Testing Intergroup Threat Theory: Realistic and Symbolic Threats, Religiosity and Gender as Predictors of Prejudice

The complex phenomenon of prejudice has been the focus of interest among social psychologists since the mid-20th century. The Intergroup Threat Theory (ITT) is one of the most efficient theoretical frameworks to identify the triggers of prejudice. In this study, using experimental design, we examined the effects of symbolic and realistic threats on prejudice that was measured by means of a modified social distance scale. The study participants were 611 undergraduate students from the country of Georgia. In addition to providing further support for ITT, the study showed that the level of religiosity moderated the effects between both types of threats and prejudice, although it had different indications for realistic and symbolic threats, while gender interacted only with symbolic threat. Implications of the findings are discussed.

ITT authors (Stephan et al., 2009) stress that the theory is concerned with perceived rather than actual threats, as "perceived threats have real consequences, regardless of whether or not the perceptions of threat are accurate" (Stephan et al., 2009, p. 45). The study of attitudes toward immigrants in Germany, for example, found that the actual proportion of immigrants did not predict negative attitudes toward them, but the perceived proportion of immigrants did (Semyonov, Raijman, Tov, & Schmidt 2004). However, studies (e.g., Quillian, 1995) and historical experience (Aronson, 2012) also suggest that actual threats -poor economy, a large portion of minorities or immigrants -enhance negative attitudes toward out-groups which, according to Riek et al. (2006) means that national problems, including economic hardships, are ascribed to threat-inducing out-groups.
ITT distinguishes two types of perceived threats: realistic and symbolic threats. Perceived realistic threat, the concept that has its roots in the Realistic Group Conflict Theory (e.g. Sherif, 1966), is a threat to the actualpolitical, economic or physical -well-being (land, security, health, wealth, employment) of a group, while perceived symbolic threat is concerned with a group's values, traditions, ideology, morals, and is expected to be more prominent when an in-group believes that their cultural values and traits are different from those of an outgroup (Zárate, Garcia, Garza, & Hitlan 2004).
Whether perception of symbolic or realistic threat becomes salient depends upon the threat-invoking out-group (Stephan et al., 2009). Economically powerful out-groups or people with diseases might elicit realistic threats (Stephan et al., 2005), while socially marginalized out-groups, such as homosexuals (Haddock, Zanna, & Esses, 1993) and sects (Stephan et al., 2009), engender symbolic threats.
However, drawing a clear line between symbolic and realistic threats may be problematic in certain cases (Riek et al., 2006) as they may overlap. For example, symbolic threat posed by a religious out-group might involve realistic threat as well or evolve into the latter (Riek et al., 2006). The present study tries to address this issue by experimentally manipulating symbolic and realistic threats. Another reason to explore threats through experimental manipulation is the questionable validity of the threat instrument that has been used in most of the studies (e.g., Aberson & Gaffney, 2008;Dunwoody & McFarland, 2018;Myers et al., 2013;Stephan, Ybarra, Martinéz, Schwarzwald, & Tur-Kaspa, 1998;Stephan et al., 2002). Specifically, threats have often been measured by questionnaires including 12 (e.g., Stephan et al., 2002) or fewer (e.g., Dunwoody & McFarland, Parrillo & Donoghue, 2005) argue that males, as compared to females, are more prone to being prejudiced.
At the same time, studies show cross-cultural differences in factors contributing to prejudice. For example, while race was identified as an important source of social distances among American students, religion and the employment status was found to contribute to negative attitudes in Greek and Japanese students, respectively (Triandis, Loh, & Levin, 1966). The study of the social distances of Georgian, German and Japanese students (Javakhishvili, Schneider, Makashvili, & Kochlashvili, 2012) found no significant effects of gender and religiosity on social distance scores in any of the three samples.

