Discrimination, Life Stress, and Mental Health Among Muslims: A Preregistered Systematic Review and Meta-Analysis

Understanding the experiences of Muslims living in western countries requires knowledge of the interrelated contexts of racialization, Islamophobia, and discrimination. Herein, we discuss each.

Racialization is the process through which racial groups are formed (Garner & Selod, 2015). Race is socially constructed and situated within a system of power, often used to disadvantage certain groups within society. Racism includes stereotypes, prejudice, and discrimination,¹ and occurs overtly, covertly, intentionally, unintentionally, and institutionally (Fish & Syed, 2019; S. P. Harrell, 2000). There is little evidence for a biological basis for race; instead, the differences between groups are due to consequences of the majority imposing power onto the minority (see Syed & Kathawalla, 2018 for discussion). Thus, even though Muslims are quite diverse with respect to ethnicity, nationality, and religious orientation, the universal anti-Muslim sentiment has “racialized” Muslims, creating a racial and pan-ethnic group (Cesari, 2014; Richardson, 2012; Selod, 2015).

Conceptualizing Muslims as a racialized group can be somewhat confusing due to their heterogeneity and because they are often thought of as a religious group of members that can share some aspects of culture (e.g., tradition, language, food). It is only in the global context of Islamophobia that the group became racialized (Meer, 2019). Bolstering this conceptualization is the fact that the racialization process has extended to non-Muslims whom people think are Muslim (e.g., Sikhs, South Asians of other religious, Christian Arabs); this creates a “Muslim-like” identity group that has been racialized as Muslims (Abdulrahim et al., 2012; Meeussen et al., 2013). And indeed, the context of Islamophobia has been connected to individuals conceptualizing their own identity as Muslim (Wang et al., 2020) and working toward combating racism and prejudice (Cesari, 2014; Garner & Selod, 2015; Meer, 2019; Selod, 2015).

As noted above, the racialization of Muslims has occurred in the context of Islamophobia, which was originally defined as an irrational fear toward Muslims and Islam that results in explicit and subtle discrimination toward Muslim people (Runnymede Trust, 1997). Islamophobia involves negative beliefs about Muslim values—that they hate freedom, oppress Muslim women, and want to instill sharia law (Islam, 2018). Moreover, there is a common rhetoric that Muslim faith leads to less commitment within mainstream value systems in the U.S. and Europe (e.g., Hervik, 2019; A. S. Ahmed, 2019; Martinovic & Verkuyten, 2012). As a phobia, Islamophobia involves a sense of “justified fear” that allows for justification of discriminatory or racist actions toward Muslims for protection or survival of majority group members. Media portrayals of radicalization of Muslims and the European refugee “crisis” provide justification for the fear (Awan & Zempi, 2016; Jackson, 2010; Pew Research Center, 2017b; Shaver et al., 2017; Trevino et al., 2010; Vedantam, 2017), even in the face of contradictory evidence. For example, most of the terrorist attacks (and the deadliest) carried out in the name of Islam have occurred in the Middle East, North Africa, South Asia, and Sub-Saharan Africa (Miller, 2017). In addition, white supremacists carried out more attacks than Islamists in the U.S. from 2008–2016 (Neiwert, 2017). Most importantly, the majority of Muslims in the world do not engage in violent attacks, and thus the fear of Muslims is unjustified. Nevertheless, fear of Muslims and Islam has been connected to anti-terror initiatives, such as the anti-Muslim laws in Europe and more recently in the United States (Neiwert, 2017; Sivanandan, 2010).

Whereas Islamophobia pertains to group-level discrimination, individual-level discrimination directed towards Muslims must also be considered. In the context of Islamophobia, there has certainly been an increase in anti-Islamic or Muslim hate-related incidents in the U.S. over the past 20 years (see Figure 1; Federal Bureau of Investigation, n.d.). Studies including Muslims in the U.S., Canada, and Europe have consistently found high rates of perceived discrimination directed to oneself or toward people they know (Ashraf & Nassar, 2018; Brüß, 2008; Nadal et al., 2012; Sirin & Fine, 2007; Wilkins-Laflamme, 2018). Qualitative research has also found that Muslims report unique experiences of discrimination compared to other marginalized, minority groups in the U.S. (Nadal et al., 2012; Sirin & Fine, 2007) and Europe (e.g., Moffitt et al., 2018). In a related national study amongst Canadians, it was found that non-Muslims reported the lowest rating toward Muslims living in Canada compared to other minority groups (Wilkins-Laflamme, 2018).

Discrimination is “any behavior which denies individuals or groups of people equality of treatment” based upon social identities or characteristics outside of one’s control (e.g., religion, color of skin, gender, sexuality; Stroebe & Insko, 1989, p. 50). The overall experience is the same—unfair treatment—but the focus or perception of the unfair treatment may differ based upon domain. We use ethnic, racial, and religious discrimination interchangeably due to evidence that discriminatory events are motivated by all three types of discrimination (i.e., being called a terrorist, comments about the hijab, and being told to go back to their own country; Nadal et al., 2012) and also due to the “racialization” of this group described earlier (Garner & Selod, 2015). In a systematic review of Islamophobia and health for Muslims, findings suggest that experiencing racial, ethnic or religious discrimination is associated with increased psychological distress (Samari et al., 2018); evidence indicates similar levels of perceived discrimination and impact of Islamophobia across ethnic groups (Abu-Ras et al., 2018).

As mentioned previously, discrimination can be conceptualized on multiple levels. Individuals may experience individual-level discrimination that includes, implicit racism (e.g., unconscious biases), subtle discrimination (e.g., being told they speak English well), explicit discrimination (e.g., being hit or called names), and group/systemic discrimination (e.g., hearing others talk about how Muslims are dangerous and the Muslim immigration ban; Fish & Syed, 2019). Research has largely found negative associations between discrimination and mental health for all levels, but there is also evidence for differential associations with mediator and moderator variables between individual and group discrimination (Lui & Quezada, 2019; Schmitt et al., 2014). Research based in Europe has found that anti-Muslim prejudice is a distinct construct; even though it is related to other forms of prejudice (e.g., cultural, economic), it is not related to blatant discrimination (Elchardus & Spruyt, 2014). In contrast, meta-analytic results between subtle and overt discrimination have similar associations with health, demonstrating there may not actually be a difference in the effect on outcomes (Jones et al., 2016). In total, meta-analytic results suggest the need for further examination of the association of individual vs. group-based discrimination with mental health.

Despite the conceptual overlap, life stressors and discrimination are discussed as entirely separate constructs, which may or may not be the case. Life stressors are “conditions and stimuli where predictability and controllability are at stake” (Koolhaas et al., 2011, p. 1292). Additionally, life stressors are life events that “are negative, fateful, unpredictable, and central in the sense of being life-threatening or otherwise disruptive of the usual activities and major goal of the individual” (Dohrenwend, 2000, p. 13). What is considered a life stressor can be vague and broad, including diverse types of stressors, such as: financial hardship, unemployment, and caregiving stress. For the purposes of this review, we focus on the definition of life stressors as defined by The Traumatic Life Events Questionnaire (TLEQ; Kubany et al., 2000) and the abuse items of the Behavioral Risk Factor Surveillance System (BFRSS) Adverse Childhood Experiences (ACEs) questionnaire (Centers for Disease Control and Prevention, n.d.). The TLEQ was developed to assess exposure to a wide range of traumatic events, including 18 specific events (e.g., natural disaster, war trauma). The BFRSS ACEs questionnaire includes five abuse-related items, including physical abuse (one item), psychological abuse (one item), and sexual abuse (three items). Traumatic life events and adverse childhood experiences are commonly examined in relation to the development of psychopathology in majority groups (e.g., Hughes et al., 2017; Steel et al., 2009).

The literatures on life stressors (as defined above) and perceived discrimination are almost entirely separate. Life stressors tend to be the focus in research with majority group members, whereas discrimination is a stronger focus in research with minority group members. Research undeniably supports the notion that stress exposure is shaped by individuals’ context, which is certainly the case for discrimination; and discrimination is consistent with the definition of stress as being negative, unpredictable, and uncontrollable (Dohrenwend, 2000; Koolhaas et al., 2011). Substantial theoretical evidence suggests that discrimination is a unique stressor, given that discriminatory actions are based upon factors (e.g., ethnicity, race, stereotypes), all of which are uncontrollable; but this is rarely tested in conjunction with life stressors (S. P. Harrell, 2000).

Harrell’s (2000) model of racism-related stress is one of the few—if not the only—model that equally emphasizes both discrimination and general life stressors in the development of psychopathology. Additionally, the model of racism-related stress postulates the moderator role of internal characteristics. The model of racism-related stress (S. P. Harrell, 2000) separates racism-related stress into six different types of stressors: 1) racism-related life events, 2) vicarious racism experiences, 3) daily racism micro-stressors, 4) chronic-contextual stress, 5) collective experience of racism, and 6) transgenerational transmission of group traumas (S. P. Harrell, 2000). The model posits that racism-related stress is differentially associated with physical, psychological, social, functional, and spiritual outcomes due to antecedent variables (e.g., race, gender, age), familial and socialization influences (e.g., family characteristics, racial socialization), and internal and external mediators (e.g., self-esteem, coping, cultural values, spirituality, identity; S. P. Harrell, 2000). Although the model’s focus is on racism-related stressors, S. P. Harrell (2000) emphasizes the importance of also considering the stressors “common to the mainstream of society” and outside of the minority status, as the combined effect may be especially detrimental (S. P. Harrell, 2000; p. 47).

