Abstract
Why are religious conflicts more violent than non-religious conflicts? Research has argued that religion pushes partisans toward violence. However, existing research suffers from widespread problems of measurement validity and fails to confront the possibility of endogeneity in the relationship between religion and violence. This article develops a more precise measure of the relevance of religion to conflict based on the use of religious rhetoric by political organizations. With this approach in mind, this article disentangles the causal sequence linking religious rhetoric and violence using annually coded data on the rhetoric of 495 organizations worldwide from 1970 through 2012. The analysis finds a strong general correlation between religious rhetoric and violence. However, past use of religious rhetoric does not increase the likelihood that an organization will participate in violence or the overall intensity of conflict. On the contrary, previous participation in violence makes an organization more likely to adopt religious rhetoric for mobilization. Indeed, religious rhetoric becomes more likely as violence increases in intensity and conflict continues for longer periods of time. These findings suggest that violent actors adopt religious rhetoric to solve the logistical challenges associated with violence, including access to mobilizing resources and recruitment and retention of members. This article contributes to the study of religious conflict by providing evidence of endogeneity in the relationship between religion and violence and highlighting the need for temporally sensitive measures of religious mobilization.
The past half-century has seen a significant increase in the public presence of religion worldwide. Especially critical has been the dramatic rise in the use of religious rhetoric to promote violence. Between the 1960s and 1990s, the percentage of civil wars involving religion increased from 21% to 43% (Toft, 2007). In the Middle East and North Africa, the proportion of armed conflicts involving religious claims increased from 0% in 1975 to 75% in 2011 (Svensson, 2013).
Research has suggested that conflict involving religion is more violent and longer lasting than other forms of conflict (Henne, 2012a; Horowitz, 2009; Pearce, 2005; Piazza, 2009; Toft, 2007). Religious conflicts are more likely to concern issues seen as indivisible, less likely to be resolved through negotiated settlement, and more likely to relapse once a settlement has been reached (Hassner, 2009; Svensson, 2007, 2012; Walter, 2011). These challenges can be seen worldwide, from enduring religious hostilities in Afghanistan and Israel to the rise of religious violence in Burma, Iraq, and Nigeria.
Though considerable research has highlighted the relationship between religion and violence, little scholarship has examined the causal sequence linking the two (Grzymala-Busse, 2012). While some scholars maintain that religious beliefs encourage violence, there is also reason to believe that violence encourages religion. Specifically, religion may help solve the logistical challenges facing violent organizations, including resource shortages and recruitment problems. To overcome this ambiguity, this article develops a temporally sensitive measure of the relevance of religion to conflict and systematically examines temporal sequencing in the relationship between religion and violence.
The article proceeds as follows. First, I discuss two weaknesses in the literature on religious conflict: poor measurement validity and a failure to confront endogeneity in the relationship between religion and violence. Second, I present a rhetoric-based measure of the relevance of religion designed to provide a precise proxy for the public mobilization of religion in conflict. With this measure in mind, I examine the temporal sequencing of religious rhetoric relative to participation in violence and intensity of violence.
Analysis confirms that religion and violence are broadly correlated. However, I find no evidence that prior religious rhetoric encourages organizations to participate in violence or to increase the intensity of violent tactics. On the contrary, findings support the notion that violent actors adopt religious rhetoric to solve the logistical challenges associated with violence, including access to mobilizing resources and recruitment and retention of members. I conclude by noting the limitations of this analysis and the implications of these findings for the study of religious conflict.
Weaknesses in the literature
This study responds to two weaknesses in the literature on religious conflict. The first weakness concerns the ways in which scholars measure the relevance of religion to conflict. The second weakness concerns the possibility of endogeneity in the relationship between religion and violence.
Measurement validity
Existing literature has reached few clearcut conclusions regarding the relationship between religion and violence. On one hand, research has suggested that conflict involving religion is more violent and longer lasting than other forms of conflict (Henne, 2012a; Horowitz, 2009; Pearce, 2005; Roeder, 2003; Toft, 2007). Henderson (1997) finds that religious differences increase the likelihood of interstate conflict, while Henne (2012b) finds that religion tends to intensify interstate conflicts caused by other factors.
On the other hand, some research has found that religious factors do not exert an independent effect on conflict (Ellingsen, 2000). Piazza (2009) argues that the high casualty rates of terrorist groups can be explained by organizational factors rather than religious motivations. At the interstate level, considerable scholarship has failed to find evidence that religious differences increase the likelihood of militarized disputes (Chiozza, 2002; Henderson & Tucker, 2001; Tusicisny, 2004). 1
A growing body of scholarship has also offered more complex arguments, including that religion can encourage violence through interaction with the state (Grim and Finke, 2007, 2011; Henne, 2012b; Philpott, 2007), religious institutions can provide resources to facilitate violence (Fox, 1999; Smith, 1996), or certain religious traditions or ideologies are more prone to conflict than others (Fox, 2000; Philpott, 2007).
