School Racial Composition as a Moderator of the Effect of Discrimination on Mental Health and Substance use Among American Indian Adolescents

Abstract

Purpose:

To examine the relationships between individual-level perceived racial/ethnic discrimination and mental health and substance use outcomes by school-level racial composition among American Indian (AI) adolescents.

Method:

Self-reported survey data on individual-level variables come from a sample of AI adolescents (n = 510) living in or near the Cherokee Nation during the fall of 2021. School-level data come from publicly available databases. Multilevel linear and logistic regression analyses were performed to test for and examine the interaction between perceived racial/ethnic discrimination and school racial composition in relation to symptoms of anxiety and depression, past 30-day use of alcohol and marijuana, and misuse of prescription opioids.

Results:

Adjusted analyses showed a significant interaction effect between discrimination and racial composition on anxiety symptoms, such that the effect of discrimination was more pronounced at lower % AI (10th percentile) than at more equivalently mixed (50th percentile) or higher % AI (90th percentile) school settings. No significant interactions were observed with depressive symptoms or substance use outcomes.

Discussion:

School racial compositions of higher percentage AI may buffer the adverse effect of racial/ethnic discrimination on anxiety symptoms among AI adolescents.

Discrimination against American Indian and Alaska Native (AI/AN) people in the United States is rooted in entrenched and pervasive legacies of Western colonialism, genocide, forced migration of Tribal communities, and resultant mass trauma [1]. Racial/ethnic discrimination, or discrimination on the basis of race or ethnicity, has affected more than one in three AI or AN adults in the United States [2]. Of those, approximately 39% have reported experiencing racial/ethnic discrimination in the form of microaggressions, 35% in the form of racial slurs, and 23% in the form of sexual harassment due to their AI or AN identity [2]. Despite the current surge in scientific attention devoted to racism as a driver of social inequities and mental health outcomes, AI or AN communities continue to be underrepresented in the literature [3]. An improved understanding of racial/ethnic discrimination among diverse Tribal and cultural groups could have implications for the design of tailored measurement and intervention strategies [4].

Across cultures, settings, and the life course, racial/ethnic discrimination predicts adverse mental and behavioral health outcomes [5]. Racial/ethnic discrimination increases the risk of depressive and internalizing symptoms, anxiety, and conduct, aggression, and other externalizing problems among members of minority racial/ethnic groups [5,6]. Perceived racial/ethnic discrimination has come under closer scrutiny as an adverse childhood experience with significant co-occurrence alongside other adverse childhood experiences attributable to a young person’s social environment, such as food and housing insecurity, parent or caregiver divorce, death, or imprisonment, family and neighborhood violence, and living with someone who has a mental illness or a substance use problem [7–9]. Racial/ ethnic discrimination as an adverse childhood experience has demonstrably larger effect sizes on mental, emotional, and behavioral health than other adverse childhood experiences among certain groups, which may indicate its relative clinical significance as a social determinant of psychosocial health among youth [7]. Similar adverse effects have been found among young people who identify as AI or live in and near Tribal reservations [2,10–12].

Racial/ethnic discrimination has also been found to predict substance use early onset, frequency, and intensity during adolescence among AI adolescents and those living in or near the Cherokee Nation Reservation [11,13]. The population-level problem of substance use is considered a crisis in Indian country and a pressing issue for native communities and public health interventionists who focus on improving the future for AI adolescents [14]. Relative risk ratios of past 30-day and lifetime substance the use including alcohol, marijuana, and other illicit substances, revealed significantly higher risk among AI adolescents compared to a national sample of adolescents in the United States [15]. As a complex problem, the relationship between mental illness and substance use is often mutually reinforcing, such that self-medication with alcohol and other substances may be used as a coping strategy to alleviate anxiety, depression, or other symptoms of mental illness, but self-medication may exacerbate symptoms over the long term [16,17].

