Soc 3

Data Set Feature

Getting Religion Right in Civil Wars

Monica Duffy Toft1

Abstract Surveying civil war in the world today is striking in terms of how often religious cleavages and grievances have become central to armed conflict. How are the causes and outcomes of religious civil wars different than other civil wars, if at all? Is Islam implicated for the contemporary surge in religious civil war? The first section reviews the literature and addresses the importance of religion for civil war. I then introduce a dataset and describe key trends in religious civil war in the third section, while in the fourth section I present tests of whether Muslim or Arab Muslim societies in particular are more prone to religious strife. The paper concludes with a discussion of the implications of the main findings.

Keywords civil wars, religion, ethnicity, dataset

Surveying contemporary civil wars, it is striking how often religious cleavages and

grievances become central to armed conflict. The new religious civil war (RCW)

dataset employed here documents 28 civil wars ongoing as of the end of 2014,1 with

twenty conflicts, making up 71 percent of the sample, featuring religion as a central

or peripheral factor. By contrast, only eight ongoing civil wars in 2014 had no

religious dimension, of which only five were classic ethnic wars involving identity

politics but not involving a significant religious dimension As a result, religious civil

wars today pose some of the biggest strategic challenges to the global order, from the

1Tufts University, Fletcher School of Law and Diplomacy, Medford, MA, USA

Corresponding Author:

Monica Duffy Toft, Tufts University, Fletcher School of Law and Diplomacy, 160 Packard Avenue,

Medford, MA 02155, USA; Global Scholar, Peace Research Institute Oslo, Norway.

Email: [email protected]

Journal of Conflict Resolution 2021, Vol. 65(9) 1607-1634

ª The Author(s) 2021 Article reuse guidelines:

sagepub.com/journals-permissions DOI: 10.1177/0022002721997895

journals.sagepub.com/home/jcr

intractable conflicts over Kashmir and Palestine, to the Sunni insurgencies in Afgha-

nistan and Iraq, to the sectarian Syrian bloodbath that has claimed an estimated 300

to 400,000 lives since 2011.

How are the causes and outcomes of religious civil wars different than other civil

wars, if at all? Is Islam implicated for the contemporary surge in religious civil war?

Does the rise of religious civil war indicate a “clash of civilizations?” The next five

sections take up these questions. The first section contains a review of the literature,

noting that although there have been some advances made, there is much work to be

done. I then introduce the RCW dataset and describe several key trends in religious

civil war in the third section, while in the fourth I present tests of whether Muslim

societies in general or Arab Muslim societies in particular are more prone to reli-

gious civil war. The final section summarizes the major findings and relates the

implications of those findings.

The Literature

The literature on religion and civil war is not much more developed than the liter-

ature on ethnic civil war when van Evera (1994) codified hypotheses on nationalism

and war or Sambanis (2001) asked whether ethnic and non-ethnic civil wars have the

same causes. It has also been much less studied than modern religious terrorism.2

Just over a decade ago Toft (2007) made an early contribution to the study of religion

and civil war by theorizing how bidding dynamics among religious elites contribute

to the rise of religious civil war. That article also proposed an explanation for why

Islam in particular appeared to account for so many contemporary religious civil

wars. Since then, both the phenomenon of religious civil wars and the literature on

religion and civil war has grown quickly. However, heated debate on the most basic

questions persists. This article investigates several major claims often made about

religion and civil war. This empirical “brush-clearing” exercise seeks to move the

literature forward both theoretically and empirically, and to point to a more pro-

gressive research paradigm in this area.

Literature on the topic of religion and conflict can be methodologically broken

down into two broad categories. The relevance of religion can be measured as an

issue or as an identity (Pearce 2005; Isaacs 2016). Toft (2007) defines a religious

civil war “as a war in which religious belief or practice is either a central or per-

ipheral issue in the conflict,” therefore falling in the issue-based methodological

group (p. 97).

Further studies have analyzed the relationship between religion and violence

looking at both issue and identity definitions of religion. Generally, identity was

not found to have a significant effect on the onset or characteristics of civil war,

while issue-based definitions of religious conflict are found to have a significant

relationship. Isak Svensson (2007) found that civil conflict where at least one side

made overt religious claims—an issue-based definition—were less likely to be

resolved by negotiated settlements. However, when looking at identity-based

1608 Journal of Conflict Resolution 65(9)

definitions, neither religious nor ethnic demographic dissimilarities were shown to

impact the likelihood of termination via negotiated settlement. Similarly, Maoz and

Henderson (2020) found that civil wars coded as religious are more severe, lengthy,

and less likely to be terminated through negotiated settlements compared to non-

religious civil conflicts (similar to findings presented in Toft [2007]). They further

concluded that the demographic dominance of a particular religious group in a state

does not impact that states propensity for intrastate violence (Maoz and Henderson

2020).3

Some authors have criticized the measurement validity inherent in issue-based

classification of religious conflict because religious classification could be limited to

conflicts in which major actors express religious demands while conflicts with less

prominent religious influence are left out. Isaacs specifically points to the Serbian–

Bosnian war as an example of a conflict that lacks religious demands and therefore

could be coded as non-religious despite the influence of religious mythology (2016).

It should be noted that this study aims to remedy this concern by classifying civil

conflicts as featuring religion as either a central or peripheral factor. In fact, the

RCW dataset classifies the Serbian–Bosnian conflict as a religious civil war where

religion is a peripheral issue.

The grievance-based theory first put forth by Gurr (1970) asserts that discontent

due to relative deprivation is a condition for political violence and that the level of

violence will increase as the level of relative deprivation increases. This theory has

been tested quantitatively using the Minorities at Risk (MAR) dataset which con-

tains information about politically active ethnic minorities. Fox built on Gurr’s

theory as well as mobilization theory to study the role of religion in ethnoreligious

violence in his 2002 book. The study found that if a minority group is seeking

autonomy or independence, then religious grievances increase the likelihood of

rebellion significantly (Fox 2002). Further studies found that the level of religious

discrimination against an ethnoreligious group is a predictor of civil war (Akbaba

and Taydas 2011). Due to a perception selection bias of the MAR data, other

researchers have studied the grievance theory using ethno-linguistic fractionaliza-

tion and polarization indices (ELF/ELP).

Many studies using a demographic or identity-based definition of religion rely on

ELF/ELP. Some of these studies failed to find a robust effect of ethnic heterogeneity

on civil war onset and further found that neither the percentage of Muslims or

Middle East affected civil war onset (e.g. Fearon and Laitin 2003). But subsequent

studies indicated that ethnic grievances—rooted in horizontal ethnic inequality and

specific patterns of political and economic privilege and exclusion—help explain

the outbreak of ethnic civil wars (e.g. Cederman, Wimmer, and Min 2010;

Cederman, Gleditsch, and Buhaug 2013).4 When investigating the relationship

between ethnic and religious identities, Basedau et al. (2015) found that the pres-

ence of overlapping religious and ethnic identities increase the likelihood of the

onset of civil violence.

Toft 1609

The view that ethnic groups are the most likely to be aggrieved and the easiest

to mobilize and cohere as rebels has become a dominant view in the field (e.g.

