Obesity Disparities in California Vary by School Ethnic Segregation and School Neighbourhood Income

ABSTRACT

Objective

This study examines how ethnic disparities in childhood obesity differ across combinations of school ethnic composition and school neighbourhood income in California, to identify the contexts in which disparities are greatest. Prior studies have been unable to study this due to the high correlation between segregation and income.

Methods

This observational study focuses on overweight/obesity data objectively measured among California public school children in 2018/2019 and school neighbourhood data from the American Community Survey, using a ‘within versus between’ differences analysis.

Results

Black and Latino children had higher overweight/obesity prevalence than White children, but disparities varied across schools. Disparities were smallest in integrated low‐income schools: the Black versus White (B/W) and the Latino versus White (L/W) disparities were 3% and 10%, respectively. In integrated high‐income schools, these disparities were 8% and 13%, respectively. Disparities between segregated low‐income schools and segregated high‐income schools were 11% (B/W) and 14% (L/W), and 22% (B/W) and 26% (L/W). The largest disparity was between minority children in minority‐segregated low‐income schools compared to White children in White‐segregated high‐income schools (B/W disparity 23% and L/W 28%).

Conclusions

Place‐based interventions to reduce childhood obesity should prioritize low‐income, ethnically segregated schools, where the largest disparities were observed.

1. Introduction

In the United States, eliminating health disparities and addressing the social determinants of health are major public health goals [1, 2, 3]. To advance efforts in these areas, it is critically important to focus on ethnic disparities that develop early in life. Among these, childhood obesity is a particularly important target because marked ethnic disparities in child obesity persist across the life course, contributing to downstream health inequities. For example, compared with White children, Black and Latino children have nearly twice the rates of obesity [4, 5]; and these higher obesity rates persist into adulthood, leading to a greater lifelong risk of comorbid diseases, such as Type II diabetes and cardiovascular disease [6, 7].

Although multiple factors contribute to childhood obesity, broader structural inequities across racial and ethnic groups are important drivers of disparities [8]. The ‘place, not race’ theoretical framework argues that health disparities are driven primarily by the social, economic and environmental conditions of the places where people live rather than by individual‐level race itself [9, 10]. It proposes that historical and structural forms of racism, such as residential segregation and discriminatory housing policies, have concentrated many ethnic minority populations into neighbourhoods with fewer resources and greater health risks, making place a key pathway through which health inequities emerge [9, 10]. For example, children living in segregated neighbourhoods with underfunded schools, limited access to safe parks and recreational spaces, and less access to healthy foods are more likely to have poor health [11, 12, 13, 14]. Hence, the framework emphasizes that reducing place‐level inequities is essential for improving health disparities [9, 10].

Building on this framework, previous research has demonstrated that two key place‐level factors associated with ethnic disparities in childhood obesity are school ethnic segregation and school neighbourhood income [11, 12, 13, 14, 15]. Black and Latino children who attend minority‐segregated schools have a much greater risk of obesity than children attending White‐segregated schools, whereas disparities are substantially reduced when children attend the same integrated schools [11, 12, 13, 14]. Similarly, children attending schools in low‐income neighbourhoods have a higher risk of obesity than those attending schools in high‐income neighbourhoods [15]. However, school ethnic segregation and neighbourhood income are highly correlated due to the effects of structural racism and disinvestment in neighbourhoods with greater ethnic minority populations [1, 2, 16, 17]. As such, disentangling the independent and joint contributions of these place‐level factors on childhood obesity disparities has been methodologically challenging due to data limitations.

Consequently, it remains unclear whether disparities attributed to school ethnic segregation primarily reflect disparities related to differences in neighbourhood income, versus whether both factors independently contribute to childhood obesity disparities. To address this gap in the literature, we use data from California, a state that is uniquely diverse, to expand on our previous work showing that ethnic disparities in childhood obesity vary across levels of school integration [11]. This ecological study uses obesity prevalence data, disaggregated to ethnic groups within schools, to address the following questions: (1) Are there ethnic disparities in obesity within ethnically integrated schools (i.e., ‘within‐school disparities’), and how do these disparities vary further by school neighbourhood income? (2) Are there ethnic disparities in obesity when comparing schools of varying ethnic composition (i.e., ‘between‐school disparities’) in neighbourhoods of similar income levels? And, how do these disparities vary across different levels of school neighbourhood income? By addressing these research questions, this study seeks to identify which combinations of school ethnic composition and neighbourhood income result in the biggest disparities in child obesity, and disentangle the independent role of school ethnic segregation.