The diversity of these findings originates in part from a lack of agreement over how to measure the relevance of religion in conflict. Studies of religious conflict generally adopt one of two methods to measure the relevance of religion: issue-oriented or identity-oriented (Pearce, 2005). While the relatively restrictive issue-oriented approach overlooks a number of ways in which religion can influence conflict, the more permissive identity-oriented approach includes many cases in which religion has little influence on conflict.
First, the issue-oriented approach defines conflict as religious if the issues at stake between conflicting groups are religious in nature. Scholars adopting this perspective often posit that religious conflict is especially prone to violence because it concerns issues that do not lend themselves to compromise. Hassner (2009) emphasizes the conflictual nature of holy sites, while Svensson (2007, 2012) finds that conflicts involving religious incompatibilities are less likely than other conflicts to reach a negotiated settlement. One strength of this approach is its focus on the mechanisms linking religion with violence. Several scholars posit that the salience of religious issues tends to link conflict goals with the beliefs of participants, increasing the value of success and justifying the use of violence (Almond, Appleby & Sivan, 2003; Horowitz, 2009; Juergensmeyer, 2003; Rapoport, 1984).
However, the issue-oriented approach limits religious conflicts to those in which major actors express religious demands, most often control over a holy site or the establishment of a state governed by religious law. Although this measure undoubtedly includes conflicts in which religion is significant, it may omit many conflicts in which religion has a substantial influence. Sells (1998), for instance, points to the infusion of religious mythology into the Serbian–Bosnian conflict through articulation of the story of Christ as a metaphor to justify anti-Muslim violence. Although religion undoubtedly influenced this conflict, the issue-oriented approach would classify this as a non-religious conflict due to a lack of religious demands or issues.
There is reason to believe that this selection method biases estimations of the effect of religion on conflict. Because issue-based religious mobilization is often more overt than the use of religious metaphor or symbolism, highly salient religious issues may have a larger effect on violence. By selecting only these cases, the issue-oriented approach offers a most-likely test of the effect of religion on conflict. A more liberal measure of the relevance of religion would allow for a more nuanced analysis of the relationship between religion and violence.
On the other end of the spectrum, the more permissive identity-oriented perspective defines conflict as religious when it occurs across religious boundaries. This approach emphasizes the demographic relationship between conflicting groups over the salience of religious issues (Nordås, 2010; Pearce, 2005). Fox (2002), for instance, includes in the concept of ‘ethnoreligious conflict’ any conflict in which opposing groups have different religious affiliations. Many studies adopting this approach have found that religious conflict is more violent than conflict between groups of the same religion (Fox, 2002; Henderson, 1997; Roeder, 2003).
By definition, this approach characterizes conflict as religious whenever conflicting groups differ in religious affiliation. In practice, however, it is often counter-intuitive to classify conflict across religious boundaries as religious in nature. Though Greek Cypriots are generally Christian and Turkish Cypriots are generally Muslim, few observers would identify the Cyprus dispute as a case of religious conflict (Fisher, 2001).
The identity-oriented approach also precludes any measure of religion that varies across groups or over time while demography remains constant. This limitation is clear when examining the Sri Lankan civil war. Though Sri Lanka’s Sinhalese majority is largely Buddhist and the island’s Tamil minority is predominantly Hindu, religion only recently gained salience among Sinhalese political leaders. Tamil leaders, on the other hand, have rarely drawn attention to the religious dimension of communal difference (Little, 1994). This within-conflict variation is difficult to capture with a demographic measure.
In sum, these measures offer a poor foundation for assessing the relevance of religion to conflict or developing informed insights into the relationship between religion and violence. While the issue-oriented approach classifies some religiously influenced conflicts as negative cases, the identity-oriented approach classifies some non-religious conflicts as positive cases. If the relevance of religion contributes to conflict intensity, a more fine-grained measure of religious relevance is required to develop practical policy aimed at mitigating religious violence.
Endogeneity
Research on religious conflict has also frequently overlooked the possibility of endogeneity in the relationship between religion and violence. Endogeneity describes a situation in which two variables cause each other (Grzymala-Busse, 2011; Pierson, 2004). Although considerable scholarship has examined how religion affects violence, few scholars have systematically considered the possibility that violence affects religion.
However, scholarship offers several reasons to expect a high degree of reverse causation in this relationship. Hoffman & McCormick (2004) and Lichbach (1995), among others, note that violent political organizations are in a precarious position relative to the state. By choosing to engage in violence, organizations accept an anti-state stance that undercuts access to resources and increases the costs of individual participation. As a result, violent organizations tend to face difficulty securing material resources and recruiting members (Hegghammer, 2013).