Blattman and Miguel 2010). The importance of ethnicity for many civil wars begs

the question of what it is about ethnicity and ethnic groups that is important for the

frequency, severity, and duration of identity-based civil wars. Are some dimensions

of ethnicity more important than others?

It is precisely this question that Bormann, Cederman, and Vogt (2017) recently

took up by compiling data on the linguistic and religious composition of ethnic

groups. Bormann and colleagues argue that the presence of linguistic cleavages

between potential ethnic challengers and the dominant ethnic group in a state are

more likely to lead to the outbreak of armed ethnic conflict than the presence of

religious cleavages between these ethnic groups. However, there are several limita-

tions in their data for testing the importance of religious cleavages on civil war onset.

First, they “do not identify the political meaning of religion and language” (p. 13).

Second, their study excludes all intra-ethnic religious civil wars—which comprise a

majority of civil wars in which religion was central—involving religious conflicts

within linguistically or culturally defined ethnic groups or inter-ethnic civil wars

between ethnic groups that share the same religion, but have competing ideas about

religion in public life.5 Finally, the Bormann et al. findings are biased against even

inter-ethnic religious cleavages because they rely on a permissive definition of “civil

war,” including all cases of armed conflict with at least 25 battle deaths. In fact, their

main result—that linguistic cleavages dominate religious cleavages as a determinant

of ethnic civil war onset—is much weaker when using a more traditional definition

of civil war with at least 1,000 deaths.6

Additional analyses aim to uncover potential causal mechanisms in the relation-

ship between religion and violent conflict. Isaacs found that organizations that have

previously used violence are more likely to use religious rhetoric, and that such

rhetoric may help organizations mobilize resources and discourage defection (2016).

Conversely, organizations that have recently used religious rhetoric are no more

likely to use violence than organizations who have not (Isaacs 2016). An additional

study found that cultural affinity, including religious identity, between a group and

an external actor leads the group to anticipate external support. This anticipation

increases the likelihood the organization will rebel and that the rebellion will resort

to violent methods (Jackson, San-Akca, and Maoz 2020). Looking at state determi-

nation conflict more specifically, Breslawski and Ives (2019) found that religious

factions were more likely to use violence than non-religious ones, theorizing that

violence is used to visibly outbid for the support of transnational religious supporters

(also see, Toft 2007).

Scholarship on the association between Islam and violence has also been mixed.

In his influential The Clash of Civilizations, Huntington put forth that Islam has

bloody borders and bloody innards (1996). Following this claim, scholars have

conducted quantitative assessments to uncover the relationship between Islam and

civil conflict. Toft found that Muslim combatants were present in 81 percent of

1610 Journal of Conflict Resolution 65(9)

religious civil wars and hypothesized this was due to a lack of historical internecine

religious wars, the geographic proximity of Muslim majority countries to petroleum

reserves, and jihad (2007). When assessing all intrastate conflict, Karakaya (2015)

found that while Muslim-plurality countries are disproportionately involved, after

controlling for other factors such as GDP, repression, oil dependency, and youth

populations, Islam was no longer a statistically significant predictor of the onset of

civil conflict. Gleditsch and Rudolfsen (2016) found that nearly all civil wars in the

past years have taken place in Muslim majority countries and that the majority of

those involve Islamist insurgencies. Yet, Fish, Jensenius, and Michel (2010) did not

find any correlation between Muslim population and episodes of large-scale political

violence.

In addition to the association between Islam and violence, it is also worth inves-

tigating the causes of civil conflict specifically in Arab countries. In a report on civil

conflict in the Middle East and North Africa, Gause (2014) encourages readers to

look beyond the unnuanced narrative of a Sunni versus Shia sectarian fight and

instead understand that conflicts arise from the weakening of Arab states and

intra-Sunni disputes. Additionally, Hägerdal (2021) notes that non-separatist ethnic

wars have clustered in the Arab world due to identity cleavages, often religious ones.

Using the Ethnic Power Relations (EPR) dataset, Vogt et al. finds that ethnic dis-

crimination in the MENA region is significantly higher than other regions since the

1980s and that excluded ethnic groups are more likely to start ethnic rebellions.

In short, quantitative studies on both issue and identity-based definitions of

religion and civil wars have provided mixed results. This article aims to further

develop the literature forward both theoretically and empirically.

The Religious Civil War Data Set

Designed to help arbitrate the core ongoing debates on religion and civil war, this

article employs the new civil war (RCW) dataset.7 An earlier version of a dataset

employed by Toft included 133 civil wars over the 1940 to 2000 period, of which 42

(32 percent) were religious civil wars. Based on that data set, this newer dataset

covers a longer period from 1945 to 2014, the last year for which we have data on

battle deaths and definitive conflict outcomes and more significant analyses of the

dynamics of the cases.8

Over this 1945 to 2014 period, there were 208 civil wars, which were defined as

armed conflicts between an independent state’s9 central government (or a loyal

military faction10) and one or more organized non-state combatants11 (or dissident

military factions) over either territory or who governs the state, in which there were

(a) at least 1,000 war-related total deaths, (b) at least an average of 100 war-related

total deaths per year,12 and (c) in which at least 5 percent of the total war-related

deaths were suffered by each side.13

An identity-based civil war is a civil war in which there are ethnic, linguistic, or

religious cleavages in the conflict. A religious civil war is a subset of such civil wars

Toft 1611

in which religious belief or practice is an issue in the conflict. A non-religious civil

war may be one of two types: (a) a non-identity-based civil war or (b) non-religious

identity-based civil war. Religious centrality in civil war is the degree to which

religious belief or practice is a cleavage in the conflict. Religion is central to civil

war if the rebels fight over whether the state or a rebel-dominated region will be

ruled according to a specific religious tradition. For example, civil wars where rebel

groups have a main stated goal to institute Sharia law (even if the government was

led by Muslims) are included in this category. Other possibilities include the man-

aging of religious institutions, privileging certain religious institutions with

resources and legal monopolies, controlling dress and education, apostasy and mar-

riage laws, and laws restricting minorities. An example of a non-Muslim group that

took up arms against a government in order to govern a region according to a

religious tradition is the Holy Spirit Movement insurgency in Uganda that started

in 1986.

Religion is peripheral to a civil war if the rebels identify with a specific religious

tradition and group themselves accordingly, but the rule of this specific religious

tradition is not a major reason why they have rebelled. Peripheral wars thus often

pertain to those wars in which the conflict parties adopt religious rhetoric to mobilize

resources and recruitment and retention of members (Isaacs 2016). Because religion

is not a central issue in these wars, peripherally coded civil wars can often be

understood as an identity-based classification of religion.

Religion is irrelevant for all other civil wars. In categorizing religious civil war

this way, religion is not epiphenomenal to civil war when religion is central as

opposed to peripheral to civil war.14 Of the 208 wars, 90 (43 percent) were religious

civil wars.15 Of these 90 religious civil wars, religion was a peripheral issue in 49

wars and was a central issue in 41 wars.

