Doctrines of Proxies: How Rebel Ideology and Foreign Support Affect Violence Against Civilians in Civil Wars

Full text

Doctrines of Proxies: How Rebel Ideology and
Foreign Support Affect Violence Against Civilians
in Civil Wars
Vujo Ilic1and Milos Popovic∗2
1Central European University
2Columbia University
Abstract
Existing research suggests that extreme ideologies and foreign support, respec-
tively, make rebel groups particularly violent against civilians. However, scholars
have paid less attention to the interplay of group ideology and external support in
producing varying levels of violence. In this paper, we examine whether certain
political goals of armed groups such as the transformation of society versus the
preservation of the pre-existing order, may encourage militant groups in civil con-
flicts to exercise restraint, curtailing their indiscriminate violence against civilians,
despite the access to external state support. We draw on cross-country data for
1989–2013 from UCDP and EACD as well as other sources to test the hypotheses. We
find that when revolutionary groups receive support they are less likely to engage in
civilian targeting than their conservative counterparts. Our findings remain robust
to the disaggregation of foreign support into military and non-military. Our findings
speak to debates in the wider literature on political violence and external support in
civil conflicts, and contribute to human rights and mediation efforts of third parties
in ongoing civil conflicts.
Keywords
civil war, violence, ideology, foreign support
Word count: 7940
∗Vujo Ilic
is a PhD Candidate in the Department of Political Science at Central European University
and a former pre-doctoral fellow in the Program on Order, Conflict, and Violence at Yale University.
Address: Department of Political Science, Central European University, Budapest, Hungary. email:
ilic_vujo@phd.ceu.edu
Milos Popovic is a Research Consultant for Columbia University on a Minerva project.
Corresponding author: Milos Popovic, Columbia University, New York City, USA;
email: milos.agathon@gmail.com

Introduction
Emerging research on violence in civil wars suggests that externally-backed insurgencies
are a harbinger of civilian targeting (Humphreys and Weinstein 2008;Malaquais 2001;
Salehyan et al. 2014;Weinstein 2006;Wood et al. 2012). While foreign support may
fuel the rise of brutal warlords across the globe, there is a plethora of foreign-sponsored
rebel groups that refrained from civilian victimization. We argue that foreign support
alone cannot account for this variation. Explaining the variation in violence requires
understanding rebel goals, too. We suggest that the interaction between means (foreign
support) and goals (ideology) helps explain why the capacity to do harm is not always
employed.
The capacity of rebels may come from two sources — from within and from outside.
First, rebels can increase their power by strengthening their ranks and surpassing the
organizational capacities of the state. This is largely affected by what goals they are
trying to achieve. On the other hand, assistance to rebels can come from outside the
internal conflict, through third party involvement, which might up the means they have
on their disposal. The ideologies of rebel groups can be associated with the quality of
relations they maintain with civilian population. Some ideologies prescribe behavior
and institutions, which lead to more restraint towards civilians than others. Similarly,
foreign assistance, which comes in different shapes, affects the power relations and
could also impose political constraints on the behavior of armed groups. In this paper,
we aim to understand the interaction between these factors, and the effect it has on
violence against civilians.
To test our argument, we draw on cross-national data for 1989–2013 from UCDP
and EACD, as well as other sources. Our key finding is that foreign-backed revolutionary
groups are less likely to engage in civilian targeting than their conservative counterparts.
This finding remains robust to the disaggregation of foreign support into military and
non-military.
2

The main contribution of this paper is that—unlike much of the literature on political
violence in civil wars—it seeks to bring together ideology and capabilities in explaining
civilian targeting. We suggest that foreign support often comes with ideological strings
attached, which may have a profound effect on the behavior of rebel groups. Our
secondary contribution is in the policy area. This paper offers a counter-intuitive
suggestion that revolutionary groups such as ISIS might have been as brutal as they
were because of the absence of overt backing by foreign sponsors.
The structure of the paper is the following: we first briefly summarize the literature on
ideology, foreign support, and violence, then we move to hypotheses and the description
of our data, and follow with the presentation of the models and results. We finish the
paper with discussion and with suggestions for future research.
Ideology and Violence
The end of the Cold War saw a shift from numerous Marxist insurgencies to the wave
of ethnic conflicts in the 1990s, which for a while made research on ideology stagnant.
As ethnicity was losing its explanatory appeal (Fearon and Laitin 2003), the economic
dimensions of civil war came to the forefront (Collier 2000). However, recent advances
in the civil war literature have emphasized the organizational and institutional aspects
or rebellions (Mampilly 2011;Staniland 2014), which in turn paved the way for a new
assessment of the role of ideology in civil wars (Sanín and Wood 2014), perhaps also
driven by considerations of ideologies of armed groups such as ISIS (Kalyvas 2015).
In the past research, the role of ideology was often seen as nothing more than
’window dressing’ of other motivations, but contemporary analyses increasingly point to
the relevance of ideology for armed conflicts (Ugarriza and Weintraub 2015). There are
several different ways in which ideologies can affect civil war dynamics. For instance,
they can prescribe certain types of military strategies. Marxist ideology is usually associ-
ated with a certain type of warfare (irregular), or certain institutional configurations,
such as having political and military wings. Ideology can have an instrumental value for
3

