Bad Religion? Religion,Collective Action,and the Onset of Armed Conflict in Developing Countries

Abstract

Anecdotal evidence from many armed conflicts suggests that religion incites violence. Theoretically speaking, several facets of religion can create motives and opportunities to overcome the collective action problems associated with organized violence. However, empirical research has hitherto found no conclusive answer on the extent to which religion is connected to armed conflict onset. Contributing to the filling of this gap, we use a new database that incorporates important religious factors that previous studies left largely untested. The data set covers 130 developing countries for the period 1990 to 2010. Results from logistic regressions confirm our expectation that certain religious factors fuel armed conflict—in particular, the overlap of religious and other identities, religious groups’ grievances, and religious leaders’ calls for violence. We also find that religious determinants vary in their impact according to whether conflicts are religious or not in origin.

Keywords

internal armed conflict rebellion religion collective action civil wars

Anecdotal evidence from many theaters of conflict around the world suggests that religion fuels organized violence. In Nigeria, for example, Boko Haram rebels have taken up arms in order to push for the introduction of Islamic law throughout the whole country. The Syrian civil war has increasingly developed sectarian overtones. Even when religious claims are seemingly largely absent, warring factions often differ in their religious affiliation. In the Sri Lankan civil war, Buddhists legitimized the fight against Tamils on the ethno-religious basis of Sinhala Buddhism. Protestants and Catholics have fought in Northern Ireland, and Baptists and the central Hindu government of India have faced off in Nagaland.

From a theoretical perspective, the concept of collective action may explain why religion can facilitate armed conflict onset. Religion—or more precisely several facets of it—can create opportunities and motives for rebellion, thus overcoming the collective action problems of organized violence: religious identity structures offer opportunities for mobilization while grievances over discrimination provide motives for the aggrieved parties to take up arms. Despite religious peace norms, religious leaders have the capacity to justify the use of violence or even to make its enactment imperative; religious beliefs may also help combatants overcome the fear that they experience in battle.

Despite ample anecdotal evidence and strong theoretical plausibility, empirical research has found no conclusive answer regarding the extent to which religion is connected to armed conflict onset. We argue that this inconclusiveness is in large part due to the focus of previous studies having been on the readily available data about religious diversity, and the literature has thus left several theoretically important religious conditions as yet untested.

The unique data set that we have compiled on Religion and Conflict in Developing Countries (RCDC) contributes to the filling of this gap. RCDC includes a number of widely untested but theoretically important variables such as the overlap of religious and other social identities, grievances over discrimination, and religious leaders’ calls for violence. Moreover, RCDC comprises a recoding of existing demographic indices and new codings of types of religious conflict onsets. Results of logistic regressions support our initial suspicion that the lack of answers provided in previous studies is linked to the presence of several widely untested but theoretically important religious factors.

The article proceeds as follows: we first review the existing quantitative literature on the religion–conflict link, showing that results are rather inconclusive and important religious factors have thus far remained largely untested. We then elaborate on our conceptual framework, which builds on an examination of the collective action problems of rebellion. Particular facets of religious structure (diversity and identity overlaps) and religious practice (grievances over discrimination and religious calls for violence) are likely to offset these intrinsic challenges—by providing either opportunity or motive. Moreover, we assume that the relevance of religious variables increases when warring factions differ in their religious affiliation and/or when the armed conflicts are fought over theological differences.

The following section, first, discusses in depth the codings and variables of the RCDC data set, which covers 130 countries for the period 1990 to 2010. Second, we outline our empirical strategy, which employs logit regressions and a number of robustness tests. Results confirm our expectation that specific religious factors fuel armed conflict—in particular, the overlap of religious and other identities as well as grievances over discrimination. We also find that armed conflicts differ from religious conflicts. Interreligious conflicts are mainly driven by identity overlaps, while conflicts over theological issues result from an aggressive religious discourse. In concluding, we summarize our findings and also lay down some pointers for future research.

Previous Research on the Link between Religion and Armed Conflict

Since the terrorist attacks of September 11, 2001, the link between religion and violent conflict has attracted increasing scholarly attention (e.g., Toft, Philpottt, and Shah 2011). A growing body of quantitative literature has investigated a number of causal mechanisms that link religion to armed conflict onset. Though not always made explicit, almost all works adopt the idea that several aspects of religion can create the necessary opportunity and/or motive to help overcome the collective action problems faced in trying to perpetrate organized violence (Fearon and Laitin 2003). As “religion” is a phenomenon notoriously difficult to define,¹ it makes sense to adopt a distinction used in the study of ethnicity and to disaggregate the concept into “religious structure” and “religious practice” (Chandra and Wilkinson 2008). Religious structure in our understanding refers to rather static religious demographic constellations and is mainly connected to the opportunity for organized violence. In contrast, religious practice comprises those aspects related to human agency—such as the characteristics of the current religious discourse, which can be subject to rapid change. Religious practice more frequently provides the motive for violence.

The most obvious element of religious structure is religious diversity. Religious diversity refers to a constellation wherein several religious groups are simultaneously present in a given society—it does not, however, make any judgment about the quality of the relationships existing between these groups. Comparable to the study of ethnicity and conflict (Seul 1999; Stewart 2009), these works expect that diversity may result in the intensification of hostile intergroup dynamics—mainly through offering opportunities to potential rebels. Academic studies usually distinguish between three forms of religious diversity, such as fractionalization, polarization, and dominance. Fractionalization should make armed conflict less likely, as the presence of more and smaller groups increases the cost of coordinating joint action for diverse religious groups. In contrast, religious polarization and religious dominance should increase the likelihood of armed conflict occurring, as such constellations favor in-/out-group distinctions (Reynal-Querol 2002; Montalvo and Reynal-Querol 2005). The dominance of only one or two religious groups in a country, moreover, might result in a strengthening of the internal cohesion of the smaller groups due to the (perceived) threat that the dominant group will otherwise discriminate against them. Empirical results, however, are mixed and yield, possibly with the exception of dominance that mostly increases risks, no evidence of robust relationships (e.g., Rummel 1997; Ellingsen 2000; Reynal-Querol 2002; Montalvo and Reynal-Querol 2005; Pearce 2005).

Other works look at several politically relevant cleavages in a given country (e.g., Selway 2011b). According to these studies, the risk of armed conflict increases when religious differences are reinforced by other differences—such as ethnicity, socioeconomic status, or geographic location (Stewart 2008; Selway 2011a). The in-/out-group mechanism—whereby humans by nature privilege in-group members over out-group ones—is intensified through the overlap with other cleavages. Religious groups can overcome collective action problems by highlighting the similarities and differences between the various identity groups in a given country (Seul 1999; Stewart 2009). Notorious examples are the entrenched conflicts between the Muslim-Arab North Sudan and the Christian/Animist-African South Sudan or the fighting between Catholic Croats and Orthodox Serbs, with religious and ethnic identities reinforcing each other in both cases.

Quantitative studies on this determinant are rare, as data for global cross-country analysis is scarcely available—though recent works do support the validity of this assumed relationship. Basedau et al. (2011) find that armed conflict in sub-Saharan Africa is particularly likely when religious and ethnic boundaries run parallel to each other. In a similar vein, Gubler and Selway (2012; also Selway 2011a) look at the relationship between ethnicity and conflict in 102 countries worldwide for which opinion poll data are available; their study yields evidence that civil war becomes less likely when ethnic cleavages cut across frontiers of income, geography, and religion.

Religious practice comprises a huge number of variables, such as religious institutions and all facets of agency and discourse. In line with the literature on relative deprivation and horizontal inequalities (Gurr 1970; Stewart 2008), the existence of institutions that discriminate against particular religious communities increases the risk that discontent will arise. State restrictions on religious education or difficulties to find a job because of one’s personal religious identity can be a strong motive for participating in organized violence as a way to combat such injustices. Whether religious groups actually engage in violence, however, may also depend on the political system in place. When they are excluded from politics and do not trust the regime’s policies the likelihood of them perpetrating violence increases (Hafez 2003; De Soysa and Nordås 2007). For example, prior to the outbreak of armed conflict in Côte d’Ivoire in the early 2000s, many Muslims had become aggrieved that they were accused of being foreigners and denied equal citizenship rights. Political and religious entrepreneurs can capitalize on such grievances, and the escalation of violence becomes more likely in such contexts.

So far, empirical support for this assumption has been mixed. In a study on political violence perpetrated by Muslims and Jews in Israel, Canetti et al. (2010) find that relative economic and political deprivation results in support for violence only when religious actors provide an inflammatory interpretation of these deprivations. Akbaba and Taydas (2011) find in a global sample of countries that religious discrimination by the state against ethno-religious minorities slightly increases the risk of violence (see also Fox, James, and Li 2009).