The updated RCW dataset improves on the original version created by the author

in “Getting Religion?” (2007) in three central ways. First, a more comprehensive

definition of civil war leads to the inclusion of more conflict cases. The 2007 version

of the dataset was limited to conflicts with more than 1,000 average annual battle

fatalities, however, the current RCW includes conflicts with an average of 100

deaths per year as long as 1,000 fatalities are recorded over the duration of the

conflict. Second, the dataset is extended temporally to include civil conflicts through

the year 2014. Finally, while the original dataset only provided summaries of civil

conflicts coded as religious, the updated RCW provides case narratives, multiple

references, and coding justifications for every conflict in the dataset.

The RCW also provides unique value in comparison to existing datasets on armed

conflict. Integral to the RCW is an issue based coding of religion in each conflict.

Two of the most prominent civil conflict datasets, the UCDP/PRIO armed conflict

dataset and Correlates of War intra-state war dataset, lack variables coding either an

identity or issue based classification of religion in civil conflicts.16 The COW dataset

is accompanied by a book titled A Guide to Intra-state Wars which provides cases

and coding justification, but the armed conflict dataset provides no such appendices

1612 Journal of Conflict Resolution 65(9)

(Dixon and Sarkees 2015). The RCW provides a 600-page appendix containing

original case summaries, coding justifications, potential definitional issues, and

comparisons with other key datasets for every included conflict. Additionally, the

RCW provides summaries and explicit justification for civil wars that appear in

other datasets but are excluded from RCW based on its own definition of civil

conflict.

The RCW dataset also improves upon religious-specific civil conflict databases.

Four of the prominent religious conflict classification datasets include the Religion

and Armed Conflict dataset (RELAC) created by Svensson and Nilsson (2018), the

Religion and Conflict in Developing Countries dataset (RCDC) created by Vüllers,

Pfeiffer, and Basedau (2015), the supplemented Political Instability Task Force

dataset (PITF) created by Fox (2012), and the Running on Faith dataset created

by Lindberg (2008). The RCW dataset is uniquely valuable compared to these

datasets in three primary ways. First, the RCW provides the broadest geographic

and temporal coverage of religious civil conflicts.17 Second, the RCW bases the

inclusion of cases and coding of variables on a broad survey of historiography of

each civil war and the literature. By contrast, Fox (2004, 2014) relies solely on

conflict descriptions by the Political Instability Taskforce, the RCDC dataset relies

solely on country reports of the Economist Intelligence Unit and US government

(Vüllers, Pfeiffer, and Basedau 2015), the RELAC dataset relies heavily on the

UCDP Encyclopedia (Svensson and Nilsson 2018), while only the Running on Faith

dataset utilizes a wider variety of sources (Lindberg 2008). Finally, the RCW

includes extensive documentation in an appendix providing both original case sum-

maries and detailed coding justifications. Saliently, none of the preceding datasets

offer either.

The RCW further differentiates itself from certain existing datasets in other

significant ways. Although scholars may dispute RCW’s classification of certain

civil wars, its coding is more reliable and substantively meaningful than a pure

demographic coding of religious cleavage, such as that of Lindberg (2008). If no

evidence that a religious cleavage was used to recruit or mobilize combatants, that

civil war was coded as non-religious even if there were differences in the religions of

the combatants. For example, even though Kurds are primarily Sunni and Iran’s

government is dominated by Shia, the 1946 Iranian Kurdish rebellion is coded as

non-religious because it was sponsored by the Soviet Union and the separatist

Kurdish Mahabad Republic was led by secular elites who did not engage in religious

bidding (Toft 2007). Additionally, the coding of religious centrality does not arbi-

trarily partition religious civil wars between inter-religious and intra-religious con-

flicts. For example, religion is coded as central to the Caucasus Emirates insurgency

in Russia since 2007, because jihadists sought imposition of Sharia law. By contrast,

Fox (2014) codes this case as a “religious identity conflict” and not a “religious war”

since it pits Muslims against Christians.

Toft 1613

The Significance of Religious Civil War

This section revisits claims that religion in civil war matters.18 The data first show

that religious civil wars are becoming more frequent over time. Worldwide, all kinds

of civil wars—non-ethnic, ethnic, and religious—display a general upward trend in

the number of ongoing wars from the 1950s through the end of the Cold War. In the

late 1980s, there was a brief spike in both ethnic and religious civil wars (both

peripheral and central) whereas non-identity based civil wars began to decline.

However, whereas all other kinds of civil war declined since the early 1990s, after

2001 the number of civil wars in which religion featured centrally has nearly tripled

(See Figure 1). Other studies of religious civil war fail to capture this twenty-first

century upsurge.19

Second, centrally religious civil wars are increasingly concentrated within the

Muslim world. Although the Middle East and North Africa (MENA) is home to only

5 percent of the world’s population (and 20 percent of the world’s Muslims), this

region accounts for about 40 percent (sixteen of forty-one) of centrally religious civil

wars since 1945 (see Table 1).

By contrast, centrally religious civil wars are entirely absent in Western Europe,

the Americas, and Oceania, with other regions in between. In addition, no more than

5 percent of non-Muslim ruled countries have ever experienced a centrally religious

civil war in any given year. By contrast, by 2014, there was a centrally religious civil

war ongoing in almost a quarter of Muslim-ruled countries (see Figure 2).

Figure 1. Religious and Nonreligious civil wars, 1945–2014.

1614 Journal of Conflict Resolution 65(9)

T a b

le 1 .

G eo

gr ap

h ic

R eg

io n s

an d

R el

ig io

u s

C en

tr al

it y

in C

iv il

W ar

si n ce

1 9 4 5 .

R el

ig io

us C en

tr al

ity

N o n re

lig io

u s

R el

ig io

n P er

ip h er

al R

el ig

io n

C en

tr al

R eg

io n

C o ld

W ar

P o st

– C

o ld

W ar

C o ld

W ar

P o st

– C

o ld

W ar

C o ld

W ar

P o st

– C

o ld

W ar

A m

er ic

as 1 6

2 0 %

0 0 %

0 0 %

0 0 %

0 0 %

0 0 %

W ._

E u ro

p e

1 1 %

0 0 %

3 1 0 %

4 2 1 %

0 0 %

0 0 %

E . E u ro

p e

5 6 %

4 1 1 %

0 0 %

4 2 1 %

0 0 %

3 1 4 %

SS A

2 1

2 6 %

2 3

6 1 %

7 2 3 %

6 3 2 %

2 1 0 %

4 1 9 %

M E N

A 1 5

1 9 %

6 1 6 %

4 1 3 %

1 5 %

7 3 5 %

9 4 3 %

So u th

A si

a 9

1 1 %

3 8 %

5 1 7 %

2 1 1 %

6 3 0 %

3 1 4 %

E as

t A

si a

1 3

1 6 %

2 5 %

1 0

3 3 %

2 1 1 %

5 2 5 %

2 1 0 %

O ce

an ia

0 0 %

2 0 %

1 3 %

0 0 %

0 0 %

0 0 %

T o ta

l O

n se

t 8 0

1 0 0 %

3 8

1 0 0 %

3 0

1 0 0 %

1 9

1 0 0 %

2 0

1 0 0 %

2 1

1 0 0 %

N ot

e: B ec

au se

o f ro

u n d in

g th

e p er

ce n ta

ge , co

lu m

n s

d o

n o t

ad d

to 1 0 0 %

.