A separate body of research has pointed to the mobilizing resources associated with religion, including finances, communication channels, and leadership (Smith, 1996; Stark & Bainbridge, 1980). Iannaccone (1994) argues that religious organizations are able to solve recruitment problems by penalizing activities that compete for the resources of members. Recent scholarship has suggested that religion improves the effectiveness of violence by providing access to social networks, delivering spiritual goods to participants, and undercutting incentives to defect (Basedau, Pfeiffer & Vüllers, 2016; Berman, 2011; Gates & Nordås, 2014; Hassner, 2011). If religious rhetoric helps solve resource and recruitment problems, it follows that violent political organizations may have strategic incentives to adopt religion as a basis for mobilization.
Examples of this type of strategic religious mobilization can be found worldwide. Bakke (2013) argues that the proliferation of Islamic rhetoric in the Second Chechen War was an attempt to encourage the flow of resources from the Muslim world. Similarly, Toft (2007) argues that embattled Sudanese political leaders adopted religious rhetoric to gain political legitimacy and external support. Ganiel & Dixon (2008) point to the pragmatism of Protestant fundamentalist rhetoric in the conflict in Northern Ireland, while Sugunnasil (2006) notes that militants adopted religious rhetoric in the South Thailand insurgency to access the communication networks embedded in religious schools.
In the long term, the strategic adoption of religious rhetoric may result in a strong general correlation between religion and violence even if religion does not exert an independent effect on violence. By failing to distinguish instances in which religion contributes to violence from those in which violence contributes to religion, scholarship that does not control for this reverse causation may overestimate the independent effect of religion on violence.
This problem is closely related to the question of measurement validity. Scholarship adopting the issue-oriented perspective has tended to use non-temporal measures of religion. Svensson (2007), for instance, measures religious issue incompatibility once per conflict-dyad. Piazza (2009) codes organizations as ‘Islamist’ once per organization. As described above, the identity-oriented approach precludes the possibility of examining temporal sequencing by focusing on demographic measures that vary little over time. To offer a nuanced assessment of the relationship between religion and violence, scholars require a temporally sensitive measure of religious mobilization capable of distinguishing between a sequence in which prior religious mobilization contributes to subsequent violence and one in which prior violence introduces strategic incentives for religious mobilization.
Conceptualizing religious conflict
To overcome these weaknesses, I measure the relevance of religion to conflict based on the explicit use of religious rhetoric by organizations involved in conflict. This approach draws on a small body of literature arguing that the effect of religion on conflict is mediated through the active rhetoric of conflict actors (Nordås, 2010).
To overcome the challenges of previous research, I focus on political organizations as the unit of analysis and measure religious rhetoric on an annual basis. According to this approach, religion is highly relevant to conflict when a group’s political representatives employ the language of religion to encourage collective mobilization. By contrast, religion has low relevance when a group’s political representatives use other forms of rhetoric to encourage mobilization.
This rhetoric-based measure is both more empirically reliable and more theoretically precise than the measures adopted in previous research. This measure captures greater variation in the relevance of religion than the issue-oriented approach by including all public references to religion, including symbolic religious rhetoric as well as explicit religious goals and demands. This measure is more restrictive than the identity-oriented approach by requiring active appeals for religious mobilization. Furthermore, this measure offers more theoretical precision by capturing variation across organizations as well as variation over time.
In assessing the relevance of religion to conflict between Catholics and Protestants in Northern Ireland, for instance, the issue-based approach would point to a lack of religious demands by major actors as evidence that this is not a case of religious conflict. The identity-based approach, on the other hand, would classify the conflict as religious due to the religious boundary between conflicting groups. By contrast, the rhetoric-based approach would point to the growth of religious rhetoric among Protestant organizations in the 1960s relative to the secular nationalist rhetoric of Catholic organizations throughout the conflict (Ganiel & Dixon, 2008).
With this approach in mind, the first question asked in this analysis is: does past religious rhetoric have a significant effect on the likelihood that an organization will participate in violence? If framing conflict in the language and symbolism of religion has an independent effect on the use of violence, it follows that organizations should be more likely to participate in violence after using religious rhetoric. I consider the following hypothesis:
Hypothesis 1: Organizations that have used religious rhetoric in the past are more likely to participate in violence than those that have not used religious rhetoric.
Alternatively, I consider the possibility that organizations already engaged in violence may adopt religious rhetoric for strategic reasons. As described above, violent organizations tend to face resource shortages and difficulty in recruitment and retention of members. A growing body of literature suggests that religion provides militants with powerful mobilizing resources while undercutting incentives to defect. For these reasons, violent organizations may have strategic motivations to adopt religious rhetoric. If the logistical challenges associated with violence encourage organizations to frame conflict in religious terms, it follows that organizations should be more likely to adopt religious rhetoric after participating in violence. I test the following hypothesis:
Hypothesis 2: Organizations that have participated in violence in the past are more likely to adopt religious rhetoric than those that have not participated in violence.
These hypotheses both outline processes that cumulatively predict a correlation between religion and violence. However, each hypothesis offers distinct observable implications regarding the temporal sequencing of this relationship. While Hypothesis 1 posits that religious rhetoric should come prior to violence, Hypothesis 2 posits that violence should come prior to religious rhetoric. These hypotheses are not mutually exclusive. It may be the case that religious rhetoric causes violence and violence causes religious rhetoric.