The sociocultural environments in which AI adolescents grow, learn, and thrive are distinct, with population-level race-related traumas and triumphs undergirding communities’ legacies, informing conceptions of mental health and coping approaches, and providing essential context for understanding etiologic and intervention-focused research findings [18]. In and near the Cherokee Nation Reservation, which encompasses the northeastern 14-county region of the state of Oklahoma, higher levels of perceived racial/ethnic discrimination among adolescents have been linked to elevated depressive symptoms and the use of alcohol, misuse of prescription drugs, and use of other illicit substances [10,11].

While these studies have examined individual-level exposures, examination of higher-level variables (i.e., sociocultural and environmental influences on adolescent mental health and substance use behavior) is needed to contextualize these effects and identify potential extra-individual intervention targets. Multilevel analysis is an underutilized but powerful tool that can be leveraged to better understand social and contextual variables that predict mental health and substance use among adolescents. For example, multilevel analysis of data from the National Longitudinal Study of Adolescent Health revealed that school-level random effects were significant and salient cluster-level predictors of depressive symptoms [19]. Students in Oklahoma spend an average of over 40% of waking hours in school during the academic year [20], which is especially important considering schools provide exposure to social relationships, support, and norms [21,22].

The racial composition of a young person’s school environment may play a significant role in the relationship between their experiences of racial/ethnic discrimination and mental health and substance use. Social environments with a higher concentration of one’s racial/ethnic group may offer protection against racism in the form of emotional and practical support, as well as a heightened sense of belonging and commitment to shared cultural identity [23]. The literature proposes two competing hypotheses and mixed empirical evidence regarding the impact of the racial composition or ethnic density of an individual’s social context on perceived racial/ethnic discrimination and prejudiced beliefs: (1) the relationship is curvilinear, such that discrimination peaks when racial composition is roughly mixed (e.g., 50:50 ratio of racial/ethnic groups) and tapers when racial composition is predominantly one racial/ethnic group or the other [24]; or (2) the relationship is linear inverse i.e., as racial composition (% one racial/ethnic group) increases, discrimination toward this group decreases [24–26]. Empirical evidence has not kept pace with theoretical developments in this area. Most empirical research on racial composition or ethnic density in social contexts and discrimination has examined the likelihood of discrimination as a function of racial/ethnic composition [27] or has tested discrimination as a mediator of the effect of racial composition on mental health outcomes [28]. These research pursuits have largely bypassed the investigation of the buffering offered by racial composition or ethnic density on discrimination’s adverse effects among minority groups; however, there are a few exceptions among adults in different settings [23,29,30]. Emerging findings reveal trending [30] or significant protective attenuation by community-level racial composition on adult health outcomes, including depression [23,29]. Reasons for inconsistent evidence across studies might include varying levels of racial compositions, distinct settings, distinct racial/ethnic groups, adjustment for different covariates, and concurrent measurement of overall health or other outcomes with insufficient lag to detect effects [30]. No prior research has examined the protective moderation on mental health and substance use among AI adolescents.

The main opposing hypotheses of racial composition as a protective buffer merit exploration in the context of high schools, where adolescents spend a significant amount of time and their development may be affected by ethnic-racial socialization, the process of learning about race, ethnicity, and racial discrimination [31,32]. Ethnic-racial socialization tends to occur at home initially, but as young people frequent school and spend more time with a mix of peers, teachers, and community members, the variety of socialization cues, content, and messaging becomes increasingly complex [32]. Given the importance of school environments to adolescent mental health, substance use, and racial/ethnic minority health, research is needed to reconcile emerging and competing theories about the role of school racial composition on AI adolescents’ experiences with discrimination, mental health, and substance use.

The present study builds on this growing body of research and theory on the racial composition of social environments and discrimination experiences as they apply to adolescents living on and near the Cherokee Nation Reservation (K. A. Komro et al, unpublished data, 2023) [10,11,13]. In this set of analyses, we incorporated school racial composition as a feature of the school environment to better understand the relationships between racial/ethnic discrimination and mental health and substance use outcomes among AI students. Using cross-sectional data, we performed multilevel linear and logistic regressions to examine cross-level moderation by school racial composition of the effect of individual-level perceived racial/ethnic discrimination on mental health (anxiety symptoms and depressive symptoms) and substance use (past 30-day use of alcohol, marijuana, and prescription opioids off-label). We analyzed the main effects of perceived racial/ethnic discrimination on these outcomes when interactions were non-significant.

