How does an emphasis on shared religion affect attitudes and behavior toward a marginalized group? And if religion matters, how much does it matter relative to other considerations? We study these questions in the context of attitudes and behavior of Turkish citizens toward Syrian refugees. In 2013, when we conducted our study, the UN High Commissioner for Refugees reported that the number of internally displaced persons, refugees, or asylum-seekers exceeded 50 million for the first time since World War Two (United Nations High Commissioner for Refugees (UNHCR) Reference Salehyan and Gleditsch2014). Population displacement has large repercussions for the political dynamics in the recipient countries. For instance, previous research has demonstrated that the presence of refugees and displaced persons can increase the risk of subsequent conflict in host countries (Salehyan and Gleditsch Reference McCleary2006). Yet, little is known about individual-level perspectives on refugee flows—a surprising feature, given the extensive political science literature on opposition to immigration (for reviews, see Dancygier and Laitin Reference Colby and Damon2014; Hainmueller and Hopkins Reference Gaertner and Dovidio2014). Although refugees and immigrants are distinct groups, we argue that research on the attitudes and behavior toward refugees can inform the long-standing debate on the strength of economic considerations and cultural bonds in forming individual treatment of outgroups. In our study, we attempt to experimentally test these two factors via a unified theoretical framework rooted in the common in-group identity model (Gaertner and Dovidio Reference Druckman2000) and its associated recategorization approach.
Recategorization encourages members of different groups to see themselves as part of an encompassing, superordinate identity (Gaertner and Dovidio Reference Druckman2000). With some recent exceptions (Charnysh, Lucas and Singh Reference Carkoglu and Kalaycioglu2015), political scientists have rarely studied the effects of recategorization. This is a missed opportunity, particularly as political elites extensively emphasize group categories in their rhetoric. For instance, Recep Tayyip Erdogan, Turkey’s prime minister at the time of our study, called Syrian refugees as “our brothers,” a reference that implicitly emphasized the common religion between Turks and Syrians (Karaveli Reference Stürmer, Snyder and Omoto2013). However, in spite of the government’s emphasis on shared religion and need for solidarity, Turkish citizens exhibit a high level of prejudice toward Syrian refugees.
Our study aims to systematically test the effect of religious recategorization on reducing individual-level prejudice toward Syrian refugees in Turkey. Relatively few studies examine the impact of religious appeals in spite of the prominent role of religion in politics. In our view, religious appeals can be a strong basis for recategorization due to their invocation of moral obligations and a powerful norm present in major world religions that all divisions within the community of believers—whether national, ethnic, or tribal—are sinful.
In addition to focusing on how religion can reduce out-group prejudice, our study expands on existing studies of recategorization in four important ways. First, the extant literature is limited in that it only tests the effect of a single unifying identity category relative to a distinct group identity. By contrast, we introduce two alternative categories, allowing us to examine if and when one identity type might be more effective at reducing out-group prejudice than another type. In particular, subjects were exposed to priming of either the common Muslim identity of Turkish citizens and Syrian refugees, or the common Sunni identity between the two groups. We theorize that the Sunni prime may have a stronger effect because in the context of our study, it redirects prejudice that would be directed toward refugees toward a new out-group non-Sunnis (in particular, the Alawite-dominated Syrian regime of Bashar al-Assad). The Muslim prime, by contrast, creates a far less salient outgroup in an environment where virtually all Turks and Syrians are Muslim.
Our second contribution is to analyze the relative strength of recategorization. Previous studies have not considered how competing information might alter the effect of recategorization. Our study analyzes whether the hypothesized prejudice reduction effect of emphasis on shared identity still holds when subjects are provided with information that is likely to induce prejudice. We do so via an additional experimental manipulation that randomly provides a factual statement on the economic cost of the refugees to Turkey’s national budget. Economic cost information serves as a prime of group competition for scarce resources—a condition that negatively affects the process and success of recategorization (Gaertner and Dovidio Reference Druckman2000). By cuing subjects to a common religious identity as well as providing potentially prejudice-inducing information, our design approximates a situation of competing frames (Druckman Reference Dovidio, Johnson, Gaertner, Pearson, Saguy and Ashburn-Nardo2004).
Third, because extant evidence on recategorization effects has mostly come from the laboratory (Kramer and Brewer 1984; Kurzban, Tooby and Cosmides Reference Karaveli2001), there is a dearth of evidence on how recategorization affects prejudice in real-world settings (Paluck and Green Reference List and Price2009). To help meet the demand for such research, we conducted a survey experiment in the field. Our design further took strides to incorporate real-world dynamics by randomly sampling respondents in areas with different levels of expected contact with Syrian refugees and different levels of religiosity. To this end, we surveyed Turkish citizens in three districts of Istanbul as well as in Gaziantep, a Turkish city and now refugee hub that is just 30 miles from the Syrian border.
Fourth, our study expands the literature by obtaining both attitudinal and behavioral measures of prejudice. In doing so, we join a trend in studies on opposition to immigration (Hainmueller and Hangartner Reference Gaertner, Dovidio and Bachman2013) that analyze both types of outcomes. Our behavioral measure was presented in the form of voluntary charitable donations to the outgroup. Thus, we are able to examine if our manipulations yield different attitudinal and behavioral responses and consider the implications for recategorization theory.