Each hypothesis also has distinct observable implications beyond this temporal relationship. If religious rhetoric causes violence, as Hypothesis 1 suggests, we should also expect to see a relationship between past use of religious rhetoric and current intensity of violence. Alternatively, if violence encourages religious rhetoric, as posited by Hypothesis 2, we should also expect organizations to adopt religious rhetoric when they are involved in more intense violence or longer lasting conflict.
Research design
I assess these hypotheses by analyzing the behavior of organizations representing religiously distinct groups on an annual basis from 1970 through 2012. Below I describe rules regarding the universe of cases, operationalization of variables, and statistical methods used in this analysis.
Universe of cases
The purpose of this analysis is to identify as closely as possible the temporal sequence linking religious rhetoric and violence. One challenge in assessing temporal sequencing is identifying a universe of cases in which all variables of interest exhibit variation. I limit the universe of cases to political organizations for which both religious rhetoric and violence are possible but not predetermined (Mahoney & Goertz, 2004). I accomplish this task in two steps.
First, I include only organizations representing groups that share a religious identity and another type of ethnic commonality (language, race, tribe, or nation), both of which distinguish the group from the remainder of the state. By focusing on overlapping ethnic cleavages, I omit all cases in which religious rhetoric is favored for demographic reasons. For instance, political leaders may be predisposed to use religious rhetoric when they represent a group defined only by religious commonality (Posner, 2004). I include any organization making some form of ethnic rhetoric central to its political platform (Chandra, 2011).
Second, I include all organizations engaged in political activity, including political parties, protest organizations, and paramilitaries. This focus allows me to identify all organizations that have the possibility of participating in violence without selecting only those already engaged in violence. I identify organizations based on inclusion in annual editions of the Europa World Year Book (Europa Productions, 2012) and the Political Handbook of the World (Banks et al., 2014), as well as information in the Global Terrorism Database (National Consortium for the Study of Terrorism and Responses to Terrorism, 2012), the Minorities at Risk Organizational Behavior dataset (Asal, Pate & Wilkenfeld, 2008; Minorities at Risk Project, 2009), and the Pro-Government Militias Database (Carey, Mitchell & Lowe, 2013). 2
This limited universe of cases entails a trade-off between precision and generalizability. By holding demographic factors constant and including organizations of all types, this universe offers an ideal foundation for assessing the independent effect of religious rhetoric on violence and vice versa. However, this universe omits many organizations employing religious rhetoric in the absence of an overlapping ethnic cleavage. Because demography alone may encourage such organizations to use religious rhetoric, the temporal sequencing of religious rhetoric and violence may differ beyond the narrow universe of cases examined. Nonetheless, there is no reason to expect that findings regarding the strategic adoption of religious rhetoric are not generalizable beyond this limited universe of cases.
Variables of interest
The hypotheses presented above posit distinct relationships between two primary concepts: religious rhetoric and violence. Religious rhetoric entails explicit and public reference to the religious identity or beliefs of a group. To capture this concept, I code Religious rhetoric, a binary variable measuring whether or not an organization claims to represent religious interests, advocates religious policies, or otherwise publicly references the religious identity of its constituents each year. This variable does not distinguish between types of religious rhetoric, such as rhetoric focused on theology or collective religious identity. As described above, this measure is more inclusive than issue-based measures of religious mobilization and more restrictive than identity-based measures.
This information is drawn from annual editions of the Europa World Year Book and Political Handbook of the World from 1970 through 2012. For each source, organizations are coded on an annual basis as using religious rhetoric if they (a) include reference to the religious identity of their group in their name, or (b) were identified as making explicitly religious rhetorical appeals by one or both sources. 3
I test the reliability of this measure against two alternative sources of information on religious mobilization. Comparison of this measure with a more restrictive measure of ‘religious organizational ideology’ drawn from the Minorities at Risk Organizational Behavior dataset (Asal, Pate & Wilkenfeld, 2008) yields a correlation coefficient of 0.74. Comparison of this measure with a more permissive coding of ‘religious organizational membership’ drawn from the Pro-Government Militias Database (Carey, Mitchell & Lowe, 2013) yields a correlation coefficient of 0.97. 4
I capture violence using three measures: Participation in violence, Intensity of violence, and Duration of conflict. Participation in violence measures whether or not an organization intentionally uses lethal tactics, regardless of the target. I code this variable based on information contained in annual editions of the Europa World Year Book and Political Handbook of the World. I supplement these sources with information on terrorist events from the Global Terrorism Database and data from the Mapping Militants Project (Crenshaw, 2010). I code this binary variable yearly as 1 if any of these sources indicates that the organization participates in violence and 0 if there is no indication of violent activity. 5
Importantly, this variable does not distinguish between levels of violence. To overcome this shortcoming, I include Intensity of violence, drawn from the Armed Conflict Dataset (Gleditsch et al., 2002; Pettersson & Wallensteen, 2015). For all violent organizations, this binary variable is coded on an annual basis as 0 if armed conflict involving the organization results in between 25 and 999 battle-related deaths in a year and 1 if the organization is involved in armed conflict resulting in over 1,000 battle-related deaths in a year.