Aims and Hypotheses
Considering the real-life relevance of threats (e.g., Quillian, 1995), the aim of the current study is to show that threat-evoking contexts elicit prejudice.
Drawing upon ITT, we test the effects of both, symbolic and realistic threats on prejudice and by manipulating the two threatening situations through exposing the participants to information about fictitious out-groups, we examine whether they have different effects as compared to the situation where threat is absent (control condition).
Furthermore, we explore the demographic variables that have been shown to be associated with prejudice.
These variables include the participants' gender and the level of religiosity. Expecting that both factors are related to prejudice, their interactions with threats are also tested to examine whether any of the two moderates the relation between threats and prejudice and whether the interaction is different for realistic and symbolic threats.
Given that symbolic threats are related to traditions, customs and values (Stephan et al., 2009), we assume that the interaction effect between religiousness and this type of threat will be stronger than for the realistic threat.
were females, with the mean age of 20.37 (4.984), ranging from 18 to 25 years. The uneven gender distribution reflects the situation in the population of students from which the sample has been taken: this population consists of students from the humanities, social sciences and natural sciences, where females dramatically outnumber males. At the same time, the main aim of our design was to ensure equivalency between the three groups, which was met (see Table 1).
A test of homogeneity of variance was performed in order to check the equivalence of age distribution across experimental (symbolic and realistic threats) and control ("no threat") conditions. Group mean (p = .193) and median (p = .680) comparisons showed no significant differences between the groups (see Table 1). Similarly, a test of homogeneity of variance was used to look at the equivalence of distribution of the level of religiosity across the three groups. Again, no significant differences were found between experimental and control conditions based on the mean (p = .059) and median (see Table 1).
Crosstabs showed that gender distribution, although with predominantly more women in all the three groups, was still similar across the groups with 141 women versus 56 men in realistic threat condition, 131 women versus 65 men in symbolic threat condition and 151 women versus 45 men in control condition ("no threat).
Data collection took place between December 2015 to February 2017.

Procedure
Before starting data collection, a detailed research project was submitted to the ethics committee of Ilia State University. Data collection started only after the ethical approval was obtained.
Having obtained approval from the universities, we entered the classes and asked the students if they wished to take part in the study of the role of mass media in a daily life. Those who agreed to participate were led to believe that the study was concerned with the daily coverage of events and the role of media in shaping public attitudes. They were asked to read a copy of a newspaper article and answer several questions, after which they were to fill out an eight-item attitude questionnaire about an out-group mentioned in the article and to provide demographic data. After completing all of the tasks, the participants were thoroughly debriefed: the true objectives of the study were disclosed and the students were given sufficient time to ask questions.
Testing Intergroup Threat Theory 468

Design
The study used a between-subjects factorial design with one factor, perceived threat, which was examined at three levels: realistic (n = 197) and symbolic threat (n = 196) in experimental conditions, and the absence of threat in the control condition (n = 196).

Measures Threat Manipulation
In order to manipulate realistic and symbolic threats, several vignettes were composed -a page length newspaper-type articles about a fictitious out-group called "Abirians" (invented solely for research purposes).
The articles designed for experimental conditions (i.e., symbolic and realistic threats) contained information that the UN planned to settle in Georgia an ethnic group of Abirians, who currently resided in the Gaza Strip, and described potential threats of the plan for Georgia.
a. Realistic threat manipulation: in one version of the articles, the settlement of the Abirians in Georgia was associated with realistic threats: competition for jobs and higher education, increased taxes and the possible spread of diseases such as hepatitis C. An excerpt from the article read as follows:

As reported [by the organization], Abirians will immediately need workplaces on the local labor market upon their arrival in Georgia. According to the International Index of Education and Employability (IIEE),
Abirians are quite similar to leading European nations. This means that they can easily compete with the local population in terms of both, education and labor market.
b. Symbolic threat manipulation: another version described symbolic threats and focused on the types of cultural differences which, according to a number of recent opinion polls, Georgians are particularly sensitive to: the out-group was reported to have a non-uniform religion, favor same sex marriage and adhere to such traditions and values that are clearly uncommon in Georgia. c. Control condition: the third version, designed for the control group ("no threat" condition), was similar in structure and style but described the out-group (Abirians) in neutral terms, with no references to threats whatsoever (it did not mention the prospects of the Abirians' settlement in Georgia). For example:

As reported by PCW, the census of an ethnic group Abiri residing in Gaza Strip is already available.
Historical sources suggest that the Abirians have been residing in the territory since the 1830s when Egypt conquered Palestine, and later, as a result of the British intervention, the Ottoman Empire regained control over the region.
To ensure that the participants read and comprehended the articles they were given several questions; for example, the Likert-scale items: "Please evaluate the style of the article. Did you find it objective?"

