Participation in Communal Religious Practices

There is a growing and substantial literature that examines the effects of participating in communal religious practices upon political motivation and beliefs across faiths, employing different research methods and empirical commitments. Arikan and Bloom (Reference Arikan and Bloom2019), using data from the World Values Survey, found that involvement in religious social networks fosters individuals' propensity to protest. Calhoun-Brown (Reference Calhoun-Brown1996) found that among African Americans, attending a political church was a strong predictor of political involvement and motivation to engage in political concerns. Harris (Reference Harris1994), studying religion and political behavior among African Americans, found that participation in church acts as both a psychological resource for individuals and facilitates collective action by decreasing the transactional costs of organizing (see also Layman Reference Layman1997; Jones-Correa and Leal Reference Jones-Correa and Leal2001). Campbell (Reference Campbell2004), in his study of evangelical Christian political behavior, found that the tight social networks formed through intensive church participation and other church-related activities facilitate rapid and intense political mobilization.

Clingingsmith et al. (Reference Clingingsmith, Khwaja and Kremer2009) studied Pakistanis who undertake the Hajj (pilgrimage to Mecca) to exposit how participation in the Hajj effected communal, sectarian, and ethnic attitudes. They found that Hajjis (those who did the Hajj) were more inclined toward peace and harmony with non-Muslims, with Muslims of different sects, and different ethnic groups. While undertaking Hajj is correlated with inclinations toward concord among different Pakistanis; Ginges et al. (Reference Ginges, Hansen and Norenzayan2009), using survey data for Palestinians, found that attending communal religious services positively correlated with support for suicide attacks, likely because it enhanced coalitional commitments among attendees. They also studied the impact of participating in regular religious service in six religions across six countries and found that while regular, individual prayer had no impact upon beliefs about martyrdom or out-group hostility, attending communal religious services did. Hoffman and Nugent (Reference Hoffman and Nugent2017), using survey data from Lebanon, found communal prayer increased individual support for arming political parties among those groups engaged in conflict while religious practice tended to foster opposition to militarization among non-combatants. They propose that communal religious practice increases the salience of group interests both through informational mechanisms as well as identity formation.

Ahsan Butt (Reference Butt2016) examined the impact of communal Friday prayers upon street mobilization in the large Pakistani port city of Karachi. He found that street mobilizations were far more likely on Friday than any other day of the week. One of the reasons Butt proposes for this is the content of the khutba, or Friday prayer sermons.Footnote ¹ It is common for Friday sermons to be provocative. Sometimes they focus upon international actors (i.e., various U.S. invasions of Muslims countries, Israeli oppression of Palestinians, various European papers' publications of cartoons of the Prophet, France's ruling on Islamic head coverings, etc.). At other times, the khutba may focus upon domestic political actors or issues (the government, the government opposition, specific policies such as the content of curricular or textbook reforms, sales of alcohol, etc.). The khutba may dilate upon conflicts with sectarian or communal minorities such as the Ahmadis, Sufis, Hindus, etc.Footnote ²

While the general literature reviewed above suggests that participation in Friday communal prayers may increase individual support for Islamist militant groups' goals or means, Bangladesh is one of many countries which has sought to regulate the content of khutba as a tool to mitigate Islamist violence. Prime Minister Hasina has long worried about Islamist extremism in her country in part because of HuJI-B's attempt to assassinate her. Following the IS' Holey Bakery attack in July 2016, Hasina's government began regulating these sermons across hundreds of thousands of mosques in the country. Her government shuttered television channels that promoted ideologies that were sympathetic to that of Islamist militant groups and began monitoring the sermons as well as social media to prevent dissemination of messages in support of Islamist militant groups' goals and means. The government enlisted the state-run Islamic Foundation, which is an oversight group for mosques and religious establishments, to prepare the Friday sermon in the national mosque and asked other mosques to use it as well (Express Tribune 2016, also see U.S. Department of State 2018). Additionally, in December 2015, the country's ulema issued a specific fatwa (religious decree) against the IS to counter the threat it posed. According to Maulana Fariduddin Masoud, chairman of Bangladesh Jamiatul Ulama (BJU), a national body of Islamic scholars in the country, “The fatwa is to be jointly signed off by 100,000 Islamic scholars from across the country…It identifies IS and its local Islamist militant supporters—who are killing people and indulging in terrorist activities in Bangladesh and elsewhere—as not just the ‘enemies of Islam’ but also ‘enemies of the Muslims’” (Asian News 2015). Moreover, the BJU took the lead in organizing and training clerics and scholars to support the fatwa and take the campaign to the grassroots level.