Other works assume that religious ideas and discourse provide motive and thus can explain why religious groups engage in violence. Religious ideas are shared values and norms that commonly include appeals for appropriate behavior on the part of believers. These ideas are legitimized by a (presumably) transcendental source, and therefore, they are barely subject to negotiation and compromise given their accepted supernatural origin (Horowitz 2009). As a result, religious ideas can increase the likelihood of armed conflict onset if conflict-oriented ideas become the guiding principles of one religious community (Seul 1999). In Sri Lanka, for example, nationalist monks introduced an interpretation of Sinhala Buddhism that legitimized discrimination against Tamils and other minorities. This religious idea has become dominant in the Buddhist community in the country in recent decades and was used to justify the war against the Tamils.

While quantitative studies do not test the impact of religious ideas directly, they nevertheless find that armed conflicts fought over religious ideas last longer and cause more fatalities than those armed conflicts between parties divided only by religious identities do (Fox 2004, 45; Pearce 2005; Svensson 2007; Toft 2007).

Connected to ideas, religious symbols can potentially have the same effect as religious discourse. Hassner’s (2011a) comparative assessment shows that in conflicts fought along religious divides, leaders often initiate the conflict on dates or at times that have a sacred meaning. Moreover, a threat against a particular religious symbol—such as the Danish Mohammed cartoon caricatures of 2005 (Hassner 2011b)—can result in violence because a religious group feels offended by the degradation of its sacred symbol/symbols.

In sum, the quantitative literature on religion and armed conflicts has failed to test the whole “menu” of possible religious determinants. Given the limitations with regard to available data, the measurement of religious variables has thus far been mainly confined to the testing of simple religious diversity measures and the religious characteristics of armed conflict. Further variables—including almost all aspects of religious practice—have been investigated only sporadically, or in limited samples at best.

How Religion Can Overcome the Collective Action Problems of Organized Violence

Our argument is simple: we claim that religion—or, more precisely, certain aspects of religion in a given setting—can provide both the opportunity and motive for armed conflict, and hereby can help overcome the collective action problems faced by those wishing to rebel against the state (Olson 1965; Lichbach 1998). Following on, we identify the religion-specific conditions that are likely to enable or assist rebels to successfully overcome these challenges.

Intrastate-armed conflict is defined as a violent conflict occurring between state security forces and nonstate actors. For the latter, it is a significant challenge to successfully organize rebellion against the state. Potential rebels need to overcome the collective action problems that are connected to the necessity of having (a) the opportunity to launch collective military action and (b) enough people sufficiently motivated to join a rebel group and enact violence (Collier and Hoeffler 2004).

Specific religious factors can help potential rebels to overcome these collective action problems. More precisely and as shown in Figure 1, we argue that specific aspects of religious structures—such as religious diversity and the overlap of religious and other identities—are suited to ease the collective action problems related to opportunity, while religious practice—mainly in the form of grievances or calls for violence—will be able to mitigate motivation-related collective action problems.

Figure 1.

Religious factors, causal mechanisms, and violent collective action.

First, specific demographic constellations vis-à-vis religious diversity in a given society can provide the opportunity for rebellion (Seul 1999; Stewart 2009). Religious groups offer recruitment pools of individuals with strong social ties. As already partly discussed in the literature review, scholars differ in opinion regarding what exact constellations of religious diversity are most conducive to violent collective action. Some argue that higher fractionalization may be less suitable for successful mobilization—as higher diversity makes violent action more difficult to coordinate. The cost of coordinating various groups will increase while the strength of any one particular religious group will be relatively low in such a setting. Consequently, the costs of taking up arms will exceed the expected benefits of doing so (Collier and Hoeffler 2004; Reynal-Querol 2002). In contrast, a demographic structure characterized by two more or less equally sized religious groups (polarization) will be more conducive to the undertaking of collective action (Reynal-Querol 2002; Montalvo and Reynal-Querol 2005). Such a social structure will tend to make ascriptions along religious identity boundaries easier; as a result, religious identities will be more salient. Examples are the sectarian violence between Sunni and Shia Arabs in Iraq or the Northern Ireland conflict between Catholics and Protestants. Similarly, the dominance of one group is likely to increase the chance of collective action problems being overcome (Collier and Hoeffler 2004). Religious identities will be strengthened in a societal setting that is marked by the presence of a dominant religious group and one or more smaller ones.² The civil war between the Buddhism majority and the Hindu minority in Sri Lanka exemplifies such a constellation. Regarding the three classic variables of religious diversity, we expect the following to be true:

Hypothesis 1a: Religious fractionalization reduces the opportunity for collective action and thus decreases the likelihood of armed conflict onset.

Hypothesis 1b: Religious polarization facilitates the opportunity for collective action and thus increases the likelihood of armed conflict onset.

Hypothesis 1c: Dominance of one religious group facilitates the opportunity for collective action and thus increases the likelihood of armed conflict onset.

Another aspect of religious structure stresses the relationship of religious to other social identities, which also seemingly increases opportunity. Political science has traditionally related the cleavage structure to voting behavior and the party system (Lipset and Rokkan 1967; Taylor and Rae 1969). Recent work shows that crosscutting cleavages reduce ethnic voting in developing countries (Dunning and Harrison 2010). Important works relate crosscutting or parallel cleavages and identities to ethnic and religious conflict (Horowitz 1985; Stewart 2009; Gubler and Selway 2012): when religious and other identities overlap, the identity differences between the various religious groups become more salient—thereby reducing opportunity costs for potential rebels. In addition, tensions in other identity dimensions automatically spill over into the religious domain. At least three dimensions can reinforce religious identities: first, ethnic and religious identities can mutually reinforce one another when most believers are also members of a specific ethnic group (Seul 1999; Stewart 2009; Selway 2011a). Examples are the rebellion by the Muslim Moro minority against the mostly Christian government of the Philippines or the conflict between the mostly Hindu government of India and the Christian Nagas.

Second, the literature on ethnic conflict indicates that the concentration of groups in a specific geographical area creates the opportunity for collective action (Toft 2003). The area of settlement of a religious group may also work as—in the famous words of Mao Zedong—“the water in which the rebel fish can swim.” In line with these assumptions, the mobilization of religious groups will thus be easier if religious and regional identities overlap. Third, studies on horizontal inequalities have shown that differences in political, economic, or other types of status between identity groups may enhance differences and thus increase the opportunity for mobilizing the believers (Stewart 2008; Østby, Nordås, and Rod 2009). In war-torn African countries such as Côte d’Ivoire, Nigeria, and Sudan, the pronounced overlaps of religious and other identities has been widely said to have contributed to violence. With regard to the relationship between religious and other identities, we assume the following:

Hypothesis 2a: Overlapping religious and ethnic identities increase the opportunity for collective action and thus the likelihood of armed conflict onset.

Hypothesis 2b: Overlapping religious and regional boundaries increase the opportunity for collective action and thus the likelihood of armed conflict onset.

Hypothesis 2c: Overlapping religious and economic boundaries increase the opportunity for collective action and thus the likelihood of armed conflict onset.

Economic or other inequalities between religious groups make differences more salient and thus increase the chance to mobilize—but they are also likely to provide motive, in the form of grievances. A string of recent empirical studies have found that the objective political deprivation of ethnic groups increases conflict risks (Cederman, Weidmann, and Gleditsch 2011); there is also preliminary evidence that the same holds true for ethno-religious minorities as well (Akbaba and Taydas 2011). Possibly more often than is assumed, however, pronounced inequalities between groups do not translate into subjective grievances. Religious structures do not determine religious practice. Recently, the Minorities at Risk project has produced a list of minorities that did not mobilize politically. When Gurr (1970, 2000) developed his concept of “relative deprivation,” he stressed that objective discrimination becomes politically relevant only when it is subjectively felt by a given group (and sometimes grievances may emerge even without objective deprivation taking place). In such cases, the subjective marginalization of religious groups represents a powerful motivation for the rank and file; this, in conjunction with calling on the organizational resources of religious groups, can be capitalized on by religious political entrepreneurs (Toft 2007; De Juan 2009).

In sum, objective inequalities are not unlikely to create grievances, but it is first and foremost the subjective feeling of being discriminated against that makes believers choose rebellion. The numerous aggrieved religious groups in India (Christians, Muslims, and Sikhs), Northern Ireland (Catholics), and the Philippines (Muslims) that took up arms exemplify this point. We thus expect the following:

Hypothesis 3a: Grievances over discrimination against a particular religious group increase the motivation for violent collective action and thus the likelihood of armed conflict onset.

