Research suggests that experiences of discrimination and life stressors are associated with negative mental-health outcomes for Muslim populations in western countries. The current study reports on two meta-analyses based on 295 correlations from 130 unique samples and 27,725 individuals, examining the associations of discrimination and life stressors, both separately and jointly, with mental health. Discrimination was significantly associated with negative mental-health outcomes (rs = .22–.23). Between-study variability in effects sizes was explained by discrimination level, mental-health outcome, number of discrimination measure items, and refugee status. Life stressors were also significantly associated with negative mental-health outcomes (rs = .32–.37). Between-study variability in effect sizes was explained by publication bias, sample population, number of life stressor measure items, continent, and ethnicity. Both omnibus effect-size estimates were robust to tests of publication bias, outliers, and within-study dependence. Results suggest unique associations between both discrimination and life stressors with mental health. In the current sociopolitical climate, this study is an important step to better serve the mental health needs of the growing global Muslim community.
Muslims have a long history of being “othered” and subjected to discrimination, dating from the colonization of land ruled by the Ottoman Empire to U.S. Executive Order 13769 (also known as the ‘Muslim ban;’ in 2017), a law banning individuals from Muslim-majority countries from attaining a U.S. visa (Poynting & Mason, 2007; Runnymede Trust, 1997). Indeed, Islamophobia is widespread in contemporary society, evidenced by anti-Muslim stereotypes and propaganda across North America, Europe, New Zealand, and Australia (A. S. Ahmed, 2019; Trevino et al., 2010). As a result, a pressing question has emerged: how are the patterns of Islamophobia impacting the psychological well-being of Muslims living in western countries?
Racialization, Islamophobia, and Discrimination
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
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).
Islamophobia
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
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.
Life Stressors and Discrimination
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.
Stressors and Health
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).
Discrimination
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.
Life Stressors
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.
Potential Moderators
Demographic
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.
Identity
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.
Study-Related Moderators
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.
Present Study
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.
Research Questions
All hypotheses described below do not include control variables.
RQ1: What is the bivariate effect-size estimate between discrimination (combined individual, group, and individual-group discrimination measures) and negative mental health for Muslims living in Western countries?
Hypothesis: We expected a positive relation between increased discrimination and negative mental-health outcomes.
RQ2: What study-related, demographic-related, and identity-related variables moderate the relation between discrimination and negative mental-health outcomes?
Study-Related Variables
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.
Demographic-Related Variables
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.
Identity-Related Variables
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.
RQ3: What is the bivariate effect-size estimate between life stressors and mental-health outcomes for Muslims living in Western countries?
Hypothesis: We expected a positive relation between increased life stressors and negative mental-health outcomes.
RQ4: What study-related, demographic-related, and identity-related variables moderate the relation between life stressors and negative mental-health outcomes?
Study-Related Variables
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.
Demographic-Related Variables
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.
Identity-Related Variables
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.
RQ5. Is stressor type (discrimination vs. life stressors) a significant predictor of the association between stressors and 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.)
RQ6. Do discrimination and life stressors uniquely predict negative mental-health outcomes when in the same model?
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.
Method
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.
Literature Search
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).
Inclusion and Exclusion Criteria
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.
Coding
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.
Data Analysis Plan
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).
Random-Effects Model
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).
Robust Variance Estimation
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).
Heterogeneity
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).
Publication Bias
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).
Moderation
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/).
Results
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.
Discrimination
Sample characteristics
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).
Sample | N | Age | %F | Ctry | Muslim type | %Ethnicity | Sample Type | Pub | Design | Quality | r | # corr /sample | Stress Types | MH Types | Identity |
Adam & Ward, 2016 | 167 | 32 | 65 | NZ | Muslim | 32.9 IN/SAS, 17.4 EA, 16.8 MDE, 16.2 W, 9 AFR 4.8 PASF | comm. | P | C | fair | 0.32 | 2 | P.DIS | W.B, DIS | Y |
Ahmed, 2006 | 240 | 16 | 48 | USA | Muslim | 48.3 LB, 19.2 IQ, 15.8 YE, 16.7 OTHA | school | T | C | good | 0.27 | 4 | C.DIS | EXT, INT | Y |
Aichberger et al., 2015 | 205 | 37 | 100 | DE | Turkish | 100 TR | comm. | P | C | fair | 0.34 | 1 | P.DIS | DIS | |
Alemi et al., 2017 | 133 | 26 | 64 | USA | Afghan | 100 AF | comm. | P | C | fair | 0.36 | 1 | P.DIS | DEP | Y |
Alemi & Stempel, 2018 | 259 | 49 | 51 | USA | Afghan | 100 AF | comm. | P | C | fair | 0.27 | 1 | P.DIS | DIS | Y |
Ameline et al., 2019 | 88 | 32 | 53 | FR | Muslim | 40 TR/MA/DZ, 60 born FR | comm. | P | C | poor | -0.17 | 1 | G.DIS | SE | Y |
Brenick et al., 2018 | 297 | 12 | 49 | DE | Turkish | 100 TR | school | P | C | fair | 0.29 | 2 | P.DIS | DEP, W.B | |
Cakir & Guneri, 2011 | 248 | 34 | 100 | UK | Turkish | 100 TR | comm. | P | C | poor | 0.2 | 1 | P.DIS | DIS | |
East et al., 2018(s1) | 198 | 10 | 44 | USA | Muslim | 100 SO | comm. | P | C | good | 0.67 | 1 | C.DIS | DEP | |
Ellis et al., 2008 | 129 | 15 | 38 | USA | Muslim | 100 SO | comm. | P | C | fair | 0.51 | 2 | P.DIS | PTSD, DEP | |
Every & Perry, 2014 | 49 | 30 | 43 | AU | Muslim | NR | comm. | P | C | poor | -0.16 | 2 | G.DIS, P.DIS | SE | |
Feddes et al., 2015 | 46 | 17 | 22 | NL | Muslim | 85 MA, 11 TR, 1 SR, 1 PK | comb. | P | I | poor | -0.18 | 2 | G.DIS, P.DIS | SE | |
Goforth et al., 2014 | 128 | 16 | 45 | USA | Muslim | 100 AB | clinic | P | C | fair | 0.31 | 1 | C.DIS | DIS | |
Hashemi et al., 2019 | 275 | 30 | 48 | AU | Muslim | 100 MDE | comm. | P | C | fair | 0.23 | 1 | P.DIS | W.B | Y |
Hassouneh & Kulwicki, 2007 | 30 | 38 | 100 | USA | Muslim | 33 PAL, 30 IQ, 23 LB, 7 SY, 3 YE, 3 AA | comm. | P | C | poor | 0.31 | 6 | P.DIS | ANX, DEP | |
Hakim et al., 2018(s1) | 228 | 42 | 47 | USA | Muslim | 100 MDE/NRAFR | comm. | P | C | fair | 0.05 | 1 | C.DIS | W.B | Y |
Hakim et al., 2018(s2) | 422 | 42 | 49 | USA | Muslim | 100 AB | comm. | P | C | fair | 0.05 | 1 | C.DIS | W.B | Y |
Hernandez, 2012 | 127 | 12 | 63 | USA | Muslim | 62 AB, 16 EA, 13 OTH, 4 AFR, 3 AA, 1 H, 2 W | school | T | C | fair | -0.03 | 1 | P.DIS | SE | Y |
Hosseini et al., 2017 | 182 | 37 | 46 | AU | Iranian | 100 IR | comm. | P | C | fair | 0.15 | 1 | P.DIS | DEP | |
Husain, 2008 | 131 | 15 | 54 | USA | Muslim | 100 SO | org. | T | C | fair | 0.04 | 1 | P.DIS | SE | Y |
Ikizler & Szymanski, 2018 | 58 | 35 | 60 | USA | Muslim | 100 MDE/AB | comm. | P | C | fair | 0.37 | 1 | P.DIS | DIS | |
Jasperse et al., 2012 | 153 | 28 | 100 | NZ | Muslim | 60 A, 24 MDE, 7 AFR, 7 WEST, 3 PASF | comm. | P | C | fair | -0.06 | 2 | P.DIS | W.B, DIS | Y |
Jibeen, 2011 | 308 | 36 | 47 | CA | Pakistani | 100 PK | comm. | P | C | fair | 0.48 | 2 | P.DIS | W.B, DIS | |
Thijs et al., 2018 | 707 | 10 | 57 | NL | Muslim | 43 MA, 28.4 TR, 1.4 NL, 23.5 OTH | school | P | C | good | 0.25 | 2 | G.DIS, P.DIS | SE | Y |
Matheson et al., 2008/Jorden et al., 2009 | 169 | 29 | 64 | CA | Muslim | 100 SO | comm. | P | C | fair | 0.38 | 2 | P.DIS | PTSD, DEP | |
Kheirieh, 2019 | 91 | 30 | 78 | USA | Muslim | 100 MDE | comm. | T | C | poor | -0.06 | 1 | G.DIS | DIS | |
Kira et al., 2010/2012(s2) | 501 | 36 | 45 | USA | Muslim | 100 IQ | comm. | P | C | fair | 0.34 | 2 | C.DIS | DIS, PTSD | |
Knauss et al., 2015 | 46 | NR | NR | DE | Turkish | 100 TR | school | P | C | poor | 0.73 | 2 | C.DIS | DIS, SOM | |
Kunst et al., 2013(s1) | 167 | 24 | 58 | DE | Muslim | 100 DE-AB | comm. | P | C | good | 0.25 | 2 | G.DIS, P.DIS | DIS | |
Kunst et al., 2013(s2) | 184 | 24 | 57 | DE | Muslim | 100 DE-TR | comm. | P | C | good | 0.24 | 2 | G.DIS, P.DIS | DIS | |
Kunst et al., 2013(s3) | 205 | 24 | 43 | UK | Muslim | 100 UK-PK | comm. | P | C | good | 0.29 | 2 | G.DIS, P.DIS | DIS | |
Kunst et al., 2013(s4) | 262 | 25 | 47 | DE | Muslim | 100 DE-TR | comm. | P | C | good | 0.16 | 1 | G.DIS | DIS | |
Kunst et al., 2013(s5) | 277 | 25 | 50 | FR | Muslim | 100 FR-NRAFR | comm. | P | C | good | 0.22 | 1 | G.DIS | DIS | |
Kunst et al., 2013(s6) | 249 | 25 | 42 | UK | Muslim | 100 UK-PK | comm. | P | C | good | 0.23 | 1 | G.DIS | DIS | |
Lindencrona et al., 2008 | 115 | 34 | 38 | SE | Middle Eastern | 100 MDE | org. | P | C | poor | 0.29 | 2 | P.DIS | DIS, PTSD | |
Lindert et al., 2008 | 232 | 38 | 54 | NL | Iranian | 100 IR | comm. | P | C | fair | 0.37 | 1 | P.DIS | W.B | |
Lowe et al., 2019 | 141 | 21 | 74 | USA | Muslim | NR | school | P | C | fair | 0.34 | 2 | P.DIS | ANX, DEP | Y |
Mewes et al., 2015 | 214 | 33 | 63 | DE | Turkish | 100 TR | comb. | P | C | fair | 0.28 | 2 | P.DIS | DIS | Y |
Musa, 2015 | 202 | 25 | 76 | USA | Muslim | 38 MDE, 37 SAS, 11 W, 5 AA, 4 A, 4 M | comm. | T | C | fair | 0.5 | 2 | P.DIS | ANX, DEP | Y |
Musa, 2018 | 138 | 26 | 69 | USA | Muslim | 43 SAS, 35 MDE, 7 W, 2 AA, 5 M, 3 EA | comm. | T | C | fair | 0.39 | 2 | G.DIS, P.DIS | INT | Y |
Oppedal, 2011(s1) | 90 | 16 | 50 | NO | Somali | 100 SO | comb. | P | C | fair | 0.09 | 1 | P.DIS | INT | |
Oppedal, 2011(s3) | 103 | 16 | 50 | NO | Turkish | 100 TR | comb. | P | C | fair | 0.22 | 1 | P.DIS | INT | |
Oppedal, 2011(s2) | 74 | 12 | 48 | NO | Somali | 100 SO | comb. | P | C | fair | 0.4 | 1 | P.DIS | INT | |
Oppedal, 2011(s4) | 96 | 12 | 48 | NO | Turkish | 100 TR | comb. | P | C | fair | 0.34 | 1 | P.DIS | INT | |
Rippy & Newman, 2006 | 152 | 34 | 39 | USA | Muslim | 44 AB, 22 EA, 18 AA, 16 W | comm. | P | C | fair | 0.08 | 1 | C.DIS | ANX | |
Schaafsma, 2013 | 243 | 35 | 45 | NL | Moroccan | 57 MA, 43 TR | comm. | P | C | good | 0.21 | 1 | P.DIS | W.B | |
Schaafsma, 2011 | 320 | 30 | 52 | NL | Turkish | 52 TR, 48 MA | comm. | P | C | good | 0.13 | 2 | P.DIS | W.B | Y |
Sedighdeilami, 2003 | 205 | 36 | 50 | CA | Iranian | 100 IR | comm. | T | C | fair | 0.25 | 3 | C.DIS | EXT, W.B, INT | |
Shah, 2018 | 123 | NR | 71 | USA | Muslim | NR | comb. | T | C | fair | 0.17 | 1 | P.DIS | SE | Y |
Shawahin, 2016 | 272 | 33 | 61 | USA | Muslim | 10.3 W, 23.2 AA, .4 H, .4 A, 26.8 SAS, 26.8 MDE, .7 NAA, 11.4 OTH | comm. | T | C | good | 0.24 | 8 | G.DIS, P.DIS | INT, DEP, INTP, SOM | |
Stevens et al., 2005a/2005b+Stevens & Thijs, 2018 | 361 | 14 | 52 | NL | Muslim | 100 MA | comm. | P | C | fair | 0.13 | 15 | G.DIS, P.DIS | EXT, INT, SE | Y |
Stuart & Ward, 2018 | 155 | 20 | 70 | NZ | Muslim | 52 A 27 MDE 21 AFR | comm. | P | C | fair | 0.24 | 2 | P.DIS | DEP, W.B | Y |
Tolaymat, 2014 | 217 | 32 | 100 | USA | Muslim | 36 A, 31 AB/MDE ,12 W, 9 AA, 4.6 M, 6 OTH, 1 H | comm. | T | C | good | 0.25 | 1 | C.DIS | INT | |
van de Beek et al., 2017 | 267 | 24 | 87 | NL | Moroccan | 100 MA | comm. | P | C | fair | 0.25 | 2 | P.DIS | PSYC, DEP | |
Vedder et al., 2016 | 303 | 15 | 55 | NL | Muslim | 34 MA 46 MA | school | P | C | fair | 0.27 | 1 | P.DIS | INTP | |
Verkuyten & Nekuee, 1999 | 67 | 32 | 40 | NL | Iranian | 100 IR | org. | P | C | poor | 0.28 | 4 | P.DIS | INT, W.B, SE | Y |
Vedder et al., 2017 | 369 | 15 | 52 | NL | Muslim | 54.5 TR 29 MA | school | P | C | fair | 0.23 | 1 | P.DIS | INTP | |
Verkuyten, 1998(s1) | 92 | 14 | 55 | NL | Turkish | 100 TR | school | P | C | fair | 0.24 | 2 | G.DIS, P.DIS | SE | |
Verkuyten, 1998(s2) | 78 | 14 | 55 | NL | Moroccan | 100 MA | school | P | C | fair | 0.21 | 2 | G.DIS, P.DIS | SE | |
Verkuyten & Thijs, 2004(s1) | 161 | 15 | 49 | NL | Turkish | 100 TR | school | P | C | fair | 0.17 | 1 | P.DIS | SE | |
Verkuyten & Thijs, 2004(s2) | 112 | 12 | 53 | NL | Turkish | 100 TR | school | P | C | fair | 0.08 | 1 | C.DIS | SE | |
Wadsworth et al., 2007 | 206 | 29 | 30 | UK | Bangladeshi | 100 BD | comm. | P | C | fair | 0.14 | 4 | P.DIS | DEP, INTP, SOM, ANX | |
Wilson, 2017 | 43 | 23 | 45 | USA | Middle Eastern | 100 MDE | comm. | T | C | poor | 0.17 | 2 | P.DIS | SE, DEP | Y |
Abuelezam & El-Sayed, 2018 | 354 | 40 | 47 | USA | Arab American | 100 AB | comm. | P | C | poor | 0 | 2 | P.DIS | DEP, DIS | |
Gorst-Unsworth & Goldenberg, 1998 | 84 | 39 | 0 | UK | Muslim | 100 IQ (45 KUR) | clinic | P | C | poor | 0.32 | 1 | P.DIS | DEP | |
Ikram et al., 2015 | 834 | 40 | 53 | NL | Turkish | 100 TR | comm. | P | C | good | 0.16 | 2 | P.DIS | DEP | |
Sheridan, 2006 | 222 | 22 | 53 | UK | Muslim | 61.7 PK, 12.9 IN, 8.1 BD, 7.7 W, 4.8 OTHAS, 3.8 AB, .5 M .5 AA | comm. | P | C | poor | 0.13 | 1 | P.DIS | DIS | |
Lecerof et al., 2016 | 587 | 35 | 53 | SE | Iraqi | 100 IQ | comm. | P | C | fair | 0.27 | 1 | P.DIS | DIS | |
Tinghog et al., 2010(s1) | 259 | 42 | 47 | SE | Iraqi | 100 IQ | comm. | P | C | good | 0.11 | 1 | P.DIS | INT | |
Tinghog et al., 2010(s2) | 250 | 42 | 44 | SE | Iranian | 100 IR | comm. | P | C | good | 0.05 | 1 | P.DIS | INT | |
Tinghog et al., 2017 | 1215 | 34 | 37 | SE | Syrian | 100 SY | comm. | P | C | fair | 0.2 | 3 | P.DIS | INT, PTSD, W.B | |
von Haumeder et al., 2019 | 127 | 32 | 34 | DE | Syrian | 100 SY | org. | P | C | fair | 0.43 | 1 | P.DIS | PTSD |
Sample | N | Age | %F | Ctry | Muslim type | %Ethnicity | Sample Type | Pub | Design | Quality | r | # corr /sample | Stress Types | MH Types | Identity |
Adam & Ward, 2016 | 167 | 32 | 65 | NZ | Muslim | 32.9 IN/SAS, 17.4 EA, 16.8 MDE, 16.2 W, 9 AFR 4.8 PASF | comm. | P | C | fair | 0.32 | 2 | P.DIS | W.B, DIS | Y |
Ahmed, 2006 | 240 | 16 | 48 | USA | Muslim | 48.3 LB, 19.2 IQ, 15.8 YE, 16.7 OTHA | school | T | C | good | 0.27 | 4 | C.DIS | EXT, INT | Y |
Aichberger et al., 2015 | 205 | 37 | 100 | DE | Turkish | 100 TR | comm. | P | C | fair | 0.34 | 1 | P.DIS | DIS | |
Alemi et al., 2017 | 133 | 26 | 64 | USA | Afghan | 100 AF | comm. | P | C | fair | 0.36 | 1 | P.DIS | DEP | Y |
Alemi & Stempel, 2018 | 259 | 49 | 51 | USA | Afghan | 100 AF | comm. | P | C | fair | 0.27 | 1 | P.DIS | DIS | Y |
Ameline et al., 2019 | 88 | 32 | 53 | FR | Muslim | 40 TR/MA/DZ, 60 born FR | comm. | P | C | poor | -0.17 | 1 | G.DIS | SE | Y |
Brenick et al., 2018 | 297 | 12 | 49 | DE | Turkish | 100 TR | school | P | C | fair | 0.29 | 2 | P.DIS | DEP, W.B | |
Cakir & Guneri, 2011 | 248 | 34 | 100 | UK | Turkish | 100 TR | comm. | P | C | poor | 0.2 | 1 | P.DIS | DIS | |
East et al., 2018(s1) | 198 | 10 | 44 | USA | Muslim | 100 SO | comm. | P | C | good | 0.67 | 1 | C.DIS | DEP | |
Ellis et al., 2008 | 129 | 15 | 38 | USA | Muslim | 100 SO | comm. | P | C | fair | 0.51 | 2 | P.DIS | PTSD, DEP | |
Every & Perry, 2014 | 49 | 30 | 43 | AU | Muslim | NR | comm. | P | C | poor | -0.16 | 2 | G.DIS, P.DIS | SE | |
Feddes et al., 2015 | 46 | 17 | 22 | NL | Muslim | 85 MA, 11 TR, 1 SR, 1 PK | comb. | P | I | poor | -0.18 | 2 | G.DIS, P.DIS | SE | |
Goforth et al., 2014 | 128 | 16 | 45 | USA | Muslim | 100 AB | clinic | P | C | fair | 0.31 | 1 | C.DIS | DIS | |
Hashemi et al., 2019 | 275 | 30 | 48 | AU | Muslim | 100 MDE | comm. | P | C | fair | 0.23 | 1 | P.DIS | W.B | Y |
Hassouneh & Kulwicki, 2007 | 30 | 38 | 100 | USA | Muslim | 33 PAL, 30 IQ, 23 LB, 7 SY, 3 YE, 3 AA | comm. | P | C | poor | 0.31 | 6 | P.DIS | ANX, DEP | |
Hakim et al., 2018(s1) | 228 | 42 | 47 | USA | Muslim | 100 MDE/NRAFR | comm. | P | C | fair | 0.05 | 1 | C.DIS | W.B | Y |
Hakim et al., 2018(s2) | 422 | 42 | 49 | USA | Muslim | 100 AB | comm. | P | C | fair | 0.05 | 1 | C.DIS | W.B | Y |
Hernandez, 2012 | 127 | 12 | 63 | USA | Muslim | 62 AB, 16 EA, 13 OTH, 4 AFR, 3 AA, 1 H, 2 W | school | T | C | fair | -0.03 | 1 | P.DIS | SE | Y |
