Cultural threat perceptions predict violent extremism via need for cognitive closure

Significance

Triangulating methodological approaches (correlational, experimental, meta-analysis, and field research) with diverse samples (i.e., general populations and former violent extremists), we show that cultural threat perceptions lead to extremist tendencies because they activate a desire for definitive answers and solutions (i.e., a need for cognitive closure). Whereas approaches to violent extremism often focus primarily on situational influences and sometimes psychopathology, the present conceptual integration illuminates a key nonpathological psychological process underpinning violent extremism by demonstrating the pathway from cultural threat to a greater need for cognitive closure to greater endorsement of extremism. These findings contribute to a better understanding of the processes involved in the development of violent extremism, which can help limit its spread.

Abstract

Understanding the psychological processes that drive violent extremism is a pressing global issue. Across six studies, we demonstrate that perceived cultural threats lead to violent extremism because they increase people’s need for cognitive closure (NFC). In general population samples (from Denmark, Afghanistan, Pakistan, France, and an international sample) and a sample of former Mujahideen in Afghanistan, single-level and multilevel mediation analyses revealed that NFC mediated the association between perceived cultural threats and violent extremist outcomes. Further, in comparisons between the sample of former Afghan Mujahideen and the general population sample from Afghanistan following the known-group paradigm, the former Mujahideen scored significantly higher on cultural threat, NFC, and violent extremist outcomes. Moreover, the proposed model successfully differentiated former Afghan Mujahideen participants from the general Afghan participants. Next, two preregistered experiments provided causal support for the model. Experimentally manipulating the predictor (cultural threat) in Pakistan led to higher scores on the mediator (NFC) and dependent variables (violent extremist outcomes). Finally, an experiment conducted in France demonstrated the causal effect of the mediator (NFC) on violent extremist outcomes. Two internal meta-analyses using state-of-the-art methods (i.e., meta-analytic structural equation modeling and pooled indirect effects analyses) further demonstrated the robustness of our results across the different extremist outcomes, designs, populations, and settings. Cultural threat perceptions seem to drive violent extremism by eliciting a need for cognitive closure.

The rise of violent extremism has become an alarming global phenomenon, transcending geographical and cultural boundaries (1, 2) posing a significant threat to international peace and security. In recent years, high-profile attacks carried out by extremists, such as Jihadists, right-wing extremists, and other groups (1, 3, 4) have highlighted the urgent need for a comprehensive understanding of the psychological processes driving individuals toward this violent behavior.

Triangulating methodological approaches (correlational, experimental, meta-analysis and field research) with diverse samples (i.e., general populations and former violent extremists) living in various parts of the world, we empirically investigated a key psychological process that may underpin violent extremism. Specifically, we show that cultural threats predict violent extremist tendencies because they activate a desire for definitive answers and solutions, as reflected in higher levels of need for cognitive closure (i.e., NFC) (5). While we acknowledge that violent extremism exists across ethnic and religious groups, here we focus on the case of Jihadist extremism, which has been underexplored in psychological studies (6). Hence, we examined our hypotheses in various cultural contexts marked by specific societal and political events representing flashpoints for recent Jihadist terror attacks in Europe (1).

We focus on Jihadist extremism for several reasons. First, while countries such as Afghanistan and Pakistan are heavily affected by terrorism (1), there is limited psychological research into the causes of violent extremism among non-Western populations. Second, recent terrorist attacks by individuals or groups claiming to act in the name of Islam have fueled widespread fear and suspicion of Muslims and the Islamic faith, leading to harmful stereotypes that Jihadist extremism is inherently rooted in broader Islamic cultural and religious values (7). By identifying how contextual factors (i.e., threat perceptions) influence violent extremist intentions, we aim to demystify the violent extremism of non-Western groups by attributing it, at least in parts, to basic psychological processes and personality traits (i.e., NFC) that are likely to be prevalent across groups and cultures. Thus, the findings from the present research may be expected to generalize to any form of violent extremism driven by threat perceptions, including Western groups such as right-wing extremists who perceive their status, power, and group dominance as threatened by minorities and immigration (8).

Perceived Cultural Threat

Social identity and self-categorization theories critically distinguish between the dynamics of personal and group identities generally (9, 10), as well as in responses to threats. With respect to social identity threat, according to intergroup threat theory (11), perceiving that other groups threaten one’s group’s culture not only lead to avoidance, negativity, and prejudice, but also to more extreme forms of outgroup hostility (12, 13). Indeed, such threat perceptions have been positively associated with extremist tendencies in various populations. For example, Obaidi et al. (14) documented that individuals in Europe, the US as well as Afghanistan and Turkey express higher levels of hostility toward the respective out-group (e.g., toward Muslims or the West) when they perceive that their cultural practices and values are being threatened. Similarly, threatening conspiracy theories such as “The Great Replacement” have been shown to predict violent extremism among non-Muslim majority-group members in a set of Western countries (8). Other scholars specifically focusing on non-Western groups have shown that outgroup hostility toward the West among Muslims living in Europe or the Middle East is embedded in cultural threats originating from perceived value incompatibility (15), criticism of Islamic culture, and irreconcilable ways of life (16).

Accordingly, in the present paper, we hypothesized that Muslim minorities in the West (e.g., Denmark and France) as well as Muslims in Muslim-majority countries (e.g., Afghanistan and Pakistan) who perceive a greater cultural threat would also display greater intentions to engage in and support anti-Western violence (H₁). Previous cross-cultural studies have shown that a common psychology of outgroup hostility and violence was explained by perceived cultural threat across various cultural groups and contexts (14). Therefore, we expect support for the proposed model across different populations and cultural contexts, but we are agnostic about the variation in effect sizes.

Need for Cognitive Closure.

NFC is a psychological construct that reflects a desire for structure, predictability, and decisiveness (17) (for a review, see ref. 18). People high in NFC tend to find ambiguity unpleasant and adhere to black-and-white, authoritarian and dogmatic beliefs (19). It is often argued that NFC constitutes a key component of the extremist mindset (20, 21), and extremist ideologies that are marked by a relatively simplistic, black-and-white understanding of the social world are often appealing to people with a heightened NFC (22). For instance, compared with political moderates, extremists categorize the social world into simpler, homogenous, and rigid categories (23). These theoretical insights therefore suggest a link between NFC and extremism (24). Indeed, greater NFC predicts more extreme positions relative to outgroups, including religious fundamentalism (22) and support for torture (25). It is important to note that these biases and positions are generally associated with right-wing orientations, and empirical research directly linking NFC to violent Jihadism is limited (cf. ref. 26). Given that existing research has primarily been conducted with Western samples and within Western contexts, we tested the prediction that higher levels of NFC would be associated with and causally predict more violent extremist intentions among non-Western samples (H₂).

NFC as Mediator of the Relationship between Cultural Threat and Violent Extremism.

