Faith and Friendship: Religious Bonding and Interfaith Relations in Muslim Countries

Abstract Studies have documented more negative attitudes and a higher level of social hostilities toward religious minorities in Muslim than in non-Muslim countries. I seek to explain what contributes to these poor interfaith relations. Diverging from the mainstream approaches that focus on cultural, institutional, or psychological explanations, I argue that the poorer interfaith relations in Muslim countries are driven by high levels of religious bonding or religiously homogeneous friendships among Muslims in these countries. Analyzing a global survey of more than 17,000 Muslims and a report documenting how religious groups in a country restrict or discriminate against each other, I show that religious bonding is related to more negative attitudes toward religious minorities, that a country's level of religious bonding is positively related to its level of social hostilities, and that religious bonding is indeed higher among Muslims in Muslim countries than among Catholics in Catholic-majority Latin American countries.

What, then, contributes to these poor interfaith relations? Diverging from the mainstream approaches that focus on cultural-theological, institutional, and individual-level antecedents, I argue that the more negative interfaith relations in Muslim countries are driven by the religiously homogeneous friendship networks of Muslims in these countries. Drawing from two cross-national surveys, I show that Muslims in Muslim countries have higher levels of religious bonding or ties with co-religionists than non-Muslims in non-Muslim countries do. These homogeneous networks are theorized to affect interfaith relations through reinforcement of group identity and confirmation bias.
The paper is structured as follows. I first review studies on interfaith relations in Muslim countries, highlighting a gap in the quality of interfaith relations within Muslim countries compared to non-Muslim countries. This is followed by a discussion on how other scholars have explained this gap. I then highlight how networks matter and support the argument by presenting three pieces of evidence rooted in the social networks literature.
First, I show that levels of religious bonding among Muslims, i.e., Muslims' ties with fellow Muslims, significantly predict more negative attitudes toward non-Muslims. Second, I show that religious bonding is positively related to the social hostilities index (SHI) at the country level-a measure of ways in which individuals and groups in a society infringe upon religious beliefs and practices (Pew Research Center 2018, 1). Third, I show that Muslims in Muslim-majority countries indeed have more religiously homogeneous friendship networks compared to Catholics in Catholic-majority Latin American countries.

INTERGROUP AND INTERFAITH RELATIONS IN THE MUSLIM WORLD
The current study focuses on interfaith relations broadly defined. At the individual level, I relate interfaith relations to individuals' attitudes toward religiously different others. At the country level, interfaith relations are defined in the context of social hostilities or how the majority religious group persecutes or discriminates against religious minorities. Because the focus is on interfaith relations or social discrimination, the current study is therefore different from studies that focus on state discrimination of religious minorities (Fox 2016).
A number of studies have examined interfaith relations in Muslim countries. Some focus on socioeconomic or cultural factors (Ojo and Lateju 2010;Menchik 2016;Ventura 2018) whereas others look at how actors or institutions shape the patterns of interfaith relations (Scott 2010). Muslim-Christian relations command the most attention, but there are also studies that examine Muslim-Buddhist (Stewart 2014) or Muslim-Hindu relations (Varshney 2001), indicating scholarly and public interests in these topics. Although these studies have contributed valuable insights, most focus on single countries and rely on qualitative case studies. Such an approach often provides a rich picture of interfaith dynamics in the country studied but offers few clues as to how these dynamics vary across countries. Comparative studies of interfaith or intergroup relations are relatively rare. 1 Studies that engage in a comparative approach have found an intriguing pattern of interfaith relations across countries, documenting less positive intergroup relations in Muslim than in non-Muslim countries. Inglehart (2003, 56; also see Spierings 2014) finds that levels of trust and tolerance in Muslim countries are generally lower than ones in more established democracies. Bohnet, Herrmann, and Zeckhauser (2010) find lower trust between individuals and between groups in the Gulf than in Western countries. Gu and Bomhoff (2012) arrive at the same conclusion when comparing Muslim-majority to Catholic-majority countries. Muslims are found to be less trustful of people of another religion and of another nationality, and are more likely to believe that ethnic diversity erodes a country's unity.
There is also evidence that these differences between Muslims and non-Muslims hold up outside Muslim countries. Djupe and Calfano (2012) find that Muslims in the United States are more intolerant of people who commit acts against religion. Comparing Muslims and Christians in six Western European countries, Koopmans (2015) finds higher hostilities toward homosexuals and Jews among Muslims than among Christians ( p. 47). Analyzing the World Values Survey, Milligan, Andersen, and Brym (2014) find that Muslims have higher objections than non-Muslims do when it comes to having religiously different neighbors. Gu and Bomhoff (2012) arrive at the same conclusion when comparing individuals in Muslim-and Catholic-majority countries.
The poorer interfaith relations in Muslim countries are also evident when we look at measures of social hostilities (Pew Research Center 2018). Based on Grim and Finke's (2006) study, the SHI is concerned with religious restrictions imposed by the society (including individuals and social groups) as distinct from the government. Rather than attitudes, the index reflects incidents in which religious groups (predominantly the majority groups) discriminate against other groups (predominantly the minority groups). Figure 1 compares levels of social hostilities in Muslim and non-Muslim countries between 2007 and 2016. As the figure shows, Muslim countries consistently have higher social hostilities than Christian-majority countries or non-Muslim countries in general. At the same time, it also shows a degree of social hostilities in non-Muslim countries. This suggests that minority groups everywhere continue to face discriminations, albeit to varying degrees. The findings presented here, therefore, must be understood in the context of a comparison as opposed to an argument of the superiority of one culture or religion over the others.