Finally, I also measure Duration of conflict. This variable is operationalized as the number of years since the beginning of armed conflict involving an organization, ranging from 0 to 37. This variable is coded based on information in the Armed Conflict Dataset (Gleditsch et al., 2002; Pettersson & Wallensteen, 2015).
Additional variables
I also include several additional variables that have been shown to contribute to political violence. Research has demonstrated that exclusion from political institutions increases the likelihood of political violence (Birnir, 2007; Cederman, Wimmer & Min, 2010; Gurr, 1993). This is particularly true for religious groups (de Soysa & Nordås, 2007; Nordås, 2014). I split this variable into three levels of analysis: the state, the group, and the organization. At the state level, I measure the characteristics of a state’s political system as a proxy for political exclusion. I code Regime based on the Polity IV Project ‘combined polity score’. This variable measures regime type on a scale from –10 (strongly autocratic) to +10 (strongly democratic) (Marshall, Gurr & Jaggers, 2013). 6
Within a state, politically excluded groups may be more likely to participate in violence. I incorporate a measure of Group status drawn from the Ethnic Power Relations dataset (Wimmer, Cederman & Min, 2009). This variable is measured on an ordinal scale between 0 and 7 with higher values indicating greater political inclusion.
At the organizational level, legal status may have a strong influence on participation in violence. I incorporate a measure of Legality, a binary variable coded as 1 if an organization is formally legal in the year of observation and 0 if not. This variable is coded based on information in annual editions of the Europa World Year Book and the Political Handbook of the World.
Beyond measures of political exclusion, Fearon & Laitin (2003) and Gurr (1993) identify a negative correlation between GDP per capita and political violence. I include an annual measure of GDP per capita, logged and measured in 2012 US dollars (World Bank, 2013). Cederman, Wimmer & Min (2010), Fearon & Laitin (2003), and Hegre & Sambanis (2006) find that more populous states are more likely to experience political violence. To account for this, I include the state’s Population, logged and measured on an annual basis (World Bank, 2013). Finally, Ellingsen (2000) suggests that violence may be influenced by a group’s size relative to the rest of the state. I include Group proportion, measured yearly as a decimal value capturing the proportion of the state made up by the group (Central Intelligence Agency, 2012).
In robustness checks, I consider a number of additional variables that may influence violence. Scholars have argued that the effect of religion on conflict is influenced by connections between religion and the state (Grim & Finke, 2007, 2011; Henne, 2012b; Philpott, 2007). I include Government involvement in religion, an ordinal variable measured on an annual basis ranging from 0 to 14 with higher values corresponding to closer ties between religion and the state. This variable is drawn from the Religion and State Dataset (Fox, 2008, 2011). I also consider the effect of Ethnic homeland, a binary variable measuring whether or not the group occupies a territory it considers its homeland (Toft, 2005). Finally, scholars have suggested that organizational competition can encourage violence (Bloom, 2005; Cunningham, Bakke & Seymour, 2012). I consider a measure of Fragmentation, operationalized as the number of organizations representing a group in a given year and ranging from 1 to 15. Summary statistics for all variables are provided in the Online appendix.
Methods
For each hypothesis I conduct a logistic regression clustering standard errors by organization. Importantly, panel data such as those used in this article tend to violate the temporal independence assumption of logistic regression models (Beck & Katz, 2011). I correct for the effect of each organization by including a measure of the number of years since the last positive value on the dependent variable by organization (Beck, Katz & Tucker, 1998). This method controls for the lagged value of the dependent variable, allowing the effect of the independent variable to be isolated from any omitted long-term trends.
I assess the explanatory power of each hypothesis using a lagged measure of the relevant independent variable. For Hypothesis 1, I include a measure of an organization’s recent use of religious rhetoric. For Hypothesis 2, I include a measure of an organization’s recent participation in violence and intensity of violence. I conduct this analysis using five-year lagged measures for each independent variable.
This analysis serves as a strong test of the hypotheses outlined above. In order for either hypothesis to receive support, the effect of the independent variable must remain statistically significant while controlling for demographic difference between groups, the control variables listed above, and the past behavior of the organization.
I conduct numerous robustness checks to reinforce the strength of this analysis. I replicate all analyses using a fixed effects regression with a lagged dependent variable and clustering standard errors by organization and a generalized estimation equation with robust standard errors. I also replicate these analyses using one-year lags instead of five-year lags of independent variables. Further robustness checks include the additional variables described above (Government involvement in religion, Ethnic homeland, and Fragmentation). Another set of models isolates the behavior of organizations representing Muslim groups. In assessing Hypothesis 1, I also substitute Participation in violence with a measure of violence drawn from Cunningham, Bakke & Seymour (2012) and a measure of Support for violence. In assessing Hypothesis 2, I substitute Participation in violence with a measure of Cumulative violence (Gleditsch et al., 2002), a measure of organizational Age, and Fragmentation. The Online appendix contains a full description of these variables and findings from these robustness checks.
