Does Mindfulness Improve Intergroup Bias,Internalized Bias,and Anti-Bias Outcomes?: A Meta-Analysis of the Evidence and Agenda for Future Research

Intergroup Bias

Dovidio and Gaertner (2010) define intergroup bias as “an unfair evaluative, emotional, cognitive, or behavioral response toward another group in ways that devalue or disadvantage the other group and its members either directly or indirectly by valuing or privileging members of one’s own group” (p. 1084). Thus, embedded in the concept of intergroup bias is an ethical position, specifically that our evaluations and responses toward ingroup and outgroup members should be balanced, equitable, and nonharming.

Bias has typically been examined in terms of stereotypes (overgeneralized beliefs), prejudice (biased attitudes/affect), and discrimination (unfair treatment) (Dovidio & Gaertner, 2010). Stereotypes refer to those qualities, such as social roles and traits, that are perceived to distinguish particular categories of individuals. When stereotypes are activated, individual group members are judged by the cognitive representations of that group, which include trait associations (e.g., Asians are good at math) as well as affective reactions to the group (i.e., low warmth) (Cuddy et al., 2008). Prejudice refers to biased attitudes, which contain cognitive and affective elements, as well as behavioral inclinations to respond in particular ways toward an outgroup (Dovidio & Gaertner, 2010). As prejudice has shifted in modern times to be more subtle and less overtly hostile, it may manifest as aversive or avoidant reactions, policy views that would adversely affect the outgroup (e.g., “colorblind” policies), and implicit associations.

Whereas stereotypes and prejudice are conceptualized as intrapsychic phenomena, spanning cognitive and affective dimensions, discrimination refers to unfair treatment of individuals because of their group membership. Discrimination may include direct and indirect actions that cause harm, including violence, hostility, negligence, and exclusion, as well as less positive behavior or consideration than would be directed toward an ingroup member. Examples include failing to help someone in need, devaluing the qualifications of a job candidate, or ostracizing someone from one’s social network because of their race or some other characteristic.

In their seminal work, Cuddy et al. (2007) proposed the Behaviors from Intergroup Affect and Stereotypes (BIAS) map, which models a set of predictable relationships between cognitive (stereotypes), affective (emotional prejudice), and behavioral (discrimination) dimensions of bias. Empirical tests of the BIAS map confirm that while both stereotype content and emotions predict particular behavioral action tendencies, emotions (Esses & Dovidio, 2002; Esses et al., 1993) mediate the relationship between stereotypes and action tendencies (Bye & Herrebrøden, 2018; Cuddy et al., 2007). In other words, stereotype content activates certain emotions, and those emotions lead to certain behavioral tendencies (to actively or passively facilitate or harm the target group’s goals). These findings are consistent with other studies suggesting that affect often predicts discriminatory behavior better than stereotypes (e.g., Esses & Dovidio, 2002; Stangor et al., 1991). Taken together, research confirms the distinctive but related influence of cognitive and affective dimensions of bias in predicting discriminatory behaviors, suggesting distinct targets for intervention. For example, given the more direct influence of emotions on discriminatory behavior, improving emotional self-regulation in response to stereotype activation may serve to weaken the emotion–behavior link. While less closely associated with behavior, cognitive dimensions of bias (e.g., stereotypes, beliefs, attitudes) also predict discriminatory intentions, creating what Talaska et al. (2008) propose as a “conscious ideological system.” Therefore, strengthening self-awareness and cognitive flexibility also may contribute to decreases in explicit and implicit attitudes and greater behavioral intentions to challenge bias and discrimination in self and others (Lillis & Hayes, 2007). To quantify the relative strength of associations with these different bias components, we draw on the Cuddy et al (2007) BIAS framework to separately explore the effects of mindfulness on the cognitive, affective, and behavioral dimensions of bias (see Figure 1).

Effects of Bias on Targeted Groups

Whereas much of the research on intergroup bias focuses on ways to reduce bias among higher-status groups, less is known about the experiences and impact of bias from the perspective of the target group members themselves (Fox et al., 2018; Pyke, 2010). The sociological concept of stigma describes the status loss that occurs when particular individuals or groups become designated as marked, tainted, or morally polluted as a result of some socially salient attribute (Goffman, 1963; Link & Phelan, 2001). Examples of stigmatized statuses in North America include being a member of a racial, ethnic, sexual, gender, or religious minority; being elderly, overweight, homeless, or having an illness or disability (e.g., mental illness, HIV/AIDS). Those with the power to define and reinforce ideas of differentness and abnormality systematize the process of devaluation into norms, practices, and policies (Crocker et al., 1998; Link & Phelan, 2001). In this way, bias at the interpersonal level is reinforced by and contributes to systematic bias in the institutions and structures of society, with harmful consequences for targeted groups.

For example, chronic exposure to racist stereotypes, prejudice, and discrimination is a feature of everyday life for Black Americans and is associated with low educational achievement, poverty, poor health, police shootings, incarceration, and decreased life expectancy (Mesic et al., 2018; Morris & Perry, 2016; Williams et al., 2019). Over time, they may come to internalize ideas about themselves as inferior and accept racial inequities as deserved (David et al., 2018; Pyke, 2010), resulting in feelings of self-hatred and shame, devaluation of one’s culture of origin and idealization of whiteness (Fanon, 1963; Freire, 1970b).

Like bias itself, internalized bias has been operationalized in terms of its cognitive, affective, and behavioral components. Examples include endorsement of negative attitudes, beliefs, and judgments about one’s own (stigmatized) group; negative affect related to being a member of the group; stereotype-consistent behaviors, discriminating against members of their own group, and supporting institutional policies that maintain inequality (Corrigan et al., 2006; David et al., 2018; Lillis et al., 2010; Steele & Aronson, 1995). However, most studies focus on its cognitive and affective dimensions. For example, the widely used Weight Bias Internalization Scale (Durso & Latner, 2008) assesses respondents’ endorsement of negative views about persons with overweight and obesity (e.g., “I am less attractive than most other people because of my weight”). Studies also typically find that while marginalized groups view their own group more favorably than do dominant groups, they still evidence more positive implicit associations to the (dominant) outgroup, with some exceptions (e.g., Black Americans) (Nosek et al., 2007).