Although NFC has typically been conceived as a stable individual difference, situational influences and associated motivations can influence trait variables in systematic ways (27). It has long been noted that threat perceptions and motives to resolve ambiguity can prompt various cognitive and/or motivational strategies that are relatively rigid and simple (18, 28). Indeed, substantial evidence suggests that various threat stimuli decrease cognitive capacity and motivation such as cognitive flexibility (29), information processing (30), and information searches (18, 31). Furthermore, when people feel threatened, they respond by becoming more closed-minded (32). Studies have shown that various threat stimuli (e.g., existential, mortality, economic, and terrorist) elicit dogmatic beliefs and hostile worldview defenses (33–35), resembling a cognitive response to threat and uncertainty (36). Situations requiring difficult or unpleasant information processing or strong group adherence may increase NFC (37).

A major focus of the current research is on how perceived cultural threat, a form of social identity threat, can increase NFC, which in turn can promote violent extremism. Our predictions are reconcilable with several major social-cognitive models of aggression and violence that illustrate how situational factors influence cognitions, which in turn influence behavioral outcomes. In line with the General Aggression Model (38, 39) and the Cognitive-Neoassociation Theory of Aggression (40), contextual perceptions (i.e., threat) are expected to influence cognitive processing (i.e., NFC), which in turn can enhance aggressive inclinations (i.e., violent behavioral intentions). Several findings are supportive of this reasoning. In a series of studies among majority populations by Orehek et al. (25), reminders of the physical threat of terrorism activated NFC, and elevated NFC led to out-group negativity such as support for tough and decisive counterterrorism policies (see also ref. 24). Moreover, whereas threat potentially can increase NFC and extremism, the nature of the threat (for example, personal vs. group threat (10); symbolic vs. realistic threat 11) critically shapes the strength and type of responses. For example, Webber et al. (26) found in a correlational study with Muslim participants from the Philippines that a personal loss of significance that creates shame and uncertainty about the self is related to a greater need for closure and, ultimately, more extremism.

As previously noted, in contrast to existing research, we focus on the role of group-related cultural threat. We concentrate our work on violent extremism in the Middle East, South Asia, and the West, thus offering a conceptual insight into an issue of pressing practical importance. Specifically, we tested in multiple ways a mediation model in which cultural threat increases NFC, which mediates the effect of cultural threat on violent extremism (H₃).

Study 1

We tested our three hypotheses in six studies. Study 1 was conducted among 260 Muslims living in Denmark. We chose Denmark because relationships between Muslims and the ethnic majority population have become highly salient in the country, particularly following Jyllands Posten’s publication of cartoons depicting the Prophet Muhammad and the recent violent attack on the Prophet Muhammed cartoonists in Copenhagen. We assessed participants’ perceptions of cultural threat, NFC, and intentions to defend one’s in-group and religion. For an overview of the variables and items included in each study, see SI Appendix, Tables S4–S10. Descriptive statistics, reliability coefficients, and correlations for study 1 through study 6 are presented in SI Appendix, Tables S12–S17.

To evaluate the fit of our models with the data, ensure the robustness of our findings, and reduce confounding, we conducted the analyses including the available covariates in correlational studies 1–4. These covariates include basic demographic variables (e.g., age, gender, socioeconomic status), as well as individual differences related to key elements of identity (religiosity, Muslim identification), and perceptions related to intergroup relations (religious discrimination, perceived injustice) that could confound the exposure-mediator, exposure-outcome, and mediator-outcome relationships.

The fit of the single-level mediation model was satisfactory, χ² (10) = 18.302, P = 0.050, Standardized Root Mean Square Residual (SRMR) = 0.029, Comparative Fit Index (CFI) = 0.949, Root Mean Square Error of Approximation (RMSEA) = 0.058, and supported our hypotheses. As displayed in Fig. 1, NFC mediated the relationship between threat perceptions and willingness to defend one’s in-group and religion. The bootstrapped indirect association of perceived cultural threat on willingness to defend one’s in-group and religion was significant; B = 0.02, SE = 0.02, 95% CI [0.002, 0.063]. Please note that all indirect associations in this and the remaining correlational studies (studies 2–4) held when we conducted the analysis without covariates. We present these results in SI Appendix, Fig. S1.

Fig. 1.

Single-level mediation (study 1), multilevel mediation models (studies 2–5), and internal meta-analysis (studies 1–6). Note. c′ = direct effect; Unstandardized coefficients are displayed and 95% CI are reported in brackets. The correlational studies 1–4 control for a set of covariates. Findings held when not controlling for covariates, see SI Appendix.

Study 2

Having obtained evidence from a context where Muslims form a stigmatized minority, we set out to test whether we could replicate our model in the Muslim-majority country of Afghanistan. Following 9/11, Afghanistan was invaded by the US and its Western allies, and since then the country has become the epicenter of intergroup clashes between various domestic and regional factions and the West, in particular between NATO forces, Mujahedeen, and the Taliban. In this context, we investigated Afghans’ perceived cultural threat and the mediating effect of NFC on violent extremism outcomes. In study 2, we also addressed one shortcoming of study 1. Instead of measuring the willingness to defend one’s in-group and religion, which by no means necessarily involves violence, we measured four different types of violent extremist outcomes, see SI Appendix, Table S6.

In this and all remaining studies that tested mediation models and included multiple violent extremist outcomes, we estimated 2-2-1 two-level models (41) using multilevel structural equation modeling (see SI Appendix, for a technical description of this analysis, p. 14). These models, which estimate effects across the dependent outcomes that are nested within participants, have the advantage of providing robust effect estimates that generalize across specific measures. However, for full transparency, we also present single-level mediation models estimated separately for each dependent measure for this and the remaining studies in SI Appendix, Table S19–S22 and Figs. S2–S5.

The model had an excellent fit, χ² (6) = 8.649, P = 0.194, SRMR_within = 0.005, SRMR_between = 0.033, CFI = 0.990, RMSEA = 0.023, and supported the prediction that perceived cultural threat is indirectly linked to violent extremist outcomes through higher NFC (Fig. 1). Bayesian credible intervals showed that the indirect association was significant: B = 0.71, posterior SD = 0.04, 95% CI [0.004, 0.144]. Of note, effects in the single-level mediation model were also significant for all dependent variables (SI Appendix, Table S19 and Fig. S2).

Study 3

Although the findings from studies 1 and 2 are valuable for understanding violent extremism, they were conducted among general populations. Hence, we aimed at increasing the ecological validity of these results with a sample of former Mujahideen in Afghanistan. Many Afghan Mujahedeen were accused of acts of violent extremism, including war crimes and human rights abuses in the civil war of the 1990s (42). As in study 2, we assessed participants’ perceptions of cultural threat, NFC, and four violent extremist outcomes, see SI Appendix, Table S7.

The multilevel mediation model had an excellent fit, χ² (3) = 4.152, P = 0.246, SRMR_within = 0.001, SRMR_between = 0.054, CFI = 0.968, RMSEA = 0.037, and supported our predictions as the perceived cultural threat was positively related to violent extremist outcomes through higher NFC, see Fig. 1. The resulting indirect association was significant according to Bayesian credible intervals: B = 0.10, posterior SD = 0.05; 95% CI [0.024, 0.220]. The indirect associations were also significant for all dependent outcomes except for support for violent extremism, when single-level mediation models were estimated separately (SI Appendix, Table S20 and Fig. S3).

