adopt terrorism when
similarly excluded groups in the region also adopt terrorism, I rely on a two-pronged
research design focusing on multilevel and spatial econometric models. First, given
the clustered nature of the data, with group-years nested in groups nested in coun-
tries, I employ a series of multilevel spatial logit models in which the intercepts vary
as a function of clustered group- and country-level variables. These models account
1928 Journal of Conflict Resolution 64(10)

for different forms of spatial dependence between observations and for the variance
among groups’ in terms of their susceptibility to adopting terrorism (i.e., unit het-
erogeneity). Empirically, this is modeled by including a spatial lag and by allowing
the group intercepts to vary according to specific group-level attributes conducive to
terrorism.
In the multilevel models, the spatial lag is obtained as the product of the weights
matrix WExclusionProximity , as previously defined, and a temporally lagged dependent
variable (yt1 ).14 In the models, I rely on two different observational spatial lags.
The main difference between them is given by the measure of geographic proximity
adopted in the construction of the W matrix. In the first case, the spatial lag is
calculated considering as neighbors all politically similar (i.e., excluded) ethnic
groups within a minimum distance of 200 km. In the second case, only the politically
similar groups in first-order neighbor countries are considered; hence, I exclude
from the neighbors’ list all groups within the same country. This is because the
observed diffusion effects could be due not only to emulation between groups but
also to the alternative mechanism of competition. However, if this mechanism is
indeed at work, it should only affect ethnic groups fighting against the same gov-
ernment and possibly sharing similar audiences. The two observational spatial lags
also differ in that the one which only considers ethnic groups in neighboring coun-
tries takes the form of a dummy variable measuring whether at least one group in the
neighborhood has adopted terrorist tactics in the previous year (rather than their
weighted average or sum).
The multilevel model described above provides an important first test of
terrorist diffusion. However, it does not allow to fully estimate the strength of
interdependence between groups’ tactical choices and how spatial effects pro-
pagate through the groups in the system (see also Franzese, Hays, and Cook
2016). Therefore, to provide a more complete analysis of terrorism diffusion that
incorporates these important aspects, I rely on a more sophisticated spatial
probit model, estimated by Bayesian Markov Chain Monte Carlo (MCMC;
Wilhelm and Godinho de Matos 2013; LeSage and Pace 2009) on a subset of
the data set.
The estimation of a spatial probit model represents a more complex approach.
Suppose we have the following spatial autoregressive model:15

y ¼ rWy þ Xb þ E; E*N ð0; s2E In Þ; ð1Þ

where y is the continuous latent outcome variable, W is an N  N connectivity

matrix which captures the spatial interdependence between units, the parameter r is
the coefficient for the effects of other units’ outcome through this type of connec-
tivity as specified in the W matrix, X is an N  k matrix of covariates, and b is a
k  1 vector of coefficients associated with the k covariates. In this model, the latent
variable is unobserved. Instead what is observed are the binary outcomes (0, 1) as:
yi ¼ 1 if yi  0 and yi ¼ 0 if yi < 0.
Polo 1929

The reduced form of equation (1) is:

y ¼ ðIn  rWÞ1 X b þ ðIn  rWÞ1 E; ð2Þ

where ðIn  rWÞ1 E is the reduced-form error term, and where the errors are no
longer independent and identically distributed due to the spatial multiplier
ðIn  rWÞ1 . The jointly determined error terms represent a considerable estimation
challenge due to the need to compute an n-dimensional integral which becomes
analytically intractable even for relatively small n (see also Franzese, Hays, and
Cook 2016).
The Bayesian MCMC approach is a simulation-based method. The basic idea in
Bayesian estimation is to sample from the posterior distribution of the model para-
meters pðy ; b; rjyÞ given the data and some prior distributions for the parameters.
The sampling from the posterior distribution can be realized by an MCMC and
Gibbs sampling scheme (Wilhelm and Godinho de Matos 2013).
The Bayesian estimator of the spatial probit introduced by Wilhelm and Godinho
de Matos (2013) follows the Bayesian Gibbs sampling approach proposed by
LeSage (2013) and LeSage and Pace (2009) with some modifications to facilitate
implementation. For computational efficiency, the spatial probit is estimated on a
subset of the main data set, that is, two cross sections of all ethnic groups from 1991
to 1999 and from 2000 to 2009, where values of the dependent variable are averaged
over each time period. To avoid posttreatment bias, the average values of the group-
level covariates are calculated using only the years up to the first observed terrorist
attack for each ethnic group. An additional step would be to estimate the models
considering values of the dependent and independent variables in each year. The size
of data set, which has more than 22,000 observations, makes it very computationally
intensive to estimate a Bayesian spatial probit over the full data set.
In both the multilevel and spatial probit models, I consider a number of additional
influences on groups’ incentives for terrorism. In particular, I control for the non-
spatial effect of political exclusion, that is, the effect of political exclusion in the
absence of diffusion (i.e., predynamic effects; Franzese, Hays, and Cook 2016).
While the theory posits that political grievances are more likely to lead to terrorism
when they are activated by the terrorist choices of similar groups elsewhere, it is
important to test whether exclusion by itself is sufficient to induce terrorism. There-
fore, all models include dummy variables for the political status of ethnic groups
which reflect the type and degree of political exclusion, with included groups as
reference category. I also control for the relative size of the ethnic group, to account
for latent mobilization potential. Data on political status and group size are taken
from the EPR-ETH data set. Even though the analysis is at the group-level, con-
textual factor may still affect groups’ choice of terrorism. I therefore control for a
country’s level of democracy (Vreeland 2005) and its gross domestic product per
capita (Gleditsch 2002). In addition, since civil war is frequently associated with
higher levels of domestic terrorism (Findley and Young 2012; Smith and Zeigler
1930 Journal of Conflict Resolution 64(10)