EXPLAINING INTERFAITH RELATIONS
Three major perspectives have been widely used to understand interfaith relations. The first two views-cultural-theological and institutionalfocus on Muslim societies; whereas the third focuses on individual-level antecedents.

Cultural-Theological
The cultural-theological perspective includes approaches that emphasize the importance of Islamic theology (Huntington 1997) and the effects of modernization on societal values (Inglehart and Welzel 2005). These approaches contend that Muslims' lack of openness to other faiths can be traced back to the absence of certain liberal values in the society. The theological approach looks at Islamic doctrines and argues that the all-encompassing nature of the Qur'an (March 2015, 106) encourages Muslims to be obedient to religious authority. Such a sweeping perception overlooks the reality that Muslims across the world live and understand their religion differently (Sadowski 2006). A theological approach that attributes poor interfaith relations in Muslim countries to the Qur'an may inadvertently give legitimacy to a certain view in Islam that is intolerant and discredit the more tolerant views as illegitimate.
The modernization theory regards culture as important but adds another explanatory variable: socioeconomic development. Higher socioeconomic development leads to cultural changes, one of them being more positive intergroup attitudes. An obvious limitation of this theory is that it cannot explain the Gulf countries that are less tolerant than one would expect given their levels of development. That is why Inglehart and Welzel (2005, 19) regard cultural starting points as insurmountable, positing that "The fact that a society was historically Protestant or Orthodox or Islamic or Confucian manifests itself in coherent cultural zones with distinctive value systems that persist even when one controls for the effects of socioeconomic development." This muddles what we should expect regarding the effects of modernization on interfaith relations in Muslim countries. As Muslim societies become more modernized, should we expect interfaith relations in the countries to be at least as strong as those in non-Muslim countries, or should we expect them to improve only relative to their state in the past?

Institutional
Two institutional factors are of interest in the context of the current study: secularism and political competition. These factors shape interfaith relations through a similar mechanism, namely open competition. The more open a polity is, the greater the opportunity for citizens to encounter diverse viewpoints. This exposure to diversity, along with the learning process individuals experience in responding to that diversity, in turn helps to internalize tolerance in the society (Peffley and Rohrschneider 2003). The level of secularism is concerned with how well separated the state and religious institutions are. 2 A state's neutrality in religious affairs opens up space for religions to compete in attracting believers (Iannaccone 1998). The resulting religious life is more vibrant with believers actively engaging each other. A strict regulation of religion, on the other hand, may privilege certain faiths over others or privilege believers over unbelievers (Grim and Finke 2010). It would be difficult for citizens to tolerate each other when even the government demonstrates through its laws that not all citizens are equal. As such, the poor interfaith relations in Muslim countries are driven by laws that advantage Islam over the other faiths.
The logic is the same with the perspective that emphasizes political competition. The perspective suggests that Islamists become more moderate the more they participate in electoral competition (see Schwedler 2011, for a review). Electoral incentives induce Islamists to moderate their platforms to attract more votes and political reality forces parties to compromise. According to this perspective, then, the poor interfaith relations in Muslim countries are due to the lack of political competition, which means that parties have few incentives to build inclusive coalitions.
What both perspectives miss, however, is that although free competition can induce moderation, it can also induce religious populism (Tanuwidjaja 2010;Lorch 2019). As opposed to moderating themselves, parties may choose to outflank each other. Another limitation of the institutional explanations is that their level of analysis makes them more suited to explain country-level differences than individual-level ones. They are mute when it comes to answering why different individuals in the same country have different interfaith attitudes despite being exposed to the same institutional arrangements. The social relationships-based argument that I propose, on the other hand, is capable of explaining variations in interfaith relations at both the individual and country levels.

