______
Who’s Allowed in Your Lifeboat?
How Religious Identity Altered Life-Saving Priorities in Response to COVID-19
MATTHEW R. MILES AND JUSTIN A. TUCKER
Not only do Republicans and Democrats disagree about politics; the more ardent partisans express disdain for supporters of the other party. These negative feelings move beyond the political realm into hopes that their children will not marry someone from the opposing political party and beliefs that members of the opposing party are less intelligent and more selfish (Iyengar et al. 2019; Iyengar and Hahn 2009; Iyengar, Sood, and Lelkes 2012; Iyengar and Westwood 2015). One cause of this interparty antipathy is that social identification with a party motivates emotional responses to information. Negative information about the out-group is more easily accepted than positive information (Huddy 2001; Tajfel and Turner 1979). In the United States, some religious groups are closely aligned with political parties (Campbell 2020), which exacerbates in-group and out-group dynamics. Social identity theorists posit that the more salient the group affiliation, the more biased an individual’s belief about out-group members will be. As such, it is possible that what appears to be partisan hatred for the other side may simply be another manifestation of religious divisions that spill over into politics. Strong religious identifiers may associate the opposing political party with members of religious groups they dislike. If so, expressing concern that one’s child might marry a Democrat could simply be an expression of concern that their child might marry outside of their religion. Is the coronavirus pandemic any different? Do people prioritize religion over politics in the pandemic?
In this chapter, we exploit an opportunity presented by the COVID-19 pandemic to test the extent to which individuals prioritize saving certain individuals. We embedded a conjoint experiment into a national survey conducted in the middle of June 2020—the point at which U.S. governors were trying to weigh whether their states should reopen while novel coronavirus cases seemed to be on the rise. By experimentally manipulating numerous demographic traits simultaneously and asking participants to indicate which individual they would prioritize saving if the choice were theirs alone, we can estimate the precise amount of interparty hostility that is the result of religious difference. In the case of COVID-19, we find that differences in religious identity explain more of the hostility than do partisan identity differences.
Religious Social Identity and Out-Group Antipathy
The dynamics we are exploring in this chapter fit under the umbrella of political tolerance—putting up with those whom you dislike, perhaps even detest. What causes political intolerance in the United States? Some argue that differing religious beliefs are the primary cause of intolerance (Eisenstein and Clark 2015; Gibson 2010). Other scholars find that belonging and behaving have a strong influence on political intolerance. Religious competition for adherents motivates churchgoing Americans to be less tolerant of nonbelievers (Cox, Jones, and Navarro-Rivera 2015). Others emphasize the importance of values and traits. Individuals who feel disgust are less tolerant (Ben-Nun Bloom and Courtemanche 2015), and because people with exclusive religious values are more likely to feel threatened by religious out-groups (Schaffer, Sokhey, and Djupe 2015), people with religious values that emphasize distinctiveness are less tolerant than those who value religious inclusiveness (Djupe and Mockabee 2015). People with exclusive religious values are more likely to feel threatened by religious out-groups, which in turn motivates greater intolerance toward those groups (Schaffer, Sokhey, and Djupe 2015).
Yet scholars also note that depending on how political tolerance is measured, religious beliefs may not be the primary influence motivating political intolerance (Eisenstein and Clark 2015). Perceptions of threat are consistently the strongest (and least understood) predictors of political intolerance (Gibson 2006). Because threat is a multidimensional concept, Gibson argues that it can be difficult to isolate the conditions under which people perceive a threat. Yet perceived group-level threats—for example, the belief that Democrats will strip Christians of their liberties—tend to be stronger predictors of intolerance than do perceived personalized threats (which are rare in any event). As such, perceived threats to religious group identity could be strong predictors of antipathy toward out-group members. Strong, positive religious in-group identification leads to strong religious out-group negativity. This negativity leads to antipathy toward religious opponents, which heightens perceptions of threat from religious opponents and causes religious intolerance (Gibson and Gouws 2005; Miles 2019).
Muslim religious identity in the United States has also become increasingly racialized (Lajevardi and Oskooii 2018; Selod and Embrick 2013) and conflated with Arab national identity (Calfano and Lajevardi 2019). Thus, non-Muslims may perceive threat as more than competition for adherents but an actual physical or existential threat to Christian American lives and values. Whereas before 9/11 most discrimination faced by Muslims would have largely been based on national origin (e.g., Iranian, Lebanese, etc.), post 9/11 Muslims have been reframed as an Arab and religious out-group that is a threat to American society (Ayers and Hofstetter 2008; Jamal and Naber 2008; Selod 2015).