2. Materials and Methods

2.1. Study Population

This study utilizes cross‐sectional data on Black, Latino and White children enrolled in fifth, seventh or ninth grade in California public schools during the 2018/2019 year, who participated in the California FitnessGram test [18]. The de‐identified FitnessGram datasets are available from the California Department of Education (CDE) and provide data at the school level, which is also broken down by various subgroups including ethnicity [19]. As such, data are at the school‐subgroup level (i.e., a measure for all White children in that school, a separate prevalence measure for Black children in the same school, etc.). All available data were used, and students were only excluded from analysis if they had missing FitnessGram data, which was primarily due to data masking by the CDE. To ensure confidentiality, the CDE marks data as ‘missing’ when there are fewer than 10 children within each subgroup of grade and ethnicity at a given school. Based on enrolment data on the CDE website [20], in SY 2018/2019 there were 1 204 070 Black, Latino or White children enrolled in fifth, seventh and ninth grades in California public schools (78 119 Black children, 797 292 Latino children and 328 659 White children). FitnessGram data were available for 1 063 737 (88%) of these children (54 170 Black children, 719 911 Latino children and 289 656 White children). Given that this study utilizes publicly available, de‐identified data, it does not meet the definition of human subjects research.

2.2. Measures

2.2.1. Outcome

The FitnessGram assessment is a mandatory test that included measures of body mass index (BMI) until the 2018/2019 school year [18]. FitnessGram assessments were administered by trained school staff, who followed standard procedures to objectively measure students' height and weight. Students typically remove their shoes, and measurements are recorded to the nearest whole inch and pound. Height is measured twice for accuracy. Data are entered into the FitnessGram online platform using unique student identification numbers and height and weight data are used to calculate BMI using standard formulas [21]. These body weight measures are valid and reliable: > 80% agreement between teacher and trained experts has been found [22]. Using the Centers for Disease Control and Prevention's (CDC) age‐ and sex‐specific BMI reference distributions, children in the FitnessGram dataset are categorized into three groups that correspond to the CDC's categories of ‘normal weight’, ‘overweight’ and ‘obesity’, respectively [18]. The term ‘overweight/obesity’ is employed to refer to the overweight and obesity categories combined.

2.2.2. School Ethnic Segregation

We used three different segregation measures to compare overweight/obesity rates for each group of children of colour (Black and Latino children) with those of White children across segregation levels, thus helping ensure that study findings are robust to the definition of segregation measures. Notably, the measures we chose cover a range of extremes in segregation. For our first measure, we maximize the use of available FitnessGram data and use a threshold for segregation that is most extreme, representing a very low threshold for integration and very high threshold for segregation, as we have done before [11]. In this first, purely empirical approach, we defined: (1) ‘Integrated schools’ as those with more than 10 students from both the minoritized group and White students. This approach ensures sufficient (unmasked) data for both groups and also represents a low threshold for being categorized as ‘integrated’ because a school could, for example, have 11 Black children and 1000 White children and still receive that label. (2) ‘Minority‐segregated schools’ were defined as those that had more than 10 students from the minoritized group but 10 or fewer White students. This approach ensures that all schools that do not have enough data for White students (due to masking) are included as part of the minority‐segregated schools—that is, it ensures inclusion of the likely most segregated schools into the ‘minority‐segregated’ category given that these schools do not have sufficient White population to ascertain obesity prevalence among White children. (3) ‘White‐segregated schools’ had more than 10 White students but 10 or fewer students from the minoritized group; in this category, the reversed argument for maximizing inclusion of minoritized schools applies to White‐segregated schools. Schools with 10 or fewer students in both groups were excluded from the analysis due to masked outcome data.

The second measure of segregation was developed using the index of concentration at the extremes (ICE) [23] that was computed separately for each ethnic group. This index is a widely used measure of polarization contrasting more privileged versus most social deprived groups, and has been used as a proxy of structural racism [24]. It ranges from −1 to 1; in our case, −1 reflects that no White children are enrolled in the school, and 1 is interpreted as only White children are enrolled. The ICE was then categorized into tertiles to classify schools as segregated toward children of colour (lowest tertile), integrated (middle tertile) and White‐segregated (upper tertile).

This resulted in a more moderate threshold for segregation versus integration: Taking the Latino versus White comparison as an example, schools in the lowest (most minority‐segregated) tertile had an average of 83% Latino and 3% White students; schools in the middle (integrated) tertile had an average of 56% Latino and 19% White students and schools in the upper (most White‐segregated) tertile had an average of 23% Latino and 52% White students.

Third, we used a definition of segregation that is based on simple majorities [25] and represents the most strict criteria for integration. We categorized schools as integrated when neither ethnic group being compared (e.g., neither White nor Latino) had a majority (> 50%) of the school's enrolled student population. Segregated schools required enrolment of the respective ethnic group to be a majority (> 50%).

2.2.3. School Neighbourhood Income

The median annual household income of each school's neighbourhood was obtained from the 2015 American Community Survey [26], at the census tract level. School neighbourhood income was categorized into tertiles based on the median household income distribution of the neighbourhoods of all schools included in the analysis to represent low, medium and high school neighbourhood income. Categorization was chosen because the study aimed to compare obesity disparities across substantively distinct neighbourhood income contexts rather than estimate a continuous dose–response relationship. Tertiles also ensured adequate numbers of schools within each segregation category while facilitating direct comparisons across combinations of school segregation and neighbourhood income.

To further examine the robustness of the income categorization, we also examined an alternative definition that both accounts for potential income outliers and examines more stringent definitions of low‐ and high‐income. Low‐income schools were defined as those falling between the 1st and 10th percentile of the median household income distribution (i.e., excluding outliers < 1st percentile and lowering the threshold to be considered low‐income from the 33rd to the 10th percentile of the income distribution). Similarly, but on the opposite side of the income distribution, the high‐income schools were defined as those that had median household income between the 90th and 99th percentile.