Bangladesh is not the only country to attempt to regulate the content of khutba to align it with state goals: states across the Middle East and North Africa as well as Central, South, and Southeast Asia have also done so (e.g., Wiktorowicz Reference Wiktorowicz1999; Epkenhans Reference Epkenhans2009; Errihani Reference Errihani2011). Niger, since 1974, has done it with varying degrees of success over the ensuing decades and regimes (Elischer Reference Elischer2015). In Iran, the government determines the content of the khutba and clerics are directly employed by the state while marginalizing or even prosecuting dissident clerics in the Special Court for Clerics (Tezcür Reference Tezcür2012, 124). Turkey has variously regulated Friday prayers to promote Islamist populism and violence under Erdoğan (Yilmaz and Erturk Reference Yilmaz and Erturk2021) as well as Kemalist secularism (Köker Reference Köker2010). Malaysia too has developed legal and administrative infrastructure to shape the ways in which Malaysians understand Islamic law itself including exclusive control over all existing mosques and the erection of new ones, the appointment and discipline of imams (religious leaders) while federal and state agencies dictate the content of Friday sermons (Moustafa Reference Moustafa2014).

While such efforts may be popular among governments seeking to assert control over religious activities, there is limited evidence attesting to the efficacy of these efforts much less evidence of salubrious impacts upon individuals. After all, persons who distrust state-sponsored khutba can always choose not to attend or, where possible, find alternative arrangements beneath the prying eyes of the state. Conversely, those who do attend may do so because they support the state's effort. Equally notable are the conclusions of Henne (Reference Henne2019), who combined data from the Religion and State Round 3 (RAS3) dataset, which provides measures of several types of government control over religious institutions, and data from the Global Terrorism Database's (GTD), which measures the impact of government interference upon incidence of terrorism, to study the effects of the former upon the latter. He finds that government efforts to exert control over religious institutions increase the incidence of terrorism. This finding complements that of Saiya and Scime (Reference Saiya and Scime2015) who present evidence that religious terrorism derives from a paucity of religious liberty.

Because it is not obvious a priori what—if any—correlation may obtain in the case of Bangladesh, we posit the following Hypothesis:

This discussion requires an additional caveat: in South Asia, as a rule, women do not pray in mosques. Since most mosques do not have reserved places for women to pray inside, the rare woman who chooses to participate must pray upon carpets which are distributed outside of the mosque and endure the immodesty of this arrangement. This does not mean that women cannot listen to the khutba. In Bangladesh—as is the norm throughout South Asia—mosques loudly broadcast the khutba through speakers and/or other public address systems. Both because mosques tend to be small and distributed across the localities they serve and because they employ such public-address systems, many women can hear the khutba, despite the distance and the often-garbled nature of the sound. They may also hear second-hand accounts of the contents of the khutba from the male household members who attended. They may also perform their prayers albeit in privacy or with other female household members. Nonetheless, most women will not experience the communal aspects of Friday prayer in a mosque. Our survey data comport with this general phenomenon in South Asia: only 11 of the 1,752 who attended congregational Friday prayers were female.