Religion is functionally equivalent to political ideologies and the promotion of any Weltanschauung. Religion—particularly one wherein a religious tradition claims to have universal, and moreover exclusive, validity—can thus strongly spur political violence (Appleby 2000). The Christian Crusades and Islamic Jihads, for example, sadly underscore the fact that religious ideas can stimulate aggressive action (Horowitz 2009; Latham 2011).

However, the relationship of religious norms and discourse to violent collective action is a deeply ambiguous one (Appleby 2000). Religious norms demanding peaceful behavior, tolerance, and love—“thou shalt not kill,” “love thy neighbor as you love thy self”—can be found in all Abrahamic faiths, as well as in many other religious traditions besides; one may argue, however, that this is only the flip side of the coin. Religious ideas are subject to selective interpretation, and theology can thus also be used to instigate violence.

This can be achieved in many different possible ways, but predominantly through the promotion of the idea by leaders that hostile action against out-group members is theologically justified, beneficial, or even imperative: The “sword” might be used to impose religious norms and beliefs on others; zealots and bigots may demand the punishment of blasphemers, heretics, and pagans. Both Crusaders and recent Islamic terrorist groups—such as Al Qaida or Boko Haram—alike made reference to such religious ideas to legitimize violence.

Participation in violence may be spurred by the belief that a particular military campaign is divinely endorsed: Pope Urban II justified his call for the first Crusade by claiming that Deus lo vult (“God wills it”). In both the First and the Second World Wars, German soldiers wore the words Gott mit uns (“God with us”) on their belt buckles. Violent action may be further motivated by the belief in otherworldly rewards for martyrdom—“72 virgins,” “fallen warriors go to Valhalla and sit at Odin’s table”—or by specific religious practices, such as using magic spells that combatants believe mean that firearms and other weapons can do them no harm.

As ideas are obviously ambiguous, we assume that discourse by religious leaders will make the key difference regarding the impact of religious ideas:

Hypothesis 3b: Calls for violence by religious elites increase the motivation for engaging in violent action and thus the likelihood of armed conflict onset.

Our argument does not necessarily work for all types of armed conflict equally. Armed conflict in general covers all fighting between rebel groups and governments. Religion thus constitutes just one among a multitude of different possible sources of incompatibility between actors. Hence, we expect that the effects of the religious variables outlined in Hypotheses 1 through 3 will be more significant in conflicts with objective religious issues—namely, in what we term interreligious and theological armed conflicts (Pearce 2005; Svensson 2007; Toft 2007).

An interreligious armed conflict is one wherein warring factions differ in terms of their religious affiliation. Examples are the confrontations in India (Hindu government vs. Christian rebels in Nagaland), in the Philippines (Christian government vs. Muslim rebels), and in Sri Lanka (Buddhist government vs. Hindu rebels). Such interreligious conflict is not necessarily fought over theological ideas; rather, it may just be the expression of underlying tensions existing between identity groups due to their secular incompatibilities such as in terms of power or wealth. Given the identity character of interreligious conflict, it is not unlikely that parallel other identities, such as ethnicity, will particularly spur this type of conflict.

Theological armed conflict refers to an incompatibility over religious ideas between the state and the rebel group, such as regarding the introduction of religious law or over a particular state religion. A typical example of this is the demand of certain Islamic groups for the introduction of Sharia in a hitherto secular state. We have to consider a number of aspects for this type of conflict: First, labeling an armed conflict “theological” does not mean we claim that there is solely a theological incompatibility present. In most cases, multiple sources of incompatibilities exist. Also, the theological nature of the conflict can change over time and may vanish during the conflict. At the onset of a conflict (our dependent variable), however, a theological goal ensures the widespread mobilization of believers, and it is not unlikely that this goal will be prominent at this particular moment in time. Second, it is sufficient for only one side to articulate theological goals. The other conflict party, almost inevitably, will react to these and hence theological ideas become a core issue in the conflict.

Finally, it should be noted that religious discourse and the theological goals of warring factions exist independent of each other. A theological conflict requires that at least one side has explicit theological goals that differ from those of the other conflict party. In contrast, religious calls for violence refer to statements made by religious leaders about whether violence can be legitimately used or not; these calls are not necessarily connected to the conflict parties themselves and often emerge before conflict onset, potentially doing so in isolation from the rest of the political discourse. Hence, a call for violence by a religious leader does not warrant a conflict being labeled “theological.”³ Rather, religious calls for violence will increase the likelihood that theological conflicts subsequently occur. For instance, while the Islamist Front Islamique du Salut (FIS) in Algeria had long been demanding Islamic law, it only turned to violence in 1992, once religious leaders had declared this choice legitimate. While we do not expect that inflammatory religious discourse will inherently increase the likelihood of armed conflict, theological armed conflict should, at least, be more affected by such discourse. We expect the following:

Hypothesis 4: Religious factors have a more significant and intense impact on armed conflict onset when conflicts possess an interreligious or theological component.

Hypothesis 4a: Religious identity overlaps increase the risk of interreligious armed conflict onset.

Hypothesis 4b: Conflict-prone religious discourse increases the risk of theological armed conflict onset.

Data and Empirical Strategy

Our new RCDC data set is particularly useful for testing the theoretically important religious factors between which we hypothesize a relationship. RCDC contains a number of new religion-specific variables, such as overlapping group boundaries, grievances over discrimination, or types of religious armed conflict. Moreover, we recoded existing religious demographic variables using uniform sources. RCDC covers the years 1990 to 2010, and includes 130 developing countries in Asia, Latin America, the Middle East, and sub-Saharan Africa.⁴

Data were coded by country year and we have a maximum of 2,730 observations per variable. We were keen to maintain uniform, consistent sources to avoid distorted information. Except for the diversity measures and the controls, we thus used annual Religious Freedom Reports and Human Rights Practices Country Reports (both series of reports compiled by the US Department of State), as well as the Economist Intelligence Unit (EIU) country reports that are published on (at least) a quarterly basis.⁵ Every coding was checked and discussed by the project team, including time-consuming conferences to ensure the quality of data and uniform coding.

Concerning data quality, we relied on the information available from the previously mentioned sources. Certainly, not every possible relevant phenomenon or event will have been coded by us due to the somewhat limited information provided by these sources. Arguably, the coding sources may be biased toward more easily accessible information or events that are considered particularly relevant both domestically and internationally. However, this suggests, at least, that the events included in our data set represent the incidents deemed most significant in the country itself and—most importantly—by the respective religious groups themselves.

Moreover, the reliability and validity of our sources—and especially of the Religious Freedom Reports—might be questionable given the possible ulterior motives of the US government (the publisher). However, Grim and Finke (2011), as well as Fox (2011), have cross-checked the validity of the reports, and they all conclude that they represent a useful and valid coding source for religious factors. Moreover, possible bias in the reports due to the specific interests of the US government should be compensated for by the use of other coding sources (EIU Country Reports).

Dependent Variables

In order to test our hypotheses, we employed three different dependent variables. The variable conflict onset, taken from the Uppsala Conflict Data Program/Peace Research Institute Oslo (UCDP/PRIO) Armed Conflict Onset Dataset (Version 4/2011), was used to measure intrastate armed conflict onset. The dichotomous variable conflict onset, thus, has a value of 1 if there is a conflict onset resulting in more than twenty-five annual battle-related deaths; otherwise, it is 0.⁶ A total of 138 onsets happened within the period under consideration.⁷ In order to capture religious subtypes of armed conflicts (Hypotheses 4a and 4b), we employed two dichotomous dependent variables (interreligious conflict onset and theological conflict onset). We coded each variable based on the UCDP/PRIO conflict onset variable. Interreligious conflict onset takes a value of 1 if at least one intrastate conflict onset happened wherein, according to the RCDC sources mentioned previously, the warring factions differed greatly in their religious affiliation, and a value of 0 otherwise. The variable encompasses a total of sixty onset years in episodes spanning twenty-four different countries. Examples are, for instance, armed conflict onsets in Azerbaijan (1991 [year of onset]) and Côte d’Ivoire (2002 [year of onset]) in which Christians and Muslims confronted each other.⁸

In order to capture the theological dimension of religious conflicts, we created a variable (theological conflict onset) that encompasses those armed conflict onsets based on religious incompatibility. As discussed previously, we coded a conflict as theological when at least one warring faction had explicit religious aims vis-à-vis state policies according to our sources and when these differed from those of the other side in the conflict dyad. This variable encompasses a total of forty-one onset years in episodes spanning twenty-two different countries. Examples include Algeria (onset 1991, rebel group with religious goals: FIS) and Nigeria (2009, Boko Haram).⁹

Independent Variables

The explanatory religious variables included in this study are separated out according to whether they represent religious structure (diversity and overlaps) or practice (calls and grievances). We included three variables describing the classical features of religious diversity (Hypotheses 1a–1c). Existing data sets use different coding sources for their indices. In order to ensure comparability of our measurements of religious diversity, we coded all three indices with data from the World Christian Database (WCD). WCD provides data for sixteen major religions for the period from 1990 to 2010, as well as the percentage of people who are unaffiliated with any religion. In contrast to other cross-national data sets on religious composition, WCD includes information for all 130 countries in our data set. WCD is highly correlated, close to 0.90, to other data sets on religious demographics (Hsu et al. 2008). Lacking more detailed data on demographic dynamics, we use constant values for all three variables of religious diversity.