armed groups, although normative commitments also matter — members of some armed
groups, as well as whole groups, act based on normative concerns prescribed by their
ideology (Sanín and Wood 2014). But the ideology matters for outcomes, too. Balcells
and Kalyvas (2015) explain the ’Marxist paradox’ in which, contrary to expectations,
Marxist groups ended up more often on a losing side, as the type of threat posed by
these rebellions indirectly strengthened the states they were set against.
It should not come as a surprise that ideology is shown to be related to patterns
in targeting of military or civilian objectives. More ideologically committed fighters
in Columbia were less likely to attacks civilians and more likely to attack military
targets (Ugarriza and Weintraub 2015). Based on a case study of Mozambique, Thaler
(2012) shows that groups which practice restraint towards civilians can experience an
increase in victimization based solely on a breakdown in ideological commitment among
elites. The effects of ideology-induced restraint among rank and file is lost through
indiscriminate recruitment, the lack of socialization, and a loss of discipline caused by
ideological inconsistency of the leadership.
Drawing on the comparison of nationalist and Islamist groups in the North Caucasus
Toft and Zhukov (2015) show that counterinsurgency based on selective tactics elicits
compliance from nationalist rebels, but it fails to corner Islamist rebels. This is arguably
due to different structures of rebel social bases. While the nationalists rely on local
sources of support, the Islamists are more dependent on foreign backing. This makes
the latter more capable to continue fighting even when counterinsurgency targets local
supporters. However, this study finds no evidence of difference in Islamist and nationalist
patterns of violence.
In sum, existing evidence points to an ambiguous conclusion: group ideology may
and may not affect restraint toward civilians. Reliance on local or foreign support
appears to play a role in group survival, but the mechanism leading to violence is still
blurry. To properly address this issue, we first discuss the effect of external support on
violence against civilians.
4

External Support and Civilian Victimization
Some of the most brutal civil wars since 1989 have featured externally-backed rebel
groups. For instance, with the direct military support of Rwanda and Uganda, Tutsi
Alliance of Democratic Forces for the Liberation of Congo-Kinshasa (AFDL) killed more
than 35,000 civilians in the rural regions of Zaire in only two years of fighting. Similarly,
Serbian Republic of Bosnia’s forces — which received weapons, para-military and regular
forces from the Yugoslav government — killed more than 3,000 civilians in 1992/1992,
and massacred another 8,000 in Srebrenica in a few days.
Weinstein (2006) suggests that external resources attract low-committed individ-
uals who are primarily interested in profit and who have little regard for the lives or
livelihoods of civilians, or in establishing a stable system of long-term tax collection.
In contrast, rebel groups that operate in resource-poor environments absent foreign
support are highly dependent on the local population for survival. These will employ
a "stationary bandit" approach, and attract high-commitment recruits who — driven
by a strong belief in the organization’s ideology — will employ violence selectively.
For instance, the access to foreign support alienated UNITA from the local population
making it more hostile to civilians (Malaquais 2001).
Existing research demonstrates that rebel groups that receive material support from
foreign sponsors are more prone to violence toward civilians (Humphreys and Weinstein
2008;Wood 2010;Salehyan et al. 2014). The prevailing explanation portrays externally-
backed rebel groups as disinterested in civilian support, and even hostile toward the
population because they have means to do so. This account suggests that civilians are
passive bystanders and literally useless to the rebels, which is rarely the case in civil wars
(Kalyvas 2006). It is unclear how rebel groups use resources, and why opportunistic
organizations choose coercion over other strategies. It is possible that the loss of foreign
support may also encourage violence. For instance, Metelits (2009) shows how the loss
of support from Ethiopia matched with SPLA attacks on civilians at the end of the Cold
5

War.
One stream of this argument focuses on the supply side of this relationship. Sponsors
may encourage rebel groups to brutalize civilians. In this scenario, violence either serves
rebels to show their usefulness to the sponsor or the sponsor demands their clients to
spread violence (Salehyan et al. 2014). Either way, violence is driven by the need to
maintain the flow of resources from the sponsor who may or may not support such a
policy depending on its political regime. Democracies are likely to discourage the use
of violence because they are more committed to human rights protection, constrained
by domestic institutions and sensitive to public opinion (Ibid). Absent such domestic
constraints, authoritarian sponsors, on the other hand, may either turn a blind eye to
rebel abuses or support them.
Another approach to explaining the link between external support and violence
sheds light on the demand side. Accordingly, foreign support improves rebel capabilities,
encouraging them to change their strategy toward civilians. Military intervention on
the side of insurgents should decrease their costs associated with fighting, territorial
control and civilian loyalty. In turn, this may diminish the need for excessive coercion
toward civilians since insurgents can invest more resources into governance structures
(Wood et al. 2012, 653). Deprived of direct military support or faced with a foreign-
backed government, weakened rebels are more likely to use coercion to maintain civilian
cooperation. Therefore, military intervention in favor of insurgents should decrease
their indiscriminate violence against civilians.
We acknowledge that the presence of foreign support may affect the strategies of
combatants by providing them the means to shape their relationship with civilians. The
Greek Civil War shows, for example, that the Communists were more inclined toward
civilian victimization before 1944 when they lacked foreign sponsors than after World
War II when they received outside assistance from Albania, Bulgaria, and Yugoslavia
(Kalyvas 2007). But we also suggest that to fully understand how rebel behave toward
civilians we need to understand their goals. To paraphrase the constructivists, how
6

one views oneself is likely to affect what one wants. Similarly, rebels have different
expectations about their role in a conflict. Some are interested in keeping the dominant
social order while seizing the power over the entire or a part of the territory. Others
want to radically change the social, political, and economic system epitomized in their
opponent. We suggest that foreign support (means) in interaction with these views
(goals) should produce different behavior toward civilians.
In the subsequent section, we discuss: 1) why some groups are more restrained in
their interaction with civilians due to particular ideologies they adhere to; 2) how much
this restrain depends on the content of the ideology, and how much this is affected by
groups’ reliance on domestic or foreign support in achieving their military-political goals.
External Support, Ideology and Civilian Victimization
Depending on whether they are trying to transform the society, or trying to maintain
the existing relations, we differentiate between revolutionary and conservative groups.
We argue that these different societal transformation goals lead to different preferences
when it comes to political and military objectives in civil wars, which in turn make
groups prioritize their resources in different ways, and also varies their reliance on
foreign support.
The revolutionary group’s goal is the radical transformation of society. The first impli-
cation is that the political outcomes of the war might be equally significant, or even take
precedence over direct military outcomes. Taken further, these groups invest significant
resources in building their relations with the parts of the society they claim to represent,
or whose political transformation they aim to induce. Take for example the political
commissars of various Marxist armed groups, or various institutions run by Islamist
groups, which are based on Quranic prescriptions for functioning of social endowments.
Since the revolutionary group prioritizes political objectives in circumstances where it
cannot rely on tapping in existing domestic pools of support, it should exercise high
restraint and have high need for foreign assistance.
7