1615

Thus, although only 25 percent (five of twenty) of centrally religious civil wars

involved both a Muslim-dominated government and predominantly Muslim rebel

groups during the Cold War, that figure has shot up to 71 percent (fifteen of twenty-

one) after the Cold War.

In total, some three-quarters of all religious civil wars involve a Muslim govern-

ment or Muslim rebel group. In the post-Cold War era, the Islamic world has had a

monopoly on religious civil war, when 95 percent (twenty of twenty-one) of cen-

trally religious civil wars involve at least one Muslim combatant (see Table 2).

In contrast, non-Muslims account for the majority of non-religious civil wars

after the Cold War, with nearly 40 percent alone involving both a Christian-led

government and rebel group (see Table 2). Civil wars in which religion is peripheral

are distinctive, with a majority of all such wars pitting Muslims against Christians.

Islamist rebel groups are implicated in all 14 centrally religious civil wars that have

broken out since 2000; non-Muslim rebel groups are now implicated in none.

Third, there is a “ratchet effect” in the centrality of religion when civil wars in

countries recur, implying that conflict actors have found increasing returns to mak-

ing religious bids over time. Of 43 civil wars that recur, there are 10 cases of either

non-religious civil wars becoming religious civil wars or religion shifting from

peripheral to central, but only one clear example of a religious civil war recurring

without religion being just as central as it was previously.20 Thus, in 2014, there is

not a single region of the world where there are more non-religious ethnic civil wars

ongoing than religious civil wars. In MENA, religious civil wars now outnumber

Figure 2. Trends and data on Muslim rulership and Civil War incidence since 1945. (a) Muslim-ruled countries. (b) Non-Muslim-ruled countries.

1616 Journal of Conflict Resolution 65(9)

T a b

le 2 .

A ra

b ve

rs u s

M u sl

im Id

en ti ty

an d

R el

ig io

u s

C en

tr al

it y

in C

iv il

W ar

si n ce

1 9 4 5 .

R el

ig io

us C en

tr al

ity

N o n re

lig io

u s

R el

ig io

n P er

ip h er

al R

el ig

io n

C en

tr al

A ny

Si de

Id en

tit y

C o ld

W ar

P o st

– C

o ld

W ar

C o ld

W ar

P o st

– C

o ld

W ar

C o ld

W ar

P o st

– C

o ld

W ar

N o

M u sl

im 5 8

7 3 %

2 2

5 8 %

9 3 0 %

5 2 6 %

5 2 5 %

1 5 %

M u sl

im 2 2

2 8 %

1 6

4 2 %

2 1

7 0 %

1 4

7 4 %

1 5

7 5 %

2 0

9 5 %

N on

A ra

b 8

1 0 %

1 2

3 2 %

1 8

6 0 %

1 3

6 8 %

1 0

5 0 %

1 0

4 8 %

A ra

b 1 4

1 8 %

4 1 1 %

3 1 0 %

1 5 %

5 2 5 %

1 0

4 8 %

T o ta

l O

n se

t 8 0

1 0 0 %

3 8

1 0 0 %

3 0

1 0 0 %

1 9

1 0 0 %

2 0

1 0 0 %

2 1

1 0 0 %

N ot

e: B ec

au se

o f ro

u n d in

g th

e p er

ce n ta

ge , co

lu m

n s

d o

n o t

ad d

to 1 0 0 %

.

1617

non-religious civil war by a ratio of four-to-one. In Sub-Saharan Africa, the ratio is

five-to-zero; in East Asia, three-to-one. Such data undermine the conclusion of

Bormann, Cederman, and Vogt (2017) that “civil wars are more likely to be fought

along linguistic cleavages than religious ones.”

Fourth, centrally religious civil wars are bloodier than other civil wars. The

average number of total civil war-related deaths is about 55,000 for when religion

is irrelevant, 25,000 when peripheral, and 160,000 when central (see Table 3).

Similarly, the mean battle deaths in civil war total is about 24,000 when religion

is irrelevant, 12,000 when peripheral, and 40,000 when central (see Table 3).21

These findings contrast with Pearce (2005), who also found that conflict intensity

was higher for religious than non-religious conflict in the UCDP dataset from 1946

to 2001, but who found that intensity was actually lower when religion had “high”

rather than “low” relevance.22 After the Cold War, centrally religious civil wars have

become less deadly in absolute terms (66,000 average total deaths) than those waged

during the Cold War, but remain more deadly than other civil wars (25,000 average

total deaths) in relative terms.

Fifth, the deadliest religious civil wars involve Muslim combatants. Civil wars

with religion as central and involving Muslims result in an average of 180,000 total

deaths (43,000 battle deaths), three times (50 percent) more than any other type of

civil war including Muslims or not (see Table 3).23 By contrast, Fish, Jensenius, and

Michel (2010) attempted to argue that Islamic states have not suffered dispropor-

tionate deaths from “Major Episodes of Political Violence” from 1946 to 2007

compared to non-Islamic states. The RCW data indicate that Fish, Jensenius, and

Michel’s (2010) finding is only half-true: civil wars involving Muslims are indeed

no more deadly than those without Muslim combatants, but only when religion is

irrelevant or peripheral. These data also echo Gleditsch and Rudolfsen (2016), who

observed that 94 percent of all civil war battle deaths since 2011 are in Muslim

states.

Sixth, centrally religious civil wars, but not other civil wars, have become sig-

nificantly more intractable after the Cold War. Religious civil wars did not differ

Table 3. Religious Centrality and Deaths in Civil War since 1945.

Religious Centrality

Religion Nonreligious Peripheral Central

No Muslim Combatant 30,845 (n ¼ 80)

11,482 (n ¼ 14)

14,000 (n ¼ 6)

Any Muslim Combatant 10,169 (n ¼ 38)

12,040 (n ¼ 35)

43,908 (n ¼ 35)

Total 24,187 (n ¼ 118)

11,881 (n ¼ 49)

39,532 (n ¼ 41)

Note: The number in parentheses is the total number of civil wars of that kind (denominator).

1618 Journal of Conflict Resolution 65(9)

much from non-religious civil wars in terms of outcome, duration, or recurrence

during the Cold War. In fact, if anything a larger share of religious central wars

initiated during the Cold War ended in a negotiated settlement than non-religious

civil wars. The post-Cold War story, however, is entirely different: whereas only

5 percent of civil wars where religion was irrelevant remain ongoing as of 2014, 71

percent of civil wars where religion was central remain ongoing. By contrast,

whereas a third to a quarter of post-Cold War civil wars where religion was irrele-

vant or peripheral have reached a negotiated settlement, only a single (5 percent) of

post-Cold War civil war where religion was central has reached a negotiated settle-

ment (see Table 4).