Table 1. Probability of Terrorism for Ethnic Groups (Full Sample 1970 to 2009).

Wy: Subnational Wy: Subnational Wy: Transnational

and Transnational and Transnational Only

Wyt1 ExclusionProximity 1.967*** (0.300) 1.990*** (0.300)

Wyt1 ExclusionProximity (binary) 2.022*** (0.133)
Excluded 0.541** (0.203)
Discriminated 0.951*** (0.233) 0.863*** (0.222)
Powerless 0.150 (0.229) 0.168 (0.220)
Separatist 1.399*** (0.396) 1.292** (0.399)
Regional autonomy 0.349 (0.282) 0.425 (0.277)
Group size 0.401 (0.591) 0.472 (0.578) 0.066 (0.529)
Civil war 1.918*** (0.134) 1.838*** (0.135) 1.694*** (0.133)
Xpolity 0.062*** (0.018) 0.060** (0.018) 0.043* (0.018)
GDPpc log 0.387** (0.128) 0.422** (0.130) 0.130 (0.118)
Constant 8.228*** (1.148) 8.351*** (1.159) 6.061*** (1.027)
s country 0.432 0.417 0.266
s groups 0.644 0.599 0.466
Wald w2 720.33*** 735.08*** 843.37***
Log-likelihood 2,045.73 2,035.66 1.990.40
Number of observations 21,949 21,949 23,500

Note: Cubic polynomials are not shown in the table. GDP ¼ gross domestic product.
*p < .05.
**p < .01.
***p < .001.

2017), I include a variable for the presence of civil war to account for possible
domestic spillover effect. Civil war data are culled from the Uppsala Conflict Data
Program using a threshold of twenty-five battle-related deaths. Finally, I include a
cubic polynomial of time since the last terrorist attack to control for temporal
dependence (Carter and Signorino 2010).16

Empirical Analysis and Discussion

The empirical results for the multilevel models are reported in Table 1. The results
are consistent with the theoretical expectation that an ethnic group is more likely to
adopt terrorist tactics if other politically marginalized and geographically prox-
imate ethnic groups also use similar tactics. More specifically, the coefficient for
the spatial lag of terrorism from excluded groups in neighboring regions, regard-
less of country, is positive and highly significant. The coefficient for the spatial lag
of terrorism from excluded groups in neighboring countries is also positive and
significant.
The results from the Bayesian spatial probit are reported in Table 2.17 Again, the
results provide strong support for the hypothesis. The coefficient for r, the spatial
autocorrelation parameter, is positive and highly significant. This indicates
Polo 1931

Table 2. MCMC Spatial Autoregressive Probit (20,000 Iterations, Burn-in ¼ 2,000, Diffuse
Priors for b Parameters and Uniform Prior for r).

1991 to 1999 2000 to 2009

rWy 0.490*** (0.095) .536*** (.090)

Discriminated 0.803*** (0.197) .728*** (.208)
Powerless 0.029 (0.162) .129 (.169)
Separatist 1.385*** (0.352) .989** (.364)
Regional autonomy 0.066 (0.213) .112 (.196)
Group size 0.071 (0.293) .283 (.335)
Civil war 0.147 (0.156) .148 (.150)
Xpolity 0.029 (0.019) .003 (.017)
GDPpc log 0.110þ (0.063) .037 (.058)
Constant 0.024 (0.499) .466 (.494)
Number of observations 623 630
Note: Coefficients indicate posterior mean. Standard deviation is given in parentheses. GDP ¼ gross
domestic product.
þ
p < .1.
*p < .05.
**p < .01.
***p < .001.