Conservative rebel groups have no strategic goals of transforming society. They do
not call for any drastic changes to the societal relations, introduction of new institutions,
or policies; there is a relative equilibrium between state and society that should be
maintained. Their conservatism is more pragmatic than ideological. This, in turn —
compared to revolutionary groups — should free them from investing resources in
transforming the society. In other words, military objectives come first, political comes
second. We understand conservative groups as a wide category: from armed groups
countering insurgency by revolutionary actors, to the bulk of nationalist-secessionist
groups. If the group’s goal is primarily to carve a new country, but to maintain the same
or similar social structure as before, then we regard this as a conservative group. As
conservative groups tend to feed on existing societal networks as well as institutions,
rather than trying to build new ones, we predict a higher proclivity toward civilian
victimization and lower dependence on foreign assistance.
Since the reliance on foreign support should be higher for revolutionary than the
conservative groups, we argue that military support should make the former more
restrained towards civilians. The foreign military support to revolutionary groups makes
them able to channel resources to their political project, and invest resources in a
transformation of the society. If such a support is mainly political, we might expect
this to have an opposite effect, in two ways. The non-military support to revolutionary
groups does not directly boost the war effort, which might lead to the weakening of their
relative position vis-à-vis other parties in the war. The need to prioritize the military
aspect might lead to either a breakdown in ideological commitment of the army, thus
leading to less restraint towards civilians as a resource in the war, or to a radicalization of
the group as an attempt to bolster internal cohesion. The latter, in turn, might generate
radical approaches to the transformation of the society, making civilians more exposed
to drastic measures of social manipulation.
For conservative groups, military support is likely to decrease even more the need for
reliance on the population for political objectives, thus further decreasing the restraint
8

towards civilians. However, we expect non-military support for conservative groups
to come primarily from conservative nation states. Such support should entail greater
political constraints to their behavior, than what we would expect from anti-systemic
sponsors. Based on these theoretical expectations, we make the following hypotheses:
H1: Revolutionary groups are less likely to target civilians when they receive
foreign support than conservative groups that receive foreign support.
H2: Military support is likely to increase restraint of revolutionary groups
and decrease restraint of conservative groups.
H3: Non-military support is likely to increase restraint of conservative and
decrease restraint of revolutionary groups.
Data and Research Design
To test our hypotheses, we examine data on one-sided violence in civil conflicts from
the Uppsala Conflict Data Project (UCDP) (Eck and Hultman 2007). These data are
appropriate for our analysis because they measure violence against civilians as the
number of intentional and direct killings of non-combatants by rebel groups. Figure
1 shows the natural logarithm of civilian victims in our data; when displayed in raw
numbers the distribution is highly skewed in favor of no violence or no victims (790 out
of 1113 observations have ’zero’ outcome). The data include all armed clashes between
a government and a rebel group that have resulted in 25 or more battle-related deaths
from 1989 to 2013 (Harbom et al. 2008). In total, we identify 168 such dyads for the
given period, resulting in a total of 1113 dyad-years. To our knowledge, this is one of
the rare analysis involving an up-to-date dataset on civilian victims. Previous analyses,
including the most recent ones such as Salehyan et al. (2014) and Wood (2014), use
UCDP data on violence until 2009 and 2008, respectively.
We expect that our reliance on the most recent UCDP dataset (version 2014) might
yield slightly different but more robust results.
Figure 1 here
9

Regarding our main independent variable, we create our measure of conservative/revolutionary
ideology in relational terms: to what extent the government and rebels share their ideas
about the existing socio-political order? We first determine the organizational ideology
of a group based on the concept from the Minorities at Risk Organizational Behavior
(MAROB) dataset (Asal et al. 2008). MAROB distinguishes between four ideal types
based on whether the focus is on material (right-wing vs. left-wing ideologies) or nor-
mative goals for social change (religious vs. ethno-nationalist ideologies). Right-wing
ideologies advocate free market policies or policies that benefit economic/financial
elites with or without calls for democratic rule, while left-wing groups promote the
re-distribution of wealth. On the other hand, religious groups fight for religious reasons
and/or to establish the state based on a religious doctrine, whereas ethno-nationalist ide-
ologies seek to carve out a territorial space for an ethnic group. Since MAROB provides
information only for insurgencies in the Middle East and North Africa, we also use other
fairly compatible and widely used datasets such as EACD and START. We compare our
findings with another frequently used database, the UCDP Conflict Database, in search
of contradictory or unclear cases. Finally, we also double-check our coding of religious
movements using a PRIO dataset on Islamist groups (Gleditsch and Nordås 2014). Figure
2 indicates that ethno-nationalist ideology is by far the most dominant rebel doctrine
with religious groups placed second in front of the material-driven doctrines. This is
most likely due to a selection effect as our data covers the post-1989 period in which
the dilemma between socialism and capitalism is not the main geopolitical repertoire in
town.
Figure 2 here
Next, we identified the ideology of the government, drawing on ’execrlc’ variable from
the Database of Political Institutions (Beck et al. 2001). The advantage of this variable
is that it is composed of the very same four types as our variable rebel ideology. This
allows us to easily determine whether the sponsor and rebel ideology match. Next,
we compared the ideology of the group with the ideology of the government. If the
10