For both practical and ethical reasons, our outcomes of interest in the studies that we have reported thus far were extremist attitudes and violent intentions, not actual engagement in violent behavior. Therefore, to examine the known-group validity (43, 44)—a type of construct validity—of our self-report measures, we pooled the sample of former Mujahedeen (from this study) with the male sample of Afghans from study 2. This male sample was selected for comparability with the former Mujahedeen sample; all participants in the former Mujahedeen sample were men. The comparison between former Mujahedeen participants and the general sample of Afghans allowed us to conduct two critical tests. First, it enabled us to compare the mean differences on our variables of interest between the two groups. We proposed that if the former Mujahideen participants scored higher, on average, on threat, NFC, and violent attitudes and intentions than those in the general sample, this would provide evidence that the results go beyond self-reported findings. Second, the pooled data allowed us to test whether our proposed mediation model could statistically distinguish former Mujahideen from the Afghan general population. In this model, membership in an extremist group was thus treated as the dependent variable. This outcome is of particular interest as it assesses the behavioral validity of our theoretical model that in the other studies only was tested with self-report outcomes.

The reported mean differences were theory-consistent—the former Mujahideen scored significantly higher on each violent extremist outcome. Furthermore, the former Mujahideen scored significantly higher on perceived cultural threat and NFC (Table 1).

Table 1.

Means on key variables for the in the former Mujahideen and Non-Mujahideen samples (male participants only)

Variables	Former Mujahideen			Non-Mujahideen			t (162)	Cohen’s d	P (2-tailed)
	M	SD	SE	M	SD	SE
Willingness to defend Islam/Muslims	4.95	1.03	0.12	4.11	1.21	0.12	4.69	0.74	<0.001
Violent behavioral intentions (VBI-2)	4.19	1.21	0.15	3.76	1.11	0.11	2.37	0.37	0.019
Violent behavioral intentions (VBI-7)	4.20	0.94	0.11	3.64	0.93	0.10	3.81	0.60	<0.001
Support for violent extremism	4.23	1.46	0.18	3.82	0.99	0.10	2.08	0.33	0.039
Need for cognitive closure	4.70	0.94	0.11	3.90	0.96	0.10	5.29	0.84	<0.001
Perceived cultural threat	4.58	0.99	0.12	3.56	1.16	0.12	5.96	0.95	<0.001

Note. VBI-2 scale measured with two items.

VBI-7 scale measured with seven items.

Next, we estimated a structural equation model with a binary outcome variable (0 = general population, 1 = former Mujahideen), NFC as a mediator, and cultural threat as the independent variable (Fig. 2). In this fully saturated mediation model, cultural threat predicted NFC, and both threat, OR = 2.06, 95% CI [1.456, 2.927], and NFC, OR = 2.09, 95% CI [1.318, 3.307], predicted a higher likelihood that participants would be part of the former Mujahedeen group. In other words, scoring higher on the variables predicted a twice as high likelihood of being a former Mujahedeen. Bootstrapping further showed that cultural threat indirectly predicted, through greater NFC, a higher likelihood of belonging to the former Mujahedeen [β = 0.13, 95% CI (0.043, 0.212)]. Thus, the fact that we find converging evidence from a “known-group paradigm” (general population vs. former Mujahideen) offered evidence of the construct validity of our measures and our proposed model.

Fig. 2.

Validation model (male general population and former Mujahideen).

Note. c’ = direct effect; Standardized coefficients are displayed and 95% CI are reported in brackets.

Study 4

Studies 1 through 3 were conducted in two settings where intergroup relations between Muslims and the Western majority populations have deteriorated in recent years. This could suggest that such findings may be tied to the specific context in which Muslims are enmeshed in intergroup conflict with Western populations, limiting the generalizability of our findings across other cultural contexts. In study 4, we therefore aimed to replicate our theoretical model with a diverse and multinational sample of Muslims. We assessed participants’ perceptions of cultural threat, NFC, and two violent extremist outcomes (SI Appendix, Table S8).

The multilevel mediation model had a satisfactory fit, χ² (10) = 28.583, P = 0.002, SRMR_within = 0.000, SRMR_between = 0.060, CFI = 0.91, RMSEA = 0.068, and supported our predictions (Fig. 1). The indirect association of perceived cultural threat on violent extremist outcomes through NFC was significant: B = 0.03, posterior SD = 0.02, 95% CI [0.006, 0.070]. When estimating single-level mediation models for each violent extremist outcome separately, the indirect association was significant for violent behavioral intention (VBI-2, measured with two items) but fell below significance for violent behavioral intention (VBI-7, measured with seven items)^* (SI Appendix, Table S21 and Fig. S4).

As studies 1 through 4 relied on correlational mediation models, we tested alternative models in which a) NFC was the predictor and perceived cultural threat the mediator or b) NFC moderated the effect of cultural threat on the dependent outcomes. With two exceptions (e.g., VBI-7 and Support for violent extremism in study 3), no evidence for alternative moderation models was obtained for the remaining 11 dependent variables (SI Appendix, pp. 25–26). However, the alternative mediational models in which NFC promotes violent extremist outcomes through cultural threat perceptions obtained some, albeit less consistent support (SI Appendix, Tables S23–S26 and Figs. S6–S9). The finding that cultural threat perceptions mediated the effect of NFC on violent extremist outcomes indicates that the theoretical propositions reflecting both the dispositional and the situational conception of NFC are valid and emphasize the importance of obtaining causal evidence for our hypothesized model through experiments.

Study 5

Whereas the previous studies correlationally supported our three hypotheses in diverse and hard-to-reach populations, they are not suited for drawing causal conclusions (see ref. 45). We, therefore, conducted a preregistered study in Pakistan that experimentally manipulated the predictor cultural threat perception and examined its causal effect on violent extremist outcomes via NFC. Pakistan was chosen because it is one of the largest Muslim-majority countries (with a population of over 220 million), hosting conservative Islamic values that often clash with Western values.

Participants were randomly assigned to read either a) statements ostensibly written by Western politicians, military personnel, and the public that were designed to be culturally threatening to Muslims (experimental condition) or b) a neutral text about mangrove forests (control condition), see SI Appendix for details, p.  30. Having read the text, participants completed a manipulation check (i.e., threat perceptions), the mediator (i.e., NFC) and three dependent outcomes (i.e., violent extremist outcomes; see SI Appendix, Table S9).

In addition to support for violent extremism and violent behavioral intentions to fight against the West, we included a measure of violent behavioral intentions against a non-Western group that is westernized yet not involved in the value conflict between the Western and Muslim worlds (i.e., Japanese people). The inclusion of this variable allowed us to test the discriminant validity of our model.

Specifically, we expected our model, in which we hypothesized that cultural threat perceptions promote violent extremist outcomes through NFC, to explain violent extremism against the West but not against Japanese people (i.e., an unrelated group).