Makashvili, Vardanashvili, & Javakhishvili 469
Level of religiosity was measured by a single 10-points scale item with 1 = not at all religious and 10 = highly religious.
Prejudice was measured by an updated version of the Social Distance Scale (Bogardus, 1925 (Javakhishvili, 2005;Javakhishvili, 2011;Javakhishvili et al., 2012;Javakhishvili et al., 2016;Vardanashvili & Javakhishvili, 2016), the scale is a good measure of explicit prejudice as it is assumed to tap into the behavioral component of prejudice, versus, for instance, the feeling thermometer that tests the affective aspect of prejudice (Henry, 2008;Stefaniak & Bilewicz, 2016).
A revised version of the scale included eight 5-points Likert-scale items, e.g., "I would accept Abirians as my child's teacher" or "I would make friends with Abirians," with 1 = strongly agree and 5 = strongly disagree, higher scores indicating higher levels of social distance, and hence, prejudice, and lower scores indicating lower levels of prejudice.
The validity of the measure was examined using the confirmatory factor analysis (CFA) in Mplus (Muthén & Muthén, 2010). The first model considered one latent variable (social distance as a single construct) predicted by the respective eight items. The model did not fit the data. We then divided the scale into two parts, each representing a latent factor. The first one included three items, while the second factor covered the remaining five items.
Therefore, the Social Distance Scale incorporated two factors. For the first factor the Cronbach's Alpha was .829, while for the second factor it was .879. Consequently, two separate scores of the dependent variable for the first factor and the second factor were used in the further analysis. Data of factor loadings are provided in Table 2.

Results
Descriptive statistics showed that threat manipulation had an effect on both factors of prejudice toward Abirians (see Table 3). The differences between the mean scores of prejudice across realistic threat and control ("no threat") condition as well as symbolic threat and control ("no threat") condition were verified by the one-way analysis of variance (ANOVA). No significant differences were found between the realistic and symbolic threat conditions. Mean scores of prejudice where higher for the first factor of prejudice, F(2, 586) = 25.45, p < .001 than for the second factor, F(2, 586) = 24.24, p < .001 (see Table 3). The level of religiosity positively correlated with both forms of prejudice: a higher correlation was found for the first factor of prejudice, r(582) = .40, p < .001 as compared to the second factor, r(582) =.28, p < .001.
Since the level of religiosity was positively related to both factors of prejudice and gender had a significant effect on the second factor, simple regression was performed with threats and these variables as predictors.
Initially, we examined the predictors of the first factor of prejudice. The latter was entered as a dependent variable in the regression equation. Threat was turned into a dummy variable so that we were able to enter realistic and symbolic threats as predictors in Model 1. Both types of threats significantly contibuted to the variance of prejudice, R 2 = .08, F(2, 581) = 24.87, p < .001. The level of religiosity was added in Model 2, resulting in ΔR 2 of 15% with threats retaining significance and religiosity predicting prejudice, R 2 = .22, F(3, 581) = 46.68, p < .001. Results of regression are shown in Table 4.
Likewise, we ran simple regression for the second factor of prejudice. Realistic and symbolic threats (turned into dummy variables) were entered in Model 1. Both of them again predicted prejudice, R 2 = .08, F(2, 580) = 24.01, p < .001. Gender (turned into a dummy variable) and the level of religiosity were added in Model 2.
Makashvili, Vardanashvili, & Javakhishvili 471  22% of the variance of the first factor of prejudice was explained by symbolic and realistic threats and religiosity, while 16% of the variance of the second factor was explained by symbolic and realistic threats and religiosity plus gender. According to β-coefficients, for both, the first and the second factors, symbolic threat was a better predictor than realistic threat.
Since religiosity and gender contributed substantially to this association, we performed a moderation analysis by means of a special software Process, which is a supplement to SPSS (Hayes, 2013). This statistical technique uses regression to provide model outputs.
For the first factor of prejudice, the moderation of the effect of realistic and symbolic threats was tested by the level of religiosity, while for the second factor we examined the moderation of the effect of threats by the level of religiosity and gender. In total, four models were tested: 1. Realistic threat → Religiosity → The first factor of prejudice 2. Symbolic threat → Religiosity → The first factor of prejudice 3. Realistic threat → Religiosity → Gender → The second factor of prejudice 4. Symbolic threat → Religiosity → Gender → The second factor of prejudice Testing Intergroup Threat Theory 472 Model 1: Realistic threat → Religiosity → The first factor of prejudice. Overall, the model was significant: F(3, 578) = 47.12, p < .001, R 2 = .20. b coefficients are presented in Table 6.

Conditional effect of realistic threat on the first factor of prejudice at 3 different values of the moderator -level of religiosity
Level of religiosity 1 SD below mean -3.95 0.58*** 5.02 Level of religiosity at the mean -6.36 0.26*** 3.20 Level of religiosity 1 SD over mean -8.77 0.06 -0.53 ***p ≤ .001.
All of the effects -the main effects of realistic threat, the level of religiosity as well as their interaction -were significant, which indicates the moderating effect of religiosity. The most interesting information is provided in the second, lower part of the table: b coefficient decreased as the level of religiosity increased and, finally, it became nonsignificant (p = .595).
As in the case of realistic threat, religiosity moderated the association. Main effects of symbolic threat, the level of religiosity and their interaction were significant, although the direction of association was reversed: b coefficient and the level of religiosity increased simultaneously.