Hosseini et al., 2017 | 182 | 37 | 46 | AU | Iranian | 100 IR | comm. | P | C | fair | 0.15 | 1 | P.DIS | DEP | |
Husain, 2008 | 131 | 15 | 54 | USA | Muslim | 100 SO | org. | T | C | fair | 0.04 | 1 | P.DIS | SE | Y |
Ikizler & Szymanski, 2018 | 58 | 35 | 60 | USA | Muslim | 100 MDE/AB | comm. | P | C | fair | 0.37 | 1 | P.DIS | DIS | |
Jasperse et al., 2012 | 153 | 28 | 100 | NZ | Muslim | 60 A, 24 MDE, 7 AFR, 7 WEST, 3 PASF | comm. | P | C | fair | -0.06 | 2 | P.DIS | W.B, DIS | Y |
Jibeen, 2011 | 308 | 36 | 47 | CA | Pakistani | 100 PK | comm. | P | C | fair | 0.48 | 2 | P.DIS | W.B, DIS | |
Thijs et al., 2018 | 707 | 10 | 57 | NL | Muslim | 43 MA, 28.4 TR, 1.4 NL, 23.5 OTH | school | P | C | good | 0.25 | 2 | G.DIS, P.DIS | SE | Y |
Matheson et al., 2008/Jorden et al., 2009 | 169 | 29 | 64 | CA | Muslim | 100 SO | comm. | P | C | fair | 0.38 | 2 | P.DIS | PTSD, DEP | |
Kheirieh, 2019 | 91 | 30 | 78 | USA | Muslim | 100 MDE | comm. | T | C | poor | -0.06 | 1 | G.DIS | DIS | |
Kira et al., 2010/2012(s2) | 501 | 36 | 45 | USA | Muslim | 100 IQ | comm. | P | C | fair | 0.34 | 2 | C.DIS | DIS, PTSD | |
Knauss et al., 2015 | 46 | NR | NR | DE | Turkish | 100 TR | school | P | C | poor | 0.73 | 2 | C.DIS | DIS, SOM | |
Kunst et al., 2013(s1) | 167 | 24 | 58 | DE | Muslim | 100 DE-AB | comm. | P | C | good | 0.25 | 2 | G.DIS, P.DIS | DIS | |
Kunst et al., 2013(s2) | 184 | 24 | 57 | DE | Muslim | 100 DE-TR | comm. | P | C | good | 0.24 | 2 | G.DIS, P.DIS | DIS | |
Kunst et al., 2013(s3) | 205 | 24 | 43 | UK | Muslim | 100 UK-PK | comm. | P | C | good | 0.29 | 2 | G.DIS, P.DIS | DIS | |
Kunst et al., 2013(s4) | 262 | 25 | 47 | DE | Muslim | 100 DE-TR | comm. | P | C | good | 0.16 | 1 | G.DIS | DIS | |
Kunst et al., 2013(s5) | 277 | 25 | 50 | FR | Muslim | 100 FR-NRAFR | comm. | P | C | good | 0.22 | 1 | G.DIS | DIS | |
Kunst et al., 2013(s6) | 249 | 25 | 42 | UK | Muslim | 100 UK-PK | comm. | P | C | good | 0.23 | 1 | G.DIS | DIS | |
Lindencrona et al., 2008 | 115 | 34 | 38 | SE | Middle Eastern | 100 MDE | org. | P | C | poor | 0.29 | 2 | P.DIS | DIS, PTSD | |
Lindert et al., 2008 | 232 | 38 | 54 | NL | Iranian | 100 IR | comm. | P | C | fair | 0.37 | 1 | P.DIS | W.B | |
Lowe et al., 2019 | 141 | 21 | 74 | USA | Muslim | NR | school | P | C | fair | 0.34 | 2 | P.DIS | ANX, DEP | Y |
Mewes et al., 2015 | 214 | 33 | 63 | DE | Turkish | 100 TR | comb. | P | C | fair | 0.28 | 2 | P.DIS | DIS | Y |
Musa, 2015 | 202 | 25 | 76 | USA | Muslim | 38 MDE, 37 SAS, 11 W, 5 AA, 4 A, 4 M | comm. | T | C | fair | 0.5 | 2 | P.DIS | ANX, DEP | Y |
Musa, 2018 | 138 | 26 | 69 | USA | Muslim | 43 SAS, 35 MDE, 7 W, 2 AA, 5 M, 3 EA | comm. | T | C | fair | 0.39 | 2 | G.DIS, P.DIS | INT | Y |
Oppedal, 2011(s1) | 90 | 16 | 50 | NO | Somali | 100 SO | comb. | P | C | fair | 0.09 | 1 | P.DIS | INT | |
Oppedal, 2011(s3) | 103 | 16 | 50 | NO | Turkish | 100 TR | comb. | P | C | fair | 0.22 | 1 | P.DIS | INT | |
Oppedal, 2011(s2) | 74 | 12 | 48 | NO | Somali | 100 SO | comb. | P | C | fair | 0.4 | 1 | P.DIS | INT | |
Oppedal, 2011(s4) | 96 | 12 | 48 | NO | Turkish | 100 TR | comb. | P | C | fair | 0.34 | 1 | P.DIS | INT | |
Rippy & Newman, 2006 | 152 | 34 | 39 | USA | Muslim | 44 AB, 22 EA, 18 AA, 16 W | comm. | P | C | fair | 0.08 | 1 | C.DIS | ANX | |
Schaafsma, 2013 | 243 | 35 | 45 | NL | Moroccan | 57 MA, 43 TR | comm. | P | C | good | 0.21 | 1 | P.DIS | W.B | |
Schaafsma, 2011 | 320 | 30 | 52 | NL | Turkish | 52 TR, 48 MA | comm. | P | C | good | 0.13 | 2 | P.DIS | W.B | Y |
Sedighdeilami, 2003 | 205 | 36 | 50 | CA | Iranian | 100 IR | comm. | T | C | fair | 0.25 | 3 | C.DIS | EXT, W.B, INT | |
Shah, 2018 | 123 | NR | 71 | USA | Muslim | NR | comb. | T | C | fair | 0.17 | 1 | P.DIS | SE | Y |
Shawahin, 2016 | 272 | 33 | 61 | USA | Muslim | 10.3 W, 23.2 AA, .4 H, .4 A, 26.8 SAS, 26.8 MDE, .7 NAA, 11.4 OTH | comm. | T | C | good | 0.24 | 8 | G.DIS, P.DIS | INT, DEP, INTP, SOM | |
Stevens et al., 2005a/2005b+Stevens & Thijs, 2018 | 361 | 14 | 52 | NL | Muslim | 100 MA | comm. | P | C | fair | 0.13 | 15 | G.DIS, P.DIS | EXT, INT, SE | Y |
Stuart & Ward, 2018 | 155 | 20 | 70 | NZ | Muslim | 52 A 27 MDE 21 AFR | comm. | P | C | fair | 0.24 | 2 | P.DIS | DEP, W.B | Y |
Tolaymat, 2014 | 217 | 32 | 100 | USA | Muslim | 36 A, 31 AB/MDE ,12 W, 9 AA, 4.6 M, 6 OTH, 1 H | comm. | T | C | good | 0.25 | 1 | C.DIS | INT | |
van de Beek et al., 2017 | 267 | 24 | 87 | NL | Moroccan | 100 MA | comm. | P | C | fair | 0.25 | 2 | P.DIS | PSYC, DEP | |
Vedder et al., 2016 | 303 | 15 | 55 | NL | Muslim | 34 MA 46 MA | school | P | C | fair | 0.27 | 1 | P.DIS | INTP | |
Verkuyten & Nekuee, 1999 | 67 | 32 | 40 | NL | Iranian | 100 IR | org. | P | C | poor | 0.28 | 4 | P.DIS | INT, W.B, SE | Y |
Vedder et al., 2017 | 369 | 15 | 52 | NL | Muslim | 54.5 TR 29 MA | school | P | C | fair | 0.23 | 1 | P.DIS | INTP | |
Verkuyten, 1998(s1) | 92 | 14 | 55 | NL | Turkish | 100 TR | school | P | C | fair | 0.24 | 2 | G.DIS, P.DIS | SE | |
Verkuyten, 1998(s2) | 78 | 14 | 55 | NL | Moroccan | 100 MA | school | P | C | fair | 0.21 | 2 | G.DIS, P.DIS | SE | |
Verkuyten & Thijs, 2004(s1) | 161 | 15 | 49 | NL | Turkish | 100 TR | school | P | C | fair | 0.17 | 1 | P.DIS | SE | |
Verkuyten & Thijs, 2004(s2) | 112 | 12 | 53 | NL | Turkish | 100 TR | school | P | C | fair | 0.08 | 1 | C.DIS | SE | |
Wadsworth et al., 2007 | 206 | 29 | 30 | UK | Bangladeshi | 100 BD | comm. | P | C | fair | 0.14 | 4 | P.DIS | DEP, INTP, SOM, ANX | |
Wilson, 2017 | 43 | 23 | 45 | USA | Middle Eastern | 100 MDE | comm. | T | C | poor | 0.17 | 2 | P.DIS | SE, DEP | Y |
Abuelezam & El-Sayed, 2018 | 354 | 40 | 47 | USA | Arab American | 100 AB | comm. | P | C | poor | 0 | 2 | P.DIS | DEP, DIS | |
Gorst-Unsworth & Goldenberg, 1998 | 84 | 39 | 0 | UK | Muslim | 100 IQ (45 KUR) | clinic | P | C | poor | 0.32 | 1 | P.DIS | DEP | |
Ikram et al., 2015 | 834 | 40 | 53 | NL | Turkish | 100 TR | comm. | P | C | good | 0.16 | 2 | P.DIS | DEP | |
Sheridan, 2006 | 222 | 22 | 53 | UK | Muslim | 61.7 PK, 12.9 IN, 8.1 BD, 7.7 W, 4.8 OTHAS, 3.8 AB, .5 M .5 AA | comm. | P | C | poor | 0.13 | 1 | P.DIS | DIS | |
Lecerof et al., 2016 | 587 | 35 | 53 | SE | Iraqi | 100 IQ | comm. | P | C | fair | 0.27 | 1 | P.DIS | DIS | |
Tinghog et al., 2010(s1) | 259 | 42 | 47 | SE | Iraqi | 100 IQ | comm. | P | C | good | 0.11 | 1 | P.DIS | INT | |
Tinghog et al., 2010(s2) | 250 | 42 | 44 | SE | Iranian | 100 IR | comm. | P | C | good | 0.05 | 1 | P.DIS | INT | |
Tinghog et al., 2017 | 1215 | 34 | 37 | SE | Syrian | 100 SY | comm. | P | C | fair | 0.2 | 3 | P.DIS | INT, PTSD, W.B | |
von Haumeder et al., 2019 | 127 | 32 | 34 | DE | Syrian | 100 SY | org. | P | C | fair | 0.43 | 1 | P.DIS | PTSD |
NR = Not reported; Country: NZ = New Zealand; SE = Sweden; FR = France; DE = Germany; UK = United Kingdom; CA = Canada; NL = Netherlands; AU = Australia; DK = Denmark; NO = Norway; CH = Switzerland; BE = Belgium; Ethnicity: A = Asian or Asian American; AA = African American or Black; AAA= African Arab; AB = Arab or Arab American; AF = Afghanistan; AFR = Africa; AL = Albania; BD = Bangladesh; DE = Germany; DZ = Algeria; EA = East Asian; EG = Egypt; FR = France; H = Hispanic or Latinx; IN = India; IQ = Iraq; IR = Iran; KUR = Kurdish; LB = Lebanon; M = Multiethnic; MA = Morocco; MDE = Middle East; NAA = Native American; NL = Netherlands; NRAFR = North Africa; OTH = Other; OTHA= Other Arabic; OTHAS = Other Asian; OTHNA= Other non-Arabic; PAL = Palestine; PASF = Pasifika; PK = Pakistan; SAS = South Asian; SO = Somalia; SR = Suriname; SY = Syria; TR = Turkey; UK = United Kingdom; W = White; WEST = Western; YE = Yemen; Study-Deisgn: C = crosssectional; L = longitudinal; I = intervention; Publication Status: P = published; T = thesis; Stressor: P.DIS = personal discrimination; G.DIS = group discrimination; C.DIS = combination discrimination; Mental health: INT = internalizing; PTSD = PTSD; W.B = wellbeing; DIS = psychological distress; EXT = externalizing; INTP = interpersonal; DEP = depression; ANX = anxiety; SE = self-esteem; SOM = somatization; PSYC = psychotic; Sample Type: comm. = community; comb. = combination; org. = organization
Sample characteristics | Details |
k | 72 |
kc | 138 |
N (range) | 244 (30-1215) |
Age (M(sd)) | 26.13 (9.73) |
% Female (sd) | 54.62 (17.77) |
Data from published sources (%) | 85 |
Country | |
- Australia | k=3 |
- Canada | k=3 |
- France | k=2 |
- Germany | k=8 |
- Netherlands | k=15 |
- New Zealand | k=3 |
- Norway | k=4 |
- Sweden | k=5 |
- United Kingdom | k=6 |
- USA | k=23 |
Majority Ethnicity | |
- Asian/South Asian | k=10 (14.5%) |
- Middle Eastern | k=42 (60.9%) |
- Middle Eastern/North African | k=1 (1.4%) |
- Middle Eastern/North African/South Asian | k=2 (2.9%) |
- North African | k=8 (11.6%) |
- Somali | k=6 (8.7%) |
Majority Ethnicity (detailed) | |
- Afghani | k=2 |
- Arab | k=5 |
- Asian/South Asian | k=5 |
- Bangladeshi | k=1 |
- British-Pakistani | k=2 |
- French-Maghrebi | k=1 |
- German-Arab | k=1 |
- German-Turkish | k=2 |
- Iranian | k=5 |
- Iraqi | k=4 |
- Lebanese | k=1 |
- Middle Eastern | k=5 |
- Middle Eastern/Arab American | k=1 |
- Middle Eastern/North African | k=1 |
- Middle Eastern/South Asian | k=1 |
- Moroccan | k=7 |
- Pakistani | k=2 |
- Palestinian | k=1 |
- Somali | k=6 |
- Syrian | k=2 |
- Turkish | k=13 |
- Turkish/Moroccan/Algerian | k=1 |
Sample characteristics | Details |
k | 72 |
kc | 138 |
N (range) | 244 (30-1215) |
Age (M(sd)) | 26.13 (9.73) |
% Female (sd) | 54.62 (17.77) |
Data from published sources (%) | 85 |
Country | |
- Australia | k=3 |
- Canada | k=3 |
- France | k=2 |
- Germany | k=8 |
- Netherlands | k=15 |
- New Zealand | k=3 |
- Norway | k=4 |
- Sweden | k=5 |
- United Kingdom | k=6 |
- USA | k=23 |
Majority Ethnicity | |
- Asian/South Asian | k=10 (14.5%) |
- Middle Eastern | k=42 (60.9%) |
- Middle Eastern/North African | k=1 (1.4%) |
- Middle Eastern/North African/South Asian | k=2 (2.9%) |
- North African | k=8 (11.6%) |
- Somali | k=6 (8.7%) |
Majority Ethnicity (detailed) | |
- Afghani | k=2 |
- Arab | k=5 |
- Asian/South Asian | k=5 |
- Bangladeshi | k=1 |
- British-Pakistani | k=2 |
- French-Maghrebi | k=1 |
- German-Arab | k=1 |
- German-Turkish | k=2 |
- Iranian | k=5 |
- Iraqi | k=4 |
- Lebanese | k=1 |
- Middle Eastern | k=5 |
- Middle Eastern/Arab American | k=1 |
- Middle Eastern/North African | k=1 |
- Middle Eastern/South Asian | k=1 |
- Moroccan | k=7 |
- Pakistani | k=2 |
- Palestinian | k=1 |
- Somali | k=6 |
- Syrian | k=2 |
- Turkish | k=13 |
- Turkish/Moroccan/Algerian | k=1 |
k = number of samples; kc = number of outcomes or correlations.
RQ1: Bivariate effect-size estimate between discrimination and negative mental health
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, kc = 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.
Measure | Random-Effects Model | RVE ⍴ = .8 | RVE ⍴ = 0 | RVE ⍴ = .2 | RVE ⍴ = .4 | RVE ⍴ = .6 | RVE ⍴ = 1 |
r | 0.234 | 0.215 | 0.215 | 0.215 | 0.215 | 0.215 | 0.215 |
95% CI | [0.197, 0.271] | [0.18, 0.249] | [0.18, 0.249] | [0.18, 0.249] | [0.18, 0.249] | [0.18, 0.249] | [0.18, 0.249] |
z-value | 11.945 | ||||||
t-value | 11.960 | 11.961 | 11.961 | 11.961 | 11.96 | 11.96 | |
tau-squared | 0.022 | 0.018 | 0.018 | 0.018 | 0.018 | 0.018 | 0.018 |
Q-stat | 338.295 | ||||||
I-squared | 82.773 | 79.510 | 79.439 | 79.457 | 79.474 | 79.492 | 79.528 |
Index of Dispersion | [-0.055, 0.487] |
Measure | Random-Effects Model | RVE ⍴ = .8 | RVE ⍴ = 0 | RVE ⍴ = .2 | RVE ⍴ = .4 | RVE ⍴ = .6 | RVE ⍴ = 1 |
r | 0.234 | 0.215 | 0.215 | 0.215 | 0.215 | 0.215 | 0.215 |
95% CI | [0.197, 0.271] | [0.18, 0.249] | [0.18, 0.249] | [0.18, 0.249] | [0.18, 0.249] | [0.18, 0.249] | [0.18, 0.249] |
z-value | 11.945 | ||||||
t-value | 11.960 | 11.961 | 11.961 | 11.961 | 11.96 | 11.96 | |
tau-squared | 0.022 | 0.018 | 0.018 | 0.018 | 0.018 | 0.018 | 0.018 |
Q-stat | 338.295 | ||||||
I-squared | 82.773 | 79.510 | 79.439 | 79.457 | 79.474 | 79.492 | 79.528 |
Index of Dispersion | [-0.055, 0.487] |
Robust variance estimation (RVE) model at ⍴ = .8 is the weight used in all RVE models. 95% CI: 95% confidence intervals around the weighted mean correlation.
Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) | M(sd)/% | Range |
Publication Status | 72 | 0.664 | 0.189(1) | |||||||
Published (reference) | 61 | 0.242/0.238 | 0.022 | 11.103 | ||||||
Unpublished | 11 | -0.025/0.214 | 0.057 | -0.435 | 0.664 | |||||
Type of Sample Population | 72 | 0.870 | 0.279(2) | |||||||
Community (reference) | 47 | 0.231/0.227 | 0.025 | 9.385 | ||||||
Combination | 13 | 0.015/0.242 | 0.056 | 0.274 | 0.784 | |||||
School | 12 | 0.028/0.253 | 0.056 | 0.496 | 0.620 | |||||
Correlation Type | 72 | 0.273 | 1.204(2) | |||||||
Pearson r (reference) | 63 | 0.247/0.242 | 0.021 | 11.551 | ||||||
non-Pearson r | 9 | -0.063/0.182 | 0.058 | -1.097 | 0.273 | |||||
Discrimination Measure: | ||||||||||
Discrimination-level | 138 | |||||||||
Personal discrimination (reference) | 92 | 0.233/0.229 | 0.018 | 12.918 | 50.281 | |||||
Combination discrimination | 18 | 0.033/0.259 | 0.075 | 0.670 | 0.435 | 14.423 | ||||
Group discrimination | 28 | -0.137/0.096 | 0.046 | 0.008 | -2.969 | 17.765 | ||||
Respondent- Discrimination Measure | 138 | |||||||||
Self-report (reference) | 112 | 0.201/0.198 | 0.016 | 12.578 | 59.343 | |||||
Interview/Combination | 26 | 0.128/0.318 | 0.087 | 0.166 | 1.473 | 11.996 | ||||
Measure Quality Moderators | ||||||||||
Mean Discrimination Measure Items | 67 | 0.006 | 0.003 | 2.458 | 0.014 | 9.56 (7.96) | (1-36) | |||
Mean Discrimination Measure Alpha | 55 | 0.158 | 0.299 | 0.528 | 0.597 | 0.84 (0.08) | (0.59-0.96) | |||
Mental Health Measure: | ||||||||||
Mental Health Outcome Direction | 138 | |||||||||
Negative outcome (reference) | 103 | 0.256/0.251 | 0.021 | 12.398 | 50.191 | |||||
Positive outcome | 35 | -0.135/-0.121 | 0.035 | <.001 | -3.87 | 38.218 | ||||
Respondent- Mental Health Measure | 138 | |||||||||
Self-report (reference) | 108 | 0.203/0.201 | 0.016 | 12.515 | 59.073 | |||||
Interview/Combination | 30 | 0.102/0.296 | 0.084 | 0.244 | 1.219 | 13.586 | ||||
Measure Quality Moderators | ||||||||||
Mean Mental Health Measure Items | 67 | 0.002 | 0.001 | 1.678 | 0.093 | 16.62 (19.82) | (1-115) | |||
Mean Mental Health Measure Alpha | 58 | -0.168 | 0.27 | -0.624 | 0.533 | 0.84 (0.08) | (0.6-0.96) | |||
Mean Mental Health Response Period | 38 | 0 | 0.001 | 0.045 | 0.964 | 49.16 (64.62) | (3-180) |
Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) | M(sd)/% | Range |
Publication Status | 72 | 0.664 | 0.189(1) | |||||||
Published (reference) | 61 | 0.242/0.238 | 0.022 | 11.103 | ||||||
Unpublished | 11 | -0.025/0.214 | 0.057 | -0.435 | 0.664 | |||||
Type of Sample Population | 72 | 0.870 | 0.279(2) | |||||||
Community (reference) | 47 | 0.231/0.227 | 0.025 | 9.385 | ||||||
Combination | 13 | 0.015/0.242 | 0.056 | 0.274 | 0.784 | |||||
School | 12 | 0.028/0.253 | 0.056 | 0.496 | 0.620 | |||||
Correlation Type | 72 | 0.273 | 1.204(2) | |||||||
Pearson r (reference) | 63 | 0.247/0.242 | 0.021 | 11.551 | ||||||
non-Pearson r | 9 | -0.063/0.182 | 0.058 | -1.097 | 0.273 | |||||
Discrimination Measure: | ||||||||||
Discrimination-level | 138 | |||||||||
Personal discrimination (reference) | 92 | 0.233/0.229 | 0.018 | 12.918 | 50.281 | |||||
Combination discrimination | 18 | 0.033/0.259 | 0.075 | 0.670 | 0.435 | 14.423 | ||||
Group discrimination | 28 | -0.137/0.096 | 0.046 | 0.008 | -2.969 | 17.765 | ||||
Respondent- Discrimination Measure | 138 | |||||||||
Self-report (reference) | 112 | 0.201/0.198 | 0.016 | 12.578 | 59.343 | |||||
Interview/Combination | 26 | 0.128/0.318 | 0.087 | 0.166 | 1.473 | 11.996 | ||||
Measure Quality Moderators | ||||||||||
Mean Discrimination Measure Items | 67 | 0.006 | 0.003 | 2.458 | 0.014 | 9.56 (7.96) | (1-36) | |||
Mean Discrimination Measure Alpha | 55 | 0.158 | 0.299 | 0.528 | 0.597 | 0.84 (0.08) | (0.59-0.96) | |||
Mental Health Measure: | ||||||||||
Mental Health Outcome Direction | 138 | |||||||||
Negative outcome (reference) | 103 | 0.256/0.251 | 0.021 | 12.398 | 50.191 | |||||
Positive outcome | 35 | -0.135/-0.121 | 0.035 | <.001 | -3.87 | 38.218 | ||||
Respondent- Mental Health Measure | 138 | |||||||||
Self-report (reference) | 108 | 0.203/0.201 | 0.016 | 12.515 | 59.073 | |||||
Interview/Combination | 30 | 0.102/0.296 | 0.084 | 0.244 | 1.219 | 13.586 | ||||
Measure Quality Moderators | ||||||||||
Mean Mental Health Measure Items | 67 | 0.002 | 0.001 | 1.678 | 0.093 | 16.62 (19.82) | (1-115) | |||
Mean Mental Health Measure Alpha | 58 | -0.168 | 0.27 | -0.624 | 0.533 | 0.84 (0.08) | (0.6-0.96) | |||
Mean Mental Health Response Period | 38 | 0 | 0.001 | 0.045 | 0.964 | 49.16 (64.62) | (3-180) |
k = number of samples; kc = number of outcomes or correlations. Note: Random-effects models and mixed-effects models display z statistics, while robust variation estimation (RVE) models display t statistics. Z values are based upon the population mean while t statistics are based upon the sample mean. As random-effects models and mixed-effects model estimate the population means, those models use z values.