Demonstrating the effectiveness of our manipulation, participants in the threat condition reported more perceived cultural threat, M = 4.83, SD = 1.57, 95% CI [4.630, 5.038], than those in the control condition, M = 4.08, SD = 1.33, 95% CI [3.882, 4.281], F(1, 399) = 26.90, P < 0.001, η² = 0.06, 95% CI [0.025, 0.114]. Supporting our predictions, the experimental manipulation affected both the dependent variables, violent behavioral intentions, and support for violent extremism, as well as the mediator, NFC. That is, as predicted, participants who read about threats against one’s in-group and religion expressed more violent behavioral intentions against the West, M = 4.02, SD = 1.53, 95% [3.838, 4.195], than those in the control group, M = 3.28, SD = 0.96, 95% CI [3.102, 3.451], F(1, 399) = 33.93, P < 0.001, η² = 0.08, 95% CI [0.035, 0.132] (Fig. 3). Participants in the cultural threat condition also reported more support for violent extremism, M = 3.87, SD = 1.74, 95% [3.645, 4.096], than those in the control condition, M = 3.26, SD = 1.46, 95% CI [3.038, 3.479], F(1, 399) = 14.60, P < 0.001, η² = 0.04, 95% CI [0.008, 0.077]. By contrast, and supporting the discriminant validity of the observed effects, no significant differences were observed in violent intentions toward Japanese people between the cultural threat, M = 2.56, SD = 1.32, 95% CI [2.364, 2.758], and the control conditions, M = 2.42, SD = 1.48, 95% CI [2.229, 2.615], F(1, 399) = 0.985, P = 0.321, η² = 0.002, 95% CI [0.000, 0.021].

Fig. 3.

Response distributions for the dependent and mediating variables in study 5. Note. The violin charts show the response distributions. The upper and lower horizontal lines of the box plots show the 75th to 25th percentiles of the interquartile range. The median score is indicated by the bolded horizontal line. The largest value within 1.5 times the interquartile range above the 75th and below the 25th percentile is shown by the two vertical lines. Each participant’s response is plotted as a data point.

In terms of the mediator, participants in the cultural threat condition had higher levels of NFC, M = 4.74, SD = 0.92, 95% [4.618, 4.871], than those in the control condition, M = 4.45, SD = 0.88, 95% CI [4.323, 4.570], F(1, 399) = 11.02, P < 0.001, η² = 0.03, 95% CI [0.004, 0.065]. We also note that, consistent with our predictions, higher levels of NFC related to more violent behavioral intentions, r = 0.23, P < 0.001 and greater support for violent extremism, r = 25, P < 0.001.

Next, we examined a multilevel mediation model in which NFC mediated the effects of the cultural threat condition (0 = control, 1 = cultural threat; treated as a between-participants variable) on both violent behavioral intentions against the West and support for violent extremism (treated as within-participants variables). In this model, cultural threat predicted NFC, which in turn predicted violent behavioral intentions against the West and support for violent extremism (Fig. 1). Bayesian credible intervals indicated a significant indirect effect of the experimental condition on violent extremist outcomes [B = 0.10, posterior SD = 0.04, 95% CI (0.037, 0.180)] through NFC. Also, in single-level mediation analyses, the indirect effects were significant for both violent extremist outcomes (SI Appendix, Table S22 and Fig. S5). The single-level mediation model in which NFC mediated the effects of the cultural threat condition on anti-Japanese violent behavioral intentions was not significant (SI Appendix, Table S22 and Fig. S5).

Study 6

Having obtained causal support for the first leg of our predicted model (i.e., from threat to NFC), we set out to experimentally test the effect of the mediator on the dependent variables (i.e., from NFC to violent extremist outcomes) as recommended by Spencer et al. (45). We chose to conduct the study in France because of the increasingly tense relationship between Muslims and non-Muslims in the country, and because Jihadists have been found to pose a greater threat to France (and Belgium) than to the rest of Europe (46).

We employed a time-pressure manipulation, which in previous research has been shown to increase the mediator NFC (47, 48). Specifically, participants were randomly assigned to a time-pressure or control condition. In the time-pressure condition, they were asked to rate a set of paintings under time pressure, and in the control condition, participants rated the same paintings without time pressure.

Supporting the effectiveness of the experimental procedure, the time-pressure manipulation had a significant effect on NFC, F(1, 237) = 102.82, P < 0.001, η² = 0.30, 95% CI [0.210, 0.387]. Participants in the time-pressure condition showed a higher level of NFC, M = 4.81, SD = 0.70, 95% CI [4.652, 4.963], than did those in the control condition, M = 3.68, SD = 0.98, 95% CI [3.531, 3.837]. In terms of the dependent variables, the time-pressure manipulation also had a significant effect on VBI-7, F(1, 237) = 108.87, P < 0.001, η² = 0.32, 95% CI [0.222, 0.399], VBI-2, F(1, 237) = 70.12, P < 0.001, η² = 0.23, 95% CI [0.141, 0.314], and on support for violent extremism, F(1, 237) = 116.21, P < 0.001, η² = 0.33, 95% CI [0.236, 0.413]. Compared to the control condition, participants in the time-pressure condition showed higher VBI-7, M = 4.39, SD = 1.84, 95% CI [4.110, 4.677] vs. M = 2.28, SD = 1.23, 95% CI [2.004, 2.564], VBI-2, M = 4.33, SD = 1.39, 95% CI [4.086, 4.567] vs. M = 2.89, SD = 1.27, 95% CI [2.651, 3.126] and more support for violent extremism, M = 4.79, SD = 1.62, 95% CI [4.518, 5.058] vs. M = 2.71, SD = 1.35, 95% CI [2.442, 2.976]. See Fig. 4 for response distributions.

Fig. 4.

Response distributions for the dependent variables in study 6. Note. The violin charts show the response distributions. The upper and lower horizontal lines of the box plots show the 75th to 25th percentiles of the interquartile range. The median score is indicated by the bolded horizontal line. The largest value within 1.5 times of the interquartile range above the 75th and below the 25th percentile is shown by the two vertical lines. Each participant response is plotted as a data point.

Internal Meta-analysis

To further validate the indirect association of perceived cultural threat on violent extremist outcomes via NFC, we synthesized and estimated the magnitude of the indirect associations across all studies meta-analytically using a structural equation meta-analysis. Specifically, following procedures outlined by Cheung (49, 50), we fitted a meta-analytic mediation model in which the different measures of violent extremist outcomes were treated as indicators of a latent violent extremist outcome variable (Fig. 1). The model showed satisfactory fit [χ² (3) = 99.97, P < 0.001, TLI = 0.92, CFI = 0.98, SRMR = 0.07], which provided important information because the estimation of saturated models in several of the primary studies prevented us from assessing the theoretical model’s fit to the data (note, however, that all unsaturated models that included covariates showed close fit to the data). Sobel, Aroian, and Goodman^† tests supported the meta-analytical indirect association (B = 0.12, all SEs = 0.01, all ps < 0.001), demonstrating that the indirect association of perceived cultural threat on violent extremism via NFC was significant for all dependent variables. Additionally, we repeated this analysis, including only correlational studies 1–4 as a form of sensitivity analysis to ensure that the study design (correlational vs. experimental) did not affect the results. Again, the fit of the model was satisfactory [χ² (3) = 49.78, P < 0.001, TLI = 0.91, CFI = 0.98, SRMR = 0.06], and the indirect association was significant according to Sobel, Aroian, and Goodman tests (B = 0.14, all SEs = 0.02, all ps < 0.001).