Although internalized bias is detrimental to the mental and physical well-being of marginalized groups (Graham et al., 2016; Hatzenbuehler, 2009; Hwang, 2021; Livingston & Boyd, 2010; Turan et al., 2017), few interventions have been designed specifically to address it. As research has predominantly focused on reducing intergroup bias and stigma among dominant group members (e.g., White college students), centering the experiences of marginalized groups emphasizes both the deleterious psychological effects of bias and oppression as well as intragroup efforts to resist, confront, and heal from it (David, 2013; French et al., 2020; Ma et al., 2019). For these reasons, in the present review and meta-analysis, we broaden our definition of bias outcomes to include both intergroup bias (directed toward outgroup members) and internalized bias (directed toward ingroup members), with a shared focus on their cognitive, affective, and behavioral components (see these dimensions in Figure 1).

From Bias to Anti-Bias

A second distinctive aspect of our approach is to conceptualize bias on a continuum from bias to anti-bias. This view is consistent with critical race scholars who argue that the antidote to racism is not simply race-neutrality or colorblindness, but rather anti-racism (Delgado & Stefancic, 2001; Kendi, 2019). Racial colorblindness, which provides justification for practices that maintain racial inequities, such as race-blind admissions, police brutality, and housing discrimination, has been found to be associated with internalized oppression (Neville et al., 2005) and less engagement in anti-racist actions (Yi et al., 2022). In contrast, an anti-racist position may include awareness that White supremacy and other systems of oppression shape individuals’ life outcomes, and behaviors aimed at disrupting racist ideas and policies to promote racial equity. We extend this basic premise to other forms of bias and anti-bias—including that related to gender identity, sexual orientation, social class, nativity, ability, body size, religion, and illness.

For minoritized groups, becoming aware of and resisting biased and inferiorizing messages from society has been proposed as a crucial component of well-being (David, 2013; French et al., 2020). The influential Brazilian educator Paulo Friere advanced the concept of critical consciousness (CC) or conscientization, which describes the process whereby marginalized or oppressed groups learn to identify and analyze the structural forces (such as racism, classism, and sexism) that produce inequitable social conditions (critical reflection) and take action to change them (critical action) (Freire, 1970a). Resistance to the dominant messages that normalize one’s subordination provides a pathway to reclaiming one’s humanity and changing society. Thus, like bias itself, the concept of critical consciousness encompasses cognitive (awareness), affective (self-compassion), and behavioral (action) components. Among diverse minoritized groups, critical consciousness has been associated with positive psychological, educational, and health outcomes and the reduction of negative consequences associated with oppression (see Jemal, 2017, for a review). Other intervention frameworks for internalized bias focus on the development of self-love, self-compassion, positive ethnic identity, and collective resistance (David et al., 2018; French et al., 2020). Therefore, in addition to examining outcomes related to implicit and explicit cognitions, affect, and behavior regarding stigmatized groups, we also include studies that target outcomes related to anti-bias/ anti-oppressive and inclusive cognitions, emotions, and behaviors (Figure 1). Examples include intercultural attentiveness, sensitivity, engagement, and respect for cultural differences (Chen, 2000), multicultural counseling competence (Holcomb-McCoy & Day-Vines, 2004), and prosocial actions to benefit marginalized groups specifically (e.g., Ashar et al., 2016).

Mindfulness

Given the widespread harm that bias causes to marginalized groups, determining how best to decrease bias and increase anti-bias efforts is a public health priority. In recent years, researchers have begun to explore the possible benefit of mindfulness for reducing bias and promoting anti-bias outcomes. Mindfulness is a term with different meanings in Buddhist and contemporary contexts, but in a basic sense refers to “a self-regulated attentional stance oriented towards present-moment experience that is characterized by curiosity, openness, and acceptance” (Dahl et al., 2015, p. 515). Mindfulness has been operationalized in numerous ways, including as (a) a naturally occurring dispositional trait (e.g., Baer et al., 2006; Brown & Ryan, 2003; Lau et al., 2006), (b) an attentional skill, developed through meditation, of awareness of what is arising moment-to-moment in sensory experience, and (c) a group of meditative practices used to develop a set of skills or qualities to reduce self-processing biases, reduce suffering, and cultivate a healthy mind (Vago & Silbersweig, 2012).

Most studies applying mindfulness to the problem of bias are informed by Buddhist philosophical understandings of mindfulness and its cultivation through different forms of meditation (see Dahl et al., 2015; Lutz et al., 2007). Focused attention (FA) practices involve training one’s attention on a sensory target (e.g., breath) or other object (Dahl et al., 2015). As the task requires noticing when one’s mind has wandered, FA practices cultivate attentional control and self-monitoring skills (Lutz, Slagter, et al., 2008; Vago & Silbersweig, 2012). For example, in Lueke and Gibson’s (2015) study examining the effects of mindfulness on implicit age and race bias, participants listened to a 10-min mindfulness recording that guided them to become aware of any physical sensations they were experiencing in the present moment and to accept those sensations without trying to alter, resist, or judge them. A foundational practice, FA is frequently employed in empirical studies of mindfulness and bias, alone or in combination with other techniques. Building on the attentional control cultivated by FA practices, open-monitoring (OM) practices enhance the person’s ability to monitor dimensions of experience (thoughts, feelings, etc.) with less judgment, discernment, or reactivity (Dahl et al., 2015). OM practices often begin with FA but then broaden the focus of attention while emphasizing the monitoring aspect of awareness. For example, the mindfulness condition in Berry et al.’s (2018) study of prosociality toward ostracized strangers consisted of listening to an 8½-min recording of a guided practice to induce a state of attention to moment-to-moment somatic, cognitive, and emotional experiences (adapted from Segal et al., 2002).

Finally, others have studied the effects of constructive practices, or what Vago and Silbersweig (2012) refer to as “ethical enhancements” to mindfulness interventions. These practices typically begin with foundational training in mindful attention and acceptance of present-moment experience, before advancing to the active cultivation of prosocial emotions, such as lovingkindness, empathy, and compassion (Dahl et al., 2015; see Galante et al., 2014, for a review) that may be specifically directed toward marginalized groups (e.g., Kang et al., 2014). Finally, clinical researchers also have tested the effects of multi-component mindfulness-based interventions (MBIs), such as adaptations of Mindfulness-based Stress Reduction (MBSR) and Acceptance and Commitment Therapy (ACT), that integrate various aspects of mindfulness training with other contemplative, psychological, or relational methods to specifically address intergroup bias and internalized bias (e.g., Lillis & Hayes, 2007; Skinta et al., 2015; Taylor et al., 2016). As applications of MBIs to the problem of bias are in its early stage, this review includes both types of studies (e.g., “mindfulness-only” and multi-component, “mindfulness-plus” studies) and compares results in moderator analyses to identify promising areas for future study.