We decided to study the effects of recategorization on prejudice in Turkey for several reasons. First, the influx of Syrian refugees—400,000 refugees were in Turkey at the time of our study—was and continues to be a highly salient issue in Turkish domestic politics (Balci 2014). Second, in Turkey, as in many countries with refugee or immigrant influxes, identity appeals and economic considerations are both crucial to elite political discourse (see the Social Context section for details). Third, we wanted to conduct our work in a venue where it was less likely that our results were driven by social desirability bias that inhibits the expression of prejudice. In Turkey, the overt expression of prejudice toward Syrian refugees does not contravene social expectations. Finally, relative to other countries affected by the Syrian refugee crisis, Turkey was appealing for our study because attitudes toward refugees there are far less entrenched. In Jordan and Lebanon, for instance, local populations living alongside Palestinian refugees for decades and in higher refugee-to-inhabitant ratios have developed strongly held attitudes toward refugees as outgroups. In Turkey, where the refugee flow is more recent and the refugee-to-inhabitant ratio is far lower, we expected to find a context where our experimental manipulations were better positioned to affect prejudice.Footnote 1
Our study finds that emphasis on shared religion can reduce prejudice, but that its effects are limited. We find that the Sunni and Muslim primes reduced prejudicial behavior as well as certain attitudes toward refugees. In particular, the Muslim prime more effectively reduced prejudicial behavior, whereas the Sunni prime had a stronger effect of reducing prejudice as measured by our composite index of all attitudinal and behavioral measures. Consistent with our hypothesis, we found that joint provision of economic cost information alongside a religious prime removed the pro-refugee effect of the religious prime. This finding persisted even for the highly religiously observant.
The remainder of the paper unfolds in five sections. The next section presents our theoretical framework and hypotheses. We then describe our study’s social context, research design, and empirical findings, and close with our interpretation of the results.
Theoretical Framework and Hypotheses
Intergroup relations play a crucial role in attitude formation and political behavior (Gaertner and Dovidio Reference Druckman2000; Kinder and Kam Reference Hardy and Carlo2010). The tendency to favor in-group members and treat out-group members negatively has served as a basic building block of several concepts in political psychology, such as prejudice, ethnocentrism, and discrimination. Negative bias toward a social category of people includes cognitive, affective, and behavioral components. The strength of out-group bias varies depending on the context and can be triggered or diminished by external stimuli (Kinder and Kam Reference Hardy and Carlo2010). Social identity and self-categorization theories (Tajfel and Turner Reference Özden1979; Brewer and Miller Reference Brader, Valentino and Suhay1984) proposed that out-group prejudice can be reduced by breaking down or rearranging social boundaries. One particular intergroup approach toward prejudice reduction recommends recategorization, in which targeted subjects are encouraged to think of themselves as part of a superordinate, all-encompassing category.
The common in-group identity model, proposed by Gaertner and Dovidio (Reference Druckman2000), establishes a theoretical link between recategorization and prejudice reduction. According to the model, recategorization alters perceptions of in-group boundaries by seeking to incorporate the outgroup into the ingroup. By inducing members of distinct groups to perceive themselves as part of a single superordinate category, recategorization can help reshape perception of those who were formerly part of an outgroup. In short, the approach encourages people to abandon the “us and them” cognitive dichotomy in favor of a new inclusive “we” category.
We rely on religion as the basis for recategorization in our study. Appeals to a common religion have been rarely used in the research on social categorization. However, we argue that belonging to a common religion should be a strong basis for recategorization. First, appeals to common religion invoke moral obligations (Hardy and Carlo Reference Hainmueller and Hopkins2005; Colby and Damon Reference Chong and Druckman2010). Second, major world religions, including Islam in our study’s context, often dictate that any divisions within the community of believers are sinful, whether those divisions are racial, tribal, national, or otherwise. Lastly, in the Turkish context, religious primes are “meaningful, relevant, and strong,” ensuring that they satisfy the requisite criteria for successful priming (Chong and Druckman Reference Charnysh, Lucas and Singh2007).
In contrast to existing recategorization studies that only test one unifying category per study, we theorize and test potentially divergent effects of two alternative recategorizations. We generate the first recategorization via an all-encompassing religious appeal to the Muslim identity and the second recategorization via a more narrow sectarian appeal to the Sunni identity. Both appeals emphasize social bonds between the majority of the Turkish population and the majority of the Syrian refugees. However, they differ in an important respect. In the context of our study, an appeal to the Sunni identity creates a new outgroup. It emphasizes that Syrian refugees fled from the Alawite (non-Sunni) regime of Syrian President Bashar al-Assad, and thus makes salient the Sunni-non-Sunni divide. In contrast, the Muslim prime, in the context of our study, does not create a new meaningful outgroup, as virtually all Turks and Syrians are Muslims.Footnote 2 In this differentiation, we highlight that our recategorization approach does not only create a new “we” (i.e., Muslims) but also provides a new “them” (i.e., non-Sunni).