When using an alternative meta-analytic approach that pools the indirect associations from studies 1–5 following the procedure by Cheung (49), the meta-analytic indirect association was also significant (B = 0.08, SE = 0.02, P = 0.025, 95% CI [0.018, 0.139], I² = 0)^‡. Moreover, a meta-regression indicated that the size of the indirect association was unaffected by the specific measures used to assess violent extremist outcomes, or by correlational versus experimental study design (all ps for moderation effects > 0.27; see SI Appendix, p. 27 and Tables S27 and S28).

Discussion

Anecdotal observations suggest that various terrorist attacks were motivated by actors who perceived serious threats to their group, which may have triggered a narrow, black-and-white mindset. The present research provides empirical support for this notion using correlational, experimental, and field data collected among various non-Western, Educated, Industrialized, Rich, and Democratic (WEIRD) groups living in diverse contexts. Approaches to understanding violent extremism in recent years began to focus on personality, with the field of extremism research regaining some of its early interest in the role of psychopathology (51), or social and situational influences that operate independently of personality or other individual differences (52). Theoretically, the present research integrates these lines of work, suggesting how personality and situational approaches can be mutually informative. Specifically, our research shows how the study of threat perceptions and the resulting need for cognitive closure can inform our understanding of violent extremism.

Supporting the cross-cultural validity of our hypothesized model, the results demonstrated that perceiving a greater cultural threat was associated with violent extremist outcomes across different cultural contexts (studies 1 through 4), or led, experimentally, to a higher level of NFC (in study 5). Higher levels of NFC were in turn positively associated with (in studies 1 through 5) or produced (in study 6) more violent behavioral intentions and greater support of violent extremism. The meta-analytic results, the results of multilevel analyses in each study, and most of the single-level indirect associations of cultural threat on the extremist outcomes were significantly mediated by NFC.

To provide convergent support for the proposed mediation model, we tested our hypotheses with different attitudinal and violent behavioral intention measures of extremism (e.g., willingness to violently defend one’s group and/or religion, and support for violent extremism) and demonstrated significant effects across these measures. The consistent multilevel results across studies, designs, national contexts, and measures increase our confidence in the generality of the hypothesized model. In study 5, we also examined the alternative explanation that Muslims under threat hold negative attitudes against all modern and economically developed nations. Such a finding would suggest that this hostility is not necessarily rooted in ideological, political, and intergroup conflicts, such as cultural threat perceptions associated with the West. This interpretation was not supported: Cultural threat perceptions did not statistically significantly predict anti-Japanese violent behavioral intentions. Thus, the results supported the discriminant validity of our model, indicating that violent extremism in the groups of interest can be primarily attributed to cultural conflicts that lay at the core of the threat perceptions in question.

Although the general pattern of single-level mediation results across the samples showed considerable consistency, two exceptions were observed. In study 3, the indirect association was nonsignificant for support for violent extremism (one out of four measures). This measure was arguably the least extreme as it dealt with understanding but explicitly not the condonement of violence committed by other group members. As such, the results suggest that our model has the most explanatory value for the type of extremist outcomes that most closely resemble violent extremist activity (i.e., violent intergroup warfare by the Mujahedeen). Nevertheless, it is important to note that study 3 focused on a very hard-to-reach population and therefore had the smallest sample size of all studies. The nonsignificant indirect association which trended in the predicted direction may thus have resulted from insufficient power and must be interpreted with caution. Critically, the multilevel analysis demonstrated the robustness of effects across the measures assessed in the study. In study 4, one of two indirect associations trended in the predicted direction but fell below significance (with the lower bound of the CI being −0.005). However, the multilevel effect across the measures was significant, giving us confidence in the results also for this study.

Although previous research indicates that behavioral intentions reliably predict actual behavior (53), we further addressed a limitation of our work (as well as most other research on extremism), namely the reliance on self-reports of extremist attitudes and violent intentions. We did so by testing the construct validity of our measure using the known-group paradigm (43, 44). The comparison between former Mujahideen and the general Afghan samples showed that the former Mujahideen on average scored higher on all violent extremist outcomes as well as NFC and perceived cultural threat. Moreover, pooled analyses further supported the ecologic validity of our model, showing that it could successfully differentiate between former Mujahedeen and the general population of Afghan men.

The contribution of the present research is threefold. First, this study investigates the effect of NFC and cultural threat perceptions in the context of Jihadist extremism among both Muslim minority and majority populations rather than among Western populations (6, 25). Second, placing the current study in the broader literature on political extremism, researchers have increasingly recognized that both right-wing and left-wing extremists are characterized by a relatively simplistic (high need for cognitive closure), black-and-white perception of the social world (23). However, research based on many comparative studies from different countries still needs to convincingly map whether such features also characterize Jihadist extremists. The idea that general individual differences in psychological traits, needs, motivations, feelings, and thoughts predict people’s attitudes toward different social and political outcomes has long been recognized in social psychology (54). Nonetheless, previous research on violent extremism has traditionally downplayed the role of nonclinical, individual-level variables (see ref. 55). Our work addressed this shortcoming. Hence, our findings contribute to the empirical and theoretical insights proposing a link between cognitive simplicity and political extremism (56) by extending these propositions to Jihadist extremism, testing them directly with different designs and in different settings.

Third, previous research has predominantly focused on threat perception and outgroup derogation among Westerners and high-power groups (e.g., ref. 57). Conventional wisdom holds that threat perceptions are important bases of right-wing orientations and attitudes (24). However, recent research shows that threat perception increases intolerance and prejudice on both the left and the right (58) and, compared to moderates, both the left and the right extreme experience stronger threat in general (59). Our work contributes to this growing field, showing that cultural threat perceptions are potent predictors of violent extremism among some Muslim minority populations in the West, as well as among some Muslim majority populations in the Middle East and South Asia.

Limitation and Future Direction.

Our focus was on influences on Jihadist extremism, and thus all six of our studies involved Muslim participants. We chose our objective based, practically, on its importance for current international relations and, theoretically, because it is still a surprisingly understudied topic involving non-Western populations that are generally underrepresented in psychological research (6). However, the rationale behind our hypotheses about the cultural threat, NFC, and mediational effects suggests that our findings would generalize more broadly to intergroup relations. Indeed, the effects of cultural threat on bias and conflict have been demonstrated in a range of settings and internationally (11, 14), and NFC has been validated cross-culturally as well (e.g., ref. 60). Nevertheless, additional studies of how NFC may mediate the effects of various forms of threat on diverse expressions of negative intergroup orientations across a variety of cultural contexts could illuminate important moderating and boundary conditions not considered in our current research.

A second limitation is that our set of studies fully tested only one causal model, the mediating role of NFC in the relationship between cultural threat and violent extremism outcomes. As noted, we correlationally tested other models as well, and bidirectional influences remain possible. For instance, another hypothesis, which is anchored in a personality approach, would suggest an alternative casual sequence in which people higher in NFC would be more sensitive to perceived cultural threat (treated as alternative mediator), which would positively predict violent extremism. Indeed, supplementary analyses provided some support for this model in our correlational studies (i.e., studies 1–4; see SI Appendix). We did not experimentally test this additional model as we focused our studies on our hypothesized model in studies 5 and 6. Thus, although our mediation model was based on previous theorizing (e.g., refs. 25, 26, 34, and 35) and was causally supported (e.g., studies 5 and 6), our findings and conclusions do not challenge the validity of this alternative mediational model. However, we did not find consistent support for an alternative model that takes a personality-based perspective on extremism, assuming NFC to moderate the effect of cultural threat. Still, such an effect might be possible under other conditions, such as situations in which perceived threat is more ambiguous or less intense (61).

