Identity threats and ideas of superiority as drivers of religious violence? Evidence from a survey experiment in Dar es Salaam, Tanzania

Religion has become increasingly contentious in recent years. Faith-based discrimination, hostility and violence seem to have increased worldwide. But how can faith lead to conflict? In this article, we test the impact of two important dimensions of religion that have been neglected in previous research: the belief in ‘one true religion’ and perceptions of threats by other religious groups. Putting these two potential drivers to the test, we conducted a representative survey experiment with 972 respondents in Dar es Salaam, Tanzania. Results show that one of the tested dimensions, perceptions of threats by others, increases the support to use violence to defend one’s own group. This is particularly the case for religiously intolerant respondents with characteristics such as pre-existing threat perceptions, unfavorable views on intermarriage, or belief in the superiority of their own faith. In contrast, we find relatively weak evidence that the prime of ‘one true religion’ increases the readiness to use violence. Our findings have important implications for policy: We conclude that appeals by leaders to threats by others and intolerance toward other faiths can contribute to more conflict. Political and religious leaders should refrain from capitalizing on such notions and should promote tolerance towards other faiths instead.


Introduction
In recent years, religion has become increasingly contentious. Terrorist attacks, armed conflict and xenophobia based on religious identity or ideology have been making headlines almost on a daily basis and in virtually all parts of the world (see e.g. Asal & Rethemeyer, 2008;Obaidi et al., 2018;Sandler, 2014). With little doubt, religious violence is on the rise (e.g. Svensson & Nilsson, 2018;Toft, Philpott & Shah, 2011). Armed conflict with religious overtones occurs mainly in Africa, Asia, and the Middle East. Notorious cases include Islamist fundamentalist insurgencies in Afghanistan, Nigeria, or Syria (Walter, 2017). Bloody confrontations between religious identity groups have occurred in the Central African Republic or Iraq (Walter, 2017). In Myanmar and many other countries, religious minorities have suffered from violent repression (e.g. Fox, 2018). Europe and North America have also been affected. Terrorist attacks with apparent fundamentalist Islamist motivation have occurred in many European countries. At the same time, right-wing populist movements like 'Pegida' in Germany and other countries have mobilized against perceived threats by 'the Islam' (e.g. Obaidi et al., 2018). 1 How can religion increase the likelihood of violence in the first place? This article concentrates on two potential and theoretically highly relevant drivers of religiously motivated violence: Threats to religious group identities and the belief in the superiority of one's own faith. Both ideas have been discussed widely in the literature (e.g. Appleby, 2000;Assmann, 2009;Fox, 2004Fox, , 2012Henne, Saiya & Hand, 2020) but have only partially been tested. In particular, studies with causal leverage that go beyond correlation and plausibility arguments are lacking. We use unique data from a representative survey experiment in Dar es Salaam, Tanzania, to study whether priming respondents with threat perceptions to their group and the belief in 'one true religion' results in increasing support for the use of violence to defend one's own group. By using experimental priming techniques, we are able to isolate the causal effect of activating these ideas on the support for violence. Tanzania is chosen because of its mixed religious demography and its relatively peaceful interreligious relations -that however have a potential to escalate.
We find that activating the idea of threat perceptions substantially increases support for the use of violence to defend one's own group. Priming the superiority of one's own belief only increases support for violence when we control for other individual characteristics. The effect by the idea of threat perception especially works on the condition that respondents are interreligiously intolerant including characteristics and attitudes such as preexisting threat perceptions, unfavorable views on intermarriage, and few friends with other faiths.
The contribution of this article is threefold. First, we show a causal effect of activating the idea of threat perceptions on the acceptance of violence to defend one's own group. Second, we can show that this effect is at least partially driven by existing unfavorable views towards other religious groups. Third, we believe that our findings have substantial policy implications: if even comparatively subtle religious primes increase support for violence, outright incitement by leaders, peers or media may have much stronger effects -at least when intolerance pre-exists.
The remainder of this article proceeds as follows: In the following section (section 2), we discuss previous works and related research gaps. We then develop our hypotheses in section 3. Section 4 outlines our empirical strategy, especially the logic of the survey experiment. In the fifth section, we report the main results, including the 'raw' experimental evidence, regressions with theoretically relevant control variables, as well as the test of heterogeneous (interaction) effects. In the final section, we summarize our findings and our contribution; we also discuss policy implications as well as challenges and opportunities for future research.