interdependence in ethnic groups’ decision to adopt terrorist tactics based on shared

political exclusion and degree of geographic proximity.
Turning to the nonspatial effects of political exclusion, while excluded groups are
generally more likely to resort to terrorism than included ones, when disaggregating
exclusion only discriminated and separatist groups are likely to adopt terrorism in
the absence of diffusion. In contrast, in all the models, powerless and regional
autonomy status have no significant independent effect on the choice of terrorism.
These results support the theoretical argument that while domestic grievances are
important they are often insufficient on their own to motivate terrorism. In order to
spur terrorism, such grievances need to be activated by the terrorist choice of similar
groups elsewhere. Interestingly, these results on the exclusion variables differ from
the findings in the civil war literature, where different forms of exclusion are all
strong predictors of civil war onset independent of diffusion effects (Cederman,
Gleditsch, and Buhaug 2013).
I now turn to examine the substantive implications of the models. Figure 3 depicts
the substantive effects from the multilevel model based on estimates from Table 1.
Panel A illustrates the effect of terrorism from geographically proximate excluded
groups on the probability that an excluded group also adopts terrorism. The results
are shown for discriminated groups, which face the highest level of political exclu-
sion and are likely to have very strong domestic motives for violence even in the
absence of terrorism from other groups. The presence of at least one politically
excluded ethnic group which used terrorism in the previous year increases the
1932 Journal of Conflict Resolution 64(10)

A B
0.100

0.075
Probability of terrorism

Probability of terrorism
0.075

●
0.050

0.050

0.025
0.025
●

0.000
0.000
No Terrorism Terrorism 0.00 0.25 0.50 0.75 1.00

Politically excluded reference groups' tactic Spatial lag of terrorism − Wy (exclusion*proximity)

Figure 3. Predicted probability of terrorism as a function of reference groups’ terrorism with

95 percent confidence intervals.

likelihood of adoption for a discriminated group by a factor of 5. Conversely, in the

absence of diffusion, the probability that a discriminated group adopts terrorism is
much lower and close to zero. Panel B shows how an increase in the proportion of
politically excluded and geographically proximate terrorism adopters affects the
likelihood of terrorism for a discriminated group. As the spatial lag increases from
0 to 0.75, indicating 75 percent terrorism adopters among the reference groups, the
predicted probability of terrorism for a discriminated group increases by a factor of
3. Taken together, these results reveal a substantial diffusion effect between polit-
ically excluded and geographically proximate ethnic groups.
Marginal effects for the Bayesian spatial probit require more complex calcula-
tions but allow us to directly estimate the magnitude of terrorism spillovers, or
spatial feedbacks, between groups. Following LeSage and Pace (2009) and LeSage
(2013), scalar summary measures of direct, indirect (i.e., spillover/diffusion), and
total effects are presented in Table 3. As before, I focus specifically on discriminated
groups. Marginal effects reflect changes in the probability of terrorism following an
exogenous shock in an independent variable, in this case discrimination. However,
in the spatial probit, a shock toward political discrimination in ethnic group i will not
only increase the probability that this group adopts terrorist tactics but will also
affect the probability that other, excluded neighboring groups j resort to terrorism.
The magnitude of this effect depends on the degree of connectivity between i and j
(as defined in the WExclusionProximity matrix) and on the strength of spatial depen-
dence (as measured by the r parameter). Discrimination of group i will then have a
direct impact on the probability that group i adopts terrorist tactics as well as an
indirect or spatial spillover impact on all connected groups j. In other words, group
i’s use of terrorism influences groups j adoption, which feeds back again to group i in
Polo 1933

Table 3. Average Direct, Indirect, and Total Effects from Spatial Probit 1991 to 1999 and
2000 to 2009 (95 Percent Credible Intervals).

Variable Posterior Mean Lower Bound Upper Bound

Direct effects (domestic)

Discriminated 1991–1999 .144 .077 .218
Discriminated 2000–2009 .121 .055 .197
Indirect effects (diffusion)
Discriminated 1991–1999 .130 .066 .211
Discriminated 2000–2009 .128 .059 .216
Total effects
Discriminated 1991–1999 .274 .160 .401
Discriminated 2000–2009 .249 .124 .388

a recursive process. As Table 3 illustrates, the total effects of discrimination are

about twice as large as the direct effects, which means that half the average total
probability of terrorism for a group is actually due to spatial feedbacks from neigh-
boring units, namely, to diffusion. Failing to take into account these diffusion effects
through a spatial econometric analysis can therefore lead to a substantial overesti-
mation of the direct effect of domestic factors relative to diffusion effects from other
units.
Moreover, in all the estimated models, two types of exclusion, regional autonomy
and powerless, do not have a significant direct effect on terrorism (i.e., independent
of diffusion). As I further illustrate below, groups in similar circumstances are still
subject to diffusion from other excluded groups and will therefore emulate terrorist
tactics even when domestic incentives alone are insufficient. All in all, these results
provide additional, strong evidence in support of the theory.
The spatial probit allows us to estimate spatial effects for individual
groups. Since every group has a different network of connections, spatial
effects vary for each individual group. Figure 4, panel A, and Figure 5 present
disaggregated spatial effects for two ethnic groups, namely, the Kurds in
Turkey and the Sunni in Iraq, and their respective neighborhoods.18 They
show how an increase in the probability of terrorism for these groups posi-
tively influences other groups’ adoption of terrorist tactics, with different
colors indicating different probabilities of terrorism, based on the model esti-
mates. Moreover, while panel A in Figure 4 presents the predicted probability
of terrorism for the Kurds and their neighborhood based on the model, panel B
reports the ethnic groups who actually adopted terrorist tactics in the region
during the same time period. Several of the ethnic groups associated with an
increased probability of diffusion-related terrorism based on the model have in
fact adopted terrorist tactics.
I also examine the probability of terrorism for an ethnic group conditional on the
actual adoption of terrorist tactics by another excluded group in the region using the
1934 Journal of Conflict Resolution 64(10)