A third limitation is that all the current studies examined the impact of cultural threat on extremist support for violence and on willingness to defend or use violence to defend one’s in-group and its members. According to intergroup threat theory (11), cultural threat, which we studied, represents a form of symbolic threat. Although symbolic (e.g., traditions, values) and realistic (e.g., political, economic) threats are highly related empirically and thus difficult to disentangle practically, the theoretical distinction may be important for the study of extremism. Whereas material conflicts (involving realistic threat) can be alleviated by compromises over the distribution of resources between groups, cultural conflict (evoking symbolic threat) focuses on moral issues of right and wrong, for which compromise is more difficult to attain (see ref. 62). Because of less potential to compromise, cultural threats may rally support for outgroup violence and extremism to a greater extent than realistic threats such as physical violence (e.g., threat of terrorism) or economic loss as suggested by previous research (14). Nevertheless, future research may profitably replicate our model with different types of threat perceptions as predictor; different types of threats (e.g., from perceived internal or external sources) (63) can have very different psychological impacts, and many sources of threat can have complex influences. For instance, realistic threats resulting from the adverse effects of climate change on intergroup conflict are of growing importance. Gradual as well as rapid climate changes may lead to resource scarcity and parochial competition, which nurture realistic threat perceptions and subsequent intergroup conflict and violence (64, 65). Climate change may also evoke existential threats, increasing the mortality salience for many people (66). Some people respond to such mortality threats with a heightened cognitive rigidity (similar to the heightened NFC in the present research), which in turn can increase aggression against worldview-threatening others (e.g., ref. 67).

Another shortcoming of our research is that we relied on snowball sampling, which can be prone to selection bias and may limit the generalizability of our results. Although this recruitment strategy has limitations, snowball sampling remains one of the few ways to access hard-to-reach populations such as minority groups. To address the limitations, we combined several sampling techniques, which provided enhanced methodological rigor. For example, in addition to snowball sampling, we relied on contact with key community gatekeepers to facilitate access to the target population, which previous research identified as an effective strategy for recruiting hard-to-reach populations (68).

Finally, while our work focused specifically on the pathway from threat to stronger engagement in violent extremism through greater NFC, it is important to note that our findings can contribute to and complement a growing body of empirical and theoretical approaches to violent extremism. For example, research has found that identity fusion, which represents a visceral feeling of oneness with a group, contributes to violent extremism in ways similar to the ways of NFC. For instance, Israeli Jews who were more strongly fused with Judaism more strongly endorsed hostile and retaliatory action against Palestinians in response to perceived terror threats (69). Moreover, with respect to mediation, Gómez et al. (70) showed that threatening circumstances (e.g., the threat of Jihadist terrorism) produced identity fusion, which in turn motivated extreme sacrifices for ingroup members among Spanish undergraduates (see also ref. 71). Future research might thus examine directly whether and how NFC and identity fusion relate in the pathway from group threat to violent extremism both among Muslims against the West and more generally. For instance, given that we observed partial mediation in most studies, it is possible that identity fusion functions as an alternative mediator to NFC. Alternatively, it is possible that a greater threat may, as we found, increase NFC because it activates a quest for certainty and the avoidance of ambiguity. The activation of this mindset may then lead to greater identity fusion to reduce uncertainty (72), which may elicit greater violent extremism. It would thus be particularly valuable for future work to integrate these separate lines of research, thereby contributing to a comprehensive understanding of the dynamics of violent extremism.

Concluding Remarks.

In sum, the present research shows how threat perceptions and NFC can shape conditions that propel some individuals toward violent extremism. The theory and our findings contribute to a better understanding of the processes involved in the development of violent extremism and thus can contribute to ways to limit its spread. Interventions can aim at limiting the kinds of statements that stir cultural threat, for example, by educating individuals about the impact of their rhetoric on intergroup relations. In addition, interventions might be directed toward limiting the impact of exposure to such cultural threats among groups at risk. For example, such interventions might encourage greater openness and tolerance of ambiguity (i.e., by addressing processes associated with NFC), which could facilitate opportunities for constructive intergroup dialogue and ultimately produce outcomes more constructive than violent, extremist reactions.

Method

Study 1.

Participants and procedure.

Data were collected in 2013. We sampled 260 Muslims living in Denmark (68.3% female). Respondent ages ranged from 18 to 74 y, with the majority being between 18 and 34 y old (i.e., 94% of the total sample). Participants were asked to take part in a study that investigated “the relationship between social and personality factors and political action and attitudes.” The sample was recruited via key community gatekeepers and, subsequently, snowball and chain referral sampling, which are established methods for recruiting people from religious minority groups (68). Community gatekeepers, well-respected religious and elected members of Muslim networks, played a key role in establishing trust and representative outreach in this sample. For instance, we contacted spokespersons for Muslims, such as the Imam (prayer and religious leader) for the main mosque in Copenhagen and the president of the largest Muslim association in Denmark. In order to build trust and recruit diversity beyond gatekeepers, researchers also personally attended the studied populations’ social and cultural events.

All studies were conducted in accordance with the national and university regulations of the authors conducting the study. All studies were approved by the Institutional Review Boards of the first and second authors, namely the University of Oslo and the Institute of Business Administration Karachi, respectively. However, study 1 was exempt from review as it did not require or qualify for review under Danish laws at that time^§. All respondents in this and the remaining studies provided fully informed consent at the beginning of the study and were provided with the opportunity to contact the researchers in the case of questions or if they were interested in the results of the study.

For this and the other studies, detailed demographic information such as socioeconomic status, education, gender, and age is presented in SI Appendix, Tables S1–S3. A power analysis using the continuously varying sample size approach to Monte Carlo power analysis (73) suggested that approximately 190 participants would provide a 90% chance to detect the hypothesized indirect association. For this and the following studies, only those reporting to be Muslim (practicing or nonpracticing) were included in the analyses. Of the total sample, twelve participants were excluded because they either did not identify as Muslims or failed to indicate their religious affiliation. The study materials were translated into Danish using forward-back translation. While access to the data has been made available to those directly involved in the editorial and review process, due to the sensitive nature of our studies and the vulnerability of participants, additional access to the datasets will be granted by the corresponding author upon reasonable request.

Measures.

For a full overview of the variables and items included in each study, see SI Appendix, Tables S4–S10. Please note that all correlations, reliability coefficients, means, and SDs for studies 1 through 6 are presented in SI Appendix, Tables S12–S17.

Our samples were recruited from hard-to-reach Muslim populations among which data collection is time consuming, resource intensive and often fraught with difficulties. Given these challenges, we often collected data for several projects at the same time, such that studies 1–4 were part of larger datasets collected to address different research questions.