RQ2: Moderating effect of study-related, demographic-related, and identity-related variables
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, Qbetween (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 (kc = 112), followed by an interview (kc = 21), and a combination of self-report and a parent interview (kc = 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 (kc11 = 92), followed by group discrimination (kc = 28), and then a combination measure of both personal and group discrimination (kc = 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.
k | kc | r | 95% CI | t.value | p | Tau.squared | I.squared | |
Group Discrimination | 16 | 28 | 0.096 | [0.008, 0.183] | 2.332 | 0.035 | 0.015 | 74.762 |
Personal Discrimination | 56 | 92 | 0.229 | [0.196, 0.262] | 13.387 | <.001 | 0.011 | 71.197 |
Combination Discrimination | 11 | 18 | 0.268 | [0.106, 0.416] | 3.64 | 0.005 | 0.053 | 91.809 |
k | kc | r | 95% CI | t.value | p | Tau.squared | I.squared | |
Group Discrimination | 16 | 28 | 0.096 | [0.008, 0.183] | 2.332 | 0.035 | 0.015 | 74.762 |
Personal Discrimination | 56 | 92 | 0.229 | [0.196, 0.262] | 13.387 | <.001 | 0.011 | 71.197 |
Combination Discrimination | 11 | 18 | 0.268 | [0.106, 0.416] | 3.64 | 0.005 | 0.053 | 91.809 |
k = number of samples; kc = number of outcomes or correlations.
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 (kc = 103), and only 25% of all correlations measured positive mental health (kc = 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 (kc = 6), depression (kc = 22), externalizing (kc = 9), internalizing (kc = 23), interpersonal (kc = 5), psychological distress (kc = 26), psychotic (kc = 1), PTSD (kc = 6), self-esteem (kc = 20), somatization (kc = 4), and well-being (kc = 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; kc = 22), followed by psychological distress (r = .23; kc = 26), interpersonal outcomes (r = .22; kc = 5), externalizing outcomes (r = .22; kc = 9), anxiety (r = .22; kc = 6), and well-being (r = -.17; kc = 16). These results suggest that discrimination has a stronger relationship with negative, psychopathological mental-health outcomes.
r | 95% CI | kc | k | t.value | p | Tau.squared | I.squared | |
Negative Mental Health | 0.254 | [0.215, 0.292] | 103 | 55 | 12.712 | <.001 | 0.017 | 79.596 |
Positive Mental Health | -0.124 | [-0.18, -0.067] | 35 | 26 | -4.468 | <.001 | 0.016 | 75.161 |
Depression | 0.298 | [0.211, 0.381] | 22 | 17 | 6.977 | <.001 | 0.028 | 85.072 |
Anxiety | 0.217 | [0.006, 0.409] | 6 | 5 | 2.942 | 0.046 | 0.021 | 73.747 |
Externalizing | 0.218 | [0.143, 0.29] | 9 | 3 | 12.287 | 0.007 | 0.02 | 82.67 |
Internalizing | 0.204 | [0.136, 0.271] | 23 | 14 | 6.364 | <.001 | 0.013 | 75.355 |
Interpersonal | 0.224 | [0.129, 0.315] | 5 | 4 | 7.457 | 0.005 | 0.005 | 56.235 |
Psychological distress | 0.231 | [0.165, 0.295] | 26 | 22 | 7.098 | <.001 | 0.019 | 80.477 |
Self-esteem | -0.074 | [-0.152, 0.005] | 20 | 14 | -2.034 | 0.064 | 0.018 | 71.006 |
Somatization | -0.299 | [-0.857, 0.581] | 4 | 3 | -1.381 | 0.303 | 0.073 | 91.216 |
Wellbeing | -0.168 | [0.094, 0.24] | 16 | 14 | 4.908 | <.001 | 0.011 | 75.262 |
r | 95% CI | kc | k | t.value | p | Tau.squared | I.squared | |
Negative Mental Health | 0.254 | [0.215, 0.292] | 103 | 55 | 12.712 | <.001 | 0.017 | 79.596 |
Positive Mental Health | -0.124 | [-0.18, -0.067] | 35 | 26 | -4.468 | <.001 | 0.016 | 75.161 |
Depression | 0.298 | [0.211, 0.381] | 22 | 17 | 6.977 | <.001 | 0.028 | 85.072 |
Anxiety | 0.217 | [0.006, 0.409] | 6 | 5 | 2.942 | 0.046 | 0.021 | 73.747 |
Externalizing | 0.218 | [0.143, 0.29] | 9 | 3 | 12.287 | 0.007 | 0.02 | 82.67 |
Internalizing | 0.204 | [0.136, 0.271] | 23 | 14 | 6.364 | <.001 | 0.013 | 75.355 |
Interpersonal | 0.224 | [0.129, 0.315] | 5 | 4 | 7.457 | 0.005 | 0.005 | 56.235 |
Psychological distress | 0.231 | [0.165, 0.295] | 26 | 22 | 7.098 | <.001 | 0.019 | 80.477 |
Self-esteem | -0.074 | [-0.152, 0.005] | 20 | 14 | -2.034 | 0.064 | 0.018 | 71.006 |
Somatization | -0.299 | [-0.857, 0.581] | 4 | 3 | -1.381 | 0.303 | 0.073 | 91.216 |
Wellbeing | -0.168 | [0.094, 0.24] | 16 | 14 | 4.908 | <.001 | 0.011 | 75.262 |
k = number of samples; kc = number of outcomes or correlations. Note: Estimates are exploratory when kc/k < 10.
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 (kc = 108), followed by an interview (kc = 21), interview with parent (kc = 6), and teacher-report (kc = 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, Qbetween (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/).
Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) | M(sd)/% | Range |
Mean Age | 70 | -0.002 | 0.002 | -1.235 | 0.217 | 26.13 (9.73) | (10.02-48.75) | |||
% Female | 71 | 0.0001 | 0.001 | 0.092 | 0.927 | 54.62 (17.77) | (0-100) | |||
% Bachelors+ | 33 | -0.0001 | 0.001 | -0.096 | 0.924 | 40.04 (24.40) | (0-87.4) | |||
% Born in Western country | 33 | 0.0001 | 0.001 | 0.053 | 0.958 | 59.67 (25.85) | (0-100) | |||
% First generation | 34 | -0.001 | 0.001 | -0.73 | 0.465 | 60.57 (29.71) | (11-100) | |||
% Refugee | 9 | 0.002 | 0.001 | 2.463 | 0.014 | 76.33 (36.10) | (20-100) | |||
Continent | 72 | 0.191 | 3.315(2) | |||||||
Europe (reference) | 40 | 0.228/0.224 | 0.026 | 8.634 | ||||||
North America | 26 | 0.048/0.27 | 0.042 | 1.139 | 0.255 | |||||
Australia/New Zealand | 6 | -0.085/0.143 | 0.075 | -1.135 | 0.257 | |||||
Ethnicity (80+) | 93 | |||||||||
Middle Eastern (reference) | 53 | 0.226/0.222 | 0.021 | 10.614 | 32.721 | |||||
Asian/South Asian | 9 | 0.038/0.258 | 0.073 | 0.630 | 0.523 | 3.792 | ||||
North African | 22 | -0.085/0.14 | 0.055 | 0.189 | -1.531 | 4.773 | ||||
Somali | 8 | 0.152/0.361 | 0.125 | 0.265 | 1.218 | 6.564 | ||||
Middle Eastern/North African | 1 | -0.176/0.05 | 0.021 | <.001 | -8.261 | 32.721 | ||||
Ethnicity | 133 | |||||||||
Middle Eastern (reference) | 70 | 0.22/0.216 | 0.02 | 10.939 | 39.264 | |||||
Asian/South Asian | 19 | 0.005/0.221 | 0.048 | 0.913 | 0.112 | 13.839 | ||||
North African | 26 | -0.038/0.18 | 0.038 | 0.352 | -0.979 | 9.472 | ||||
Somali | 8 | 0.158/0.361 | 0.125 | 0.250 | 1.268 | 6.278 | ||||
Other | 10 | -0.177/0.043 | 0.096 | 0.192 | -1.846 | 2.24 | ||||
Muslim sample type | 72 | 0.571 | 0.322(2) | |||||||
Majority Muslim country (reference) | 33 | 0.251/0.246 | 0.03 | 8.445 | ||||||
Muslim | 39 | -0.023/0.224 | 0.04 | -0.567 | 0.571 |
Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) | M(sd)/% | Range |
Mean Age | 70 | -0.002 | 0.002 | -1.235 | 0.217 | 26.13 (9.73) | (10.02-48.75) | |||
% Female | 71 | 0.0001 | 0.001 | 0.092 | 0.927 | 54.62 (17.77) | (0-100) | |||
% Bachelors+ | 33 | -0.0001 | 0.001 | -0.096 | 0.924 | 40.04 (24.40) | (0-87.4) | |||
% Born in Western country | 33 | 0.0001 | 0.001 | 0.053 | 0.958 | 59.67 (25.85) | (0-100) | |||
% First generation | 34 | -0.001 | 0.001 | -0.73 | 0.465 | 60.57 (29.71) | (11-100) | |||
% Refugee | 9 | 0.002 | 0.001 | 2.463 | 0.014 | 76.33 (36.10) | (20-100) | |||
Continent | 72 | 0.191 | 3.315(2) | |||||||
Europe (reference) | 40 | 0.228/0.224 | 0.026 | 8.634 | ||||||
North America | 26 | 0.048/0.27 | 0.042 | 1.139 | 0.255 | |||||
Australia/New Zealand | 6 | -0.085/0.143 | 0.075 | -1.135 | 0.257 | |||||
Ethnicity (80+) | 93 | |||||||||
Middle Eastern (reference) | 53 | 0.226/0.222 | 0.021 | 10.614 | 32.721 | |||||
Asian/South Asian | 9 | 0.038/0.258 | 0.073 | 0.630 | 0.523 | 3.792 | ||||
North African | 22 | -0.085/0.14 | 0.055 | 0.189 | -1.531 | 4.773 | ||||
Somali | 8 | 0.152/0.361 | 0.125 | 0.265 | 1.218 | 6.564 | ||||
Middle Eastern/North African | 1 | -0.176/0.05 | 0.021 | <.001 | -8.261 | 32.721 | ||||
Ethnicity | 133 | |||||||||
Middle Eastern (reference) | 70 | 0.22/0.216 | 0.02 | 10.939 | 39.264 | |||||
Asian/South Asian | 19 | 0.005/0.221 | 0.048 | 0.913 | 0.112 | 13.839 | ||||
North African | 26 | -0.038/0.18 | 0.038 | 0.352 | -0.979 | 9.472 | ||||
Somali | 8 | 0.158/0.361 | 0.125 | 0.250 | 1.268 | 6.278 | ||||
Other | 10 | -0.177/0.043 | 0.096 | 0.192 | -1.846 | 2.24 | ||||
Muslim sample type | 72 | 0.571 | 0.322(2) | |||||||
Majority Muslim country (reference) | 33 | 0.251/0.246 | 0.03 | 8.445 | ||||||
Muslim | 39 | -0.023/0.224 | 0.04 | -0.567 | 0.571 |
k = number of samples; kc = number of outcomes or correlations. Note: Random-effects models and mixed-effects models display z statistics, while robust variation estimation (RVE) models display t statistics. Z values are based upon the population mean while t statistics are based upon the sample mean. As random-effects models and mixed-effects model estimate the population means, those models use z values.
Country | k | r | 95% CI | z | p | Tau squared | I squared |
Australia | 3 | 0.113 | [-0.082-0.299] | 1.137 | 0.255 | 0.021 | 74.809 |
Canada | 3 | 0.377 | [0.236-0.502] | 4.976 | <.001 | 0.014 | 75.666 |
Germany | 8 | 0.332 | [0.218-0.436] | 5.493 | <.001 | 0.025 | 81.773 |
France | 2 | 0.038 | [-0.333-0.398] | 0.193 | 0.847 | 0.069 | 89.981 |
Netherlands | 15 | 0.206 | [0.164-0.247] | 9.446 | <.001 | 0.003 | 42.024 |
Norway | 4 | 0.262 | [0.128-0.387] | 3.768 | <.001 | 0.009 | 43.547 |
New Zealand | 3 | 0.17 | [-0.058-0.381] | 1.467 | 0.142 | 0.035 | 84.296 |
Sweden | 5 | 0.184 | [0.102-0.264] | 4.335 | <.001 | 0.006 | 70.784 |
United Kingdom | 6 | 0.207 | [0.152-0.26] | 7.251 | <.001 | <.001 | 0.004 |
USA | 23 | 0.254 | [0.169-0.335] | 5.69 | <.001 | 0.040 | 87.933 |
Country | k | r | 95% CI | z | p | Tau squared | I squared |
Australia | 3 | 0.113 | [-0.082-0.299] | 1.137 | 0.255 | 0.021 | 74.809 |
Canada | 3 | 0.377 | [0.236-0.502] | 4.976 | <.001 | 0.014 | 75.666 |
Germany | 8 | 0.332 | [0.218-0.436] | 5.493 | <.001 | 0.025 | 81.773 |
France | 2 | 0.038 | [-0.333-0.398] | 0.193 | 0.847 | 0.069 | 89.981 |
Netherlands | 15 | 0.206 | [0.164-0.247] | 9.446 | <.001 | 0.003 | 42.024 |
Norway | 4 | 0.262 | [0.128-0.387] | 3.768 | <.001 | 0.009 | 43.547 |
New Zealand | 3 | 0.17 | [-0.058-0.381] | 1.467 | 0.142 | 0.035 | 84.296 |
Sweden | 5 | 0.184 | [0.102-0.264] | 4.335 | <.001 | 0.006 | 70.784 |
United Kingdom | 6 | 0.207 | [0.152-0.26] | 7.251 | <.001 | <.001 | 0.004 |
USA | 23 | 0.254 | [0.169-0.335] | 5.69 | <.001 | 0.040 | 87.933 |
k/kc | b | SE | z | p | M(sd)/% | Range | |
Ethnic Identity | 14 | -0.001 | 0 | -0.19 | 0.85 | 68.64 (17.18) | (39.8-85) |
Religious Identity | 8 | 0.0004 | 0.01 | 0.03 | 0.97 | 78.13 (6.95) | (65-85) |
k/kc | b | SE | z | p | M(sd)/% | Range | |
Ethnic Identity | 14 | -0.001 | 0 | -0.19 | 0.85 | 68.64 (17.18) | (39.8-85) |
Religious Identity | 8 | 0.0004 | 0.01 | 0.03 | 0.97 | 78.13 (6.95) | (65-85) |
k = number of samples; kc = number of outcomes or correlations. Note: Random-effects models and mixed-effects models display z statistics, while robust variation estimation (RVE) models display t statistics. Z values are based upon the population mean while t statistics are based upon the sample mean. As random-effects models and mixed-effects model estimate the population means, those models use z values
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.
Summary of meta-analytic results of the association between discrimination and mental health
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.
Life stressors
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.
Sample Characteristics
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
RQ3: Bivariate effect-size estimate between life stressors and mental health
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.