Piety and Support for Islamist Militancy

Despite some lingering enthusiasts for Samuel Huntington's “clash of civilizations” thesis (Huntington Reference Huntington1984; Reference Huntington1993; Reference Huntington1996), which posits a fundamental conflict between the Christian West and the so-called Islamic World and asserts that Muslim support for Islamist violence is moored in adherence to Islam itself (Laqueur Reference Laqueur1999; Stern Reference Stern2003; Mendelsohn Reference Mendelsohn2005), as Jonathan Fox (Reference Fox2005, Reference Fox and Haynes2021) observed, most empirical studies of support for Islamist violence find little association between simply believing in Islam and supporting violent politics (Tessler and Nachtwey Reference Tessler and Nachtwey1998; Esposito Reference Esposito2002; Tessler and Robbins Reference Tessler and Robbins2007; Clingingsmith et al. Reference Clingingsmith, Khwaja and Kremer2009; Ginges et al. Reference Ginges, Hansen and Norenzayan2009; Fair et al. Reference Fair, Malhotra and Shapiro2012; Ciftci et al. Reference Ciftci, O'Donnel and Tanner2017; Fair et al. Reference Fair, Littman, Malhotra and Shapiro2018b; Piazza Reference Piazza2021). In fact, several studies using different data and research methodologies consistently find that Muslim respondents who described themselves as more religiously observant were less supportive of Islamist militant groups and/or Islamist violence (Jo Reference Jo2012; Zhirkov et al. Reference Zhirkov, Verkuyten and Weesie2014; Ciftci et al. Reference Ciftci, O'Donnel and Tanner2017; Piazza Reference Piazza2021). At least two scholars have presented limited evidence that individuals with greater knowledge of Islam, obtained through Quranic study groups and other pietic practices, are better able to resist the arguments of militant thought leaders and thus less likely to support Islamist militant politics (Wiktorowicz Reference Wiktorowicz2005; Fair et al. Reference Fair, Goldstein and Hamza2016).

When a positive correlation between embracing Islam and violence does exist, the relationship is driven by a particular understanding or interpretation of Islam and the concomitant personal obligations they believe Islam establishes for Muslims (Haddad Reference Haddad2004; Fair et al. Reference Fair, Malhotra and Shapiro2012; Fair et al. Reference Fair, Littman and Nugent2018a; Berger Reference Berger2019).

This literature gives rise to our second testable hypothesis:

Shari'a and Support for Islamist Militancy

Empirical studies come to mixed conclusions about the posited relationship between preferences for Shari'a and support for Islamist violence. Fair et al. (Reference Fair, Littman and Nugent2018a) identify and tweeze out several factors that are usually collapsed into survey items querying respondent “support for Shari'a.” They found that support for the physical punishment prescribed by Shari'a explains support for Islamist militancy.

Piazza (Reference Piazza2021) has further elaborated upon the connections between support for Shari'a and support for Islamist violence. He theorizes that a person's beliefs about the compatibility of Shari'a and democracy would be a key predictor in respondent support for Islamist militant groups. He employed 2017 survey data from six Arab countries to understand the support for violence perpetrated by the IS. He found that respondents who favor both the implementation of Shari'a and believe Islam is incompatible with democratic principles are more likely to support the IS. In contrast, those who believe that Shari'a is compatible with democracy are significantly less likely to support the IS.

In the context of Bangladesh, Bangladeshi militants make specific appeals to Shari'a generally and Hudood punishments as part of the IS they wish to establish. For these three groups, democracy and Shari'a are not mutually compatible. To achieve their goals of establishing Shari'a in Bangladesh, all the groups under study here have used violence in hopes of coercing the state to abandon its judicial system and democratic character and replace both with Shari'a law and institutions to enforce it (Bhattacharyya Reference Bhattacharyya2017) while generally making no appeals to public service provisions as the Afghan Taliban claim to do.Footnote ³ This discussion gives rise to our third hypotheses:

As noted above, Bangladesh's Islamist militant groups have been arguing for a government based upon their interpretation of Shari'a for decades and they have specifically targeted Bangladesh's secularists and liberal activists presumably because they oppose the Islamist agenda of rendering Bangladesh an Islamist state. It is important to understand what secularism means in the Bangladesh context. In Bangla, secularism is reflected in the phrase “dharmo niropekhkhota,” which literally means “religiously neutral.” Sheikh Mujibur Rehman, who led Bangladesh's independence movement, explained that “secularism [dharma niropekhkhata] does not mean absence of religion. Muslims will observe their religion and nobody in this state has the power to prevent that. Hindus will observe their religion, and nobody has the power to prevent that. Buddhists and Christians will observe their respective religions, and nobody can prevent that. Our only objection is that nobody will be allowed to use religion as a political weapon” (Government of Bangladesh 1972, cited by Uddin Reference Uddin2015, 44).