How Mindfulness May Reduce Bias

In Vago and Silbersweig’s (2012) widely cited integrative conceptual framework, mindfulness is proposed to reduce biases and distortions in three main ways: by developing meta-awareness of one’s moment-to-moment experience without judgment or reactivity (self-awareness), increasing one’s ability to manage emotions and responses in adaptive ways (self-regulation), and cultivating a prosocial orientation toward others that transcends personal needs and desires (self-transcendence). The self-processing focus of their model is supported through converging evidence of the benefits of mindfulness and its underlying neurobiological mechanisms (Hölzel et al., 2011). Experimental, phenomenological, and clinical studies on mindfulness describe a range of cognitive and psychological benefits including improvements in self-awareness, cognitive flexibility, attentional functioning, emotional and behavioral regulation (Hodgins & Adair, 2010; Moore & Malinowski, 2009). These outcomes are mirrored by mindfulness-induced changes in brain regions involved in attention, learning and memory, self-referential processing, perspective-taking, and emotion regulation (for a review, see Tang et al., 2015). Beyond the individual mind, brain, and body, mindfulness also has been shown to have significant interpersonal benefits (Berry et al., 2020; Donald et al., 2019). We predict that these cognitive, affective, and behavioral benefits of mindfulness will extend to the bias domain to reduce biases directed toward marginalized outgroup members as well as biases directed toward the self (internalized bias). This leads to our first prediction:

The Present Study

In this review, we draw on prior theorizing (e.g., Cuddy et al., 2007; David, 2013; Dovidio & Gaertner, 2004; French et al., 2020; Shapiro et al., 2006; Vago & Silbersweig, 2012) to present an integrative three-dimensional conceptual framework for exploring the effects of mindfulness on (a) manifestations of bias at the level of implicit attitudes, explicit attitudes, affect, and behavior; (b) bias target, whether members of an outgroup (bias) or ingroup (internalized bias); and (c) intergroup orientation toward both bias or anti-bias. We make the case for broadening the scope of attention beyond bias enacted by dominant groups to include internalized bias and anti-bias outcomes and their cognitive, affective, and behavioral expressions. Second, we meta-analyze the associations between mindfulness and cognitive, affective, and behavioral dimensions of bias and anti-bias to identify which outcomes are most influenced by mindfulness as a trait and as an intervention. We also examine potential moderators, including features of the sample, bias target, and study design.

Finally, as this field is still in its early stages of development, we contribute an evidence and gap map to catalog how mindfulness has been studied in relation to our three-dimensional framework and identify gaps in need of further investigation. Thus, in addition to testing our separate hypotheses regarding the pooled and separate associations between mindfulness (as a trait and as an intervention) on implicit attitudes, explicit attitudes, affect, and behavior, we also seek to examine two exploratory research questions:

Methods

Procedures, including definitions of key constructs, search terms and strategy, inclusion and exclusion criteria, data extraction procedures, risk of bias assessment, and analytic plan, were preregistered with PROSPERO International Prospective Register of Systematic Reviews after piloting the search criteria (Chang et al., 2018). The datafile, codebook, and script used for analyses are available on the Open Science Framework Online Repository at https://osf.io/w9yxu/?view_only=8122f5fb70de4f87927a80edec60b631.

Data Collection Process

Two reviewers independently extracted data from the 66 eligible records. Any disagreements were resolved via discussion among the first, second, and final authors. Study screening was conducted using the Covidence software (www.covidence.org). In cases where there was insufficient data provided to calculate effect sizes, the authors were contacted with a request to provide the missing data. Failure to respond or inability to provide additional data resulted in the additional exclusion of two studies. Figure 2 is the PRISMA diagram, showing the data screening process, which resulted in the retention of 72 eligible studies from 64 records. This includes 42 intervention studies (n= 41 records), and 30 correlational studies (n = 26 records), with three reports including both intervention and correlational studies.

OPEN IN VIEWER

Figure 2.

PRISMA Flow Diagram.

Results

Study Characteristics

Of the 70 eligible studies from 62 reports included in the meta-analysis, we identified 42 studies (178 effect sizes; N = 3,229 participants) that measured the effect of an MBI on indices of bias and anti-bias; six were unpublished studies. We identified another 30 correlational studies (150 effect sizes; N = 6,002 participants) that examined the relationship between self-reported mindfulness, bias, and anti-bias outcomes. Twelve studies were unpublished as of January 2020 when we completed the search; however, two have since been published, specifically Berry et al. (2021) (from Berry, 2017) and Kang and Falk (2020). Our search did not identify any longitudinal studies that met our inclusion criteria. The characteristics of the 70 included studies are summarized in Supplemental Material and shown in Tables S1 and S2.

Table 1 presents our model-based evidence and gap map, specifically the studies and number of effect sizes for each of the categories in our theoretical model in Figure 1 (this addresses RQ2). For bias directed toward outgroups, the number of effect sizes were respectable for outcomes focusing on attitudes, affect, and behavior, except there were only two effect sizes for implicit attitudes. For bias directed toward ingroups, there were no effects for either internalized bias or anti-bias for either implicit attitudes or behavioral measures. The number of effect sizes for internalized bias for explicit attitudes and affect were respectable to proceed with a meta-analysis.

OPEN IN VIEWER

Table 1.

Model-Based Evidence and Gap Map.