Figure 4. Spatial effects. (A) Spatial diffusion of the probability of terrorism following a shock
to the political status of the Kurds (Turkey) toward discrimination (model predictions 1991 to
1999). (B) Ethnic groups that actually adopted terrorist tactics between 1991 and 1999 in the
same region (for comparison with the model predictions).

Figure 5. Spatial diffusion of the probability of terrorism following a shock to the political
status of the Sunni (Iraq) toward discrimination (model predictions 2000 to 2009).

parametric simulation approach introduced by Franzese, Hays, and Cook (2016).19

Focusing specifically on Sunni groups in Iraq and Syria, I find that the actual use of
terrorism by Sunni organizations in Iraq increases the probability of terrorism for
Sunni in Syria by 28 percent. This finding is remarkable also because the political
status of the Sunni in Syria is powerless and all models suggest that a powerless
status by itself (i.e., in the absence of diffusion) is insufficient for spurring the choice
of terrorism. Yet, as I argued above, powerless groups are responsive to, and emu-
late, the terrorist choice of other, proximate excluded groups.
Polo 1935

Alternative Mechanisms and Robustness

To probe the diffusion mechanism and check the robustness of the results, I conduct
several additional tests. These tests are summarized in this section and discussed in
greater detail in the Online Appendix.
First, I have considered a series of competing mechanisms of diffusion. For
instance, terrorist diffusion may be the product of purely nonrelational (i.e., non-
spatial) channels, particularly those related with the media. To examine this, I rely
on a placebo test and reestimate the spatial probit using a connectivity matrix of
political exclusion between groups in the absence of geographic proximity. In this
model, the spatial autocorrelation coefficient r is negative and not significant. There
is therefore no empirical support for a global terrorism bandwagon among ethnic
groups. This provides further support for the combined effect of shared exclusion
and geographic proximity in fostering emulation.
To further probe the theorized mechanism, I conduct additional tests which
examine the role of political status in shaping groups’ responses to terrorism from
geographically proximate groups. Specifically, I interact a spatial lag of terrorism
from proximate ethnic groups with several variables indicating a recipient group’s
type and degree of political exclusion/inclusion. In line with the expectations, as
groups become more excluded, they also become increasingly responsive to prox-
imate terrorist examples and thus more likely to adopt terrorist tactics. In contrast,
terrorism from proximate groups that are politically included has no significant
effect on the likelihood that excluded groups emulate this tactic. Taken together,
these results help us rule out the possibility that terrorism arises from unobserved
characteristics shared by geographically proximate groups. This provides additional
evidence that mere geographic proximity is insufficient to generate emulation.
Instead, mutual identification through shared political exclusion is a critical condi-
tion for the diffusion of terrorism between groups.20
The apparent diffusion of terrorism may also result from generic spillover effects
from geographically proximate groups or conflicts, or from some omitted, spatially
clustered determinants of terrorism. To control for this, I include a spatial lag of all
terrorism in a group’s neighborhood. Consistent with the theory, this spatial lag is
not a significant predictor of an ethnic group’s adoption of terrorism. To further
demonstrate that geographic proximity is insufficient, on its own, to generate emu-
lation, I introduce two alternative spatial lags of terrorism by geographically prox-
imate nonethnic organizations, namely, leftist and purely religious groups.
Arguably, mutual identification between these organizations and ethnic ones should
be relatively low. In line with the theory, terrorism by geographically proximate
leftist and religious groups is not systematically associated with an ethic group’s
decision to adopt terrorism.
Another mechanism of diffusion, alternative to emulation, is represented by
competition between groups. To control for this, I employ a spatial lag which
excludes from the neighbors’ list all groups within the same country and only
1936 Journal of Conflict Resolution 64(10)

includes groups in neighboring countries. If group competition does occur, it should