The historical role that the struggle over secularism has played in Bangladesh's civil and not-so-civil society (Khondker Reference Khondker2006) along with the scholarly works discussed above gives rise to our fourth testable hypotheses:

Gendered Support for Islamist Violence

There is a growing and increasingly sophisticated body of scholarship that seeks to explain women's support for religious and political ideologies that undermine women's equality of access to opportunities and/or outcomes, inclusive of violence against women and girls. Setzler and Yanus (Reference Setzler and Yanus2018); Tien (Reference Tien2017), among others, used survey data to understand why some women—overwhelmingly white—supported Donald Trump despite his overt misogyny and appeals to male supremacy. Others have sought to explain why some women support restrictions on women's access to reproductive justice (e.g., Jelen et al. Reference Jelen, Damore and Lamatsch2002; Swank Reference Swank2021), equal rights legislation (Burris Reference Burris1983; Young Reference Young2007), or fundamentalist versions of religious beliefs which actively promote theological justifications for depriving women and girls of equal opportunities and even violence against the same (e.g., Blaydes and Linzer Reference Blaydes and Linzer2008; Gerami Reference Gerami2012). Many of these studies conclude that women derive social capital or other private benefits from supporting policies that undermine the collective interests of women and girls. There is also an ever-growing literature on female direct and indirect participation in the perpetration of violence (inter alia, Parashar Reference Parashar2011; Lahoud Reference Lahoud2014; Pearson and Winterbotham Reference Pearson and Winterbotham2017; Bodziany and Netczuk-Gwoździewicz Reference Bodziany and Netczuk-Gwoździewicz2019; Huey et al. Reference Huey, Inch and Peladeau2019; Sela-Shayovitz and Dayan Reference Sela-Shayovitz and Dayan2019). These studies too draw out the ways in which such participation in violence bestows upon women different forms of agency as well as social capital.

In contrast, there is a paucity of scholarly studies of respondent-level, female support for political violence generally or Islamist violence specifically. In fact, gender remains both under-theorized and under-studied as a potential explanatory factor in support for violence in empirical studies. If scholars include gender as a variable in their empirical studies at all, almost all do so as a control variable rather than a study variable. Depending upon the countries studied, the research methodology employed, the dependent variable selected, and the nature and character of the survey data used in these studies, among other considerations, scholars exposit differing impacts of gender as a control variable. The various scholars who have used Pew's surveys of the Muslim World to study respondent support for suicide attacks that kill civilians find that when gender is statistically significant, males are more supportive than females (e.g., Shafiq and Sinno Reference Shafiq and Sinno2010; Cherney and Povey Reference Cherney and Povey2013; Fair et al. Reference Fair, Hamza and Heller2017; Fair et al. Reference Fair, Hwang and Majid2019). Notable exceptions include Wike and Samaranayake (Reference Wike and Samaranayake2006) and Fair and Shepherd (Reference Fair and Shepherd2006).

Several scholars who have used data from non-Pew surveys to examine support for various kinds of Islamist violence arrive at different conclusions when they include gender as a control variable. LaFree and Morris (Reference LaFree and Morris2012), using START data to study support for anti-American violence in Egypt, Morocco, and Indonesia, found no statistically significant gender relationship while Kaltenthaler et al. (Reference Kaltenthaler, Miller and Christine Fair2015), using a novel 2013 dataset for Pakistan, found females were statistically more likely to support some Islamist terrorist groups than males. Unfortunately, none of these studies dilate upon gender as a significant consideration.