Data
This article examines the behavior of 495 organizations from 1970 through 2012. Of these 495 organizations, 160 have used religious rhetoric and 103 have participated in violence at some point during the period examined. Figure 1 provides an overview of participation in violence and religious rhetoric across these organizations over time. Participation in violence shows a general increase beginning in the 1970s and peaking in the early 1990s. This trend is consistent with data on civil wars worldwide (Fearon & Laitin, 2003). Consistent with previous research, religious rhetoric also displays a positive trend over time (Nordås, 2010; Svensson, 2013; Toft, 2007).
Participation in violence and religious rhetoric per year, 1970–2012
Alongside these visual trends, I estimate the correlation between religious rhetoric and violence. Table I presents the bivariate correlation between the organizations in the data that have used religious rhetoric and those that have participated in violence. Approximately one-third (33.8%) of organizations that have used religious rhetoric have also participated in violence, while just 14.6% of organizations that did not use religious rhetoric participated in violence. Taking only these factors into account, religious rhetoric has a statistically significant positive correlation with participation in violence.
Correlation between religious rhetoric and participation in violence
Parentheses show percentage of row. Pearson x 2 = 24.031, p = 0.000.
However, there are several reasons to discount the utility of this correlation. First, a bivariate correlation overlooks the effect of other variables known to influence violence. Without considering these variables, we have no way of knowing whether this correlation is caused by religious rhetoric or by the coincidence of religious rhetoric with other determinants of violence. Second, this correlation overlooks the passage of time, a crucial factor distinguishing the hypotheses presented above. Although Table I provides evidence that religious rhetoric is often associated with violence, it does not indicate whether religious rhetoric generally precedes violence or vice versa. This correlation thus serves as a hoop test for both hypotheses: neither is confirmed but neither can be thrown out.
Does religious rhetoric increase participation in violence?
Hypothesis 1 posits that organizations that have used religious rhetoric are more likely to subsequently participate in violence. To assess this hypothesis, I use several logistic regressions to estimate the effect of an organization’s average use of religious rhetoric over the five-year period prior to each observation (Religious rhetoric t–5) on that organization’s current likelihood of participating in violence. This analysis includes a measure of the years since an organization has participated in violence. By including this lagged dependent variable, this analysis controls for omitted time trends such as the general increase in Participation in violence displayed in Figure 1. Table II shows the results of this analysis.
The effect of religious rhetoric on participation in violence
† p < 0.10; *p < 0.05, **p < 0.01. Standard errors in parentheses are clustered by organization. All models include cubic splines for duration spells.
Table II provides little evidence for Hypothesis 1. As expected, the length of time since an organization has participated in violence is the strongest predictor of current Participation in violence across all four models. Group proportion appears to hold a negative relationship with violence, suggesting that groups forming a larger share of the state are less likely to participate in violence, although this relationship is not significant across all models.
Models 2 through 4 in Table II incorporate each of the three measures of political exclusion described above: Regime, Group status, and Legality, respectively. As expected, the negative coefficient on Regime in Model 2 indicates that greater levels of democracy are negatively correlated with participation in violence. Model 3 replaces Regime with Group status, which does not appear to have a significant effect on participation in violence. Finally, Model 3 indicates that Legality has a negative effect on participation in violence. These findings reinforce the notion that political exclusion makes an organization more likely to participate in violence.
Most importantly, this analysis consistently fails to provide evidence for Hypothesis 1. The correlation between Religious rhetoric and Participation in violence identified in Table I disappears when including control variables and considering variation over time. Across all four models, religious rhetoric in the last five years has no significant effect on an organization’s current likelihood of participating in violence. Organizations that have used religious rhetoric in the recent past are no more likely to participate in violence than organizations that have not.
These findings suggest that religious rhetoric does not independently contribute to an organization’s decision to participate in violence. Instead, participation in violence is more heavily influenced by an organization’s degree of political exclusion. These findings are confirmed in numerous robustness checks, including alternative statistical methods, alternative measures of key variables, analysis focusing on Muslim groups, and inclusion of additional control variables. 7
Does religious rhetoric increase intensity of violence?
The above analysis indicates that past religious rhetoric does not encourage participation in violence. But what if religion has more complex effects on violence? Previous research has suggested that religion may influence the intensity of violence rather than the onset of violence (Henne, 2012b). Even if religious rhetoric does not push organizations to adopt violence, religion may contribute to the intensity of the tactics used by violent organizations. I test this possibility by estimating the relationship between lagged Religious rhetoric and Intensity of violence. To isolate the causes of the intensity of violence from the causes of participation in violence, I limit this analysis to violent organizations.