Our main hypothesis is that HYPOTHESIS 1: Priming Sunni or Muslim identities will decrease prejudice toward Syrian refugees. Regarding the relative strength of these appeals, we expect that the Sunni prime redirects prejudice from refugees to a new outgroup (i.e., non-Sunni). Consequently, we hypothesize that HYPOTHESIS 2: The Sunni prime may have a larger effect on reducing prejudice than the Muslim prime.
We assume that the effects of our manipulations may not be uniform across different subpopulations. First, we hypothesize that the effects of recategorization are conditional on out-group contact.Footnote 3 We assume that in-group members with more out-group contact hold more entrenched attitudes toward the outgroup than do individuals with less out-group contact. Consequently, we expect that HYPOTHESIS 3: Recategorization will have weaker effects on individuals with more out-group contact.
In addition, we expect the effect of different recategorizations to be conditional on the respondent’s level of religious observance. Religious appeals are more likely to resonate with individuals for whom religion plays an important role in their daily lives. In particular, we hypothesize that HYPOTHESIS 4: The Sunni or Muslim recategorization will be more effective in reducing prejudice among highly religiously observant individuals.
Our final hypothesis concerns the relative strength of recategorization: how much does it matter relative to other considerations? Existing studies have not considered how the positive effect of recategorization might interact with an intervention that provides prejudice-inducing information. We seek to fill this research gap by relying on the competing frames approach, particularly by introducing a cross-cutting factor of economic cost considerations. Our decision to do so follows a leitmotif in the literature on attitudes toward immigration, namely, the opposing effects of economic competition and cultural bonds (Brader, Valentino and Suhay Reference Blaydes and Gillum2008; Hainmueller and Hopkins Reference Gaertner and Dovidio2014). Locals may be concerned that their national government’s spending on sheltering, feeding, and otherwise caring for refugees will harm the national balance sheet and lead to increases in the prices of goods and housing rents. In addition, economic cost information can be interpreted as an indicator of group competition for scarce resources. Intergroup competition for scarce resources has been shown to reduce the effectiveness of recategorization; subjects are more hesitant to embrace the superordinate identity because it carries the cost of potential resource transfers from the ingroup to the outgroup (Gaertner and Dovidio Reference Druckman2000). Consequently, the provision of economic cost information is likely to induce prejudice toward Syrian refugees. However, we are mostly interested not in the direct effect of prejudice-inducing information, but in the effect of joint provision of it with religious recategorizations. Based on the logic of competing frames, we hypothesize that HYPOTHESIS 5: When economic cost information and religious recategorization are both provided to the same individual, the two statements will have countervailing effects on that individual’s attitude toward refugees. In other words, we expect that the negative effects of learning about economic costs will, in expectation, counterbalance any positive pro-refugee effects of recategorization using religious primes, and therefore the joint provision of a religious prime and economic cost information will have no effect on out-group prejudice.
Social Context
Our hypotheses and research design are tailored to Turkey’s social context. This section offers a brief background of the Syrian refugee crisis in Turkey, prejudice toward refugees, and the role of religious appeals in the Turkish political discourse on the refugee issue.
As the Syrian civil war rages on, a massive humanitarian and refugee crisis has ensued. Hundreds of thousands of Syrians fled for refuge, with most settling in neighboring Turkey, Lebanon, and Jordan. Recipient countries were forced to shoulder considerable financial burdens and political shocks. The New York Times warned that the Syrian refugee crisis was “what the United Nations, governments and international humanitarian organizations describe as the most challenging refugee crisis in a generation-bigger than the one unleashed by the Rwandan genocide and laden with the sectarianism of the Balkan wars” (Onishi Reference Skitka and Morgan2013). Indeed, the crisis in Syria has created a refugee population that accounts for 23 percent of the worldwide refugee population (UNHCR Reference Salehyan and Gleditsch2014).
The influx of Syrian refugees in Turkey became one of the most salient and contested issues in national politics. By mid-2013, just before our study began, Turkey was home to ~400,000 Syrian refugees, which accounted for some 21 percent of all Syrian refugees. The Turkish government’s policy has been to welcome Syrian refugees and provide them with range of essential services, including housing facilities, food supplies, and medical treatment.
The Turkish government and pro-government media outlets have made their case to support Syrian refugee by appealing to national pride and religious solidarity. Recep Tayyip Erdogan, Turkey’s prime minister at the time of our study said that, “For our Syrian brothers who are asking when God’s help will come, I want to say: God’s help is near.” He also proclaimed, “You are now in the land of your brothers, so you are in your own home.”Footnote 4
In addition to the explicit broad appeals to Muslim solidarity, there are also many implicit and some explicit appeals to the common sectarian Sunni identity of Syrian refugees and Turkish population (US Department of State, Bureau of Democracy, Human Rights and Labor Reference Paluck and Green2012; Fehim 2013). The Syrian civil war is often processed and framed in the region as a war led by mostly Sunni rebels against an Alawite (non-Sunni) regime in Damascus. The vast majority of Syrian refugees in Turkey are Sunni Muslims, an identity they share in common with the majority of Turks. The civil war, coupled with the refugee crisis, has made such sectarian dynamics politically salient in Turkey. Given these dynamics, the government tried to emphasize the common Sunni identity with Syrian opposition and refugees to promote their agenda.Footnote 5
Although religious appeals in support of Syrian refugees resonate with some in Turkey, they also provoke opposition among others. Secular Turks, in particular, may perceive an emphasis on shared refugee identity as contradictory to the Republic’s secular foundations established by Mustafa Kemal Ataturk.Footnote 6 Appeals to a common Sunni identity, also widely used in public discourse, are also controversial. Despite the common Sunni affiliation of most Turkish citizens and Syrian refugees, Turkey also has a considerable minority of Alevis, many of whom support Syrian Alawites, including President al-Assad, and oppose the presence of Syrian Sunni refugees in Turkey (Balci 2014). Consequently, appeals to Muslim or sectarian Sunni religious identities are highly politicized and may significantly shape Turkish perceptions toward Syrian refugees.