government had no ideology or different ideology from the rebel group, goal was coded
as revolutionary or 1, and status quo or 0 otherwise. If neither the government nor
the opposition followed an identifiable ideology, goal was coded as status quo. Ethno-
nationalist groups were coded as status quo unless they simultaneously followed other
ideologies. For example, the Kurdish Democratic Party (KDP) in Iraq, which followed
both ethno-nationalist and left-wing ideologies was coded as revolutionary group. In
contrast, the Moro National Liberation Front (MNLF) in the Philippines which promoted
no ideology other than ethno-nationalist, was coded as status quo. Revolutionary
and conservative groups are almost evenly matched in our data with a total of 524
observations for revolutionary groups and 589 for conservative.
To code foreign support for insurgents, we draw on UCDP External Support Dataset,
1975–2009 (Högbladh et al. 2011). Given that our primary focus is on the provision
of material support, this dataset is the best available fit for our analysis. Another
advantage of this dataset is that the presence of foreign support is recorded for any
given year, allowing for a more fine-grained analysis. UCDP defines external support as
the provision of warring or non-warring means by a government of an internationally
recognized country with an intent of aiding a warring side in an ongoing conflict (Ibid).
This support can take a form of the provision of weapons, funding, sanctuary, logistics,
training, intelligence, regular troops and other types of material support. We draw on
these various forms of support to disaggregate them into two types of support — military
and non-military. Military support includes assistance that is lethal or can be easily
transformed into lethal such as access to military or intelligence infrastructure, weapons,
funding and troops. Non-military support involves aid that is largely non-lethal in nature
or has no direct influence on rebels’ lethal component — sanctuary, materiel/logistics,
training, intelligence and other forms of support. Figure 3 displays the distribution
of types of support in our data. Military support has been more frequent source of
assistance than non-military support even though they both are relatively rare compared
to cases without foreign interference.
11

Figure 3 here
We also use several control variables that can be roughly divided into three categories.
First, we include variables that capture conflict dynamics between the government and
rebels. One such factor is government violence. Previous research shows that government
brutalization of civilians may affect levels of rebel violence. To measure this variable,
we use the natural logarithm of the number of civilians killed by the government forces.
Similarly, multiparty conflicts should yield more civilian victims as groups should adopt
violence to outbid each other (Bloom 2004,2005). We measure this variable using
UCDP Armed Conflict Dataset.
Second, we control for the characteristics of the conflict environment. Previous
research demonstrates that combatants have greater incentives to target civilians in
more intense conflicts (Downes 2006;Hultman 2007). Intensity is coded 1 for years in
which the conflict reached or exceeded 1000 battle-related deaths, and 0 otherwise. We
borrow this measure from UCDP Dyadic Dataset (Harbom et al. 2008). It is possible
that violence against civilians might be affected by previous violence. To account for
this temporal dependence we lag our dependent variable by one year and include it in
the analysis. Violence against civilians should also decrease over time as belligerents
realize their counterproductive effects (Kalyvas 2006). Duration is measured as the
logged number of years since the outset of a conflict. Next, incompatibility of conflict
may affect the level of violence in civil war. Previous studies find a relationship between
non-territorial conflicts and more violence. Incompatibility is coded 1 for non-territorial
and 0 for territorial conflicts (Harbom et al. 2008). We expect the association between
non-territorial conflicts and violence to be positive.
Third, nature of the conflict-ridden country may fuel brutalization of civilians. Previ-
ous research has found that democracies should experience more violence because they
are casualty-averse and constrained by human rights norms; the perpetrators might
perceive these limitations as a disability of democracies to punish their acts (Goodwin
2006;Pape 2005). We should, therefore, expect more rebel violence under democratic
12

than non-democratic regimes. We include a binary variable that denotes whether the
ruling regime is democratic (takes the value of 1) or non-democratic (takes the value of
0). Using the Polity IV scale (Marshall and Jaggers 2007), the government is coded as
democratic if its score was greater than 5 in the first year of conflict, and non-democratic
when the score was below 5. We also consider the size of the conflict state’s population.
More populous countries should create greater opportunities for violence. This measure
is borrowed from Correlates of War National Capabilities dataset (Singer et al. 1972).
Model and Results
Since only a fraction of conflicts experienced civilian targeting in the last two and a
half decades (approximately 30 percent of cases), the dependent variable suffers from
overdispersion. Standard Poisson-based count model is not appropriate because our
data violates its key assumption of normality of ordinary least squares. This can lead to
estimates that are inefficient, inconsistent, and biased (Long 1997, 217–238). To account
for overdispersion, we could either use negative binomial (NB) or zero-inflated negative
binomial (ZINB). NB accounts for the non-normal distribution and overdispersion of
the dependent variable. ZINB consists of two steps: in the first, the model predicts the
number of victims similar to NB (’count model’), while in the second the probability that
there will be no victims (’inflation model’) (Walker et al. 2009). We run our models
using both NB and ZINB and compare their fit using the Vuong test. Since the results of
the tests are in favor of NB we use this model for our analysis.
In the parametrization that follows, we present exponentiated coefficients—i.e.
incidence rate ratios (IRR), and their confidence intervals
1
—to explore the relationship
between rebel violence and given predictors. IRR allow us to determine the ratio by
which violence changes for a unit change in any explanatory variable; it is an intuitive
approach to interpreting the regression coefficients in count models (Hilbe 2011).
1
Following American Statistical Association’s (ASA) suggestion to avoid p-values in favor of other
approaches (Wasserstein and Lazar 2016), we choose confidence intervals to present our findings.
However, we also present coefficients with standard errors and p-values for each of our models in the
Appendix.
13

In Figure 4, we first assess H1 by interacting foreign support with ideology (conser-
vative groups are a baseline category), including control variables. Our figure displays
the IRR with 90 and 95 percent confidence intervals, respectively. It is worth noting that
the observed relationship is regarded as "significant" when both the lower and upper
confidence bars are located on the same side relative to the line of no effect, i.e. either
on the right or left. The findings in Model 1 offer a tentative support for H1.
Figure 4 here
As envisioned, the interaction effect ("Ideology:Support") in Model 1 suggests that
foreign support for revolutionary groups is less likely to incite more violence than
foreign support for conservative groups. In substantive terms, the effect of foreign
support for revolutionary groups is 0.31 times that for conservative groups. This
means that foreign-backed revolutionaries are expected to target nearly 70 percent
fewer civilians than their conservative counterparts, with the remaining variables held
constant. The confidence intervals of the first interaction term ("Ideology") indicate that
being a revolutionary group is not significantly different from being conservative group
regarding violence. Finally, the other interaction term ("Support") demonstrates that
violence rate is significantly higher for rebel groups with foreign support irrespective
of their ideology, but this encompasses the lowest volume of cases (widest confidence
interval).
To determine substantial effects of the interaction, we compute the interaction effects
of "Ideology:Support" from Model 1. Figure 5 presents these findings. "Ideology" is
displayed on x-axis, while the predicted counts of civilian victims are on the y-axis.
Figure 5 shows the difference in the predicted number of civilian deaths between groups
that have foreign support and those that lack it. According to the results on the left
hand side of the panel, both revolutionary and conservative groups commit less violence
against civilians when they receive no support. Revolutionary groups seem to be slightly
less violent. A more interesting result for our argument is when both types receive
external support (the right hand side of the panel). As we anticipate, foreign support
14