only affect ethnic groups fighting against the same government, who may compete
to obtain concessions, rather than groups in different countries which fight against
different governments and have different domestic audiences. The main results hold
also with this alternative spatial lag; therefore, we can be confident that the results
are not capturing just a competition logic.
Second, I examine the robustness of the results to several alternative specifica-
tions of the spatial lag. Groups may be more likely to adopt terrorism if the number
of other terrorism adopters, rather than their weighted average, increases, or if the
reference groups conduct a greater number of attacks. Therefore, I reestimate the
models with spatial lags capturing the weighted sum of terrorism from politically
excluded and geographically proximate groups. I also construct an alternative spatial
lag which takes into account the magnitude of terrorist activity from other groups by
using a threshold of at least ten attacks carried out by each connected group in the
previous year and treating lower terrorist activity in the neighborhood as no activity
at all (i.e., 0). All substantive conclusions remain unchanged.
Third, it is possible that common exposure to observed or unobserved factors that
cluster geographically may drive the results. In this regard, the main results already
account for unobserved, time-invariant group-specific and country-specific factors
which may correlate with terrorism as well as many strong predictors of terrorism
such as civil war. Moreover, to capture any residual, time-invariant, and unobserved
factors that may influence terrorism and potentially also correlate with the spatial
lags, I reestimate the models including fixed effects. The coefficients for all the main
spatial lag variables remain positive and statistically significant.
Given the prominence of civil war as an explanation for terrorism, I conduct
analyses with additional controls for the clustering of civil war in a group’s neigh-
borhood and for the simultaneous presence of domestic and neighboring civil war.
I also reestimate the models dropping civil war cases. Substantive conclusions
remain unchanged and this indicates that the results are not driven by civil war
in the sending or receiving group or by common exposure to civil war in the
broader neighborhood. Moreover, if excluded groups are spatially clustered, then
terrorism may result from the simple clustering of conditions reflecting domestic
motives for terrorism, hence another form of common exposure. To control for
this, I construct a spatially weighted measure of the number of excluded groups
surrounding each ethnic group. The effect of terrorism diffusion remains
unchanged and this helps rule out the possibility of confounding due to spatially
correlated group-level factors.
Fourth, I reestimate the multilevel models accounting for possible changes in
technology and communication habits over time, which may influence patterns of
diffusion. I include a media density index (Warren 2014) as additional covariate and
interact this with the spatial lag of terrorism. The results provide no evidence of an
interactive effect with diffusion patterns. This is consistent with recent research by
Weidmann et al. (2016) which shows that governments often engage in digital
Polo 1937

discrimination by depriving of Internet access politically excluded ethnic groups in

order to limit their mobilization potential. Finally, common shocks may produce
clusters of terrorism even in the absence of diffusion; hence, I include year-specific
intercepts to control for temporal shocks that are common to all groups in a given
year (Neumayer and Plümper 2010). Substantive conclusions remain unchanged.

Conclusion
This article shows the importance of considering interdependence between terrorist
organizations and strategic emulation as a crucial mechanism leading to the adoption
of terrorist tactics. Despite a burgeoning literature on the causes of domestic terror-
ism, the vast majority of existing studies regards terrorist organizations as indepen-
dent of each other and the adoption of terrorist tactics as a purely “domestic”
decision, rooted in country-specific or individual-level attributes. Common accounts
of domestic terrorism, based on grievances and/or opportunities for violent mobili-
zation, have mainly assumed these to be domestically determined. This study
focuses instead on how the behavior of other groups can make group grievances
politically salient and shape a group’s perceptions of efficacy by mitigating the
resource, skills, and logistical constraints on the adoption of terrorist tactics. Rather
than simply testing for interdependence of terrorist activities between ethnic groups,
I introduce a specific mechanism of diffusion, based on emulation. Such emulation
is conditional on mutual identification based on shared grievances and on percep-
tions of terrorist efficacy. These are more likely to occur when sender and receiver
groups share a status of political exclusion and are geographically proximate. All
models provide strong support for the theoretical argument based on diffusion and
for the specified conditions under which terrorist emulation is more likely to occur.
The new data set introduced in this study has allowed to identify the specific
ethnic and ethnoreligious communities from which terrorist organizations could
emerge and to provide the first quantitative evidence for the diffusion of domestic
terrorism through emulation. At this point, it is difficult to study explicitly religious
terrorism because of the lack of cross-national data on the religious affiliations of
terrorist organizations and on societal groups other than politically relevant ethnic
groups. At the same time, this study, and its mechanism of diffusion, have important
implications also for the recent phenomenon of Jihadist radicalization in Western
Europe. A recent study by the Combating Terrorism Center at West Point—which
analyzed the profiles of nearly 2,000 European jihadist foreign fighters who traveled
to Iraq and Syria—highlights that “despite the growing attention to the role of virtual
recruitment, the majority of foreign fighters still relied on some form of interperso-
nal connection to make the decision to leave their home country and join the jihadi
movement” (Perliger and Milton 2016, 26-27). The study adds that “while virtual
propaganda can provide the initial cognitive opening for adopting the jihadi narra-
tive, a human connection is necessary to push the individual to actual activism, as
well as for logistical reasons for traveling to a war zone” (p. 31). Interestingly, while
1938 Journal of Conflict Resolution 64(10)

the individual backgrounds of European foreign fighters vary considerably, their

places of origin are highly clustered geographically, which further suggests the
importance of more traditional forms of recruitment, local recruitment hubs, and
in-person facilitators. This is in line with what argued in this article regarding the
specific role of direct ties in fostering the adoption of terrorism. All in all, the role of
spatial proximity—in combination with shared grievances—cannot be discounted,
even for jihadist terrorism.
There are also a number of fruitful avenues for future research. For example,
understanding how new technologies such as social media reinforce and interact
with more traditional organizational recruitment processes in sparking diffusion is
an important topic for future research. Future studies could also better integrate the
role of the government, especially state repression. Finally, terrorism comes in
different forms. This study and its methodology can be a stepping stone for future
efforts to understand the adoption and diffusion mechanisms of specific terror tac-
tics, such as improvised explosive devices (IED) bombings, vehicle ramming, and
suicide attacks.