Even as Ankara has continued to welcome Syrian refugees with the language of religious brotherhood, the Turkish public has come to see Syrian refugees as a burden (Kirişci Reference Hart and Middleton2014). Anecdotal evidence from our study, as well as various accounts from international NGOs and media outlets (Özden Reference Kurzban, Tooby and Cosmides2013), demonstrates significant levels of prejudice against Syrian refugees in recipient countries. Anti-refugee discourse is especially prevalent in those areas that have recently become home to large refugee populations. Many Turkish citizens complain that the refugees impose costs on the national budget, raise local rent prices, hurt the labor market, and depress wages (Kirişci Reference Hart and Middleton2014). Some Turks explain their frustration by highlighting cultural differences between the Turkish population and Syrian refugees, criticizing the latter for “not having manners, speaking and laughing very loudly, being uneducated, having too many children, and harassing women on the streets” (Özden Reference Kurzban, Tooby and Cosmides2013). In the southeastern Turkish province of Hatay, home to a large refugee population, there were documented instances of institutionalized discrimination, such as prohibiting Syrians from sitting at front tables at a cafeteria in the city center. Finally, in the aftermath of the terrorist attack in the town of Reyhanli, which borders Syria, local residents blamed Syrian refugees for the attack. Clashes between locals and refugees ensued. These episodes reflect a broader decline in Turkish public support for their country’s refugee policy. A representative national poll taken in November 2013 found that 86 percent of Turks believe that no further Syrian refugees should be allowed in the country and that 30 percent of respondents also wanted the refugees to be deported to Syria (Balci 2014).
Research Design
We conducted an original survey with an embedded experiment on a sample of 1140 maleFootnote 7 respondents in Istanbul and Gaziantep. Gaziantep is a major Turkish town on the Syrian border that is known today for its proximity to many Syrian refugee camps. Our Gaziantep sample includes 240 respondents. Within Istanbul, we also sampled one district Bagcilar (N=180) that experienced a major influx of Syrian refugees beginning late 2012. The selection of these sites enables us to examine treatment effects conditional on contact with Syrian refugees. Furthermore, to facilitate a more fine-grained analysis of the treatment effects, we stratified the sample on level of religious observance by further dividing the Istanbul sample pool between one conservative district Fatih (N=360) and one more secular district Besiktas (N=360). The survey was conducted during Ramadan, when religious identities in Turkey are especially salient.
Respondents were selected via uniform sampling: enumerators were instructed to survey every fifth male adult starting from a random point on streets in each district. Subjects were approached in parks, side streets, and small shops. Each potential respondent was told that he would immediately receive 4 Turkish lira (about 2 US dollars)Footnote 8 as compensation for his time upon successful completion of the survey. Compensation was provided to encourage participation and helped us achieve a relatively high response rate of 70 percent.Footnote 9 Although we cannot claim that our sample is representative of Turkey or the populations of Istanbul and Gaziantep, we achieved considerable variation on important observable characteristics of interest, including age, education, income, employment status, previous contact with refugees, and level of religious observance. Descriptive statistics of our sample are presented in the Supplementary Table 1.
Randomization was conducted before the implementation of the surveys. Blocking on each enumerator, we generated a list of subject ID numbers with each ID number randomly assigned to one of the six experimental conditions described below. This enabled us to account for the enumerator effect, which was found to be significant in shaping sensitive surveys, especially with regard to surveys on religion in the Middle East (Blaydes and Gillum Reference Barabas and Jerit2013). We also included enumerator fixed effects in our analysis.
Our survey began with a series of background questions on demographic characteristics, education, level of income, employment status, and religious observance. Religious observance is an especially important characteristic for our study and was measured before treatment. Our measure of religious observance is based on a question that asked respondents how often they attend Friday prayers at a mosque. Attendance at Friday prayers indicates a high level of personal religiosity as well as an outward display of religiosity in a communal ritual. In addition, we asked respondents about the extent to which Sharia should be the foundation for Turkish national laws. This question was designed to obtain respondent views on political Islam.Footnote 10
After the initial battery of background questions, the survey provided an introductory statement regarding Syrian refugees. Our survey experiment manipulated this introductory statement in specific ways in line with our factorial design. Our design includes six distinct experimental conditions that represent all possible combinations of our religious primes and economic cost information.