The studies that have examined both discrimination and life stressors have indicated mixed results regarding the strength and significance of the associations (e.g., Alemi & Stempel, 2018; Hassouneh & Kulwicki, 2007; Pieterse et al., 2013; Pieterse & Carter, 2007). Some findings even suggest that perceptions of discrimination contribute almost nothing to the racial differences in health outcomes compared to the contribution of life events and traumas (Taylor & Turner, 2002). Hence, a major goal of the current study is to examine both the association between perceived discrimination and mental health, and the association between life stressors and mental health, for the purpose of better understanding both constructs’ contribution to health outcomes.

The focus of the present study is on the associations between stressors (discrimination and life stressors) and mental health. We focus on mental health due to the downstream effects of mental health on other physical health and social outcomes (e.g., E. K. Adam et al., 2017) and the well-supported evidence of the impact of discrimination and stressors on mental health (Pascoe & Smart Richman, 2009).

Meta-analytic results demonstrate racial/ethnic discrimination is associated with adverse physical and mental-health outcomes (Benner et al., 2018; Paradies et al., 2015; Pascoe & Smart Richman, 2009; Pieterse et al., 2012; Schmitt et al., 2014). The meta-analytic correlations vary slightly based upon type of discrimination, outcome, and sample. Generally, experiences of discrimination have shown small to moderate associations with negative mental health and a small negative association with positive mental health (Benner et al., 2018; Lee & Ahn, 2011; Pascoe & Smart Richman, 2009; Schmitt et al., 2014). There is evidence of similar associations between different types of racism (e.g., direct, group, vicarious) and negative mental-health outcomes (Paradies et al., 2015). It is important to note that there seem to be differences based upon study design, with the association being nonsignificant for experimental studies and significant, but attenuated, in longitudinal studies (Schmitt et al., 2014).

A recent systematic review of Islamophobia and health summarizes the results of 34 articles examining mental health in Western countries (out of 53 in the review; Samari et al., 2018). The review found support for a consistent relationship between experiences of discrimination and poorer mental health among Muslims and Muslim-like populations (Samari et al., 2018). Moreover, the association between discrimination and psychological distress (e.g., depressive symptoms, anger, and anxiety) was consistent for discrimination due to race, ethnicity, and religion (Samari et al., 2018). However, there was heterogeneity in effect sizes and significance levels between discrimination and various types of outcomes, which were not explored further for potential moderators (e.g., Sawssan R. Ahmed et al., 2011; Assari & Lankarani, 2017; Ellis et al., 2008; Hakim et al., 2017; Rippy & Newman, 2006). To summarize, although there seems to be consistent evidence for the negative associations of various types of discrimination with mental-health outcomes for Muslims in Western countries, there is variation in the literature potentially due to study-related decisions, demographic, or other factors.

A body of evidence suggests that traumatic events (e.g., Steel et al., 2009) and adverse childhood experiences (ACEs; Hughes et al., 2017; Kessler et al., 2010) are associated with negative mental-health outcomes. A meta-analysis on the effects of torture and potentially traumatic events among populations exposed to mass conflict and displacement from 40 different countries, found that torture was associated with PTSD and depression (Steel et al., 2009), and that potentially traumatic events were also associated with PTSD and depression (Steel et al., 2009). Additionally, previous trauma, general childhood adversity, and childhood abuse demonstrate meta-analytic effects with PTSD (Brewin et al., 2000). With respect to ACEs, individuals with at least four ACEs were at greater risk for anxiety, depression, and low life satisfaction (Hughes et al., 2017). These meta-analyses provide evidence for a small to moderate association between negative life events and negative mental-health outcomes.

Studies that include both discrimination and general life stressors for Muslim samples suggest mixed results regarding the association of life stressors and mental health. For example, perceived discrimination and life stressors are both unique predictors of mental-health outcomes (e.g., Alemi & Stempel, 2018; Mölsä et al., 2017 ²), but effect sizes varied across studies. Conversely, studies also found that both predictors were not significantly associated with specific negative mental-health outcomes (e.g., Hassouneh & Kulwicki, 2007; Mölsä et al., 2017). These differential results not only emphasize the need for rigorous statistical methods to clarify the effect, but also suggest the importance of both discrimination and general life stress as predictors for mental-health outcomes.

In sum, there is research on the impact of discrimination and life stressors, but results are somewhat varied for Muslims, and very few studies examine both predictors (with minority groups generally or Muslims specifically). Thus, to address this, in the proposed study, we examine the associations of both perceived discrimination (individual and group) and life stressors with mental health. Examining both discrimination and life stressors may be important based upon preliminary qualitative interviews of second-generation Muslim Americans (n = 53; Kathawalla & Syed, 2019). Qualitative work indicated most second-generation Muslim Americans experienced discrimination or systematic racism in the United States but seemed to be differentially affected by the discrimination based upon life adversity, demographic factors, and multiple intersecting identities (Kathawalla & Syed, 2019). Additionally, some participants experienced individual-level discrimination, whereas others experienced group discrimination (Kathawalla & Syed, 2019; Wheeler et al., 2019). Thus, based upon the current state of the literature and previous qualitative work, there are added benefits for further investigation of discrimination and life stress as predictors of mental health.

Meta-analyses and systematic reviews consistently support investigating age, gender, immigration status, education, and region as moderators of stressors and mental health due to the theoretical support of their potential impact (e.g., S. P. Harrell, 2000). In the recent systematic review of health effects of Islamophobia, gender moderated the relation, such that discrimination and psychological distress were significant only for men in some studies (Assari & Lankarani, 2017; Rippy & Newman, 2006; Samari et al., 2018). However, a meta-analysis examining the relation between racism and mental health did not find any significant moderation effects with age, sex, birthplace, or education level (Paradies et al., 2015). Conversely, meta-analytic results for PTSD demonstrate consistent support for lack of education being as a risk factor, but results vary based upon sample and method (Brewin et al., 2000). There is some evidence that generational status or years living in the U.S. affects the association between stressors and mental health (Abu-Ras et al., 2018). Thus, evidence suggests a moderation effect for gender and generational status, but there is less clarity regarding education. Results from a systematic review demonstrated that the associations between discrimination and mental health were not affected by racial identification, skin color, or area of residence (Samari et al., 2018). In the current study, we aim to test potential demographic moderators including, but not limited to gender, age, immigrant status, and country/region.

Harrell’s (2000) model of racism-related stress includes identity domains or personal factors as potential moderators of the association between stressors and mental-health outcomes. Identity is especially relevant to Muslims, given that the context of Islamophobia may lead individuals to want to distance themselves from their identity and negative Muslim stereotypes (Sadek, 2017). Thus, there is not only a negotiation between Muslim values and Western values, but there is also a negotiation between Muslim religious values, Muslim secular values, and Muslim fundamentalism (Martinovic & Verkuyten, 2012; Sadek, 2017). For example, in a study of Muslims living in New Zealand, stronger Muslim identification was associated with increased susceptibility to the negative effects of perceived religious discrimination (Jasperse et al., 2012).

Erikson’s (1968) theory of identity development asserts that at various stages of development, healthy development involves negotiation of tensions in order to build a coherent and healthy sense of self. Life stressors and adversity create obstacles that make negotiating these tensions potentially more challenging. Erikson (1968) describes psychopathology as severe symptoms of identity confusion. Thus, Erikson’s theory highlights the importance of culture and social context on identity development.

The literature suggests multiple domains of cultural identity, including, but not limited to ethnic identity, national identity, identity integration, religious identity, and visible identity. However, there are many domains of identity, and most studies include only one identity construct while excluding others. It is unclear which aspects of identity may be more protective or whether there are differential mitigating factors of identity for ethnic minorities (e.g., ethnic identity commitment is a protective factor, while ethnic identity exploration is an exacerbating factor; Phinney, 1990; Yip et al., 2019).

For example, ethnic identity may be an important protective factor for Muslims that may mitigate some of the negative effects of discrimination and stress on the development of mental health (e.g., Sawssan R. Ahmed et al., 2011; El Bouhaddani et al., 2019; Sheldon et al., 2015). However, the evidence also suggests that stronger ethnic identities may be associated with more experiences of perceived personal and group discrimination for Muslims (Verkuyten & Nekuee, 1999). In relation, there is support for the impact of American identity (e.g., Sirin et al., 2008), identity integration (e.g., Britto & Amer, 2007; Isik-Ercan, 2015; Lowe et al., 2018; Saleem et al., 2018; Saroglou & Galand, 2004; Sirin & Fine, 2007), religious identity (e.g., Abu-Ras et al., 2018; Z. Adam & Ward, 2016; Sawssan R. Ahmed et al., 2011; Peek, 2005), and visible identity (e.g., Abdulrahim et al., 2012; Awan & Zempi, 2016; Gulamhussein & Eaton, 2015; Rahmath et al., 2016; Selod, 2015) on mental health or experiences of perceived discrimination for Muslims.

The major methodological limitations in the current state of literature for Muslim mental health include publication bias, homogeneous samples, inconsistent measurement of predictors and outcomes, and neglect of relevant constructs. The current study addresses these limitations.

Publication bias. The most recent systematic review conducted by Samari and colleagues (2018) did not include dissertations or unpublished studies. This is a major limitation, given that publication bias (i.e., selective publication based on arbitrary factors such as statistical significance) leads to biased estimates of psychological phenomena; therefore, publication status (unpublished vs published studies) is an important consideration (Siddaway et al., 2019). Moreover, most studies examining mental health among Muslims include small sample sizes and are underpowered. Underpowered studies lead to Type 1 and Type 2 errors and therefore can bias findings. Furthermore, there has been an emphasis on statistical significance and p-values in research without careful consideration of the effect sizes (Siddaway et al., 2019). Meta-analytic methods can account for some of these concerns.