Methods

Data sources and participants

Individual-level data are cross-sectional and have been collected in collaboration between Emory University and Cherokee Nation Behavioral Health as part of a prevention trial to reduce substance use among high school students attending public schools within the boundaries of the Cherokee Nation Reservation (k = 17) or near its boundaries (k = 3 within the boundaries of a neighboring Tribal Nation Reservation), which are located in a rural region of 14 counties in the state of Oklahoma [33]. During the fall of 2021, administrators from area high schools agreed to participate (k = 20) and undergo randomization to receive intervention strategies for prevention and support. Participating 10th-grade students (n = 919) completed surveys during school hours using tablets. Survey items asked about sociodemographics, experiences with discrimination, mental health, and substance use. For this set of analyses, the sample was restricted to the 54.9% of students who self-identified as AI (AI; n = 510). School-level data were retrieved from the National Center for Education Statistics (NCES) and the Oklahoma School Report Cards system (OKSRC) [34,35]. The Transparent Reporting of Evaluations with Nonrandomized Designs statement for observational research and CARE principles for Indigenous Data Governance guide the reporting of methods and findings for this study [36,37]. All study protocols and this manuscript have been approved by the Cherokee Nation Institutional Review Board, which serves as the single institutional review board for the study, with a reliance agreement with Emory University and a letter of agreement from the neighboring Tribal Nation.

Measures

Perceived racial/ethnic discrimination.

Perceived racial/ethnic discrimination (individual-level predictor) was measured using a single item that asked, “How often have you experienced any kind of discrimination due to your race or ethnicity?” [38,39]. Responses were indicated on a 5-point scale where 0 = never, 1 = hardly ever, 2 = a few times a year, 3 = monthly, and 4 = daily. Small cell counts at higher levels of discrimination required dichotomization of this predictor to yield a binary variable where one and 0 respectively correspond to ever and never having experienced racial/ethnic discrimination.

School racial composition.

School racial composition (school-level predictor) was the percentage of each school’s student body that was AI (% AI). Percentage AI was chosen as it relates to the competing theories of the effects of the racial composition of one’s social environment on discrimination experienced among members of the focal racial/ethnic group. School racial composition was retrieved from the NCES, a publicly available database [34].

Anxiety symptoms.

Anxiety symptoms were assessed with the generalized anxiety disorder scale, a 7-item measure (generalized anxiety disorder [GAD]-7) [40]. The GAD-7 asked about past two-week experiences of being bothered by feelings of nervousness/anxiousness, inability to control worry, worrying too much, having trouble relaxing, restlessness, annoyance, and irritability, and fear of something awful happening. Responses were indicated on a 4-point scale where 0 = not at all, 1 = several days, 2 = more than half the days, and 3 = nearly every day. According to the scoring protocol, responses were summed to yield a total score for each respondent that could range from 0 to 21. Scores of 10 and 15 are the clinical cut points for moderate and severe risk of clinical anxiety, respectively. Cronbach’s alpha for the GAD-7 among AI students was 0.91.

Depressive symptoms.

Depressive symptoms were assessed using the patient health questionnaire, an 8-item measure (PHQ-8) [41]. The PHQ-8 asked about past two-week experiences feeling bothered by lack of interest or pleasure, hopelessness, sleep problems, lack of energy, poor appetite, feeling bad about one-self, poor concentration, and movement concerns. Responses were indicated on a 4-point scale where 0 = not at all, 1 = several days, 2 = more than half the days, and 3 = nearly every day. Per the scoring protocol, responses were summed to yield a total score for each respondent that could range from 0 to 24. Scores ≥10 reflect a moderate risk of clinical depression. Cronbach’s alpha for the PHQ-8 in this sample was 0.89.