Author’s Note
The data and code to reproduce the empirical results can be found on the Journal of Conflict
Resolution website http://jcr.sagepub.com/.

Acknowledgments
I am deeply grateful to Victor Asal, Navin Bapat, Jessica Maves Braithwaite, Chad Clay, Xun
Cao, Songying Fang, Federica Genovese, Kristian Gleditsch, Imke Harbers, Ryan Kennedy,
Ashley Leeds, Quan Li, Cliff Morgan, David Siegel, Hugh Ward, and the anonymous
reviewers for their helpful comments and suggestions and also grateful for support from the
Economic and Social Research Council (grant number ES/J500045/1).

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, author-
ship, and/or publication of this article.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship,
and/or publication of this article: This work was supported by Economic and Social Research
Council (grant number ES/J500045/1).

ORCID iD
Sara M. T. Polo https://orcid.org/0000-0002-4965-6022

Supplemental Material
Supplemental material for this article is available online.
Polo 1939

Notes
1. Domestic terrorism constitutes the vast majority of global terrorism. It is homegrown in
that the venue, main target, and perpetrators all belong to the same country. Transnational
terrorism involves more than a single country, either through its victims, targets, or
perpetrators (Enders and Sandler 2012). A complete definition of terrorism is provided
in the next section.
2. The French Basques and Corsicans both started in the late 1970s, the Baloch in the mid-
2000s, and the Kurds in the 1980s.
3. For an excellent discussion of differences between domestic and transnational terrorism,
see Enders, Sandler, and Gaibulloev (2011).
4. Research on civil war contagion has also focused primarily on linkages between countries
(e.g., Buhaug and Gleditsch 2008; Maves and Braithwaite 2013; Metternich, Minhas, and
Ward 2017).
5. Organizations with a religious ideology that do not make ethnic claims (e.g., al-Qaida in
the Islamic Maghreb, al-Shabaab) fall outside the scope conditions of this study.
6. The theoretical argument seeks to explain the group-level, organizational decision to
adopt terrorism rather than the choices of isolated individuals. Therefore, the so-called
lone wolf attacks are outside the scope of the theory.
7. This process applies not only to the choice of terrorism but also to specific terrorist tactics
such as bombings and suicide attacks.
8. Focusing on the case of refugees, Schmidt (2016) highlights that radicalization to violent
extremism and recruitment into terrorist groups becomes more likely where refugee
camps are in direct contact with fighters from an ongoing conflict (emphasis added).
9. This is an important departure from Tobler’s First Law of Geography, according to which
“everything is related to everything else but near things are more related than distant
things.” In other words, not all terrorist choices are related; mutual identification between
politically excluded groups plays a central role in terrorist diffusion processes.
10. For additional details on the coding rules and the data set, see Online Appendix C.
11. In Online Appendix A, I discuss the appropriateness, in this study, of binary measures of
terrorism versus attack counts.
12. I have used two different thresholds to define neighbors (see below).
13. The results are not affected by whether or not the matrix is row-standardized, see Online
Appendix A.
14. The dependent variable is temporally lagged to avoid simultaneity bias.
15. For identification, s2E is set to s2E ¼ 1 for probit.
16. In the spatial probit, which is cross-sectional, I control for a group’s history of terrorism
(Online Appendix B).
17. Online Appendix B describes all the convergence diagnostics performed.
18. The calculation of these spatial effects is particularly intensive. See Online Appendix B
for further details.
19. See Online Appendix B. Unfortunately, it is not possible to estimate these effects for all
groups simultaneously because this would require evaluating an n-dimensional integral.
1940 Journal of Conflict Resolution 64(10)