In the control condition, the introductory statement read: “These next questions address your opinion toward the Syrian refugees in Turkey, who suffered from the Assad regime.” In the Sunni prime treatment, the statement was modified to read: “… the Syrian refugees in Turkey, most of whom are Sunnis, who suffered from the Assad regime.” In the Muslim prime treatment the statement read: “… the Syrian refugees in Turkey, most of whom are Muslims, who suffered from the Assad regime.”
The second factor is a statement about the economic cost of Syrian refugees to the Turkish government. This statement read: “The Turkish government has spent almost 3 billion Turkish lira providing food, shelter, and other humanitarian support to these refugees.” The control condition of this factor is represented by the absence of the statement on economic cost.Footnote 11 Combination of these experimental manipulations leaves us with 3×2 factorial design.
In all six experimental conditions, the introductory statement is followed by attitudinal questions designed to gauge each respondent’s attitude toward Syrian refugees. These questions include an informational question as to how frequently the respondent sees Syrian refugees as well as questions on attitudes toward: (1) Turkish government spending on Syrian refugees; (2) acceptance of Syrian refugees working in the respondent’s neighborhood; (3) permitting refugees to stay in Turkey if the war in Syria continues; (4) and trust toward the refugees. With the exception of the question on how often the respondent sees Syrian refugees, all of these social proximity questions are used as attitudinal indicators of support for refugees. These measures are thus used as dependent variables in our ensuing analysis.
Next, our survey included a question designed to provide a behavioral measure of the treatment effect. All respondents were asked whether they would like to donate any part of their survey compensation to a charity supporting Syrian refugees in Turkey.Footnote 12 Charitable giving may provide evidence of one’s altruism, concern for reputation, or certain psychological costs and benefits. At its base, though, charitable donation is a monetary contribution to a particular group or cause that shows a donor’s attitude and social connectedness toward a recipient (List and Price Reference Kinder and Kam2009). In other ways, our donation request also mimics the widely used dictator game. Respondents were given four one-lira coins, enabling them to easily donate different amounts. To minimize the chance that respondents might have felt compelled to donate, enumerators assured respondents that the decision to make a donation to the charity was entirely voluntary and that respondents were by no means required to donate any part of their four lira compensation.
Empirical Analysis
The outcome measures of the control group, which serves as a baseline indicator, show that there is significant prejudice toward Syrian refugees. More than half of our respondents (51 percent) think that Syrian refugees should not be allowed to stay in Turkey even if the war in Syria continues; 57 percent of the sample agree that government spending on refugees is excessive and should be cut; 61 percent indicated that they do not trust the refugees; and more than half of the respondents (53 percent) decided to not donate any amount to the refugees; and 39 percent of respondents stated that they would not like a Syrian refugee to work in their neighborhood.
The survey also provides strong support for our initial claim that religion plays a principal role in determining attitudes and behavior toward refugees. When we analyze the full sample we find that, ceteris paribus, religious observance has a strong positive impact on support for refugees across all five outcome measures (see Table 2A in the Appendix for the results of the regression analysis). For example, people who report weekly attendance at Friday prayers (i.e., the highly religiously observant) on average donate 0.4 Turkish Lira more to Syrian refugees than do people who report that they never or only sometimes attend Friday prayers. These positive effects remain when we use potential alternative measures of religiosity, such as support for Sharia law or residence in Fatih, a religiously conservative district of Istanbul.
In contrast, other predictors’ explanatory value holds only for certain outcomes. For example, our data show that, educated people are more likely to allow refugees to stay in Turkey and more likely to donate money to support them, yet are no different from the less educated in their views on governmental spending or trust toward the refugees. Income has a positive effect on support for government’s spending, donation, and the likelihood of acceptance of the refugees into the neighborhood. Contact with refugees is associated with lower donations. The variable capturing employment remained insignificant in all models. Finally, residents of Gaziantep tend to favor more government spending on refugees, express more trust and are more willing to accept a refugee in their neighborhood, but on average give significantly less in their donations. Although we do not attribute any causal effects regarding these patterns, we interpret these findings as broadening our understanding of trends in attitude formation.
Experimental Results
Our main objective is to examine the effect of religious primes and economic cost information on levels of support for Syrian refugees. To do so, we compare average survey responses and average amounts of liras donated among subjects who were randomly assigned to the control and the five treatment conditions. In order to estimate the general effect of our manipulations and address multiple comparisons problem,Footnote 13 we construct a summary index of support for the refugees.Footnote 14 The index aggregates all five dependent variables. Figure 1 graphically presents the means for each experimental condition across all outcomes.