Homogeneous samples. Current research of Muslim experiences of discrimination and mental health tends to be location- and group-specific, and therefore very homogeneous. This limits generalizability to the broader Muslim population in Western countries. There is a need to capture the experiences of diverse Muslims in terms of geographic location and ethnicity (Samari et al., 2018). For example, many studies use data from the Detroit Arab American Study (Baker et al., 2006). Although the study provides a rich dataset, there is a need for studies to use other samples of Muslims living in Western countries. Others may argue that homogeneous samples lead to less statistical ‘noise’ in the data; while that point has merit, due to the racialization of Muslims, there is a benefit for more research to examine processes on a broader scale and test for relevant moderators.

Inconsistent measurement. Variation in measurement and definitions of discrimination lead to challenges in the field. Researchers use many different measures of discrimination that range from validated measures with 40 or more items to a single item question (Kathawalla et al., 2021). This is further elaborated on in a forthcoming systematic review of measures of Islamophobia and discrimination for Muslims (Kathawalla et al., 2021). For example, a study measured Islamophobia using two items that have not been validated in other studies (Abu-Ras et al., 2018). Additionally, studies measure individual or group discrimination, but tend to discuss the results without indicating the discrimination type. Even in terms of potential moderators and mediators, different measures (e.g., ethnic identity, Black identity, importance of identity, collective self-esteem) are used, and it is challenging to identify clear and robust effects. This is evident even in the literature reviewed above where effect sizes change depending on the measure (e.g., direct exposure vs. indirect exposure; Paradies et al., 2015).

Neglect of other relevant constructs. Another limitation is the general neglect of life stressors along with discrimination—a major limitation that the proposed project seeks to address. Harrell’s (2000) model of racism-related stress offers alternative possibilities that emphasize both discrimination and life stressors. As described above, meta-analytic results do show a significant association between (1) discrimination and mental-health outcomes, and (2) potential moderators and mediators (e.g., Paradies et al., 2015; Yip et al., 2019); but individual studies that include both discrimination and life stressors demonstrate varied results (e.g., Mölsä et al., 2017). Research on the etiology of mental-health problems for the general population tend to include life stressors and should be investigated in minority populations as well. Investigating the influence of life stressors may help in understanding differential negative effects of discrimination among mental-health outcomes.

Broadly, the aim of the current meta-analysis is to examine the associations between (1) discrimination and mental health, and (2) life stressors and mental health for Muslims living in Western countries (i.e., United States, Canada, Europe—including France, UK, Germany, Austria, Belgium, Netherlands, Denmark, Norway, Sweden, Bulgaria, Greece—New Zealand, or Australia). We further explored the difference between these two stressors—discrimination and life stressors. Additionally, we examined various moderators of the two associations, including study-related (e.g., publication status, level of discrimination, outcome type), demographic-related (e.g., age, gender, education level, country of origin, region/continent), and identity-related (e.g., ethnic identity, religious identity, identity integration) variables.

All hypotheses described below do not include control variables.

Hypothesis: We expected a positive relation between increased discrimination and negative mental-health outcomes.

2a) Is publication status a significant predictor of the association between discrimination and negative mental-health outcomes? What are the effect-size estimates for the separate publication status predictors?

Hypothesis: We expected publication status to be a significant predictor where effect-size estimates for published studies would be stronger than unpublished studies.

2b) Are characteristics of discrimination measures (i.e., measure name, number of items) significant predictors of the association between discrimination and negative mental-health outcomes?

Hypothesis: We expected discrimination-measure characteristics to be a significant predictor. There were not enough consistent findings in the literature for us to propose a specific directional hypothesis.

2c) Is outcome type (i.e., positive vs. negative mental health) a significant predictor of the association between discrimination and mental-health outcomes? What are the effect-size estimates for the two outcome types?

Hypothesis: We expected outcome type to be a significant predictor, with the effect-size estimate for discrimination and negative mental health being stronger than discrimination and positive mental-health outcomes.

2d) Is level of discrimination (i.e., individual, group, or individual-group discrimination) a significant predictor of the association between discrimination and mental-health outcomes? What are the effect-size estimates for the separate level of discrimination predictors?

Hypothesis: We expected level of discrimination to be a significant predictor, with the effect-size estimate for individual discrimination and negative mental-health outcomes being stronger than group and individual-group discrimination.

2e) Are age, gender, refugee/immigration status, country of origin, region/continent, and Muslim sample type (Muslim sample vs. Muslim-majority country sample) significant predictors of the association between discrimination and negative mental-health outcomes?

Hypothesis: We expected gender to be a significant predictor, with the association between discrimination and negative mental-health outcomes being stronger for men than it would be for women. Based on past literature, there were not enough consistent results for us to hypothesize the effects of other demographic predictors.

2f) Are identity-related variables (e.g., ethnic identity, religious identity, national identity) significant moderators of the association between increased discrimination and negative mental-health outcomes?

Hypothesis: We expected identity-related variables to be significant moderators, with higher levels of identity attenuating the relation between discrimination and negative mental-health outcomes.

Hypothesis: We expected a positive relation between increased life stressors and negative mental-health outcomes.

4a) Is publication status a significant predictor of the association between life stressors and negative mental-health outcomes? What are the effect-size estimates for the separate publication status predictors?

Hypothesis: We expected publication status to be a significant predictor, with the effect-size estimate for published studies being stronger than unpublished studies.

4b) Are the characteristics of life-stressor measures (i.e., measure name, number of items) a significant predictor of the association between life stressors and negative mental-health outcomes?

Hypothesis: We expected life-stressor-measure characteristics to be a significant predictor. There were not enough consistent findings in the literature for us to propose a specific directional hypothesis.

4c) Is outcome type (i.e., positive vs. negative mental health) a significant predictor of the association between life stressors and mental-health outcomes? What are the effect-size estimates for the two outcome types?

Hypothesis: We expected outcome type to be a significant predictor, with the effect-size estimate for life stressors and negative mental-health outcomes being stronger than life stressors and positive mental-health outcomes.

4d) Are age, gender, refugee/immigration status, country of origin, region/continent, and Muslim sample type (Muslim sample vs. Muslim-majority country sample) significant predictors of the association between life stressors and negative mental-health outcomes?

Hypothesis: We expected gender to be a significant predictor, with the association between discrimination and negative mental-health outcomes being stronger for men than for women. Based on past literature, there were not enough consistent results for us to hypothesize effect of other demographic predictors.

4e) Are identity-related variables (e.g., ethnic identity, religious identity, national identity) significant moderators of the association between increased life stressors and negative mental-health outcomes?

Hypothesis: We expected identity-related variables to be significant moderators, with higher levels of identity weakening the relation between life stressors and negative mental-health outcomes.

Hypothesis: We expected stressor type not to be a significant predictor, with similar effect-size estimates across the different kinds of stressors. (Note: We combined individual, group, and individual-group combined discrimination measures.)

Hypothesis: We expected discrimination to uniquely predict negative mental health while controlling for life stressors. We expected life stressors to uniquely predict negative mental-health outcomes while controlling for life discrimination.

This study was accepted as a Stage 1 Registered Report on November 14, 2019, and registered on the OSF Registered Report Registry (https://osf.io/4wsg3); therefore, all decisions were made a priori unless otherwise noted. The accepted Stage 1 manuscript is available at https://osf.io/d7bzs.

We searched electronic databases to locate both published and unpublished studies. These included PsycINFO, PubMed, Academic Search Premier, ATLA Religion Database with ATLASerials, Index Islamicus, JSTOR, Sociological Abstracts, Ovid Medline, SHARE, and ProQuest Dissertations and Theses Global. The search was conducted between December 10, 2019 – January 29, 2020, by the first author after consultation with a university Librarian. We searched using the title, abstract, keywords, and subject headings. Search terms included a combination of discrimination terms (e.g., discrimination, Islamophobia, anti-Muslim), combined with OR, life stressor terms (e.g., life stress, childhood adversity, trauma), combined with AND, mental health terms (e.g., mental health, psychological distress, self-esteem), combined with AND identifier of Muslim or Muslim-like identity terms (e.g., Muslim, Islam, Arab, Pakistani). See Appendix A for exact search terms and supplemental materials (https://osf.io/nvhxt/) for reproducible search information for each database.

To search for additional unpublished studies, researchers in the field of Muslim mental health were contacted using the procedures outlined by Siddaway et al. (2019). First, we identified 16 authors across the United States, Canada, Europe, and Australia/New Zealand with one or more publications on the topic; we emailed each of the authors to see if they had unpublished studies or forthcoming data. Authors were given one month to respond and were sent one follow-up email after two weeks. Lastly, the reference lists of related systematic reviews were examined to conduct a forward search.

The database searches were combined in EndNote Desktop (citation management software) and resulted in 9,333 records, with an additional eight records identified through other sources (see Figure 2 for PRISMA flowchart; EndNote X9 Desktop, Version 19.3.2.15235). After duplicates were removed, the search resulted in 4,820 records for screening. An online systematic review software, Rayyan QCRI, was used to assist in study selection and track the reliability of inclusion for articles identified through the search procedure outlined (Rayyan QCRI; Ouzzani et al., 2016). Two researchers independently evaluated articles through a two-part process. First, using an iterative screening process, two researchers screened 10 articles and discussed discrepancies; then, they screened another 10 articles and discussed discrepancies before continuing to screen the whole set of articles. The first author screened the titles and abstracts of all 4,820 records and retained those that appeared to meet the following inclusion/exclusion outlined below for the next stage of screening. A research assistant double-screened 22% of records (1,063 records) with a percent agreement of 92%. Second, the 572 records were marked for full text review; then, they were reviewed and evaluated for inclusion and exclusion criteria. The first author reviewed all potential articles, and an undergraduate research assistant double-evaluated 27% of the articles (155 articles) resulting in 90% agreement. All discrepancies were discussed and resolved by consensus. Reasons for study exclusion (e.g., qualitative study, not Western country) were tracked (see Figure 2). The screening process resulted in 130 unique samples (i.e., datasets), with 295 correlations, from 120 articles. The PRISMA Statement and Checklist were used to transparently document the research procedure (Liberati et al., 2009; see Figure 2).