Previous work
Since 9/11 and the ensuing 'war on terror', the body of literature on the religion-conflict link has grown significantly. Related works deal with religious terrorism (e.g. Henne, Saiya & Hand, 2020;Sandler, 2014), religion and armed conflict (e.g. Basedau, Pfeiffer & Vüllers, 2016;Fox, 2004;Svensson, 2007;Toft, 2007) or Islamophobia and antiWestern sentiment in the Muslim world (e.g. Obaidi et al., 2018). Religion can affect violence through its socially relevant sub-dimensions, of which group identity and ideology -or content, ideas or theology 2 -are the most prominent (e.g. Appleby, 2000;Fox, 2012Fox, , 2018Fox & Sandler, 2004). 3 If we conceptualize violence as a special form of collective action, sub-dimensions of religion can provide both capacity and motive for such group behavior. In this article, we focus on the dimensions of group identity and ideology of religion as potential motivational drivers of violence.

Religious identity threats and violence
Religious group identities can lead to violence, like any other group identities, as they are subject to in-and outgroup dynamics (Tajfel et al., 1971). Out-group biases by themselves increase conflict risks -and relative deprivation (Gurr, 1970) or horizontal inequalities (Stewart, 2008) can intensify them. Empirical studies support that parallel religious and ethnic group boundaries increase conflict risks (e.g. Selway, 2011). While there is evidence on political and economic inequalities as drivers of ethnic group violence (e.g. Cederman, Gleditsch & Wucherpfennig, 2017), limited support exists that this effect also holds true for religious group violence. On the country level, it has been shown that grievances lead to a higher propensity to engage in interreligious conflict, but discrimination does not (Basedau, Pfeiffer & Vüllers, 2016) or only to a limited extent (Akbaba & Taydas, 2011). Basedau et al. (2017) find that discrimination of religious groups significantly affects subjective dissatisfaction, but no link can be established to the use of violence. This also holds true for contentious reactions below the level of organized violence (Fox, Bader & McClure, 2019).
At the level of individuals, some evidence points to the fact that especially threat perceptions drive identity-related support for the use of violence. The effect on violence by threats is particularly convincing in theory (e.g. Fox, 2018: 113-115;Obaidi et al., 2018). Threats imply danger and danger may trigger defense, e.g. fighting. Violence is not the only possibly reaction to threats, others being submission and flight. 4 The effect of threats will also depend on other conditions like the previous experience with victimization or the status of a group. However, surveys support an increased likelihood of contentious reactions to threat. Obaidi et al. (2018) review five recent individual data level studies on threat perceptions among Muslims and Christians in different countries. They report solid evidence that threat perceptions increase the likelihood of engaging in conflictual behavior such as participating in 'hostile' demonstrations, defending one's faith or the use of violence. While studies on threats of ethnic and other identity groups have established causal links to more conflictive views and behavior in experimental settings (Branscombe & Wann, 1994;Grant, 1993;Obaidi et al., 2018), this is not the case for religious groups. Evidence is correlational (Obaidi et al., 2018) and the effect might be bidirectional (i.e. pro-violence attitudes lead to threat perceptions as rationalizations). Obaidi et al. (2018) conclude that 'experimental studies [ . . . ] remain to be conducted' (2018: 580).