Fig. 1 Experimental results (a) Refugees can stay (b) Support for government spending (c) Trust (d) Acceptance of the refugees (e) Average donation (f) Composite index
The results indicate that the Sunni prime has a positive effect on several outcome measures, namely acceptance of refugees in the subjects’ communities, levels of charitable donations and general support for the refugees. In turn, the Muslim prime has a positive effect on donations, but remains statistically insignificant in shaping attitudes toward the refugees. To calibrate the effect size of the treatments, we divide the difference-in-means estimates by the pooled standard deviation of the control group and the relevant treatment group. Among all manipulations, the largest magnitude is the impact of the Muslim prime on donations (Cohen’s d=0.29). Using the mean exchange rate for the month of July 2013 (1 USD=2 YTL), respondents primed with the Muslim cue donated on average 30 US cents more than respondents in the control. The effect of the Sunni prime on the donation measure was smaller (Cohen’s d=0.17), roughly equivalent to 17 US cents. The Sunni prime had the largest effect on the acceptance of refugees in the subjects’ communities (Cohen’s d=0.26). The magnitude of the Sunni prime effect on the composite index of support for Syrian refugees on average equals 0.22 SD (i.e., Cohen’s d=0.22). In turn, the Muslim prime effect on the composite index is non-distinguishable from 0 (Cohen’s d=0.03).
Regarding the competing frame factor, we found that the provision of economic cost information reduces trust toward refugees but does not have a consistent negative impact for all other outcomes of the study. Nonetheless, when subjects receive both the religious prime and economic cost information, the positive effects of both religious primes vanish. Table 1 presents results obtained with fully saturated regression models that allow us to incorporate all treatments into a unified estimation framework and test treatment-by-treatment interactions. Formally, we estimate the following equation:
$$Y_{i} =\alpha {\plus}\beta _{a} A_{i} {\plus}\beta _{b} B_{i} {\plus}\beta _{c} C_{i} {\plus}\delta _{{ac}} A_{i} C_{i} {\plus}\delta _{{bc}} B_{i} C_{i} {\plus}{\varepsilon}_{i} ,$$
where A is the Sunni prime treatment, B the Muslim prime treatment, and C the economic cost information treatment. The coefficients of the interaction terms δ signify the effects of the treatment condition that exposed respondents to both a religious prime and economic cost information conditional on the direct effects of a religious prime and economic cost prime.
Table 1 Treatment Effects of Religious Primes, Economic Cost Information and Their Interactions on Measures of Prejudice Against Syrian Refugees
Note: *p<0.1, **p<0.05, ***p<0.01.
Our results strengthen when we adjust for predictive covariates such as age, level of education, employment status, income level, level of religiosity, contact with the refugees, city, and neighborhood (see Table 2). In particular, using covariates allows us to increase precision as we isolate the positive effect of the Sunni prime on (1) support for an increase in governmental spending; and (2) the summary index of support for refugees. Covariate adjustment also increases the positive effect of the Muslim prime on donations. The same patterns hold when we include enumerator fixed effects. Specifications with enumerator fixed effects are presented in Appendix.
Table 2 Treatment Effects with Covariate Adjustment
Note: *p<0.1, **p<0.05, ***p<0.01.
Heterogeneous Treatment Effects
By City: Istanbul versus Gaziantep
In this section, we analyze how treatment effects may vary among specific subsamples of the respondents, namely between Istanbul and Gaziantep; between those who had more and had less contact; and between those who are more and are less religious.Footnote 15
First, we look at the difference in response to treatments across two cities in our study, Istanbul and Gaziantep (Figure 2). In Gaziantep, the Sunni prime had a large positive impact on the general index of support for the refugees (Cohen’s d=0.78); and some less clear, but distinguishable positive effects on allowing refugees to stay, support for government spending, acceptance in the neighborhood and donations. In contrast, the Muslim prime had no statistically distinguishable effect in Gaziantep.
Fig. 2 Heterogeneous treatment effects (Istanbul versus Gaziantep) (a) Refugees can stay (b) Spend (c) Trust (d) Acceptance of the refugees (e) Donation (f) Composite index
In Istanbul, neither the Sunni prime nor the Muslim prime had a statistically significant effect on the general index of prejudice. But the Muslim prime caused an increase in donations (Cohen’s d=0.36). Although the Sunni prime had no clear effect on donations, it did increase respondents’ willingness to accept refugees in their communities. Nonetheless, we should interpret these findings with caution, as the Gaziantep sample size is much smaller (N=240) than the Istanbul one (N=900).
By Contact
Second, we estimate the effects of experimental manipulations conditional on respondent’s contact with refugees (Figure 3). We categorize respondents, who report “almost never see” refugees as the group of no contact; and those who report seeing them “everyday” or “once per week” as the group of contact. This categorization splits our sample in two almost equal parts. The results show that among those who report contact with refugees, the Sunni prime has a positive effect on support for governmental spending (Cohen’s d=0.22), acceptance of refugees in the neighborhood (Cohen’s d=0.20), donations (Cohen’s d=0.34), and the general index of support for refugees (Cohen’s d=0.17); the Muslim prime increased donations (Cohen’s d=0.26). By contrast, among those who report no contact with refugees, the Sunni prime does not produce any statistically significant results, whereas the Muslim prime has a larger positive effect on donations (Cohen’s d=0.36).
Fig. 3 Heterogeneous treatment effects (by contact) (a) Refugees can stay (b) Support for government spending (c) Trust (d) Acceptance of the refugees (e) Donation (f) Composite index
By Religiosity
Finally, we estimate average treatment effects conditional on respondent religiosity (Figure 4). To do so, we first divided our sample by responses to the survey question on Friday prayer attendance. Our sample has roughly similar percentages of respondents who stated that they always attend Friday prayers and respondents who indicated that they sometimes or never attend Friday prayers. We classify the former group as exhibiting a high level of religious observance and the latter group as exhibiting a low level of religious observance.