Individual-Level Antecedents
In addition to the aforementioned country-level factors, individual-level variables also affect how individuals approach intergroup relations. Three of these variables have been heavily studied (Sullivan and Transue 1999). The first relates to cognitive ability and includes factors such as education and political expertise. More politically sophisticated people tend to be more tolerant (Fish 2011), presumably because they are exposed to dissonant views more often and have a better internalization of democratic values.
The second variable is threat perception (Marcus et al. 1995;Muluk, Sumaktoyo, and Ruth 2013). Individuals are more likely to have positive attitudes toward an outgroup if they perceive the group as posing little or no threat to them or their way of life. Politicization of social cleavages can affect this threat perception. When certain groups are portrayed as ancient enemies, members of each group would perceive the other group as threatening, making intolerance more likely.
Third, in terms of personality type, negative intergroup attitudes are linked to closed-mindedness and dogmatism (Anderson and Koc 2015). These predispositions increase the likelihood of individuals conforming to established social norms while rejecting dissenting views. Since religiosity involves a strong adherence to beliefs (Saroglou 2002), these predispositions can also explain why religious people tend to have more negative interfaith attitudes (Gibson 2010). Thus, Muslim societies' high levels of religiosity might explain their poor interfaith relations.

WHY SOCIAL TIES MATTER
The aforementioned approaches have contributed to our understanding of interfaith relations and intergroup relations in general. However, as discussed above, they have certain limitations in their explanatory powers. They also are incomplete because they overlook an important component of social life: ties among the individuals themselves. Social ties, including religious ties, affect not only the individuals' political behavior (Mckenzie 2004;Lewis, MacGregor, and Putnam 2013) but also how the society functions (Putnam 1993(Putnam , 2000Brehm and Rahn 1997;Coleman 1988).
Not all social ties are equal, however. Ties that are inward looking (bonding relationships, i.e., relationships with ingroup members) are more likely to produce undesirable effects on intergroup attitudes compared to ties that are outward looking (bridging relationships or relationships with outgroup members). There are three reasons why bonding relationships may be related to more negative evaluations of an outgroup and, by implication, why religious bonding may be related to poor interfaith relations.
The first relates to the flow of norms and information (Huckfeldt and Sprague 1987). Individuals receive political and social cues from their surroundings (Berelson, Lazarsfeld, and McPhee 1954;Granovetter 1973). Psychological research on social learning theory (Bandura 1976) also suggests that one's social environment shapes what are considered appropriate or inappropriate behavior and attitudes. As such, the higher one's level of bonding relationships, the more likely it is for the social network to facilitate the flow of norms and information that are favorable to the ingroup and unfavorable to the outgroup (Huckfeldt, Mendez, and Osborn 2004).
A second mechanism involves ingroup identity. In an experimental setting Levendusky, Druckman, and McLain (2016) show that interactions with politically similar others enhance partisan identity. Coupled with preferential treatments for ingroup members, such an ingroup attachment may lead to rejection of the outgroup (Brewer 1999). In the religious context, religious social identity has been shown to lead to the rejection of religious outgroups (Rhodes 2012;Ben-Nun Bloom, Arikan, and Courtemanche 2015).
The third reason why bonding may lead to negative intergroup attitudes concerns bridging relationships. Since maintaining relationships is costly in that one has to devote time to one's friends (Wellman et al. 1997, 36) or one's group (Campbell 2004), it is likely that there would be a trade-off between bonding and bridging. The more the resources that one devotes to one's ingroup, the less that one can devote to one's relationships with outgroup members.
Bridging relationships, on the other hand, have been linked to tolerance and acceptance of outgroups (Allport 1954;Cigler and Joslyn 2002;Mutz 2006;Ikeda and Richey 2009;Harell 2010;Putnam and Campbell 2010;Rapp and Freitag 2015). Positive interactions with people from a different group reduce anxiety about the group and increase the ability of individuals to empathize with and understand the other group's viewpoints (Pettigrew and Tropp 2008). That high levels of bonding may take up the time and resources needed to develop bridging relationships might, in turn, hinder the developments of these positive effects.
Based on the preceding review, I argue that one of the drivers for poor interfaith relations in Muslim countries is the countries' high levels of religious bonding. Supporting this argument requires three pieces of evidence. First, given the review above about how social ties shape attitudes, I need to show that religious bonding is related to more negative interfaith attitudes.
Hypothesis 1 (Negative Attitudes): At the individual level, greater religious bonding is associated with more negative attitudes toward members of religious outgroups.
Second, I need to show that religious bonding is also positively related to a more behavioral measure of interfaith relations, namely a country's level of social hostilities. Studies have shown how attitudes should shape behavior especially if they are salient and involve one-sided information (Glasman and Albarracín 2006). As bonding keeps religious identity salient and influences the type of information individuals are exposed to, it is plausible that religious bonding is related not only to more negative interfaith attitudes but also to higher levels of social hostilities.
Hypothesis 2 (Social Hostilities): At the country level, greater religious bonding is associated with higher levels of social hostilities.
Finally, I need to show that religious bonding in Muslim countries is indeed higher than in non-Muslim countries. This is a plausible hypothesis considering the central roles of religious organizations and mosque networks in shaping the political dynamics of Muslim countries (Sadowski 2006). These organizations arguably are more conducive to bonding than bridging. Furthermore, the idea of a global ummah (Roy 2006) that envisions Muslims around the world as one big family may also lead to higher bonding among Muslims.
Hypothesis 3 (Higher Bonding): Religious bonding is higher in Muslim than in non-Muslim countries.
DATA I analyze data from two surveys. The first is a study of "The World's Muslims" (TWM; Pew Research Center 2013), which was conducted between October 2011 and November 2012 and surveyed 32,604 Muslims in 26 countries. I excluded Bosnia-Herzegovina, Russia, and Thailand from the analysis because they are not Muslim majority (Pew Research Center 2014b).
The second dataset is a survey of "Religion in Latin America" (RILA; Pew Research Center 2014a) that covers 30,326 respondents in 19 Latin American countries and was conducted between October 2013 and February 2014. This dataset is primarily used to enable a comparison between Muslims and non-Muslims necessary for testing the Higher Bonding hypothesis.
I focus on 17 countries that are Catholic majority and on respondents who were Catholics-Catholics in Catholic-majority Latin American countries. This criterion is justified in light of my goal to find a comparison for Muslims in Muslim-majority countries. "Catholics" is relatively 642 Sumaktoyo well defined compared to "Protestants" or "Evangelicals" that encompass different denominations (Steensland et al. 2000). There is also a practical reason in that the TWM and RILA surveys are the only comparative surveys that include a question on social ties. Both surveys employed face-to-face interviews and were designed to be nationally representative.
Wording for all the questions below is available in the online Appendix, along with sample size from each country.