Religious ideology and violence
It is subject to debate whether or not and to what extent religious ideology is an independent driver of violence (e.g. Fox, 2004Fox, , 2012Fox, , 2018). An influential approach conceptualizes religious and other ideology as a more or less replaceable resource of mobilization or even a 'myth' (e.g. Cavanaugh, 2009;Walter, 2017). Regarding religious ideology, some evidence supporting this claim is provided by Isaacs (2016) who shows that political groups tend to use religious rhetoric more frequently after starting to be involved in conflict than before this threshold has been crossed. However, it is unlikely that we are dealing with an either/or logic of true believers and cold-blooded cynics. Isaacs also identifies cases in which religious rhetoric had been used before violence was applied, and some correlational studies suggest a mobilizing potential of religion. Fox (2004Fox ( , 2018 discusses several pathways from religious ideology to violence. For instance, fundamentalist convictions correlate with more support for the use of violence, especially in defense of one's religious identity group (e.g. Koopmans, 2015). Studies on religious leaders in South Sudan yield evidence that a lack of secularism, that is, the idea that religious laws are more relevant than state laws and intolerance towards other faiths, increases the support for faith-based violence (Basedau & Koos, 2015).
There are close to no studies that are able to provide causal evidence of the influence of religious ideology on violence and conflict (see also Fox, 2018: 117). A related body of literature has, however, investigated the effect of religion on prosociality, that is, generosity in distributional games such as the 'dictator game' (e.g. Norenzayan et al., 2016). Shariff et al. (2016) find a small to medium positive effect of religious primes on prosocial behavior in a meta-study including 93 experimental studies. While some studies find that religious outgroups tend to be discriminated (e.g. Chakravarty et al., 2016;Chuah et al., 2014Chuah et al., , 2016Parra, Joseph & Wodon, 2016), other works find no evidence for intergroup discrimination across religious groups (Johansson-Stenman, Mahmud & Martinsson, 2009;Tan & Vogel, 2008).
Such contradicting findings are, at least in part, due to the 'ambivalence of the sacred' (Appleby, 2000). Different religious contents might have adverse effects on prosocial behavior and discrimination (Hoffmann et al., 2020; see also Appleby, 2000;Philpott, 2007) and likely also on support for violence. 5 While the idea of universal love that informs most theologies (such as Christianity, Islam, Judaism, and other faiths) may decrease conflict, the belief in the universal validity of one's own faith might be particularly conducive to intergroup conflict (Hoffmann et al., 2020). There is lively debate around whether adherents of the missionary monotheistic faiths, in particular Christianity and Islam, tend to denigrate other faiths and are thus more likely to engage in conflict with religious out-groups (e.g. Assmann, 2009;Dawkins, 2006). The history of both faiths lends evidence to this claim, but such effects have rarely been rigorously tested, to the best of our knowledge. Supporting negative effects by dominant religions, Henne, Saiya & Hand (2020) find that in countries with a dominant faith that is favored and supported by the state, the number of terrorist attacks by adherents to the dominant faith increases, not by members of the minority.

Summary: Persisting gaps
The main gaps in the existing body of literature can be summarized as follows: Theoretically, both (1) threat perceptions linked to religious group identity and (2) religious ideology that claims universal validity or even superiority and missionary imperatives can be plausibly linked to an increased likelihood of aggressive and violent action. However, both claims lack causal empirical support, at least for the case of religious ideas and religious groups. Studies on threat perceptions by religious groups rely only on correlational evidence and often lack direct measures for violence; claims for 'one true religion' have hardly been tested as a source of conflict at all and not directly in connection to the use of violence.

Hypotheses
Our hypotheses follow the theorizing outlined in the previous section. Our first hypothesis focuses on a religious ideology of superiority, namely the belief that one's own faith is the 'true religion' and that everybody should follow this faith. The belief in the universal validity and superiority of one's own religion may not only denigrate religious out-groups (see Assmann, 2009;Hoffmann et al., 2020) but also make the use of violence against those groups more acceptable.

H1 (Ideology: True Religion):
Priming respondents with the belief of one's own faith as superior and universally valid increases their support for the use of violence.
Our second hypothesis deals with threat perceptions to group identity. Levels of threats can range from existential threats to fears to milder ones like the simple increase in number by outgroups (see e.g. Obaidi et al., 2018). We believe it makes sense to hypothesize a medium level of threat in terms of unspecified 'domination' and 'harm'. When religious (and other) groups feel threatened in this way, it might trigger the instinct of self-defense. Violence is neither the only possible nor a necessary reaction. Others can be flight or submission as outlined earlier. However, 'defense can often be violent' (Fox, 2018: 213). We assume that, ceteris paribus, threat perceptions will increase the likelihood of supporting the use of violence: H2 (Identity: Threat Perception): Priming respondents with threats against the religious in-group by a religious out-group increases their support for the use of violence.
Finally, we cannot necessarily assume that all respondents react to our treatments the same way. Generally, both the belief in one true religion and threat perceptions can have heterogeneous effects conditional on preexisting attitudes and characteristics, especially those related to religion. Specifically, we assume that it is not the extent (i.e. 'how much') to which people believe but what they believe vis-à-vis other groups. Theoretically, especially pre-existing intolerance towards other faiths seems very likely to further increase or, if absent, neutralize primes. A final set of hypotheses thus expects:

H3a (Conditional/Heterogeneous Effects -Intolerance):
Participants who hold intolerant views towards other faiths are more likely to increase their support for the use of violence when being confronted with both primes. H3b (Conditional/Heterogeneous Effects -Religiosity): Strong religiosity will not affect the impact of the primes, as it does not directly refer to relations to other religious groups. H3c (Conditional/Heterogeneous Effects -State-Religion Perceptions): Unfavorable perceptions of statereligion relations will not affect the impact of the primes either, as they do not directly refer to relations to other religious groups.