Only two measures (the 7-item violent behavioral intentions scale in studies 3 and 4) have been previously reported in a manuscript (currently under review) that focuses on the role of HEXACO personality traits, thereby addressing a different research question.

Data used in study 1 were based on a larger dataset. Other parts of the data are included in another study, Obaidi et al., 2018, study 1a (74) to address different research questions. The present variables and findings have not been presented previously. Other variables collected as part of study 1, but not included in the main analysis concerned, for example, perceived injustice, antigay sentiment, anger, nonviolent intentions, religiosity, religious discrimination, and Muslim identification.

Need for cognitive closure.

We selected the ten items with the highest factor loadings from Roets and Van Hiel’s Need for Closure Scale (75) to measure NFC (e.g., “I don’t like situations that are uncertain”_; α = 0.78). Unless stated otherwise, all responses were rated on 7-point Likert scales, ranging from 1 (strongly disagree) to 7 (strongly agree).

Perceived cultural threat.

Perceived cultural threat was assessed using three items (e.g., “Non-Muslim Westerners hold values that conflict with the values of people like me”; see 14_, α = 0.77).

Willingness to defend one’s in-group and religion.

An objective and unifying definition of violent extremism does not exist (76). Similarly, there is no validated and agreed-upon method to measure violent Jihadist extremism. In the current paper, we used several extremism scales to test the robustness of our results across measures, including a) willingness to defend one’s in-group and religion (e.g., studies 1 through 3), b) violent behavioral intentions (e.g., studies 2 through 6), and c) support for violent extremism (e.g., studies 2, 3, 5, and 6).

For this first study, willingness to defend one’s ingroup and religion was assessed using two items (e.g., “I am prepared to do everything in my power to defend my religion”; see ref. 74_, r = 0.73).

Analyses.

For this study that contained only one outcome variable, single-level linear regression analyses with linear predictors (i.e., cultural threat and NFC) were conducted using the PROCESS macro v.4.0 (77) to test the mediation model. Bootstrapping with 10,000 random resamples was used to estimate indirect associations. In this and all remaining studies, P values are two-tailed.

Study 2.

Participants and procedure.

Data were collected in 2019. We sampled 250 Muslims living in Kabul, Afghanistan (M_age = 25.5, SD_age = 8.22; 53% women), using an online survey. The participants were recruited through personal contacts using online convenience and snowball sampling. Based on the power analysis from study 1, we aimed to recruit 190 participants. To avoid poor data quality, we decided to recruit 60 additional participants in case some participants failed the attention check. Of the total sample, 48 participants failed the attention checks (i.e., “To assure that you are reading everything, please respond with the number three for this item”; “If you are paying attention, answer with the response option furthest to the right”) and were excluded a priori from analyses, leaving a sample of N = 202. All participants identified as Muslims. The study materials were translated from English into Dari using forward-back translation.

Measures.

The survey for study 2 included the perceived cultural threat (α = 0.91) and NFC (α = 0.91) measures from study 1. As a dependent variable, in the current study, we used several attitudinal and violent behavioral intention measures of violent extremism to test the robustness of our results, see SI Appendix, Table S6. In addition to willingness to defend one’s in-group and religion (r = 0.82), we assessed:

Violent behavioral intentions measured with 7 items (VBI-7, e.g., “I’m ready to go and fight against Westerners for the sake of Muslims in another country”; α = 0.71) (14);

Violent behavioral intentions measured with 2 items (VBI-2, e.g., “I am prepared to use violence to defend other Muslims”; r = 0.82) (74), and

Support for violent extremism measured with four items adapted from Tausch et al. (78) and Obaidi et al. (15) (e.g., “I can understand why some Muslims might have wanted to commit acts of violence in Europe, even though I do not condone the violence itself”; α = 0.73).

Analyses.

For studies 2 through 5, multilevel structural equations modeling in Mplus v. 8 with linear predictors (i.e., cultural threat and NFC) was used to estimate a 2-2-1 mediation model (41). This model assumes that the predictor (perceived cultural threat) and mediator (NFC) is measured between participants, whereas the outcomes are measured within participants; that is, the different outcome variables are treated as repeated measures of the same construct, namely violent extremism. Because bootstrapping is not available for such models, we tested the indirect association using the Bayesian estimator with 10,000 iterations and 95% Bayesian credible intervals, with the indirect association deemed significant if the interval does not contain zero (79). Further details are presented in SI Appendix. Inferential estimates are reported for all paths of the model, whereas Bayesian estimates are reported for the indirect association. The model is fully saturated at the between-level, hence fit indices are not informative and are not reported.

Study 3.

Participants and procedure.

Data were collected via a pencil and paper survey in Afghanistan in 2019. In this field study, participants were recruited through a nonprobability chain-referral sampling technique, which is one of the few methods to recruit hidden and very hard-to-reach populations such as former Mujahideens (68). Due to challenges of collecting data from this hard-to-reach population in a conflict zone, we did not attain the suggested sample size in this study. A decision was made to terminate the data collection after three months. Participants included 69 former Mujahideen (M_age = 46.00, SD_age = 8.31; 100% men) living in Kabul. With the help of personal connections living in Afghanistan, we first established contact with some former Mujahideens who had agreed to participate in the study. These individuals assisted us in identifying other potential subjects to whom we could reach out, after they had agreed to be contacted by us. All participants identified as Muslims, and with either Pashtun or Tajik ethnic groups belonging to different factions (e.g., Jamayat-E-Islami, Harakat-I Inqilab-I Islami, Hizb-I Islami). When the US and NATO forces completed their pullout from Afghanistan in fall 2021, many former Mujahideen took up arms again and joined the uprising (80), but we do not have information about whom this applied to in this study.

Measures.

The survey for study 3 included all measures from study 2—perceived cultural threat (α = 0.88), willingness to defend one’s in-group and religion (r = 0.77), VBI-7 (α = 0.76), VBI-2 (r = 0.76)—with the exception of NFC, which was measured with a 15-item scale (α = 0.84) (75) and support for violent extremism which was measured with 3 items (α = 0.77), see SI Appendix, Table S7.

Study 3 was part of a larger battery of variables, collected to answer different research questions within the same surveys. Variables not included here concerned, for example, altruism, emotionality, openness to experience, and nonviolent intentions.

Analyses.

In addition to multilevel mediation models, t tests were conducted to compare the male general population (study 2) to Mujahedeen Afghanis (study 3). A logistic regression approach in Mplus Version 8 (81) was used to estimate the model with grouping as a dependent variable, to behaviorally validate our model.

Study 4.

Participants and procedure.

Study 4 (N = 240, 42.5% women) was an online study conducted in 2020 among Muslims living in different countries (e.g., Europe, Asia, Australia, Africa, and the Middle East; see SI Appendix, and SI Appendix, Table S11 for detailed information about their place of birth, country of residency, and nationality) through purposive nonrandom sampling. Respondent ages ranged from 18 to 74 y, with the majority being between 25 and 34 y old (i.e., 34% of the total sample). Based on the power analysis in the previous studies, we aimed to recruit 190 participants. To account for poor data quality, we recruited 50 additional participants. All participants identified as Muslims. We included the same attention checks as in study 2 and anticipated excluding respondents who failed these checks. Of the total sample, 40 participants failed the attention check and were excluded a priori from the analyses, leaving a sample of N = 200.