radicalizes revolutionary groups much less than their conservative competitors. In fact,
foreign-backed conservative groups are likely to brutalize approximately 4 times more
civilians than foreign-backed revolutionaries. A small change in predicted violence for
revolutionary groups on both sides of the panel suggests that external support is less
important for these groups.
Figure 5 here
Moving to conflict-related variables, we find that our model mostly supports previous
studies. As expected, more government violence is linked to more rebel violence. Ceteris
paribus, for a one-unit increase in government violence, the rate of rebel violence
increases by 10 percent. In contrast, multiparty conflicts are associated with less violence
in accord with previous research. We find that—all else equal—multiparty conflicts are
expected to produce one third fewer civilian victims than single-party conflicts.
Among the situational variables, our findings offer support to previous research.
Conflict intensity has a much higher effect on violence than any other predictor, in-
dicating that as battlefield costs increase, rebels are likely to increase their coercion
against civilians. Previous violence is also associated with subsequent levels of civilian
victimization and relevant in the base model. In contrast, for every increase in duration
there seems to be no significantly higher rise in violence. As suggested by previous re-
search, incompatibility demonstrates an effect on civilian targeting in that non-territorial
conflicts are associated with more violence. In particular, we find that civilian targeting
is more than 3 times higher for non-territorial conflicts.
Country-related specifics offer similar results to previous research. We find that demo-
cratic regimes are likely to experience 2.3 times more violence than are authoritarian
regimes. In contrast, we find—congruent to previous research—that an increase in
population holds no practical significance for understanding civilian targeting.
We now turn to our last set of hypotheses related to disaggregated support and
ideology displayed in Figure 6. Models 2–3 explore whether there is any difference in
the impact of military/non-military support interacted with group ideology on civilian
15

targeting. These models are used to test H2 and H3, which expected that military
support will have a varying impact on violence depending on a group’s ideology.
Figure 6 here
The findings in Model 2 and Model 3 fail to corroborate these hypotheses. Each of these
models suggests that there is only a slight difference between the impact of military
and non-military support on violence. Even though military support yields a somewhat
higher estimate, both types of support are essentially associated with more civilian
deaths. We notice, however, that unlike the aggregated variable, the constitutive term
for military and non-military support alike yield much greater effects. When we interact
the disaggregated support with ideology, the results are similar to those Model 1, in
that support for revolutionary groups decreases violence, while support for conservative
groups increases violence irrespective of their type.
Robustness Checks
While the above analysis provides mixed support for our argument, we are aware that
the revolutionary/conservative variable is derived from four distinct ideologies, i.e.
religious, left-wing, right-wing and ethno-nationalist. Some of these ideologies may be
driving the outcome. For example, we coded the bulk of ethno-nationalist movements
and right-wing groups as conservative, and religious and left-wing as revolutionary. We
can reasonably expect that some of these variables may be associated with more or less
civilian targeting.
Figure 7 here
In Figure 7, we present the results of interaction between the aggregated foreign support
and each of the ideologies. The interaction effect ("Religious:Support") in Model 4
indicates significantly lower rate of violence for foreign-backed religious groups than
for other rebel groups. On the other hand, religious groups without foreign support
("Religious") barely fail to achieve a significantly different impact on civilian targeting
16

than other groups although the direction of the term is in line with expectations regarding
such violent groups as ISIS. In Model 5, we interact left-wing groups with foreign support
to find that the interaction effect ("Left-Wing:Support") has no practical significance for
our study. Nevertheless, we find that left-wing groups without support ("Left-Wing") are
likely to commit 87 percent less violence than other groups, all else being equal. This
is in accordance with our expectation that the Marxist focus on domestic mobilization
should curtail violent behavior of such groups even when they lack outside sponsors.
Interestingly, Model 6 shows a similar pattern among right-wing groups, indicating that
this ideology is less violent than others. This result generally speaks to our suggestion
that conservative groups need not be more violent than others unless they have access
to foreign support. In Model 7, we interact ethno-nationalist ideology with foreign
support. While the direction of the interaction term ("Nationalist") is expected, we find
no discernible effect on violence given that the confidence intervals overlap the line of
no effect. We similarly find that the interaction effect ("Support:Nationalist") has no
practical significance for our study, indicating that externally-backed ethno-nationalist
groups are no more likely to brutalize civilians than other rebel groups.
We additionally tested for sponsor state ideologies and their matching to rebel
ideologies only to find no significant results. Neither did we find a significant effect on
violence against civilians when rebel groups were receiving assistance from different
sponsor states. The way we interpret these negative findings is that an overall effect of
outside assistance is not contingent on the sponsor’s ideology, but on the material aspect
of the sponsor’s aid. This seems to be in line with our argument about goals of rebels
and means of sponsors as the driving forces in explaining violence.
The finding that religious rebel groups substantially reduce violence against civilians
when they receive external support, and that foreign-backed left-wing groups do not,
might be related to the international system. Earlier analyses showed that the end of
the Cold War had a transformative impact on the way civil wars are fought (Kalyvas and
Balcells 2010). We argue that the structural characteristics of the Cold War made many
17

