The end of court-ordered desegregation and US children’s health: quasi-experimental evidence

Abstract

School racial segregation significantly affects racial disparities in US children’s health. Recently, school segregation has been increasing, partially due to Supreme Court decisions since 1991 that have made it easier for school districts to be released from court-ordered desegregation. We investigated the association of the end of court-ordered desegregation with child health, using the 1997-2018 waves of the National Health Interview Survey (n = 8182 Black children; n = 16 930 White children). We exploited quasi-random variation in the timing of school districts’ releases from court orders to estimate effects on general health, body weight, mental health, and asthma, using difference-in-differences and event-study methods (including traditional and heterogeneity-robust estimators). Heterogeneity-robust difference-in-differences analyses show that release was associated with increased school segregation, improved mental health among Black children, and better self-reported health among White children. For heterogeneity-robust event-study analyses, school segregation increased steadily over time after release, with worse self-reported health and higher risk of asthma episodes among Black children aged 18 years or older after release. Black children’s mental health temporarily improved in the short term. In contrast, White children had improved self-reported health, mental health, and risk of asthma episodes in some years. Interventions to address the harms of school segregation are important for reducing racial health inequities.

Introduction

Recent studies have provided new evidence that school racial segregation worsens health for Black children during childhood and later in life. Black children exposed to higher levels of school racial segregation have greater risk of behavioral problems and alcohol use during childhood, with higher teen pregnancy rates, poorer self-rated health, and increased alcohol use during adulthood.^1-5 Suggested mechanisms include systematic disinvestment from majority-Black schools as a manifestation of structural racism, resulting in reduced school quality that constrains educational and occupational attainment later in life,^6-9 as well as increased racial isolation, exposure to discrimination, and stress that lead to poorer mental health and greater adoption of unhealthy coping behaviors.^10-14 Conversely, school segregation also could plausibly improve Black children’s health by reducing exposure to interpersonal racism from White peers or teachers.^15,16 More evidence is needed to unveil the association of school segregation with Black children’s health, because the overall impacts may be ambiguous, depending on which mechanisms dominate.

This issue is particularly important in light of recent US court decisions that increased levels of school racial segregation. In 1954, the Supreme Court’s landmark decision in Brown v Board of Education found that school segregation was unconstitutional, and more than 1000 school districts nationwide subsequently underwent court-ordered desegregation.^17-19 This desegregation improved access to school resources, educational and occupational attainment, and later-life health among Black children.^3,20 Beginning in 1991, however, the Supreme Court issued several decisions that made it easier for school districts to be released from court-ordered desegregation. About 400 of the approximately 650 school districts that were still under court oversight in 1991 have since been released.^17,19,21 School racial segregation grew steadily within 10 years of release.^19,22 Subsequently higher levels of segregation were associated with higher dropout rates among Black students and high rates of preterm birth among Black women.^22,23

Other than these studies, evidence on the health impacts of the end of court-ordered desegregation is sparse. Moreover, most work has examined the average change in health outcomes after the releases, rather than assessing trends over time, which would help us understand the timing of how such exposures “get under the skin” to affect health. In this study, we addressed this gap in the literature, using data from a large national survey and quasi-experimental methods intended to robustly assess the effects of policy changes. In particular, we used difference-in-differences (DiD) methods that leveraged the varying timing of court releases across districts. The latter variation is effectively quasi-random, because many arbitrary factors affected the timing of release procedures (eg, unequal court caseloads and varying duration of the release process across districts).^18,19,22 Besides traditional DiD methods, we also used novel DiD methods robust to heterogeneous treatment effects to address the increasing concern that traditional DiD models may generate biased estimates when treatment effects vary across groups or over time. Our study contributes important evidence to inform ongoing discussions of educational and public health policies that address school segregation and its impacts on racial health inequities.