20. I also control for terrorism diffusion through transborder kinship ties. Substantive con-
clusions remain unchanged.

References
Asal, Victor, Mitchell Brown, and Marcus Schulzke. 2015. ““Kill Them All-Old and Young,
Girls and Women and Little Children”: An Examination of the Organizational Choice of
Targeting Civilians.” Political Science Research and Methods 3 (03): 589-607.
Asal, Victor H., Hyun Hee Park, R. Karl Rethemeyer, and Gary Ackerman. 2016. “With
Friends Like These . . . Why Terrorist Organizations Ally.” International Public Manage-
ment Journal 19 (1): 1-30.
Asal, Victor, and Karl R. Rethemeyer. 2008. “The Nature of the Beast: Organizational
Structures and the Lethality of Terrorist Attacks.” The Journal of Politics 70 (2): 437-49.
Bapat, Navin A., and Sean Zeigler. 2016. “Terrorism, Dynamic Commitment Problems, and
Military Conflict.” American Journal of Political Science 60 (2): 337-51.
Bove, Vincenzo, and Tobias Böhmelt. 2016. “Does Immigration Induce Terrorism?” The
Journal of Politics 78 (2): 572-88.
Braithwaite, Alex, and Tiffany S. Chu. 2018. “Civil Conflicts Abroad, Foreign Fighters, and
Terrorism at Home.” Journal of Conflict Resolution 62 (8): 1636-60.
Braithwaite, Alex, and Quan Li. 2007. “Transnational Terrorism Hot-spots: Identification and
Impact Evaluation.” Conflict Management and Peace Science 24 (4): 281-96.
Bueno de Mesquita, Ethan, and Eric S. Dickson. 2007. “The Propaganda of the Deed: Terror-
ism, Counterterrorism, and Mobilization.” American Journal of Political Science 51 (2):
365-81.
Buhaug, Halvard, and Kristian Skrede Gleditsch. 2008. “Contagion or Confusion? Why
Conflicts Cluster in Space.” International Studies Quarterly 52 (2): 215-33.
Carter, David B., and Curtis S. Signorino. 2010. “Back to the Future: Modeling Time
Dependence in Binary Data.” Political Analysis 18 (3): 271-92.
Cederman, Lars-Erik, Kristian Skrede Gleditsch, and Halvard Buhaug. 2013. Inequality,
Grievances, and Civil War. New York: Cambridge University Press.
Cederman, Lars-Erik, Brian Min, and Andreas Wimmer. 2010. “Why Do Ethnic Groups
Rebel? New Data and Analysis.” World Politics 62 (1): 87-119.
Checkel, Jeffrey T., ed. 2013. Transnational Dynamics of Civil War. Cambridge, UK:
Cambridge University Press.
Choi, Seung-Whan, and James A. Piazza. 2014. “Ethnic Groups, Political Exclusion and
Domestic Terrorism.” Defence and Peace Economics 27 (1): 37-63.
Cunningham, Kathleen, Marianne Dahl, and Anne Frugé. 2018. “Strategies of Resistance:
Diversification and Diffusion.” American Journal of Political Science 61 (3): 591-605.
Cunningham, Kathleen Gallagher, and Katherine Sawyer. 2017. “Is Self-determination Con-
tagious? A Spatial Analysis of the Spread of Self-determination Claims.” International
Organization 71 (3): 585-604.
Enders, Walter, and Todd Sandler. 2012. The Political Economy of Terrorism. Cambridge,
MA: Cambridge University Press.
Polo 1941

Enders, Walter, Todd Sandler, and Kusrav Gaibulloev. 2011. “Domestic versus Transnational
Terrorism: Data, Decomposition, and Dynamics.” Journal of Peace Research 48 (3):
319-37.
Findley, Michael G., and Joseph K. Young. 2011. “Terrorism, Democracy, and Credible
Commitments.” International Studies Quarterly 55 (1): 1-22.
Findley, Michael G., and Joseph K. Young. 2012. “Terrorism and Civil War: A Spatial and
Temporal Approach to a Conceptual Problem.” Perspectives on Politics 10 (2): 285-305.
Franzese, Robert J. Jr., Jude C. Hays, and Scott J. Cook. 2016. “Spatial and Spatio-temporal
Autoregressive Probit Models of Interdependent Binary Outcomes.” Political Science
Research and Methods 4 (1): 151-73.
Galton, Francis. 1889. “Comment on Edward B. Tylor’s ‘On a Method of Investigating the
Development of Institutions, Applied to Laws of Marriage and Descent’.” Journal of the
Royal Anthropological Institute 18 (3): 245-69.
Gleditsch, Kristian Skrede. 2002. “Expanded Trade and GDP Data, 1946-99.” Journal of
Conflict Resolution 46 (5): 712-24.
Hill, Stuart, Donal Rothchild, and Colin Cameron. 1998. “Tactical Information and The
Diffusion of Peaceful Protests.” In The International Spread of Ethnic Conflict: Fear,
Diffusion, and Escalation, edited by David Lake and Donald Rothchild, 61-88. Princeton,
NJ: Princeton University Press.
Horowitz, Michael C. 2010. “Nonstate Actors and the Diffusion of Innovations: The Case of
Suicide Terrorism.” International Organization 64 (1): 33-64.
Krueger, Alan. 2007. What Makes a Terrorist: Economics and the Roots of Terrorism. Prin-
ceton, NJ: Princeton University Press.
Kuran, Timur. 1998. “Ethnic Dissimilation and its International Diffusion.” In The Interna-
tional Spread of Ethnic Conflict: Fear, Diffusion, and Escalation, edited by David Lake
and Donald Rothchild, 35-60. Princeton, NJ: Princeton University Press.
Lake, David A., and Donald Rothchild. 1998. The International Spread of Ethnic Conflict:
Fear, Diffusion, and Escalation. Princeton, NJ: Princeton University Press.
LeSage, James P. 2013. “Bayesian Estimation of Limited Dependent Variable Spatial Auto-
regressive Models.” Geographical Analysis 31 (1): 19-35.
LeSage, James P., and R. Kelley Pace. 2009. Introduction to Spatial Econometrics. Boca
Raton, FL: CRC Press/Taylor and Francis.
Maves, Jessica, and Alex Braithwaite. 2013. “Autocratic Institutions and Civil Conflict Con-
tagion.” The Journal of Politics 75 (2): 478-90.
Metternich, Nils W., Shahryar Minhas, and Michael D. Ward. 2017. “Firewall? or Wall on
Fire? A Unified Framework of Conflict Contagion and the Role of Ethnic Exclusion.”
Journal of Conflict Resolution 61 (6): 1151-73.
Midlarsky, Manus, Martha Crenshaw, and Fumihiko Yoshida. 1980. “Why Violence Spreads.
The Contagion of International Terrorism.” International Studies Quarterly 24 (2):
262-98.
Nemeth, Stephen C., Jacob A. Mauslein, and Craig Stapley. 2014. “The Primacy of the Local:
Identifying Terrorist Hot Spots Using Geographic Information Systems.” The Journal of
Politics 76 (2): 304-17.
1942 Journal of Conflict Resolution 64(10)