This reframing influences how the U.S. public views Muslims in the United States. The following figures illustrate the partisan/religious-based hostility toward Muslims in the United States. Pew asked respondents to indicate their feelings toward members of various religious groups in 2014 (N = 35,000) and again in 2017 (N = 4,284). A score of 100 on the thermometer represents very warm, positive feelings toward members of the group, with a score of 0 indicating very cold, negative feelings. A score of 50 means that the person has moderate feelings toward members of the described group (see Figure 14.1).
In 2014, Republican Protestants1 had slightly warmer than neutral attitudes toward Muslims (53.98), while Democratic Protestants reported feelings 16 points warmer on the 100-point scale (69.96). That is a considerable partisan gap in attitudes toward Muslims. Yet, by 2017 the gap is even larger. In 2017, Republican and independent Protestants report cold feelings toward Muslims (39.75) while Democratic Protestants register cooler feelings toward Muslims (64.38) in 2017 compared to 2014. The large decline in Republican Protestant attitudes toward Muslims yields a large partisan gap of nearly 25 points on the 100-point scale. We are reluctant to attribute all of this change in attitudes to President Trump, in part because Republican presidential candidates have a long history of incorporating anti-Muslim messages in their rhetoric. In 2012, Herman Cain said that he would not appoint a Muslim to his cabinet, Rick Santorum argued that the concept of equality does not exist in Islam, and almost every 2016 candidate for the Republican presidential nomination made anti-Muslim comments (Bush 2015). Clearly, President Trump’s harsh anti-Muslim rhetoric exacerbated religious divisions, but he was also tapping into sentiment that existed before he entered politics.
Figure 14.1 The Change in Protestant Attitudes toward Religious Group Members by Partisan Affiliation from 2014 to 2017 (Source: Pew 2014 and 2017 surveys.)
In addition, Democratic Protestants report much lower ratings of evangelical Protestants in 2017 than they did in 2014, while Republican Protestants report warmer ones, and Democrat Protestants report colder feelings toward Mormons in 2017 compared to 2014. This suggests that partisanship influences how religious Americans feel about members of religious groups in the United States. Perceptions of religious and other out-groups is contextual and made in comparison to other groups (Calfano, Lajevardi, and Michelson 2019; Kalkan, Layman, and Uslaner 2009). Thus, individual perceptions of others is in part a reference to one’s own position as well as a multitude of factors, including, but not limited to, religious affiliation.
To demonstrate the strength of partisan influence on attitudes about Muslims, Figure 14.2 shows the same thermometer rating about Muslims from Catholics, agnostics, and those who do not affiliate with a religion. Not only do Republican Protestants report more negative feelings toward Muslims in 2017 than they did in 2014, Republicans in virtually every religious category did the same. Republican agnostics report a nearly 50-point decline in their attitudes toward Muslims in this three-year period, while Democratic agnostics have much more favorable attitudes toward them. To a lesser degree, those who are not affiliated with any particular religion show an identical pattern. Democrats become more favorable toward Muslims, while Republicans become much less so. In 2014, there were no real partisan differences in Catholic feelings about Muslims, but by 2017, Republican Catholics report feelings almost 20 points lower than Democratic Catholics.
Figure 14.2 Feelings toward Muslims by Religious and Partisan Affiliation (Source: Pew 2014 and 2017 surveys.)
Strongly identifying with one’s religion leads individuals to view members of some groups in the United States more like members of their in-group and other group members more like an out-group. Religious in-groups and out-groups are determined not solely by religious beliefs but also by political alignment. When religious groups compete politically, it magnifies the perceived distance between “us and them” and creates intergroup hostility. Because white evangelical Protestants align with the Republican Party and atheists align with Democrats, there is strong antipathy between the two groups (Campbell 2020). Similarly, as religious groups are perceived to be on one’s own side politically, there will be less antipathy between the two religious groups, despite theological divergence.
Social identities are not as stable as beliefs and values, and as the strength of one’s identification changes in response to societal conditions, antipathy expressed at one point in time can quickly dissipate as individuals adjust the importance of their various social identities (Miles 2019). That is, prejudice is context specific. Even if religious Americans exhibit prejudice toward members of some disadvantaged groups in society, it is unlikely that it represents a general, stable view.