Methods

Data and sample

We linked individual-level health outcome and demographic data from the 1997-2018 National Health Interview Survey (NHIS) with exposure data about school districts from several sources. The NHIS is an annual, nationwide, serial cross-sectional survey. We did not include earlier and more recent survey waves, because of differences in NHIS data collection techniques.^24,25 We linked respondents’ residential census blocks with data on school districts, using a crosswalk that mapped school district boundaries onto census blocks. School district data included school districts’ court order status (compiled by Reardon et al²⁶ and ProPublica²⁷), school district segregation measures derived from the Common Core of Data compiled by the National Center for Education Statistics,²⁸ and relevant demographic characteristics such as student racial composition and residential racial segregation from the decennial census. Nearly 90% of US children attend their assigned public school and thus are unlikely to be misclassified.²⁹

The sample was restricted to Black children and White children who were school-aged (5-17 years) and resided in school districts that were under court-ordered desegregation as of 1991. We focused on these 2 groups because the court orders were designed to address segregation among Black children and White children. Furthermore, other racial/ethnic groups have smaller sample sizes that would leave analyses underpowered. We further restricted the sample to those with at least 1 health outcome measured and no missing values for demographic and district characteristics. The final sample included 25 112 children (n = 8182 Black children; n = 16 930 White children; see sample selection flowchart and power calculation in Figure S1 and Appendix S1).

Exposure

Since the 1991 Supreme Court ruling on Dowell v Board of Education, school districts previously under court-ordered desegregation have gradually been released from oversight. Figure S2 illustrates the temporal variation in the number of school districts that were released each year during 1991-2018. The timing of these releases was effectively quasi-random, because many arbitrary factors affected release procedures, including unequal court caseloads across districts and the varying duration of the release process.^18,19,21,22 Previous studies showed that released school districts had similar characteristics (eg, baseline segregation levels, student racial composition) as unreleased districts.¹⁹ Therefore, releases from court-ordered desegregation may introduce exogenous variation in school segregation, and thus children’s health, with the association less likely to be confounded by individual and neighborhood factors. We considered children to be “exposed” if, at the time they were interviewed, they lived in a school district that had been released from court orders.

Outcomes

Outcomes included the level of school segregation in the school district and measures of children’s health that could plausibly be affected by school segregation. School segregation was measured by the Black-White dissimilarity index (range 0-1), a standard measure of segregation.^19,30 The index represents the proportion of Black or White students who would have to move to a different school within the district to achieve a uniform Black-White distribution across schools. Higher values represent more segregation.

Health outcomes included self-reported health (dichotomized as poor/fair/good vs very good/excellent), body weight captured using a z-score of age- and sex-adjusted body mass index (BMI), having mental health problems determined using standard cutoffs of the 6-item Strengths and Difficulties Questionnaire, and whether the child had an asthma attack or episode during the past 12 months (Appendix S1).

Several pathways may link school segregation with the outcomes we examined, based on epidemiologic theory and empirical work (Figure 1). For example, school segregation may reduce school quality: Highly segregated non-White schools tend to have poorer physical infrastructure (eg, less funding, fewer resources, less-skilled teachers)^31-33 that limits health-promoting activities (eg, exercise, health education, health services).^31,34-36 School segregation may also increase mental and emotional challenges for children exposed to discrimination and a stressful environment; previous evidence has shown that Black children in racially segregated schools experienced harsher disciplinary treatment and more frequent police contact,^37-39 which, in turn, may be linked with higher stress levels.⁴⁰ However, it is also plausible that attending segregated schools can improve Black children’s health by reducing exposure to interpersonal racism from White peers or teachers.^15,16 For White children, it is less clear a priori how school segregation may affect health, because there is little empirical work on this subject, although the concentration of resources in majority-White districts would likely provide health-promoting advantages.

Figure 1.

Causal diagram linking school racial segregation and child health. Boxes with double borders represent the key exposures and outcomes of interest.

Covariates

We adjusted regressions for covariates that may confound the relationship between residence in a particular school district and our outcomes. Individual characteristics included child’s age and sex; parent’s education, marital status, and employment; and family income. School district covariates correspond to the district’s characteristics in 1991 (at baseline prior to the 1991 Supreme Court decision), including total student enrollment, student racial composition, percentage of students eligible for free or reduced-price lunch, and residential racial segregation (measured by the Black-White dissimilarity index across census tracts within each district). We also included fixed effects (ie, indicator variables) for state of residence to adjust for any time-invariant state-level confounding, and fixed effects for year to adjust for underlying secular trends in the outcomes nationwide, as well as period-specific “shocks,” such as economic and social conditions, which, according to the literature related to court-ordered desegregation releases, are associated with educational outcomes.⁴¹ We do not include fixed effects at the district level because of the small number of observations in each cell (Appendix S1).