Conditional effect of realistic threat on the second factor of prejudice at different values of the moderators -3 levels of religiosity and two levels of gender
Level of religiosity 1 SD below mean -3.95 -for women 0.42** 3.81 Level of religiosity at the mean -6.35 -for women 0.13 1.68 Level of religiosity 1 SD over mean -8.76 -for women 0.15 -1.44 Level of religiosity 1 SD below mean -3.95 -for men 0.37** 2.69 Level of religiosity at the mean -6.35 -for men 0.09 0.68 Level of religiosity 1 SD over mean -8.76 -for men 0.20 -1.27 **p ≤ .01.
For this model, only the level of religiosity proved to be a moderator in the association between realistic threat and the second factor of prejudice. All main and interaction effects were significant except the interaction of realistic threat and gender. For men as well as for women, effect of threat decreased in line with the increase of religiosity; in fact, it became nonsignificant.

Conditional effect of symbolic threat on the second factor of prejudice at different values of the moderators -3 levels of religiosity and two levels of gender
Level of religiosity 1 SD below mean -3.95 for women -0.18 -1.64 Level of religiosity at the mean -6.35 -for women 0.16* 2.01 Level of religiosity 1 SD over mean -8.76 -for women 0.50*** 5.13 Level of religiosity 1 SD below mean -3.945 -for men 0.12 0.89 Level of religiosity at the mean -6.35 -for men 0.46*** 3.83 Level of religiosity 1 SD over mean -8.76 -for men 0.80*** 5.73 *p ≤ .05. ***p ≤ .001.

Testing Intergroup Threat Theory 474
The main effects of symbolic threat and the level of religiosity were significant, while that of gender was not.
Both interactions were significant, which indicates the moderating effect of religiosity as well as gender. The effect of symbolic threat grew along with the level of religiosity both for men and women. However, effects of symbolic threat were stronger for men than for women.