Sample | N | Age | %F | Ctry | Muslim type | %Ethnicity | Sample Type | Pub | Design | Quality | r | # corr /sample | Stress Types | MH Types | Idenity |
Albdour, 2017 | 150 | 14 | 30 | USA | Muslim | 45.3 YE, 21.3 LB, 8 IQ, 4 SY, 6 JO, 12 PAL, 3.3. OTH | org. | T | C | fair | 0.5 | 2 | BUL | DIS, SOM | |
Aldhalimi, 2015 | 48 | 37 | 0 | USA | Iraqi | NR | comb. | T | C | poor | 0.31 | 1 | WR.TR | PTSD | |
Alemi & Stempel, 2018 | 259 | 49 | 51 | USA | Afghan | 100 AF | comm. | P | C | fair | 0.49 | 1 | WR.TR | DIS | Y |
Alhasanat et al., 2017 | 50 | 30 | 100 | USA | Muslim | 48 YE, 2 SA, 2 QA, 4 PAL, 30 LB, 2 JO, 10 IQ, 2 EG | org. | P | C | poor | 0.3 | 1 | G.LIFE | DEP | |
Almqvist & Broberg, 1999 | 39 | 8 | 26 | SE | Iranian | 100 IR | comm. | P | L | poor | 0.33 | 4 | WR.TR, BUL | W.B, DIS | |
Altawian, 2016 | 246 | 36 | 100 | USA | Muslim | 100 SA | school | T | C | poor | 0.84 | 4 | PH.AB, V.AB | PTSD, DEP | |
Arnetz et al., 2012 | 350 | 46 | 44 | USA | Iraqi | 100 IQ | comm. | P | C | fair | 0.42 | 4 | WR.TR | SOM, PTSD, DEP | |
Arnetz et al., 2014 | 298 | 33 | 46 | USA | Iraqi | 100 IQ | comb. | P | C | fair | 0.23 | 4 | WR.TR, PH.AB, G.LIFE | PTSD, DEP | |
Arnetz et al., 2013 | 128 | 41 | 59 | USA | Iraqi | 58.6 IQ, 41.4 MDE (LB, JO, EG, and YE) | comm. | P | C | poor | 0.37 | 2 | AB | PTSD, DIS | |
Bentley et al., 2012 | 74 | 39 | 26 | USA | Somali | 100 SO | comm. | P | C | fair | 0.57 | 8 | WR.TR, MIG.TR | SOM, PTSD, DEP, ANX | |
Bronstein et al., 2012+Bronstein & Montgomery, 2013 | 222 | 16 | 0 | UK | Afghani | 100 AF | org. | P | C | poor, fair | 0.48 | 2 | TR | PTSD, DIS | |
Dadouch, 2019 | 19 | 35 | 45 | USA | Muslim | 100 SY | org. | T | C | poor | -0.12 | 2 | TR | SOM, PTSD | |
East et al., 2018(s1) | 198 | 10 | 44 | USA | Muslim | 100 SO | comm. | P | C | good | 0.5 | 2 | BUL, TR | DEP | |
East et al., 2018(s2) | 198 | 39 | 100 | USA | Muslim | 100 SO | comm. | P | C | good | 0.25 | 2 | WR.TR | DEP | |
Elsayed et al., 2019 | 103 | 8 | 53 | CA | Syrian | 100 SY | comm. | P | C | fair | -0.07 | 2 | G.LIFE | EXT, INT | |
Ellis et al., 2015 | 79 | 21 | 0 | USA | Somali | 100 SO | comm. | P | C | fair | 0.41 | 2 | G.LIFE, PH.AB | PTSD | |
Ellis et al., 2008 | 129 | 15 | 38 | USA | Muslim | 100 SO | comm. | P | C | fair | 0.49 | 2 | WR.TR | PTSD, DEP | |
Ellis et al., 2013 | 30 | 13 | 37 | USA | Somali | 100 SO | school | P | L | good | 0.38 | 2 | WR.TR | PTSD, DEP | |
Groen et al., 2019 | 57 | 36 | 35 | NL | Iraqi | 51 IQ, 49 AF | clinic | P | C | fair | 0.27 | 2 | WR.TR | INT, PTSD | |
Hassouneh & Kulwicki, 2007 | 30 | 38 | 100 | USA | Muslim | 33 PAL, 30 IQ, 23 LB, 7 SY, 3 YE, 3 AA | comm. | P | C | poor | -0.02 | 3 | WR.TR | ANX, DEP | |
Hellemans et al., 2015 | 392 | 34 | 50 | BE | Muslim | 100 TK | comm. | P | C | good | 0.14 | 2 | PH.AB, V.AB | W.B | |
Jakobsen et al., 2017 | 138 | 16 | 0 | NO | Afghan | 74 AF, 23 SO, 2 IR, 1 DZ | org. | P | L | fair | 0.07 | 1 | TR | DIS | |
Jobson et al., 2016 | 43 | 42 | 40 | UK | Iranian | 100 IR | comm. | P | C | poor | -0.14 | 2 | TR | PTSD, DEP | |
Matheson et al., 2008/Jorden et al., 2009 | 156 | 28 | 64 | CA | Muslim | 100 SO | comm. | P | C | fair | 0.29 | 12 | WR.TR, TR, PH.AB, G.LIFE | PTSD, DEP | |
Jorgenson, 2017 | 80 | 30 | 62 | USA | Somali | 100 SO | org. | T | C | fair | 0.38 | 2 | WR.TR | PTSD, INT | Y |
Kia-Keating & Ellis, 2007 | 76 | 16 | 46 | USA | Somali | 100 SO | comm. | P | C | fair | 0.53 | 2 | WR.TR | PTSD, DEP | |
Kira et al., 2010/2012(s2) | 501 | 36 | 45 | USA | Muslim | 100 IQi | comm. | P | C | fair | 0.27 | 1 | TR | DIS | |
Kira et al., 2012(s1) | 286 | 40 | 62 | USA | Iraqi | 60.1 IQ, 19.8 LB, 10.1 YE, 6.9 OTHA, and 6 OTHNA | clinic | P | C | fair | 0.16 | 1 | WR.TR | DIS | |
Koch et al., 2019 | 74 | 20 | 0 | DE | Afghan | 100 AF | comb. | P | C | fair | 0.25 | 3 | TR | ANX, PTSD, DEP | |
Kulwicki et al., 2015 | 312 | 36 | 100 | USA | Muslim | 43.6 LB, 38.5 IQ, 8.3 YE, 5.8 PAL, 3.8 OTH | comm. | P | C | good | 0.44 | 1 | PH.AB | DEP | |
Areba, 2014 | 156 | 21 | 75 | USA | Muslim | 100 SO | school | T | C | fair | 0.38 | 3 | G.LIFE | ANX, DEP, W.B | Y |
Lincoln et al., 2016 | 135 | 15 | 38 | USA | Somali | 100 SO | school | P | C | fair | 0.49 | 2 | WR.TR | PTSD, DEP | |
Lindencrona et al., 2008 | 115 | 34 | 38 | SE | Middle Eastern | 100 MDE | org. | P | C | poor | 0.33 | 2 | G.LIFE | PTSD, DIS | |
Lembcke et al., 2020 | 31 | 3 | 55 | DE | Syrian/Iraqi | 100 SY or IQ | comm. | P | C | poor | -0.12 | 1 | WR.TR | PTSD | |
Mghir et al., 1995 | 38 | 18 | 45 | USA | Afghani | 100 AF | comm. | P | C | poor | 0.48 | 1 | WR.TR | DIS | |
Muller et al., 2012/2018 | 334 | 43 | 60 | DE | Turkish | 100 TK | clinic | P | C | poor, fair | 0.23 | 2 | MIG.TR | W.B, PTSD | |
Mohlen et al., 2005 | 10 | 13 | 40 | DE | Kozovo | 90 KOS-AL, 10 RO | org. | P | I | poor | 0.89 | 4 | WR.TR | PTSD, W.B, ANX, DEP | |
Montgomery & Foldspang, 2006 | 311 | 8 | 49 | DK | Iraqi | 54 IQ, 7 LB, 10 IR, 4 SY, 24 PAL, .003 TR | org. | P | C | fair | 0.4 | 5 | WR.TR | DIS | |
Nilsson et al., 2008 | 62 | 38 | 100 | USA | Muslim | 100 SO | org. | P | C | poor | 0.76 | 2 | PH.AB, V.AB | DIS | |
Oberoi, 2014 | 177 | 17 | 67 | USA | Muslim | 58 A, 23 W, 17 OTH, 3 AA, 1 H | school | T | C | fair | 0.58 | 1 | G.LIFE | DIS | Y |
Lien et al., 2010(s1) | 602 | 42 | 40 | NO | Iranian | 100 IR | comm. | P | C | fair | 0.19 | 2 | WR.TR | DIS | |
Lien et al., 2010(s2) | 448 | 45 | 44 | NO | Pakistani | 100 PK | comm. | P | C | fair | 0.21 | 2 | WR.TR | DIS | |
Lien et al., 2010(s3) | 426 | 42 | 46 | NO | Turkish | 100 TR | comm. | P | C | fair | 0.13 | 2 | WR.TR | DIS | |
Perera et al., 2013 | 437 | 38 | 49 | USA | Somali | 58 SO, 42 ORO | org. | P | L | good | 0.44 | 3 | WR.TR, G.LIFE | PTSD | |
Mghir & Raskin, 1999 | 15 | 17 | 29 | USA | Pashtun | 100 PAS | comm. | P | C | poor | 0.69 | 1 | WR.TR | PTSD | |
Stevens et al., 2005a/2005b+Stevens & Thijs, 2018 | 376 | 14 | 52 | NL | Muslim | 100 MA | comm. | P | C | fair | 0.19 | 4 | G.LIFE | EXT, INT | Y |
Turner et al., 2003 | 842 | 38 | 53 | UK | Kosovan | 100 KOS | org. | P | C | fair | 0.38 | 4 | WR.TR | DIS, PTSD, DEP, ANX | |
Woods, 2004 | 220 | 36 | 41 | UK | Somali | 100 SO | comm. | T | C | fair | 0.4 | 3 | WR.TR | INT, EXT, PTSD | |
Zoellner et al., 2018 | 39 | NR | 36 | USA | Muslim | 100 SO | org. | P | C | poor | 0.4 | 1 | WR.TR | PTSD | |
Abuelezam & El-Sayed, 2018 | 354 | 40 | 47 | USA | Arab American | 100 AB | comm. | P | C | poor | 0.01 | 6 | PH.AB, V.AB, SE.TR | DEP, DIS | |
Ahmed et al., 2017 | 12 | 27 | 100 | CA | Syrian | 100 SY | org. | P | C | poor | 0.37 | 1 | MIG.TR | DEP | |
Almqvist & Brandell-Forsberg, 1997 | 50 | 6 | 28 | SE | Muslim | 100 IR | comm. | P | L | poor | 0.2 | 1 | TR | PTSD | |
Chaudhry, 2007 | 202 | 37 | 100 | UK | Muslim | 100 PK | comm. | T | L | fair | 0.4 | 1 | G.LIFE | DEP | |
Chung et al., 2018 | 564 | 35 | 32 | SE | Syrian | 100 SY | comm. | P | C | fair | 0.24 | 4 | WR.TR, G.LIFE | PTSD | |
Daud et al., 2008 | 67 | 43 | 0 | SE | Iraqi | 54 IQ, 46 RB | clinic | P | C | fair | 0.69 | 1 | WR.TR | SOM | |
Fox & Johnson-Agbakwu, 2020 | 879 | 31 | 100 | USA | Somali | 68 SO, 26 SOB | comm. | P | C | fair | 0.22 | 1 | TR | DIS | |
Ghazinour et al., 2004 | 100 | 37 | 34 | SE | Iranian | 100 IR | comb. | P | C | poor | 0.25 | 1 | WR.TR | DEP | |
Gilgen et al., 2002 | 98 | 36 | 60 | CH | Turkey | 63 TR, 37 BA | clinic | P | C | poor | 0.35 | 1 | WR.TR | DIS | |
Gorst-Unsworth & Goldenberg, 1998 | 84 | 39 | 0 | UK | Muslim | 100 IQ (45 KUR) | clinic | P | C | poor, fair | 0.41 | 2 | SE.TR, WR.TR | PTSD, DEP | |
Hondius et al., 2000 | 156 | 26 | 34 | NL | Turkish/Iranian | 100 TR/IR | org. | P | C | poor | 0.14 | 2 | WR.TR | DEP, ANX | |
Husain, 2010 | 237 | 28 | 100 | UK | Muslim | 100 PK | clinic | T | L | good | 0.28 | 1 | G.LIFE | DEP | |
Husain et al., 1997 | 77 | 36 | 73 | UK | Pakistani | 100 PK | clinic | P | C | fair | 0.39 | 1 | G.LIFE | DEP | |
Mirza, 2014 | 57 | 15 | 60 | USA | Muslim | 50.9 SAS, 38.7 M, 10.5 AAA, 35.1 MDE | comm. | T | C | fair | 0.43 | 1 | TR | EXT | Y |
Momartin et al., 2003 | 86 | 47 | 61 | AU | Muslim | 100 BA | comb. | P | C | fair | 0.42 | 1 | WR.TR | PTSD | |
Tinghog et al., 2017 | 1215 | 34 | 37 | SE | Syrian | 100 SY | comm. | P | C | fair | 0.18 | 3 | WR.TR | INT, PTSD, W.B | |
Tomlins, 2016 | 74 | 44 | 100 | USA | Iranian | 43 IR, 32 IQ, 24 AF | org. | T | C | fair | 0.93 | 2 | SE.TR | PTSD, DEP | |
Westermeyer et al., 2010 | 622 | 37 | 48 | USA | Somalian | 100 SO | comm. | P | C | fair | 0.32 | 1 | WR.TR | SOM | |
Willard et al., 2014 | 497 | 42 | 44 | USA | Iraqi | 100 IQ | org. | P | C | poor | 0.4 | 1 | WR.TR | SOM | |
Ai et al., 2002 | 129 | 35 | 45 | USA | Muslim | 98 AL/KOS | org. | P | C | fair | 0.39 | 1 | WR.TR | PTSD |
Sample | N | Age | %F | Ctry | Muslim type | %Ethnicity | Sample Type | Pub | Design | Quality | r | # corr /sample | Stress Types | MH Types | Idenity |
Albdour, 2017 | 150 | 14 | 30 | USA | Muslim | 45.3 YE, 21.3 LB, 8 IQ, 4 SY, 6 JO, 12 PAL, 3.3. OTH | org. | T | C | fair | 0.5 | 2 | BUL | DIS, SOM | |
Aldhalimi, 2015 | 48 | 37 | 0 | USA | Iraqi | NR | comb. | T | C | poor | 0.31 | 1 | WR.TR | PTSD | |
Alemi & Stempel, 2018 | 259 | 49 | 51 | USA | Afghan | 100 AF | comm. | P | C | fair | 0.49 | 1 | WR.TR | DIS | Y |
Alhasanat et al., 2017 | 50 | 30 | 100 | USA | Muslim | 48 YE, 2 SA, 2 QA, 4 PAL, 30 LB, 2 JO, 10 IQ, 2 EG | org. | P | C | poor | 0.3 | 1 | G.LIFE | DEP | |
Almqvist & Broberg, 1999 | 39 | 8 | 26 | SE | Iranian | 100 IR | comm. | P | L | poor | 0.33 | 4 | WR.TR, BUL | W.B, DIS | |
Altawian, 2016 | 246 | 36 | 100 | USA | Muslim | 100 SA | school | T | C | poor | 0.84 | 4 | PH.AB, V.AB | PTSD, DEP | |
Arnetz et al., 2012 | 350 | 46 | 44 | USA | Iraqi | 100 IQ | comm. | P | C | fair | 0.42 | 4 | WR.TR | SOM, PTSD, DEP | |
Arnetz et al., 2014 | 298 | 33 | 46 | USA | Iraqi | 100 IQ | comb. | P | C | fair | 0.23 | 4 | WR.TR, PH.AB, G.LIFE | PTSD, DEP | |
Arnetz et al., 2013 | 128 | 41 | 59 | USA | Iraqi | 58.6 IQ, 41.4 MDE (LB, JO, EG, and YE) | comm. | P | C | poor | 0.37 | 2 | AB | PTSD, DIS | |
Bentley et al., 2012 | 74 | 39 | 26 | USA | Somali | 100 SO | comm. | P | C | fair | 0.57 | 8 | WR.TR, MIG.TR | SOM, PTSD, DEP, ANX | |
Bronstein et al., 2012+Bronstein & Montgomery, 2013 | 222 | 16 | 0 | UK | Afghani | 100 AF | org. | P | C | poor, fair | 0.48 | 2 | TR | PTSD, DIS | |
Dadouch, 2019 | 19 | 35 | 45 | USA | Muslim | 100 SY | org. | T | C | poor | -0.12 | 2 | TR | SOM, PTSD | |
East et al., 2018(s1) | 198 | 10 | 44 | USA | Muslim | 100 SO | comm. | P | C | good | 0.5 | 2 | BUL, TR | DEP | |
East et al., 2018(s2) | 198 | 39 | 100 | USA | Muslim | 100 SO | comm. | P | C | good | 0.25 | 2 | WR.TR | DEP | |
Elsayed et al., 2019 | 103 | 8 | 53 | CA | Syrian | 100 SY | comm. | P | C | fair | -0.07 | 2 | G.LIFE | EXT, INT | |
Ellis et al., 2015 | 79 | 21 | 0 | USA | Somali | 100 SO | comm. | P | C | fair | 0.41 | 2 | G.LIFE, PH.AB | PTSD | |
Ellis et al., 2008 | 129 | 15 | 38 | USA | Muslim | 100 SO | comm. | P | C | fair | 0.49 | 2 | WR.TR | PTSD, DEP | |
Ellis et al., 2013 | 30 | 13 | 37 | USA | Somali | 100 SO | school | P | L | good | 0.38 | 2 | WR.TR | PTSD, DEP | |
Groen et al., 2019 | 57 | 36 | 35 | NL | Iraqi | 51 IQ, 49 AF | clinic | P | C | fair | 0.27 | 2 | WR.TR | INT, PTSD | |
Hassouneh & Kulwicki, 2007 | 30 | 38 | 100 | USA | Muslim | 33 PAL, 30 IQ, 23 LB, 7 SY, 3 YE, 3 AA | comm. | P | C | poor | -0.02 | 3 | WR.TR | ANX, DEP | |
Hellemans et al., 2015 | 392 | 34 | 50 | BE | Muslim | 100 TK | comm. | P | C | good | 0.14 | 2 | PH.AB, V.AB | W.B | |
Jakobsen et al., 2017 | 138 | 16 | 0 | NO | Afghan | 74 AF, 23 SO, 2 IR, 1 DZ | org. | P | L | fair | 0.07 | 1 | TR | DIS | |
Jobson et al., 2016 | 43 | 42 | 40 | UK | Iranian | 100 IR | comm. | P | C | poor | -0.14 | 2 | TR | PTSD, DEP | |
Matheson et al., 2008/Jorden et al., 2009 | 156 | 28 | 64 | CA | Muslim | 100 SO | comm. | P | C | fair | 0.29 | 12 | WR.TR, TR, PH.AB, G.LIFE | PTSD, DEP | |
Jorgenson, 2017 | 80 | 30 | 62 | USA | Somali | 100 SO | org. | T | C | fair | 0.38 | 2 | WR.TR | PTSD, INT | Y |
Kia-Keating & Ellis, 2007 | 76 | 16 | 46 | USA | Somali | 100 SO | comm. | P | C | fair | 0.53 | 2 | WR.TR | PTSD, DEP | |
Kira et al., 2010/2012(s2) | 501 | 36 | 45 | USA | Muslim | 100 IQi | comm. | P | C | fair | 0.27 | 1 | TR | DIS | |
Kira et al., 2012(s1) | 286 | 40 | 62 | USA | Iraqi | 60.1 IQ, 19.8 LB, 10.1 YE, 6.9 OTHA, and 6 OTHNA | clinic | P | C | fair | 0.16 | 1 | WR.TR | DIS | |
Koch et al., 2019 | 74 | 20 | 0 | DE | Afghan | 100 AF | comb. | P | C | fair | 0.25 | 3 | TR | ANX, PTSD, DEP | |
Kulwicki et al., 2015 | 312 | 36 | 100 | USA | Muslim | 43.6 LB, 38.5 IQ, 8.3 YE, 5.8 PAL, 3.8 OTH | comm. | P | C | good | 0.44 | 1 | PH.AB | DEP | |
Areba, 2014 | 156 | 21 | 75 | USA | Muslim | 100 SO | school | T | C | fair | 0.38 | 3 | G.LIFE | ANX, DEP, W.B | Y |
Lincoln et al., 2016 | 135 | 15 | 38 | USA | Somali | 100 SO | school | P | C | fair | 0.49 | 2 | WR.TR | PTSD, DEP | |
Lindencrona et al., 2008 | 115 | 34 | 38 | SE | Middle Eastern | 100 MDE | org. | P | C | poor | 0.33 | 2 | G.LIFE | PTSD, DIS | |
Lembcke et al., 2020 | 31 | 3 | 55 | DE | Syrian/Iraqi | 100 SY or IQ | comm. | P | C | poor | -0.12 | 1 | WR.TR | PTSD | |
Mghir et al., 1995 | 38 | 18 | 45 | USA | Afghani | 100 AF | comm. | P | C | poor | 0.48 | 1 | WR.TR | DIS | |
Muller et al., 2012/2018 | 334 | 43 | 60 | DE | Turkish | 100 TK | clinic | P | C | poor, fair | 0.23 | 2 | MIG.TR | W.B, PTSD | |
Mohlen et al., 2005 | 10 | 13 | 40 | DE | Kozovo | 90 KOS-AL, 10 RO | org. | P | I | poor | 0.89 | 4 | WR.TR | PTSD, W.B, ANX, DEP | |
Montgomery & Foldspang, 2006 | 311 | 8 | 49 | DK | Iraqi | 54 IQ, 7 LB, 10 IR, 4 SY, 24 PAL, .003 TR | org. | P | C | fair | 0.4 | 5 | WR.TR | DIS | |
Nilsson et al., 2008 | 62 | 38 | 100 | USA | Muslim | 100 SO | org. | P | C | poor | 0.76 | 2 | PH.AB, V.AB | DIS | |
Oberoi, 2014 | 177 | 17 | 67 | USA | Muslim | 58 A, 23 W, 17 OTH, 3 AA, 1 H | school | T | C | fair | 0.58 | 1 | G.LIFE | DIS | Y |
Lien et al., 2010(s1) | 602 | 42 | 40 | NO | Iranian | 100 IR | comm. | P | C | fair | 0.19 | 2 | WR.TR | DIS | |
Lien et al., 2010(s2) | 448 | 45 | 44 | NO | Pakistani | 100 PK | comm. | P | C | fair | 0.21 | 2 | WR.TR | DIS | |
Lien et al., 2010(s3) | 426 | 42 | 46 | NO | Turkish | 100 TR | comm. | P | C | fair | 0.13 | 2 | WR.TR | DIS | |
Perera et al., 2013 | 437 | 38 | 49 | USA | Somali | 58 SO, 42 ORO | org. | P | L | good | 0.44 | 3 | WR.TR, G.LIFE | PTSD | |
Mghir & Raskin, 1999 | 15 | 17 | 29 | USA | Pashtun | 100 PAS | comm. | P | C | poor | 0.69 | 1 | WR.TR | PTSD | |
Stevens et al., 2005a/2005b+Stevens & Thijs, 2018 | 376 | 14 | 52 | NL | Muslim | 100 MA | comm. | P | C | fair | 0.19 | 4 | G.LIFE | EXT, INT | Y |
Turner et al., 2003 | 842 | 38 | 53 | UK | Kosovan | 100 KOS | org. | P | C | fair | 0.38 | 4 | WR.TR | DIS, PTSD, DEP, ANX | |
Woods, 2004 | 220 | 36 | 41 | UK | Somali | 100 SO | comm. | T | C | fair | 0.4 | 3 | WR.TR | INT, EXT, PTSD | |
Zoellner et al., 2018 | 39 | NR | 36 | USA | Muslim | 100 SO | org. | P | C | poor | 0.4 | 1 | WR.TR | PTSD | |
Abuelezam & El-Sayed, 2018 | 354 | 40 | 47 | USA | Arab American | 100 AB | comm. | P | C | poor | 0.01 | 6 | PH.AB, V.AB, SE.TR | DEP, DIS | |
Ahmed et al., 2017 | 12 | 27 | 100 | CA | Syrian | 100 SY | org. | P | C | poor | 0.37 | 1 | MIG.TR | DEP | |
Almqvist & Brandell-Forsberg, 1997 | 50 | 6 | 28 | SE | Muslim | 100 IR | comm. | P | L | poor | 0.2 | 1 | TR | PTSD | |
Chaudhry, 2007 | 202 | 37 | 100 | UK | Muslim | 100 PK | comm. | T | L | fair | 0.4 | 1 | G.LIFE | DEP | |
Chung et al., 2018 | 564 | 35 | 32 | SE | Syrian | 100 SY | comm. | P | C | fair | 0.24 | 4 | WR.TR, G.LIFE | PTSD | |
Daud et al., 2008 | 67 | 43 | 0 | SE | Iraqi | 54 IQ, 46 RB | clinic | P | C | fair | 0.69 | 1 | WR.TR | SOM | |
Fox & Johnson-Agbakwu, 2020 | 879 | 31 | 100 | USA | Somali | 68 SO, 26 SOB | comm. | P | C | fair | 0.22 | 1 | TR | DIS | |
Ghazinour et al., 2004 | 100 | 37 | 34 | SE | Iranian | 100 IR | comb. | P | C | poor | 0.25 | 1 | WR.TR | DEP | |
Gilgen et al., 2002 | 98 | 36 | 60 | CH | Turkey | 63 TR, 37 BA | clinic | P | C | poor | 0.35 | 1 | WR.TR | DIS | |
Gorst-Unsworth & Goldenberg, 1998 | 84 | 39 | 0 | UK | Muslim | 100 IQ (45 KUR) | clinic | P | C | poor, fair | 0.41 | 2 | SE.TR, WR.TR | PTSD, DEP | |
Hondius et al., 2000 | 156 | 26 | 34 | NL | Turkish/Iranian | 100 TR/IR | org. | P | C | poor | 0.14 | 2 | WR.TR | DEP, ANX | |
Husain, 2010 | 237 | 28 | 100 | UK | Muslim | 100 PK | clinic | T | L | good | 0.28 | 1 | G.LIFE | DEP | |
Husain et al., 1997 | 77 | 36 | 73 | UK | Pakistani | 100 PK | clinic | P | C | fair | 0.39 | 1 | G.LIFE | DEP | |
Mirza, 2014 | 57 | 15 | 60 | USA | Muslim | 50.9 SAS, 38.7 M, 10.5 AAA, 35.1 MDE | comm. | T | C | fair | 0.43 | 1 | TR | EXT | Y |
Momartin et al., 2003 | 86 | 47 | 61 | AU | Muslim | 100 BA | comb. | P | C | fair | 0.42 | 1 | WR.TR | PTSD | |
Tinghog et al., 2017 | 1215 | 34 | 37 | SE | Syrian | 100 SY | comm. | P | C | fair | 0.18 | 3 | WR.TR | INT, PTSD, W.B | |
Tomlins, 2016 | 74 | 44 | 100 | USA | Iranian | 43 IR, 32 IQ, 24 AF | org. | T | C | fair | 0.93 | 2 | SE.TR | PTSD, DEP | |
Westermeyer et al., 2010 | 622 | 37 | 48 | USA | Somalian | 100 SO | comm. | P | C | fair | 0.32 | 1 | WR.TR | SOM | |
Willard et al., 2014 | 497 | 42 | 44 | USA | Iraqi | 100 IQ | org. | P | C | poor | 0.4 | 1 | WR.TR | SOM | |
Ai et al., 2002 | 129 | 35 | 45 | USA | Muslim | 98 AL/KOS | org. | P | C | fair | 0.39 | 1 | WR.TR | PTSD |
NR = Not reported; Country: NZ = New Zealand; SE = Sweden; FR = France; DE = Germany; UK = United Kingdom; CA = Canada; NL = Netherlands; AU = Australia; DK = Denmark; NO = Norway; CH = Switzerland; BE = Belgium; Ethnicity: A = Asian or Asian American; AA = African American or Black; AAA= African Arab; AB = Arab or Arab American; AF = Afghanistan; AL = Albania; BA = Bosnia; DZ = Algeria; EG = Egypt; H = Hispanic or Latinx; IQ = Iraq; IR = Iran; JO = Jordan; KOS = Kosovo; KUR = Kurdish; LB = Lebanon; M = Multiethnic; MA = Morocco; MDE = Middle East; ORO = Oromo; OTH = Other; OTHA= Other Arabic; OTHNA= Other non-Arabic; PAL = Palestine; PAS = Pashtun; PK = Pakistan; QA = Qatar; RO = Romania; SA = Saudi Arabia; SAS = South Asian; SO = Somalia; SOB = Somali Bantu; SY = Syria; TR = Turkey; W = White; YE = Yemen; Study-Deisgn: C = crosssectional; L = longitudinal; I = intervention; Publication Status: P = published; T = thesis; Stressor: TOR = torture; WR.TR = war truama; MIG.TR = migraiton trauma; TR = trauma; G.LIFE = general life; PH.AB = phsyical abuse; V.AB = verbal abuse; BUL = bullying; AB = abuse; SE.TR = sexual trauma; Mental health: INT = internalizing; PTSD = PTSD ; W.B = wellbeing; DIS = psychological distress; EXT = externalizing; INTP = interpersonal; DEP = depression; ANX = anxiety; SE = self-esteem; SOM = somatization; PSYC = psychotic; Sample Type: comm. = community; comb. = combination; org. = organization
Sample characteristics | Details |
k | 69 |
kc | 157 |
N (range) | 236 (10-1215) |
Age (M(sd)) | 29.59 (11.82) |
% Female (sd) | 49.05 (25.11) |
Data from published sources (%) | 83 |
Country | |
- Australia | k=1 |
- Belgium | k=1 |
- Canada | k=3 |
- Denmark | k=1 |
- Germany | k=4 |
- Netherlands | k=3 |
- Norway | k=4 |
- Sweden | k=7 |
- Switzerland | k=1 |
- United Kingdom | k=8 |
- USA | k=36 |
Majority Ethnicity | |
- Asian/South Asian | k=6 (8.7%) |
- Bosnian Peninsula | k=4 (5.8%) |
- Middle Eastern | k=41 (59.4%) |
- North African | k=1 (1.4%) |
- Somali | k=17 (24.6%) |
Majority Ethnicity (detailed) | |
- Afghani | k=5 |
- Albanian/Kosovar | k=1 |
- Arab | k=1 |
- Asian/South Asian | k=2 |
- Bosnian | k=1 |
- Iranian | k=6 |
- Iraqi | k=11 |
- Kosovan | k=2 |
- Lebanese | k=1 |
- Middle Eastern | k=2 |
- Moroccan | k=1 |
- Pakistani | k=4 |
- Palestinian | k=1 |
- Pashtun | k=1 |
- Saudi Arabia | k=1 |
- Somali | k=17 |
- Syrian | k=5 |
- Syrian/Iraqi | k=1 |
- Turkey | k=1 |
- Turkish | k=3 |
- Turkish/Iranian | k=1 |
- Yemen | k=1 |
Sample characteristics | Details |
k | 69 |
kc | 157 |
N (range) | 236 (10-1215) |
Age (M(sd)) | 29.59 (11.82) |
% Female (sd) | 49.05 (25.11) |
Data from published sources (%) | 83 |
Country | |
- Australia | k=1 |
- Belgium | k=1 |
- Canada | k=3 |
- Denmark | k=1 |
- Germany | k=4 |
- Netherlands | k=3 |
- Norway | k=4 |
- Sweden | k=7 |
- Switzerland | k=1 |
- United Kingdom | k=8 |
- USA | k=36 |
Majority Ethnicity | |
- Asian/South Asian | k=6 (8.7%) |
- Bosnian Peninsula | k=4 (5.8%) |
- Middle Eastern | k=41 (59.4%) |
- North African | k=1 (1.4%) |
- Somali | k=17 (24.6%) |
Majority Ethnicity (detailed) | |
- Afghani | k=5 |
- Albanian/Kosovar | k=1 |
- Arab | k=1 |
- Asian/South Asian | k=2 |
- Bosnian | k=1 |
- Iranian | k=6 |
- Iraqi | k=11 |
- Kosovan | k=2 |
- Lebanese | k=1 |
- Middle Eastern | k=2 |
- Moroccan | k=1 |
- Pakistani | k=4 |
- Palestinian | k=1 |
- Pashtun | k=1 |
- Saudi Arabia | k=1 |
- Somali | k=17 |
- Syrian | k=5 |
- Syrian/Iraqi | k=1 |
- Turkey | k=1 |
- Turkish | k=3 |
- Turkish/Iranian | k=1 |
- Yemen | k=1 |
k = number of samples; kc = number of outcomes or correlations.