	Bias directed toward outgroup
	Attitudes		Affect	Behavior
Intergroup orientation	Implicit	Explicit
Bias
Number of studies	15	24	5	5
Number of effect sizes	39	114	13	9
Anti-bias
Number of studies	1	17	6	6
Number of effect sizes	2	95	29	19
	Bias directed toward ingroup
	Attitudes		Affect	Behavior
	Implicit	Explicit
Internalized bias
Number of studies	0	10	7	0
Number of effect sizes	0	24	13	0
Anti-bias
Number of studies	0	2	4	0
Number of effect sizes	0	2	5	0

Intervention Studies

Primary Analyses

In Table 2 and Figures 3 and 4, we show the results of the meta-analyses of the 42 intervention studies (within 40 separate records) included in this review. As noted, following recommendations by Cheung, (2014b), we used “record ID” as the clustering variable in all meta-analyses. Consistent with Hypothesis 1, and as Table 2 shows, there was a medium-sized negative effect of mindfulness interventions on intergroup bias outcomes (g = −0.56, CI 95% [−0.72, −0.40]). To test Hypothesis 2, the extent to which mindfulness influences bias along the four dimensions of attitudes (implicit and explicit), affect, and behavior, we conducted subset analyses. We calculated the meta-analytic effect of mindfulness interventions separately for each of these four dimensions of intergroup bias. Consistent with Hypothesis 2, mindfulness interventions produced a small-to-medium sized negative effect on implicit attitudes (H2(a); g = −0.35, CI 95% [−0.51, −0.19]), a medium-sized, negative effect on explicit attitudes (H2(b); g = −0.51, CI 95% [−0.68, −0.34]) and affect (H2(c); g = −0.55, CI 95% [−0.74, −0.36]), and a large negative effect on behaviors that reflect intergroup bias (H2(d); g = −0.90, CI 95% [−1.53, −0.26]). These findings, and the effect sizes from the studies underpinning them, are shown in Figure 3A and 3B.

OPEN IN VIEWER

Table 2.

Meta-Analytic Effects and Moderation Tests for Mindfulness Intervention Studies on Bias Outcomes.

Outcome types	k	ES	g [95% CI]	S E	R ( 2 ) 2	R ( 3 ) 2	p
All outcomes (H1) ( I ( 2 ; 3 ) 2 : 0.39; 0.48)	40	178	–0.56 [–0.71, –0.41]	0.08	—	—	—
Subset analyses
Implicit attitudes (H2a)	10	24	–0.35 [–0.51, –0.19]	0.08	—	—	—
Explicit attitudes (H2b)	22	100	–0.51 [–0.68, –0.34]	0.09	—	—	—
Affect (H2c)	14	32	–0.55 [–0.76, –0.34]	0.10	—	—	—
Behavior (H2d)	9	22	–0.90 [–1.53, –0.26]	0.32
Moderation analyses—Hypotheses
Controlled vs. implicit bias (H3)	40	178	—	—	.000	.047	.707
Bias target (outgroups vs. the self) (H4)	40	178	—	—	.000	.064	.215
Moderation analyses—Exploratory
Minority group
Racial/ethnic minorities	40	178	—	—	.001	.030	.373
Clinically stigmatized groups	40	178	—	—	.000	.005	.700
Other bias target groups	40	178	—	—	.000	.061	.271
Study population	40	178	—	—	.000	.160	.057^
Helping professionals	7	33	–0.26 [–0.40, –0.12]	0.07	—	—	—
General population	34	145	–0.63 [–0.82, –0.45]	0.09	—	—	—
Gender	40	178	—	—	.008	.000	.431
Study design	40	178	—	—	.00	.15	.032*
RCTs	30	117	–0.52 [–0.71, –0.34]	0.10	—	—	—
Non-RCTs	10	61	–0.66 [–0.96, –0.37]	0.15	—	—	—
Intervention type			—	—	.00	.00	.916
“Mindfulness-only” interventions	16	52	–0.57, [–0.91, –0.23]	0.17	—	—	—
“Mindfulness-plus” interventions	25	126	–0.54, [–0.69, –0.39]	0.08	—	—	—
Intervention duration in minutes (log)			—	—	.00	.00	.745
Measurement lag			—	—	.01	.00	.415
Risk of bias			—	—	.00	.053	.195

Note. k = number of records (i.e., articles and dissertations); ES = number of effect sizes; g = Hedge’s g; CI = confidence interval; R²₍₂₎ = within-record variance in the pooled effect explained by the moderator; R²₍₃₎ = between-record variance in the pooled effect explained by the moderator.

*

p < .05.

OPEN IN VIEWER

Figure 3.

(A) Forrest Plot of Effects of Mindfulness Interventions on Biased Implicit Attitudes; (B) Forrest Plot of Effects of Mindfulness Interventions on Biased Explicit Attitudes.

OPEN IN VIEWER

Figure 4.

(A) Forrest Plot of Effects of Mindfulness Interventions on Biased Affect; (B) Forrest Plot of Effects of Mindfulness Interventions on Biased Behavior.

We next tested Hypothesis 3, that mindfulness interventions would have larger effects on forms of controlled bias (i.e., explicit attitudes, emotions, and behavior) than implicit attitudes. We did not find evidence for this, however (R²₂ = 0.00; R²₃ = 0.047; p = .71). The pooled effect for controlled forms of bias was large in magnitude (g = −0.62, CI 95% [−0.82, −0.43]), while the pooled effect for implicit attitudes was moderate in size (g = −0.35, CI 95% [−0.51, −0.19]); however, the confidence intervals surrounding these effects overlapped.

Finally, we tested Hypothesis 4, that mindfulness interventions would have similar-sized effects on bias toward outgroups relative to their effect on internalized bias. We found support for this prediction, with no evidence that effects of mindfulness interventions differed between these two different forms of bias (R²₂ = 0.00; R²₃ = 0.065; p = .215). Effects for both sets of targets were of a similar, large magnitude (for outgroup bias, g = −0.51, CI 95% [−0.71, −0.33]; for internalized bias, g = −0.70, CI 95% [−0.88, −0.51]).Notably, the confidence intervals of the two pooled effects overlap substantially, which increases our confidence in H4. As an additional step, we conducted sensitivity analyses, with one study by Langer et al. (1985) excluded from the analysis. Mindfulness interventions such as the one conducted by Langer et al. (1985) have been argued to be conceptually somewhat different from more Buddhist conceptualizations of mindfulness (e.g., Brown, Ryan, & Creswell, 2007). Excluding this study did not alter the pattern of meta-analytic findings (H1, among all bias outcomes, g = −0.55, CI 95% [−0.71, −0.40]; H2, among biased behaviors, g = −0.85, CI 95% [−1.47, −0.24]; and for H3 and H4, there was no evidence of moderation).

Moderation Analyses

We identified moderate levels of both within- and between-study heterogeneity for the overarching meta-analytic effect of mindfulness interventions on bias outcomes (R²₂ = 0.39; R²₃ = 0.48; see Table 2). We therefore proceeded with moderation analyses to probe potential sources of this heterogeneity. Based on a priori theorizing, we examined the following moderators: bias target, sample population, gender, study design, and publication year. We also examined several methodological moderators (i.e., intervention type, duration, and timing of assessment) and risk of bias.