Marxist groups rely heavily on foreign sponsors. With the dissolution of the Soviet Union,
not only had the main source of foreign support plummeted, but also the ideology lost
its domestic appeal. Under these conditions, the social revolutionary ambitions of these
groups experienced a setback and even though the ideology might not have formally
changed, the realistic objectives of such groups changed from transforming societies to
surviving. On the other hand, it might be the case that the end of the Cold War had an
opposite effect on sponsoring religious groups. In the aftermath of the Cold War, the
only ’real’ revolutionaries have seemingly been the religious rebels. As Figure 2 indicates,
Islamist rebel groups have been on the rise since 1989. This might explain why foreign
support for religious groups had an effect we theoretically predicted for revolutionary
groups.
Conclusion
How foreign support affects rebel propensity for civilian targeting? In this paper, we sug-
gest that foreign support is likely to have varying impact on rebel violence depending on
their goals. To better understand why some externally-backed groups commit violence
against civilians, while other do not, we first examine whether rebels aim to radically
transform the society, or maintain the existing balance of state-society relations. We
argue that the former, revolutionary groups, should invest more resources in building
their relations with those parts of the society they claim to represent, or whose political
transformation they aim to induce. In contrast, conservative groups, with an intent
of preserving the pre-existing order, should be less interested in investing resources
in transforming the society. Supplying foreign support to revolutionary groups should
encourage them to focus on political objectives and develop further cooperation with
civilians. Regarding conservative groups, military support is likely to decrease their
reliance on popular support for political objectives. We, therefore, expect conservative
groups to be more violent against civilians, than their revolutionary counterparts, es-
pecially when they receive military support. In contrast, we argue that non-military
18

support is likely to constrain conservatives, and make revolutionaries more violent.
The argument we advance in this paper engenders mixed explanatory power. We
find that foreign support for the revolutionaries is associated with less civilian targeting,
while support for the conservatives is linked to more violence. On the other hand,
we could not corroborate our hypotheses about the interaction between ideology and
types of support. When interacted with revolutionary/conservative groups, military and
non-military support have the identical effect on violence as the interaction between the
aggregated support and revolutionary/conservative groups.
However the general findings could have both policy and theoretical implications.
We suggest that revolutionary groups such as ISIS might have been as brutal as they
were because of the absence of overt backing by foreign sponsors. Similarly, this paper
suggests that sponsoring conservative groups may lead to serious human rights atrocities;
sponsoring states searching for militant groups to destabilize a regime or tip the scales
in favor of one side of a civil war would especially do well showing caution with military
assistance. How rebel goals and means interact in the context of armed conflict is a topic
that must be addressed if we are to further understand the conditions for prevention of
human right violations.
On a theoretical level, Straus (2015) makes a similar claim by synthesizing strategic
and ideological arguments in explaining genocide. In his account, preexisting ideological
frameworks, or ’founding narratives’ in which leaders define primary political community
and the main project of the state are key to understanding genocides. But to be
executed, genocide requires effective domination over a territory, and a material capacity
to maintain ’coalitions of violence’ or access to population. He finds that restrains
matters, too. They are seen as ideational, such as counter-narratives that emphasize
accommodation and compromise, or as the capacity to inflict violence, where external
actors can impose costs on perpetrators. Even though genocide is different from rebel
violence against noncombatants in several ways, our findings resonate with Straus’ and
supports the theoretical wedding of ideational and material factors in explaining political
19

violence phenomena.
20

Figures
Figure 1: Distribution of Violence
Number of Civilian Deaths (log)
0
20
40
60
80
100
25 100 1000 10000
Observations
Figure 2: Distribution of Types of Rebel Ideology
Religious Left−Wing Right−Wing Ethnonationalist
Ideology
Observations
0
100
200
300
400
500
600
700
21

Figure 3: Distribution of Types of Military Support
None Non−Military Military
Foreign Support
0
200
400
600
800
Observations
Figure 4: Foreign Support and Ideology
●
●
●
●
●
●
●
●
●
●
●
Ideology:Support
Multiparty
Ideology
Population
Duration
Government Violence
Democracy
Incompatibility
Support
Intensity
Previous Violence
0 5 10 15 20
Incidence Rate Ratios
Model 1
Note: The dot shows the mean point estimate of Incidence Rate Ratio, the bold line
displays a 90 percent confidence interval and the thin line indicates a 95 percent
confidence interval.
22

Figure 5: Interaction Effects of Ideology and Support on Rebel Violence
Ideology
Predicted Civilian Victims
20
40
60
80
100
120
140
Conservative Revolutionary
●
●
: Support Absent
Conservative Revolutionary
●
●
: Support Present
Note: Interaction effects. Predicted rebel violence for groups with access to foreign
support relative to those without it over group ideology. The 95 percent confidence
intervals are denoted by bars. Control variables held at mean values.
23

Figure 6: Disaggregated Support and Ideology
●
●
●
●
●
●
●
●
●
●
●
Ideology:Military Support
Multiparty
Ideology
Population
Duration
Government Violence
Democracy
Incompatibility
Military Support
Intensity
Previous Violence
0 5 10 15 20
Incidence Rate Ratios
Model 2
●
●
●
●
●
●
●
●
●
●
●
Ideology: ~Military Support
Multiparty
Ideology
Population
Government Violence
Duration
Democracy
Incompatibility
~ Military Support
Intensity
Previous Violence
0 5 10 15 20
Incidence Rate Ratios
Model 3
Note: The dot shows the mean point estimate of Incidence Rate Ratio, the bold line displays a 90 percent confidence interval and the thin
line indicates a 95 percent confidence interval.
24