Article type. Unpublished, published, and dissertations were included. If the exact same sample was used in a dissertation and a published article, only the published article was included.

Variables. Quantitative and mixed-method studies that had a measure of discrimination, Islamophobia, prejudice, racism, life stressors, life adversity, or ACEs and a measure of mental health (e.g., depression, anxiety, PTSD, paranoia, internalizing symptoms, psychological distress, well-being, self-esteem) were included. For this study, life stressors were defined by TLEQ (Kubany et al., 2000) and the BFRSS ACEs questionnaire (Centers for Disease Control and Prevention, n.d.) and not as other types of stress (e.g., racism-related stress, financial strain, relationship conflict). When it was unclear whether a measure fit the inclusion criteria based on the information provided in the article, we then looked at the original measure article for clarification.

Any study from which we were able to extract the effect size (i.e., zero-order correlation coefficient) was included. If the statistic was not included in the study, we calculated the correlation coefficient using other statistics provided in the article (detailed process outlined in the coding section) or contacted the study author for more information. One hundred thirty studies were eligible, but did not contain the necessary statistics; therefore, those authors were contacted for additional information and provided instructions on how to conduct analyses in R. Approximately two email attempts were made to author(s) in a four-week period. Ten studies were included after the author provided additional information. Studies were excluded when zero-order correlation coefficients could not be calculated or were not provided by authors.

Participants and geographic location. Relevant studies included a sample of Muslim or Muslim-majority-country (i.e., a country that is 90% or above Muslim (see Appendix B); e.g., Arab, Middle Eastern, Iranian, Afghani, Pakistani) participants who lived in the United States, Canada, Europe (including: France, UK, Germany, Austria, Belgium, Netherlands, Denmark, Norway, Sweden, Bulgaria, Greece), New Zealand, or Australia. These European countries were included, as they each have at least a five percent (5%) Muslim population (Pew Research Center, 2017a). Relevant studies included samples collected within the Western countries listed. With respect to the Muslim-majority-country sample, we also included Indonesian individuals who live in the Western countries delineated, given that—although not a Muslim-majority (by 90% or above)—Indonesia has the largest population of Muslims as of 2015 (Diamant, 2019).

Studies with samples that were not exclusively Muslim or Muslim-majority, or in which the results were not stratified by religion, were still included when the sample was 90% Muslim (90% Muslim or Muslim-majority country of origin; e.g., 90% Pakistani). Authors were contacted for a Muslim-specific correlation when the study sample was at least 70% Muslim/Muslim-majority (n = 13).

We included Muslim-majority groups, given that, even though their religious identity may not have been measured, participants likely experience negative group discrimination due to the racialization process described earlier. Studies were included regardless of participants’ age or gender.

Language. We included articles only published in English.

Research design. Cross-sectional, longitudinal, mixed-method, and experimental studies were included. We included intervention studies when they included baseline measures of required variables. Otherwise, we excluded interventions and randomized control trials, as the intervention can affect the magnitude of the correlation. We also excluded systematic reviews and meta-analyses, qualitative studies, commentaries, editorials, book chapters, and corrections. However, we cross-referenced all systematic reviews and meta-analyses for relevant studies.

Time period. There were no exclusion criteria based on the date of publication.

The first author and research assistant coded at least 10 articles together to refine the coding spreadsheet and manual. Data-extraction notes for each article were documented following Hartgerink’s (2014) recommendations. Extracted data were saved in an Excel spreadsheet (see supplemental folder titled “Coding & Data” on OSF for all data extraction materials; https://osf.io/enb97/).

Reliability. Each article was coded independently by the first author and an undergraduate research assistant. The first author and research assistants discussed discrepancies and resolved them by consensus. The percent agreement across all items, a measure of interrater reliability, was 99%, which met our predetermined level of 90% (Siddaway et al., 2019; Syed & Nelson, 2015).³

Zero-order correlation coefficients. We extracted the study effect sizes (Pearson’s r coefficient) between discrimination and mental health or life stressors and mental health. The correlation coefficients were extracted in one of four different ways: (1) the Pearson’s correlation coefficient was included in the article; (2) the standardized beta from a bivariate regression model was included in the article; (3) a phi correlation was calculated using a contingency table of two dichotomous outcomes; (4) when the bivariate correlation was not reported—Chi-squared statistics with one degree of freedom, unadjusted odds ratio, and sample means and standard deviation were used to calculate the Pearson’s r coefficient using compute.es package in R (Del Re, 2013; Johnson & Eagly, 2014; R Core Team, 2020). We did not calculate correlation coefficients from regression coefficients with covariates. Authors were contacted for additional statistic information, if necessary, and were given one month to respond. All effect sizes retrieved from studies were coded so that increased discrimination is related to worse mental health/negative mental health. The effect sizes of increased discrimination and positive mental health were reverse-coded in estimating the combined summary effect size.

When a study included correlations for multiple outcomes, all effect sizes were included as a unique entry. We included only one study per dataset; when two studies used the exact same predictor and outcome variables, we included the most recent study with the largest sample size. When two studies use the same dataset, but one study focuses on adults and another on children, both samples were included; similarly, when studies included different variables but used the same dataset, all studies were included.

We used meta-analytic procedures to maintain independence of effect sizes that are outlined below. Notably, although most meta-analyses average effect sizes within a study or across studies with the same sample size, these methods have been found to be unreliable (Exec. Order No. 13769, 2017; Hedges et al., 2010; Moeyaert et al., 2017); therefore, we used alternative methods as described below.

Moderators. Moderators that fall under study-related, demographic-related, and identity-related variables were coded.

Study-related. Consistent with other meta-analyses, we extracted conventional publication characteristics (e.g., measures, study design; Siddaway et al., 2019; see the “Coding & Data” folder on the OSF project page for data extraction spreadsheet, https://osf.io/enb97). Additionally, we coded for discrimination-level (individual, group/systematic, and individual-group combined items), life-stressor type (e.g., trauma, general life, abuse, torture) and extracted the stressor measure used. We extracted measure information for both predictor and outcome measures: number of items, Cronbach’s alpha, response type, time period. Outcome was coded for type (e.g., depression, PTSD, self-esteem) and for positive or negative mental health measures. We coded for publication status of the article (i.e., published or unpublished).

Demographic-related. Various sociocultural characteristics—including mean age, gender, mean years of education, country of origin, region/continent, refugee/immigration status, and Muslim sample type (Muslim vs. Muslim-majority sample)—were coded to examine variations of the associations of sociocultural characteristics in the literature. Average age was computed when the mean age was reported for different genders or ethnic/racial groups. When the range was provided instead of the mean, we used the midpoint of the reported range.

Identity-related. Twenty-eight unique samples in the full combined discrimination and life-stressor dataset included identity-type variables that fit under ethnic identity (k⁴ = 18), national identity (k = 6), religious identity (k = 16), identity integration (k = 3), or visible identity (k = 2). The mean, standard deviation, and range were extracted for each identity measure.

Quality ratings. The quality of the articles was assessed using a modified version of National Institutes of Health’s (NIH) Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies (https://www.nhlbi.nih.gov/health-topics/study-quality-assessment-tools; the author modified version can be found in the “Coding & Data” folder at https://osf.io/enb97). The quality ratings included three areas of evaluation: (1) sample size (e.g., clearly specified study population, power analysis), (2) measurement (e.g., reliability of measures, exposure measured prior to outcome), and (3) robust disclosures (e.g., transparent analytical choices, disclosure of non-significant findings).

Studies rated as good included many of the following factors across the three areas of evaluation: moderate to large sample sizes or justification of the sample size with a power analysis, samples that were clearly defined, randomly recruited participants to increase generalizability, validated measures, pre-registration, transparent reporting of results, and alternative analysis models. Studies rated as fair included some of the factors listed under good or exceled in only one of the three areas of evaluation. Studies rated as poor included studies with small samples sizes, non-validated measures, selective reporting of significant results, evidence of p-hacking, or analyses without justification. The quality of the studies (good, fair or poor) are presented but are not used to exclude studies per recommendations from Siddaway et al. (2019). The quality of the studies was assessed by two raters with a percent agreement of 82%. All discrepancies were discussed by the research team and resolved by consensus.

To answer RQ1-RQ4, analyses were conducted separately for each stressor type (discrimination and life stressors), and to answer RQ5–RQ6, analyses were conducted with the combined data. To account for within-sample dependence, we conducted the main effect-size estimation analyses in two ways: (1) random-effects model (aggregating effect sizes across sample), and (2) robust variance estimation (RVE).

We used a random-effects model with the restricted maximum likelihood (REML) estimator to calculate the weighted mean correlation using the metafor package in R (R Core Team, 2020; Viechtbauer, 2010). Random-effects models assume the average effect size in the population varies randomly from study to study and, thus, the study effect sizes will be heterogenous (Field & Gillett, 2010). Based upon the small number of studies from preliminary literature searches, REML estimator is generally recommended (Veroniki et al., 2016; Viechtbauer, 2005). Additionally, random-effects models allow for generalizability of the findings to the population (Field & Gillett, 2010).

Because a random-effect model assumes independent samples, we created an aggregate effect size per sample using the agg function of MAc package in R for samples with multiple effect sizes (Hunter & Schmidt, 2004; R Core Team, 2020, see pp. 435-438; Del Re & Hoyt, 2010; D. S. Quintana, 2015). The extracted aggregated sample effect sizes were then converted to Fisher’s Z scores for calculations and later converted back to Pearson’s r coefficient for interpretation. Forest plots of study effect sizes and the overall random-effect model effect-size estimate are presented (see Figures 3 & 6).