Bias Target Group

First, we tested whether the meta-analytic effect varied between biases directed at three bias target groups: (a) racial/ethnic minorities, (b) clinically stigmatized groups, and (c) other bias target groups. We did not find evidence that effects varied as a function of any of these different categories (for racial/ethnic minorities, R²₂ = 0.00; R²₃ = 0.030; p = .373; for clinically stigmatized groups R²₂ = 0.00; R²₃ = 0.005; p = .700; for other groups R²₂ = 0.00; R²₃ = 0.061; p = .271).

Sample Population

Next, we tested whether the pooled effect differed between studies among helping professionals compared with studies of students and the general population. We found evidence of marginally significant moderation (R²₂ = 0.00; R²₃ = 0.16; p = .057). The pooled effect of mindfulness interventions on intergroup bias among helping professionals was medium in magnitude (g = −0.26, CI 95% [−0.40, −0.12]), while the effect among students and the general population was large (g = −0.63, CI 95% [−0.82, −0.45]). Notably, the confidence intervals around these estimates were non-overlapping, indicating mindfulness interventions had a larger impact on intergroup bias among the general population than among helping professionals.

Gender

We next tested whether the percentage of female study participants would explain differences in effect sizes across studies, with expectations that studies with majority female samples may find larger effects than those with majority male samples. However, we did not find evidence for this (R²₂ = 0.067; R²₃ = 0.030; p = .554).

Study Design

As Table 2 shows, meta-analytic effects varied as a function of study design (R²₂ = 0.00; R²₃ = 0.15; p = .032). The pooled effect for randomized controlled trials (RCTs) was smaller (g = −0.53, CI 95% [−0.71, −0.34]) than the effect for non-RCTs (g = −0.66, CI 95% [−0.96, −0.37]), though the confidence intervals around these effects were overlapping. These findings provide further support for Hypothesis 1, with a medium-to-large pooled effect of randomized controlled mindfulness interventions on all bias types. For parsimony, we tested Hypotheses 2, 3 and 4 among RCT studies only. In support of Hypothesis 2, we found that among RCT studies, mindfulness interventions resulted in medium-sized attenuations in implicit attitude biases (H2(a); g = −0.33, CI 95% [−0.51, −0.16]), medium-to-large reductions in explicit attitude biases (H2(b); g = −0.41, CI 95% [−0.55, −0.23]), and large reductions in biased affect (H2(c); g = −0.58, CI 95% [−0.87, −0.29]) and behaviors (H2(d); g = −0.90, CI 95% [−1.53, −0.26]). In relation to Hypothesis 3, we did not find evidence that mindfulness interventions had larger effects on controlled forms of bias (i.e., explicit attitudes, affect, and behavior) than on implicit attitudes (R²₂ = 0.00; R²₃ = 0.192; p = .901). Regarding Hypothesis 4, and consistent with prior findings, we did not find differences in effects of mindfulness interventions on biases toward outgroups and internalized biases, among RCTs (R²₂ = 0.004; R²₃ = 0.815; p= .213).

Other Potential Moderators

Effects did not vary as a function of publication year or the methodological moderators intervention type (i.e., interventions with mindful awareness as their singular focus [“mindfulness-only” interventions] vs. multi-component [or “mindfulness-plus”] interventions), intervention duration (including when this variable was log- and quadratic-transformed for possible non-linear effects), or timing of assessment. Regarding intervention type, we used a +1 (mindfulness-only) versus −1 (multi-component) coding scheme and tested whether this dichotomous variable explained heterogeneity in the pooled effect. We also conducted sensitivity analysis with multi-component mindfulness interventions removed from the dataset and obtained a meta-analytic effect (g = −.57, CI 95% [−0.91, −0.23]) that was almost identical to the pooled effect with all interventions included (see also Table 2). We also found no evidence that risk of bias was a moderator of the pooled meta-analytic effect.

Tests for Influential Outliers and Corrections

We examined funnel plots for outliers and tested for influential outliers using the “influence” function in the “metafor” package in R. Funnel plots are displayed in Supplemental Material. Our formal test for influential outliers identified two influential outliers among intervention studies. We conducted subset analyses with these outliers removed. The exclusion of these outliers did not substantially change the pattern of meta-analytic findings. Among intervention studies, the outlier-adjusted meta-analytic effect of mindfulness interventions on all bias outcomes was g = −0.47, CI 95% [−0.57, −0.36]. We also tested for influential outliers among effects for each of the four bias categories (i.e., implicit attitudes, explicit attitudes, explicit affect, and biased behavior), but there were none.

Publication Bias Tests and Corrections

Tests of moderation by publication status identified evidence of publication bias for affect, but not for any of the remaining three outcome types (i.e., implicit attitudes, explicit attitudes, and prejudicial behavior). Meta-analytic effects of mindfulness interventions on measures of affect were larger for published (g = −0.62, CI 95% [−0.81, −0.43]) than unpublished studies (g = −0.10, CI 95% [−0.31, −0.11]), indicating potential publication bias. Second, the 3PSM tests found evidence of potential publication bias for implicit attitudes, χ²(df = 1) = 7.62, p = .006, and explicit attitudes, χ²(df = 1) = 6.19, p = .013. Based on this evidence, we used the PET-PESSE method to generate bias-corrected meta-analytic estimates for implicit and explicit attitudes, and affect. Bias-corrected estimates for implicit attitudes became non-different from zero (g = −0.12, 95% CI [−1.40, 1.16]), while those for explicit attitudes and affect reduced to a medium-sized effect (for explicit attitudes, g = −0.37, 95% CI [−0.56, −0.17]; for affect, g = −0.38, 95% CI [−0.63, −0.12]). There was no evidence of publication bias among intervention studies of prejudicial behavior, or when all outcome types were included in a single meta-analytic model, so no corrections were made to these estimates.

Correlational Studies

Primary Analyses

Meta-analytic findings from the 30 correlational studies (nested within 25 records) included in this review are shown in Table 3 and Figures 5 and 6. As Table 3 indicates, and in support of Hypothesis 1, there was a negative association between dispositional mindfulness and indices of intergroup bias (r = −0.17 [−0.27, −0.03]). Per guidelines provided by Funder and Ozer (2019) and Gignac and Szodorai (2016), this can be interpreted as a small-to-medium sized correlation: one that is consequential over time and of potential explanatory and practical value in the short term (e.g., in relation to specific instances of intergroup bias (Funder & Ozer, 2019).