A NOTE ON CAUSATION
This study analyzes observational data, which means that some limitations commonly found in an analysis of such data also apply here. The first concerns causal ordering. It is possible that a relationship between bonding and interfaith attitudes is driven by self-selection or people with negative interfaith attitudes being more likely to befriend religiously similar others in the first place. Second, unlike in an experimental study where randomization voids the effects of confounding factors, both observed and unobserved, an analysis of observational data relies on the observables. Only those that are measured can be controlled. I employ three strategies to address these concerns. First, I rely on insights from previous studies that have demonstrated the causal effect of social relationships on attitudes and behavior (Van Laar et al. 2005;Lazer et al. 2010;Putnam and Campbell 2010). These insights suggest that self-selection, although possible, does not constitute the whole story. Even if people build their networks based on attitudinal similarities, these networks would still have an effect on their attitudes. The homogeneous friendship networks of Muslims in Muslim countries, in turn, should lead to more negative interfaith attitudes, even if these homogeneous networks are in part a product of negative interfaith attitudes. Second, I leverage the temporal ordering of the variables in the Social Hostilities hypothesis. Here, the dependent variable was measured 1 year after the independent variable. This correct temporal ordering provides us with a stronger causal identification. Finally, I take alternative explanations into account, as far as the data allow, by including an extensive set of control variables in all of the analyses.
I nonetheless acknowledge that without experimental data it is impossible to completely rule out the possibilities of reverse causation and spurious correlations. As such, I position the current study as a pioneering examination of a relationship between bonding and interfaith relations Faith and Friendship 643 in Muslim countries. By proposing a theory grounded on the literature on social networks and political behavior, and by testing this theory using the best available data, I aspire to make a case for the plausibility of the theory as well as encourage further research on the topic.