Heterogeneity. The data showed a statistically significant and large degree of between-study variability, Q(68) = 701.86, p < .001, τ2 = 0.068, I2 = 93.52% (see Table 12). The 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 life stressors and mental-health outcomes can be attributed to variances in the true effect rather than sampling error. The index of dispersion for the correlation of interest was -0.13 to 0.72; this shows that, within the population, bivariate correlations would fall between this range 95% of the time. Therefore, in some cases, there is a small, negative effect, and in others, there are worse outcomes. Given the large index of dispersion, it is important to examine potential moderators of the association between life stressors and mental-health outcomes.
Influential cases and outliers. Figure 6 shows the seven influential samples identified through examining DFFITS values, Cook’s distances, hat values, DFBETAS values, and effect sizes that are 2.5 absolute deviations below/above the median (Viechtbauer & Cheung, 2010). As a sensitivity analysis, we removed the seven influential samples and recalculated the effect-size estimates (k = 62, kc = 140). The weighted random-effects and RVE estimates were unchanged and still statistically significant with high heterogeneity (albeit, lower heterogeneity), r = .333 (95% CI [.30, .37]) and r = .303 (95% CI [ .26, .33]), respectively (see Figure 6 for estimates without influential cases, and Figures S7, https://osf.io/baesz/, & S8, https://osf.io/3uwfk/, for all outlier statistics). The leaveout analysis in R demonstrated that, when removed, each influential case either slightly attenuated or increased the omnibus effect size, but it did not alter the statistical significance of the effect, and heterogeneity remained high (see Table S4, https://osf.io/gkt8u/). Thus, findings showed that influential cases and outliers did not alter the omnibus weighted mean correlation nor account for a significant amount of heterogeneity between studies.
Sensitivity analysis for between- and within-study variance. Table 12 also summarizes the results from the RVE method using different levels of ρ. 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 13 summarizes publication status moderation results; 83% of samples were from published (k = 57) vs. unpublished studies (k = 12). We conducted meta-regressions with random-effect model using both REML (mixed-effects model) and RVE; consistent with our hypothesis, we found publication status to be a significant moderator, z = 2.902, p = .004 (mixed-effects model); t (15.1) = 2.3, p = .036 (RVE). However, the moderation result was not in the direction we hypothesized; rather, unpublished studies (k = 12) had an increased estimate by b = .243. Using a random-effects model, the effect-size estimates for published data was r = .325 (95% CI [.28, .37]); for unpublished data, it was r = .516 (95% CI [.30, .68]). Second, we created two funnel plots to visualize a) the effect size plotted against sample size, and b) the effect size plotted against standard error. Visually, both funnel plots appeared slightly skewed with slightly more studies on the right (see Figures 7 & 8). However, the symmetry of the two funnel plots were quantified by the Egger’s test, and both tests were not significant (sample size: z = -1.725, p = .085; standard error: z = 1.266, p = .205).
For an additional visual representation of publication bias, forest plots were created and sorted by publication status and sample size; both displayed visual patterns consistent with evidence of bias, with larger confidence intervals having been found with unpublished studies and smaller sample sizes (see Figure S9, https://osf.io/q9uer/, & S10, https://osf.io/rg6em). The sensitivity analysis proposed by Vevea and Wood (2005) showed the adjusted effect-size estimate was r = .329, which is slightly attenuated, but within the confidence interval of the omnibus effect size. Thus, even though the results suggest some evidence of publication bias, the effect-size estimate between life stressors and mental health seems to be robust to publication bias with the current methods used to detect and correct for publication bias.
RQ4: Moderating effect of study-related, demographic-related, and identity-related variables
We examined study-related (i.e., study-design characteristics, stress measure, and mental health measure), demographic-related, and identity-related moderators when there were at least 10 effect estimates per one covariate or per group. We use mixed-effects model, unless specified otherwise. When results differed between the mixed-effects and RVE approaches, both models are presented. Tables S5 and S6 include results from the model that is not presented (https://osf.io/gkt8u/).
Study-design characteristics. Table 13 presents study-design characteristic moderators. In terms of study design, most studies used a cross-sectional design (k = 61), seven used a longitudinal design, and one used an intervention design. We combined longitudinal and intervention design studies so that there were nearly ten samples per category, however, the analysis was still slightly underpowered. Study design did not explain the between-variance in the effect-size estimate (see Table 13). We also examined to see if the type of sample population was a moderator of the association between life stressors and mental health (see Table 13). The majority of samples were community-based samples (k = 33), followed by official organization (k = 18), clinic (k = 8), school (k = 5), and a combination of samples (k = 5). We grouped school, 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 was a significant moderator of the association between life stressors and mental health, Qbetween (2) = 6.105, p = .047, whereas community samples, r = .295, p <.001, had a significantly lower estimate compared to samples recruited from official organizations, r = .454 (z = 2.303, p = .021). Using RVE meta-regression with the nonaggregate data—which allowed for more categories—the type of sample population also was a significant moderator (see Table S5, https://osf.io/gkt8u/). Samples from schools, r = .497, p = .045 (t (4.95) = 2.66, p = .045), resulted in a stronger estimate. Samples from organizations were approaching significance, r = .378, p = .06 (t (29.46) = 1.954, p = .06), when compared to community samples, r = .263, p < .001 (t (30.01) = 8.90, p < .001; Table S5, https://osf.io/gkt8u). Although the p-values varied slightly between models, the beta estimates were nearly the same.
We also examined when correlation type was a moderator (see Table 13). The majority of samples included a Pearson correlation (k = 46); however, we did calculate a Pearson correlation for some of the samples from various statistics: chi-squared (k = 1), means (k = 2), and phi correlation (k = 13). Since three samples included a combination of correlation types, we used RVE to test when correlation type was a significant moderator. We grouped chi-squared and mean-calculated correlations (kc = 14), and correlation type was not a significant moderator (see Table 13 and Table S5, https://osf.io/gkt8u/). The evidence suggests study design and correlation type were not significant moderators, while the type of sample population explains some of the between-study variance in the omnibus estimate.
Measure | Random-Effects Model | RVE ⍴ = .8 | RVE ⍴ = 0 | RVE ⍴ = .2 | RVE ⍴ = .4 | RVE ⍴ = .6 | RVE ⍴ = 1 |
r | 0.366 | 0.32 | 0.32 | 0.32 | 0.32 | 0.32 | 0.32 |
95% CI | [0.307, 0.423] | [0.273, 0.365] | [0.273, 0.365] | [0.273, 0.365] | [0.273, 0.365] | [0.273, 0.365] | [0.273, 0.365] |
z-value | 11.303 | ||||||
t-value | 12.99 | 12.991 | 12.991 | 12.991 | 12.991 | 12.99 | |
tau-squared | 0.068 | 0.032 | 0.032 | 0.032 | 0.032 | 0.032 | 0.032 |
Q-stat | 701.855 | ||||||
I-squared | 93.52 | 87.031 | 86.966 | 86.983 | 86.999 | 87.015 | 87.047 |
Index of Dispersion | [-0.13, 0.716] |
Measure | Random-Effects Model | RVE ⍴ = .8 | RVE ⍴ = 0 | RVE ⍴ = .2 | RVE ⍴ = .4 | RVE ⍴ = .6 | RVE ⍴ = 1 |
r | 0.366 | 0.32 | 0.32 | 0.32 | 0.32 | 0.32 | 0.32 |
95% CI | [0.307, 0.423] | [0.273, 0.365] | [0.273, 0.365] | [0.273, 0.365] | [0.273, 0.365] | [0.273, 0.365] | [0.273, 0.365] |
z-value | 11.303 | ||||||
t-value | 12.99 | 12.991 | 12.991 | 12.991 | 12.991 | 12.99 | |
tau-squared | 0.068 | 0.032 | 0.032 | 0.032 | 0.032 | 0.032 | 0.032 |
Q-stat | 701.855 | ||||||
I-squared | 93.52 | 87.031 | 86.966 | 86.983 | 86.999 | 87.015 | 87.047 |
Index of Dispersion | [-0.13, 0.716] |
Note: Robust variance estimation (RVE) model at ⍴ = .8 is the weight used in all RVE models.
Stress measure. Type of life stressor was examined using RVE meta-regression approach, as 13 samples included correlations for multiple stressors. Out of the 157 correlations, the majority included a measure of war trauma (kc[^19] = 78), followed by general life stress (kc = 23), abuse (kc = 22), trauma (kc = 17), migration trauma (kc =7), sexual trauma (kc =5), and bullying (kc = 5). We re-grouped the stressors to form categories of at least 10 stressors: abuse (abuse and bullying; kc = 27), general life stress (kc = 23), trauma (trauma, sexual trauma, and migration trauma; kc = 29), and war trauma (kc =78). Inconsistent with our hypothesis, life-stressor type was not a significant moderator of the estimate between life stressors and negative mental health (see Table 13). All life-stressor types were significantly associated with negative mental-health outcomes (see Table 14).
We also examined life-stressor measure quality-related moderators, which included: number of items, the measure’s internal reliability or Cronbach’s alpha, and measure respondent (see Table 13). The average number of life-stressor measure items was 18.79 (SD = 19.43; range = 1–104; k = 51). Aligned with our hypothesis, number of items was a significant moderator, z = 2.236, p = .025, with a one-item increase above the mean number of items, resulting in a r = .003 increase from r = .344 (Table 13). However, in terms of p-value significance, this result did not replicate using RVE meta-regression, although the beta estimate was the same (see Table S6, https://osf.io/gkt8u/). The average Cronbach’s alpha or measure internal reliability was 0.82 (SD = 0.09; range = 0.63–0.97; k = 17) and did not explain the between-study variance in the estimate between life stressors and mental health, z = 1.295, p = .195. Measure respondent was examined using RVE meta-regression approach, as one sample included life-stress reports from multiple respondents. Out of the 157 correlations, the majority included a self-report measure (kc = 70), followed by an interview (kc = 45), then, clinical interview with professional (kc = 18), parent-report (kc = 16), medical record (kc = 4), and a combination of reporters (kc = 4). Due to the numbers per group, we grouped together medical record and combination of reporters and found measure respondent did not explain between-sample variation in effect size (see Table 13). Based upon this data, the results were mixed as it relates to our hypotheses for stress measure moderators; aligned with our hypothesis, increased number of measure items significantly explained some between-study variance between life stressors and mental health, but other life-stressor measure quality-related variables did not.
Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) | M(sd)/% | Range |
Publication Status | 69 | 0.004 | 8.42(2) | |||||||
Published (reference) | 57 | 0.341/0.329 | 0.035 | 9.793 | ||||||
Unpublished | 12 | 0.243/0.526 | 0.084 | 2.902 | 0.004 | |||||
Study Design | 69 | 0.908 | 0.013(2) | |||||||
Crosssectional (reference) | 61 | 0.385/0.367 | 0.036 | 10.753 | ||||||
Longitudinal/Intervention | 8 | -0.013/0.356 | 0.11 | -0.115 | 0.908 | |||||
Type of Sample Population | 69 | 0.047 | 6.105(2) | |||||||
Community (reference) | 33 | 0.304/0.295 | 0.047 | 6.522 | ||||||
Combination | 18 | 0.128/0.407 | 0.079 | 1.63 | 0.103 | |||||
Organization | 18 | 0.186/0.454 | 0.081 | 2.303 | 0.021 | |||||
Correlation Type | 157 | |||||||||
Pearson r (reference) | 119 | 0.344/0.331 | 0.028 | 12.239 | 44.182 | |||||
Phi | 24 | -0.088/0.25 | 0.048 | 0.081 | -1.844 | 18.516 | ||||
Chi-squared/Means | 14 | 0.036/0.362 | 0.116 | 0.764 | 0.308 | 11.075 | ||||
Life Stressor Measure: | ||||||||||
Mean Life stressor Measure Items | 51 | 0.003 | 0.001 | 2.236 | 0.025 | 18.79 (19.43) | (1-104) | |||
Mean Life stressor Measure Alpha | 17 | 0.706 | 0.545 | 1.295 | 0.195 | 0.82 (0.09) | (0.63-0.97) | |||
Life Stressor Type | 157 | |||||||||
Abuse (reference) | 27 | 0.438/0.412 | 0.086 | 5.085 | 9.42 | |||||
General Life | 23 | -0.139/0.29 | 0.102 | 0.189 | -1.362 | 19.079 | ||||
Trauma | 29 | -0.146/0.284 | 0.114 | 0.213 | -1.287 | 19.594 | ||||
War trauma | 78 | -0.111/0.316 | 0.09 | 0.239 | -1.226 | 15.018 | ||||
Respondent- Life stressor Measure | 157 | |||||||||
Clinical interivew (reference) | 18 | 0.288/0.28 | 0.051 | 5.674 | 5.913 | |||||
Interview | 45 | 0.073/0.345 | 0.067 | 0.306 | 1.085 | 9.142 | ||||
Other | 8 | 0.064/0.338 | 0.075 | 0.417 | 0.853 | 8.435 | ||||
Parent-report | 16 | -0.157/0.13 | 0.087 | 0.102 | -1.812 | 9.503 | ||||
Self-report | 70 | 0.066/0.339 | 0.068 | 0.358 | 0.965 | 9.465 | ||||
Mental Health Measure: | ||||||||||
Mean Mental Health Measure Items | 53 | 0.007 | 0.004 | 1.889 | 0.059 | 17.44 (10.68) | (1-72) | |||
Mean Mental Health Measure Alpha | 35 | 1.007 | 0.577 | 1.745 | 0.081 | 0.87 (0.07) | (0.62-0.97) | |||
Mean Mental Health Response Period | 37 | -0.002 | 0.001 | -1.751 | 0.08 | 22.93 (35.83) | (7-180) |
Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) | M(sd)/% | Range |
Publication Status | 69 | 0.004 | 8.42(2) | |||||||
Published (reference) | 57 | 0.341/0.329 | 0.035 | 9.793 | ||||||
Unpublished | 12 | 0.243/0.526 | 0.084 | 2.902 | 0.004 | |||||
Study Design | 69 | 0.908 | 0.013(2) | |||||||
Crosssectional (reference) | 61 | 0.385/0.367 | 0.036 | 10.753 | ||||||
Longitudinal/Intervention | 8 | -0.013/0.356 | 0.11 | -0.115 | 0.908 | |||||
Type of Sample Population | 69 | 0.047 | 6.105(2) | |||||||
Community (reference) | 33 | 0.304/0.295 | 0.047 | 6.522 | ||||||
Combination | 18 | 0.128/0.407 | 0.079 | 1.63 | 0.103 | |||||
Organization | 18 | 0.186/0.454 | 0.081 | 2.303 | 0.021 | |||||
Correlation Type | 157 | |||||||||
Pearson r (reference) | 119 | 0.344/0.331 | 0.028 | 12.239 | 44.182 | |||||
Phi | 24 | -0.088/0.25 | 0.048 | 0.081 | -1.844 | 18.516 | ||||
Chi-squared/Means | 14 | 0.036/0.362 | 0.116 | 0.764 | 0.308 | 11.075 | ||||
Life Stressor Measure: | ||||||||||
Mean Life stressor Measure Items | 51 | 0.003 | 0.001 | 2.236 | 0.025 | 18.79 (19.43) | (1-104) | |||
Mean Life stressor Measure Alpha | 17 | 0.706 | 0.545 | 1.295 | 0.195 | 0.82 (0.09) | (0.63-0.97) | |||
Life Stressor Type | 157 | |||||||||
Abuse (reference) | 27 | 0.438/0.412 | 0.086 | 5.085 | 9.42 | |||||
General Life | 23 | -0.139/0.29 | 0.102 | 0.189 | -1.362 | 19.079 | ||||
Trauma | 29 | -0.146/0.284 | 0.114 | 0.213 | -1.287 | 19.594 | ||||
War trauma | 78 | -0.111/0.316 | 0.09 | 0.239 | -1.226 | 15.018 | ||||
Respondent- Life stressor Measure | 157 | |||||||||
Clinical interivew (reference) | 18 | 0.288/0.28 | 0.051 | 5.674 | 5.913 | |||||
Interview | 45 | 0.073/0.345 | 0.067 | 0.306 | 1.085 | 9.142 | ||||
Other | 8 | 0.064/0.338 | 0.075 | 0.417 | 0.853 | 8.435 | ||||
Parent-report | 16 | -0.157/0.13 | 0.087 | 0.102 | -1.812 | 9.503 | ||||
Self-report | 70 | 0.066/0.339 | 0.068 | 0.358 | 0.965 | 9.465 | ||||
Mental Health Measure: | ||||||||||
Mean Mental Health Measure Items | 53 | 0.007 | 0.004 | 1.889 | 0.059 | 17.44 (10.68) | (1-72) | |||
Mean Mental Health Measure Alpha | 35 | 1.007 | 0.577 | 1.745 | 0.081 | 0.87 (0.07) | (0.62-0.97) | |||
Mean Mental Health Response Period | 37 | -0.002 | 0.001 | -1.751 | 0.08 | 22.93 (35.83) | (7-180) |
k = number of samples; kc = number of outcomes or correlations. Note: Random-effects models and mixed-effects models display z statistics, while robust variation estimation (RVE) models display t statistics. Z values are based upon the population mean while t statistics are based upon the sample mean. As random-effects models and mixed-effects model estimate the population means, those models use z values.
k | r | 95% CI | kc | t.value | p | Tau.squared | I.squared | |
Abuse | 12 | 0.401 | [0.245, 0.536] | 27 | 5.357 | <.001 | 0.084 | 94.169 |
General life stress | 14 | 0.271 | [0.181, 0.356] | 23 | 6.361 | <.001 | 0.018 | 79.091 |
Trauma | 17 | 0.267 | [0.13, 0.394] | 29 | 4.081 | 0.001 | 0.041 | 86.545 |
War trauma | 39 | 0.307 | [0.258, 0.354] | 78 | 12.173 | <.001 | 0.022 | 83.302 |
k | r | 95% CI | kc | t.value | p | Tau.squared | I.squared | |
Abuse | 12 | 0.401 | [0.245, 0.536] | 27 | 5.357 | <.001 | 0.084 | 94.169 |
General life stress | 14 | 0.271 | [0.181, 0.356] | 23 | 6.361 | <.001 | 0.018 | 79.091 |
Trauma | 17 | 0.267 | [0.13, 0.394] | 29 | 4.081 | 0.001 | 0.041 | 86.545 |
War trauma | 39 | 0.307 | [0.258, 0.354] | 78 | 12.173 | <.001 | 0.022 | 83.302 |
k = number of samples; kc = number of outcomes or correlations
Mental health measure. We examined mental-health measure quality-related moderators,18 which included: number of items, the measure’s internal reliability or Cronbach’s alpha, measure response time period, and respondent (Table 13). The average number of mental-health measure items was 17.44 (SD = 10.68; range = 1–72; k = 53). Number of mental-health measure items was a not a significant moderator, z = 1.889, p = .059. The average Cronbach’s alpha or measure internal reliability was 0.87 (SD = 0.07; range = 0.62–0.97; k = 35) and was not a significant moderator of the association between discrimination and mental health, z = 1.745, p = .081. The average response time period was 23 days (SD=35.83, range = 7-180) and was not a significant moderator. Measure respondent was examined using RVE meta-regression approach, as four samples included mental-health outcomes from multiple respondents. Out of the 157 correlations, the majority included a self-report measure (kc = 74), followed by an interview (kc = 44), then, clinical interview (kc = 16), parent-report (kc = 16), medical record (kc = 3), and combination of reports (kc = 4). Due to the numbers per group, we grouped together medical records and combination reports, and we found that measure respondent did not account for between-study variance in the association between life stressors and negative mental health (see Table 13). Based upon this data, no mental-health measure quality significantly moderated the association between life stressors and negative mental health.