The data therefore partially confirm findings on the intractability of religious

conflict by Svensson (2007, 2012), but only for the post-Cold War period.

Reexamining Islamic Exceptionalism in Religious Civil War

In summary, the RCW data show that after the Cold War civil wars in which religion

is central have been on the rise, have been particularly bloody and intractable, and

over the last several decades have clustered in the Muslim world. What drives this

conflict pattern: Islam, Arab culture, or something else?24 Toft (2007) explained the

over-representation of religious civil wars in Islamic countries by arguing that

religious bids have a higher utility for Islamic elites than non-Islamic elites. Why?

Islam has yet had no equivalent to the separation of religion and state and the decline

of Christian religious-political authority as seen in the West after the Thirty Years

War and Treaty of Westphalia.25 The Islamic doctrine of Salafist jihad may make

Muslims, whether locally or as transnational jihadists, more likely to fight in the

name of Islam out of a sense of religious obligation.26 Toft (2007) further argued that

Arab states are more prone to religious conflict since 1948 since Islam’s holy sites

are located in the region, there is a concentration of oil reserves, and the Arab-Israel

conflict has festered Arab grievance.

Table 4. Religious Centrality and Total Deaths in Civil War since 1945.

Religious Centrality

Religion Nonreligious Peripheral Central

No Muslim Combatant 67,610 (n ¼ 80)

17,057 (n ¼ 14)

36,167 (n ¼ 6)

Any Muslim Combatant 29,158 (n ¼ 38)

27,651 (n ¼ 35)

179,734 (n ¼ 35)

Total 55,227 (n ¼ 118)

24,624 (n ¼ 49)

158,724 (n ¼ 41)

Note: The number in parentheses is the total number of civil wars of that kind (denominator).

Toft 1619

The relationship between religion and civil wars can be explained by a series of

grievances that are mobilized among group dynamics. If a group is feeling discri-

minated against or is repressed by the state, it may turn to protest or violence.

Despite the promotion of a secularization thesis, the political influence on religion

has dramatically increased over the past forty years (Toft, Philpott, and Shah 2011).

While this phenomenon has been observed worldwide, Toft et al. found that the post-

colonial Arab Regimes have largely seen a conflictual consensus between religious

and political authority as well as a low degree of independence between religious

and political authority (2011). This general relationship between the state and reli-

gion in the Arab world may explain the propensity for religiously-based grievances.

Islam’s correlation with autocracy27 and terrorism28 has long been debated.

Ignited by Toft (2007), a similar debate on Islam and civil war has arisen (cf. Fish,

Jensenius, and Michel 2010; Sørli, Gleditsch, and Strand 2005; Karakaya 2015).

After reviewing this literature, Gleditsch and Rudolfsen (2016) concluded that

“More sophisticated causal models are needed to assess whether religion itself plays

a role in armed conflict.” This section is a first step to fill this void through the

testing of three alternative hypotheses:

Hypothesis 1a (Islamic Religious Civil War Exceptionalism): Both Arab and

Non-Arab Islamic populations are positively associated with the onset of civil

wars where religion is central and where grievances exist, ceteris paribus.

Hypothesis 1b (Arab Religious Civil War Exceptionalism): Arab Muslim

populations, but not non-Arab Muslim ones, are positively associated with

the onset of civil wars where religion is central and where grievances exist,

ceteris paribus.

Hypothesis 1c (Islamic Religious Civil War Myth): Neither Arab nor non-

Arab Muslim populations are positively associated with the onset of civil wars

where religion is central and where grievances exist, ceteris paribus.

Populations are coded as Arab Muslim if >50 percent of the population is Muslim

and >50 percent of the population is ethnically Arab. The RCW dataset indicates that

out of 118 non-religious civil wars, non-Arab Muslims are involved in twenty

(17 percent), Arab Muslims in eighteen (15 percent), and non-Muslims in eighty

(68 percent). By contrast, out of forty-nine peripheral religious wars, thirty-one

(63 percent) involve non-Arab Muslims, whereas only four (8 percent) involve Arab

Muslims and fourteen (29 percent) involve non-Muslims. Finally, out of forty-one

central religious wars, only six (15 percent) involve non-Muslims, whereas the

remaining are split more evenly with twenty (49 percent) involving non-Arab

Muslims and fifteen (37 percent) involving Arab Muslims (see Table 2).

Of course, descriptive statistics fail to provide baseline conflict propensities for

Muslims, Arab Muslims, or other groups. For this, we can rely on our global version

of the EPR-ED data.29 Out of 853 different ethnic groups in the world since 1945,

some 247 are predominantly Muslim. Of those 247 Muslim ethnic groups, about 80

1620 Journal of Conflict Resolution 65(9)

percent (200) are non-Arab whereas only 20 percent (forty-seven) are Arab. Hence,

Arabs comprise only about 5 percent of all ethnic groups. Yet fifteen of forty-one (36

percent) central religious wars involve Arab Muslim groups, almost as many

(twenty; 49 percent) as involve non-Arab Muslim groups. This fact would lead us

to suspect that Arab Muslims are particularly prone to religious conflict onset more

than any other non-Arab Muslim groups in the world. To assess whether this corre-

lation is spurious, I employ a cross-country regression model.

Dependent Variables

For each kind of civil war—non-identity based, identity-based but non-religious,

peripherally religious, and centrally religious—the dependent variable is equal to

“1” if a civil war of that type was initiated in that country-year, and is “0” otherwise.

Independent Variables

Earlier studies measured the relationship between Islam and civil conflict using a

measure of the total Islamic share of the population or a dummy variable for

Muslim-plurality countries (e.g. Fish, Jensenius, and Michel 2010; Karakaya

2015).30 However, such aggregate measures by themselves do not enable us to

distinguish potential independent effects that Arab and Non-Arab Muslim popula-

tions have on conflict propensities. To include these disaggregated demographic

measures, the RCW dataset updates and derived data on the relative size of ethnic

groups as found in the Ethnic Power Relations (EPR) Ethnic Dimensions (EPR-ED)

dataset. This enables us to measure the total number of ethnic groups (and percent-

age of the population) that is included or excluded from power in the central gov-

ernment that is non-Muslim, Arab Muslim, and non-Arab Muslim. The percent of a

population that is excluded from power affords a more specific measure of

grievance.

Control Variables

In the base model, the following correlates of civil war onset identified in the

literature that could confound the Islam-conflict relationship are employed: the level

and annual change in the log of GDP per capita,31 logged population, democracy

(polity2) score32 and its square, a dummy measure of political instability,33 a dummy

for the Cold War (1945–1990), a cubic polynomial of peace years (Carter and

Signorino 2010), and a series of region dummy variables.34

Results

Table 5 shows the base model logit regressions of civil war onset for the 1947 to

2014 period for both types of non-religious and religious civil war.

Toft 1621

T a b

le 5 .

C o rr

el at

es o f R

el ig

io u s

an d

N o n re

lig io

u s

C iv

il W

ar O

n se

t, 1 9 4 7

to 2 0 1 4 .