Statistical analyses

We leveraged geographic and temporal variation in districts’ releases from court-ordered desegregation to examine the association of the end of court-ordered desegregation with school segregation levels and child health. We used quasi-experimental analytical methods, including traditional DiD and novel DiD methods intended to address possible biases with traditional DiD.

Association of releases with children’s health: traditional DiD analyses

We first used traditional DiD analysis, a commonly used quasi-experimental method for examining the effect of a policy change. Traditional DiD examines pre-post differences in outcomes for the treatment group (ie, released districts), while “differencing out” secular trends in the control group (ie, districts not released). The underlying assumption is that trends in outcomes would be similar between treatment and control groups without the intervention (Appendix S1 and Table S1).

Specifically, we regressed each outcome on a binary variable indicating whether the district where the child resided when interviewed had been released from court-ordered desegregation, adjusting for covariates described above. School districts that were released in the last year of our study period (2018) were considered to be never treated. As is standard in DiD analyses, we estimated multivariable linear models for both binary and continuous outcomes. Logistic models were not used, because of differences in the interpretation of interaction terms in nonlinear models.⁴² For binary outcomes, coefficients can be interpreted as the percentage-point change in risk. We clustered SEs at the district level to account for correlated outcomes within treatment units. We stratified all models by race.

To further investigate how the association of court orders with the outcomes evolved over time, we carried out event-study analyses, an extension of DiD.⁴³ Rather than a binary exposure variable indicating whether a student’s district had been released, event-study models involved regressing the outcomes of interest on a set of leads and lags to the event time (eg, −2, −1, 0, 1, 2 years from district release). By comparing outcomes in different periods with a baseline reference period (set to −1, by convention), the coefficients on the leads and lags allow for estimation and visual presentation of impacts over time. With estimated coefficients of leads (ie, the pre-event period), event-study analyses evaluate differences in pre-event outcome trends by exposure as a test of whether the parallel trends assumption of DiD analyses may be satisfied (Appendix S1). As above, we estimated linear models stratified by race.

Novel DiD approach to account for potential biases

Recent work has shown that traditional DiD estimators may be biased when different units implement a treatment at different times and treatment effects vary across units or across time.^44,45 This is mainly due to the use of inappropriate comparator groups (ie, already-treated units). In the context of this study, the traditional DiD estimator will be biased if the effects of being released from court-ordered desegregation were different across school districts or across time. Recent work has suggested the former to be true: released districts in the South had larger subsequent increases in segregation.⁴⁶

Several novel estimators have been proposed recently.^47-50 Each uses slightly different methods to create more valid control groups for treated units (ie, not-yet-treated units and/or never-treated units), and the various methods generate roughly similar estimates.^51,52 In this study, we used the approach developed by Borusyak, Jaravel, and Spiess (BJS).⁵⁰ See Appendix S1 for details.

Secondary analyses

The health effects of release from court-ordered desegregation may differ by key characteristics such as age and gender. For example, disruptions in schooling and neighborhood may produce more harms to older children and boys, whose development may be more reliant on peers and social network.^53,54 Therefore, we performed BJS DiD analyses using samples stratified by age (5-10 vs 11-17 years) and sex (as a proxy for gender identity and exposure to gendered psychosocial experiences).⁵⁵ We also conducted secondary analyses to account for missing data, multiple hypothesis testing,⁵⁶ and a placebo test (Appendix S1).

Results

Descriptive summary

About 49% of the sample were girls (Table 1). The mean age at interview was 11 years. Black children had less-advantaged family backgrounds than White children, with lower family income and parents who were less likely to have a high school degree or be married or employed. Generally, Black children were more likely to have worse health than White children, with a lower percentage reporting very good/excellent health, a higher mean BMI, and a higher percentage reporting asthma episodes in the past 12 months. The mean level of school segregation to which children were exposed, measured by the Black-White dissimilarity index, was 0.52 among Black children and 0.50 among White children. That is, for Black children, about 52% of Black or White children in their districts would have to move to another school in the same district to reach even racial distribution across schools.