OPEN IN VIEWER

Table 3.

Meta-Analytic Associations and Moderation Tests for Self-Reported Mindfulness and Bias Outcomes.

Outcome types	k	ES	r [95% CI]	z-score	S E	R ( 2 ) 2	R ( 3 ) 2	p
All outcomes (H1)( I ( 2 ; 3 ) 2 : 0.11; 0.83)	26	150	–0.16 [–0.26, –0.05]	0.16	0.05	—	—	—
Subset analyses
Implicit attitudes (H2a)	5	10	–0.10 [–0.26, 0.02]	–0.10	0.06	—	—	—
Explicit attitudes (H2b)	20	113	–0.12 [–0.18, –0.06]	–0.12	0.03	—	—	—
Explicit affect (H2c)	8	24	–0.19 [–0.33, –0.04]	–0.19	0.08	—	—	—
Behavior (H2d)	2	3	–0.67 [–0.90, –0.15]	–0.81	0.33	—	—	—
Moderation analyses—Hypotheses
Controlled vs. implicit bias (H3)	26	150	—	—	—	.000	.047	.707
Bias target (outgroups vs. the self) (H4)	26	150	—	—	—	.002	.000	.935
Moderation analyses—Exploratory
Minority group
Racial/ethnic minorities	26	150	—	—	—	.000	.032	.195
Clinically stigmatized groups	26	150	—	—	—	.000	.040	.373
Other bias target groups	26	150	—	—	—	.024	.000	.420
Study population	26	150	—	—	—	.00	.00	.868
Gender	26	150	—	—	—	.037	.057	.067
Mindfulness measure	20	116	—	—	—	.000	.140	.031*
MAAS	6	14	–0.21 [–0.54, 0.17]	–0.21	0.20	—	—	—
Multi-dimensional measures	16	102	–0.17 [–0.24, 0.10]	–0.21	0.04	—	—	—
Risk of bias	26	150	—	—	—	.00	.00	.936

Note. k = number of records (i.e., articles and dissertations); ES = number of effect sizes; d = Cohen’s d; CI = confidence interval; R²₍₂₎ = within-record variance in the pooled effect explained by the moderator; R²₍₃₎ = between-record variance in the pooled effect explained by the moderator; MAAS = Mindful Attention and Awareness Scale (Brown & Ryan, 2003).

*

p < .05.

OPEN IN VIEWER

Figure 5.

(A) Forrest Plot of Correlation Between Self-Reported Mindfulness and Biased Implicit Attitudes; (B) Forrest Plot of Correlation Between Self-Reported Mindfulness and Biased Explicit Attitudes.

OPEN IN VIEWER

Figure 6.

(A) Forrest Plot of Correlation Between Self-Reported Mindfulness and Biased Affect; (B) Forrest Plot of Correlation Between Self-Reported Mindfulness and Biased Behavior.

We next tested Hypothesis 2, conducting subset analyses on the associations between self-report mindfulness and the four indices of intergroup bias. As Table 3 shows, we found mixed support for this prediction among correlational studies: The meta-analytic effect for the association between mindfulness and implicit attitudes was non-different from zero (H2(a); r = −0.12 [−0.27, 0.03]). There were small-to-medium sized, negative associations between mindfulness and both explicit attitudes (H2(b); r = −0.12, CI 95% [−0.18, −0.06]) and affect (H2(c); r = −0.19, CI 95% [−0.33, −0.04]), but insufficient data (per the Fu et al., 2011 guidelines) to conduct subset analyses for discriminatory behavior (H2(d); k = 2; ES = 3).

Regarding Hypothesis 3, that mindfulness would have a more negative association with controlled forms of biases (i.e., explicit attitudes, emotions, and behaviors), than with implicit attitudes, we did not find evidence for this (R²₂ = 0.018; R²₃ = 0.082; p = .102). The pooled association between mindfulness and controlled forms of bias was medium-sized and negative (r = −0.18, CI 95% [−0.28, −0.07]), while the pooled effect for implicit attitudes was non-different from zero (r = −0.12 [−0.27, 0.03]).

Finally, we found evidence in support of Hypothesis 4, that dispositional mindfulness would be similarly associated with decreases in bias directed toward others as well as the self (R²₂ = 0.002; R²₃ = 0.000; p = .940). The association between mindfulness and bias toward outgroups was of a small magnitude (r = −0.13, CI 95% [−0.25, −0.01]), while the association between mindfulness and internalized bias was medium-sized (r = −0.23, CI 95% [−0.37, −0.08]).

Moderation Analyses

As Table 3 shows, there was substantial heterogeneity between studies (I²₂ = 0.83), indicating that 83% of the variation in effect sizes was unexplained by the baseline meta-analytic model. Within-study heterogeneity was negligible (I²₃ = 0.11). To probe this unexplained heterogeneity, we conducted moderation analyses testing whether effects varied as a function of the bias target group, study population, gender, mindfulness measure, publication year, and risk of bias.

Mindfulness Measure

As shown in Table 3, the type of mindfulness measure was a marginally significant moderatorof the pooled effect (R²₂ = 0.08; R²₃ = 0.00; p = .054). Subset analyses showed that meta-analytic associations for studies with attentional/awareness measures of mindfulness (i.e., studies using the MAAS; Brown & Ryan, 2003) were non-different from zero (r = −0.21, CI 95% [−0.54, 0.17]), while associations based on multidimensional mindfulness measures were negative, and small-to-medium in size (r = −0.17, CI 95% [−0.24, −0.10]). However, we note the limited number of studies and effects based on the MAAS (k = 6; effects = 14), limiting the inferences that can be drawn from these moderation findings.

Other Potential Moderators

We found no evidence that the meta-analytic effect varied as a function of the bias target group, study population, gender, publication year, or risk of bias status (Table 3).