Figure 7: Robustness Checks
●
●
●
●
●
●
●
●
●
●
●
Religious:Support
Multiparty
Population
Government Violence
Duration
Democracy
Religious
Incompatibility
Support
Intensity
Previous Violence
0 5 10 15
Incidence Rate Ratios
Model 4
●
●
●
●
●
●
●
●
●
●
●
Left−Wing
Multiparty
Population
Duration
Government Violence
Left−Wing:Support
Support
Democracy
Incompatibility
Intensity
Previous Violence
0 5 10 15 20
Incidence Rate Ratios
Model 5
●
●
●
●
●
●
●
●
●
●
●
Right−Wing
Multiparty
Right−Wing:Support
Population
Government Violence
Duration
Democracy
Support
Incompatibility
Intensity
Previous Violence
0 5 10 15 20
Incidence Rate Ratios
Model 6
●
●
●
●
●
●
●
●
●
●
●
Multiparty
Nationalist:Support
Population
Duration
Government Violence
Nationalist
Democracy
Support
Incompatibility
Intensity
Previous Violence
0 5 10 15 20
Incidence Rate Ratios
Model 7
Note: The dot shows the mean point estimate of Incidence Rate Ratio, the bold line displays a 90 percent confidence interval and the thin
line indicates a 95 percent confidence interval.
25

Appendix
This section displays the regression results from corresponding figures in the
main text. Table 1 shows the interaction between ideology and foreign support
along with controls (Model 1). In Table 2, we present the interaction between
disaggregated support and ideology (Model 2 and Model 3). Finally, Table
3 displays the regression coefficients from the robustness checks where we
interact each of the four ideologies with foreign support (Models 4-7).
Table 1: Foreign Support and Ideology
Model 1
Support 1.47∗∗
(0.35)
Ideology −0.14
(0.31)
Ideology:Support −1.16∗
(0.52)
Government Violence 0.09∗
(0.04)
Democracy 0.83∗∗
(0.30)
Population −0.12
(0.08)
Incompatibility 1.32∗∗
(0.28)
Intensity 2.03∗∗
(0.31)
Multiparty −0.39
(0.25)
Duration 0.07
(0.13)
Previous Violence 2.54∗∗
(0.52)
AIC 5637.79
BIC 5702.90
Log Likelihood -2805.90
Deviance 653.92
Num. obs. 1106
Presented are coefficients with standard errors in parentheses. P-value: ∗∗ p < 0.05,∗p < 0.1
26

Table 2: Disaggregated Support and Ideology
Model 2 Model 3
Ideology −0.31 −0.18
(0.28) (0.30)
Military Support 1.82∗∗∗
(0.39)
Ideology:Military Support −1.39∗
(0.57)
Non-Military Support 1.55∗∗∗
(0.36)
Ideology:Non-Military Support −1.16∗∗
(0.53)
Government Violence 0.07 0.10∗∗
(0.04) (0.04)
Democracy 1.09∗∗ 0.88∗∗
(0.30) (0.30)
Population −0.13 −0.11
(0.08) (0.08)
Incompatibility 1.49∗∗ 1.28∗∗
(0.27) (0.28)
Intensity 1.87∗∗ 2.01∗∗
(0.31) (0.31)
Multiparty −0.52∗∗ −0.45∗
(0.25) (0.25)
Duration −0.02 0.11
(0.13) (0.13)
Previous Violence 2.59∗∗ 2.55∗∗
(0.27) (0.27)
AIC 5635 5636.92
BIC 5700.11 5702.03
Log Likelihood -2804.50 -2805.46
Deviance 654.10 653.98
Num. obs. 1106 1106
Presented are coefficients with standard errors in parentheses. P-value: ∗∗ p < 0.05,∗p < 0.1
27

Table 3: Robustness Checks
Model 4 Model 5 Model 6 Model 7
Support 1.47∗∗ 0.95∗∗ 0.98∗∗ 1.06∗∗
(0.31) (0.28) (0.26) (0.40)
Religious 0.75∗∗
(0.33)
Religious:Support −1.49∗
(0.58)
Left-Wing −2.01∗∗
(0.38)
Left-Wing:Support 0.72
(0.65)
Right-Wing −1.78∗
(1.03)
Right-Wing:Support −0.29
(1.28)
Nationalist 0.78∗∗
(0.36)
Nationalist:Support −0.17
(0.52)
Government Violence 0.13∗∗ 0.13∗∗ 0.14∗∗ 0.14∗∗
(0.05) (0.04) (0.05) (0.05)
Democracy 0.59∗∗ 1.24∗∗ 0.67∗∗ 0.81∗∗
(0.30) (0.31) (0.30) (0.30)
Population −0.08 −0.08 −0.11 −0.14∗
(0.08) (0.08) (0.08) (0.08)
Incompatibility 1.22∗∗ 1.71∗∗ 1.14∗∗ 1.57∗∗
(0.26) (0.28) (0.27) (0.34)
Intensity 2.09∗∗ 2.15∗∗ 2.23∗∗ 2.09∗∗
(0.32) (0.31) (0.31) (0.32)
Multiparty −0.41∗−0.45∗−0.51∗∗ −0.36
(0.25) (0.25) (0.26) (0.25)
Duration 0.18 0.12 0.14 0.10
(0.13) (0.13) (0.13) (0.13)
Previous Violence 2.42∗∗ 2.61∗∗ 2.55∗∗ 2.55∗∗
(0.27) (0.27) (0.27) (0.27)
AIC 5635.37 5621.78 5638.50 5639.852
BIC 5700.48 5686.89 5703.61 5704.96
Log Likelihood -2804.69 -2797.89 -2806.25 -2806.93
Deviance 654.13 654.43 653.85 653.68
Num. obs. 1106 1106 1106 1106
Presented are coefficients with standard errors in parentheses. P-value: ∗∗ p < 0.05,∗p < 0.1
28

Acknowledgements
A previous version of this paper was presented at the 2016 International Studies Associa-
tion (ISA) Annual Meeting in Atlanta. We are thankful for comments from the discussant
and participants on this occasion.
Funding
Parts of this research have benefited from financial support from The United States
Department of Defense Minerva Research Initiative.
29