Affective Polarization
Recent scholarship notes the alarming trend in which members of one political party express surprising negativity toward members of the opposing political party. Partisans report being upset if their progeny were to marry someone from the opposing political party (Iyengar, Sood, and Lelkes 2012), and they are less likely to have friends from the opposing political party (Iyengar et al. 2019). It is tempting to combine this scholarship with findings presented previously and conclude that antipathy toward Muslims is simply the result of partisan affective polarization.
We think that this explanation is too simple and does not adequately explain increased hostility toward Muslims for two reasons. First, the percent of Muslims who identified as Republican in 2007 is roughly the same as in 2017 (Mohamed 2018). If the number of Republican Muslims stayed constant in the three-year period, there is no reason for the affective gap to widen if it is solely driven by partisan affiliations. Second, the gap in partisan affect toward Muslims between Democrats and Republicans is nearly twice as large as it is for other religious groups who are aligned with the Democratic Party. As such, a widening affective polarization gap between Democrats and Republicans between 2014 and 2017 may account for some of the change in attitudes toward Muslims but cannot explain why Republican affect toward Muslims declines more than affect toward other religious groups aligned with Democrats.
As noted previously, out-group antipathy and hostility are context dependent. That which motivates hostility toward an individual who identifies with an out-group in one context would not do so in a different context. Context raises the salience of one identity over another. Something happened within the Republican Party in the three-year period between 2014 and 2017 that caused Republicans to have much more negative feelings about Muslims. As noted previously, many Republican presidential candidates used anti-Muslim rhetoric in their campaigns (Bush 2015), and by the 2018 midterm campaigns, seventy-one Republican candidates for office used anti-Muslim rhetoric in their messaging. Anti-Muslim candidates came from every region of the country, in progressive, conservative, and swing districts, and at every level of government (Muslim Advocates 2018).
Some Protestant clergy employed strong anti-Muslim rhetoric. When Pope Francis declared in 2016 that Islam is not terroristic and that all religions want peace, Franklin Graham, perhaps an extreme example, argued that as individuals “behead, rape, and murder in the name of Islam,” they are following the teachings of the Koran (Gibson 2016). The divide has only intensified since the election of President Trump. He tweeted about the threat of radical Islam to the American way of life about once per month.2 One of the first things he did in office was to ban Muslims from entering the United States. When political leaders of one’s own party employ rhetoric that matches that of their religious leaders, it has a synergistic influence on their attitudes (Campbell 2020; Egan 2020; Nacos, Nacos, and Torres-Reyna 2007; Ocampo, Dana, and Barreto 2018). President Trump’s anti-Muslim rhetoric, along with any potential anti-Muslim teachings of evangelical Protestant clergy, causes Republican evangelical Protestants to have much more negative attitudes toward Muslims than they otherwise would. This is not necessarily because of differing values, cultures, or xenophobic tendencies; rather it is because group leaders in both movements use rhetoric to frame Muslims as members of the out-group. And since negative information about the out-group is more easily accepted than positive (Huddy 2001; Tajfel and Turner 1979), this rhetoric might cause some Americans to feel considerable hostility toward Muslims.
How much do some Americans dislike Muslims? How much of the anti-Muslim sentiment is spillover from widening affective partisan gaps? To answer these questions, we employed an experiment during the COVID-19 pandemic to measure intergroup hostility precisely by assessing how Americans would save some Americans rather than others.
Study Design
Participants. A sample of 1,997 subjects was recruited by Lucid to participate in a national political study during June 10–26, 2020. Lucid is an aggregator of survey respondents from many sources, and its respondents are widely used in academic research. It collects basic demographic information from all subjects who flow through their doors, facilitating quota sampling to match U.S. census demographic margins (Coppock and McClellan 2019). Because convenience sample participants might not pay close attention to the survey questions, we included an attention item and filtered out those who were not paying attention. In all, 4,445 began the survey, 2,340 (52.6 percent) correctly answered the attention question, and, of those, 1,997 (85.3 percent) completed the entire survey. A comparison in the appendix shows that our sample is similar to the U.S. population on several key demographic variables.
Procedure. Rather than creating several separate experiments manipulating demographic profiles of people who the individual might prioritize protecting from the virus in turn, we opted for a conjoint experimental design because it allows us to see how each demographic variation works in conjunction with each other (Hainmueller, Hopkins, and Yamamoto 2013). This design allows us to isolate the effects of the different treatments while maintaining a balance of internal and external validity. Randomization means that regression can be used to recover the treatment effects, and the experiment will have higher degrees of realism compared to other experiments that simply vary a single dimension (Bansak et al. 2017, 2018). Table 14.1 describes each treatment to which a participant might be randomly exposed, and Figure A14.1 in the appendix provides an example of what might have been seen by a participant in the survey. The survey was fielded as states were reopening after an initial shutdown but right as numbers of COVID-19 infections were rising once again. At this moment, the public was engaged in genuine debate about whether to prioritize restarting the economy or protecting vulnerable individuals from infection.