Dependent Variable
To test the Negative Attitudes hypothesis, I relied on three questions from the TWM survey. The first two questions tap into interfaith marriage and asked respondents "How comfortable would you be if a son (daughter) of yours someday married a Christian?" I recoded the responses into a 5point scale ranging from "very comfortable" to "not at all comfortable." The third question taps into perception of similarity and asked respondents "From what you know, do you think that the Muslim religion and the Christian religion have a lot in common, or do you think that the Muslim religion and the Christian religion are very different?" Possible responses were "have a lot in common" (coded as "0") and "are very different" (coded as "1").
A factor analysis based on the variables' polychoric correlation matrix reveals that these questions have a unidimensional structure that explains 64% of the variance. The eigenvalues are 1.93, 0.90, and 0.17; and the loadings for son intermarriage, daughter intermarriage, and perception of similarity are 0.94, 0.93, and 0.43, respectively. The correlation between the two intermarriage variables is r = 0.83, and the correlations between son and daughter intermarriage on the one hand and perception of similarity on the other are r = 0.23 and r = 0.19, respectively. Given this evidence for unidimensionality, I use as the dependent variable the factor score generated from these loadings by the regression method. The method calculates each respondent's factor score by multiplying their responses on the three interfaith items with the corresponding items' factor loadings and then taking the sum of the resulting values. 3 As the reader may have noticed, the interfaith variables specify Christians as the target group. Does the relationship apply to other minority groups? Although this is an understandable concern, specifying Christians as the target group should not harm the theory's generalizability for two reasons. First, Gibson (1992) finds that an a priori specification of the target group affects studies that focus on descriptive statistics (e.g., how many people have positive attitudes toward a particular group) but not studies that examine antecedents of intergroup attitudes. Second, although religious conflicts are rarely purely religious, several conflicts between Muslims and Christians carry with them strong religious sentiments (Ayoub 1989), making Muslims-Christians relations an important topic to study. 4 The importance of this topic becomes even more obvious once we consider how the war on terror is sometimes framed as a Christian West's war on Islam.

Independent Variable
As the independent variable, I use a question that asked respondents the level of religious bonding in their friendship networks. 5 The question read, "How many of your close friends are Muslim?" Possible responses were "none of them," "hardly any of them," "some of them," "most of them," or "all of them." Higher scores indicate higher levels of religious bonding.
Two concerns regarding the variable are worth discussing. The first relates to the variable not capturing network density or how well the friends themselves know each other (Everton 2018). This means that I cannot examine whether the effects of bonding on interfaith attitudes are constrained to dense religious networks or are more generalizable across levels of network density. On the one hand, it is possible that dense social networks more strongly reinforce group identity than sparse networks do through the development of strong ties. On the other hand, it is possible that sparse networks more effectively facilitate the flow of information that can advantage the ingroup or put the outgroup in negative light (Granovetter 1973). Future studies will benefit from exploring the plausibility of such interactions by collecting and analyzing whole network data.
The second concern relates to whether the independent variable itself is a measure of interfaith attitudes like the dependent variables. If that was the case, then a relationship between bonding and interfaith attitudes would be given. There are three reasons to argue why this is not the case. First, from the wording alone, we can see that the dependent variables measure perceptions or attitudes and the independent variable measures network composition. Specifically, the bonding question is an example of a global estimate measure of social network (Eveland, Hutchens, and Morey 2013).
Second, a connection between social relationships and attitudes is not given and subject to various conditions (Pettigrew 1998). This means that the question of whether or not religious bonding would lead to more negative interfaith attitudes is also an empirical one. This is particularly true among Muslims as some scholars have argued that religion would be the primary driver of Muslim political behavior (Huntington 1997).
Finally, even if the concern was plausible at the individual level, it would be less likely so at the country level where the dependent variable is the level of social hostilities. As mentioned above, the SHI captures actual incidents of social discriminations and is based on a coding of various human rights reports and news sources. As such, it is not an attitudinal measure and is even more conceptually different from religious bonding than the interfaith attitudes measures are.