Use of alcohol, marijuana, and prescription opioids off-label.

Past 30-day use of alcohol, marijuana, and prescription opioids in an off-label manner (i.e., differently than how a doctor or medical provider instructed) was assessed in three separate items of similar format from the Youth Risk Behavior Survey [42]. Students reported their alcohol and substance use from 0 to 30 days. Due to small cell counts, past 30-day use responses were collapsed to yield dichotomous outcomes (1 = used, 0 = did not use).

Individual- and school-level covariates.

Individual-level covariates included age, gender, and enrollment in a free or reduced-price lunch program in school, which served as a proxy for socioeconomic status defined by household income, according to the federal Income Eligibility Guidelines by the United States Department of Agriculture [43]. School-level covariates were retrieved from NCES and OKSRC [34,35]. The number of students in ninth grade during the academic year 2020–21 (the most recently available data), which corresponds to the 10th-grade sample analyzed, and the percentage of students who were directly certified were obtained from NCES. The percentage of students with direct certification is a school-level indicator of economic status, such that based on household receipt of supplemental nutrition assistance or other assistance, students are eligible for the National School Lunch Program. This indicator has been cross-validated with community measures of low income, such as the percentage of families living in poverty, the median household income, and percentage of households in poverty [44]. School four-year graduation rates, rates of chronic absenteeism (>10% of school days missed), and student-teacher ratios were obtained from OKSRC.

Analyses

All analyses were performed in SAS 9.4 [45]. Univariate statistics are reported in Table 1 for individual- (racial/ethnic discrimination, age, gender, free or reduced-price lunch, anxiety symptoms, and depressive symptoms) and school-level variables (grade enrollment, school racial composition as % AI, percentage of students who are directly certified, four-year graduation rate, chronic absenteeism, and student-teacher ratio). To estimate the association between racial discrimination on AI adolescents, we analyzed a series of generalized linear models. For each outcome, we began by estimating a maximal model of the following general form:

$$\text{g}\left({\text{y}}_{\text{ij}}\right)={\beta }_0+{\beta }_2{\text{D}}_{\text{ij}}+{\beta }_3{\text{AI}}_{\text{j}}+{\beta }_4{\text{D}}_{\text{ij}}{\text{AI}}_{\text{j}}+{\text{X}}_{\text{ij}}+{\text{Z}}_{\text{j}}$$

where g() is the appropriate link function chosen based on the outcome distribution, y_ij is the outcome of interest for each individual, D_ij is the individual-level reported racial/ethnic discrimination, AI_j is the school-level percentage of students identified as AI (% AI), X_ij is the set of individual-level covariates, and Z_j is the set of school-level covariates. If the estimated interaction was found to be statistically non-significant, the interaction term was dropped from the model, and the covariate-adjusted and unadjusted associations between discrimination and each outcome were estimated. For continuous outcomes, unstandardized and standardized effect sizes (Table A2) are reported with the mean set to 0 and the standard deviation set to 1.

Table 1.

Individual- and school-level characteristics (n = 510, k = 20)

	n or M	% Or SD
Individual-level variables
Racial/ethnic discrimination	217	42.55
Age
≤15	238	46.76
16	245	48.13
≥17	26	5.11
Male	243	47.93
Free or reduced-price lunch	386	76.13
Anxiety symptoms	6.81	5.90
Depressive symptoms	7.59	6.14
Past 30-d use: alcohol	103	20.40
Past 30-d use: marijuana	82	16.24
Past 30-d misuse: prescription opioids	19	3.83
School-level variables
School racial composition (% AI)	47.69	11.77
Grade enrollment	61.30	20.07
% Directly certified	30.06	9.71
Graduation rate (four-year)	87.66	7.30
Chronic absenteeism (>10% missed)	15.96	20.63
Student/teacher ratio	13.41	1.12

n = number of students; k = number of schools. Percentages were calculated based on available data. Depressive and anxiety symptoms were measured using the PHQ-8 and GAD-7 respectively. Prescription opioid misuse refers to the use of prescription opioids without a doctor’s prescription or differently than how a doctor or medical provider intended. School-level data represent academic year 2020–21, the current sample’s first year. School racial composition: the percentage of students who are American Indian (% AI). Directly certified: the un-duplicated number of students who are eligible for the National School Lunch Program through direct certification (e.g., children in households that receive Supplemental Nutrition Assistance or other assistance).