First, we calculate a religious fractionalization index applying the formula of Alesina et al. (2003; fractionalization):

\sum f r a c t i o n a l i z a t i o n_{j} = 1 - \sum_{i = 1}^{N} s_{i j}^{2} .

In contrast to Alesina et al. (2003), we use more recent data from 2010 and concentrate on major religious persuasions.¹⁰

Second, we construct a new polarization index of the interreligious structure, ranging from 0 to 1 (polarization). We use the Reynal-Querol index, but calculate values based on the WCD data from 2010:

p o l a r i z a t i o n = 1 - \sum_{j = 1}^{N} {(\frac{0.5 - s_{j}}{0.5})}^{2} s_{j} .

As WCD data are available for all countries in the sample, our data cover more countries than Montalvo and Reynal Querol (2005) do.¹¹

Additionally, we coded a new variable on religious dominance, with it taking a value of 1 if at least 60 percent of the country in question’s population adheres to one particular denomination (dominance). In order to distinguish between polarization and dominance, we used the 60 percent threshold to demarcate the existence of a dominant religious group. In contrast, Collier and Hoeffler (2004) use a range of 45 percent to 90 percent of the population and have consequently been criticized for including constellations of polarization (e.g., by Schneider and Wiesehomeier 2008).¹²

We included three binary variables from RCDC that capture various identity overlaps (Hypotheses 2a–2c). This is not the first attempt to measure parallel cleavages in a cross-country manner (e.g., Selway 2011b), but RCDC covers more, particularly conflict-ridden countries and measures the overlaps at the group and not the individual level. We coded whether the sources indicate that religious identities overlap at least partially with ethnic or regional boundaries (ethnic overlap and regional overlap) as well as with economic differences (economic overlap).¹³ For instance, we coded ethnic and religious overlap positively in the case of the Côte d’Ivoire. The Religious Freedom Report (2006) of the US State Department states that “religious affiliations tended to follow ethnic lines. For example, the Mende and Voltaic groups, which included the Malinke and Senufo people, were largely Muslim. The Akan ethnic group, which included the Baoule and Agni people, tended to be Catholic.”

Simultaneously, regional and economic overlaps were positively coded when such parallel identities were explicitly mentioned in the sources. In this sense, our coding may rather reflect the public discourse on such overlaps, not actual differences. As public perception will determine the political impact of such differences, we believe that our codings are a valid approach for capturing this dimension. All three overlap variables are time invariant for the whole period, given that the available sources for the whole sample do not allow for more fine-grained assessments. Group boundaries may have changed in some cases, but they are probably relatively stable over time. Ethnic, regional, and economic overlaps differ: while ethnic and regional overlaps show a correlation about 0.48, the other pairs of variables are less strongly correlated (about 0.37).

With regard to properties of religious discourse (Hypotheses 3a and 3b) we created three new variables. The binary variable grievances over discrimination measures whether or not a religious community feels discriminated against. The variable was only coded positive if one of the RCDC sources indicates that at least one religious community or its adherents in a particular country claim that they are discriminated against in a given year. We find a total of 908 cases in ninety-six countries in which religious groups expressed such grievances. The correlation with objective discrimination (by the state according to laws or official regulations) is rather low with a correlation coefficient of 0.19 (Fox 2004).¹⁴

Furthermore, the variable religious calls for violence was coded when religious actors legitimize or incite violence in a given year. The variable is dichotomous and was coded positive if the sources reported at least one such call. One caveat to this operationalization might be that we do not distinguish between the number of calls within one year or between the importance of the calls in terms of their societal impact. On average, we find fewer than 30 of the 130 countries in a given year. This rather low number suggests that our sources reported mainly calls that represent the most prominent calls.

Control Variables

Control variables were chosen in accordance with sensitivity analyses performed by Hegre and Sambanis (2006). They cover properties of demographics and of the economy, political system, and topography. All of them are theoretically suited to affect collective violent action in one way or another, as they may provide opportunity and/or motive.¹⁵ Logarithm of total population (log population) and rough terrain (terrain) proxy opportunities. Standard economic variables include gross domestic product (GDP) growth (gdp growth), logarithm of per capita GDP (log gdppc), and oil exports (oil exports),¹⁶ being mainly connected to motive. More ambiguous variables include regime durability (durable), measuring the years since the most recent regime change, and the level of democracy (polity2), as measured by the combined polity score (data for both taken from the Polity IV Project). Both may increase opportunity and motive. In addition, a variable reflecting the time lapse since the last event/onset (time since last conflict) was included in all models to minimize any problems of temporal dependence on a history of conflict (Beck, Katz, and Tucker 1998). Except for time since last conflict, all other independent variables were lagged by one year. Table 1 provides an overview of the different dependent and explanatory variables included in our analysis.¹⁷

Table 1.

Summary Statistics, 1990 to 2010.

Variable	Obs	Mean	SD	Min	Max
Armed conflict onset	2,688	0.05	0.22	0.00	1.00
Interreligious conflict onset	2,688	0.02	0.15	0.00	1.00
Theological conflict onset	2,688	0.02	0.12	0.00	1.00
Nonreligious conflict onset	2,688	0.03	0.16	0.00	1.00
Full religious conflict onset	2,688	0.01	0.11	0.00	1.00
Ethnic armed conflict	2,688	0.12	0.32	0.00	1.00
Fractionalization	2,727	0.30	0.22	0.01	0.76
Polarization	2,727	0.47	0.29	0.01	0.98
Dominance	2,727	0.77	0.42	0.00	1.00
Dominance Christ	2,727	0.41	0.49	0.00	1.00
Dominance Muslims	2,727	0.30	0.46	0.00	1.00
Share Christ	2,727	44.20	39.10	0.05	97.40
Share Muslims	2,727	33.90	39.30	0.00	99.80
Economic overlap	2,727	0.25	0.43	0.00	1.00
Ethnic overlap	2,727	0.47	0.50	0.00	1.00
Regional overlap	2,727	0.58	0.49	0.00	1.00
Demographic change	2,727	0.24	0.43	0.00	1.00
Tensions	2,178	0.15	0.36	0.00	1.00
Grievances over discrimination	2,706	0.34	0.47	0.00	1.00
Minority discrimination	2,378	0.77	0.42	0.00	1.00
Religious calls for violence (religious actors)	2,727	0.13	0.33	0.00	1.00
Population	2,673	38.90	144.00	0.07	1330.00
gdp growth	2,562	4.17	6.59	−51.00	106.00
gdp per capita	2,540	3651.00	6597.00	54.50	40837.00
Oil exports	2,673	66.70	190.00	0.00	1580.00
Durability of political regime	2,520	19.20	22.30	0.00	133.00
Polity2	2,486	1.04	6.61	−10.00	10.00
Rough terrain	2,370	2.19	1.44	0.00	4.56
Time since last conflict onset	2,688	7.86	6.06	0.00	20.00
Time since last interreligious conflict onset	2,688	8.94	6.12	0.00	20.00
Time since last theological conflict onset	2,688	9.19	6.11	0.00	20.00
Time since last nonreligious conflict onset	2,617	8.81	6.09	0.00	20.00
Time since last full-religious conflict onset	2,688	9.41	6.11	0.00	20.00
Time since last ethnic conflict onset	2,688	8.05	6.39	0.00	20.00

Model Specification

In our analysis, we focus on the drivers of the probability (p) of conflict onsets, rather than on the probability (1 − p) of the absence of conflict onsets. We specify a logit model¹⁸ as a standard binary choice model (Greene 2008):

\begin{aligned} p (y = 1 | x) = F (β_{0} + β_{1} x_{1} + . . . + β_{k} x_{k}) \\ p (y = 0 | x) = 1 - F (β_{0} + β_{1} x_{1} + . . . + β_{k} x_{k}), \end{aligned}

where y is a Boolean dependent variable, x is a set of control variables (as described previously), and F represents a logistic function that takes a value between 0 and 1. The parameters of the model are estimated using maximum likelihood (ML), assuming therein a logistic distribution. All control variables except for the number of years of peace are lagged by one year. In all specifications, we apply cluster (country) robust standard errors, which specify that the observations are independent across countries but not over time. We report the logit coefficients after exponentiation (odds ratios) to facilitate the interpretation of the coefficients, as the log of the odds ratio $(ln \{p / (1 - p)\})$ is linear in the explanatory variables and for the coefficients in the logit model (Cameron and Trivedi 2009).¹⁹ We also plot the relationship between the predicted probabilities of outcome variables and the predictors of different model specifications.