One of the rare empirical studies which include gender as a primary study variable is Afzal (Reference Afzal2012), who used 2008 START data for Pakistan. Afzal sought to exposit the relationship between gender and support for terrorism with education as a mediating variable. Afzal found that as women's education increased, they became less likely to support Islamist violence relative to comparably educated men all else equal. In contrast, uneducated women were more likely than comparably uneducated men to support Islamist militancy, all else equal. However, women in Pakistan are far less likely to be educated than men: 63.9% of females over the age of 10 are illiterate compared to 32.2% of men (Pakistan Bureau of Statistics, n.d.).

Another exception is a study by Fair and Hamza (Reference Fair and Hamza2018) in which the authors employed data derived from a large national survey of Pakistanis fielded in 2011 to exposit the impacts of gender upon support for two militant groups operating in and from Pakistan, namely: the sectarian group Sipha-e-Sahaba-Pakistan (SSP) and the Afghan Taliban. They found that while women were more likely to support the SSP, gender did statistically explain support for the Taliban. They too fail to propose a mechanism through which gender would exert different effects upon support for violence apart from considering the varying opportunity structures afforded by the groups.

Blaydes and Linzer (Reference Blaydes and Linzer2008), in their study of female support for Islamist fundamentalism and concomitant support for practices that are anti-women, posit that women are making a rational trade-off between returns in the labor and marriage markets. Women with fewer labor market options may invest in their attractiveness in the marriage market, in which fundamentalist views may result in better marriage options. Women with greater labor market opportunities may eschew fundamentalism because some behaviors associated with fundamentalism will diminish labor market opportunities (i.e., wearing a face covering). This differentiation is strongest in countries where the disparity between male and female wages is greater. While Blaydes and Linzer are interested in female support for Islamic fundamentalism rather than Islamist terrorist violence, their framework makes a compelling case for including gender as a study variable rather than a control.

Finally, while Littman et al. (Reference Littman, Marks and Zakayo2021) have also sought to theorize gendered differences in support for violence. They put forward a “female guardian hypothesis,” to explain why women are often observed to be aggressive toward outgroups. They contend that “women's psychology can be characterized by social vigilance and indirect aggression (exclusionary attitudes and behavior) when they perceive a communal threat.” Using data from a survey of 6,000 Nigerians they find compelling support for their theory: women were more likely to perceive a threat and those who have such perceptions are more likely to support discriminatory behaviors toward outgroups. While our data do not permit us to test for gendered threat perceptions, there is an extensive literature on the role of South Asian women policing communal boundaries often by controlling women, whose bodies are the bearer of respectability of the family and community, variously defined (Jayawardena and de Alwis Reference Jayawardena and De Alwis1996). In post-colonial South Asia, women continue to be seen as the repositories of religious beliefs and the “keepers of the purity and integrity of the community” (Basu Reference Basu, Jeffery and Basu2012, 4). While some women have mobilized against such communal politics and the salience attached to women's bodies and behaviors, others have derived power from being invigilators of the same.

This discussion prompts the following hypothesis:

The Data

We use a dataset derived from a face-to-face, nationally representative survey of 4,067 Bangladeshis, fielded in Bangladesh's national language (Bangla), by gender-appropriate teams. We conducted this survey under IRB supervision on behalf of the United States Agency for International Development (USAID) and executed by the RESOLVE NETWORK, under the auspices of the United States Institute of Peace (USIP). Two co-principal investigators (XXX and YYY) led the survey effort, developed the instrument, oversaw the quality control of the translation, identified, and worked with a highly regarded Bangladeshi survey firm (which wishes to remain un-named due to Bangladesh's political environment) to conduct focus groups about the instrument and pre-test it to ensure that it performed as expected. The instrument they developed collected demographic information for the respondents as well as their beliefs about an array of issues including religion, governance, and violent extremism. The firm conducted the survey between April 12 and 30, 2017.