Survey experiment
We use a survey experiment to test our hypotheses. This method integrates randomly assigned treatments into a standard survey. As dependent variable, we use the answers to the following statement that many other studies use: 'It is sometimes acceptable to use violence to defend the group you belong to'. Answers range from strongly agree (5) to strongly disagree (1). We apply a betweensubject design and randomly assign the survey participants to one of three groups with different treatments. As treatments, we change the order of the questions and vary the question asked before our dependent variable (see Figure 1). In order to answer H1, we place a statement on the universal validity of the respondents' religion before the dependent variable: 'I am convinced that my own religion is the true religion, and I wish that all mankind followed this faith'. This treatment is referred to as 'one true religion'. Treatment 2, referred to as 'threat perception', tests H2 and uses a medium level of threat in the form of the following statement before the dependent variable: 'Some people say that there is a risk that other religious groups want to dominate our community or harm us otherwise'. In the control group, we follow the original order of the questionnaire. The questions asked before the dependent variable refer to interreligious and same-sex marriage.
Because of the random assignment of the survey participants to the three groups, any variation in the outcome of the question following the treatments can be attributed to the treatment (assuming that both groups are otherwise equal). Thus, using a survey experiment eliminates threats to the internal validity of the results and facilitates causal inference.

Case selection and data collection
Case selection: Our survey was conducted between 3 July and 28 July 2017 in Dar es Salaam, the biggest city in Tanzania. Selecting Tanzania and Dar es Salaam respectively is useful for several reasons: Tanzania can be considered a medium-likely case for religious tensions and Dar es Salaam, as the biggest city of Tanzania, constitutes a melting pot with a mixed interreligious demography. Besides, practical considerations and the fact that little experimental work has been done on the case played a role for our case selection.
Religious demography: Tanzania's religious demography is characterized by a Christian majority, with Islam as the biggest minority religion. Census data on the distribution of faith in Tanzania is not available. To this day, the state avoids carrying out any religious census. A recent estimation suggests that about 60% of believers are Christians while about 30% are Muslims (Afrobarometer, 2018). According to a US State Department report, local commenters maintain that there are roughly equal numbers of Christians and Muslims in the country (Religious Freedom Report, 2016).
There is no clear geographic separation between the two main religious communities, although Muslims form a strong majority on Zanzibar and are more concentrated in coastal areas. Originally, Islam spread from the coastal regions of the country into the mainland, but European colonialism introduced Christianity throughout the whole territory. The policies of Tanzania's first president, Julius Nyerere, relocated citizens and thus presumably strongly reduced if not eliminated religious and ethnic geographic boundaries until the 1980s on the mainland (excluding Zanzibar). Dar es Salaam can be considered as a melting pot for the whole country, attracting migrants from more rural areas. It presumably shows a comparably large share of Muslim believers compared to the rest of the country.
Interreligious relations: In general, interfaith relations in Tanzania are relatively peaceful and tolerant with comparatively little religious violence (Basedau, Vüllers & Körner, 2013). There have, however, been recurring conflicts between Muslims and Christians since the 1980s (Heilman & Kaiser, 2002). In particular, an increasing number of attacks on churches and mosques have been noted prior to our field research (Religious Freedom Report, 2016). However, Tanzania has not experienced an organized religious or other violent intrastate conflict and does not constitute a 'most likely case' for contentious interreligious relations. At the same time, both the mixed demography and some past tensions show a potential for escalation that should be taken seriously. 6 In sum, Tanzania represents a medium-likely case for large-scale violent conflict and seems therefore particularly promising to study.
Sample selection: Dar es Salaam is divided into three districts, which each consist of several wards. We applied a stratified sampling procedure by which we randomly selected 20 wards from a complete list of wards in Dar es Salaam. Each day, three different wards were covered by a team of three to four interviewers. Within each ward, a central starting point was identified by the research team. Each interviewer covered a randomly selected street starting from this point. Interviewers knocked on every fifth door and interviewed the person who opened the door. 7 Quality checks of the interviews were conducted by the research team on a daily basis.
Questionnaire: The questionnaire is structured around various clusters of items, including classical demographic and 'mundane' attitudes and characteristics, but it focuses on religious beliefs and on attitudes reflecting how religion influences worldly conduct. In particular, we distinguish between three theoretically distinct and relevant dimensions of religion, namely (1) the respondents' religiosity, (2) their more or less tolerant views on interreligious relations and (3) the relation between religion and the state (for details, see later in the article). When possible, we used established phrasings from the Afrobarometer or the World Value surveys. Regarding our key variables of interest -support of violence to defend one's group, threat perceptions and 'one true religion' -we have made a special effort to address them cautiously in order not to create any harm and also not to produce interviewer effects. The survey was conducted using tablets by a gender-and faithmixed team of ten local interviewers from the University of Dar es Salaam in the national language, Swahili. Interviewers received a two-day interview training before the start of the survey.