Results

Table 2 gives an overview of the results, while Tables 3 through 5 show the detailed regression models—including odds ratios and levels of significance for our three dependent variables. Whereas Table 3 presents the results for armed conflict onset as the dependent variable, Tables 4 and 5 show the findings for interreligious conflict onset and for theological conflict onset, respectively. In all the tables, we present nine different models, as per our (groups of) explanatory variables. Tables include odds ratios that also allow assessing the substantive effects of the results not just their statistical significance.

Table 2.

Overview of Results.

		Armed conflict onset	Interreligious conflict onset	Theological conflict onset
Religious structure I	Fractionalization	−−−
	Polarization	−−−
	Dominance	+++
Religious structure II	Ethnic overlap	++	+++
	Economic overlap		++	+
	Regional overlap		+++
Religious practice	Religious calls for violence		+	+++
Religious practice	Grievances over discrimination		+	++

Note: − /+ indicate negative/positive relationship, number of signs indicates level of significance, see Tables 3 through 5.

Table 3.

Regression Results for Armed Conflict Onset as Dependent Variable.

Dependent variable: armed conflict onset, odds ratios
Type of explanatory variable		Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
Structure I: religious diversity	Fractionalization	0.130*** (0.085)
	Polarization		0.294** (0.142)
	Dominance			2.566*** (0.878)
Structure II: religious identity overlaps	Ethnic overlap				1.863** (0.457)
	Economic overlap					1.112 (0.279)
	Regional overlap						1.263 (0.351)
Practice: religious discourse/theological ideas	Grievances over discrimination							1.172 (0.289)
Practice: religious discourse/theological ideas	Religious calls for violence								1.334 (0.350)
Other control variables	Log population	1.391*** (0.128)	1.369*** (0.124)	1.376*** (0.117)	1.312*** (0.107)	1.347*** (0.121)	1.346*** (0.114)	1.344*** (0.126)	1.336*** (0.119)
	gdp growth	1.018 (0.015)	1.017 (0.015)	1.018 (0.015)	1.011 (0.013)	1.014 (0.014)	1.013 (0.014)	1.012 (0.014)	1.013 (0.014)
	Log gdp per capita	0.527*** (0.077)	0.533*** (0.078)	0.562*** (0.076)	0.598*** (0.076)	0.607*** (0.081)	0.616*** (0.082)	0.610*** (0.079)	0.616*** (0.078)
	Durable	1.000 (0.009)	0.999 (0.009)	0.995 (0.008)	0.992 (0.009)	0.993 (0.009)	0.994 (0.009)	0.993 (0.009)	0.994 (0.009)
	Polity2	1.009 (0.022)	1.011 (0.023)	1.003 (0.020)	1.021 (0.023)	1.012 (0.022)	1.013 (0.023)	1.012 (0.024)	1.012 (0.022)
	Terrain	0.998 (0.089)	1.005 (0.089)	1.031 (0.093)	1.076 (0.094)	1.050 (0.091)	1.062 (0.094)	1.036 (0.091)	1.044 (0.087)
	oil exports	1.001** (0.001)	1.001** (0.001)	1.001** (0.001)	1.001** (0.001)	1.001** (0.001)	1.001** (0.001)	1.001** (0.001)	1.001** (0.001)
	Time since last conflict onset	0.900*** (0.024)	0.897*** (0.024)	0.899*** (0.025)	0.894*** (0.024)	0.889*** (0.026)	0.889*** (0.025)	0.888*** (0.025)	0.889*** (0.025)
	Cons	0.043** (0.069)	0.054* (0.086)	0.009*** (0.015)	0.019*** (0.029)	0.017** (0.029)	0.013*** (0.021)	0.017** (0.029)	0.018** (0.029)
	Number of observations	2,103	2,103	2,103	2,103	2,103	2,103	2,094	2,103
	Adjusted R ²	.153	.148	.151	.146	.137	.138	.138	.139

Note: Logistic regression using cluster robust standard errors. All explanatory variables—except “time since last conflict onset” are estimated with a one-year lag. ***p < .01. **p < .05. *p < .1.

Table 4.

Regression Results for Interreligious Conflict Onset as Dependent Variable.

Dependent variable: interreligious conflict onset, odds ratios
Type of explanatory variable		Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
Structure I: religious diversity	Fractionalization	0.977 (0.869)
	Polarization		1.655 (1.304)
	Dominance			1.253 (0.543)
Structure II: religious identity overlaps	Ethnic overlap				5.902*** (3.191)
	Economic overlap					2.362** (0.939)
	Regional overlap						6.656*** (4.621)
Practice: religious discourse/theological ideas	Grievances over discrimination							1.788* (0.625)
Practice: religious discourse/theological ideas	Religious calls for violence								1.946* (0.754)
Other control variables	Log population	1.340*** (0.136)	1.343*** (0.131)	1.337*** (0.137)	1.291*** (0.121)	1.242** (0.111)	1.305*** (0.129)	1.255** (0.126)	1.282** (0.138)
	GDP growth	1.025*** (0.008)	1.024*** (0.008)	1.027*** (0.008)	1.017*** (0.006)	1.025*** (0.008)	1.022*** (0.008)	1.021*** (0.007)	1.025*** (0.008)
	Log GDP per capita	0.519*** (0.099)	0.541*** (0.100)	0.512*** (0.098)	0.531*** (0.127)	0.494*** (0.106)	0.584*** (0.121)	0.529*** (0.100)	0.538*** (0.103)
	Durable	1.013* (0.008)	1.012 (0.008)	1.013* (0.008)	1.010 (0.008)	1.010 (0.008)	1.013 (0.008)	1.012 (0.007)	1.012 (0.008)
	Polity2	1.058** (0.029)	1.059** (0.030)	1.054* (0.029)	1.074** (0.032)	1.047* (0.027)	1.053** (0.027)	1.057** (0.028)	1.050* (0.028)
	Terrain	0.796 (0.138)	0.807 (0.139)	0.796 (0.131)	0.802 (0.130)	0.799 (0.141)	0.843 (0.129)	0.789 (0.124)	0.802 (0.129)
	Oil exports	1.001 (0.001)	1.001 (0.001)	1.001 (0.001)	1.001 (0.001)	1.001* (0.001)	1.001 (0.001)	1.001 (0.001)	1.001 (0.001)
	Time since last interreligious conflict onset 1	0.886*** (0.034)	0.886*** (0.035)	0.886*** (0.034)	0.912** (0.034)	0.896** (0.040)	0.897*** (0.034)	0.883*** (0.035)	0.886*** (0.034)
	Cons	0.029 (0.064)	0.016* (0.035)	0.028 (0.061)	0.011** (0.022)	0.099 (0.192)	0.003*** (0.007)	0.064 (0.126)	0.044 (0.094)
	Number of observations	2,103	2,103	2,103	2,103	2,103	2,103	2,094	2,103
	Adjusted R ²	.164	.166	.165	.206	.177	.190	.170	.170

Note: Logistic regression using cluster robust standard errors. All explanatory variables—except “time since last conflict onset” are estimated with a one-year lag.

***p < .01. **p < .05. *p < .1.

Table 5.

Regression Results for Theological Conflict Onset as Dependent Variable.