Control Variables
To account for alternative explanations, I included in all individual-level models nine individual-level and four country-level control variables. The country-level models include only country-level covariates. The individual-level covariates include age, gender, whether living in a rural or urban area, education, political efficacy, personal economic condition, support for democracy, frequency of prayer, and believing that Islam is the one true faith. Education is intended to capture cognitive sophistication and is coded as less than high school, at least high school, or at least college degree. Political efficacy is intended to capture political interest and involvement, which should be related to more positive attitudes toward diversity (Galston 2001). Personal economic condition serves as an indicator for existential security (Norris and Inglehart 2012), which should be positively related to outgroup attitudes. Support for democracy was measured with a binary variable that asked respondents whether they thought that the people should rely on a democratic government or a strong leader to solve the country's problems. Finally, frequency of prayer and believing that one's own religion is the one true faith tap into different types of religiosity. The first relates to piety, whereas the latter is more about orthodoxy and theological exclusivity. The two may have different consequences when it comes to how individuals translate their faiths into actions (Muluk, Sumaktoyo, and Ruth 2013). Specifically, Merino (2010) finds that theological exclusivity among Americans predicts less acceptance of Muslims and Hindus. The four country-level covariates include logged GDP per capita; religious diversity index (RDI; Pew Research Center 2014a); government restrictions on religion (GRI; Pew Research Center 2018) 6 ; and Polity IV score. All variables except for the RDI are from 1 year before the survey (i.e., the 2010 values for the Muslim countries and the 2012 values for the Catholic countries). The RDI is from 2010 as this is the only data available.
GDP per capita is used as an indicator of modernization corresponding to the modernization explanation outlined above. The RDI, based on the Herfindahl-Hirschman formula, measures the relative sizes of religious traditions in a country. It takes into account not only the size of the majority but also the number of minority groups and their sizes. The index is coded so that higher values represent more diversity. Finally, the GRI and Polity IV scores capture the effects of institutional factors. Interfaith relations in a country should be negatively related to how tightly the government regulates religion and positively to its level of democracy.

Analysis and Results
I employed a multilevel linear regression to test the hypothesis, which is appropriate because respondents are nested within countries and respondents within the same country are likely more similar to each other than they are to respondents from other countries. 7 I standardized all nonbinary variables to have mean zero and standard deviation one.
The regression coefficients are presented in Table 1. The table supports the prediction that higher levels of religious bonding are related to more negative interfaith attitudes. The coefficient of religious bonding is also one of the largest among the individual-level predictors, comparable to how often one prays and only weaker than believing that Islam is the one true faith.
Higher education and better personal economic condition are related to less negative interfaith attitudes. This is consistent with earlier observations about how education is positively related to tolerance (Sullivan and Transue 1999). Religiosity and belief orthodoxy, on the other hand, are related to more negative interfaith attitudes, affirming previous findings on the importance of religiosity and theological exclusivity in shaping negative interfaith attitudes (Gibson 2010).
The positive relationship between political efficacy and negative interfaith attitudes is likely related to the high levels of government restriction Faith and Friendship of religion in Muslim countries. Those with high efficacy are more likely to be interested in politics, immersing themselves in a system that generally favors Islam at the expense of other religions. As opposed to inoculating in them a respect for diversity, this activism does the opposite. This highlights the complex relationships between political system, political participation, and political attitudes (Jamal 2007). In a secular democracy, political activism might lead to more positive interfaith attitudes. Activism in a less democratic, less secular societies, on the other hand, might inoculate just the opposite attitudes. In terms of country-level covariates, government restriction on religion is the only statistically significant variable. The higher the level of government restriction in a country, the more negative the interfaith attitudes of respondents in the country are. This highlights the importance of institutional settings in shaping individuals' attitudes (Peffley and Rohrschneider 2003;Dunn and Singh 2014). 8

THE SOCIAL HOSTILITIES HYPOTHESIS Variables
The Social Hostilities hypothesis extends the Negative Attitudes hypothesis by showing that the relationship between religious bonding and negative interfaith relations holds up at the country level. The dependent variable in this analysis is the SHI score published by the Pew Research Center (2018). I use the score from the year after the survey ended (i.e., the 2014 score for the Muslim countries and the 2015 score for the Catholic-majority Latin American countries). The independent variable is the country-level score of religious bonding, calculated as the average of the levels of religious bonding of respondents in the country, accounting for sampling weight. This operationalization provides a stronger causal identification by establishing the correct temporal order in which the independent variable (country-level bonding score) was measured before the dependent variable (the country's social hostilities score 1 year after the survey). As controls, I use the same country-level covariates as in the Negative Attitudes hypothesis.