Three samples included both correlations for positive and negative mental health. Out of the 157 correlations, the majority included a measure of negative mental health (kc = 152), and only five measured positive mental health (kc = 5). Due the low number of positive mental health correlations, we were unable to conduct a meta-regression analysis to test our hypothesis. Table 15 summarizes the RVE model estimates for life stressors with negative mental-health outcomes and positive health outcomes.19
r | 95% CI | kc | k | t.value | p | Tau.squared | I.squared | |
Negative Mental Health | 0.322 | [0.275, 0.368] | 152 | 68 | 12.831 | <.001 | 0.033 | 87.54 |
Positive Mental Health | -0.188 | [-0.411, 0.055] | 5 | 4 | -3.431 | 0.079 | 0.01 | 50.005 |
Depression | 0.331 | [0.241, 0.416] | 40 | 28 | 7.215 | <.001 | 0.042 | 87.026 |
Anxiety | 0.264 | [0.093, 0.42] | 8 | 7 | 3.995 | 0.012 | 0.015 | 65.62 |
Internalizing | 0.234 | [0.071, 0.384] | 7 | 6 | 3.78 | 0.015 | 0.013 | 75.232 |
Externalizing | 0.146 | [-0.197, 0.457] | 5 | 4 | 1.375 | 0.266 | 0.026 | 80.281 |
Psychological distress | 0.299 | [0.214, 0.379] | 31 | 20 | 7.093 | <.001 | 0.024 | 87.481 |
PTSD | 0.360 | [0.283, 0.431] | 49 | 34 | 9.017 | <.001 | 0.043 | 88.951 |
Somatization | 0.412 | [0.252, 0.551] | 9 | 7 | 6.059 | 0.001 | 0.028 | 85.828 |
Wellbeing | -0.153 | [-0.279, -0.022] | 8 | 6 | -3.48 | 0.033 | 0.01 | 66.126 |
r | 95% CI | kc | k | t.value | p | Tau.squared | I.squared | |
Negative Mental Health | 0.322 | [0.275, 0.368] | 152 | 68 | 12.831 | <.001 | 0.033 | 87.54 |
Positive Mental Health | -0.188 | [-0.411, 0.055] | 5 | 4 | -3.431 | 0.079 | 0.01 | 50.005 |
Depression | 0.331 | [0.241, 0.416] | 40 | 28 | 7.215 | <.001 | 0.042 | 87.026 |
Anxiety | 0.264 | [0.093, 0.42] | 8 | 7 | 3.995 | 0.012 | 0.015 | 65.62 |
Internalizing | 0.234 | [0.071, 0.384] | 7 | 6 | 3.78 | 0.015 | 0.013 | 75.232 |
Externalizing | 0.146 | [-0.197, 0.457] | 5 | 4 | 1.375 | 0.266 | 0.026 | 80.281 |
Psychological distress | 0.299 | [0.214, 0.379] | 31 | 20 | 7.093 | <.001 | 0.024 | 87.481 |
PTSD | 0.360 | [0.283, 0.431] | 49 | 34 | 9.017 | <.001 | 0.043 | 88.951 |
Somatization | 0.412 | [0.252, 0.551] | 9 | 7 | 6.059 | 0.001 | 0.028 | 85.828 |
Wellbeing | -0.153 | [-0.279, -0.022] | 8 | 6 | -3.48 | 0.033 | 0.01 | 66.126 |
k = number of samples; kc = number of outcomes or correlations. Note: Estimates are exploratory when kc/k < 10.
Thirty-two samples included correlations for multiple mental-health outcomes. Out of the 157 correlations, the breakdown of mental-health outcomes was as follows: anxiety (kc = 8), depression (kc = 40), externalizing (kc = 5), internalizing (kc = 7), psychological distress (kc = 31), PTSD (kc = 49), somatization (kc = 9), and well-being (kc = 8). Because many outcomes included fewer than 10 correlations, we were unable to perform a moderation analysis. However, we conducted subgroup analyses using RVE for the mental-health outcomes with nearly 10 or more correlations (see Table 15). These results show that life stress was most strongly related to somatization, r = .412 (95% CI [.25, .55]), and PTSD, r = .360 (95% CI [.28, .43]), followed by, depression, r = .331 (95% CI [.24, .42]), psychological distress, r = .299 (95% CI [.21, .38]), anxiety, r = .264 (95% CI [.09, .42]), internalizing, r = .234 (95% CI [.07, .38]), and well-being, r = -.153 (95% CI [-.28, -.02]).
Demographic characteristics.Table 16 summarize the meta-regressions for sociocultural characteristics. Sample mean age was tested as a continuous moderator in the mixed-effects meta-regression model and did not explain between-sample variation in effect size (z = .393, p = .694). Percent women (M = 50% women; SD = 27.83) was also tested as a continuous moderator and—inconsistent with our hypothesis—it did not explain between-sample variation in effect size (z = 1.611, p = .107). In terms of education, 15 percent of the sample had a bachelor’s degree or higher; however, education data was only extracted from 30 samples. Where education was tested as a continuous moderator, it proved not to be a significant moderator of the association between life stressors and mental health (z = .564, p = .573).
The majority of samples were from North America (k = 39; kc = 91), followed by Europe (k = 29; kc = 65) and Australia (k = 1; kc = 1). Continent was tested as categorical moderator and nearly explained between-study variance, Qbetween (2) = 5.952, p = .051. The results showed that samples living in North America (r = .425) had a significantly larger association by b = .162, compared to samples living in Europe (r = .283; see Table 16).20 Because there were not 10 samples per country, we were unable to perform moderation analysis. Instead, we present the random-effect model estimates between life stressors and negative mental health for each country (see Table 17). It is worth noting that only the estimate for the United States is a priori analysis, as the estimates for countries with less than 10 samples are exploratory analyses. Using the random-effects model approach, we found a statistically significant summary effect between life stressors and negative mental health for samples in the United States (k = 36), r = .440 (95% CI [.35, .52]).
In order to test when ethnicity or country of origin was a significant moderator, we conducted our analyses with the entire life-stressor dataset and also only with samples that were ethnically homogenous (had 80% or more of the same ethnic group; k = 54, kc = 131). Both approaches resulted in underpowered meta-regression analyses with fewer than 10 samples per ethnic group (see Table 16 and Table S5, https://osf.io/gkt8u/). We combined ethnic groups so that there were at least 10 samples per group, and we found ethnicity to explain between-study variance in the effect-size estimate, whereas the Somali sample (r = .370–.381) had a stronger effect, b = .095–.141, than the Middle Eastern sample (r = .254–.286; see Table 16 and Table S5, https://osf.io/gkt8u/). The beta estimation for the Somali sample was similar across analytic approaches, but the p-value level of significance was not consistent (see Table 16 and Table S5, https://osf.io/gkt8u/). Overall, results suggest that ethnicity may explain some between-study variance in the meta-analytic estimate, specifically where the effect between life stressors and mental health is stronger for the Somali sample. We also tested to see when Muslim sample type (Muslim vs majority-Muslim country) was a significant moderator Table 16. There were 24 Muslim samples and 45 majority-Muslim-country samples. Muslim sample type was not a significant moderator, z = .824, p = .41.
Table 16 summarizes the results of immigration status. Immigration status was examined through percent born in a western country (28%; k = 19), percent first-generation immigrants (80%; k = 18), and percent refugee (93%; k = 38). Percent born in a Western country tested as a continuous moderator did not explain between-study variance in effect size (z = -1.465, p = .143), nor did percent first-generation immigrant explain between-study variance in effect size (z = -1.251, p = .211). Percent refugee was tested as a continuous moderator and did not explain the between-study variance in effect size. Notably, there was very little variation among the studies that presented percent refugee status; from the 38 samples with refugee status information, most included a 100% refugee sample population. These results suggest that none of the indicators of immigration status explained between-study variance in the effect-size estimate; however, there was little variation among the studies that presented immigration status information.
Identity. Six unique samples included identity variables in the life stressors dataset (see Table 10 to identify the six samples). Identity variables included: ethnic identity (k = 3), national identity (k = 2), religious identity (k = 4), and identity integration (k = 2). There were not enough samples with identity variables to conduct meta-regression analyses. The qualitative review of the six unique samples revealed that none of the studies included analyses to determine if identity moderated the association between life stressors and mental health; therefore, no conclusions can be made regarding the moderating effect of identity.
Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) | M(sd)/% | Range |
Mean Age | 68 | 0.001 | 0.003 | 0.393 | 0.694 | 29.59 (11.82) | (2.8-48.75) | |||
% Female | 69 | 0.002 | 0.001 | 1.611 | 0.107 | 50.17 (27.83) | (0-100) | |||
% Bachelors+ | 30 | 0.002 | 0.003 | 0.564 | 0.573 | 14.53 (21.26) | (0-90.2) | |||
% Born in Western country | 19 | -0.003 | 0.002 | -1.465 | 0.143 | 27.62 (37.57) | (0-100) | |||
% First generation | 18 | -0.002 | 0.001 | -1.251 | 0.211 | 80.27 (31.68) | (0-100) | |||
% Refugee | 38 | -0.002 | 0.002 | -1.136 | 0.256 | 93.21 (18.22) | (25.7-100) | |||
Continent | 69 | 0.051 | 5.952(2) | |||||||
Europe (reference) | 29 | 0.291/0.283 | 0.05 | 5.815 | ||||||
North America | 39 | 0.162/0.425 | 0.067 | 2.428 | 0.015 | |||||
Australia/New Zealand | 1 | 0.156/0.42 | 0.278 | 0.563 | 0.574 | |||||
Ethnicity (80+) | 131 | |||||||||
Middle Eastern (reference) | 66 | 0.26/0.254 | 0.039 | 6.646 | 26.613 | |||||
Somali | 46 | 0.141/0.381 | 0.053 | 0.012 | 2.689 | 27.191 | ||||
Bosnian Peninsula | 10 | 0.158/0.395 | 0.073 | 0.122 | 2.149 | 2.95 | ||||
Other | 9 | 0.023/0.275 | 0.069 | 0.749 | 0.336 | 5.607 | ||||
Ethnicity | 157 | |||||||||
Middle Eastern (reference) | 86 | 0.294/0.286 | 0.039 | 7.599 | 37.568 | |||||
Bosnian Peninsula | 10 | 0.128/0.398 | 0.075 | 0.191 | 1.711 | 2.836 | ||||
Somali | 50 | 0.095/0.37 | 0.05 | 0.070 | 1.882 | 29.641 | ||||
Other | 11 | 0.065/0.344 | 0.08 | 0.444 | 0.803 | 8.308 | ||||
Muslim sample type | 69 | 0.410 | 0.679(2) | |||||||
Majority Muslim country (reference) | 45 | 0.364/0.348 | 0.042 | 8.628 | ||||||
Muslim | 24 | 0.059/0.399 | 0.071 | 0.824 | 0.410 |
Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) | M(sd)/% | Range |
Mean Age | 68 | 0.001 | 0.003 | 0.393 | 0.694 | 29.59 (11.82) | (2.8-48.75) | |||
% Female | 69 | 0.002 | 0.001 | 1.611 | 0.107 | 50.17 (27.83) | (0-100) | |||
% Bachelors+ | 30 | 0.002 | 0.003 | 0.564 | 0.573 | 14.53 (21.26) | (0-90.2) | |||
% Born in Western country | 19 | -0.003 | 0.002 | -1.465 | 0.143 | 27.62 (37.57) | (0-100) | |||
% First generation | 18 | -0.002 | 0.001 | -1.251 | 0.211 | 80.27 (31.68) | (0-100) | |||
% Refugee | 38 | -0.002 | 0.002 | -1.136 | 0.256 | 93.21 (18.22) | (25.7-100) | |||
Continent | 69 | 0.051 | 5.952(2) | |||||||
Europe (reference) | 29 | 0.291/0.283 | 0.05 | 5.815 | ||||||
North America | 39 | 0.162/0.425 | 0.067 | 2.428 | 0.015 | |||||
Australia/New Zealand | 1 | 0.156/0.42 | 0.278 | 0.563 | 0.574 | |||||
Ethnicity (80+) | 131 | |||||||||
Middle Eastern (reference) | 66 | 0.26/0.254 | 0.039 | 6.646 | 26.613 | |||||
Somali | 46 | 0.141/0.381 | 0.053 | 0.012 | 2.689 | 27.191 | ||||
Bosnian Peninsula | 10 | 0.158/0.395 | 0.073 | 0.122 | 2.149 | 2.95 | ||||
Other | 9 | 0.023/0.275 | 0.069 | 0.749 | 0.336 | 5.607 | ||||
Ethnicity | 157 | |||||||||
Middle Eastern (reference) | 86 | 0.294/0.286 | 0.039 | 7.599 | 37.568 | |||||
Bosnian Peninsula | 10 | 0.128/0.398 | 0.075 | 0.191 | 1.711 | 2.836 | ||||
Somali | 50 | 0.095/0.37 | 0.05 | 0.070 | 1.882 | 29.641 | ||||
Other | 11 | 0.065/0.344 | 0.08 | 0.444 | 0.803 | 8.308 | ||||
Muslim sample type | 69 | 0.410 | 0.679(2) | |||||||
Majority Muslim country (reference) | 45 | 0.364/0.348 | 0.042 | 8.628 | ||||||
Muslim | 24 | 0.059/0.399 | 0.071 | 0.824 | 0.410 |
k = number of samples; kc = number of outcomes or correlations. Note: Random-effects models and mixed-effects models display z statistics, while robust variation estimation (RVE) models display t statistics. Z values are based upon the population mean while t statistics are based upon the sample mean. As random-effects models and mixed-effects model estimate the population means, those models use z values.
Country | k | r | 95% CI | z | p | Tau squared | I squared |
Australia | 1 | 0.42 | [0.228-0.58] | 4.079 | <.001 | 0 | 0 |
Belgium | 1 | 0.144 | [0.046-0.24] | 2.867 | 0.004 | 0 | 0 |
Canada | 3 | 0.161 | [-0.134-0.43] | 1.069 | 0.285 | 0.046 | 75.403 |
Germany | 4 | 0.35 | [-0.165-0.715] | 1.345 | 0.179 | 0.255 | 93.619 |
Netherlands | 3 | 0.183 | [0.103-0.26] | 4.449 | <.001 | 0 | 0 |
Norway | 4 | 0.17 | [0.122-0.217] | 6.866 | <.001 | 0 | 0.014 |
Sweden | 7 | 0.321 | [0.169-0.458] | 4.029 | <.001 | 0.037 | 88.769 |
United Kingdom | 8 | 0.364 | [0.293-0.432] | 9.373 | <.001 | 0.007 | 58.515 |
USA | 36 | 0.44 | [0.354-0.519] | 9.026 | <.001 | 0.086 | 94.314 |
Country | k | r | 95% CI | z | p | Tau squared | I squared |
Australia | 1 | 0.42 | [0.228-0.58] | 4.079 | <.001 | 0 | 0 |
Belgium | 1 | 0.144 | [0.046-0.24] | 2.867 | 0.004 | 0 | 0 |
Canada | 3 | 0.161 | [-0.134-0.43] | 1.069 | 0.285 | 0.046 | 75.403 |
Germany | 4 | 0.35 | [-0.165-0.715] | 1.345 | 0.179 | 0.255 | 93.619 |
Netherlands | 3 | 0.183 | [0.103-0.26] | 4.449 | <.001 | 0 | 0 |
Norway | 4 | 0.17 | [0.122-0.217] | 6.866 | <.001 | 0 | 0.014 |
Sweden | 7 | 0.321 | [0.169-0.458] | 4.029 | <.001 | 0.037 | 88.769 |
United Kingdom | 8 | 0.364 | [0.293-0.432] | 9.373 | <.001 | 0.007 | 58.515 |
USA | 36 | 0.44 | [0.354-0.519] | 9.026 | <.001 | 0.086 | 94.314 |
Summary of meta-analytic results of the association between life stressors and mental health
The summary correlation between life stressors and mental health for Muslims living in western countries was consistent with our hypothesis—it was positive and falls within the moderate to large range. There was evidence of publication bias where unpublished studies were found to have a stronger and positive effect-size estimate compared to published studies. However, the meta-analytic estimate appeared to be robust for 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 a majority Middle Eastern, majority North American, lower-educated, and young Muslim population.
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 study-, demographic-, and identity-related moderators tested. However, a few study- and demographic-related variables did explain some between-study variance in the meta-analytic estimate. Interestingly, beta estimates of moderators were consistent across analyses models, but p-value significance varied.
Briefly, evidence suggested that the type of sample population was a significant moderator with a) samples from organizations and b) samples from schools, as these have stronger associations between life stressors and mental health compared to samples from the community. Additionally, some evidence suggested that number of life-stressor measure items was a significant moderator; specifically, as the number of life-stressor measure items increased, the association between life stressors and mental health was stronger. In terms of demographic-related variables, ethnicity appeared to be significant moderator of the association between life stressors and negative mental-health outcomes, specifically where Somalis were found to have a stronger association between life stressors and negative mental health compared to other ethnic groups. Additionally, some evidence suggests that continent explained between-study variance, specifically where samples in North America were found to have a stronger association between life stressors and mental-health outcomes compared to samples from Europe.
Discrimination and Life Stressors
Fourteen samples measured both experiences of discrimination and life stressors. The analytic subset that included both measures of discrimination and life stressors includes 11 unique samples,21 with 75 correlations and 3,416 individuals. The 11 unique samples had a mean sample size of 310 (range = 30–1,215) and represented individuals living in five different Western countries (see Table 18). The mean age of participants was 31 years old (range = 10–49 years old), with 47% women and a (collapsed) majority ethnic breakdown of 64% Middle Eastern, 27% Somali, and 9% North African (see Table 18 for a detailed ethnicity breakdown).
Sample characteristics | Details |
k | 11 |
kc | 75 |
N (range) | 310 (30-1215) |
Age (M(sd)) | 30.73 (12.45) |
% Female (sd) | 46.82 (23.69) |
Data from published sources (%) | 100 |
Country | |
- Canada | k=1 |
- Netherlands | k=1 |
- Sweden | k=2 |
- United Kingdom | k=1 |
- USA | k=6 |
Majority Ethnicity (detailed) | |
- Afghani | k=1 |
- Arab | k=1 |
- Iraqi | k=2 |
- Middle Eastern | k=1 |
- Moroccan | k=1 |
- Palestinian | k=1 |
- Somali | k=3 |
- Syrian | k=1 |
Majority Ethnicity | |
- Middle Eastern | k=7 (63.6%) |
- North African | k=1 (9.1%) |
- Somali | k=3 (27.3%) |
Sample characteristics | Details |
k | 11 |
kc | 75 |
N (range) | 310 (30-1215) |
Age (M(sd)) | 30.73 (12.45) |
% Female (sd) | 46.82 (23.69) |
Data from published sources (%) | 100 |
Country | |
- Canada | k=1 |
- Netherlands | k=1 |
- Sweden | k=2 |
- United Kingdom | k=1 |
- USA | k=6 |
Majority Ethnicity (detailed) | |
- Afghani | k=1 |
- Arab | k=1 |
- Iraqi | k=2 |
- Middle Eastern | k=1 |
- Moroccan | k=1 |
- Palestinian | k=1 |
- Somali | k=3 |
- Syrian | k=1 |
Majority Ethnicity | |
- Middle Eastern | k=7 (63.6%) |
- North African | k=1 (9.1%) |
- Somali | k=3 (27.3%) |
k = number of samples; kc = number of outcomes or correlations
RQ5: Effect of stress type (discrimination vs. life stressors) on mental-health outcomes
Using mixed-effects modeling with the full combined dataset (k = 141, kc = 295), stress type did explain the between-study variance of the association between stressor (discrimination and life stressors) and mental health, Qbetween (2) = 14.08, p <.001, where both stress types were significantly associated with mental health. However, the association was stronger for life stressors and mental-health outcomes (see Table 19) than for discrimination and mental-health outcomes. We tested the equivalency between the discrimination-mental-health effect-size estimate and the life stressors-mental health effect-size estimate using the TOST-p and KTOST-p test using an SESOI of r = .15, which is a small effect size difference (Counsell & Cribbie, 2015).22 Notably, Counsell & Cribbie (2015) found that TOST-p and KTOST-p are conservative tests at lower samples sizes, wherein their simulations, “the null hypothesis was never rejected for sample sizes below 250 per group” (pp. 299). Therefore, we used another equivalence test for independent correlation coefficients for further verification (AH-p), which has proven to be more reliable for smaller sample sizes; results confirmed that we could not reject the hypothesis that the coefficients were not equivalent, p = .42 (Counsell & Cribbie, 2015). Inconsistent with our original hypothesis, results of the three tests consistently found the discrimination-mental health effect-size estimate and life stressor-mental health effect-size estimate (including meta-analytic estimates without outliers or publication bias corrected) were not equivalent.