We used robust variance estimation (RVE) using the robumeta package in R to calculate an effect-size estimate that accounts for within-study and within-sample dependence (Fisher & Tipton, 2015). We used the default ⍴ = .8 setting and conducted a sensitivity analysis for within-study effects (⍴) from the interval (0,1) (Fisher & Tipton, 2015). This method has been found to be robust with small samples (Tipton, 2015). The full extracted data, including sample duplicates, were converted to Fisher’s Z scores for calculations and later converted back to Pearson’s r coefficient for interpretation. Correlation coefficients were benchmarked according to the empirically-derived cut-offs described by Hemphill (2003) as r < .20 (small effect), .20 <= r < .30 (medium effect), and r >= .30 (large effect). The p-values and confidence intervals for the effect sizes are presented (alpha = .05 level). Forest plots of study effect sizes, RVE weights, and the overall RVE effect-size estimate are presented (Figures S1 (https://osf.io/yhnur/) & S6 (https://osf.io/xmk42/) in supplemental materials on OSF).

To examine heterogeneity between-studies, Cochran’s Q-statistic, I², tau-squared, and prediction interval statistics were calculated (M. Borenstein, 2019; J. P. Higgins et al., 2003; D. S. Quintana, 2015). Cochran’s Q-statistic is the standard significance test for between-study heterogeneity that demonstrates whether the variability is greater than what is expected from sampling error (D. S. Quintana, 2015; Siddaway et al., 2019). We estimated I^2 to examine the proportion of variance in the observed effect sizes that would be attributable to the actual difference between studies, specifically where I² values represent low (25), medium (50), or high (75) heterogeneity (Michael Borenstein et al., 2017; D. S. Quintana, 2015). Tau-squared was estimated to quantify between-study variance in random-effects models, where zero indicates no heterogeneity (D. S. Quintana, 2015). Prediction intervals were calculated as an index of dispersion of how widely the true effect size varied (M. Borenstein, 2019).

In order to see if any specific studies are overly-influencing heterogeneity, we used two strategies to examine for outliers. The distribution of effect sizes was examined for extreme values (i.e., outliers). Also, we used the metafor package in R to identify outliers and potential influential cases (D. S. Quintana, 2015; Viechtbauer, 2010). As a sensitivity test, we excluded effect sizes that were larger than 2.5 absolute deviations below or above the median or were identified as an influential case (Leys et al., 2013). We then calculated meta-analytic effect-size estimates with and without outliers to compare the results (see Figures 3 & 6).

In order to assess for publication bias, differences between effect-size estimates of unpublished and published studies, funnel plots, forest plots, and sensitivity analyses with weighted correlations were conducted (Jin et al., 2015). We used meta-regression with random-effect model using REML (mixed-effects model) and RVE to test when publication status was a significant predictor (Tipton, 2015). We calculated and compared separate effect-size estimates for published and unpublished studies using random-effects models for descriptive purposes.

Publication bias was explored through a funnel plot to visualize the effect size plotted against sample size. Publication bias may be present when the graph looks asymmetrical, specifically where more data points are toward larger effects and samples (Greco et al., 2013). We also explored publication bias through a funnel plot to visualize the effect size plotted against standard error.

Although the Egger test is widely used to test for publication bias, there is some evidence that suggests it leads to an inflated Type I error rate (Lin & Chu, 2018). Thus, the symmetry of the funnel plot was quantified by the Egger’s test, but it was interpreted cautiously. For an additional visual representation of publication bias, forest plots were generated and sorted by publication status and sample size.

In order to quantify the possible impact of publication bias, we conducted a sensitivity analysis and used the “moderate one-tailed selection” weight function outlined by Vevea and Woods (2005). The method proposed by Vevea & Woods (2005) has proven to be appropriate for meta-analyses with fewer than 100 studies. We used the weightr package (Coburn & Vevea, 2009) in R. Selection model approaches outperform other traditional methods (e.g., trim-and-fill method; Carter et al., 2019).

To answer RQ2 and RQ4, moderation analyses were conducted using meta-regression approaches in a random-effect model using REML (mixed-effects model) and/or RVE. These analyses were conducted using the metafor and robumeta packages in R, which allow for continuous and categorical moderators (Michael Borenstein et al., 2009; Fisher & Tipton, 2015; Viechtbauer, 2010). Only mixed-effects model results are presented, unless RVE approaches make more theoretical sense (i.e., where correlations within a sample were aggregated and important information was lost; for example, when mental-health outcome was aggregated, information was lost) or when results differed between the two models. All meta-regression results not presented can be found in the supplemental materials on OSF (https://osf.io/gkt8u/). Moderation analyses were conducted when there were at least 10 effect estimates per one covariate or per group as recommended for sufficient power to detect the true effect (Michael Borenstein & Higgins, 2013; Harrer et al., 2019; J. P. T. Higgins & Thompson, 2004). Each covariate was examined in an individual model.

For RQ2d, we tested whether level of discrimination (individual vs. group, vs. individual-group combined discrimination) was a significant predictor. We calculated separate effect-size estimates by level of discrimination using RVE for descriptive purposes.

For RQ2a-c and RQ4a-c, we tested whether study-related, demographic-related, or identity-related variables were significant moderators for the associations between discrimination and mental health and life stressors and mental health with each predictor in a separate model. We calculated separate effect estimates by specific group variables (i.e., continent/region, Muslim sample type) using random-effects models by REML or RVE for descriptive purposes.

For RQ5, we tested whether stress type (discrimination vs. life stressors) was a significant moderator by combining the discrimination and life stress data and conducting moderation analyses using mixed-method and RVE approaches. Additionally, we used equivalence testing to more accurately test the presence or absence of an effect using two TOST methods⁴—transformed correlation coefficients (TOST-p) and the untransformed TOST (KTOST-p) (Counsell & Cribbie, 2015; Lakens, 2017). We used the two one-sided tests (TOST) procedure to test for the presence of the smallest effect size of interest (SESOI). We chose the SESOI of r = .15, where the difference between effect sizes estimates needs to be greater than .15 (-.075, .075) to be relevant (Counsell & Cribbie, 2015; Lakens et al., 2018). We used Hemphill (2003) as a guide for a small effect to determine our SESOI.

For RQ6, we used systematic review methods to provide a qualitative review of the studies that include both discrimination and life stressors as predictors. Notably, 14 samples included both predictors. Since at least 10 samples were identified, we later coded for partial correlations between 1) discrimination and mental health (controlling for life stressors), and 2) life stressors and mental health (controlling for discrimination). However, only five samples included partial correlations.⁵

The list of articles, data spreadsheet, coding process description sheet, analysis scripts, and supplemental materials are available on OSF (https://osf.io/enb97/).

The full meta-analytic dataset includes 295 correlations, from 130 unique samples and 120 articles, including 27,725 individuals (full dataset can be found at https://osf.io/enb97/). The discrimination dataset includes 138 correlations, from 72 unique samples and 62 articles, including 16,042 individuals. In terms of quality ratings, 60% of discrimination samples were rated as fair, 22% good, and 18% poor. The life-stressor dataset includes 157 correlations, from 69 unique samples and 69 articles, including 15,099 individuals. In terms of quality ratings, 54% of life stressor samples were rated as fair, 10% good, 32% poor, and 4% combined fair and poor. Fourteen samples from 18 articles,⁶ including 3,416⁷ individuals, overlap between the discrimination and life-stressor datasets.

According to an a priori power analysis, we had adequate sample sizes to continue with our analyses. We used Quintana and Tiebel’s (2019) Excel calculator, which indicated that we needed at least 10 studies for a power of .80, assuming r = .2, an average number per group = 85, and moderate heterogeneity (D Quintana, 2017).

We conducted the main effect-size estimation analyses in two ways: (1) random-effects model (aggregating effect across sample), and (2) robust variance estimation (RVE). Thus, analyses across both methods resulted in four main effect-size estimates for the association between 1) discrimination and negative mental-health outcomes, and 2) life stressors and negative mental-health outcomes (no additional covariates were included). We first present the results for the discrimination dataset, then the life-stressor dataset; and then last, we compare the two predictors. We conducted the same data-analyses plan outlined in the data-analysis section earlier for each stressor dataset. Therefore, in the section below, we aim to reduce redundant information as much as possible between the discrimination and life-stressor datasets results by removing repetitive information when presenting the life-stressor dataset results.

Table 1 summarizes the individual sample (i.e., dataset) details included in the discrimination meta-analysis. Table 2 summarizes descriptive statistics of the discrimination dataset (see supplemental materials, Table S7 (https://osf.io/gkt8u/) for additional individual sample and correlation details, including measure names). The 72 unique samples had a mean sample size of 244 (SD = 206.79; range = 30–1215) and represented individuals living in 10 different Western countries (see Table 1). The mean age of participants was 26 years old (SD = 9.73; range = 10–49 years old), and 55% were women; the collapsed majority ethnic breakdown was 61% Middle Eastern, 15% Asian/South Asian, 12% North African, 9% Somali, and 4% other combination majority (see Table 1 for individual sample information and Table 2 for a more detailed breakdown of the majority ethnicities represented by each sample). In terms of discrimination-level, the majority of samples used a measure of personal discrimination (62.5%), followed by a measure of combined personal and group discrimination items (30.6%), and last, a measure of group discrimination (6.9%).⁸ Samples measured diverse mental-health outcomes, including depression (k = 5), PTSD (k = 1), anxiety (k = 1), and self-esteem (k =11), along with broader categorizations, such as, psychological distress (k = 15), well-being (k = 6), and internalizing disorders (k = 8).⁹ Many samples also measured multiple types of mental-health outcomes, such as, anxiety and depression (k = 3), depression and well-being (k = 2), PTSD and depression (k = 2), well-being and psychological distress (k = 3).