Dependent variable: theological conflict onset, odds ratios
Type of explanatory variable		Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
Structure I: religious diversity	Fractionalization	0.380 (0.350)
	Polarization		1.008 (0.864)
	Dominance			1.703 (1.046)
Structure II: religious identity overlaps	Ethnic overlap				1.812 (0.795)
	Economic overlap					1.854* (0.621)
	Regional overlap						1.963 (1.013)
Practice: religious discourse/theological ideas	Grievances over discrimination							2.634** (1.023)
Practice: religious discourse/theological ideas	Religious calls for violence								4.270*** (1.418)
Other control variables	Log population	1.654*** (0.282)	1.633*** (0.263)	1.646*** (0.278)	1.594*** (0.253)	1.559*** (0.242)	1.612*** (0.264)	1.455** (0.230)	1.493** (0.233)
	GDP growth	1.008 (0.024)	1.004 (0.023)	1.008 (0.024)	1.000 (0.022)	1.002 (0.025)	1.002 (0.024)	1.001 (0.020)	1.000 (0.023)
	Log GDP per capita	0.684 (0.164)	0.717 (0.163)	0.699 (0.160)	0.733 (0.175)	0.713 (0.172)	0.755 (0.186)	0.734 (0.166)	0.781 (0.175)
	Durable	1.001 (0.012)	0.998 (0.013)	0.998 (0.012)	0.996 (0.014)	0.996 (0.013)	0.998 (0.014)	0.997 (0.012)	0.997 (0.012)
	Polity2	1.005 (0.036)	1.008 (0.038)	0.999 (0.034)	1.013 (0.037)	0.999 (0.034)	1.004 (0.035)	1.007 (0.036)	0.993 (0.033)
	Terrain	0.816 (0.137)	0.833 (0.141)	0.833 (0.134)	0.822 (0.133)	0.839 (0.138)	0.865 (0.132)	0.828 (0.130)	0.830 (0.116)
	Oil exports	1.001 (0.001)	1.001 (0.001)	1.001 (0.001)	1.001 (0.001)	1.001 (0.001)	1.001 (0.001)	1.001 (0.001)	1.000 (0.001)
	Time since last theological conflict onset	0.916* (0.042)	0.915* (0.042)	0.918* (0.043)	0.920* (0.042)	0.923* (0.045)	0.920* (0.042)	0.905** (0.043)	0.918** (0.040)
	Cons	0.000*** (0.000)	0.000*** (0.000)	0.000*** (0.000)	0.000*** (0.000)	0.000*** (0.001)	0.000*** (0.000)	0.000*** (0.001)	0.000*** (0.001)
	Number of observations	2,103	2,103	2,103	2,103	2,103	2,103	2094	2,103
	Adjusted R ²	.128	.125	.129	.132	.132	.131	.141	.162

Note: Logistic regression using cluster robust standard errors. All explanatory variables—except “time since last conflict onset” are estimated with a one-year lag.

***p < .01. **p < .05. *p < .1.

In our analytical discussion, we first look at the results for Hypotheses 1 through 3, in which we test the impact of religious diversity, identity overlaps, and religious discourse on the three types of armed conflict onset. Then we turn to Hypothesis 4 and discuss whether conflict types vary in the expected manner.

Regarding Hypothesis 1 (a–c), Tables 2 through 5 suggest that it is only partly confirmed. The following observations are noteworthy: first, the classical indicators for religious diversity fractionalization, polarization, and dominance are only significant for armed conflict, but not for religious armed conflicts (models 1–3 in each table). Dominance increases the risk—according to odds ratios by around 2.6 times—and fractionalization reduces the risk and hence behave as in most other studies and according to our hypotheses. However, religious polarization lowers the likelihood of armed conflict onset. In particular, the odds of armed conflict onset is about 0.29 times lower for a one-unit increase in religious polarization. This finding contradicts our expectation that demographically polarized societies are more conflict-prone, at least in the sense argued by Reynal-Querol (2002). It supports the supposition that (religious) polarization reduces conflict propensity, as roughly equally sized groups seemingly counterbalance each other.

In contrast, Hypothesis 2 (a–c) receives much more support (see models 4–6 in Tables 3–5): overlapping identities apparently have a bigger influence across conflict types. For all three types of conflict overlaps matter and, as hypothesized, they consistently increase the onset risk. It is especially parallel ethnic and religious boundaries—and, to a lesser extent, economic and regional overlap—that increase the risk of conflict. According to the odds ratios, the substantive effects are quite large. In countries where ethnicity overlaps with religion, the odds of armed conflict is 1.9 higher times higher than for countries without ethnic overlap; for interreligious conflict all three kinds of overlaps matter: a shift in the variables ethnic overlap from 0 to 1 increases the odds of interreligious conflict to 5.9. If ethnic and regional boundaries run parallel, the odds of interreligious conflict even increase to 6.7.

Real-world examples illustrate the importance of the relationship. In Côte d’Ivoire, as outlined in the previous section, the conflict between the north and the south after 2002—and renewed in 2010—featured all three “overlaps.” Religious groups differ in ethnic terms, have traditionally different areas of settlement, and the south has historically been much wealthier than the north has. Similar constellations are also found in Nigeria, the Philippines, and Sudan (prior to partition). Apparently, the inconclusive results of many former studies mirror, as argued previously, the limited availability of data on some of the empirically and theoretically more important variables and countries.

Tables 2 through 5 also document the findings on aspects of religious practice (Hypotheses 3a and 3b; see models 7–9). Again, both innovative variables—religious calls for violence as well as grievances over discrimination—turn out to be important. However, they particularly reveal the difference between armed and religious armed conflicts. Religion-based calls for violence and grievances over the discrimination of religious groups are significant only in interreligious and theological conflict. Odds ratios in Tables 4 and 5 demonstrate that grievances increase the odds for these onsets by 1.8 and 2.6 times, respectively. Calls for violence by religious leaders almost double the odds of onset in interreligious conflicts and make theological conflict more than four times likelier.

Moreover, the results show that the impact of religious factors varies greatly across different types of conflict. Hypothesis 4 expects that religious factors have a much more significant and substantive effect in religious conflicts. By and large, this expectation matches with the results obtained. Especially, variables of religious practice are significant in the religiously charged conflict only, but, more important, potential religious determinants behave in a markedly different way vis-à-vis conflict type.

Higher risks for armed conflict onset result from dominance by one group and from overlaps of religious and ethnic identities. Higher fractionalization and polarization, in contrast, reduce conflict risks. Aspects of religious practice are apparently irrelevant. In contrast, interreligious conflicts become much more likely when ethnic and regional—as well as economic—overlaps exist. These effects of identity structures are theoretically plausible, as interreligious conflicts are characterized by conflicts between identity groups; moreover, they are highly significant and very substantive. For instance, an interreligious conflict onset becomes around six times likelier when ethnic and religious—or religious and regional identities—run parallel to each other. Religious practice also plays a role, but these effects are less intense in terms of both significance and magnitude.

Variables of religious practice have the most significant and substantive effects for theological conflicts, corresponding to the ideational character of conflict. In particular, when religious leaders justify or call for violence the onset risk increases significantly, by more than four times. Grievances over discrimination play a somewhat lesser role, but are still significant at the 5 percent level—and more than double the onset risk. In contrast, demographic structures play almost no role. It is only the overlap of economic and religious boundaries that marginally affects the likelihood of a theological conflict onset. It makes sense that the overlap of religious and ethnic identities—otherwise highly influential—is insignificant in these cases. Many theological conflicts in, for instance, Algeria, Pakistan, or Somalia are fought between seculars and extremists, not between identity groups.

The differences between the types of conflict extend to the control variables (see Tables 3–5). While both population size (logged) and the time elapsed since the last conflict are significant for almost all models, other variables behave differently according to conflict type. It is particularly noteworthy that economic variables become less important the more religion is at stake. Oil only impacts conflict in general ways, but seemingly does not account for both types of religious conflict. Interestingly, economic growth is positively connected to the onset of interreligious conflict.²⁰ Theological conflicts are the least economy-driven ones: neither oil exports nor growth, nor even income per capita (gdp per capita), are significant. The only somewhat economic element is economic overlap, which is, however, only marginally significant and loses significance when controlled for religious discourse (see subsequently). Apparently, then, theological conflicts are not about the economy; they are about ideas.

In sum, our expectations have been largely confirmed by our empirical analysis: religious variables capturing identity overlaps and discourse (Hypotheses 2 and 3) that exceed the classical diversity measures (Hypothesis 1) used in most previous studies prove to be significant determinants of conflict onset. Religion accounts more for religious conflicts, but the determinants of it vary—and in a way that corresponds to the conflict type (Hypothesis 4). For interreligious conflicts, the identity dimension matters; for ideational conflicts, the discourse dimension is what counts.

Robustness Checks

In order to check for the robustness of our results, we conducted a number of additional estimations. To analyze whether our findings for religious discourse are not entirely driven by religious structures—in other words, religious diversity and identity overlaps—we also tested for the importance of identity overlaps and religious discourse simultaneously. Tables S5 through S7 in the Online Appendix show that our findings remain largely unaffected by this test—except for interreligious conflict, where grievances over discrimination turn insignificant, as per conventional levels, when controlling for overlaps. Additionally, in theological conflicts, economic overlaps lose conventional significance in a number of specifications when variables on religious discourse are included. This increases our confidence that theological conflicts are fought over ideas and not economic issues.