Conversely, using mixed-effects modeling with the subset of data that included both discrimination and life stressors (k = 11, kc = 75), stress type did not moderate nor explain the between-study variance of the association between stressor (discrimination and life stressors) and mental health, Qbetween (1) = .05, p = .822 (see Table 20). We tested the equivalency between a) the discrimination-mental health effect-size estimate (r = .301 (REM); r = .271 (RVE model)) and b) the life stressors-mental health effect-size estimate (r = . 319 (REM); r = .273 (RVE model)), which were derived from the subset of data, and we found mixed results. Using the TOST-p and KTOST-p test with an SESOI of r = .15, the estimates were not found to be statistically equivalent (Counsell & Cribbie, 2015).23 Using the equivalence test that has proven to be more reliable for smaller sample sizes (AH-p; Counsell & Cribbie, 2015), the coefficients were statistically equivalent, p = . 006 (REM) and p = .032 (RVE model). Results of the three tests provided varied results, which appears to be related to the very small sample size. Therefore, it appears the discrimination-mental-health effect-size estimate and life stressors-mental health effect-size estimate may be equivalent when examining the subset of data that measures both stressors.
In summary, the meta-analytic correlations of both a) discrimination and mental health and b) life stressors and mental health were significantly and positively associated. Results are mixed in terms of the moderation effect of stress type and equivalence of effect-size estimates. Inconsistent with our hypothesis, analyses that included the entire combined dataset show that stress type is a significant moderator, and that the effect-size estimates for the two datasets are not equivalent, specifically where the association between life stressors and mental health is stronger. However, consistent with our hypothesis, results from the subset of data that includes only the parallel samples that measure both discrimination and life stressors suggest that stress type was not a significant moderator and that the effect-size estimates appear equivalent, although noting the small sample. These findings suggest that other characteristic differences between the discrimination and life-stressor datasets may be contributing to the difference in meta-analytic correlations.
RQ6: Unique contribution of discrimination and life stressors to the association between stressors and mental health
From the 11 samples included in the analytic subset, only six samples presented partial correlations of the associations between (1) discrimination and mental health, accounting for the effect of life stressors and/or (2) life stressors and mental health, accounting for the effect of discrimination. We conducted exploratory analyses using RVE with the five samples and found a statistically significant summary effect of the partial correlation between discrimination and mental health, adjusting for the effect of life stressors, r = .19 (95% CI [ .02, .35]), t (4.98) = 2.8, p =.038. We also found a statistically significant summary effect of the partial correlation between life stressors and mental health, adjusting for the effect of discrimination, r = .22 (95% CI [ .0004, .41]), t (3.99) = 2.78, p =.0497 (see Table S9, https://osf.io/gkt8u/). Thus, consistent with our hypothesis, these underpowered analyses provide initial support that discrimination and life stressors uniquely contribute to mental-health outcomes.
Qualitative review. Due to limited quantitative data, we conducted a qualitative review of the 14 samples that included both discrimination and life-stressor measures to examine the unique contribution of discrimination and life stressors as they relate to mental-health outcomes. See Appendix C for qualitative review of eight samples that did not include partial correlations; no consensus was found among the eight samples regarding the association between discrimination and life stressors with mental-health outcomes.
Categorical Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) |
Stress Type (REM) | 141 | <.001 | 14.075(2) | |||||
Discrimination (reference) | 72 | 0.239/0.234 | 0.026 | 9.077 | ||||
Life Stressor | 69 | 0.144/0.365 | 0.038 | 3.752 | <.001 | |||
Stress Type (RVE) | 295 | |||||||
Discrimination (reference) | 138 | 0.21/0.207 | 0.017 | 12.354 | 66.191 | |||
Life Stressor | 157 | 0.125/0.323 | 0.031 | <.001 | 4.093 | 120.235 | ||
Categorical Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) |
Stress Type (REM) | 141 | <.001 | 14.075(2) | |||||
Discrimination (reference) | 72 | 0.239/0.234 | 0.026 | 9.077 | ||||
Life Stressor | 69 | 0.144/0.365 | 0.038 | 3.752 | <.001 | |||
Stress Type (RVE) | 295 | |||||||
Discrimination (reference) | 138 | 0.21/0.207 | 0.017 | 12.354 | 66.191 | |||
Life Stressor | 157 | 0.125/0.323 | 0.031 | <.001 | 4.093 | 120.235 | ||
k = number of samples; kc = number of outcomes or correlations. Note: Random-effects models and mixed-effects models display z statistics, while robust variation estimation (RVE) models display t statistics. Z values are based upon the population mean while t statistics are based upon the sample mean. As random-effects models and mixed-effects model estimate the population means, those models use z values.
Categorical Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) |
Stress Type (REM) | 11 | 0.822 | 0.05(2) | |||||
Discrimination (reference) | 11 | 0.33/0.301 | 0.063 | 5.204 | ||||
Life Stressor | 11 | -0.02/0.319 | 0.09 | -0.225 | 0.822 | |||
Stress Type (RVE) | 75 | |||||||
Discrimination (reference) | 37 | 0.278/0.271 | 0.058 | 4.8 | 8.296 | |||
Life Stressor | 38 | 0.003/0.273 | 0.05 | 0.961 | 0.05 | 9.72 |
Categorical Moderator | k/kc | b/r | SE | z | p | t | dft | Q(df) |
Stress Type (REM) | 11 | 0.822 | 0.05(2) | |||||
Discrimination (reference) | 11 | 0.33/0.301 | 0.063 | 5.204 | ||||
Life Stressor | 11 | -0.02/0.319 | 0.09 | -0.225 | 0.822 | |||
Stress Type (RVE) | 75 | |||||||
Discrimination (reference) | 37 | 0.278/0.271 | 0.058 | 4.8 | 8.296 | |||
Life Stressor | 38 | 0.003/0.273 | 0.05 | 0.961 | 0.05 | 9.72 |
k = number of samples; kc = number of outcomes or correlations. Note: Random-effects models and mixed-effects models display z statistics, while robust variation estimation (RVE) models display t statistics. Z values are based upon the population mean while t statistics are based upon the sample mean. As random-effects models and mixed-effects model estimate the population means, those models use z values.
Partial correlations of discrimination and life stressors. Briefly, from the six samples that included partial correlations, data from four samples revealed that both discrimination and life stressors were significantly associated with mental health (Alemi & Stempel, 2018; Ellis et al., 2008; Kira et al., 2010; Tinghög et al., 2017)). One sample showed that only life stressors remained significantly associated with mental health (Jorden et al., 2009), and one sample revealed varied results for both discrimination and life stressors, depending on the type of mental-health outcome measured (Abuelezam & El-Sayed, 2018).
Data from four samples showed that the adjusted discrimination and life-stressors estimates were positively and moderately associated with mental health; the adjusted life-stressor estimate appeared to be the stronger predictor. In data provided by a sample of Afghani refugees resettled in the United States, a model including other resettlement factors revealed that both life stressors and discrimination predicted negative mental-health outcomes. Here, the partial correlation of life stressors was the stronger predictor (ß = 0.30, p > .05) but did not maintain significance, and experiences of discrimination (ß = 0.26, p <.001) was the second strongest predictor (Alemi & Stempel, 2018). Similarly, in a sample of Iraqi refugees living in Michigan, discrimination (ß = 0.22, p = NR) and previous trauma (ß = 0.46, p = NR) were both associated with PTSD, but previous trauma was the stronger predictor (Kira et al., 2010). Moreover, in a sample of Somali teens, the adjusted estimates of discrimination and life stressors remained significant predictors of PTSD symptoms (discrimination, ß = 0.246, p <.01; life stressors, ß = 0.32, p <.01) and depressive symptoms (discrimination, ß = 0. 238, p <.01; life stressors, ß = 0.199, p <.05), even when accounting for other factors (Ellis et al., 2008). Although the study included only the adjusted correlation for discrimination, in a sample of Syrian refugees living in Sweden, discrimination was strongly associated with low well-being (ß = 0.43, p = NR), internalizing symptoms (ß = 0.41, p = NR24), and PTSD (ß = 0.40, p = NR), specifically when adjusted for various covariates, including traumatic events (Tinghög et al., 2017). The non-adjusted association between life stressors and mental-health outcomes varied: low well-being (r = 0.06, p = NR), internalizing symptoms (r = 0.18, p = NR), and PTSD (r = 0.20, p = NR; Tinghög et al., 2017).
Conversely, in a sample of Somali adults in Canada, the adjusted model found only various traumatic experiences significantly predicted depression (individual, ß = 0.14, p <.05; familiar assault, ß = .20, p <.01) and trauma symptoms (threat to other, ß = 0.20, p <.001; Jorden et al., 2009). And adjusted discrimination did not significantly predict depression (ß = .08, p> .05) or trauma symptoms (ß = 0.14, p >.05; Jorden et al., 2009).
In one sample, the strength and direction of adjusted correlations differed based upon type of mental-health outcome for both discrimination and life stressors. In a sample of Arabs living in Michigan, life stressors appeared more consistently related to depression than discrimination (discrimination, ß = 0.07, p = NR25; life stressors, ß = 0.11–0.26, p = NR). Discrimination—compared to life stressors—appeared more consistently related to psychological distress (discrimination, ß = 0.26, p = NR; life stressors, ß = -0.12–0.29, p = NR; Abuelezam & El-Sayed, 2018).
Interestingly, one study included results of the interaction of discrimination and life stressors in predicting mental-health outcomes. Life stressors was a significant moderator of the association between discrimination and psychological distress, specifically where increased life stressors exacerbated the effect of discrimination on psychological distress (interaction, ß = 0.403, p <.05; Alemi & Stempel, 2018).
Summary of meta-analytic results of stressor type
In summary, exploratory quantitative analyses and the qualitative review suggest that discrimination and life stressors are uniquely associated with mental-health outcomes. However, results are varied across samples. Interestingly, p-value significance and strength of the correlation appear to vary greatly within the 14 samples. It appears that the variation may be due to study-related characteristics such as sample size, type of life stressor, type of discrimination, or mental-health outcome. Overall, however, consistent with our hypothesis, there is initial support that discrimination and life stressors uniquely contribute to mental-health outcomes.
Discussion
This comprehensive preregistered meta-analysis and qualitative review, including 130 unique samples and 27,725 individuals, examined the associations between both (a) discrimination and negative mental-health outcomes and (b) life stressors and negative mental-health outcomes. Given the widespread context of Islamophobia in contemporary society, as evidenced by anti-Muslim stereotypes and propaganda across North America, Europe, New Zealand, and Australia, we addressed a pressing topic of how this context of Islamophobia may be associated with the psychological well-being of Muslims living in Western countries. Notably, this is also the first meta-analysis that focuses on Muslims living in Western countries, and furthermore, that examines the associations of both stressors—discrimination and life stressors—with mental-health. Using both meta-analytic and qualitative review approaches, we examined the associations between two types of stressors and mental-health outcomes and calculated omnibus effect-size estimates that were robust to outliers, within-study dependence, and test of publication bias. Overall, our meta-analyses for Muslims living in Western countries found (1) a statistically significant summary correlation between discrimination and mental health that is in the small to moderate range, and (2) a statistically significant summary correlation between life stressors and mental health is in the moderate to large range. In terms of strength of predictors, results are varied, and although it appears that life stressors have statistically stronger association with mental-health outcomes, there is mixed evidence that warrants future exploration.
In addition to the omnibus effect, we identified study-related and individual factors that did and did not explain some between-study heterogeneity. Our detailed sample-level tables summarize the variety of stressor- and mental-health measures and their psychometric properties that are used within the literature, which highlights gaps and shortcomings within individual research studies. In addition, we examine the partial associations of discrimination and life stressors on mental health, demonstrating that both are important and uniquely associated with Muslim mental health. Finally, our qualitative review of identity moderators contributes to the literature by summarizing the variation in the literature, and therefore, suggesting future directions.
When interpreting the findings, it is critical to consider that the measures used—for discrimination, life stressors, and mental health—were overwhelmingly based on self-report surveys. Accordingly, shared method variance due to the same reporter completing the assessments almost certainly led to an inflated correlation. The question is not if this is the case, but rather to what degree. Three pieces of evidence suggest that the associations are not entirely spurious. First, not all measures included in the current meta-analyses were based on self-report, and we did not find any differences in the effect size estimates when comparing correlations based on self-report and other types of reporting. Second, prior studies using both self-report (Lui, 2020) and genetically-informed designs (Cuevas et al., 2021) suggest that the association is not due to general negative emotionality, a leading potential confound. Third, Yap et al. (2014) found that ethnic identity—a construct that correlates with reported discrimination—still positively correlated with self-esteem and life satisfaction after accounting for shared method variance. This is admittedly tangential, but these topics that are well-studied in ethnic minority psychology rarely consider shared method variance, so this is at least useful information. To fully understand the association between discrimination and mental health, future studies need to make use of designs that take this concern into account, but doing so is not without its challenges (see Syed, 2021). In sum, even while acknowledging the need for future studies that address these limitations, current results increase to our understanding of the relationship between stressors and mental health for Muslims in western countries.
Advancing the Conceptualization of a Racialized Group and Widespread Islamophobia
A major contribution of this work is that our results support the assumption that Islamophobia is widespread, and that Muslim and “Muslim-like” groups appear to have similar experiences. Therefore, supporting previous research (e.g., Kaya, 2014; Selod, 2015), it seems that Muslim and Muslim-like groups have been racialized and may be viewed as a homogenous group—or at least have a shared racialized experience pertaining to discrimination. Muslim and Muslim-like ethnic groups both were found to have similar associations between stressors and mental-health outcomes. Moreover, ethnicity was not a significant moderator for the association between discrimination and mental-health outcomes, which suggests that deleterious associations of discrimination are present across ethnic groups, in turn, providing more evidence that Muslims are a racialized group. This provides support that individuals who may be considered to be Muslim—even if they do not religiously identify themselves as such—may also experience the negative associations of Islamophobia. Additionally, we found similar results between discrimination and mental-health outcomes across continents, highlighting that the associations between discrimination on mental health are widespread for Muslims living in Western countries.
However, there is some evidence to suggest that some demographic characteristics exacerbated the association between stress and mental-health outcomes. Specifically, some evidence shows that the association between life stressors and negative mental-health outcomes was stronger for Somali individuals compared to Middle Eastern individuals. This finding may suggest evidence of the compounding effect of being Black and Muslim, with intersecting marginalized identities exacerbating the severity of mental-health outcomes. Moreover, some evidence shows that refugee status is a significant moderator of the association between discrimination and negative mental-health outcomes, again suggesting the compounding effect of refugee status and the experience of discrimination. This aligns with the results of the interaction of discrimination and life stressors in predicting mental-health outcomes, specifically where life stressors exacerbated the effect of discrimination on psychological distress (Alemi & Stempel, 2018). Additionally, results suggest that the association between life stressors and mental-health outcomes is stronger in North America compared to Europe. Thus, this finding warrants further evaluation, but it may be due to the large number of Muslim refugee samples that have experienced cumulative trauma prior to resettlement in North America; or, it may be related to Muslims’ specific experiences living in North America (e.g., compounding effect of discrimination that is not measured, availability of resources).
Theoretical Considerations: Predictors of Negative Mental Health for Muslims
This study contributes to our understanding of important associations of negative mental health for Muslims living in Western countries. Briefly, the meta-analytic results suggest that discrimination and life stressors are significantly and uniquely associated with mental-health outcomes, as are experiences of personal and group-based discrimination. Additionally, evidence suggests that Muslims connected to organizations and Muslim youth in schools may be especially at-risk for experiencing the deleterious effects of life stressors. Importantly, as the studies included in these meta-analyses were nearly all correlational, we are not able to confirm explicitly causal claims on the relations between discrimination, life stress and mental-health outcomes. All available theoretical models postulate that experiences of discrimination and life stressors lead to worse mental health. This prediction is consistent with experimental studies on experiences of discrimination (e.g., Pascoe & Smart Richman, 2009), as well as longitudinal studies that have tested competing directions of association (e.g., English et al., 2014). To our knowledge, there are not any data that support the alternative causal relation, that individuals with worse mental health are more likely to report experiences of discrimination. That said, the causal assumptions are so strong in favor of discrimination leading to lower mental health that relatively few studies test alternatives, so the evidence base is not strong. Additionally, there is a mismatch between experimental studies, which induce discriminatory experiences in the lab, and correlational studies that are based on self-reports of life experiences. Chen & Mallory (2021) recently sought to address this problem by using propensity score weighting in a U.S. nationally representative sample. The results suggested a small causal impact of reported discrimination on indicators of physical and mental health. Thus, although we acknowledge we do not have evidence to make causal claims with the current data, the existing evidence falls in favor of the theoretical models that predict that discrimination and life stress leads to worse mental health. Nevertheless, future work should specifically seek to generate stronger evidence of the causal direction.
Mental-Health Outcomes. Experiences of discrimination and life stress are important contextual factors that can inform the development of psychopathology or our understanding of adaptive and maladaptive functioning (Causadias, 2013). Although we mainly use the umbrella term “mental-health outcomes,” we present mental-health outcome specific results in Table 6 and Table 15 that shows a) experiences of discrimination are most strongly associated with increased depression, and b) experiences of life stress are most strongly associated with somatization and PTSD. It appears that these different contexts of stress may be associated with mental-health outcomes differently. Additionally, negative mental-health outcomes (e.g., depression, anxiety) had a stronger association between discrimination and mental health compared to positive mental-health outcomes (e.g., self-esteem, wellbeing); although, both outcomes resulted in a statistically significant relation.
Discrimination and Life Stressors. An important contribution of this study is that both discrimination and life stressors were significantly associated with negative mental health. Additionally, the evidence from these meta-analyses suggest that discrimination cannot be subsumed by life stressors; instead, evidence suggests that experiences of discrimination are distinct, unique predictors and should be treated as such. Thus, our results align with Harrell’s (2000) model of racism-related stress that equally emphasizes both discrimination and general life stressors in the development of psychopathology. Harrell’s (2000) model separates different types of racism-related stressors (e.g., racism-related life events; vicarious racism experiences), and although we did not conduct analyses for all the different types of racism-related stressors, we did find that both individual and group-based discrimination were significantly associated with mental-health outcomes. Another conceptual model, specifically developed to capture the experiences of Americans of Middle Eastern and North African (MENA) descent, uses different terminology, but postulates the interaction of life-stress and discrimination experiences with identity-related factors on mental and physical health (Awad et al., 2019). Additionally, another framework—developed to understand the etiology of mental health outcomes for Black youth—describes culturally-informed Adverse Childhood Experiences (ACEs) that highlights the impact of racism in experiencing traumatic events (Bernard et al., 2020). Our meta-analytic results provide support for these different conceptual models in the development of psychopathology.
Notably, there is considerable variation in how discrimination and life stressors are measured in the literature. We were purposeful in including only measures that were captured by our definitions outlined a priori, and when necessary, we looked at the original measure article cited for further clarification of what was measured. Thus, another contribution of this study is that we did not include measures of discrimination or life stress that also included a distress rating (e.g., how much the stressor caused you distress), because that would have been conflated with negative mental health.
Personal and Group Discrimination. Furthering the systematic review by Samari and colleagues (2018), results suggest differential associations between personal and group discrimination and mental health, where personal discrimination is more strongly associated with mental-health outcomes. Importantly, group discrimination still has a significant association with mental-health outcomes. These results align with other meta-analytic results (Lui & Quezada, 2019; Schmitt et al., 2014) which also found that experiences of personal discrimination were more strongly related to mental-health outcomes more so than group-level discrimination. This may be due to the increase in ambiguity from group-based discrimination that allows an individual to rationalize that discriminatory action as not directed personally to them—resulting in utilizing different coping strategies than with personal discrimination.
At-Risk Populations. Results from the life-stressor dataset suggest that the most at-risk Muslim populations for negative mental-health outcomes are those connected to official organizations, compared to community samples. Official organizations tend to serve refugee and at-risk populations, which can help to understand the stronger association between experiences of discrimination and mental-health outcomes for samples from official organizations (e.g., Albdour, n.d.; Jorgenson, 2017). Individuals connected to official organizations may be more vulnerable to external stressors, as they do have the same resources as individuals from Western countries (e.g., Awad et al., 2019). Some evidence suggests that Muslim populations in schools may be at risk for experiencing the deleterious psychological effects of stressors, which may indicate that Muslim youth populations are at risk. The source of this risk among Muslim youth remains unclear, given that results do not reveal age to be a significant moderator of the relation between stressors and mental-health outcomes. However, previous research has found that Muslim youth may have some challenges integrating their Muslim and national identities (e.g., Sirin et al., 2008, 2008), which may limit coping strategies in the face of stress. Thus, these results highlight that it may be beneficial to target these Muslim populations to provide interventions and resources to support the impact of life stressors.
Explaining Between-Study Variance: Potential Moderators
Gaining insight to explain high heterogeneity or between-study variance was another goal of these meta-analyses, hence the examination of potential moderators. Overall, we did not identify many significant moderators, which may be due to a) not enough variation among the studies, or b) a shortcoming in the choice of potential moderators for which to search. In terms of study-related factors, both meta-analyses provide evidence to suggest that number of items in the stressor measure was a significant moderator, such that, as the number of items increased above the mean number of items, the strength of association between stressor and mental health increased. Therefore, evidence suggests number of items in the stress measure may impact the strength of the association between stressor and mental-health outcome. In terms of demographics, as discussed above, some evidence suggests refugee status (discrimination dataset), ethnicity (life-stressors dataset), and continent (life-stressors dataset) were significant moderators of the relation between stressor and mental-health outcomes.
Publication bias is an important consideration, as it is common for statistically significant results to be overrepresented in the published literature. Interestingly in the current study, publication status was a significant moderator only in the life stressors dataset, and the direction of its influence was opposite of what was expected. The association between life stressors and mental-health outcomes was actually weaker in published studies—in which larger effects seem more likely to appear—than in unpublished studies. This is likely because the unpublished studies included smaller samples (from more distressed populations) from organizations, schools and clinics, and these sample characteristics may have contributed to the stronger empirical associations and/or greater likelihood of Type 1 error. However, publication status did not appear to affect the meta-analytic estimates, likely because 83% of the samples were from published studies. Although the multiple analytical methods used to detect and correct for publication bias did not find evidence that the bias impacted the meta-analytic estimates, it is important to note that current analytic techniques for accounting for publication bias may not account for all sources of publication bias. Therefore, it is premature to confidently conclude that publication bias is not affecting the observed estimates.