Fig. 4 Heterogeneous treatment effects (by religiosity) (a) Refugees can stay (b) Support for government spending (c) Trust (d) Acceptance of the refugees (e) Donation (f) Composite index
Following this dichotomy, we found, quite surprisingly, that among less religiously observant respondents, the Sunni prime has a strong positive effect on the general index of support for Syrian refugees (Cohen’s d=0.18), on support for governmental spending (Cohen’s d=0.24) and on acceptance in the neighborhood (Cohen’s d=0.39). Among the highly religiously observant, the Sunni and Muslim primes increased donations, but importantly, neither prime had a statistically significant effect on respondents’ attitudes. For the same subsample, the effect of the Muslim prime on donations was especially large (Cohen’s d=0.48), and on average equals 46 US cents; the effect of the Sunni prime was less pronounced (Cohen’s d=0.26) and was about 25 US cents. Crucially, we found, that even among the highly religiously observant, the positive effects of religious primes on donations disappear when these primes are paired with the economic cost information.
Discussion
Our study yields several important findings. First, we establish that recategorization that appeals to a shared religion—in this case, Islam—can reduce prejudice toward a marginalized outgroup. Crucially, we found that our intervention increased charitable donations toward Syrian refugees. This finding provides rare field experimental evidence that recategorization can make individuals behave in a less prejudiced way. However, our intervention had only a limited effect on attitudes toward refugees. Second, we found that an appeal to an inclusive religious identity has a different impact on out-group prejudice relative to an appeal to a particular sect. We find that the Muslim prime had a larger effect on donations than the Sunni prime, but that overall, the Sunni prime is a more powerful approach to reduce prejudice toward the refugees. Third, we find that the effects of our manipulations are conditional on social context, contact with refugees, and respondents’ religiosity. In particular, the Sunni prime is a more powerful recategorization tool than the Muslim prime for three groups: residents of Gaziantep, those who experienced contact with refugees, and the less religiously observant respondents. In turn, the positive effect of the Muslim prime on donations is only present among more religious respondents and in Istanbul. Fourth, and crucial to understanding the relative strength of recategorization, we find that provision of economic cost information cancels the pro-refugee effect of religious primes, even among the religiously observant.
In general, our findings give empirical support to the common in-group identity model. Yet, at the same time, our results call for extensions of the model that account for: (1) diverging attitudinal and behavioral responses; (2) differential impacts of alternative recategorizations; (3) the interaction of prejudice-reducing and prejudice-inducing treatments.
To motivate possible extensions of that model, we offer two complementary mechanisms to explain why recategorization might affect attitudes and behavior differently.Footnote 16 One possible explanation comes from psychology research on empathy and helping behavior. The literature suggests that superficial helping behavior may be motivated more by empathetic concerns rather than favorable attitudes (Neuberg et al. Reference Kramer and Brewer1997; Stürmer, Mark and Omoto 2005; Dovidio et al. Reference Dancygier and Laitin2010). In this reading, behavior and attitudes are determined by two different latent variables: empathy for the former and prejudice for the latter. Another possible explanation for the observed difference in the results between attitudinal expressions and charitable giving may rest in a more fundamental difference between individual preferences and social norms. Attitudinal measures might speak to respondents’ easily accessible and pre-formed opinions, which are less likely to be manipulated by primes. By contrast, responding to a request for charitable donation may be driven by moral conviction (Skitka and Morgan Reference Neuberg, Cialdini, Brown, Luce, Sagarin and Lewis2014) social, which can be triggered by religious cues. Due to the fact that our project was conducted in Turkey, a Muslim-majority country, it is plausible that religious primes affected charitable donations because Islam places an important emphasis on making voluntary charitable donations (sadaka) to fellow Muslims (McCleary Reference Kirişci2007).
A second important finding from our project concerns the difference in the effects of the Sunni and the Muslim primes. We argue that these differences can be attributed to the nature of the identity primes. The Muslim prime primarily delivers a religious message. It may cause behavioral change by activating a religious norm of charity, yet is too weak to change individual preferences toward refugees. By contrast, in our study’s context, the Sunni sectarian appeal is likely to be processed as political message that creates a new and politically salient outgroup: non-Sunnis. Our argument of understanding divergent effects for the Muslim and Sunni primes is rooted in the point that recategorization is more effective if it introduces a more relevant out-group category. This point was raised by Tajfel and Turner (Reference Özden1979, 4) who wrote that “the social situation must be such as to allow for intergroup comparisons that enable the selection and evaluation of relevant relational attributes.”
Our study also uncovers patterns of recategorization effects conditional on the level of religious observance, city, and contact with the refugees. We focus on the most theoretically relevant findings. Regarding the level of religious observance, we find that the Muslim prime increased donations only among pious respondents. This finding can be explained by an argument rooted in the political psychology of priming (Hart and Middleton Reference Hainmueller and Hangartner2014) that a prime will be meaningful if one’s prior on an issue corresponds with that prime. By contrast, we found that the Sunni prime works only for less pious. Although our finding may at first seem counterintuitive, the result corroborates our argument that the Sunni prime conveys a sectarian message that highlights a political divide (ingroup versus outgroup) rather than a religious message.