Table 1.

Descriptive statistics.^a

	Black children	White children
	% or mean (SD)	% or mean (SD)
Student characteristic
Very good/excellent health	74.8	82.1
Body mass index z-scoreb	0.70 (1.11)	0.56 (1.10)
Mental health problemsc	9.3	8.8
Asthma episode in past 12 monthsd	8.5	5.4
Female sex	49.3	48.9
Age (years)	11.24 (3.76)	11.19 (3.79)
At least 1 parent has high school degree	77.7	83.6
Parents married	37.0	67.8
At least 1 parent employed	76.5	87.5
Family income ($US)
<20,000	31.9	19.1
20,000-45,000	34.6	30.6
45,000-75,000	17.1	19.5
>75,000	16.5	30.8
School district characteristic e
School segregationf	0.52 (0.22)	0.50 (0.22)
Total student enrollment	216 636 (299 609)	232 278 (288 213)
Black students, %	42.9	28.2
White students, %	40.0	44.1
Hispanic students, %	13.8	23.6
Eligible for free lunch, %	43.6	41.6
Residential segregationg	0.62 (0.19)	0.61 (0.19)
No. of observations	8182	16 930

^a Sample was drawn from the 1997-2018 waves of the National Health Interview Survey and includes Black and White children who resided in districts that had been under the desegregation order in 1991.

^b Body mass index represents the z-score adjusted by age and sex for children aged 12-17 years.

^c Mental health problems represent whether the child’s total Strengths and Difficulties Questionnaire score was ≥6.

^d Asthma episode represents whether the child had an asthma attack or episode during the past 12 months.

^e School district characteristics represent the 1991 characteristics for the district, and residential segregation represents the level of segregation in 1990.

^f School segregation was measured by the Black-White dissimilarity index across schools within each district.

^g Residential segregation was measured by the Black-White dissimilarity index across census tracts within each district.

Association of releases with outcomes: traditional vs BJS DiD analyses

In traditional DiD analyses (Table 2, column 1), there was no association of court-ordered releases with school segregation. In BJS DiD analyses (Table 2, column 2), releases were associated with increased segregation levels (0.068; 95% CI, 0.028-0.109): an approximately 13% increase from the baseline mean. For Black children, in traditional DiD analyses, releases were not associated with any changes in health, but in BJS DiD analyses, releases were associated with a lower probability of mental health problems (−0.037; 95% CI, −0.070 to −0.003), a 39% reduction from the baseline mean. For White children, in both traditional and BJS DiD analyses, releases were associated with a higher probability of very good/excellent health (0.020 [95% CI, 0.003-0.037] and 0.017 [95% CI, 0.000-0.034], respectively), equivalent to an approximately 2% change.

Table 2.

Association of the end of court-ordered desegregation with school segregation and children's health.^a

	Coefficient (95% CI)
Outcome	Traditional DiD	BJS DiD
School segregationb	0.015 (−0.006, 0.036)	0.068*** (0.028, 0.109)
Black children's health
Very good/excellent health	−0.01 (−0.037 to 0.017)	−0.017 (−0.046 to 0.011)
Body mass indexc	−0.08 (−0.187 to 0.028)	−0.07 (−0.164 to 0.024)
Mental health problemsd	−0.009 (−0.032 to 0.014)	−0.037* (−0.070 to −0.003)
Asthma episodee	−0.008 (−0.027 to 0.011)	−0.009 (−0.028 to 0.011)
White children's health
Very good/excellent health	0.020* (0.003, 0.037)	0.017* (0.000, 0.034)
Body mass index	−0.036 (−0.105 to 0.033)	0.002 (−0.067 to 0.070)
Mental health problems	−0.002 (−0.017 to 0.014)	−0.015 (−0.035 to 0.005)
Asthma episode	−0.003 (−0.012 to 0.005)	−0.005 (−0.014 to 0.005)

Abbreviations: BJS, Borusyak, Jaravel, and Spiess; DiD, difference-in-differences.

^* P < .05, ^**P < .01, ^***P < .001.