Weighted mean correlations. Figure 3 shows the weighted mean correlations using both a random-effects model with the restricted maximum likelihood (REML) estimator and RVE estimator, along with the univariate effect-size estimates and their 95% confidence intervals for the 72 samples in the discrimination dataset. Using the random-effects model approach, we found a statistically significant summary effect, r = .234 (95% CI [.20, .27]), z = 11.95 p < .001. Using the RVE approach, we also found a statistically significant summary effect, r = .215 (95% CI [ .18, .25]), t (68.8) = 11.96 p <.001. The RVE estimate is slightly attenuated compared to the random-effects estimate. Figure S1 (https://osf.io/yhnur/) includes the RVE weight estimates per correlation in each study. Both estimates have confidence intervals that showed the true association is unlikely to be zero, and the narrow confidence intervals show precision in this summary effect. Table 3 summarizes these statistics with three decimal places. Consistent with our hypothesis, increased experiences of discrimination are positively associated with worse mental-health outcomes for Muslims living in Western countries. The summary correlation is positive and falls between the small to moderate range using Hemphill’s (2003) guidelines (r < .20 (small effect), .20 <= r < .30 (medium effect), and r >= .30 (large effect)).

Heterogeneity. The data showed a statistically significant and large degree of between-study variability, Q(71) = 338.30, p < .001, τ² = 0.022, I² = 82.77% (see Table 3). Heterogeneity examines how much the variation between studies is true variation in the population. The statistically significant Q statistic indicated that the included studies do not share a common true effect size. The I² statistic indicated substantial heterogeneity in the true effect, meaning that a large proportion of the observed variance between discrimination and mental-health outcomes was attributed to variances in the true effect rather than sampling error. The index of dispersion for the correlation of interest was -0.055 to 0.487, which shows that, within the population, bivariate correlations would fall between this range 95% of the time. Therefore, in some cases there is no effect; in other cases, there are worse outcomes, but the data suggests that experiences of discrimination are not associated with any positive mental-health outcomes. Given the large index of dispersion, it is important to examine potential moderators of the association between discrimination and mental-health outcomes.

Influential cases and outliers. Figure 3 indicates the five influential samples identified through examining DFFITS values, Cook’s distances, hat values, DFBETAS values, and effect sizes 2.5 absolute deviations below/above the median. As a sensitivity analysis, we removed the five influential samples and recalculated the effect-size estimate (k = 67, k_c = 130). The weighted random-effects and RVE estimates were unchanged, still statistically significant with high heterogeneity, r = .231 (95% CI [.20, .26]) and r = .211 (95% CI [ .18, .24]), respectively (see Figure 3 for effect-size estimates without outliers and Figures S2, https://osf.io/kfthj/, & S3 https://osf.io/dwc8a/, for all outlier statistics). Additionally, we conducted a leaveout analysis in R where each sample was removed, and the random-effect model estimate was recalculated. Each influential case slightly attenuated the omnibus effect size but did not alter the statistical significance of the effect, and heterogeneity remained high (see Table S1, https://osf.io/gkt8u/). Thus, findings showed that influential cases and outliers did not alter the omnibus weighted mean correlation or account for the heterogeneity between studies.

Sensitivity analysis for between- and within-study variance. Table 3 also summarizes the results from the RVE method using different levels of ρ from 0.0 to 1.0 where ρ = 1 is the most conservative approach. The sensitivity analysis did not affect the summary correlation, confidence intervals, or τ². Overall, findings showed that the omnibus weighted mean correlation was not sensitive to the dependence of within-study effects.

Publication bias. Table 4 summarizes the meta-regression results for publication status; 85% of samples were from published (k = 61) versus unpublished studies (k = 11). We conducted meta-regressions using both a random-effects model using REML (mixed-effects model) and RVE; using these two methods we found—inconsistent with our hypothesis—that publication status was not a significant moderator, z = -.435, p = .664 (Table 4, mixed-effects model) and t (13.34) = -.696, p = .498 (Table S2, https://osf.io/gkt8u/, RVE model). Using random-effects models, the effect-size estimate for published data was r = .238 (95% CI [.197, .277]) and the effect-size estimate for unpublished data was r = .214 (95% CI [.109, .313]). Second, we created two funnel plots: a) to visualize the effect size plotted against sample size, and b) effect size plotted against standard error. Publication bias may be present if the graph looks asymmetrical, with more data points favoring larger effects and samples. Visually, both funnel plots seemed symmetrical (see Figures 4 - 5). The symmetry of the two funnel plots was quantified by the Egger’s test, and both results were not significant (sample size: z = .071, p = .943; standard error: z = -.393, p = .694).

For an additional visual representation of publication bias, forest plots were created and sorted by publication status and sample size, and neither displayed any obvious pattern (see Figure S4 sorted by publication status, https://osf.io/njtvk/, and Figure S5 sorted by sample size, https://osf.io/abx7t/).

We conducted a sensitivity analysis developed by Vevea & Woods (2005). Using a “moderate one-tailed selection” as proposed Vevea & Woods (2005), we created intervals of p-values. We altered the weights so that significant p-values were grouped together; there were at least four effect sizes included in each interval. The adjusted effect-size estimate was r = .207, which although slightly attenuated, is still within the confidence interval of the omnibus effect size. Thus, the effect-size estimate between discrimination and mental health seemed to be robust to publication bias based on the current methods used to detect and correct for bias.

We examined study-related (i.e., study-design characteristics, stress measure, and mental health measure¹⁰), demographic-related, and identity-related moderators when there were at least 10 effect size estimates per one covariate or per group, as recommended, in order to aim for sufficient power to detect the true effect (Michael Borenstein & Higgins, 2013; Harrer et al., 2019; J. P. Higgins et al., 2003). Most analyses were conducted using mixed-effects models unless specified that RVE meta-regression was a better approach. When results differed between the two models, both results were presented. Supplemental Tables S2 and S3 (https://osf.io/gkt8u/) include the categorical and continuous moderator results for the model not presented in this section.

Study-design characteristics. In terms of study design, all but one sample in the discrimination dataset used a cross-sectional design. Therefore, there was insufficient power to conduct moderation analyses between cross-sectional and longitudinal designs. We examined the type of sample population as a categorical moderator of the association between discrimination and mental health (see Table 4). The majority of samples were community-based samples (k = 47), followed by school (k = 12), official organization (k = 4), clinic (k = 2), and a combination of samples (k = 7). We grouped organization, clinic, and combination samples together so that there were at least 10 samples in each category. Using mixed-effects modeling, the type of sample population did not explain the between-study variance of the association between discrimination and mental health, Q_between (2) = 0.279, p = .870. We also examined when correlation type—a categorical variable—was a moderator (see Table 4). The majority of samples included a Pearson correlation (k = 63), however, we did calculate a Pearson correlation for some of the samples from various statistics: chi-squared (k = 2), odds ratio (k = 2), and phi correlation (k = 5). We grouped all non-Pearson correlations together to include almost 10 (k = 9), and correlation type was not a significant moderator, z = 1.097, p = .273. Although there was insufficient information to test if study-design was a significant moderator, the results for the type of sample population and correlation type seem to be robust (see Table S2, https://osf.io/gkt8u/, for RVE results).

Stress measure. We examined characteristics of discrimination measures as moderators, which included number of items, the measure’s internal reliability or Cronbach’s alpha, and respondent (Table 4). The average number of discrimination measure items was 9.56 (SD = 7.96; range = 1-36; k = 67), and was a significant moderator, z = 2.458, p = .014, such that a one-item increase above the mean number of items (M = 9.56) resulted in a r = .006 increase from r = .233. The average Cronbach’s alpha or measure of internal reliability was 0.84 (SD = 0.08; range = 0.59 – 0.96; k = 55), and was not a significant moderator of the association between discrimination and mental health, b = .158, z = .528, p = .597. Measure respondent was examined using RVE meta-regression approach, as one sample included discrimination reports from multiple respondents. Out of the 138 correlations, the majority included a self-report measure (k_c = 112), followed by an interview (k_c = 21), and a combination of self-report and a parent interview (k_c = 5). Due to the numbers per group, we grouped together all non-self-report measures and found respondent did not explain between-sample variation in effect size, t (12) = 1.47, p = .166 (see Table 4). Based upon this data, the results are mixed as it relates to our hypothesis, specifically where number of measure items was a significant moderator between discrimination and mental health, but other discrimination measure quality-related variables (Cronbach’s alpha and respondent) were not.

Discrimination-level was examined using RVE meta-regression approach, as 11 samples included both correlations for group and personal discrimination. Out of the 138 correlations, the majority included a measure of personal discrimination (k_c¹¹ = 92), followed by group discrimination (k_c = 28), and then a combination measure of both personal and group discrimination (k_c = 18). Discrimination-level was a significant moderator of the association between discrimination and negative mental health (Table 4). Personal discrimination, r = .23, p < .001 resulted in a significantly higher estimate than group discrimination, r = .10, (t (17.8) = -2.969, p = .008). Using a RVE approach, all three levels of discrimination were significantly associated with negative mental health (Table 5). Thus, consistent with our hypothesis, these results suggest that—although group discrimination is significantly associated with negative mental-health outcomes—personal discrimination may be more detrimental than systematic/group discrimination.

Mental health measure. Positive vs. negative mental health was examined using RVE meta-regression approach, with nine samples including both correlations for positive and negative mental health. Out of the 138 correlations, the majority included a measure of negative mental health (k_c = 103), and only 25% of all correlations measured positive mental health (k_c = 35). Consistent with our hypothesis, mental-health outcome direction was a significant moderator between the association of discrimination and mental health, with the relation between discrimination and positive mental-health outcomes, r = -.12, p <.001, being significantly weaker than discrimination and negative mental-health outcomes, r = .25, p <.001 (t (38.2) = -3.87, p < .001; Table 4).