Descriptive analysis
Our sample consists of 972 individual respondents, of which 308 belong to the control group, 327 to treatment 1 (one true religion) and 337 to treatment 2 (threat perception; see Table I). Regarding socio-economic and demographic characteristics, 51% of the total sample are women (see Table AI in the Online appendix). 8 The mean age is 32 years and for 47% of the participants, primary schooling is the highest educational level. Financial satisfaction is, with an average of 3.7 on a scale from 1 to 10, rather low. About 23% of the participants have experienced some kind of violence themselves in the past two years. There is only roughly an even share between the two major religious communities in the survey, with 42% of the respondents identifying as Christian and 58% as Muslim (see Table AI in the Online appendix). Table II summarizes the religious attitudes of our respondents, according to three theoretically distinct dimensions. In line with our theoretical considerations and to reduce the number of variables, we create composite variables for religiosity and intolerance toward other religions. 9 We test the state-religion relationship (relevant for Hypothesis 3c) with the variables 'unfair treatment by the state' and 'religion more important than laws'. 10 We choose four indicators to measure religiosity, relevant for Hypothesis 3b, regardless of the content of beliefs: (1) relative relevance of religious identity, (2) religion's influence on acting, (3) frequency of attending religious services and (4) the importance of prayer. Most of our respondents choose being Tanzanian as their most important identity group (64%) while 17% of the respondents identify first and foremost with their religious group. 11 Otherwise, levels of religiosity are rather high: 64% of the respondents state that religion has a large influence on acting in their daily lives. Moreover, the majority of respondents attend religious services at least once a week (54%) and claim that prayer is a regular part of their daily routine (88%).
To measure intolerance toward other religions, relevant for Hypothesis 3a, five indicators are included in our survey: (1) negative attitudes toward interreligious marriage, (2) not having friends from other religions, (3) being against equal rights of different religious groups, (4) agreement to the statement that other religions want to dominate and (5) agreement to the statement that there is only one true religion. Quite a large proportion, 44% of our respondents, find interreligious marriage unacceptable. On the other hand, the large majority of respondents claim that they have friends from other religions, only 12% state that they do not. In line with that, only 3% of the respondents find that religions should not have equal rights; 43% of participants fear that other religions want to dominate and 86% of respondents think that their religion is the only true religion and wish that everyone followed their faith. Taken together, even though intolerance is rather low, the majority of respondents have at least some intolerant views (e.g. thinking that their religion is the only true religion) toward other religions.
When it comes to relations between religion and the state, we observe that one quarter of the respondents feel that their religious group is treated unfairly by the government (24%). More than half of the respondents find it more important to follow religious rules as compared to state laws. Table II in the Online appendix shows that our randomization of the treatments worked quite well. The only variables where we see (slight) differences are 'married' and 'religious identity'. We seem to have fewer married respondents in the control treatment (33% in control and 41% in the other two groups). Furthermore, more participants identify first with their religious group in treatment 1 as compared to the other groups (20% in treatment 1 vs. 15% in treatment 2 and 16% in the control group). 12