AI = American Indian; GAD-7 = generalized anxiety disorder-7; PHQ-8 = patient health questionnaire-8.

Since students are nested within schools, they cannot be assumed to be independent from one another. To account for clustering of students within schools, all models were estimated with correlated residuals at the school level, assuming a compound symmetry correlation structure. Linear models for continuous outcomes were estimated in PROC MIXED, and logistic models for binary outcomes were estimated in PROC GLIMMIX. Multicollinearity was assessed for linear models by calculating variance inflation factors for the primary predictors of discrimination and the interaction of discrimination and school-level % AI. Multicollinearity was minimal, with all relevant variance inflation factors below 2.0. In alignment with current recommendations, multiplicity adjustment was not performed as statistical claims are intended for independent tests [46]; each individual test in these analyses stands independently with its corresponding alpha threshold. As a sensitivity analysis, self-identifying as multiracial (AI and at least 1 other race/ethnicity) was included as a covariate in adjusted models.

Results

Table 1 presents descriptive statistics on individual- and school-level variables. The sample consisted of 510 10th-grade students who identified as AI. The median age of respondents was 16 years old, and approximately half of the sample identified as male (47.9%). Students were predominantly from households of lower income, as indicated by enrollment in a free or reduced-price lunch program at school (76.1%). The rate of ever having experienced racial/ethnic discrimination was 42.6%. Students’ average anxiety score was 6.81 (SD 5.90). The average depression score was 7.59 (SD 6.14). Rates of past 30-day use of alcohol, marijuana, and prescription opioids in an off-label manner were 20.4%, 16.2%, and 3.8%, respectively.

Across the 20 schools, student body racial composition was 47.7% AI on average (SD 11.8, minimum 26.3%, maximum 90.7%). The mean enrollment in the sample’s grade year was 61.3 students (SD 20.1). A school-level indicator of lower socioeconomic status, as indicated by the percentage of students with direct certification or eligibility for the National School Lunch Program, was 30.1% (SD 9.7). On average, the four-year graduation rate was 87.7 (SD 7.3), chronic absenteeism (>10% missed) was 16% (SD = 20.6), and the student/teacher ratio was 13.4 (SD 1.1).

Figure 1 illustrates the significant interaction effect on anxiety symptoms for those who have and have not experienced racial/ethnic discrimination. Adjusted multilevel analyses revealed the significant interaction between racial/ethnic discrimination and school racial composition on anxiety symptoms (p = .038), such that the effect of discrimination was more pronounced at lower % AI in schools than at more equivalently mixed or higher % AI racial compositions. In other words, the adverse effect of discrimination on anxiety among AI students was cushioned by higher % AI school environments. Stratified point estimates reveal this change in effect magnitude by school % AI: the effect of discrimination on anxiety symptoms was 3.49 (95% CI 2.04, 4.93; standardized 0.59; 95% CI 0.35, 0.84) at 10th percentile racial composition (where % AI is 34.67%; p < .001), 2.12 (95% CI 1.14, 3.10; standardized 0.36; 95% CI 0.19, 0.53) at 50th percentile (where % AI is 50.49%; p < .001), and 1.54 (95% CI 0.33, 2.76; standardized 0.26; 95% CI 0.06, 0.47) at 90th percentile (where % AI is 57.14%; p = .013). There were no significant interaction effects on the other outcomes.

Figure 1.

Interaction effect between racial/ethnic discrimination and school racial composition on anxiety symptoms among American Indian adolescents. Confidence intervals (95%) are shown. Fit computed at age = 15.6, gender = 0.5, free or reduced price lunch = 0.8, grade enrollment = 61.6, direct certification = 30.1, graduation rate = 87.6, chronic absenteeism = 15.6, and student/teacher ratio = 13.4. Plotted on available sample data, with school racial composition ranging from 26.3 to 90.70% American Indian. GAD-7 = Generalized Anxiety Disorder-7.