Analysis and Results
I employed ordinary least squares (OLS) regression models to test the hypothesis, standardizing all variables to have mean zero and standard deviation one. Table 2 presents two models that test the hypothesis. The first model regresses the SHI at the country-level covariates. The second model uses the same specification but adds the SHI score from the year before the survey as another control variable. If religious bonding still has a significant effect on the SHI even after controlling for the previous SHI score, it would suggest that the variable has a unique explanatory power on the SHI that is unaccounted for by either the stability of social hostilities or other variables controlled in the model.
In model 1, we see that a country's level of religious bonding has a positive relationship with social hostilities (η 2 = 0.211). The more religiously homogeneous the friendship networks of respondents in a country, the higher the country's level of social hostilities is 1 year after the survey. The importance of government restriction on religion is again evident from its relationship with social hostilities (η 2 = 0.146). Countries with more religious restrictions have higher social hostilities. Finally, the model also suggests a positive relationship between level of democracy and social hostilities (η 2 = 0.301). As the countries' levels of democracy were relatively stable in the years of the study, this relationship is unlikely to be driven by time-specific dynamics and might reflect the more open and less repressive nature of more democratic governments. This greater openness provides more space for dissents, differences, and to an extent conflicts to be more overtly expressed. Model 2 shows that including the lagged score of SHI weakens the explanatory powers of all the variables. However, the model also shows that religious bonding is the only predictor in the model that continues to maintain its statistical significance, albeit only marginally ( p = 0.057). This suggests that although religious bonding's relationship with social hostilities is in part due to its shared variance with the lagged social hostilities ( just like the other variables' relationships are), it has sufficiently unique variance that is not due to the previous level of social hostilities. In fact, religious bonding's partial eta-squared (η 2 = 0.120), an indicator of a predictor's unique shared variance with the dependent variable, is second only to that of the lagged SHI (online Appendix S8).
This unique variance of religious bonding is at the core of this study. The current study does not argue against the possibility of a reverse causation. Rather, what the current study contends is that, just like previous studies have shown, reverse causation is unlikely to fully account for the effect of religious bonding. Religious bonding would still have a unique effect on interfaith attitudes even if interfaith attitudes also shape the composition of one's friendship network. This unique effect, in turn, should contribute to the gap in the quality of interfaith relations between Muslim and non-Muslim societies.

Variables
Having shown how religious bonding predicts more negative interfaith attitudes at the individual level and more negative interfaith relations at the country level, the next exercise is to test whether Muslim countries indeed have higher bonding than non-Muslim countries, which in this case are represented by Catholic-majority Latin American countries. In this hypothesis, the dependent variables are the individual-and the country-level religious bonding scores whereas a dummy variable that indicates whether a country is Muslim-or Catholic-majority becomes the independent variable. I include the same individual and country level covariates as in the previous two analyses. Catholic-majority countries. The x-axis represents the level of religious diversity and the y-axis represents the level of religious bonding. The figure shows that Muslim countries almost consistently have higher bonding than Catholic-majority countries with comparable shares of Catholics.

Regression Models of Religious Bonding
I run a multilevel ordinal regression and an OLS model to more formally test the hypothesis at the individual and the country levels, respectively. Table 3 presents results from the models. Models 1 and 3 include all countries that have the required variables, whereas Models 2 and 4 exclude Muslim countries that are less religiously diverse than the least religiously diverse Catholic country. This exercise tests whether the Higher Bonding hypothesis is simply driven by extremely religiously homogeneous Muslim countries.
As Table 3 shows, the Muslim dummy is statistically significant in all the models. Muslim countries indeed have higher levels of religious bonding than the Catholic-majority Latin American countries. In fact, whether or not a country is Muslim majority is the strongest predictor for the levels of religious bonding among its respondents, stronger even than the country's level of religious diversity (see online Appendix S9 for effect size).

Addressing Compatibility Concerns
Notwithstanding that the RILA survey is the only available data to use as a comparison for the Muslim countries, the curious reader might raise concerns over three issues: sectarian divide among Muslims, theological distance between the religious traditions, and characteristics of the countries studied.
The concern about sectarian divide is related to the notion that Muslims may not be a homogeneous group but rather belong to different traditions with the two largest being Sunni and Shia. This divide may influence how the respondent answered the bonding question. This is an interesting concern but it should not harm the conclusion of the current study for Second, the notion of sectarian divide actually works in favor of the current study as it makes regression models in Table 3 a conservative test of the Higher Bonding hypothesis. This is because a sectarian divide should increase the levels of religious diversity in the Muslim countries, making bridging more likely. That the current study does not take into account the possibility of a sectarian divide, therefore, means that it offers only a low-end estimate of the gap in the levels of bonding between the Muslim-and the Catholic-majority countries.
The concern about theological distance relates to religious compositions in the countries studied. Broadly speaking, the second largest religious tradition in the Muslim countries is Christianity, whereas in the Catholic countries it is not Islam but either Protestantism or Evangelicalism, which are other traditions of Christianity. Maybe it was easier for the Catholic respondents to befriend non-Catholics because it would mean befriending other Christians.
To examine this possibility we can look at the variable that asked respondents whether they thought that "the Muslim religion and the Christian religion" (the TWM survey) or "the Catholic religion and the  Thus, the theological distance argument is not supported by the data. Finally, the concern over country characteristics is related to whether it is possible to compare the Muslim-and the Catholic-majority countries given how different these countries are. To this concern, I would point out that an empirical comparison between countries is not uncommon and they often spark interesting debates in the literature (Fox and Sandler 2005;Rizzo, Abdel-Latif, and Meyer 2007;Gu and Bomhoff 2012). I also already acknowledge this concern by including in the regression models individual-and country-level covariates. As the preceding subsection has shown, even after these factors are taken into account, whether or not a country is Muslim majority is the strongest predictor for its level of religious bonding.