D V : C iv il

W ar

O ns

et

N on

re lig

io us

C iv il

W ar

R el

ig io

us C iv il

W ar

N o n -I

D ID

A ll

P er

ip h er

al C

en tr

al A

ll

A ra

b M

u sl

im (p

er ce

n t

E x cl

.) �

3 .3

2 2 .9

3 **

* 1 .6

6 *

1 .4

7 3 .3

3 **

* 2 .4

6 **

* (4

.8 1 )

(0 .8

9 )

(0 .7

4 )

(1 .1

4 )

(0 .8

1 )

(0 .7

1 )

O th

er M

u sl

im (p

er ce

n t

E x cl

.) �

0 .6

2 1 .9

2 *

0 .7

4 1 .6

7 �

1 .2

5 0 .4

7 (1

.6 4 )

(0 .8

1 )

(0 .6

8 )

(1 .0

7 )

(1 .1

7 )

(1 .0

3 )

C h ri

st ia

n (p

er ce

n t

E x cl

.) 0 .6

4 2 .0

8 **

* 1 .5

2 **

* �

2 .3

5 *

� 1 .1

4 �

1 .6

2 (0

.6 5 )

(0 .5

6 )

(0 .4

2 )

(1 .1

0 )

(1 .6

5 )

(1 .0

2 )

O th

er R

el ig

io n

(p er

ce n t

E x cl

.) �

0 .5

1 �

2 .1

2 �

0 .9

5 �

0 .2

6 �

4 .8

3 �

2 .3

6 þ

(0 .8

3 )

(1 .5

1 )

(0 .6

9 )

(1 .2

7 )

(3 .0

6 )

(1 .2

9 )

Lo g(

G D

P p er

ca p it a)

� 0 .8

2 **

* �

0 .1

9 �

0 .5

5 **

* �

0 .1

0 �

0 .4

1 þ

� 0 .2

1 (0

.2 1 )

(0 .2

2 )

(0 .1

6 )

(0 .2

4 )

(0 .2

3 )

(0 .1

6 )

D Lo

g( G

D P

p er

ca p it a)

2 .7

5 þ

1 .4

6 1 .8

6 �

2 .5

3 �

2 .6

1 *

� 2 .4

8 þ

(1 .6

4 )

(2 .5

8 )

(1 .1

7 )

(2 .1

5 )

(1 .2

7 )

(1 .2

9 )

P o lit

y Sc

o re

4 .3

1 *

2 .1

1 3 .2

5 *

5 .7

4 þ

6 .4

9 *

5 .6

6 *

(2 .1

9 )

(1 .9

0 )

(1 .3

9 )

(3 .0

9 )

(2 .9

3 )

(2 .2

5 )

P o lit

y Sc

o re

Sq .

� 4 .0

9 þ

� 1 .8

2 �

3 .1

2 *

� 5 .4

9 þ

� 5 .8

3 *

� 5 .3

7 *

(2 .0

9 )

(2 .0

7 )

(1 .3

9 )

(2 .8

9 )

(2 .8

7 )

(2 .1

6 )

P o lit

ic al

In st

ab ili

ty 0 .8

0 *

0 .9

0 **

0 .7

9 **

0 .1

6 0 .8

6 *

0 .4

4 (0

.4 0 )

(0 .3

5 )

(0 .2

8 )

(0 .4

9 )

(0 .3

9 )

(0 .2

9 )

Lo g(

p o p u la

ti o n )

0 .0

6 0 .3

6 **

0 .1

7 *

0 .5

0 **

0 .6

2 **

* 0 .5

0 **

* (0

.1 0 )

(0 .1

3 )

(0 .0

8 )

(0 .1

6 )

(0 .1

6 )

(0 .1

1 )

C o ld

W ar

0 .8

3 *

� 0 .1

1 0 .4

5 þ

0 .0

9 �

0 .2

0 0 .1

5 (0

.4 0 )

(0 .4

2 )

(0 .2

6 )

(0 .3

0 )

(0 .4

8 )

(0 .2

8 )

M id

d le

E as

t &

N . A

fr ic

a 0 .1

6 2 .2

0 **

0 .6

7 0 .2

9 2 .8

2 *

1 .2

9 *

(0 .6

9 )

(0 .7

9 )

(0 .4

3 )

(0 .6

7 )

(1 .1

3 )

(0 .6

0 )

Su b -S

ah ar

an A

fr ic

a �

0 .7

9 þ

2 .2

7 **

0 .2

3 0 .6

4 1 .0

0 0 .8

7

(c on

tin ue

d)

1622

T a b

le 5 .

(c o n ti n u ed

)

D V : C iv il

W ar

O ns

et

N on

re lig

io us

C iv il

W ar

R el

ig io

us C iv il

W ar

N o n -I

D ID

A ll

P er

ip h er

al C

en tr

al A

ll

(0 .4

2 )

(0 .8

0 )

(0 .3

2 )

(0 .5

6 )

(1 .1

7 )

(0 .5

7 )

So u th

A si

a 0 .0

5 2 .0

5 *

0 .6

9 0 .2

1 1 .9

4 1 .0

2 (0

.6 9 )

(0 .8

5 )

(0 .4

5 )

(0 .7

3 )

(1 .3

3 )

(0 .6

3 )

E as

t A

si a

0 .0

6 1 .1

2 0 .2

2 0 .4

0 1 .6

6 0 .8

7 (0

.5 0 )

(1 .2

1 )

(0 .4

0 )

(0 .6

1 )

(1 .2

1 )

(0 .5

7 )

C o n st

an t

� 0 .7

5 �

9 .6

4 **

* �

2 .9

8 *

� 9 .5

2 **

* �

1 1 .4

2 **

* �

9 .0

9 **

* (2

.0 1 )

(2 .3

9 )

(1 .5

1 )

(2 .6

4 )

(2 .8

7 )

(1 .9

6 )

P ea

ce Y

ea rs

In cl

u d ed

? Y

es Y

es Y

es Y

es Y

es Y

es #

o f C

o u n tr

ie s

1 6 5

1 6 5

1 6 5

1 6 5

1 6 5

1 6 5

# o f O

b se

rv at

io n s

8 ,3

8 0

8 ,3

8 0

8 ,3

8 0

8 ,3

8 0

8 ,3

8 0

8 ,3

8 0

P se

u d o

R -S

q u ar

e 0 .0

8 6

0 .1

8 6

0 .1

0 5

0 .1

4 4

0 .1

9 3

0 .1

5 3

R O

C C

u rv

e A

re a

0 .7

7 3

0 .8

7 9

0 .7

8 3

0 .8

3 7

0 .8

9 6

0 .8

4 5

N ot

e: C

o u n tr

y- cl

u st

er ed

ro b u st

st an

d ar

d er

ro rs

in p ar

en th

es es

. ID ¼

Id en

ti ty

-b as

ed ci

vi l w

ar .

þ p

< 0 .1

0 , *p

< 0 .0

5 , **

p <

0 .0

1 , **

*p <

0 .0

0 1 .