Results from main effects models of racial/ethnic discrimination on all outcomes were stable across models with and without adjustment for covariates. Table 2 presents results from main effects models. With adjustment, the main effect of discrimination on depressive symptoms was significant, such that among those who had experienced discrimination on the basis of race or ethnicity, depressive symptoms were elevated by 3.01 units on average (95% CI 2.02, 3.99; standardized 0.49; 95% CI 0.33, 0.65; p < .001). There were no significant main effects of discrimination on substance use outcomes, which are presented as odds ratios in Table 2. Results from the full model specifications are provided in the appendix. Sensitivity analysis with multiracial identity (AI and at least one other race or ethnicity) yielded equivalent results with no changes in primary predictor and interaction term effect sizes or p-values; multiracial identity was not a statistically significant covariate across outcomes.

Table 2.

Effect estimates of racial/ethnic discrimination on mental health and substance use outcomes from multilevel models (k = 20, n = 505)

Outcome	Racial/ethnic discrimination
	Unadjusted (95% CI)	Adjusted (95% CI)
Depressive symptoms	3.412*** (2.365, 4.459)	3.005*** (2.018, 3.992)
Past 30-d alcohol use	1.138 (0.739, 1.754)	1.045 (0.666, 1.641)
Past 30-d marijuana use	1.272 (0.791, 2.045)	1.278 (0.775, 2.109)
Past 30-d prescription opioid misuse	1.003 (0.394, 2.555)	0.919 (0.370, 2.285)

Linear regressions were performed for depressive symptoms with beta estimates shown; logistic regressions were performed for binary outcomes (past 30-day alcohol use, marijuana use, and misuse of prescription opioids) with odds ratios shown. Adjusted models included age, gender, free or reduced-price lunch, school racial composition (% AI), grade enrollment, % directly certified, graduation rate, chronic absenteeism, and student/teacher ratio. Interaction terms were dropped from shown models due to nonsignificance.

AI = American Indian.

p < .05*; p < .01**; p < .001***.

Discussion

In this set of analyses, we assessed the relationships between individual-level perceived racial/ethnic discrimination and mental health and substance use outcomes and school-level racial composition among AI adolescents. A significant interaction effect was observed on the outcome of anxiety symptoms, such that the effect of discrimination on anxiety symptoms among AI adolescents is significantly reduced when the school environment has a higher percentage of AI students. This finding could have important implications for future research and policy by adding to the evidence that school environments are crucial settings for adolescent development. Reasons for why elevated AI racial composition appears to be protective in the observed interaction may include an increased sense of belonging on the basis of race or ethnicity, Tribal affiliation, and/or cultural identity. Shared characteristics with a larger percentage of peers in one’s school environment may reflect connectedness and bonding, similar constructs to belonging, which are inversely related to or protective against substance use, delinquent behaviors, and adverse academic outcomes [21,22,47]. At the other end of the spectrum (lower % AI), the effects of discrimination on anxiety may feel more poignant when there are fewer peers in students’ school environment with whom they share important identity characteristics. Importantly, the study findings are cross-sectional, and future research is needed to explore these questions with a causal approach.