The local contractor employed a stratified random sampling design that was nationally representative at division levels. Bangladesh has eight divisions. Sample ratios were 50% male and 50% female and 75% rural and 25% urban, which are in accord with the 2011 Bangladesh Population Census. The firm defined samples at the division level which were proportionate with the distribution of the population, per the 2011 Census (Bangladesh Bureau of Statistics 2015). The survey response rate was 70%, which is comparable to other surveys in Bangladesh which have a recorded response rate of 75%. The study team originally sought to sample 8,000 respondents; however, halfway through the survey effort, local authorities objected to survey questions about the Bangladesh Jamaat-e-Islami (BJeI) and insisted that they be removed as a condition of permitting the study to continue. The team discontinued further enrolments both for scientific reasons and because the distribution of the sample that had been collected was representative of Bangladesh's eight administrative divisions with reference to gender, religion, and urban/rural residence. The original margin of error for the survey with a sample size of 8,000 was about 1.10% at a 5% level of significance. The margin of error for the reduced sample was 1.54% at a 5% level of significance. While the resultant sample is smaller than planned, it is still four times larger than other publicly available surveys, including Pew's Global Attitudes Survey (Pew 2014).

Operationalization of Variables

Independent Variables

To evaluate the impacts of attending Friday prayer upon support for Islamist violence (H₁), we use a survey item (Q051) which specifically queries the respondent “How many times did you pray Namaz in congregation in the Mosque last Friday/Jumma?” We call this variable “Pray Friday Namaz?.” Answers to this question range from zero to a maximum of five. We recode this variable as 0 for any respondent who did not pray Namaz last Friday and 1 if they did pray Namaz last Friday in the mosque at least once.

To evaluate the effects of personal pietic activities, which are distinct from participation in large congregational Friday prayers in a mosque, upon support for Islamist violence (H₂), we construct an additive index for piety that necessarily excludes participation in Friday prayer. To do so, we use the following variables: Whether the respondent attends classes for understanding the Quran (Q030); the number of times the respondent prays Namaz anywhere (Q050); and whether the respondent prays “Tahajjud Namaz,” which is a non-obligatory prayer that happens in the early hours of the morning (Q052). We scaled this index to range from 0 to 1 (higher values indicate higher piety).Footnote ⁵ Note that we also confirmed that the variables that comprise the piety index are not highly correlated with our Pray Friday Namaz variable (The correlations are all under 0.2). This measure of piety reflects private practices that are in addition to attending communal Friday prayers. We created an additional piety measure (Piety2) which combines these practices along with Friday prayer, which we also scaled from 0 to 1 with higher values indicating higher piety. We use this variety for an additional robustness check discussed below.

To evaluate the impact of support for scriptural literalism and physical punishments on the one hand and support for Islamist political violence on the other (H₃), we created an additive index of support for scriptural literalism. We did so by taking the average of each response to the following binary questions: Do you favor or oppose death penalty for Bangladesh Muslims who leave the Muslim religion? (Q975); Do you favor or oppose punishments like whippings and cutting off hands for crimes like theft and robbery (Q985); Do you favor or oppose stoning people who commit adultery? (Q990); Some people say a government under Shari'a should use physical punishments to make sure people obey the law. Do you agree or disagree with this? (Q140).Footnote ⁶ We rescaled this variable from zero to one, where zero represents complete opposition to scriptural literalism and the punishments it prescribes, while one represents complete support for the same.

To assess whether there is any relationship between support for secularism and support for Islamist militancy (H₄), we created another additive index for respondent support for secular values using several survey items, namely: In your opinion how much influence should religious leaders have in matters of political governance? (Q915); and Do you favor or oppose giving Muslim leaders such as Imams the power to decide family and property disputes? (Q970).Footnote ⁷ We transformed both variables such that zero represents support for a large role of religious leaders (Q915) and complete support for giving leaders such roles and where one represents no support at all. This secularism index ranges from 0 to 1 with higher values representing more support for secularism.

Finally, to assess the effects of gender (H₅), we use a dichotomous gender variable, which is coded as 0 for males and 1 for females.

We also include three control variables. To proxy socio-economic standing, we use the educational level of the respondent as a proxy as we were concerned about the reliability of both reported income (Moore et al. Reference Moore, Stinson and Welniak2000; Meyer and Sullivan Reference Meyer and Sullivan2003) and expenditures (Meyer and Sullivan Reference Meyer and Sullivan2003; Gibson et al. Reference Gibson, Beegle, De Weerdt and Friedman2014). Survey researchers consistently report that respondents are wary of reporting actual income while household respondents may not be knowledgeable about household expenditures. We also included respondent age and whether the person lives in a rural or urban area.