As described above, participation in violence introduces a number of logistical challenges to organizations. Once an organization has adopted violence, however, it is likely to develop norms favoring the continued used of violence (Horowitz, 2010). As a result, the costs associated with increasing the intensity of violence may diminish. We may therefore expect increases in the intensity of violence to have a lower causal threshold than initial participation in violence, other factors being equal. The effect of religious rhetoric on intensity of violence may consequently be stronger than its effect on participation in violence. In this sense, Intensity of violence offers a lower standard for testing Hypothesis 1. Table III displays the findings of this analysis.
The effect of religious rhetoric on intensity of violence
† p < 0.10; *p < 0.05, **p < 0.01. Standard errors in parentheses are clustered by organization. All models include cubic splines for duration spells.
Table III offers little support for Hypothesis 1. Once again, the most consistent predictor of Intensity of violence is the number of years since an organization experienced high intensity violence. Interestingly, Models 2 and 4 indicate that Regime and Legality, respectively, have a positive effect on Intensity of violence. These findings may be explained by the fact that violent organizations are more likely to be excluded from political institutions and thus less likely to benefit from greater political opportunity. Alternatively, legal organizations that participate in violence may do so with the support of the state, thus encouraging higher intensity violence (Mitchell, Carey & Butler, 2014).
Most importantly, an organization’s recent use of religious rhetoric consistently fails to demonstrate a positive effect on Intensity of violence. Model II suggests that lagged Religious rhetoric may even have a negative effect on Intensity of violence, though this relationship does not extend to other models. As above, these findings are confirmed in numerous robustness checks. 8
Together with the findings from Table II, this analysis provides reason to doubt the utility of Hypothesis 1. In addition to showing no relationship with participation in violence, past use of religious rhetoric does not appear to increase the intensity of armed conflict involving the organization. Given that escalation of violence is less costly than initial participation in violence, the fact that religious rhetoric continues to display no effect suggests that the causal logic posited by Hypothesis 1 does not explain the relationship between these concepts.
Does violence increase religious rhetoric?
Although this analysis provides reason to doubt existing explanations for the relationship between religion and violence, it offers few clues as to why religious rhetoric and violence are so strongly correlated. Hypothesis 2 posits that past violence increases the likelihood that an organization will adopt religious rhetoric. This hypothesis relies on the notion that violence is a costly strategy requiring material resources and encouraging defection among rank and file. Religious rhetoric may attend to these problems by granting access to mobilizing resources and increasing the effectiveness of member recruitment and retention.
If Hypothesis 2 is accurate, it follows that other factors associated with these logistical challenges should also encourage religious rhetoric, including intensity of violence and duration of conflict. As violence increases in intensity and conflict endures over time, organizations are increasingly likely to face resource shortages and widespread problems of defection. Hypothesis 2 implies that these factors should also be associated with an increase in the adoption of religious rhetoric.
To test these implications of Hypotheses 2, I employ another set of logistic regressions examining the extent to which past violence affects an organization’s current likelihood of using religious rhetoric. The dependent variable of this analysis is Religious rhetoric and the independent variables are lagged Participation in violence, lagged Intensity of violence, and overall Duration of conflict. As above, this analysis incorporates a measure of the number of years since an organization has had a positive value on the dependent variable. Table IV displays the findings of this analysis.
The effect of violence on religious rhetoric
† p < 0.10; *p < 0.05, **p < 0.01. Standard errors in parentheses are clustered by organization. All models include cubic splines for duration spells.
Table IV provides strong support for Hypothesis 2. All three measures of violence significantly increase the likelihood that an organization will use religious rhetoric. The positive coefficient on Participation in violence in Model 1 indicates that organizations that have recently participated in violence have a significantly greater likelihood of adopting religious rhetoric than those that have not, other factors being equal.
Model 2 substitutes participation with recent Intensity of violence. As expected, organizations that have recently experienced higher intensity violence are also more likely to adopt religious rhetoric. Finally, Model 3 indicates that Duration of conflict is positively associated with religious rhetoric, suggesting that religious rhetoric becomes more likely as conflict endures over time. As above, inclusion of the lagged dependent variable in this analysis controls for any omitted time trends, including the general increase in Religious rhetoric displayed in Figure 1. Measures of past violence remain statistically significant in ten of 13 robustness checks. 9
Logistic regression results are often difficult to interpret, especially when examining panel data. For this reason, I simulate the predicted probability of using religious rhetoric as a function of lagged Participation in violence. 10 Figure 2 presents a visual summary of these estimates.
Predicted probability of religious rhetoric by participation in violence t –5
As Figure 2 makes clear, the probability that an organization will use religious rhetoric increases significantly with that organization’s past Participation in violence. An organization that has not participated in violence in the previous five years has only a 7.5% chance of using religious rhetoric. However, an organization that has participated in violence in all of the five previous years has a 19% chance of using religious rhetoric.