After viewing the two scenarios, respondents were presented with a dichotomous choice of which individual to prioritize saving. Although we force a choice in this experiment, results from conjoint experiments that do not force a choice often yield similar results to those that do (Hollibaugh, Miles, and Newswander 2020). The activity was repeated three times in succession, for a total of three choices.
Methods
As we are using a conjoint experiment in our analysis, we estimate average marginal component effects (AMCEs), per the recommendations of Hainmueller, Hopkins, and Yamamoto (2013). The AMCE is an estimate of the average extent to which a particular scenario component (e.g., race, religion, gender, ideology, etc.) affects the dependent variable. Hainmueller, Hopkins, and Yamamoto (2013) show that the AMCEs can be estimated by regressing the dependent variable on sets of indicator variables measuring the levels of each attribute; for example, Age 35–45, Age 65–75 would be included as independent variables in such a regression to capture the effect of the Age treatment, with Age 18–25 as the baseline category.
Each respondent participated in the experiment three times, which gives us a sample size three times as large as the number of respondents in the survey. For all dependent variables, the forced choice operationalization is a binary variable indicating which individual they would save. Because each respondent chose from three different pairs, we use the appropriate statistical corrections.
Results
Who does the public prioritize saving from the effects of COVID-19? Figure 14.3 presents the results from the full sample of participants. The points in the figure are the estimated probability that someone with that demographic characteristic will be saved, with the lines representing the 95 percent confidence interval of the estimate. The vertical dotted line at the 0 point of the x-axis represents no effect. If the solid horizontal lines overlap with the dotted vertical line, the model predicts that there is no difference between someone with that demographic profile and someone with the baseline demographic profile.
Figure 14.3 The Estimated Effects of Attributes on the Choice to Save an Individual (Source: June 2020 survey.)
On average, Americans are significantly more likely to prioritize saving someone aged sixty-five to seventy-five compared to someone aged eighteen to twenty-five. Yet they are just as likely to prioritize saving someone aged thirty-five to forty-five as they are someone aged eighteen to twenty-five. Specifically, the model predicts that survey participants were 3 percentage points more likely to save an older American. Each of the other demographic characteristics overlaps the 0 point, which means that they do not have an effect different from zero. Broadly, this is encouraging and consistent with expectations. When Americans are forced to make a choice about who to prioritize saving, there is no systematic bias toward any particular partisan or religious identity, but there is toward those most vulnerable to the virus. Since older people were more likely to suffer extreme symptoms from COVID-19, Americans prioritize keeping them safe.
Subgroup Analyses
This experiment provides an opportunity to estimate the degree of out-group hostility some Americans express toward others. It is well documented that Republicans and Democrats are growing less tolerant of each other, and religious intolerance is nothing new. We think it is likely that the same identity-based motivations that cause members of some religious groups to dislike members of other religious groups might motivate them to prioritize saving members of their own religion and not saving members of other religions. We do this by creating subsets of the data and running the same analyses using only members of particular subgroups. Doing so reduces the overall sample size for the analyses and widens the length of the 95 percent confidence interval, which means it also requires a larger substantive effect to achieve statistical significance.
We begin by looking at the responses from people who strongly identify as Democrats. Figure A14.2 in the appendix shows that Democrats make roughly the same judgment call as other Americans. The only group with a model prediction significantly different from zero are people aged sixty-five to seventy-five years old. Democrats are about 4 percentage points more likely to prioritize saving an older person compared to a younger person. We find no evidence of partisan-motivated antipathy toward members of opposing partisan or religious groups. Democrats are just as likely to prioritize saving an evangelical Protestant or a Republican as they are a Muslim or a fellow Democrat. We find the same pattern of findings among Democrats who also have a strong religious identity.
Republicans show no systematic preference toward saving members of any particular group, including seniors. The length of the 95 percent confidence interval is fairly large among this subgroup, but the point estimates in Figure A14.3 in the appendix are pretty close to zero for members of most groups. There really is no evidence that Republicans prioritize saving members of any group over another.