Discussion
The findings provide evidence that the manipulations of realistic and symbolic threats were successful and confirm that threats alone have a potential to account for an increase in prejudice, as has been shown by previous studies (e.g., Aberson & Gaffney, 2008;Riek et al., 2006;Stephan & Stephan, 1996;Stephan et al., 2002).
Additionally, along with realistic and symbolic threats, the level of religiosity significantly predicted both factors of prejudice, while gender emerged as a fourth predictor for the second factor.
The positive relationship between religiosity and prejudice has been shown in a number of studies (Allport & Kramer, 1946;Hall et al., 2010;Rowatt et al., 2009;Scheepers et al., 2002;Ugurlu, 2013), and as Gorsuch and Aleshire (1974) argue, religiousness is linked with higher levels of conventionality and lesser acceptance of people with different norms.
However, although a number of authors (e.g., Altemeyer, 1998;Parrillo & Donoghue, 2005;see McFarland, 2010) suggest that men, compared to women, are more prone to being prejudiced, the gender is not always positively associated with prejudice (see Herek, 2002;Hughes & Tuch, 2003;Javakhishvili et al., 2012). For example, Ekehammar, Akrami, and Araya (2003) experiments revealed that males scored higher on explicit prejudice measure, while women showed higher levels of implicit prejudice (Ekehammar et al., 2003). To some extent, the inconsistency of the evidence was reflected in our findings as gender predicted only the second factor of prejudice. Since the second factor implies less delicate relations with an out-group, the result might indicate that men and women tend to have similar attitudes when it comes to more delicate relations (the first factor of prejudice) with an out-group, while the distinction comes into play with less delicate (the second factor of prejudice) relations.
The moderational analysis showed that the interaction effect between realistic threat and religiosity was similar for both factors of prejudice. Religiosity moderated the association between realistic threat and prejudice: specifically, higher levels of religiosity were linked with a lower effect of realistic threat on both factors, which suggests that people with lower levels of religiosity tend to be less sensitive to perceiving realistic threat.
According to Cohrs and Ibler (2009), a moderational approach assumes that either threat perception might happen due to preexisting levels of certain personal dispositions (religiousness in this case) or that the dispositions "are activated from memory and rendered psychologically salient and influential by contextual features" (Cohrs & Ibler, 2009, p. 82) that correspond to the content of those personal dispositions. In line with this reasoning, our findings suggest that realistic threat ("contextual feature") activates religiousness or that the preexisting levels of religiousness trigger perceptions of realistic threat.
Makashvili, Vardanashvili,& Javakhishvili 475 The interaction effect between symbolic threat and level of religiosity was also similar for both factors of prejudice and, at the same time, opposite to that between religiosity and realistic threat: religiosity moderated the association between symbolic threat and prejudice however, higher level of religiosity was associated with a higher effect of symbolic threat on both factors. The finding has an implication for our hypothesis which proposed that the interaction effect between religiosity and symbolic threat would be stronger than that between religiosity and realistic threat. Although interaction coefficients were almost similar for realistic (b = -0.13 for the first factor of prejudice and b = -0.12 for the second factor) and symbolic (b = 0.14 for both factors) threats, the finding suggests that the level of religiosity is more salient in case of symbolic threat perception, indicating the potential mediational chain between the variables (Baron & Kenny, 1986), which needs to be addressed in further analysis.
The results of the moderational analysis, altogether, correspond to the conceptualizations of the types of threats. Since symbolic threats include threats to traditions and values (Stephan et al., 2009), they are experienced on a more personal level and can be engendered by the out-groups that are perceived as having different traditions and values by people who embrace these values and traditions (Stephan, Ybarra, & Bachman, 1999). Religiosity which, in a broader sense can be regarded as an extension of traditions and values, tends to make people more vulnerable in the face of the out-group that adheres to different religious practices and norms. Indeed, according to Schwartz et al. (2012), traditionalistic values positively correlate with the importance of religion in one's life.
On the other hand, realistic threats involve more rational responses in general, such as seeking information about an out-group and negotiating with it (Stephan et al., 2009), which might account for why religiosity -a less rational phenomenon positively related to likewise less pragmatic phenomena such as rigid morality, conventionality, the need for closure (see Jost, Glaser, Kruglanski, & Sulloway, 2003) and supernatural beliefs (Oliver & Wood, 2014) -takes over realistic threat in explaining prejudice.
The only association where gender had a moderational effect was that between symbolic threat and the second factor of prejudice: the effect of symbolic threat was stronger for men than for women, a finding which is somewhat controversial since realistic threats deal with competition over power and resources (Stephan et al., 2009), and males, in general, are more likely to compete over power. As Schwartz (2006) showed, being male is positively associated with power values. Furthermore, males tend to score higher on Social Dominance Orientation (SDO) (Sidanius & Pratto, 1999) which is a personal orientation supporting hierarchies and power dominance (Sidanius & Pratto, 1999 While an insecure or, in other words, threat-arousing environment might give rise to religious sentiments (e.g., Greenberg, Pyszczynski, & Solomon, 1986;Kekelia et al., 2013), the opposite can also be true: religious sentiments might give rise to threat perceptions (Cohrs & Ibler, 2009). However, in both cases, as our results have demonstrated, threats gain importance in shaping prejudice -a finding which, again, is particularly relevant for Georgia, where the population tends to face instability, while unemployment, economic crisis and territorial integrity (one-third of the Georgian territory is occupied by Russia) are perceived to be top three concerns (International Research Institute, 2017). Given the context, people are more prone to experience threats from out-groups, which leads to prejudice (Aronson, 2012;Quillian, 1995). Real-life incidents of explicit prejudice indeed fit into this reasoning. The examples include religious discrimination having taken place in one of the Georgian villages when the Orthodox Christians attacked local Muslim population to hinder their religious service ("Axali detalebi," 2012) or, to cite a more recent example, the rally organized by the so-called Georgian March occupied the capital streets, overtly expressing ethnic discrimination and stating that their goal was to "clear Georgia from illegal immigrants" (Gvarishvili, 2017).
Finally, the present study provides further support for ITT, demonstrating that both, realistic and symbolic threats account for prejudice, while the level of religiosity and being male (only in case of the second factor of prejudice) are additional contributors.
One of the most obvious limitations of the study is the specific age-group of the sample, which prevents generalization to other age groups. This also leads to the question of external validity: the fictitious nature of the threat-inducing out-group in the current study raises a concern whether the findings can be inferred to similar situations in real-life.
Further analysis is needed to identify the mechanism of how the variables in question are ordered in explaining prejudice. The mediational analysis can clarify whether the association between symbolic threat and prejudice is mediated by the level of religiosity or if symbolic threat itself is a mediator.
Additionally, as the relationship between religiosity and prejudice is not always positive and uniform (e.g., Herek, 1987;Laythe et al., 2001), suggesting that the concept includes more than one dimensions (e.g., Laythe et al., 2001) either positively or negatively related to prejudice, it would be more informative to measure these dimensions and gain a better insight in the concept, as well as to determine which dimensions of religiosity predicts prejudice and how the latter is connected to other well-established predictors, such as right-wing authoritarianism and SDO (McFarland, 2010). The role of gender also needs further research and might be better understood in case of a more balanced distribution.