Neumayer, Erik, and Thomas Plümper. 2010. “Galton’s Problem and Contagion in Interna-
tional Terrorism along Civilizational Lines.” Conflict Management and Peace Science 27
(4): 308-25.
Pape, Robert A. 2003. “The Strategic Logic of Suicide Terrorism.” American Political Sci-
ence Review 97 (3): 343-61.
Perliger, Arie, and Daniel Milton. 2016. “From Cradle to Grave: The Lifecycle of Foreign
Fighters in Iraq and Syria.” Report—Combating Terrorism Center at West Point. West
Point, NY: Combating Terrorism Center.
Sageman, Marc. 2004. Understanding Terror Networks. Philadelphia: University of Pennsyl-
vania Press.
Schmidt, Alex P. 2016. “Links between Terrorism and Migration: An Exploration.” Report—
International Centre for Counter-Terrorism, The Hague. The Hague, the Netherlands:
International Centre for Counter-Terrorism.
Shipan, Charles R., and Craig Volden. 2008. “The Mechanisms of Policy Diffusion.”
American Journal of Political Science 52 (4): 840-57.
Simmons, Beth A., Frank A. Dobbin, and Geoffrey Garrett. 2008. The Global Diffusion of
Markets and Democracy. New York: Cambridge University Press.
Simmons, Beth A., and Zachary Elkins. 2004. “The Globalization of Liberalization:
Policy Diffusion in the International Political Economy.” American Political Science
Review 98 (1): 171-89.
Smith, Meagan, and Sean M. Zeigler. 2017. “Terrorism before and after 9/11—A more
Dangerous World?” Research & Politics 4 (4): 1-8.
Tilly, Charles. 1978. From Mobilization to Revolution. New York: McGraw-Hill.
Tilly, Charles. 2003. The Politics of Collective Violence. New York: Cambridge University
Press.
Tversky, Amos, and David Kahneman. 1974. “Judgement under Uncertainty: Heuristics and
Biases.” Science 185 (4157): 1124-31.
Vreeland, James Raymond. 2005. “A Problem with Polity: Unpacking Anocracy.” Typescript,
Department of Political Science, Yale University, New Haven, CT.
Warren, T. Camber. 2014. “Not by the Sword Alone: Soft Power, Mass Media, and the
Production of State Sovereignty.” International Organization 68 (1): 111-41.
Weidmann, Nils B., Suso Benitez-Baleato, Philipp Hunziker, Eduard Glatz, and Xenofontas
Dimitropoulos. 2016. “Digital Discrimination: Political Bias in Internet Service Provision
across Ethnic Groups.” Science 353 (64): 1151-54.
Wilhelm, Stefan, and Miguel Godinho de Matos. 2013. “Estimating Spatial Probit Models
in R.” The R Journal 5 (1): 130-43.
Wucherpfennig, Julian, Nils Metternich, Lars-Erik Cederman, and Kristian Skrede Gleditsch.
2012. “Ethnicity, the State and the Duration of Civil War.” World Politics 64 (1): 79-115.
Wucherpfennig, Julian, Nils B. Weidmann, Luc Girardin, Lars-Erik Cederman, and Andreas
Wimmer. 2011. “Politically Relevant Ethnic Groups across Space and Time: Introducing
the GeoEPR Dataset.” Conflict Management and Peace Science 28 (5): 423-37.