The null findings on substance use outcomes are worth deeper discussion. One feature of this study is that we analyzed mental health and substance use as separate outcomes in parallel models, but we did not examine the relationship between them, such as the potential mediating role of anxiety or depressive symptoms on a causal pathway from discrimination to substance use. Moreover, mental health measures were continuous, which lends granularity to these measures in comparison to the dichotomized substance use outcomes that might be more obstinate to change. Null findings on substance use outcomes might reflect this difference in measurement, a fundamental difference in reactivity to shifts in discrimination exposure, or a lack of substance use uptake among a cross-sectional sample of 10th graders. It is possible that a causal effect would emerge over time with more waves of data and increased uptake in substance use for coping, according to the typical developmental trajectory for adolescents. Additionally, measurement of discrimination was dichotomized in these analyses; a continuous scale or measure with categories of frequency or intensity of discrimination could better capture nuance in a relationship between exposure and substance use outcomes that our measures might obscure. For example, a previous study of discrimination and substance use among adolescents living in the Cherokee Nation found that higher discrimination frequency was associated with a higher risk for heavy use of alcohol, misuse of prescription drugs, and use of other illicit drugs, and that higher discrimination intensity was associated with a higher risk for the latter two substance misuse outcomes; all substance misuse outcomes were measured using the same items as in the present study, but the discrimination measure had different dimensions (frequency and intensity) and more categories (none or low, moderate, and high) [11].

Reasons for the finding of a significant interaction between school racial composition and discrimination on anxiety symptoms but not depressive symptoms might be due to mechanisms and specific psychological symptomatology. The mechanism of protection offered by a higher % AI school environment might function to alleviate heightened fight-or-flight responses in the face of threat (i.e., heightened anxiety in the face of discrimination) but less effectively buffer against sadness, hopelessness, or demoralization in the face of injustice (i.e., heightened depression in the face of discrimination). More research is needed to understand why and how school racial composition might offer protection against certain mental health conditions but not others, and how protective effects might vary on these outcomes over time for AI youth.

In light of our reported findings, this study is not without limitations. The data were cross-sectional, which means we cannot infer causality about these relationships from this study’s results alone. It is possible that heightened anxiety symptoms increase one’s sensitivity to and perception of racial or ethnic discrimination. However, existing discrimination research lends evidence to not only the temporal sequence but also the causal effects of discrimination on compromised mental health outcomes, and this evidence base is rapidly growing across populations and contexts [5,6,31,39].

An additional limitation is that the range of school racial compositions analyzed is not 0%–100%; the range is 26.28%–90.70% with an interquartile range of 19.81 around the median of 50.49%. Although a wide range is represented by schools in this sample, we could not fully test the competing theories described in the introduction and the potential protective effects of school settings with racial compositions at the tail ends of a complete distribution of possibilities (i.e., schools with >90% AI compared to schools with <26% AI). Further, we do not have data on when or where discrimination was experienced; adverse effects of discrimination events that occurred in close proximity to high % AI school contexts might be better buffered than others.

Unaccounted-for confounding remains a possibility and should be considered when interpreting study findings. While gender and enrollment in a free or reduced-price lunch program in school were statistically significant covariates in models of mental health outcomes, we encourage caution in interpretation as they cannot be construed as direct or total effects in this study [48]. Further, the lack of significant random effects may indicate school-level homogeneity; however, our data do not allow for such a conclusion, as the prevention trial from which these data were collected was designed to provide precise estimates of fixed effects in the presence of school-level clustering and not to give precise estimates of the random effects used in the models [33].

Future research could utilize longitudinal methods to examine causal effects, such as the roles of mediators and protective moderators among AI adolescents. One direction could be to explore the moderation effects of shared group identity or school belonging on the relationship between discrimination and relevant outcomes, similar to the conceptual framework and empirical findings of the Morris et al. (2020) study among Black youth [22]. If the observed moderating effect of school % AI was actually driven by a sense of shared group identity, belongingness may be the operational construct that merits attention and intervention. Shared group identity or belongingness may serve as a protective buffer in the relationship between discrimination and anxiety symptoms, such that a higher sense of belonging to one’s school environment mitigates the adverse effects of low % AI school racial composition. This would be the critically important next step to inform school-based interventions that may seek to increase a sense of belonging as a modifiable intervention target to improve mental health outcomes for AI students. Future interventions would benefit from specific causal research in considering the interaction of the school environment’s racial composition and individual discrimination experiences in relation to anxiety among students.

IMPLICATIONS AND CONTRIBUTIONS.

School racial compositions of higher percentage American Indian may buffer the adverse effect of racial/ethnic discrimination on anxiety symptoms among American Indian adolescents.