Regarding effects conditional on city, the finding that the Sunni prime has a stronger effect in Gaziantep can be explained by the intuition that its residents can more easily make sense of Syrian sectarian divides.Footnote 17 Lastly, regarding contact, we find some support for our original hypothesis that recategorization will have a stronger effect on prejudice reduction among subjects who experienced less contact with the refugees because these individuals would have less entrenched out-group attitudes. Specifically, we find that the Muslim prime had a stronger effect on donations among those who had no contact with refugees. However, our original hypothesis was challenged by our other findings. We found that the positive effect of the Sunni prime on acceptance of the refugees in the neighborhood, donations, and the composite index is only present among those who had contact with refugees. This constitutes a puzzle. A more complete accounting of this finding would require additional research in the future. We believe that one plausible explanation for this finding is that individuals with more out-group contact are likely to hold real-life images of the members of those outgroup (even if negative) that in turn make them more likely to cognize the somewhat abstract unifying categories that are experimentally induced.
Finally, our study demonstrates the sensitivity of recategorization success to one’s information environment. Although we do not find consistent evidence for the direct negative effect of exposure to economic cost information, we did find that when jointly applied with a religious prime, the economic cost information canceled out the pro-refugee effect of the religious prime. We thus find illuminating evidence that intergroup competition for resources moderates the effects of recategorization (Gaertner and Dovidio Reference Druckman2000) and, more generally, that competing frames counterbalance each other (Druckman Reference Dovidio, Johnson, Gaertner, Pearson, Saguy and Ashburn-Nardo2004).
Taken as a whole, our study demonstrates that even subtle primes in the form of identity cues and minor information provision can significantly affect behavior and to some extent attitudes toward the outgroup. Barabas and Jerit (2010) in their critique of priming studies and survey experiments emphasize the problems of strong stimuli and the lack of realism in experimental settings. Given that our study was conducted in the field and incorporates a real-world interaction during the request for charitable giving, we believe that our research meets the threshold for real-world relevance and does so with greater confidence than past laboratory or lab-in-field studies. Second, regarding stimuli strength, we reiterate that our interventions were subtle and even reminiscent of the political discourse in Turkey. We believe that researchers ought to employ specific treatments in order to identify the effects of interventions that matter and take place in everyday politics. Third, we note potential criticisms that survey experiments may overstate the effects of recategorization because such effects are likely to dissipate in a non-experimental context due to additional competing frames, time elapse, and subject attention. Consequently, we treat our results as upper bounds on the effects of recategorization in a real-world setting.
The non-representativeness of our sample—and the absence of female respondents, in particular—places important limitations on the generalizability of our results. For example, our survey was limited to male respondents, but previous work has found gender to be an important mediating factor in attitude formation, especially in the Middle (Bush and Jamal Reference Brewer and Miller2015). Our specific research interest in other mediating factors, such as religiosity and contact with refugees, coupled with resource limitations, led us to only sample male respondents. Future research could therefore produce an interesting extension of our experiment by exploring the effect of religious appeals conditional on gender. To examine the generalizability of our results in other settings, it would also be valuable to replicate our study in other countries. Lebanon would be especially appealing, especially as the sectarian nature of the Syrian refugee crisis there is even more controversial than in Turkey.
Another fruitful extension of our research includes research on the role of endorsements of particular views toward refugees (or immigrants) by prominent opinion leaders. In Turkey, for instance, one extension in this vein would be to measure the effect of appeals of support for the refugees from Recep Tayyip Erdogan or important religious leaders.
Finally, the timing of our study, the Muslim holy month of Ramadan, sets an additional scope condition for the interpretation of our findings. Our study was conducted during Ramadan because religious identities are especially salient at that time, enabling us to examine the relative import of subtle religious primes in a survey experiment. Moreover, precisely because we expect a higher baseline level of religious consciousness during Ramadan, it is all the more interesting that we found that subtle religious primes can still have an effect on behavior. That said, we might expect either stronger or weaker priming effects in another period. Indeed, one interesting extension of our study would be to conduct the study immediately before, during, and immediately after Ramadan to estimate priming effects conditional on the expected salience of religious identities.
Conclusion
We found that emphasizing the Sunni and Muslim identity of the Syrian refugees in Turkey reduced prejudice behavior as well as certain attitudes of Turkish respondents toward the refugees. Although the Muslim prime was more effective in increasing respondents’ charitable donations to refugees, we found that the Sunni prime was a more powerful means to reduce overall out-group prejudice. We also found that simultaneous exposure to information on the economic cost of Syrian refugees eliminated the pro-refugee effects of religious appeals, even for highly religiously observant Turks. Our findings advance the literature on prejudice reduction, and in particular the role of shared religious identities on out-group preferences and behavior. Our study also contributes to understanding of Turkish domestic politics as well as broader ethno-religious dynamics of the Middle East and Muslim world.
Supplementary Material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/psrm.2015.57