DISCUSSION
I argue that one of the main reasons why Muslim countries have poor interfaith relations is because of the religiously homogeneous friendship networks of Muslims in these countries. At the policy level, I aspire that this study would motivate policymakers, civil society organizers, and other concerned stakeholders to look beyond essentialist arguments that blame Muslim countries' poorer interfaith relations on Islamic theology.
Tolerance and interfaith attitudes, as the current study has shown, are shaped by social interactions, which in turn can be shaped through policies and activities that foster cooperation between religious groups. These policies and activities can take many forms, such as diversity-oriented curriculums (Rockenbach et al. 2015) or grassroots-level community meetings. Muwahidah (2008), for example, finds interfaith cooperation can be fostered through secular community meetings that are originally intended to solve land-dispute problems.
At the scholarly level, I hope that this study motivates scholars of the Muslim world to pay more attention to social relationships in explaining Muslims' political behavior. A significant improvement in this direction would be for major comparative surveys to include questions on social relationships in their questionnaires. As more data become available we will be able to gain more insights into the mechanisms and contexts that may amplify or mitigate the effects of religious bonding or bridging.
Increased data availability will also enable us to inquire about the antecedents of religious bonding. What explains Muslim countries' high levels of bonding? Or, more generally, what shapes the levels of bonding and bridging relationships in a society? To be clear, bonding is generally the norm when it comes to social relationships. Humans are more likely to form relationships with similar than dissimilar others (McPherson, Smith-Lovin, and Cook 2001). There is little reason, however, to argue that something unique about Islam's theology makes its followers more predisposed to religious bonding than followers of other religions.
Two hypotheses, briefly outlined above in the presentation of the Higher Bonding hypothesis, are particularly fruitful to explore. The first is political. Political Islam is driven more by social movements as opposed to political parties competing in an institutionalized system (Sadowski 2006, 226). Informal networks such as volunteer groups and mosques are important in enabling these movements to survive repression from the authoritarian state, the type of state many Muslims still live under today (Wickham 2002). Mosques, in particular, are more than a place of worship as they also serve as a center of political activities. Although a church can also be a nexus of political activities (Djupe and Gilbert 2008), these activities are different in that they revolve around advocacy on social issues and are not about changing the regime.
How social movements shape bonding or bridging therefore is related to what these movements are organized for. Movements that place themselves in opposition to the regime probably promote bonding more than bridging. That religious movements in Muslim countries are more political than religious movements in non-Muslim countries may contribute to the countries' higher levels of bonding.
The second hypothesis is social-psychological. Studies have discussed how some Muslims adhere to the idea of a global ummah (Roy 2006). The notion that all Muslims are one big family is unique in that it goes beyond the traditional concepts of nations, countries, and borders. Although claimed to be based on the Qur'an, the idea is largely political, driven by transnational Islamic organizations adhering to pan-Islamism and opposed to the West-dominated globalization (Roy 2006;Saunders 2008).
It is different from Catholics' relationship with the Pope in that a global ummah concerns both cultural and political matters whereas the papacy is largely a cultural symbol with no jurisdiction over political matters in Catholic countries. It is also different from Judaism in that Judaism is 656 Sumaktoyo more confined to a specific geographic area, compared to an ummah that transcends borders and continents. A perception that all Muslims are family is arguably more conducive to bonding than bridging. This effect should be more pronounced if the individuals believe that the family is under attack either militarily through "the war on terror" or culturally through globalization.

Supplementary Material
To view supplementary material for this article, please visit https://doi.org/ 10.1017/S1755048320000589.

Data
Data and code necessary to replicate the findings in the manuscript is available from the author's website (http://www.nathanael.id).