Treatment effects
Generally, more than two-thirds of the respondents (strongly) disagree with the use of violence to defend their groups and only less than one quarter (strongly) agrees (see Figure A1 in the Online appendix). Do the treatments change responses? Indeed, we observe that the treatments result in a shift of the distributions towards an increase in support for violence (see Figure A1 in the Online appendix). A lower share of respondents strongly disagree and a higher share strongly agree to the use of violence to defend one's group compared to the control group. Figure 2 displays the mean values over the three groups. A Wilcoxon rank-sum test shows a significant difference between the control group and treatment 2, 'threat perception' (p-value ¼ 0.05). The difference Figure 2. Mean values of the acceptance of violence to defend one's group per treatment 9 The composite variables religiosity and intolerance are calculated by adding up the respective dummy variables (we transformed the initial variables into dummy variables as their scale was not uniform). Religiosity ranges from 0-4, intolerance ranges from 0-5. 10 We did not use an index here as it is only two items that also might measure theoretically different aspects. 11 We cannot rule out an effect by social desirability as public discourse strongly favors a common Tanzanian identity. 12 We control for these imbalances in our regression analyses.
between treatment 1, 'one true religion', and the control treatment is not statistically significant (p-value ¼ 0.17). These findings lend evidence to support Hypothesis 2 on the effect of threat perceptions, but do not convincingly support Hypothesis 1 on the effect of 'one true religion'.