Twenty-three samples included multiple mental-health outcome correlations. Out of the 138 correlations, the breakdown of mental-health outcomes was as follows: anxiety (k_c = 6), depression (k_c = 22), externalizing (k_c = 9), internalizing (k_c = 23), interpersonal (k_c = 5), psychological distress (k_c = 26), psychotic (k_c = 1), PTSD (k_c = 6), self-esteem (k_c = 20), somatization (k_c = 4), and well-being (k_c = 16). Because many of the outcomes included fewer than 10 correlations, we were unable to perform a moderation analysis of mental-health outcome. However, we conducted subgroup analyses using RVE for the mental-health outcomes with more than 10 correlations (see Table 6).¹² The results suggest that all outcomes, except for self-esteem and somatization, are significantly associated with discrimination. To highlight the results from specific mental-health outcomes, discrimination was most strongly associated with depression (r = .30; k_c = 22), followed by psychological distress (r = .23; k_c = 26), interpersonal outcomes (r = .22; k_c = 5), externalizing outcomes (r = .22; k_c = 9), anxiety (r = .22; k_c = 6), and well-being (r = -.17; k_c = 16). These results suggest that discrimination has a stronger relationship with negative, psychopathological mental-health outcomes.

We also examined quality-related moderators of mental health measures,¹³ which included number of items, the measure’s internal reliability or Cronbach’s alpha, respondent, and response time period (see Table 4). The average number of mental-health measure items was 16.62 (SD = 19.82; range = 1–115; k = 67), and was a not a significant moderator, b = .002, z = 1.678, p = .093. The average Cronbach’s alpha (index of internal reliability) was 0.84 (SD = 0.08; range = 0.60–0.96; k = 58) and was not a significant moderator, b = -.168, z = -0.624, p = .533. Measure respondent was examined using RVE meta-regression approach, as two samples included mental-health outcomes from multiple respondents. Out of the 138 correlations, the majority included a self-report measure (k_c = 108), followed by an interview (k_c = 21), interview with parent (k_c = 6), and teacher-report (k_c = 3). Due to the numbers per group, we grouped together all non-self-report measures and found measure respondent was not a moderator of the association between discrimination and negative mental health, t (13.6) = 1.22, p = .244. Because self-report measures were the overwhelming majority, it is important to note that the shared method variance is likely leading to an inflated correlation, despite these tests suggesting no difference. The average response time period was 49 days (SD=64.62, range = 3-180) and was not a significant moderator. In sum, based upon this data, no mental-health measure-quality significantly moderated the association between discrimination and negative mental-health outcomes.

Demographic characteristics. Table 7 summarizes the meta-regressions for sociocultural characteristics. Sample mean age (M = 26.13, SD = 9.73) was tested as a continuous moderator in the mixed-effects meta-regression model and did not explain between-sample variation in effect size (b = -.002, z = -1.235, p = .217). Inconsistent with our hypothesis, the percent of women (M = 55% women, SD = 17.77), tested as a continuous moderator, did not explain between-sample variation in effect size (b = .0001, z = .092, p = .927). In terms of education, 40% of the meta-analytic sample had a bachelor’s degree or higher; however, education data was extracted from only 33 samples. Education as a continuous moderator did not significantly moderate the association between discrimination and mental health (b = -.0001, z = -.096, p = .924).

The majority of samples were from Europe (k = 40), followed by North America (k = 26), and then by Australia/New Zealand (k = 6). Continent was tested as categorical moderator and did not explain between-study variance, Q_between (2) = 3.315, p = .191. Although there were not 10 samples per continent, we performed the analysis given that there were more than 10 correlations in the nonaggregate dataset (analyses demonstrate consistent findings; see Table S2, https://osf.io/gkt8u/, for RVE meta-regression results).

Because there were not 10 samples per country, we were unable to perform moderation analysis. Instead, we present the random-effect model estimates between discrimination and negative mental health for each country (see Table 8). To note, only the estimates for the Netherlands and the United States are a priori analyses, whereas the estimates for countries with fewer than 10 samples are exploratory analyses. Using the random-effects model approach, we found a statistically significant summary effect between discrimination and negative mental health for samples in the Netherlands (k = 15), r = .206 (95% CI [.16, .25]). We also found a statistically significant summary effect for samples in the United States (k = 23), r = .254 (95% CI [.17, .34]).

In order to test if ethnicity or country of origin was a significant moderator, we conducted our analysis with the samples that had 80% or more of the same ethnic group (k = 51). Ethnicity did not account for between-study variance in the estimate using either approach. We also tested to see whether Muslim-sample type (Muslim vs majority Muslim country) was a significant moderator. There were 39 Muslim samples and 33 majority-Muslim-country samples. Muslim sample type was not a significant moderator, z = -.567, p = .571 (see Table 7 and Table S2, https://osf.io/gkt8u/, for random-effect estimates).

Immigration status was examined through percent born in the western country (60%; k = 33), percent first generation immigrants (61%; k= 34), and percent refugee (76%; k = 9; see Table 7). Percent born in a western country was tested as continuous moderator and did not explain between-study variance in effect size (b = .0001, z = .053, p = .958); percent first generation immigrant did not explain between-study variance in effect size either (b = -.001, z = -.730, p = .465). Percent refugee was slightly unpowered, as only nine samples included that information; tested as a continuous moderator, refugee status did explain between-study variance in effect size, with refugee status related to an r = .002 increase in the association between discrimination and negative mental health. These results suggest that among the indicators of immigration status, only refugee status moderated the association between discrimination and mental health (see Table 7 and Table S3, https://osf.io/gkt8u, for random-effect estimates).

Identity. Twenty-two unique samples included identity variables in the discrimination dataset. Identity variables included: ethnic identity (k = 15), religious identity (k = 12), national identity (k = 4), identity integration (k =1), and visible identity (k = 2).¹⁴ The mean, standard deviation, and range were extracted for each identity measure and used to calculate the Percentage of Maximum Possible (POMP) score, which is a useful way to standardize across different metrics of measurements (Cohen et al., 1999; Fischer & Milfont, 2010). There was only enough information to conduct the moderation analyses for ethnic identity and religious identity,¹⁵ therefore, qualitative reviews of other identity moderators were conducted.

Ethnic identity. Fourteen samples included the information needed to conduct the moderation analysis of ethnic identity on the association between discrimination and negative mental health. Inconsistent with our hypothesis, the results showed that ethnic identity (M = 68.64, SD = 17.18; k = 14) did not explain the between-study variance in the effect-size estimate, b = -.001, z = -.19, p = .85 (see Table 9 and Table S3, https://osf.io/gkt8u/).

The qualitative review that includes 15 samples, with 33 correlations, showed varied results for the moderating effect of ethnic identity on the association between discrimination and mental health. The results varied based upon the subscale of ethnic identity (e.g., centrality vs. private regard), type of discrimination (group vs. personal) and outcome type (e.g., depression vs. anxiety), but there does not appear to be a pattern among the results. Three samples found that stronger ethnic identity protects against the deleterious association between discrimination and mental health (Musa, 2015; Schaafsma, 2011; Stevens & Thijs, 2018). Four samples found that stronger ethnic identity exacerbates the deleterious effects of the association between discrimination and mental health (Alemi & Stempel, 2018; Jasperse et al., 2012; Lowe et al., 2019; Stevens & Thijs, 2018). Finally, five samples found that ethnic identity did not moderate the association between discrimination and mental health (S.R. Ahmed, 2006; Alemi et al., 2017; Lowe et al., 2019; Musa, 2015, 2018).

Three samples suggest that stronger ethnic identity protects against the deleterious association between discrimination and mental health (Musa, 2015; Schaafsma, 2011; Stevens & Thijs, 2018). In a sample of Turkish and Moroccan individuals living in the Netherlands, ethnic identity moderated the relationship between subtle and blatant discrimination and well-being. For those who reported high ethnic identity and high subtle discrimination (interaction ß = .10, p <.05) or blatant discrimination (interaction ß = .08, p <.05), they had higher well-being compared to those who reported lower ethnic identity and high subtle or blatant discrimination (Schaafsma, 2011). Additionally, in a sample of Muslim-American adults, ethnic-identity centrality was found to be a protective factor against anxiety symptoms in the context of discrimination (interaction ß = -.09, p = .04), and ethnic-identity-based private regard moderated the association between discrimination and anxiety (where lower private regard was a risk factor for increased anxiety; interaction ß = -.08, p = .04; Musa, 2015). Similarly, in a sample of Dutch Moroccan adolescents, ethnic identity moderated the association between group discrimination and internalizing outcomes, with ethnic identity being a protective factor in the association between the experience of group-based discrimination and personal internalizing outcomes (interaction ß = -.19– -.25, p <.01; Stevens & Thijs, 2018). Thus, these results suggest there is some evidence of a small moderating effect of stronger ethnic identity as a protective factor of the deleterious association between discrimination and mental health.