These findings support the notion that religious rhetoric is encouraged by the logistical challenges associated with violence. Organizations are encouraged to participate in violence for a variety of reasons, including political exclusion. After participating in violence, organizations may face difficulty securing resources and recruiting and retaining members. Religious rhetoric offers a potential solution to these challenges by granting access to mobilizing resources and discouraging defection.
Though these findings cast doubt on previous research, it is important to understand the limitations of this analysis. First, these findings do not provide evidence against the notion that religion causes violence. As discussed above, this analysis amounts to a strict test of the independent effect of religious rhetoric on violence. It does not follow that religion is never a cause of violence or that existing findings on the subject are invalid. Instead, these findings provide nuance to previous research by suggesting that the relationship between religious rhetoric and violence is characterized by a high degree of reverse causation.
Second, this analysis is limited to the behavior of organizations representing groups that differ on both religious and ethnic grounds. As described above, this limitation provides the ideal conditions for examining variation in both religious rhetoric and violence. This analysis does not consider the behavior of groups distinguished solely on the basis of religion. Because religion is the only point of differentiation for such groups, religious rhetoric may be encouraged by the need to appeal to a group’s common identity. As a result, religious rhetoric may be more likely to precede violence for such groups.
However, there is no reason to believe that the strategic benefits of religious rhetoric should be limited to the organizations examined in this article. Mobilizing resources and recruitment pathways are a pervasive feature of religion (Smith, 1996). For this reason, violent organizations may have strategic reasons to adopt religious rhetoric even beyond the specific universe of cases examined in this article.
Finally, this analysis focuses on rhetoric as an indicator of the relevance of religion to conflict. Although this measure is more precise than other measures, it is not a comprehensive proxy for the complex role of religion in conflict. Different relationships between religion and violence may hold when examining religious demography or religious goals. Nonetheless, this analysis strongly encourages a re-evaluation of previous research in light of the likelihood of endogeneity in the relationship between religion and violence.
Conclusion
Using data on 495 organizations representing religiously distinct groups from 1970 through 2012, this article finds little evidence that past religious rhetoric encourages subsequent violence. Even among violent organizations, analysis finds little indication that religious rhetoric increases intensity of violence. Alternatively, I find strong evidence that recent participation in violence, intensity of violence, and overall duration of conflict substantially increase the likelihood that an organization will adopt religious rhetoric.
Two general implications follow from this analysis. First, this analysis provides evidence that violent organizations strategically adopt religious rhetoric. I posit above that religious rhetoric meets several of the strategic needs commonly faced by violent organizations, including access to material resources and pathways of recruitment and retention. Consistent with this approach, religious rhetoric is more likely when an organization has recently participated in violence, has been associated with higher intensity violence, and has experienced longer lasting conflict. These findings suggest that religious rhetoric, like many other organizational behaviors in armed conflict, serves as a strategic tool to help political entrepreneurs gain an edge over their rivals.
Second, these findings suggest that an endogenous relationship exists between religion and violence in conflict. Although this analysis does not provide comprehensive evidence against the notion that religion encourages violence, it does provide generalizable evidence that the various logistical challenges associated with participation in violence encourage the adoption of religious rhetoric. This evidence should cast doubt on previous research on religious violence that does not control for temporal sequencing.
Importantly, this analysis does not suggest that all, or even most, religious rhetoric is a strategic tool for furthering the goals of violent actors. These findings do not give justice to the majority of religious rhetoric representing the sincere desire of political actors to express their own religious convictions or to represent those of their constituents. However, these findings do suggest that the common classification of political organizations as either ‘religious’ or ‘secular’ disguises the complex role religion plays in political behavior.
Future research should build on these findings by more precisely measuring and analyzing the role of religion as both an independent and dependent variable in organizational behavior. If religious rhetoric is sometimes adopted in response to defection, for instance, organizations should be more likely to turn to religion following a sudden decline in popular support. Similarly, if access to mobilizing resources encourages religious rhetoric, it follows that the adoption of religious rhetoric should be influenced by the structure of religious social networks and the level of resources controlled by religious institutions.
Future research should also examine the causes and effects of different types of religious rhetoric, including rhetoric focused on theology, collective religious identity, or morally sanctioned actions. More generally, if religion offers strategic resources to violent organizations, why do some violent organizations use religious rhetoric while others do not?
Footnotes
Replication data
Acknowledgements
Earlier versions of this article were presented at the 2014 American Political Science Association conference and the 2015 International Studies Association conference. The author would like to thank Matthias Basedau, Eva Bellin, Steven Burg, Jonathan Fox, Yasser Kureshi, Victoria McGroary, Ragnhild Nordås, Adam Smith, Isak Svensson, and Catherine Worsnop, as well as the editors and reviewers of the Journal of Peace Research. All remaining errors are the responsibility of the author. This work has been supported by a grant from the Research Circle on Democracy and Cultural Pluralism at Brandeis University.
Notes
References
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.