^a Sample was drawn from the 1997-2018 waves of National Health Interview Survey and includes Black and White children who resided in districts that had been under the desegregation order in 1991. All models were adjusted for 1991 school district characteristics, residential segregation, individual characteristics, and fixed effects for year and state. SEs were clustered at the district level. Person-year observations vary by health outcomes and range from 3744 to 8182 among Black children and 7622 to 16 930 among White children.

^b School segregation was measured by the Black-White dissimilarity index within each district.

^c Body mass index represents the z-score adjusted by age and sex for children aged 12-17 years.

^d Mental health problems represent whether the child’s total Strengths and Difficulties Questionnaire score was ≥6.

^e Asthma episode represents whether the child had an asthma attack/episode during the past 12 months.

Event-study results illustrate how release from court oversight was associated with school segregation levels, from 8 years before to 19 years after the release occurred (Figure 2). Reassuringly, there were no differences between treatment and control groups during the prerelease period, suggesting parallel pre-event trends. In the postrelease period, the magnitude of the association between releases and school segregation increased over time. The traditional event-study method suggests releases were positively associated with school segregation after 12 years following release, whereas the more robust BJS method suggests a positive association after 6 years. The magnitude of estimated coefficients using the BJS method was also larger than coefficients from traditional models.

Figure 2.

Association of release from court-ordered desegregation with school segregation level. Sample was drawn from the 1997-2018 waves of National Health Interview Survey and includes Black children and White children who resided in districts that had been under the desegregation order in 1991. School segregation was measured by the Black-White dissimilarity index within each district. Estimates were derived from traditional and Borusyak, Jaravel, and Spiess (BJS) event-study models adjusting for 1991 school district characteristics, residential segregation, individual characteristics, and fixed effects for year and state. SEs were clustered at the district level. The estimated coefficient for the year prior to treatment was omitted in the traditional method (default) but not in the BJS method, due to different estimation approaches.

Figure 3 shows the association of releases with children’s health over time. Among Black children (Figure 3A), releases were associated with decreased reporting of very good/excellent health about 18 years after release (BJS DiD, −0.14; 95% CI, −0.20 to −0.74). Releases had a more immediate effect on reducing Black children’s mental health problems (BJS DiD, −0.061; 95% CI, −0.095 to −0.027), but the effect faded soon after. For asthma, Black children were more likely to report having asthma episodes 19 years after release (BJS DiD, 0.082; 95% CI, 0.023-0.14). Releases were not associated with children’s BMI. Overall, the magnitude of BJS DiD estimators was slightly larger than traditional estimators for all health outcomes.

Figure 3.

Association of release from court-ordered desegregation with children’s health, by race. Sample was drawn from the 1997-2018 waves of the National Health Interview Survey and includes Black children (A) and White children (B) who resided in districts that had been under the desegregation order in 1991. Body mass index represents the z-score adjusted by age and sex for children aged 12-17 years. Mental health problems represent whether the child’s total Strengths and Difficulties Questionnaire score was ≥6. Asthma episode represents whether the child had an asthma attack or episode during the past 12 months. Estimates were derived from traditional and Borusyak, Jaravel, and Spiess (BJS) event-study models adjusting for 1991 school district characteristics, residential segregation, individual characteristics, and fixed effects for year and state. SEs were clustered at the district level. The estimated coefficient for the year prior to treatment was omitted in the traditional method (default) but not in the BJS difference-in-differences method, due to different estimation approaches.

Among White children (Figure 3B), releases were associated with increased reporting of very good/excellent health 4-5 years following release (BJS DiD, 0.053 [95% CI, 0.024-0.082] and 0.032 [95% CI, 0.0008, 0.063], respectively), but not in other years. In BJS DiD models, White children in released districts were less likely to have mental health problems compared with those in nonreleased districts, although effects were not consistent across time. Similar results were found for asthma. Releases were not associated with White children’s BMI. Again, the magnitude of BJS DiD estimators was slightly larger than traditional estimators for all health outcomes.