To recruit a diverse sample of Muslims, we engaged the help of liaisons to establish contact with public Muslim figures (gatekeepers) in various countries (e.g., Denmark, Belgium, Pakistan, Afghanistan, and England). We selected public figures with diverse backgrounds (e.g., imams, media commentators, and intellectuals) who agreed to share our survey within their networks and distribute it on various email listservs for Muslims living in the West and various Muslim-majority countries. Due to an initially low rate of participation, we encouraged the participants to share the survey with their networks. Furthermore, the survey was distributed across various English-speaking WhatsApp groups consisting of Muslims living in various countries. The survey was forward-back translated and distributed either in English, French, Swedish, Danish, Dari, Urdu, or Turkish.

Measures.

The survey for study 4 included all measures from study 3 – perceived cultural threat (α = 0.89), NFC (α = 0.84), VBI-7 (α = 0.86), VBI-2 (r = 0.83)—except for support for violent extremism and willingness to defend one’s in-group and religion, see SI Appendix, Table S8.

Study 4 was part of a larger battery of variables, collected to answer different research questions within the same surveys. Variables not included here concerned, for example, altruism, emotionality, openness to experience, agreeableness, conscientiousness, extraversion, honesty–humility, and nonviolent intentions.

Test of Alternative Models in Studies 1 through 4.

Although the proposed mediation model was based on previous theorizing (24, 25, 34, 35), the correlational designs of studies 1 through 4 allow for tests of alternate models, even if support for our hypothesized model is obtained. For example, an alternative directional model is theoretically plausible, where NFC promotes violent extremism through cultural threat perceptions. That is, because individuals displaying higher levels of NFC may react particularly strongly to threatening information that challenges their worldview (82), they may experience a greater cultural threat, which in turn predicts stronger violent extremist intentions. Given that NFC can be considered a dispositional variable, it is also possible that it moderates the relationship between threat perceptions and violent extremism. Ancillary analyses partly aligned with these alternative models are presented in SI Appendix, Tables S23–S26 and Figs. S6–S9.

Study 5.

Study 5 was preregistered at https://aspredicted.org/zz3pt.pdf.

Participants and procedure.

Data were collected in 2020 in Pakistan. Using G*Power (83), we conducted an a priori power analysis to detect a difference between two independent means with a medium-sized effect (d = 0.4) and 80% power. The expected effect was based on previously unpublished research utilizing similar priming of cultural threat among Muslims in Pakistan. Our power analysis indicated a minimum required sample size of N = 200 to detect significant effects at a significance criterion of .05. We decided to oversample to leave some room to remove participants who did not identify as Muslims or failed the attention checks. We included the same attention checks as in study 2.

The participants were recruited at various universities in Karachi in person on their university campuses’ shared areas, lawns, libraries, and classrooms. The response rate was higher than expected, and we managed to recruit a sample of N = 450. Of these, 48 participants failed the attention checks, and one participant did not identify as Muslim, leaving a sample of N = 401 (M_age = 20.9, SD_age = 2.17; 40.4% women).

Experimental manipulation.

A random half of the participants (n = 196) were presented with a brief fictitious news article including concrete statements from Western politicians, military personnel, and the public to prime the perception of cultural threat. Specifically, we focused on threats toward Islam and Muslims (i.e., “Non-Muslim Westerners view Islamic culture and traditions as inferior to Western Christian culture”; “Westerners view the ‘War on Terror’ in Muslim countries like Pakistan, Afghanistan, and Iraq as the first steps in an imperial quest to promote Western supremacy and cultural superiority in Muslim countries”; see SI Appendix for details, p. 30). The other half of the participants (n = 205) assigned to the control condition was presented with a fictional description of various environmental projects attempting to protect mangrove forests. Both texts were adapted from real newspapers to ensure the ecological validity of the materials. The study was conducted in Urdu, and materials were forward-back translated. After the manipulation, participants completed the following scales.

Measures.

Manipulation check.

Participants were asked to complete a 4-item perceived cultural threat perception measure (α = 0.87). We used the 3-item measure used in the first set of studies in combination with a fourth item “Islam is constantly under attack by the Western countries”, see SI Appendix, Table S9.

Need for cognitive closure was measured as in study 4 (α = 0.83).

Violent behavioral intentions (VBI-7). We used a 7-item scale as in studies 2 through 4 to measure participants’ behavioral intentions to commit violence in defense of one’s group (see ref. 14, α = 0.76).

Support for violent extremism. As in study 2, we used four items adopted from Tausch et al. (78) and Obaidi et al. (15) to measure support for violent extremism (α = 0.82).^¶

Anti-Japanese violent intentions. We used two items from Obaidi et al. (2018) (14) to measure violent behavioral intentions against Japanese such as, “If nothing else helps, I’m prepared to use violence against Japanese to defend Muslims” (r = 0.82).

Analyses.

In addition to the multilevel mediation models, ANOVA was used to test for group differences between the experimental conditions.

Study 6.

The study was preregistered at https://aspredicted.org/rw2nr.pdf.

Participants and procedure.

Data were collected in 2020 in France and the power rationale was the same as in study 5 (the expected effect was based on previously unpublished research utilizing an identical manipulation conducted among Muslims in Denmark). We decided to oversample to leave some room to remove participants who did not identify as Muslims or failed the attention checks. All materials were forward-back translated. Using the same procedures as in study 1, we managed to recruit a sample of 300 participants (M_age = 26.9, SD_age = 7.88; 49% women) via Facebook. We included the same attention checks as in studies 2, 4, and 5. Of the total sample, 60 participants who failed the attention checks and one participant who did not identify as Muslim were excluded a priori from analyses, leaving a sample of N = 239.

Experimental manipulation.

To experimentally induce a heightened need for cognitive closure, we followed the procedure and material from Wiersema et al. (84) (See SI Appendix for details, p. 31).

Measures.

After completing the image rating task, all participants responded to the NFC scale from study 5 to assess the intended impact of the manipulation (i.e., the manipulation check, α = 0.83), in addition to the main outcome measures of violent behavioral intentions (e.g., VBI-7, α = 0.95 and VBI-2, r = 0.88), and the support for violent extremism measure from study 5 (α = 0.89). No timer/time pressure was present for these measures.

Analyses.

Group differences were tested using ANOVA.

Meta-Analysis.

Analyses.

For the structural equations meta-analysis, we applied a two-stage approach (49, 50). First, we separately pooled all correlations between all variables in the model using the metafor package for R (85), and then we fitted a mediation model with a latent outcome to the matrix of pooled correlations using the metasem package for R (86). Because bootstrapping is not available for meta-analytical SEM, we tested the indirect association using the Sobel, Aroian, and Goodman test.

For the meta-analysis of the indirect association (49), we pooled the beta coefficients referring to indirect associations from studies 1–5, calculated separately for every measure of extremist outcomes. We used robust variance estimation with small sample correction in the robumeta package for R (87, 88) to estimate the intercept-only model and a linear meta-regression model to test whether the type of extremist intentions measure moderated the effects.

Supplementary Material

Appendix 01 (PDF)