Tests for Influential Outliers and Corrections

We examined funnel plots for outliers and tested for influential outliers using the “influence” function in the “metafor” package in R. Funnel plots are displayed in Supplemental Material. Our formal test for influential outliers identified five influential outliers among correlational studies. We conducted subset analyses with these outliers removed. Among correlational studies, the outlier-adjusted meta-analytic association between mindfulness and all bias outcomes was z = −0.14, CI 95% [−0.20, −0.07], which is very similar to the un-adjusted effect. We then separately tested for influential outliers among effects for each of the four bias categories (i.e., implicit attitudes, explicit attitudes, explicit affect, and biased behavior). For effects on explicit attitudes and explicit affect, these were almost identical to effects that were not adjusted for outliers (effects, respectively, were z = −0.11, CI 95% [−0.17, −0.05] and z = −0.15, CI 95% [−0.27, −0.03]). Effects for implicit attitudes were smaller in magnitude than the original result (z = −0.06, CI 95% [−0.11, −0.01]), while for biased behavior, there were insufficient effect sizes to generate a meaningful estimate.

Publication Bias Tests and Corrections

As our first test of publication bias, we ran moderation analyses to assess whether meta-analytic effects varied as a function of publication status. We did not find evidence for this. As a second step, we ran the 3PSM analyses, and found evidence of publication bias for the meta-analytic association between mindfulness and all prejudice outcomes, χ²(df = 1) = 4.71, p = .030, as well as the association between mindfulness and implicit attitudes, χ²(df = 1) = 4.00, p = .045. We next calculated bias-corrected effect sizes for these meta-analytic effects, using the PET-PEESE approach. The bias-corrected effect for the association between mindfulness and all prejudice outcomes was negative and medium sized (r = −0.20, 95% CI [−0.33, −0.07]). For the association between mindfulness and implicit attitudes, the PET-PEESE model failed to converge, likely due to the small number of studies (k = 5) included in this subcategory (Stanley, 2017).

Discussion

Whereas prosocial values such as being kind to your neighbor and caring for those in need are promoted in societies all over the world, empirical studies suggest a universal tendency to discriminate against outgroup members, especially those with less social power (Brewer & Brown, 1998; Fiske, 1998; Turner et al., 2001). In this meta-analysis, we explored whether mindfulness, as both a meditative practice and a dispositional factor, may reduce intergroup bias and promote anti-bias tendencies. Guided by our three-dimensional conceptual model, we examined associations between mindfulness and (a) different manifestations of bias, specifically implicit attitudes, explicit attitudes, affect, and behavior, directed toward (b) different bias targets, whether members of an outgroup (bias) or ingroup (internalized bias). We examined these relationships in a bidirectional way, examining individuals’ (c) intergroup orientation toward bias or anti-bias as assessed by measures of intergroup bias as well as measures of anti-oppressive and inclusive attitudes, affect, and behaviors. Study samples were multinational (including the United States, Canada, the United Kingdom, Italy, Portugal, Spain, Israel Palestine, India, Thailand Hong Kong, Australia, New Zealand) and spanned diverse settings and contexts (including university students, clinical populations, community members, research panels, trainees and professionals in health and mental health care, teachers, and law enforcement officers).

We found broad support for our hypothesis that mindfulness would be associated with improvements in intergroup bias/anti-bias outcomes (H1). Effects were medium sized for intervention studies (g = −0.56) and small-to-medium sized for correlational studies (r = −0.17); all effects were in the direction of reduced bias. Compared with Oyler et al. (2022), which included fewer study samples (k = 29 compared with 70 in the present study), this meta-analysis yielded a larger-sized effect of mindfulness interventions but a similar-sized effect for correlational studies on bias outcomes compared with their results (g = +.33 for intervention studies; g = +.24 for dispositional mindfulness, scaled in the direction of reduced bias).

Our hypothesis that effects would be significant across all bias outcomes (H2) was also confirmed, with MBIs having a very large effect on behaviors, a large effect on affect and explicit attitudes, and a medium-sized effect on implicit attitudes. A conservative test, including only the RCT studies, found comparable results and confirmed that effect sizes tended to be larger for biased behavior and affect, followed by explicit attitudes and implicit attitudes. Results from correlational studies also mirrored this pattern although overall effect sizes were smaller. Associations to explicit attitudes and affect were in the small-to-medium sized range, whereas there was no significant association with implicit attitudes. According to Funder and Ozer (2019), a small-to-medium sized correlation can be interpreted as one that is consequential over time and of potential explanatory and practical value in the short term (e.g., in relation to specific instances of intergroup bias). Notably, there were not enough correlational effects to examine associations separately for biased behavior.

Together, these results provide robust support that the benefits of mindfulness extend to the bias domain, with stronger effects on explicit responses and weaker effects on responses operating outside of our conscious awareness and control (e.g., implicit attitudes). These results are consistent with findings that while implicit biases can be changed, such changes are likely to be modest and rely on mechanisms that are distinct from those most effective for changing explicit biases (Forscher et al., 2019). According to dual-process theories, explicit biases stem from cognitive processes that are slower, deliberate, and controllable (Devine, 1989), such that mindfulness-induced improvements in attention, self-awareness, and cognitive control may attenuate the conscious, explicit expression of bias in its cognitive, affective, and behavioral forms. Furthermore, as mindfulness has been found to strengthen the link between intention and behavior (Chatzisarantis & Hagger, 2007), the larger effect size for behavior may be driven in part by mindfulness-induced changes in cognitive resources and motivation to control the effects of bias on behavior (Devine, 1989; Fazio & Olson, 2014).

In contrast, implicit biases reflect automatic mental processes that are fast, efficient, habitual, and uncontrolled (Devine, 1989). In a recent meta-analysis of strategies to change implicit measures, procedures that induced goal-directed motivation (e.g., making anti-racist norms salient) or directly or indirectly targeted biased associations (e.g., Black = bad) reduced implicit bias relative to a neutral control (Forscher et al., 2019). Along these lines, mindfulness-based interventions, such as lovingkindness practice, may affect implicit attitudes by emphasizing prosocial norms (Berry et al., 2020), increasing perspective-taking (Kang & Falk, 2020), and reflective awareness, which Verhaeghen and Aikman (2020) found was associated with both implicit racial bias and motivation to control prejudiced reactions.

Our third hypothesis, that mindfulness would have a substantially greater effect on controlled aspects of bias (explicit attitudes, affect, and behavior) than on automatic (implicit) bias (H3), was not supported, although results trended in the expected direction. For intervention studies, for example, the pooled effect for controlled forms of bias was large in magnitude, while the pooled effect for implicit attitudes was moderate in comparison. Further research is needed to clarify the different mechanisms through which mindfulness may affect automatic versus controlled types of bias and implications for behavior (Cuddy et al., 2007; Fazio & Olson, 2014; van Zomeren et al., 2008).