Four samples suggest that stronger ethnic identity exacerbates the association between discrimination and mental health (Alemi & Stempel, 2018; Jasperse et al., 2012; Lowe et al., 2019; Stevens & Thijs, 2018). In a sample of Afghan refugees who are resettled in the United States, ethnic identity moderated the relation between personal discrimination and psychological distress, with increased ethnic identity exacerbating the negative association between discrimination and negative mental health (interaction ß = .28, p <.01; Alemi & Stempel, 2018). Similarly, in a sample of Dutch Moroccan adolescents, ethnic identity moderated the association between personal discrimination and internalizing outcomes, with ethnic identity exacerbating the deleterious effects of personal discrimination on internalizing outcomes (interaction ß = .14–.20, p <.01–.05; Stevens & Thijs, 2018). Furthermore, in a sample of Muslim women in New Zealand, ethnic identity moderated the relation between religious discrimination and psychological symptoms (interaction ß = .27, p <.01) and life satisfaction (interaction ß = -.19, p <.05), and individuals with stronger ethnic identity contain greater risk for negative consequences of discrimination (Jasperse et al., 2012). Similarly, in a sample of Muslim-American students, ethnic identity moderated the relation between generalized anxiety disorder symptoms and discrimination; individuals with stronger ethnic identity appear to experience higher levels of discrimination and anxiety symptoms (interaction ß = .01, p <.05; Lowe et al., 2019). Thus, these results suggest there is some evidence that increased ethnic identity may exacerbate the deleterious association of discrimination and mental health.

Five samples found that ethnic identity did not moderate the association between discrimination and mental health (S.R. Ahmed, 2006; Alemi et al., 2017; Lowe et al., 2019; Musa, 2015, 2018). Then, studies from six samples did not conduct moderation analyses (Ameline et al., 2019; Hakim et al., 2018; Husain, 2008; Mewes et al., 2015; Shah, 2018; Verkuyten & Nekuee, 1999).

These mixed results from the qualitative review are parallel with results from a recent meta-analysis on the moderating role of ethnic identity on the association between discrimination and adjustment (Yip et al., 2019), with varied dimensions of ethnic identity being associated differentially with psychological outcomes. Yip and colleagues (2019) found that results also varied based on sample demographics. Therefore, this qualitative review of Muslim samples highlights a varied impact of ethnic identity on the association between discrimination and mental health. Therefore, there is a need for preregistered, high-powered studies that use validated measures of ethnic identity to examine the moderating effect of components of ethnic identity on different types of discrimination/mental-health outcomes for Muslim samples.

Religious identity. Eight samples included the information needed to conduct the moderation of religious identity on the association between discrimination and negative mental health. Inconsistent with our hypothesis, the exploratory results showed that religious identity (M = 78.13, SD = 6.95; k = 8) did not explain the between-study variance in effect size, b = .0004, z = .03, p = .97 (see Table 9 and Table S3, https://osf.io/gkt8u/).

The qualitative review of the 12 samples, which included 15 correlations, showed that in three samples, religious identity served as a protective factor that weakened the association between experiences of discrimination and mental-health outcomes (Z. Adam & Ward, 2016; Jasperse et al., 2012; Stuart & Ward, 2018); also, in three additional samples, no significant association was found (Z. Adam & Ward, 2016; S.R. Ahmed, 2006; Stuart & Ward, 2018).

Three unique samples from New Zealand found that religious identity moderated the association between discrimination and mental health, with stronger religious identity buffering against the negative association of discrimination on mental-health outcomes (Z. Adam & Ward, 2016; Jasperse et al., 2012; Stuart & Ward, 2018). In a sample of Muslim women in New Zealand, religious identity was found to moderate the relationship between a) religious discrimination and life satisfaction (interaction ß = .18, p <.05) and b) religious discrimination and psychological distress (interaction ß = -.30, p <.01), with those who were less engaged in religious practice being at greater risk for negative consequences of discrimination (Jasperse et al., 2012). Similarly, in an ethnically diverse sample of Muslims living in New Zealand, Z. Adam & Ward (2016) found that behavioral religious coping moderated the relation between personal discrimination and life satisfaction (interaction ß = .79, p <.05); specifically, the negative effects of discrimination on life satisfaction were attenuated for individuals who engage in high levels of behavioral religious coping (Z. Adam & Ward, 2016). Moreover, in a sample of New Zealand Muslims, religious identity moderated the association between discrimination and depression with religious identity buffering the detrimental influence of discrimination on depression (interaction ß = -.49, p <.001; Stuart & Ward, 2018). Thus, these findings suggest a moderate to large effect of religious identity being protective against the negative impact of discrimination on mental-health outcomes for Muslims in New Zealand.

Three samples found that religious identity did not moderate the association between discrimination and mental health (Z. Adam & Ward, 2016; S.R. Ahmed, 2006; Stuart & Ward, 2018). Seven samples did not include moderation analyses (Hakim et al., 2018; Hashemi et al., 2019; Hernandez, 2012; Husain, 2008; Musa, 2018; Stevens et al., 2005a, 2005b; Wilson, 2017). In summary, the qualitative review suggests religious identity may either protect or have no effect on the deleterious association between discrimination and mental health.

National identity. Four samples included national identity (Hakim et al., 2018 (s1,s2); Schaafsma, 2011; Thijs et al., 2018). Only Schaafsma (2011) conducted moderation analyses, and in a sample of Turkish and Moroccan individuals living in the Netherlands, Schaafmsa found that stronger national identity exacerbated the negative association between a) subtle discrimination and well-being (interaction ß = -.07, p <.05) and b) blatant discrimination and well-being (interaction ß = -.05, p <.05). Although the results were significant, the strength of association is near zero.

Identity integration. Only one sample included a measure of identity integration (Alemi & Stempel, 2018). Researchers found that, in a sample of Afghan refugees resettled in the United States, identity integration did not moderate the relation between discrimination and mental health.

Visible identity. Two samples included a measure of visible identity, and no moderating effect was found. Jasperse and colleagues (2012) did not find a significant moderating effect of visible identity between religious discrimination and life satisfaction or psychological symptoms. Musa (2015) did not conduct moderation analyses.

The summary correlation between experiences of discrimination and mental health for Muslims living in Western countries was consistent with our hypothesis—it was positive and falls within the small to moderate range. The estimate was robust as it relates to outliers, dependence within-study effects, and publication bias with the methods used to detect and correct for biases. There was high heterogeneity within the sample, but it appeared that the heterogeneity was due to true variance within the population. In terms of demographic characteristics, these meta-analytic results were from an ethnically and geographically diverse, educated, and young Muslim population.

Interestingly, both discrimination level and mental-health-outcome direction significantly explained between-study variance in the effect-size estimate. Consistent with our hypothesis, discrimination level was a significant moderator with personal discrimination having a stronger association with mental health compared to group discrimination; but, importantly, all discrimination types were significantly association with mental-health outcomes. Also consistent with our hypothesis, mental-health-outcome direction was a significant moderator between the association of discrimination and mental health, specifically where the relation between discrimination and positive mental-health outcomes was significantly weaker than discrimination and negative mental-health outcomes; but, again, both positive and negative outcomes were significantly associated with discrimination.

All moderation analyses outlined in our research questions were challenging due to limiting information available in articles and small samples per moderator. The meta-analytic estimate appeared to be unaffected by most other study-, demographic- and identity-related moderators that were tested. However, evidence suggested that the number of discrimination-measure items was a significant moderator; thus, as the number of discrimination-measure items increased, the association between discrimination and mental health grew stronger. Additionally, although slightly underpowered by one study, refugee status appeared to be a significant moderator of the association between discrimination and negative mental-health outcomes, specifically where refugee status increased the association between discrimination and negative mental health.

We conducted a parallel meta-analysis for the life-stressor dataset. We follow the same analysis plan outlined in the data analysis section. In the section below, we aim to reduce redundant information that was already explained in greater detail while presenting the meta-analytic results on discrimination.

Table 10 summarizes individual sample (i.e., dataset) details included in the life-stressor meta-analysis, and Table 11 summarizes descriptive statistics of the life-stressor dataset (see Table S8, (https://osf.io/gkt8u, for additional individual sample and correlation details, including measure names). The 69 unique samples had a mean sample size of 236 (SD = 236.07; range = 10–1,215) and represented individuals living in 11 different Western countries (see Table 11). The mean age of participants was 30 years old (SD = 11.82; range = 3–49 years old), with 49% women and a collapsed majority ethnic breakdown of 59% Middle Eastern, 25% Somali, 9% Asian/South Asian, 6% Bosnian Peninsula, and 1% North African (see Table 11 for a detailed ethnicity breakdown and Table 10 for breakdown individual majority ethnicities represented by each sample). Many life stressors were measured, including abuse (1%; k =1), bullying (1%; k =1), general life (13%; k =9), migration trauma (3%; k =3), physical abuse (1%; k =1), verbal abuse, sexual trauma (1%; k =1), trauma (1%; k =1), war trauma (46%; k =32), and a combination of these stressors (19%; k =13; e.g., war trauma and general life (k =2)).¹⁶ Samples measured diverse mental-health outcomes, including depression (k = 9), PTSD (k = 10) and anxiety, and broader categorizations, such as psychological distress (k = 13), wellbeing (k = 1), somatization (k = 3), and externalizing (k = 1). Many samples also measured multiple types of mental-health outcomes (k = 32), such as depression and PTSD (k = 10), PTSD and psychological distress (k = 3), and depression and anxiety (k = 2).¹⁷

Weighted mean correlations. Figure 6 shows the weighted mean correlations, using both a random-effects model with the restricted maximum likelihood (REML) estimator and RVE estimator, along with the univariate effect-size estimates and their 95% confidence intervals of the 69 samples in the life stressors dataset. Using the random-effects model approach, we found a statistically significant summary effect, r = .366 (95% CI [.31, .42]), z = 11.30 p < .001. Using the RVE approach, we also found a statistically significant summary effect, r = .320 (95% CI [ .27, .36]), t (64.1) = 12.99, p <.001. The RVE estimate is slightly attenuated compared to the random-effects estimate. Figure S6 (https://osf.io/xmk42/) includes the RVE weight estimates per correlation in each study. Both estimates have confidence intervals that showed the true association is unlikely to be zero, and the narrow confidence intervals show precision in this summary effect. Table 12 summarizes these statistics with three decimal places. Consistent with our hypothesis, the summary correlation between life stressors and mental health for Muslims living in Western countries falls between the moderate to large range.