Secondary analyses

In BJS DiD models stratified by age (Figure 4), releases were associated with fewer mental health problems for Black children aged 11-17 years (−0.055; 95% CI, −0.096 to −0.015) but not Black children aged 5-10 years. In BJS DiD models stratified by sex (Figure 5), releases were associated with a lower probability of mental health problems for Black female children (−0.038; 95% CI, −0.069 to −0.006) but not for Black male children.

Figure 4.

Association of release from court-ordered desegregation with children’s health, by race and age. ^*P < .05. Sample was drawn from the 1997-2018 waves of the National Health Interview Survey and includes Black children and White children who resided in districts that had been under the desegregation order in 1991. Mental health problems represent whether the child’s total Strengths and Difficulties Questionnaire score was ≥6. Asthma episode represents whether the child had an asthma attack or episode during the past 12 months. Estimates are derived from stratified Borusyak, Jaravel, and Spiess (BJS) difference-in-differences models adjusting for 1991 school district characteristics, residential segregation, individual characteristics, and fixed effects for year and state. SEs were clustered at the district level.

Figure 5.

Association of release from court-ordered desegregation with children’s health, by race and sex. ^*P < .05. Sample was drawn from the 1997-2018 waves of the National Health Interview Survey and includes Black children and White children who resided in districts that had been under the desegregation order in 1991. Body mass index represent the z-score adjusted by age and sex for children aged 12-17 years. Mental health problems represent whether the child’s total Strengths and Difficulties Questionnaire score was ≥6. Asthma episode represents whether the child had an asthma attack or episode during the past 12 months. Estimates are derived from stratified Borusyak, Jaravel, and Spiess (BJS DiD) models adjusting for 1991 school district characteristics, residential segregation, individual characteristics, and fixed effects for year and state. SEs were clustered at the district level.

After accounting for multiple hypothesis testing, the BJS DiD estimates for older and female Black children were not statistically significant (Table S2). The main findings in BJS DiD analyses event-studies remained similar (Tables S3-S10) except for the mental health effects, which were significant only in the first year after releases among Black children and in the 15th year after releases among White children. Results of other secondary analyses are reported in Appendix Tables S1 and S12).

Discussion

In this study, we used rigorous, novel, quasi-experimental methods to show that releases from court-ordered desegregation were associated with steadily increased school segregation in the past 2 decades. The longer that districts were released, the more likely Black children in those districts were to report worse self-reported health and asthma episodes. Black children’s mental health improved in the immediate aftermath, but these improvements then faded. For White children, self-reported health, mental health, and having asthma episodes in past 12 months were improved in some circumstances, but the patterns were less consistent over time.

We found a long-lasting effect of releases from court-ordered desegregation on school segregation: The Black-White dissimilarity index increased by 0.15 in districts that had been released for nearly 20 years, compared with districts that had not been released. The estimated effect is comparable to those reported in prior studies, which found that the index increased by 0.05-0.08 within 10 years of release.^19,22 This suggests that the dismantling of efforts to desegregate schools—through busing and other integration efforts—took place gradually and may still be underway in recently released districts.

Although school segregation increased shortly after releases, the effects on child health only started to emerge many years after release. For example, we observed a declining trend of very good/excellent health and an increase in asthma episodes among Black children after year 15, peaking in year 19. Several factors may explain the lag in effects. First, it may take many years for a school’s resources (eg, physical infrastructure) to deteriorate enough to adversely affect students’ health. Second, our observations indicate a gradual escalation of segregation, implying that children in districts released earlier might have attended more profoundly segregated schools, potentially leading to more adverse outcomes. Additionally, students residing in districts that were released earlier may have been exposed to segregation for a longer period. The significant effects on health may only become apparent in later periods, when the cumulative impact reaches a critical threshold. Indeed, our finding of worsened asthma rates that lagged the changes in school segregation is consistent with prior studies, showing that characteristics and indoor environmental quality of a school building may explain school-level variation in self-reported upper respiratory symptoms and missed school days.⁵⁷ Additionally, adequate health care services provided in schools (eg, nurses) reduce unmet need for health care services.³⁶ Without intervention, increasing school segregation may continue to shift educational and other resources to majority-White schools, increasing racial disparities in health.^7,31-33,58