Following the suggestions put forward by King and Zeng (2001), we employed rare events logistic estimations. While armed conflict onset takes a value of one in about 5 percent of cases in our sample, interreligious conflict onset and theological conflict onset are only positive in 2 percent and 1.5 percent of the cases, respectively. King and Zeng (2001) argue that logit coefficients may be biased, especially in small samples. Even if our sample is relatively large, we calculate to what extent the logistic estimation downplays event probabilities by applying a logit model that corrects for rare events and small sample bias. A comparison of the amended logit estimates and our baseline findings from the logistic regression show that our results remain largely intact.²¹

We also tested two alternative dependent variables. First, we estimated the effects of our models on conflicts in which both identity and theological elements were present, as coded in RCDC. Altogether, we found twenty-two conflicts of this kind. The results show that, as can be expected, these full religious armed conflicts follow a mixed logic, including determinants from both interreligious and theological conflicts. Particularly, religious calls for violence and ethno-religious overlaps increase the risk that such conflicts will occur (see Table A9 in the Online Appendix).

As interreligious conflicts are heavily influenced by such ethno-religious overlaps, one might suspect that these conflicts are in fact ethnic conflicts. We thus performed another check, using data on Ethnic Armed Conflict (EAC; Wucherpfennig et al. 2012). However, interreligious and ethnic conflict onsets are not similar. Besides a number of classical controls, they only share ethno-religious overlap as a determinant (but not other overlaps). Ethnic conflicts are more than three times more likely to occur when such overlaps exist. The latter finding increases the confidence that ethnic and religious identity overlaps have great explanatory power. Additional religious differences deepen in- and out-group salience in a number of ways. Generally, this robustness check shows that interreligious conflicts are substantially different in origins from ethnic conflicts (see Table A10 in the Online Appendix).

Finally, we employed a number of alternative independent variables. We additionally created specific dominance measures for the two principal monotheist religious traditions—that is, for Christianity (dominance_christ) and Islam (dominance_islam). We also coded for the population shares of these two denominations.²² Findings on the dominance of Christians and Muslims, as well as respective population shares, are rather insignificant or not robust.²³ As such, no major monotheist religion alone drives either religious or other forms of conflict. Finally, we used data from Fox (2004) on the objective discrimination of religious minorities. Findings from it provide little evidence that objective discrimination increases the risk of conflict onset. For no type of conflict was objective discrimination found—including in our two additional robustness checks—to increase the risk of conflict onset. It would be premature to draw far-reaching conclusions from this, given the fact that only minorities are included. However, it might well be the case that actual grievance, not objective discrimination, counts (see the Online Appendix, Tables A11–A13, models 5 in each table).²⁴

Discussion

Before drawing conclusions, we must consider a number of caveats to our study. First, our sample is limited in terms of its scope. We have looked at a time span of just twenty-one years and only at developing countries. While this may create more similarities, and thus comparability, future studies may confirm or modify our findings for other samples. Second, our data are in the traditional country–year format. We do not have exact information on subnational units. Finally, other challenges to our study may stem from nonrandom patterns of missingness. For instance, some controls (such as rough terrain and polity) are systematically missing, mainly for small countries.

Despite these caveats, however, we believe that our contribution advances the study of the religion–conflict link in several key ways: first and foremost, the new data have proved useful and furthermore supports our initial idea that missing information on theoretically important variables has been one of the most glaring gaps in quantitative studies on religion and conflict thus far. Second, our contribution may further the theorizing of how religion and conflict relate to each other. Organized violence, in the form of religious or other armed conflict, is luckily a rare occurrence due to the especially steep hurdles faced in coordinating violent collective action.

Regarding our conceptualization of opportunity and motive, we have assigned religious structures—in the form of religious diversity or identity overlaps—to opportunity (as mobilization frames become available) and religious practice—in the form of grievances over discrimination or religious ideas—to motive. With this distinction in mind, we have a clearer notion of how religion impacts various forms of violent conflict onset. With regard to armed conflict, religion works through opportunity mechanisms alone. For theological conflicts, on the other hand, motivation channels dominate. In interreligious conflicts, meanwhile, it is ultimately rather through opportunity that religion works.

Opportunity and motive, though, are not totally unrelated and, for instance, structural givens such as dominant religious groups or economic differences between religious groups may also influence motive. We should, therefore, further investigate how these two aspects relate to each other.

From a structural point of view, the most interesting result is that overlapping religious and other identities form a particular conflict risk. This is not only theoretically highly plausible and aligned with other recent findings with a somewhat different focus (e.g., Gubler and Selway 2012) but also shows that religious identities, rather than being socially insulated, are actually embedded in other social relationships. When these identities run parallel, and possibly are amplified and dichotomized through economic and political inequalities, fertile ground for conflict emerges. The relationship between ethnicity and religion in particular deserves closer inspection, both from the angle of religion and of ethnicity (Stewart 2009). In-depth case studies may look closer at how religious and ethnic identities interact in a given context.

Finally, in light of the rather insignificant weight of the economy and structural identity overlaps for theological conflict onset, another conclusion can be drawn. The findings indicate that not all conflicts result from economic or other structural conditions. Sometimes it is evidently not about the economy. For theological conflict, we have identified aspect of practice, such as grievances over discrimination and calls for violence, as determinants of conflict onset—but, simultaneously, weak support for structural determinants. This finding suggests that religious and other ideas may have a role independent from structures. In sum, we find that not all conflicts are equal. It might well be the case that the study of armed conflict has thus far largely treated phenomena as equal that are, in reality, fundamentally different.

Closing Remarks

Anecdotal evidence from many armed conflicts around the world suggests that religion incites organized violence. From a theoretical perspective, several facets of religion can create opportunities and motives that help overcome collective action problems. However, research has hitherto found no conclusive answer on the extent to which religion is connected to armed conflict onset. Previous studies have left many important religious determinants—such as overlaps of religious and other identities or grievances and calls for violence—untested. Using a unique data set on 130 developing countries, this article investigated many of the theoretically important religious variables and has tested whether the secular or religious nature of conflicts matters.

Results from logistic regressions confirm our expectation that particular religious factors fuel religious armed conflict: particularly key in this regard are the overlap of religious and other identities as well as grievances over discrimination and religious calls for violence. We also find that religious determinants substantially differ according to whether conflicts are religious or not in origin. Interreligious conflicts are mainly driven by identity overlaps, while conflicts over theological issues rather result from an aggressive religious discourse. In general, economic issues count much less as a factor in religious conflicts, and particularly in theological conflicts. A number of additional checks performed using alternative estimation techniques, as well as dependent and independent variables, have increased our confidence that our results are robust.

We believe that our study substantially advances the study of conflict and its religious determinants, especially as RCDC has identified a number of important relationships therein. Our results also reveal that religion has many different dimensions and that it is deeply embedded in nonreligious contexts. In general, our findings suggest that many challenges and opportunities for future research on the link between religion and conflict remain.

Footnotes

Authors’ Note

The data set used in this article as well as the STATA Do-files to replicate the results are provided at the journal’s website ().

Acknowledgment

The authors would like to thank Alexander De Juan and Jan H. Pierskalla, the participants of a panel at the Midwest Political Science Association Annual Meeting in April 2012 for useful comments as well as two anonymous reviewers for useful comments and suggestions. The usual caveat applies.

Declaration of Conflicting Interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Research for this article was generously funded by the Federal Ministry of Development Co-Operation (project Nr. 2010 95 850) and the Research Council of Norway, project 217995/V10.

Notes

Supplemental Material

The online [appendices/data supplements/etc] are available at .

References

Akbaba

Yasemin

Taydas

Zeynep

. 2011. “Does Religious Discrimination Promote Dissent? A Quantitative Analysis.” Ethnopolitics 10 (3-4): 271–95.

Alesina

Alberto

Devleeschauwer

Arnaud

Easterly

William

Kurlat

Sergio

Wacziarg

Romain

. 2003. “Fractionalization.” Journal of Economic Growth 8 (2): 155–94.

Appleby

Scott

. 2000. The Ambivalence of the Sacred. Religion, Violence and Reconciliation. Lanham, MD: Rowman & Littlefield.

Basedau

Matthias

Strüver

Georg

Vüllers

Johannes

Wegenast

Tim C.

. 2011. “Do Religious Factors Impact Armed Conflict? Empirical Evidence from Sub-Saharan Africa.” Terrorism and Political Violence 23 (5): 752–79.

Beck

Nathaniel

Katz

Jonathan N.

Tucker

Richard

. 1998. “Taking Time Seriously: Time-series-cross-section Analysis with a Binary Dependent Variable.” American Journal of Political Science 42 (4): 1260–88.