References
Asal, V., A. Pate, and J. Wilkenfeld (2008). Minorities at Risk Organizational Behavior
Data and Codebook. Version 9, 2008.
Balcells, L. and S. N. Kalyvas (2015). Revolutionary Rebels and the Marxist Paradox.
Beck, T., G. Clarke, A. Groff, P. Keefer, and P. Walsh (2001). New Tools in Comparative
Political Economy: The Database of Political Institutions. The World Bank Economic
Review 15(1), 165–176.
Bloom, M. (2005). Dying to Kill: The Allure of Suicide Terror. New York: Columbia
University Press.
Bloom, M. M. (2004). Palestinian Suicide Bombing: Public Support, Market Share, and
Outbidding. Political Science Quarterly 119(1), 61–88.
Collier, P. (2000). Doing Well Out of War: An Economic Perspective in Greed and
Grievance: Economic Agendas in Civil Wars. Berdal, M. and Malone, M. International
Peace Academy.
Downes, A. B. (2006). Desperate Times, Desperate Measures: The Causes of Civilian
Victimization in War. International Security 30(4), 152–195.
Eck, K. and L. Hultman (2007). One-Sided Violence Against Civilians in War Insights
from New Fatality Data. Journal of Peace Research 44(2), 233–246.
Fearon, J. D. and D. D. Laitin (2003). Ethnicity, insurgency, and civil war. American
Political Science Review 97(01), 75–90.
Gleditsch, N. P. and R. Nordås (2014). Conflicting Messages? the IPCC on Conflict and
Human Security. Political Geography 43, 82–90.
Goodwin, J. (2006). What Do We Really Know About (Suicide) Terrorism? In Sociological
Forum, Volume 21, pp. 315–330. Springer.
Harbom, L., E. Melander, and P. Wallensteen (2008). Dyadic Dimensions of Armed
Conflict, 1946–2007. Journal of Peace Research 45(5), 697–710.
Hilbe, J. M. (2011). Negative Binomial Regression. Cambridge: Cambridge University
Press.
Högbladh, S., T. Pettersson, and L. Themnér (2011). External Support in Armed Conflict
1975–2009. Presenting New Data. In 52nd Annual International Studies Association
Convention, Montreal, Canada, March, pp. 16–19.
Hultman, L. (2007). Battle Losses and Rebel Violence: Raising the Costs for Fighting.
Terrorism and Political Violence 19(2), 205–222.
Humphreys, M. and J. M. Weinstein (2008). Who Fights? the Determinants of Participa-
tion in Civil War. American Journal of Political Science 52(2), 436–455.
30

Kalyvas, S. (2015). Is ISIS a Revolutionary Group and if Yes, What Are the Implications?
Perspectives on Terrorism 9(4).
Kalyvas, S. N. (2006). The Logic of Violence in Civil War. Cambridge: Cambridge Univ
Press.
Kalyvas, S. N. (2007). Book Review of Inside Rebellion: The Politics of Insurgent
Violence, by Jeremy M. Weinstein. Comparative Political Studies.
Kalyvas, S. N. and L. Balcells (2010). International System and Technologies of Rebellion:
How the End of the Cold War Shaped Internal Conflict. American Political Science
Review 104(03), 415–429.
Long, S. (1997). Regression Models for Categorical and Limited Dependent Variables.
Advanced Quantitative Techniques in the Social Sciences 7.
Malaquais, D. (2001). Arts de Feyre au Cameroun. Politique Africaine (2), 101–118.
Mampilly, Z. C. (2011). Rebel Rulers: Insurgent Governance and Civilian Life During War.
Ithaca, NY: Cornell University Press.
Marshall, M. G. and K. Jaggers (2007). Polity IV Project: Political Regime Characteristics
and Transitions, 1800–2006.
Metelits, C. (2009). Inside Insurgency: Violence, Civilians, and Revolutionary Group
Behavior. New York: New York University Press.
Pape, R. (2005). Dying to Win: The Strategic Logic of Suicide Terrorism. New York:
Random House.
Salehyan, I., D. Siroky, and R. M. Wood (2014). External Rebel Sponsorship and Civilian
Abuse: A Principal-Agent Analysis of Wartime Atrocities. International Organization,
1–29.
Sanín, F. G. and E. J. Wood (2014). Ideology in Civil War Instrumental Adoption and
Beyond. Journal of Peace Research 51(2), 213–226.
Singer, J. D., S. Bremer, and J. Stuckey (1972). Capability Distribution, Uncertainty, and
Major Power War, 1820-1965. Peace, War, and Numbers 19.
Staniland, P. (2014). Networks of Rebellion: Explaining Insurgent Cohesion and Collapse.
Ithaca, NY: Cornell University Press.
Straus, S. (2015). Making and Unmaking Nations: War, Leadership, and Genocide in
Modern Africa. Ithaca, NY: Cornell University Press.
Thaler, K. M. (2012). Ideology and Violence in Civil Wars: Theory and Evidence from
Mozambique and Angola. Civil Wars 14(4), 546–567.
Toft, M. D. and Y. M. Zhukov (2015). Islamists and Nationalists: Rebel Motivation
and Counterinsurgency in Russia’s North Caucasus. American Political Science Re-
view 109(02), 222–238.
31

Ugarriza, J. E. and M. Weintraub (2015). The Effects of Ideology on Insurgent Targeting
Decisions: Explaining Colombian Guerrillas’ Tactical Choices. In International Studies
Association Annual Meeting, New Orleans, LA.
Walker, M., A. Hall, R. M. Anderson, M.-G. Basáñez, et al. (2009). Density-Dependent
Effects on the Weight of Female Ascaris Lumbricoides Infections of Humans and Its
Impact on Patterns of Egg Production. Parasit Vectors 2(11), 1–18.
Wasserstein, R. L. and N. A. Lazar (2016). The ASA’s Statement on P-Values: Context,
Process, and Purpose. The American Statistician.
Weinstein, J. M. (2006). Inside Rebellion: The Politics of Insurgent Violence. Cambridge
University Press.
Wood, R. M. (2010). Rebel capability and strategic violence against civilians. Journal of
Peace Research 47(5), 601–614.
Wood, R. M. (2014). From Loss to Looting? Battlefield Costs and Rebel Incentives for
Violence. International Organization 68(04), 979–999.
Wood, R. M., J. D. Kathman, and S. E. Gent (2012). Armed Intervention and Civilian
Victimization in Intrastate Conflicts. Journal of Peace Research 49(5), 647–660.
32