Multivariate regression analysis
In a next step, we test our hypotheses using multivariate regression analysis. Table III shows regressions for the acceptance of violence. We present four models: We start by estimating a baseline OLS regression including only the two dummy variables for our treatments (Model 1). In this model, the dependent variable 'acceptance of violence to defend one's group' lies between 1 (strongly disagree) and 5 (strongly agree). Model 2 presents the results of a logit regression where we code the acceptance of violence as zero if respondents strongly disagree or disagree to the use of violence, and 1 otherwise. 13 In a next step, we test a theoretically informed model that includes religious and other control variables that may influence the dependent variable, the use of violence (see for instance, Basedau & Koos, 2015). Including control variables has the advantage of reducing estimation error, which helps us to find a more efficient estimator of the treatment effects while at the same time accounting for the slight imbalances in the randomization of treatments (see Table AI in the Online appendix). Furthermore, we add interviewer controls as interviewers may affect responses to our dependent variable, especially when it comes to a sensitive topic like justification of violence. Even though interviewers are randomly assigned to different treatments, it may be the case that one interviewer systematically evokes different answers in a specific treatment. As theoretically relevant controls, we use the three distinct religious dimensions, already described earlier, namely (1) levels of religiosity regardless of content, (2) intolerance toward other religions and (3) views on the relation between religion and the state (part of Hypotheses 3a-c). Moreover, we include nominal type of faith (Christian and Muslim) and sociodemographic variables such as gender, education level, age and previous experience with violence as well as interest in politics. However, due to the limited availability of the information on all control variables, the size of the sample is reduced from 972 to 871. 14 a 0 'no', 1 'yes'; b 0 'no influence/a small influence/some influence/a fair amount of influence', 1 'a large influence'; c 0 'never/a few times a year/ about once a month', 1 'about once a week/a few times a week'; d 0 'I never pray/prayer has little importance in my life/I pray only during formal ceremonies/I usually pray in times of stress or need but rarely at any other time', 1 'Prayer is a regular part of my daily life'; e 0 'strongly agree/agree/neither nor to interreligious marriage', 1 'disagree/strongly disagree to interreligious marriage'; f 0 'strongly agree/agree/neither nor to religious groups have equal rights', 1 'disagree/strongly disagree to religious groups have equal rights'; g 0 'strongly disagree/disagree/neither nor', 1 'agree/strongly agree'; h 0 'never/rarely', 1 'sometimes/always'. 13 For robustness, we show the results of a logit regression with a different cutoff point and an ordered logit model in Table AIII in the Online appendix. While the results remain robust to an ordered logit model, choosing a different cutoff point for the logit specification leads to the T2 variable losing its significance. We explain this as agreement to violence is prone to social desirability bias and shows a skewed distribution. As the support for violence is relatively low, results lose their significance. 14 The drop in observations results from different variables with don't know answers or missing values (see Table AI in the Online appendix), but this drop is not driven by particular variables.
In the baseline regressions (Models 1 and 2), the one true religion treatment turns out positive but insignificant. Only when we add control variables (Models 3 and 4), the treatment increases in magnitude and turns significant on the 5% level (Hypothesis 1). 15 Result 1: We find relatively weak evidence that priming respondents with the belief of one's own faith as superior and universally valid increases their support for the use of violence.
Treatment 2, that is, receiving the prime that other religious groups want to dominate or harm one's group Standard errors in parentheses. Standard errors are clustered at the level of religious groups. y p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001. Models 2 and 4 are logit models where 0 means that respondents strongly disagree or disagree to the use of violence. 1 means that participants neither agree nor disagree, agree or strongly agree to the use of violence. 15 Adding control variables leads to changes in the 'one true religion' variable (as it increases in size and becomes significant). However, the direction of its impact on religiously motivated violence remains the same. One reason for the slight changes is that Model 1 uses 972 observations while the observations drop to 871 in Models 3 and 4. Running Models 1 and 2 with the reduced dataset (n ¼ 871) leads to an increase in the size of the 'one true religion' variable as compared to the full number of observations (see Table AII in the Online appendix) with a p-value of the 'one true religion' variable of p ¼ 0.125. Adding control variables, particularly sociodemographic and interviewer controls, leads to a further increase and to a p-value of below 0.10. Thus, we suggest that imbalances in controls are driving the differences between the models.
(Hypothesis 2), turns out positively significant on the 5% level in the baseline models (Models 1 and 2) as well as when adding control variables (Models 3 and 4). This treatment increases the acceptance of violence by 0.183, an increase in the acceptance of violence by about 8% (Model 1) compared to the mean value of the control group.
Result 2: We find evidence that priming respondents with threats against the religious in-group by a religious out-group increases their support for the use of violence.
When it comes to the control variables, a higher degree of religiosity is correlated with lower support of religiously motivated violence. On the contrary, content matters: participants who hold intolerant views toward other religious groups or believe that religion is more important than state laws and who feel treated unfairly tend to show a higher support for violence. This means that different aspects of religion can lead to a lower (religiosity) or higher (intolerance, valuing religious laws over state laws, feeling discriminated treated unfairly by the government) support for religiously motivated violence.
Regarding confounding effects of non-religious control variables, some 'usual suspects' such as female gender do not show the expected (significant) negative relation to supporting violence, while others do: Older respondents as well as participants with a higher financial satisfaction are less prone to support the use of violence. Having been personally attacked is (weakly) correlated with a higher support of violence. Some control variables show somewhat unexpected findings: Both better educated people and those who are more interested in politics show a higher support for violence -pointing to a potential downside of political mobilization.
Heterogeneous effects: Intolerance Next, we want to test Hypothesis 3a and look at interaction effects between the treatment variables and intolerance toward other faiths. Hypothesis 3a assumes that it is not the extent (i.e. 'how much') to which people believe but what they believe. In order to test the hypothesis, we use the composite variable of intolerant religious views, already shown in the descriptive statistics in Table II. 16 Table IV includes interaction effects with the intolerance composite variable. 17 Model 1 shows the results of an OLS regression with the acceptance of violence as dependent variable between 1 (strongly disagree) and 5 (strongly agree). Model 2 is a logit model where the dependent variable is a dummy as described earlier.
We observe that the interaction between treatment 2, the threat perception treatment, and the intolerance variable, turns out significant at the 0.1% level (Models 1 and 2). Thus, the threat perception prime strongly affects people who have rather intolerant views compared to those who do not. 18 The first graphic in Figure 3 illustrates this interaction effect: Being assigned to treatment 2 significantly increases the support for violence to defend the own  17 We present interactions for the other composite variables in Table  AVIII in the Online appendix. 18 Other model specifications with different cutoff points are presented in Table AV in the Online appendix. Again, the results are robust to the ordered logit specification but not to the different logit specification.
group (red area) compared to not being assigned to treatment 2 (blue area). Models 1 and 2 in Table IV also show the interaction effect between the 'true religion' treatment and the intolerance composite variable. It turns out to be significant at a 10% level in the OLS regression (Model 1). However, the result is not robust to a logit model (Model 2). Thus, while we find a strong effect on the 'threat perception' treatment on participants with intolerant views, we only find limited evidence for such an effect for respondents in the 'true religion' treatment. A higher religiosity is correlated with a lower support for violence. However, confirming expectations of Hypothesis 3b, this effect is independent of our treatment variables (see Table AVIII in the Online appendix). Valuing religious laws over state laws and feeling treated unfairly is correlated with a higher support for violence. In contrast to Hypothesis 3c, some heterogeneous effects seem to be at work: Participants who value religious laws over state laws and are assigned to the 'one true religion' treatment are more in favor of violence. While participants who feel that their religion is treated unfairly show higher support for violence, participants who feel treated unfairly and are assigned to the threat perception treatment seem to be less in favor of violence (see Table AVIII in the Online appendix). This result could potentially be interpreted as fear of violence which is activated by the threat perceptions treatment by those who feel treated unfairly.