The model predictions displayed in Figure A14.4 in the appendix replicate previous analyses but only on the subset of respondents who indicate that their religion is an important element of their own identity. We subset the data to include only those for whom religion is “somewhat” or “very” important to their sense of who they are. The sample does not include a lot of people from minority religions, and 56 percent of the subset with a strong religious identity are either Catholic or Protestant. The next highest group (17.76 percent) are people who did not select any of the denominational options and call themselves “something else.” Less than 3 percent of this group identifies as Muslim, about 8 percent are “nothing in particular,” and other religions combine to account for the remaining 15–16 percent of respondents with a strong religious identity. We find that individuals with a strong religious identity are more likely to prioritize saving an evangelical Christian compared to Muslims. We find no evidence that religious identity causes people to prioritize saving Republicans or conservatives from COVID-19. The only substantively large and statistically significant results from any of the subgroup analyses is that individuals with a strong religious identity are nearly 5 percentage points less likely to prioritize saving a Muslim than they are an evangelical Protestant.
Conclusion
Discrimination against Muslim Americans and anti-Muslim sentiment has grown especially acute since 2016 (Lajevardi 2020). Meanwhile, in the 2018 midterm elections, the first two Muslim women to be elected to the U.S. Congress were Democrats. Given the rising antipathy partisans express toward cross-partisans (Iyengar et al. 2019; Iyengar, Sood, and Lelkes 2012), we thought it likely that some of the increasing anti-Muslim sentiment could be explained by the increasing alignment of religious and political identities (Campbell 2020; Egan 2020). Perhaps some Republicans do not dislike Muslims per se, but they equate Islam and Democrats, and this motivates greater antipathy against Muslims because they are political opponents.
The reopening of some states and cities, combined with the rising number of infections of COVID-19 in the early summer of 2020, presented an ideal opportunity to test this hypothesis. Although President Trump and other Republican elites employed partisan rhetoric discussing infection rates, the threat posed by reopening was not clearly partisan or religious in nature. Muslims were not more likely to be infected than Christians; nor were Republicans more likely to be infected than Democrats. Willingness to prioritize saving an individual from one group over another would have to be motivated by underlying biases held by the respondent. There is no other objective, logical rationale for trying to save a Republican other than a preference for Republicans.
Our research design allows us to explore how individuals prioritize protecting some individuals from infection relative to other individuals simultaneously in an experiment. We show that Americans prioritize protecting older Americans from infection. We find no evidence of partisan differences in the willingness to protect the elderly. Furthermore, we demonstrate that partisanship did not influence one’s willingness to save a fellow partisan from infection, while allowing a cross-partisan to be infected. Both Democrats and Republicans were equally likely to save cross-partisans from infection.
When asked who they would prioritize saving from a COVID-19 infection, individuals with a strong religious identity were significantly more likely to prioritize saving an evangelical Protestant compared to a Muslim. Juxtaposed with the findings that partisanship does not change which religious groups the individual prioritizes saving, this suggests that religious identity influences life-saving priorities in the early stages of the COVID-19 pandemic. From a broader perspective, this suggests that some anti-Muslim sentiment is motivated more by religious identity than it is by religious beliefs or partisanship. Not only are Muslims closer in belief to religious Americans than are atheists (for whom there was no difference), but we replicated these analyses using belief in God (rather than religious identity) as a predictor and found no effect.
This does not mean that partisan rhetoric has no influence on anti-Muslim sentiment in the United States. Rather, we think that the findings point to the importance of elite rhetoric in guiding anti-Muslim sentiment. When this survey was conducted, Republican elites were engaging in anti-Muslim rhetoric, but it was not clearly connected to the pandemic. Elites were blaming cross-partisans for their handling of the pandemic, but at the time of the survey, it was not as prevalent in the rhetoric as it would become. As other work in this book demonstrates, elite rhetoric would eventually have a stronger influence on attitudes.
This illustrates the importance of context in the study of religion and politics. As others have noted (Abrams and Hogg 1999; Hale 2004), identities develop to fill a psychological need for certainty. Context determines which identities are salient at any given moment. The uncertainty created by the COVID-19 pandemic and the decision to begin reopening caused religious identities to be more salient than political identities were. When asked to decide who should be saved, evangelical Christians with strong religious identities chose to prioritize saving their own. Future work should examine the contexts in which religious identities become more salient than other identities.
____________
Material referencing an appendix in this chapter can be found online available here: https://dataverse.harvard.edu/dataverse/epidemic_among_my_people.