Meanwhile, Black children’s mental health was potentially briefly improved after releases, which may be explained by several competing forces. On the one hand, racial segregation may reduce exposure to interpersonal racism from White peers or teachers^15,16 and increase exposure to stronger social networks, improving Black children’s mental health. On the other hand, school racial segregation may increase stress due to increased discrimination from a sense of exclusion and/or harsher treatment and discipline of students at racially segregated schools, part of the “school-to-prison pipeline.”^37-39 Having fewer school resources (eg, adequate facilities, mental health support) may also worsen mental health. Our event-study results show that Black children’s mental health only improved in the first years after school districts were released, but not afterward. This suggests that Black children may have initially viewed the end of court-ordered desegregation as a positive movement, anticipating reduced interpersonal racism, leading to improved mental health. However, as time progressed, the benefits of avoiding interpersonal racism may no longer outweigh the harms of the structural facets of racism. Our results also show that releases were associated with a lower probability of having mental health problems for Black female children but not for Black male children, potentially due to differing mechanisms at play, which may affect genders differently. For example, reduced interpersonal racism from attending more-segregated schools could be perceived as more beneficial for the mental health of Black girls compared with Black boys. However, these findings should be interpreted with caution because they did not remain statistically significant after accounting for multiple hypothesis testing.

For White children, our study shows a generally positive effect of releases from court-ordered desegregation on self-reported health. However, this impact is not as consistent compared with the pattern among Black children (eg, no clear trend, the effect did not occur at the same time as among Black children). This difference could be due to the existence of multiple pathways linking school segregation to children’s health, which may affect racial groups differently, with varying time frames for manifestation. For instance, the physical infrastructure in schools may take several years to deteriorate, whereas the effects of discrimination may become apparent more quickly. The clearer trends among Black children, in contrast to White children, may suggest a more pronounced impact on Black children. Supporting this, a previous study that examined the effects of court-ordered desegregation during the 1960s-1980s found Black children experienced increased education resources and improved self-reported adult health, whereas White students saw no changes in these outcomes as a result of these court decisions.³

The present study has several strengths. First, we used a large nationwide survey data set with rich individual-level data to address a timely policy-relevant question related to race-based inequities in the education system, with important implications for racial inequities in health. Using more comprehensive data and a larger window within which school districts were released from court oversight, we extended previous studies that examined the policy impact on school resegregation within a limited period, we examined more granular effects over time using a larger data set, and we added evidence more generally on the policy impact on children’s health. Second, our analyses point toward the need for epidemiologists to use novel DiD methods to account for potential biases in traditional methods.⁵⁹

Our study also has limitations. First, outcomes and covariates were self-reported and, therefore, may be subject to standard reporting biases. Future studies could examine objective measures of health.⁶⁰ Second, our results could mask the dynamic experience of school segregation because we only observed school segregation and health outcomes at the time children were interviewed. Third, our main model is only powered to detect medium effect sizes. The results stratified by sex and age may be underpowered to detect small effects, and results should be interpreted cautiously. Fourth, although our event-study results did not detect a violation of the parallel trends assumption, we acknowledge that this may be due to low power.⁶¹ Methodologies have been proposed for more robust inference for DiD and event-study designs (eg, the “HonestDiD” approach⁶²). Unfortunately, such tests are not currently compatible with the BJS estimation approach we used. Finally, a limitation of all quasi-experimental analyses is the possibility of residual confounding. For example, parallel laws, policies, or practices, especially those representing manifestations of structural racism, may have co-occurred at the same time as the releases, although this is less likely given the large number of districts and the numerous sources of variation included in our sample.

To conclude, the school environment is an important determinant of child health, and our study demonstrates that school segregation contributes to long-term racial inequities in health. Future work should examine the impacts of structural interventions to improve school environments for Black children, such as providing extra resources to segregated schools, reducing discrimination and harsh treatment targeted at Black children, and promoting school racial integration like initiatives proposed in the U.S. Congress’s Strength in Diversity Act.⁶³

Supplementary material

Supplementary material is available at American Journal of Epidemiology online.

Funding

All phases of this study were supported by National Institutes of Health grant R01HL151638.

Conflict of interest

The authors declare no conflicts of interest.

Data availability statement

This study used the restricted data from the National Health Interview Survey. The authors do not have permission to share these data.