Discussion and conclusion
This article has investigated two theoretically highly relevant, potential religious drivers of violence: threats to religious group identity and the belief in the superiority of one's own faith. We have used unique data from a representative survey experiment in Dar es Salaam, Tanzania, to study whether priming respondents with threat perceptions to their group and the belief in 'one true religion' results an increased support for the use of violence.
We find that especially the 'threat perception' prime makes people more apt to condone the use of violence. We find only limited evidence that 'one true religion' increases support for violence among respondents. The prime of 'threat perception' works particularly strongly on the condition that negative interreligious attitudes Figure 3. Graphical presentation of interaction effects of religion variables and treatments. The graphs show the interaction effects of being assigned to t2 ('threat perception') vs. not being assigned to t2 as well as the effects of being assigned to t1 ('true religion') vs. not being assigned to t1. already exist, partly but not exclusively driven by actually feeling threatened by other religions.
The contribution of this article is threefold. First, we show causal effects of threat perceptions on the support for violence for religious groups. Second, we can show that already holding unfavorable, intolerant views of other religious groups is one channel through which support for violence can mobilize. Third, we believe that our findings are politically highly relevant: if even comparatively subtle primes of threat perceptions increase support for violence, outright incitement by leaders, peers and media may have much stronger effects. Policy implications are closely connected and relevant well beyond the case of Dar es Salaam -which in itself is an important add-on to a literature dominated by Western samples. We have reason to believe that a tolerant 'secular ethics' that spans across different religions (Dalai Lama, 2016) should be promoted. This has implications for world religions as universal validity is a central feature of Abrahamic, monotheist faiths (e.g. Assmann, 2009).
Our results, however, also suggest that the ability of leaders, peers, social and conventional media to incite violence is limited as the effects materialize mostly under the condition that negative views already exist. Policy implications are quite obvious: threat perceptions need to be avoided and leaders should refrain from capitalizing on or 'constructing' them and other intolerant interreligious attitudes. It will be particularly crucial to study further the determinants of religious intolerance and threat perceptions, in and outside Africa, including movements like 'Pegida' in the West.
There are several closely connected challenges and opportunities for future research: More knowledge is required on the drivers of 'toxic' religious ideas, especially intolerance and threat perceptions. It seems plausible that some of the determinants that are typically considered to reduce support for violence -such as education or interreligious personal contact -warrant closer inspection. 19 Moreover, our findings need to be replicated in other settings as we do not know if Tanzania is atypical.
Moreover, we only test attitudes and not actual actsthere are ethical and other limitations for the study of violence. Although the dependent variable directly asks for defending one's religious identity group, there are alternatives for the operationalization of 'religious violence', especially in terms of ideology-based violence. As independent variables, future studies could use alternative and more concrete threats rather than the abstract notions of 'domination' or 'harm'. One may also distinguish between the two as they do not necessarily mean the same. 20 Other aspects that should be investigated in more depth include feelings of discrimination by the state and the belief in the superiority of religious over state laws. These variables are positively correlated with the support of violence in our models. Especially believing in the superiority of religious over state laws is a conventional measure of fundamentalism that comes with support for violence.
In an African context, ethnicity may be an additional driver of the results. Future studies should explore this, but it is unlikely in Tanzania. Tanzania features as one of the countries in which ethnicity is said to be not politically salient (Posner, 2004); more importantly, it is highly ethnically fractionalized -we identified more than 40 groups in our sample -and ethnic and religious group boundaries do not run parallel.
Generally, other dimensions of religion, such as institutions and organizations, warrant closer inspection, especially when we want to determine capacity for violent collective action. The capacity and opportunity aspects of the causes of conflict have not been at the center of this article and are often overlooked in conflict research in general (see e.g. Collier & Hoeffler, 2004). It may also be interesting to study emotions, which strongly inform religious thought and behavior.
Finally, we should not forget that religion can be a source of peace, too: There are many religious peace norms such as reconciliation and forgiveness, and religious actors demonstrate engagement in peace processes worldwide (e.g. Abu-Nimer, 2001). We show that it is not religiosity, that is to say, how much people believe, that drives the acceptance of violence but what they believe. This study suggests that (in)tolerance matters: To end religious conflict and make religion work for peace, it is worthwhile to study the multifaceted effects of faith more closely.

Replication data
The dataset, codebook, and do-files for the empirical analysis in this article, along with the Online appendix, are available at https://www.prio.org/jpr/datasets/. All analyses were conducted using STATA/SE 16.1.