1623

These models provide support for Hypothesis 1b: centrally religious civil wars

are disproportionately an Arab Muslim phenomenon, even when controlling for

standard variables and potential confounders. Exclusion of other religious groups

has no effect on the probability of the onset of central religious civil war. There is no

evidence of Arab Muslim exceptionalism for other kinds of civil wars. The presence

or absence of particular religious groups has no effect for non-identity based civil

wars. The exclusion of Christian ethnic groups has just as much of an impact on the

onset of identity-based but non-religious civil wars as Muslim ethnic groups. By

contrast, the exclusion of Christian ethnic groups is negatively related to the onset of

peripherally religious civil wars.

Results using the same base specifications but for the post-Cold War period are

presented in Table 6.

After the Cold War, the size of excluded Arab Muslim ethnic groups is positively

and significantly related to the onset of centrally religious civil wars, but no other

types. By contrast, the size of excluded Christian populations is related to the onset

of identity-based but non-religious civil wars, but no other kinds of civil war.

Together, the results indicate that Arab Muslims are more likely to rebel for religious

reasons, whereas Christians are more likely to rebel for secular reasons after 1990.

So there is an Arab Muslim difference, but only where religion is central. Table 6

also indicates that some correlates of civil war differ depending on religious cen-

trality. High incomes are negatively associated with the onset of non-religious civil

war but not religious civil war after the Cold War.

Robustness Checks

A host of potential confounders to the baseline model that other studies indicate may

cause religious civil war are also included in the analysis.35 None was consistently

significant across the specifications and their addition singly or together does not

change the main results on Arab exclusion. These variables include mean years of

school,36 gender inequality in educational attainment, the mean and variability in

terrain ruggedness,37 youth bulge,38 the log of oil and gas income per capita,39 state

age, year of independence, or linguistic or religious fractionalization.40 The results

remain if we restrict the sample to the Middle East and North Africa region only: the

size of the excluded Arab population, but not non-Arab Muslim population, remains

positive and statistically significant. There is no support for the result of Karakaya

(2015) that other socio-economic and political variables—particularly youth

bulges—account for the onset of religious civil war; nor is centrally religious civil

war driven by ethnic or linguistic heterogeneity in general. Instead, the results point

to religious civil war as a distinctively Arab Muslim phenomenon after the Cold

War. This is not a result that is accounted for by any of the measures heretofore

tested. Thus, Toft (2007) should be corrected slightly: the puzzle is not one of Islam

and religious civil war so much as a puzzle of Arab Muslims and religious civil

war.41 This is a particularly worrisome finding in that it is the Arab world that is seen

1624 Journal of Conflict Resolution 65(9)

as the heartland for Islam and it from this region of the world that we see more

fundamentalist and extremist interpretations of Islam emanating.

So why should we expect more religious outbidding among Arab Muslims than

non-Arab Muslims? The answer to this question can be found in Toft’s theory on

religious outbidding (Toft 2007), which built on Jack Snyder’s (2000) argument that

Table 6. Correlates of Religious and Nonreligious Civil War Onset, 1991 to 2014.

Non-Religious Civil War Religious Civil War

DV: Civil War Onset Non-ID ID All Peripheral Central All

Arab Muslim (percent Excl.) 1.31 1.66 1.40 �49.29 3.76* 2.70* (7.56) (1.46) (1.61) (40.14) (1.52) (1.23)

Other Muslim (percent Excl.) 1.78* 1.95 1.53* 0.37 0.16 0.16 (0.88) (1.92) (0.77) (1.77) (1.62) (1.25)

Christian (percent Excl.) 0.80 2.89** 1.99* �4.77þ �3.42 �3.06 (1.34) (1.01) (0.81) (2.50) (2.72) (1.87)

Other Religion (percent Excl.) 3.76þ 2.65 2.03 �1.28 �23.21 �7.61 (2.19) (2.42) (1.48) (6.16) (18.94) (6.13)

Log(GDP per capita) �1.57** �0.25 �0.67* 0.35 �0.45 �0.02 (0.56) (0.30) (0.28) (0.35) (0.28) (0.23)

D Log(GDP per capita) 4.59 �1.89 0.13 �5.50* �2.22þ �3.22* (3.07) (1.71) (1.70) (2.46) (1.20) (1.54)

Polity Score 19.40* 0.00 3.40 15.51** 9.91* 10.72** (9.13) (3.30) (2.76) (4.95) (4.91) (3.59)

Polity Score Sq. �20.35* �0.13 �3.87 �18.4*** �9.54* �11.7*** (8.69) (3.15) (2.81) (4.35) (4.85) (3.41)

Political Instability �0.17 0.71 0.43 �0.91 0.32 �0.17 (0.73) (0.59) (0.44) (0.67) (0.62) (0.44)

Log(population) �0.45þ 0.33 0.03 0.75*** 0.78** 0.60*** (0.24) (0.23) (0.14) (0.14) (0.28) (0.15)

Middle East & N. Africa 15.7*** 2.19 2.39þ �0.76 1.60 0.41 (3.64) (1.38) (1.34) (1.15) (1.30) (0.86)

Sub-Saharan Africa 14.9*** 1.86 1.79þ 0.48 �0.12 0.34 (1.54) (1.22) (1.07) (1.06) (1.22) (0.80)

South Asia 15.7*** 1.30 1.93 0.94 0.93 0.86 (1.92) (1.57) (1.32) (0.91) (1.57) (0.87)

East Asia 16.3*** 2.46þ 1.92 �0.28 0.88 0.36 (2.08) (1.38) (1.19) (1.05) (1.55) (0.98)

Constant �9.46* �8.97** �2.31 �16.22*** �12.00** �11.6*** (3.75) (2.83) (2.42) (3.32) (3.80) (2.56)

Peace Years Included? Yes Yes Yes Yes Yes Yes # of Countries 160 160 160 160 160 160 # of Observations 3,638 3,638 3,638 3,638 3,638 3,638 Pseudo R-Square 0.256 0.258 0.195 0.234 0.196 0.169 ROC Curve Area 0.938 0.919 0.879 0.916 0.882 0.862

Note: Country-clustered robust standard errors in parentheses. ID ¼ Identity-based civil war.

þp < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001.

Toft 1625

democratization can actually exacerbate nationalist fervor and ethnic conflict

because the democratization processes increases competition between rival

groups.42 Toft extended Snyder’s argument by pointing out that elites within gov-

ernments or opposition movements can also outbid each other in order to enhance

their religious credentials both domestically and transnationally. Elites that establish

themselves as the most credible defender of the faith will then gain the support and

resources needed to maintain their tenure or defeat the existing government. The

practice of religious outbidding is particularly likely to unfold in the Arab World

given that the Arab world has been experiencing more transitions to democracy than

other regions of the world (Toft 2007).43 This makes the Arab world particularly

susceptible to practices of religious outbidding, a practice the resonates with sup-

porters at home and abroad.44

One only has to look at the events unfolding in the wake of the Arab Spring to see

these dynamics at play. For instance, strong domestic and international pressure to

democratize following the ousting of Muammar Gaddafi in Libya in 2011 paved the

way for a process of religious outbidding in which various Salafist Jihadist groups

emerged in a civil war that started in 2014, including the Islamic State (Wehrey

2014). Similarly, when Saddam Hussein was removed from power in Iraq in 2003,

the armed opposition initially consisted mainly of former members of the Ba’ath

party. Yet, Salafist Jihadist groups soon “hijacked” the insurgency in Iraq (Gerges

2017). In short, the religious civil wars in the Arab World stem from a process of

democratization that increases competition and ultimately instability.