Cameron

A. Colin

Trivedi

Pravin K.

. 2009. Microeconometrics using Stata. College Station, TX: Stata Press.

Canetti

Daphna

Hobfoll

Stevan E.

Pedahzur

Ami

Zaidise

Eran

. 2010. “Much Ado about Religiosity: Heightened Support for Political Violence by Failure to Conserve.” Journal of Peace Research 45 (5): 575–87.

Chandra

Kanchan

Wilkinson

Steven

. 2008. “Measuring the Effect of Ethnicity.” Comparative Political Studies 41 (4/5): 515–63.

Cederman

Lars-Erik

Weidmann

Nils B.

Gleditsch

Kristian Skrede

. 2011. “Horizontal Inequalities and Ethnonationalist Civil War: A Global Comparison.” American Political Science Review 105 (3): 478–95.

10.

Collier

Paul

Hoeffler

Anke

. 2004. “Greed and Grievance in Civil War.” Oxford Economic Papers 56 (4): 563–95.

11.

De Juan

Alexander

. 2009. “A Pact with the Devil? Elite Alliances as Bases of Violent Religious Conflicts.” Studies in Conflict and Terrorism 31 (12): 1120–35.

12.

De Soysa

Indra

Nordås

Ragnhild

. 2007. “Islam’s Bloody Innards? Religion and Political Terror, 1980-2000.” International Studies Quarterly 51 (4): 927–43.

13.

Dunning

Thad

Harrison

Lauren

. 2010. “Cross Cutting Cleavages and Ethnic Voting: An Experimental Study of Cousinage in Mali.” American Political Science Review 104 (1): 21–39.

14.

Ellingsen

Tanja

. 2000. “Colourful Community or Ethnic Witches’ Brew? Multiethnicity and Domestic Conflict During and After the Cold War.” Journal of Conflict Resolution 44 (2): 228–49.

15.

Fearon

James

Laitin

David

. 2003. “Ethnicity, Insurgency, and Civil War.” American Political Science Review 97 (1): 75–90.

16.

Fox

Jonathan

. 2004. Religion, Civilization, and Civil War. 1945 Through the New Millennium. Oxford, UK: Lexington.

17.

Fox

Jonathan

. 2011. “Building Composite Measures of Religion and State.” Interdisciplinary Journal of Research on Religion 7:1–39.

18.

Fox

Jonathan

James

Patrick

Yitan

. 2009. “State Religion and Discrimination Against Ethnic Minorities.” Nationalism and Ethnic Politics 15 (2): 189–210.

19.

Greene

William H.

2008. Econometric Analysis. 7th ed. Upper Saddle River, NJ: Prentice Hall.

20.

Grim

Brian

Finke

Roger

. 2011. The Price of Freedom Denied. New York: Cambridge University Press.

21.

Gubler

Joshua

Selway

Joel S.

. 2012. “Horizontal Inequality, Crosscutting Cleavages, and Civil War.” Journal of Conflict Resolution 56 (2): 206–32.

22.

Gurr

Ted

. 1970. Why Men Rebel. Princeton, NJ: Princeton University Press.

23.

Gurr

Ted

. 2000. Peoples versus States: Minorities at Risk in the New Century. Washington, DC: United States Institute of Peace Press.

24.

Hafez

Mohammed M.

2003. Why Muslims Rebel. Repression and Resistance in the Islamic World. Boulder, CO: Rienner.

25.

Harbom

Lotta

Wallensteen

Peter

. 2009. “Armed Conflict, 1946–2008.” Journal of Peace Research 46 (4): 577–87.

26.

Hassner

Ron

. 2011a. “Blasphemy and Violence.” International Studies Quarterly 55 (1): 23–45.

27.

Hassner

Ron

. 2011b. “Sacred Time and Conflict Initiation.” Security Studies 20 (4): 491–520.

28.

Hegre

Håvard

Sambanis

Nicholas

. 2006. “Sensitivity Analysis of Empirical Results on Civil War Onset.” Journal of Conflict Resolution 50 (4): 508–35.

29.

Horowitz

Donald

. 1985. Ethnic Groups in Conflict. Berkeley: University of California Press.

30.

Horowitz

Michael C.

2009. “Long Time Going: Religion and the Duration of Crusading.” International Security 34 (2): 162–93.

31.

Hsu

Becky

Reynolds

Amy

Hackett

Conrad

Gibbon

James

. 2008. “Estimating the Religious Composition of All Nations: An Empirical Assessment of the World Christian Database.” Journal of the Scientific Study of Religion 47 (4): 678–93.

32.

King

Gary

Zeng

Langche

. 2001. “Logistic Regression in Rare Events Data.” Political Analysis 9 (2): 137–63.

33.

Latham

Andrew A.

2011. “Theorizing the Crusades: Identity, Institutions, and Religious War in Medieval Latin Christendom.” International Studies Quarterly 55 (1): 223–43.

34.

Lichbach

Mark Irving

. 1998. The Rebel’s Dilemma. Ann Arbor: University of Michigan Press.

35.

Lipset

Seymour M.

Rokkan

Stein

. 1967. “Cleavage Structures, Party Systems, and Voter Alignments.” In Party Systems and Voter Alignments: Cross-National Perspectives, edited by Lipset

Seymour Martin

Rokkan

Stein

, 1–64. New York: The Free Press.

36.

Montalvo

José G.

Reynal-Querol

Marta

. 2005. “Ethnic Polarization, Potential Conflict and Civil War.” American Economic Review 95 (3): 796–816.

37.

Olson

Mancur

. 1965. The Logic of Collective Action. Public Goods and the Theory of Groups. Cambridge, MA: Harvard University Press.

38.

Østby

Gudrun

Nordås

Ragnhild

Rod

Jan Ketil

. 2009. “Regional Inequalities and Civil Conflict in Sub-Saharan Africa.” International Studies Quarterly 53 (2): 301–24.

39.

Pearce

Susanna

. 2005. “Religious Rage: A Quantitative Analysis of the Intensity of Religious Conflicts.” Terrorism and Political Violence 17 (3): 333–52.

40.

Reynal-Querol

Marta

. 2002. “Ethnicity, Political Systems, and Civil Wars.” Journal of Conflict Resolution 46 (1): 29–54.

41.

Rummel

Rudolph J.

1997. “Is Collective Violence Correlated with Social Pluralism?” Journal of Peace Research 34 (2): 163–75.

42.

Schneider

Gerald

Wiesehomeier

Nina

. 2008. “Rules that Matter: Political Institutions and the Diversity-Conflict Nexus.” Journal of Peace Research 45 (2): 183–203.

43.

Selway

Joel Sawat

. 2011a. “Cross-cuttingness, Cleavage Structures and Civil War Onset.” British Journal of Political Science 41 (1): 111–38.

44.

Selway

Joel Swat

. 2011b. “The Measurement of Cross-cutting Cleavages and Other Multidimensional Cleavage Structures.” Political Analysis 19 (1): 48–65.

45.

Seul

Jeffrey

. 1999. “Ours Is the Way of God. Religion, Identity, and Intergroup Conflict.” Journal of Peace Research 36 (5): 553–69.

46.

Stewart

Frances

. 2008. Horizontal Inequalities and Conflict: Understanding Group Conflict in Multiethnic Societies. Basingstoke, UK: Palgrave Macmillan.

47.

Stewart

Frances

. 2009. Religion Versus Ethnicity as a Source of Mobilisation. Are There Differences? MICROCON Research Working Paper 18, Institute of Development Studies at the University of Sussex, Brighton, UK.

48.

Svensson

Isak

. 2007. “Fighting with Faith: Religion and Conflict Resolution in Civil Wars.” Journal of Conflict Resolution 51 (6): 930–49.

49.

Taylor

Michael

Rae

Douglas

. 1969. “An Analysis of Crosscutting between Political Cleavages.” Comparative Politics 1 (4): 534–47.

50.

Toft

Monica Duffy

. 2003. The Geography of Ethnic Violence: Identity, Interests, and the Indivisibility of Territory. Princeton, NJ: Princeton University Press.

51.

Toft

Monica Duffy

. 2007. “Getting Religion? The Puzzling Case of Islam and Civil War.” International Security 31 (4): 97–131.

52.

Toft

Monica Duffy

Philpottt

Daniel

Shah

Timothy Samuel

. 2011. God’s Century: Resurgent Religion and Global Politics. New York: Norton & Company.

53.

Wucherpfennig

Julian

Metternich

Nils W.

Cederman

Lars-Erik

Gleditsch

Kristian S.

. 2012. “Ethnicity, the State, and the Duration of Civil War.” World Politics 64 (1): 79–115.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.27 MB

0.00 MB