We were only able to investigate identity in the discrimination dataset. The mixed results from the qualitative review parallel the results from a recent meta-analysis on the moderating role of ethnic identity on the association between discrimination and adjustment (Yip et al., 2019). We found that ethnic identity a) sometimes had no association, b) sometimes protected respondents, and c) other times exacerbated the negative effects of discrimination on mental health. Due to variation in measurement and analytic decisions, it was impossible to determine the patterns of which ethnic identity domains were protective vs. exacerbating for Muslims living in Western countries. The qualitative review of religious identity provided evidence that engaging in religious behaviors and having a stronger religious identity was protective for the deleterious effects of discrimination on mental-health outcomes. The other identity domains included only one to two studies, which makes an overall conclusion difficult as it relates to the moderating role.
Limitations
By conducting this study as a Registered Report, nearly all analytic decisions were made a prior and we received excellent feedback during the review process. Thus, our analytic strategy and decisions are transparent and not altered due to the findings. Nevertheless, the results from the current research should be interpreted in the context of the following limitations.
Two of the major limitations of this study are ones that we discussed previously, First, the studies included were overwhelmingly cross-sectional and correlational. The lack of longitudinal and experimental studies limits our ability to make causal claims. As discussed previously, we may assume that experiences of discrimination and life stress predict negative mental-health outcomes, but due to the correlational data, we cannot confirm the direction of the association. Second, measures in the discrimination dataset were overwhelmingly based on self-report. The measures in the life stress dataset were also overwhelming based on self-report (75% self-report measure or interview), but there were also parent-report measures, clinical interviews with professionals, and use of medical records. Although we did not find any differences in the estimates for self vs. other report, the shared method variance within most of the sample likely led to inflated correlations, but the degree to which this is the case remains unknown.
A major challenge, that had multiple implications, was that there were substantial differences in transparent reporting across studies. For example, some studies reported only significant correlations, and others did not provide the bivariate correlations for variables in multivariate models. We contacted authors for bivariate correlations if they were not included, but very few authors responded with the information. Therefore, we excluded the study or included only the correlations that were presented in the article, which may have led to biased estimates.
Additionally, there was variability in the way authors defined constructs and presented information, thus, making synthesis challenging. For example, researchers defined one of our moderators, generational status inconsistently (1st vs. 1.5 vs. 2nd generation). Moreover, there are differences between how European countries and the United States define the terms migrant vs. immigrant, which makes testing immigrant status challenging.
Similarly, researchers defined stress constructs inconsistently. The discrimination measures were extremely varied, such that some authors defined discrimination in terms of religious discrimination, and others defined it in terms of ethnic discrimination; however, it appeared that the two measures—religious discrimination and ethnic discrimination—measured the same construct. It would be helpful for future studies to investigate if individuals differentially experience religious discrimination vs. ethnic discrimination cognitively. The discrimination dataset included measures of Islamophobia that we categorized as group-based discrimination, but teasing apart group-based discrimination and Islamophobia warrants future exploration. There were similar problems for the life-stressor measures. For example, a common measure, TLEQ, only counts trauma that individuals say caused them fear, helplessness, or horror; this may explain the high correlation with PTSD symptoms. It is important to note that there were eight life-stress correlations that included a component of discrimination, which we included in the life-stressor dataset. We did not conduct a full review of every measure of life stress, and there is a possibility that measures of life stress also include items that measure the experience of discrimination. Both discrimination and life-stress measures also varied in terms of psychometric validity, and some authors used one item to measure experiences of stress, whereas others used validated multi-item measures. As mentioned, this difference appeared to suggest differences in association strength, as number of stress measure items was a significant moderator.
Overall, testing various moderators was also challenging due to inconsistent reporting. This resulted in small samples per group, which led to combining groups and creating unnecessary ‘noise’ in the data. Notably, to test the moderating role of identity, we used POMP scores to compare across studies. However, it would have been better if we used partial correlations, but that was not an option due to the lack of information in the studies.
These differences and inconsistencies also made it difficult to conduct the quality ratings as originally planned in the Registered Report. The articles provided varied information, and in many cases, the primary research question of an article was not related to the data we extracted. Thus, we modified the quality rating forms so that they were more relevant to our meta-analysis and also to highlight articles that were more or less transparent in terms of reporting.
Lastly, another important limitation of these meta-analyses is that many of the participant samples are from the same geographic areas. Although we excluded duplicate samples, we did include samples that were from the same geographic area; therefore, it is possible to have participant overlap. Common ethnic and geographic areas include: Somalis in Boston (three studies), Somalis in Minneapolis (three studies), Somalis in New England (two studies); Middle Easterners in Dearborn, Michigan (two studies), Iraqis in Detroit, Michigan (seven studies), and Arabs in Michigan (two studies). Even with these limitations, we believe our meta-analytic results further our understanding of the associations of discrimination and life stressors with mental-health for Muslims living in Western countries.
Clinical Considerations
Although not the primary focus of the current study, we found that experiences of discrimination and life stress are associated with different mental-health outcomes. A body of literature has examined the pathways in which different experiences may impact individuals differently (e.g., C. J. P. Harrell et al., 2011). Our results suggested that experiences of discrimination were more strongly related to depression, while experiences of life stress were more strongly related to somatization and PTSD. The specificity may be useful in deciding what clinical interventions may be useful for an individual experiencing discrimination or increased life stress. For instance, an individual experiencing discrimination and depression may benefit from acceptance and self-esteem interventions, whereas an individual experiencing life stress and somatization or PTSD may benefit from more mindfulness, mind-body, and cognitive processing of threat intervention. It may even be that certain interventions could be counter-indicated for certain experiences (e.g., the cognitive processing of threat might be counter-indicated for addressing discrimination).
It may be useful to better understand why individuals in North America have a stronger association between life stressors and mental health than individuals in Europe. It may be prudent to evaluate the resources available to Muslims, especially Muslim refugees, in North American compared to Europe. Our results suggested that there may be a gap that can be addressed to help support Muslims experiencing life stress in North America.
Conclusions and Future Directions
The current study begins to provide clarity that both discrimination and life stressors are distinctly associated with negative mental health. Our results align with Harrell’s (2000) model of racism-related stress and other theoretical models (e.g.,b Awad et al., 2019; Bernard et al., 2020) that emphasize both discrimination or cultural context and general life stressors as unique contributors in the development of psychopathology. Future work examining the impact of discrimination should also consider life stressors, and vice versa.
Our work emphasizes future directions with respect to open science practices, psychometric development, and new content areas. Researchers studying Muslim populations are encouraged to conduct preregistered analyses and to provide transparent reporting to further science and to improve our understanding of the experiences of Muslims in Western countries (see Kathawalla et al., 2021 for a tutorial on engaging with open science practices).
It would be beneficial for future studies to focus on methodological approaches that can elucidate our understanding of the relation between stress and mental health. First, longitudinal and experimental studies would provide evidence for causation in the relation between stress and mental health. It would be beneficial for studies to use non-self-report measures of discrimination, stress, and mental health for more objective measurement of the constructs. Using different methodology may also assist in our understanding of an individual’s cognitive experience and interpretation of various events. There is an important need for clear construct definitions for discrimination and life stressors and also related measures for those constructs.
Future work should examine Muslims’ experiences in other non-Western countries. The context of Islamophobia is not limited to Western countries, and Muslims are discriminated against across the world. For example, Muslim minority groups in China and India have been—and continue to be—persecuted. To further capture the experiences of Muslims, it would be helpful to also include research published in other languages, such as French and Arabic.
Our results suggest that future work should a) explore the compounding effect of life stressors and discrimination and b) to take more of an intersectional approach to see the compounding effect of: (1) race and Muslim identification or (2) refugee status and Muslim identification. Furthermore, our work suggests that Islamophobia is widespread and supports the notion that Muslim and Muslim-like groups have been racialized and may be seen as a homogenous group. Thus, future intervention work can target Muslims and Muslim-like groups, given that they appear to be similarly (and negatively) impacted by experiences of discrimination and life stressors. We hope this research can support the mental health of Muslims living in Western countries.
Contributions
Contributed to conception and design: UK, MS
Contributed to acquisition of data: UK, FC, EL
Contributed to analysis and interpretation of data: UK, MS
Drafted and/or revised the article: UK, MS
Approved the submitted version for publication: UK, MS
Acknowledgements
We would like to express our gratitude to Fei Bi Chan for her dedicated assistance in the data coding process and Elizabeth Lee for her contributions to data coding. We would like to thank Dr. Richard Lee, Dr. José Causadias, and Dr. Marti Gonzales for their comments on the project proposal and the finalized manuscript. We would also like to thank Amy Riegelman for her time and advice on refining search terms and databases, Casey Giordano for his advice regarding meta-analyses, Chris Hartgerink for his helpful comments and suggestions on the methods and analysis, and Linh Nguyen for meticulously checking that the output from the analyses matched the tables and figures.
Funding Information
This material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. (NSF 00039202) and The Society for the Psychological Study of Social Issues (SPSSI) Spring 2019 Grant-in-Aid Award.
Competing Interests
No competing interests exist.
Supplemental Material
Appendices
Appendix A. Search Terms.
Appendix B. Muslim-majority countries 2010 sorted by percent population Muslim.
Appendix C. Qualitative review of bivariate associations between stress type and mental-health outcomes for subset with measures of both discrimination and life stressors.
Supplemental Material
Please find the below listed materials on OSF: https://osf.io/enb97/
Supplemental Tables
Supplemental Table S1. Leaveout analysis for discrimination dataset.
Supplemental Table S2. Discrimination dataset categorical moderators (alternative model).
Supplemental Table S3. Discrimination dataset continuous moderators (alternative model).
Supplemental Table S4. Leaveout analysis for life stressor subset.
Supplemental Table S5. Life stressor subset categorical moderators (alternative model).
Supplemental Table S6. Life stressor subset continuous moderators (alternative model).
Supplemental Table S7. Additional individual sample and correlation details for discrimination subset.
Supplemental Table S8. Additional Individual sample and correlation details for life stressor subset.
Supplemental Table S9. Adjusted RVE Estimates of Discrimination and Life Stressors.
Supplemental Figures
Supplemental Figure S1. Forest plots of study effect sizes, RVE weights, and the overall RVE effect-size estimate for the discrimination dataset.
Supplemental Figure S2. Discrimination dataset outlier statistics.
Supplemental Figure S3. Influential samples from the discrimination dataset.
Supplemental Figure S4. Forest plot sorted by publication status for the discrimination dataset.
Supplemental Figure S5. Forest plot sorted by sample size for the discrimination dataset.
Supplemental Figure S6. Forest plots of study effect sizes, RVE weights, and the overall RVE effect-size estimate for the life stressors dataset.
Supplemental Figure S7. Life-stressor dataset outlier statistics.
Supplemental Figure S8. Influential samples from the life-stressor dataset.
Supplemental Figure S9. Forest plot sorted by publication status for the life-stressor dataset.
Supplemental Figure S10. Forest plot sorted by sample size for the life-stressor dataset.
Other Supplemental Materials
Reproducible Search Terms
Literature Search and Screening Tracking
Data Extraction Spreadsheet (password = datashare1234)
Data Extract Notes
Quality Rating Forms
Analysis Scripts
Disclaimer
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
Data Accessibility Statement
Dataset and analysis scripts are uploaded to the Open Science Framework (https://osf.io/enb97/).
Appendices
Appendix A. Search Terms
Search terms included a combination of discrimination terms: (discrimination) or (Islam#26phobi*) or (racism) or (prejudice) or (anti#Arab) or (anti#Muslim) or (anti#Islam); combined with OR, life stressor terms: (life stress*) or (stressor*) or (childhood adversit*) or (war trauma*) or (refugee*) or (ACEs) or (childhood trauma*) or (traumatic) or (life adversit*) or (abuse) or (natural disaster) or (violence) or (sexual assault); combined with AND, mental health terms: (mental health) or (mental illness) or (psychological well#being) or (psychological distress) or (depression) or (anxiety) or (PTSD) or (post#traumatic stress disorder) or (psychological adjustment) or (depressive symptom*) or (self-esteem) or (mood disorders) or (psychopathology) or (DSM diagnos*) or (somatization) or (psychotic experience*) or (well#being) or (psychological outcome*); combined with AND identifier of Muslim or Muslim-like identity terms: (Muslim*) or (Moslem*) or (Islam) or (Shia) or (Sunni) or (Arab*) or (Middle East*) or (North Africa*) or (MENA) or (South Asian*) or (Moroccan*) or (Afghani) or (Afghan) or (Tunisian*) or (Iranian*) or (Mauritania*) or (Tajikistani*) or (Yemeni*) or (Iraqi*) or (Jordanian*) or (Mayotte) or (Mahorais) or (Somali*) or (Turkish) or (Azerbaijani*) or (Maldivian*) or (Comorian) or (Nigerian) or (Algerian*) or (Palestinian*) or (Saudi Arabian*) or (Djibouti*) or (Libyan) or (Uzbekistani*) or (Pakistani*) or (Senegalese) or (Gambian*) or (Egyptian*) or (Turkmen) or (Syrian*) or (Malian*) or (Kosovan) or (Bangladeshi*) or (Indonesian*)
Appendix B. Muslim-majority countries 2010 sorted by percent population Muslim.
Morocco: 99.9
Afghanistan: 99.8
Tunisia: 99.8
Iran: 99.7
Western Sahara: 99.6
Mauritania: 99.2
Tajikistan: 99
Yemen: 99
Iraq: 98.9
Jordan: 98.8
Mayotte: 98.8
Somali: 98.6
Tukey: 98.6
Azerbaijan: 98.4
Maldives: 98.4
Comoros: 98.3
Niger: 98.3
Algeria: 98.2
Palestine: 97.5
Saudi Arabia: 97.1
Djibouti: 97
Libya: 96.6
Uzbekistan: 96.5
Pakistan: 96.4
Senegal: 95.9
Gambia: 95.3
Egypt: 94.7
Turkmenistan: 93.4
Syria: 92.8
Mali: 92.4
Kosovo: 91.7
Bangladesh: 90.4
Kyrgyzstan: 88.8
Indonesia: 88.1
Oman: 87.7
Kuwait: 86.4
Guinea: 84.2
Albania: 82.1
Bahrain: 81.2
Qatar: 77.5
United Arab Emirates: 76
Sierra Leone: 71.5
Sudan: 71.4
Malaysia: 61.4
Lebanon: 59.7
Burkina Faso: 58.9
Kazakhstan: 56.4
Chad: 55.7
Brunei: 51.9
Data retrieved from PewResearchCenter (2011). The future of the global Muslim population, projections for 2010-2030. Pew-Templeton Global Religious Futures Project. https://www.pewresearch.org/wp-content/uploads/sites/7/2011/01/FutureGlobalMuslimPopulation-WebPDF-Feb10.pdf
Appendix C. Qualitative review of bivariate associations between stress type and mental-health outcomes for subset with measures of both discrimination and life stressors.
There was no consensus found on the association between stress type and mental health after reviewing the eight samples that included measures of discrimination and life stressors (but did not report adjusted correlations). Three samples found that both discrimination and life stressors had positive associations with mental-health outcomes (East et al., 2018; Gorst-Unsworth & Goldenberg, 1998; Lindencrona et al., 2008); three samples found varied associations between stressor and mental-health outcomes (Stevens et al., 2005a, 2005b; Stevens & Thijs, 2018; Tinghög et al., 2017(s1); Hellemans et al., 2015); and, two samples found that neither discrimination nor life stressors were significantly related to mental-health outcomes (Hassouneh & Kulwicki, 2007; Tinghög et al., 2010(s2)).
Three samples that included measures of discrimination and life stressors found that both discrimination and life stressors had small to large positive associations with mental-health outcomes. Both discrimination (r = 0.67) and life stressors (peer bullying, r = 0.67; traumatic experiences, r = 0.19) were associated with worse mental-health outcomes in a sample of Somali youth (East et al., 2018). Additionally, in a sample of Middle Eastern refugees living in Sweden, discrimination and trauma were significantly associated with psychological distress (discrimination, r = 0.29; life stressors, r = 0.31) and PTSD (discrimination, r = 0.21; life stressor, r = 0.27; Lindencrona et al., 2008). In a sample of Iraqi refugees living in the United Kingdom, discrimination (r = 0.32) and life stressors (r = 0.30) were associated with depression in a similar way (Gorst-Unsworth & Goldenberg, 1998).
Three samples that included measures of discrimination and life stressors found that both discrimination and life stressors were differentially associated with mental-health outcomes depending on life-stressor type, discrimination type, or mental-health outcome. In a sample of Moroccan youth in the Netherlands, both life stressors and group discrimination were differentially associated with externalizing symptoms (discrimination, r = 0.12–0.22; life stressors, r = 0.12– 0.18) and internalizing symptoms (discrimination, r = -0.07–0.16; life stressors, r = 0.14–0.15; Stevens et al., 2005a, 2005b; Stevens & Thijs, 2018. The large variation for the association of group discrimination and mental-health outcomes is due to two different measures of group discrimination (Stevens et al., 2005a, 2005b; Stevens & Thijs, 2018). Group discrimination was consistently a predictor of externalizing problems but not internalizing problems (Stevens et al., 2005a, 2005b; Stevens & Thijs, 2018). Personal discrimination was significantly associated with externalizing symptoms (r = 0.19–0.24) and differentially associated with internalizing symptoms (r = 0.07–0.15; Stevens et al., 2005a, 2005b; Stevens & Thijs, 2018). Group discrimination was significantly associated with self-esteem (r = -0.11), but personal discrimination was not significantly associated with self-esteem (r = -0.03; Stevens et al., 2005a, 2005b; Stevens & Thijs, 2018). Comparably, in a sample of Turkish adults living in Flanders, physical abuse (r = 0.06) did not have a significant bivariate association, but psychological abuse (r = 0.19, p < .01) had a significant bivariate association with worse mental-health outcomes (Hellemans et al., 2015). No comparable analyses were conducted between racial discrimination and mental-health outcomes (Hellemans et al., 2015). In a sample of Iraqi refugees in Sweden, discrimination was associated with internalizing symptoms with a small effect (r = 0.11, p = NR; Tinghög et al., 2010 (s1)). The bivariate correlation between life stressors and internalizing symptoms could not be extracted, but people with internalizing disorders reported a significantly higher number of traumatic events compared to those without the mental health diagnosis (3.55 vs. 2.64; Tinghög et al., 2017(s1)).
Two samples found that neither discrimination nor life stressors had bivariate associations with mental health. In a sample of Palestine adults living in the United States, discrimination nor life stressors were significantly associated with depression (discrimination, r = 0.06–0.31; life stressors, r = -0.01) or anxiety (discrimination, r = 0.21; life stressors, r = -0.04; Hassouneh & Kulwicki, 2007). Similarly, in a sample of Iranian refugees in Sweden, the bivariate association between discrimination and internalizing symptoms was near to zero (r = 0.05, p = NR; Tinghög et al., 2010 (s2)). The bivariate correlation between life stressors and internalizing symptoms could not be extracted, but people with internalizing disorders reported a significantly higher number of traumatic events compared to those without the mental health diagnosis (2.53 vs. 2.17; Tinghög et al., 2017).
Footnotes
The term discrimination includes racial, ethnic, and religious discrimination.
To note, Mölsä and colleagues (Exec. Order No. 13769, 2017) is not included in the meta-analysis as we were unable to extract a bivariate correlation from the data and authors did not respond to our email request for the data and/or the bivariate correlations.
Cohen’s kappa was not used as kappa is difficult with a moving denominator, so percent agreement makes more sense (Syed & Nelson, 2015).
In our Stage 1 Registered Report we said we would use the TOSTER function by Lakens (2017), but the function is used to compare correlations to equivalence bounds instead of comparing correlations to each other.
Decided a priori, 10 studies were needed for analyses. Therefore, the analyses presented of partial correlations are exploratory.
From the 14 samples, only five included statistics required for a meta-analysis; therefore, we conducted a qualitative review.
There were 11 samples in the analytical sample and 14 in qualitative review as three samples only presented one of the two correlations. One article only presented the correlation for life stressors and mental health and did not include the correlation for discrimination and mental health. One other article (with two unique samples) only presented the correlation for discrimination and mental health and did not include the correlation between life stressors and mental health. The authors were contacted for the bivariate correlations, but they did not provide the information needed.
Percentages do not match with Table 5 as samples included different types of measures that can be analyzed separately at times and other times combined.
Number of samples do not match Table 6 due to different ways of categorizing mental-health outcomes. In order to increase the number of samples per mental-health outcome to conduct RVE analyses (see Table 6), outcomes were approximately categorized using the Hierarchical Taxonomy of Psychopathology dimensions (Kotov et al., 2017).
Mental-health measure moderators are unregistered exploratory analysis.
kc = number of outcomes or correlations
These analyses are unregistered exploratory analysis.
Mental health measure moderators are unregistered exploratory analysis.
The results for religious identity are exploratory as the analysis is underpowered (k = 8). Table S3 includes RVE analyses for both ethnic identity and religious identity that shows consistent nonsignificant results.
Number of samples do not match Table 14 as stressors were grouped differently for analyses.
Number of samples do not match Table 15 due to different ways of categorizing mental-health outcomes and the number of samples that measured multiple mental-health outcomes. When necessary, mental-health outcomes were approximately categorized using the Hierarchical Taxonomy of Psychopathology dimensions (Kotov et al., 2017).
Mental-health measure moderators are unregistered exploratory analysis.
Analyses with fewer than 10 effect size per group are exploratory.
We conducted unregistered exploratory moderation analysis between samples from North American and Europe and found continent was a significant moderator, Qbetween (1) = 5.8971, p = .02.
14 unique samples are included in the qualitative review as there were three samples that we could not extract correlation data for both stressors.
We did not use the TOST package in R (Lakens, 2017) and instead wrote our own function that can be found in our R file.
We did not use equivalence tests comparing dependent correlations outlined in Counsell & Cribbie (2015) as we were only able to extract the correlations between discrimination and life stressors from six samples, which is under our predetermined minimum of 10 studies needed to conduct a meta-analytic estimate.
Statistics calculated from odds ratios.
Statistics calculated from odds ratios.
The inclusion of ‘#’ represents a wildcard that accounts for alternative spellings or phrasing of the same word.