Conclusion

This paper has revisited the trends, causes, and dynamics of religious civil war using

new data that are current to the end of 2014. Overall, the findings confirm earlier

indications that religious identity and belief matters for the incidence and dynamics

of religious civil war, and that the politics of the Arab Muslim world lead to religious

civil wars in a way that they do not for non-religious civil war. All else equal, Islam

has bloody innards, if not bloody borders, in reference to Samuel Huntington’s astute

observation some two decades ago, although the findings reported here are more

nuanced. Furthermore, the puzzle that emerges more clearly is one about Arab

Muslims and not one of Islam and religious civil war. The results are clear that

religion is a motivator for rebellion among Arab Muslims while their Christian

counterparts are more likely to be motivated along secular lines after 1990. There

is indeed an Arab Muslim difference, but it arises only where religion is a central

feature in a civil war. Finally, the negative relationships of high income with the

onset of non-religious civil war but not religious civil war after the Cold War

provides further evidence that religion needs to be considered an important factor

in its own right in motivating people to violence (Hassner 2009, 2016).

These findings contrast starkly with the classic “opportunity” and “greed”-based

models of civil war, as well as with the view that linguistic rather than religious

1626 Journal of Conflict Resolution 65(9)

cleavages are most important for civil war onset. Religion matters for conflict and is

likely to do so into the future.

Author’s Note

Replication material will be made available on the author’s website at the time of publication.

All errors are mine.

Acknowledgments

This manuscript benefited immensely from the research and statistical assistance of John

Chin, additional research support from Jackie Faselt, fellowship support from Princeton

University, where I was the World Politics Fellows (2015–2016), insights from participants

at the 2018 International Studies Association Annual Meeting and two anonymous reviewers.

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) received no financial support for the research, authorship, and/or publication of

this article.

ORCID iD

Monica Duffy Toft https://orcid.org/0000-0001-5550-292X

Supplemental Material

Supplemental material for this article is available online.

Notes

1. See Online Appendix for the list of cases, code book of definitions, coding rules, sources,

comparative analyses with other data sets and narratives of each war.

2. The literature on religious terrorism is vast. A good place to start is Hoffman (1998) and

Juergensmeyer (2000).

3. It should be noted that some experts in the field have called into question the methodol-

ogy used in Maoz and Henderson’s 2020 book titled Scriptures, Shrines, Scapegoats, and

World Politics: Religious Sources of Conflict and Cooperation in the Modern Era.

Specifically, they point to flawed analysis in the Religion and State Dataset. The authors

have conceded that there was an error in the calculation of the base year of their dataset,

but claim that their finding holds true when correcting for this mistake. More information

on the debate can be found here: https://www.thearda.com/ras/openletter/

4. Strikingly, no significant relationship between religious grievances and the use of

violence has yet been found (Basedau et al. 2015).

5. See Online Appendix for a list of cases.

Toft 1627

6. Replication results of Bormann, Cederman, and Vogt (2017) are available in the Online

Appendix.

7. As Jonathan Fox (2014) observed, “availability of the necessary data to engage in quan-

titative research on religion and civil wars is a relatively recent development.” Indeed, a

lack of reliable and up-to-date data remains an obstacle. Svensson and Nilsson recently

published (2017) the Religion and Armed Conflict (RELAC) data set. RELAC is similar

to this one in what it allows to be tested, but is limited in its time frame to only 1975 to

2015 and its focus on Islam, thereby not allowing for an assessment of nonreligious civil

wars to religious ones and the role of religion and different faith traditions over a broader

sweep of time.

8. The universe of civil wars in the RCW data was determined by arbitrating disagreements

between major civil war datasets, including Toft (2007). See Online Appendix.

9. This criterion excludes anti-colonial armed conflicts and liberation struggles against

overseas colonial powers, which ACD labels as “extra-systemic” and COW labels as

“colonial” extra-state wars.

10. This criterion excludes what COW labels “intercommunal wars” between non-state

actors and “regional internal wars” between a regional sub-unit and a non-state actor,

but includes violent coups pitting two government actors.

11. This criterion excludes spontaneous mob actions or riots such as in the Albanian pyramid

scheme crisis in 1997.

12. Prior work included major civil wars with at least an average of 1,000 deaths per year. I

include lower-level conflicts and adopt the popular Fearon and Laitin (2003) threshold here.

13. This criterion excludes one-sided massacres and genocides.

14. See Cavanaugh (2009) and Gunning and Jackson (2011) for critiques of previous con-

cepts of religious violence.

15. For a historical review of religious civil war prior to 1946, see Bonney (2011) and

Wellman (2007).

16. See codebooks https://ucdp.uu.se/downloads/ucdpprio/ucdp-prio-acd-191.pdf; https://cor

relatesofwar.org/data-sets/COW-war/intra-state-wars-v5-1.zip/view

17. RWC dataset includes conflicts from 1945 to 2014 (sixty-nine years); RELAC dataset

includes conflicts from 1975 to 2015 (forty years); Running on faith dataset includes

conflicts from 1946 to 2004 (fifty-eight years); RCDC dataset includes conflict from

1990 to 2010 (twenty years). Additionally, it only includes conflicts in 130 developing

countries; the Fox-PITF dataset includes conflicts from 1960 to 2009 (forty-nine years).

18. See Online Appendix for models and replications. Results are robust to both.

19. Svensson (2012), for example, identifies a constant level of religious identity conflicts

from 1975 to 2010 (p. 23), whereas he shows a steady rise in religious issue conflicts

(including those where religion is central) from 1979 but no clear upward trend since

[1991] 2000 (p. 26, 41). Meanwhile, Fox (2012, 2014) identifies a wave of religious “state

failure” starting with the Iranian Revolution, which peaked in the early 1990s, declined,

and then stabilized. By contrast, the RCW data comports more closely with Burstein

(2016), who shows a rise of religious terrorist violence beginning in the 1990s and

exploding since about 2000 accompanied by a decline in secular terrorism.

1628 Journal of Conflict Resolution 65(9)

20. This is the Eritrean civil war by ELF (1961–1972) versus Eritrean civil war by ELPF

(1974–1991).

21. The two-tailed t-test of difference-in-means for total deaths between central religious

wars and others is statistically significant (p ¼ .0012). The analogous t-test for battle

deaths is insignificant overall (p ¼ .17). Nordas 2007 finds similar results.

22. De Juan (2008) explains the relative bloodiness of religious civil wars as a result of elites

that adopt religious rhetoric, which pays dividends in rebel recruitment, cohesion, and

strength. By contrast, others argue that religious bidding is endogenous to violence (e.g.

Isaacs 2016).

23. The two-tailed t-test of difference-in-means for total deaths between central religious