The survey items used in the construction of all variables are available in the Appendix. Below, in Table 1, we present the summary statistics of all dependent, independent, and control variables. Of note is that among the three groups, support for Bangla Bhai's goals is the highest. This may be due to the fact that of the three groups, Bangla Bhai is truly indigenous and is not a partner of either al-Qaeda or IS. Also of note is that overall support for the violent means is extremely low, a full order of magnitude less than support for the goals. In Table 2, we provide a correlation matrix for our independent variables.

Table 1. Summary statistics of variables

Source: In-house tabulations of survey data.

Table 2. Correlation matrix for results in Table 1

Source: In-house tabulations of survey data.

Methods

We estimate two sets of models for each of the five dependent variables. Because women do not attend Friday prayers in our data, gender and participation in communal prayers are highly correlated (see Table 2). For this reason, in one set of models (1a–5a), we include gender while excluding Friday prayer participation. In the other set of models (1b–5b), we include Friday prayer participation while excluding gender. In both sets, our five dependent variables are: support for the goals of Bangla Bhai, support for the goals of Ansar Bangla Team, support for the goals of the IS; a combined measure of support for any of the goals of any of the three groups and a combined measure for support of the means employed by any of the three groups.

We include in each model the above-noted independent variables and controls for the age and education of the respondent and whether the respondent lives in an urban area. We estimate the models using division fixed effects for a dichotomous dependent variable using a logit link. We estimate:

$$\eqalign{\log \left({\displaystyle{{\,p_{ij}} \over {1-p_{ij}}}} \right) = & {\rm \beta }_1 \times {\rm Pray}\,{\rm Friday}\,{\rm Namaz}\,{\rm OR}\,{\rm Gender} + {\rm \beta }_2 \times {\rm Piety}\,{\rm Index} \cr & \quad + {\rm \beta }_3 \times {\rm Sharia}\,{\rm as}\,{\rm Hudood} + {\rm \beta }_4 \times {\rm Secularism}\,{\rm Index} \cr & \quad + {\rm \beta }_i \times {\rm Controls} + {\rm \alpha }_i + r_{ij}, \;} \eqno ( 1) $$

where p _ij represents support for goals or means (or the probability that the dependent variable equals 1) for individual i in district j, α_i represents the fixed effects, and r _ij represents the error term.

As a robustness check, we estimate another set of models in which we use Piety2 (an index that combines the piety index and Pray Friday Namaz), described above. In this model, we run Piety2 and gender as well as the other variables noted above. As the correlation matrix for these independent variables shows, this is justified because gender is not highly correlated with Piety2. We estimate another version of this model for females only.

We place estimates for Models 1a–5a and 1b–5b in Tables 3 and 4 below.Footnote ⁸ We place estimates for Models B1–B5 in Table B21 in Appendix B (Table B2) as well as the correlation matrices for all independent variables used in those models (Table B1). The estimates for the female-only respondent model are also in Appendix B (Table B3). For ease of interpretation, we also depict predicted probability for each variable for both sets of models. Doing so permits us to compare the impact of each variable with each other. In Figures 1a and 1b, we depict the change in predicted probabilities of each variable, when its value is taken at +/− one standard deviation away from the mean while holding all other variables constant at their mean. Figure 1a is based upon the estimations for Models 1a–5a whereas Figure 1b is based upon the estimations of Models 1b–5b.

Figure 1. (a) Changes in probability of supporting violence in models that include gender (Models 1a–5a). Source: In-house tabulations of survey data. (b) Changes in probability of supporting violence in models that include Friday prayer (Models 1b–5b). Source: In-house tabulations of survey data

Table 3. Estimates for Models 1a–5a

Standard errors in parentheses.

***p < 0.001, **p < 0.01, *p < 0.05.

Table 4. Estimates for Models 1b–5b

Standard errors in parentheses.

***p < 0.001, **p < 0.01, *p < 0.05.