Given the psychological and health outcomes associated with internalized bias (Gale et al., 2020), our final hypothesis examined whether mindfulness would be associated with comparable effects on bias directed toward the (marginalized) self and outgroup members (H4). Results showed that MBIs were associated with large negative effects for both internalized bias and intergroup bias outcomes. Among correlational studies, associations between mindfulness and bias were also significant, though in the small (intergroup bias) to medium-sized (internalized bias) range. To our knowledge, this is the first meta-analysis to demonstrate that in addition to improving intergroup outcomes for dominant group members, mindfulness may enhance the well-being of marginalized individuals by increasing resilience to negative societal messages about their group. As internalized bias has both direct and indirect associations with outcomes such as distress, quality of life, self-esteem, and health risk behaviors among stigmatized groups (Lee et al., 2019), decreasing internalized bias through mindfulness may be an important part of a multi-tiered strategy for promoting health equity. At the same time, systems-level interventions are needed to dismantle the structural inequities that perpetuate and maintain population health disparities (Bailey et al., 2017).

Moderation analyses found that the effects of MBIs and dispositional mindfulness were comparable for biases directed toward racial/ethnic minorities, clinically stigmatized groups, and other bias targets, whether an outgroup or ingroup member. Effects also did not vary by gender of the respondent, year of publication, or study quality (i.e., risk of bias). For intervention studies specifically, effects also did not vary by length of the intervention or timing of assessment. Importantly, we examined the possibility, raised by Berry et al.’s (2020) critique, that the inclusion of ethical enhancements in many MBIs may introduce demand characteristics and social desirability concerns, which may promote reductions in bias beyond the direct effects of mindfulness itself. To explore this possibility, we compared the effects of mindfulness-only interventions to multi-component interventions that include other enhancements, such as compassion, values-enhancement, or other psychological skills. We also conducted “subset” analysis, to examine meta-analytic effects of mindfulness interventions with the so-called “mindfulness-plus” interventions excluded from the analysis. Together, these tests provide evidence that there was no systematic variation in effects between so-called “pure” mindfulness interventions and those with ethical enhancements. As shown in Table 2, effect sizes for both intervention types were medium sized and comparable (for “mindfulness-only” interventions, k = 16, g = −0.57, 95% CI [−0.91, −0.23], and for “mindfulness-plus” interventions, k = 25, g = −0.54, 95% CI [−0.69, −0.39]. Results for the mindfulness-only interventions are largely consistent with Berry et al.’s (2020) findings, taking into account their more restrictive selection criteria. For the studies focusing on prejudice outcomes (20.6% of the sample), they found a small to medium effect of mindfulness training only when compared with active and inactive controls, k = 6, g = .464, 95% CI [.302, .626].

Altogether, the results of moderation analyses for the intervention studies highlight that even small doses of mindfulness training, alone or packaged with other components, were sufficient to significantly improve bias outcomes for both ingroup and outgroup members. However, moderation analyses by type of mindfulness measure suggest that multiple processes are involved in bias reduction. Results from correlational studies indicated that the effect of nonjudgmental awareness (as assessed by the MAAS; Brown & Ryan, 2003) was not different from zero, whereas associations between multidimensional measures of mindfulness and bias outcomes were negative and small-to-medium sized. It seems that actively altering biased associations requires more than nonjudgmental awareness, for example, the additional skills of description, acceptance, and nonreactivity assessed by multifactor measures such as the FFMQ (Baer et al., 2006) and the KIMS (Baer et al., 2004).

A positive trend within the field of mindfulness studies is a growing focus on applications to real-world contexts. Notably, moderation analyses comparing members of the general population (predominantly university student samples) to helping professionals (e.g., teachers, counselors, and medical staff) found that MBIs had a stronger effect on bias in the general population. This finding is consistent with claims that university students represent a unique population, with less crystallized attitudes, less formulated senses of self, more tendencies toward compliance, and therefore particularly vulnerable to experimental manipulations (Sears, 1986), although empirical studies suggest they are more heterogeneous than previously assumed (Peterson & Merunka, 2014). Another possibility is that helping professionals interact more directly with individuals who are the targets of bias within institutional contexts that have a history of pathologizing and marginalizing groups with low social power. For example, systemic inequities in educational achievement and school discipline rates reflect racial, gendered, and class-based stereotypes that are often normalized and reinforced in school settings (Chin et al., 2020; Okonofua & Eberhardt, 2015). Applying MBIs within such institutional contexts likely increases the difficulty of shifting biases among helping professionals and others in positions of power, although MBIs were still associated with improved outcomes overall. More contextualized MBIs may be needed to address the complex and situated power dynamics that may reinforce bias toward stigmatized members of society (see, for example, Kanter et al., 2020). However, for correlational studies, moderation analyses found no differences in effects among the general population and helping professionals.

We found some evidence of publication bias for affect, explicit, and implicit attitudes for the intervention studies. Our bias-corrected estimates indicated a medium-sized effect of mindfulness interventions in reducing explicit biased attitudes and affect, but not in reducing implicit biases or biased behaviors. While there is arguably value in adjusting estimates for potential bias, we also note that bias-adjusted estimates need to be interpreted with caution, and not in isolation from the original estimates (Carter et al., 2019; Stanley et al., 2017).

Conclusion

This review concludes that mindfulness is a promising pathway to reducing intergroup bias and internalized bias in the direction of greater equity and inclusion of marginalized groups. We found multi-method evidence, across RCTs, other intervention designs, and correlational studies that mindfulness inhibits intergroup bias and promotes anti-bias attitudes, affect, and behavior in diverse country contexts and samples. Analysis of evidence and gaps in the literature highlights the need for future studies to adopt rigorous RCT designs, examine mechanisms of mindfulness, and target the under-researched areas of internalized bias and anti-bias attitudes, affect, and behaviors among marginalized groups specifically. Creative approaches are needed to operationalize minority experiences more thoroughly and recognize the complex intersectional ways that dynamics of power and oppression manifest in intergroup contexts. At the same time, we acknowledge that the problem of intergroup bias is complex and reinforced by systems of oppression such as racism, sexism, and classism. Structural solutions are needed to address the systemic nature of bias across life domains.

Supplemental Material