2.3. Analyses

Descriptive statistics were calculated for all schools, by school segregation and income categories and inclusion versus exclusion status due to masking. To explore the distinct patterns of overweight/obesity disparities among combinations of school segregation and income, we examined disparities by schools' segregation/integration within categories of neighbourhood income (high and low). For each combination of ethnic group and income level, two disparity measures were calculated. First, to answer Research Question 1, we estimated the ‘within‐school disparity’ by comparing the prevalence of overweight and obesity among each minoritized group to that of White students within the same integrated schools. We then computed weighted averages across all integrated schools to reflect disparities among students who share similar school environments. Specifically, using the Latino versus White analysis as an example, to calculate the weighted averages, we calculated the difference of two proportions (Latino vs. White) within each integrated school $\left(d_i=p_{\mathit{Ai}}-p_{\mathit{Wi}}\right)$ and the variance of the disparity $\left[v_{\mathit{di}}=p_{\mathit{Ai}}\left(1-p_{\mathit{Ai}}\right)/N_{\mathit{Ai}}+p_{\mathit{Wi}}\left(1-p_{\mathit{Wi}}\right)/N_{\mathit{Wi}}\right]$, where p _Ai represents the prevalence of overweight (or obesity) for Latino children in school i, N _iA represents the sample size of Latino children for school i and p _Wi represents the prevalence of overweight (or obesity) for White children in the same integrated school i. Then, the average within‐school disparity estimate is obtained as the weighted average of the within‐school disparities, where weights are proportional to sample size [$W_i=N_{\mathit{iW}}{N}_{\mathit{iA}}/\left(N_{\mathit{iW}}+N_{\mathit{iA}}\right)$, where N _iW and N _iA represent the number of Latino children and White children in school i with available data, respectively], that is, $\text{disparity}=\sum \left(w_i{d}_i\right)/\left(\sum w_i\right)$. The standard error for the disparity is estimated as sqrt(V), where $V=\sum \left(w_i^2{v}_{\mathit{di}}\right)/{\left(\sum w_i\right)}^2$ is the variance of the weighted average.

Next, to answer Research Question 2, we estimated the ‘between‐school disparity’—which is a weighted average that contrasts overweight/obesity rates among children of colour in minoritized schools (e.g., Black‐ or Latino‐segregated) with those of White students in White‐segregated schools. Again using the Latino versus White analysis as an example, the between‐school disparities are estimated by first calculating the prevalence of overweight (or obesity) among Latino children in minority‐segregated schools using a weighted average of prevalence estimates for segregated schools: $p_A=\sum \left(w_j{p}_{\mathit{Aj}}\right)/\left(\sum w_j\right)$, where $p_{\mathit{Aj}}$ is the prevalence of overweight (obesity) among Latino children in minority‐segregated school j, and $w_j=N_{\mathit{jA}}$ is the number of children with available data. The standard error of the estimate is calculated as sqrt(V _A ), where $V_A=\sum \left[w_j{p}_{\mathit{Aj}}\left(1-p_{\mathit{Aj}}\right)\right]/{\left(\sum w_j\right)}^2$. Similarly, for White children, $p_W=\sum \left(w_k{p}_{\mathit{Wk}}\right)/\left(\sum w_k\right)$, where p _Wk is the prevalence of overweight (obesity) among White children in White‐segregated school k. The disparity between segregated schools is thus, between.disparity = p _A  − p _W ; its standard error is sqrt(V _A  + V _W ).

Notably, our analysis does not utilize a multivariable approach. The ‘within‐school’ analysis does not need regression adjustment because, by computing disparities between students who attend the same school, unobserved school‐level variables are inherently controlled for. For the ‘between‐school analysis’, we do not conduct regression analyses to control for other school‐level factors because, as posited by the theory of ‘place, not race’ that motivated this research, school segregation and school neighbourhood income are jointly associated with a myriad of downstream characteristics of individual‐, school‐ and neighbourhood‐level factors that influence health. Controlling for any of these additional variables would yield the disparity that is left unexplained by these additional influences, akin to adjusting for mediators. Thus, adjustment does not serve the primary purpose of this study—that is to describe overall disparities associated with school segregation and how they vary by school neighbourhood income. Analyses were conducted using SAS version 9.4 (SAS Institute Inc.).

2.3.1. Sensitivity Analyses

Sensitivity analyses described in the Supporting Information were conducted to account for the masking of data, which is more likely among schools with lower enrolment. In these analyses, we used the first segregation measure, but only among schools with enrolment above the first quartile of enrolment across all schools, which minimized masking.

3. Results

Ethnic groups were unevenly distributed across integrated and segregated schools in low‐income and high‐income neighbourhoods. For example, among the total children with available data, 17.9% of Black and 24.1% of Latino children attended minority‐segregated schools in low‐income areas, whereas only 4.9% and 4.0%, respectively, attended minority‐segregated schools in high‐income areas (Table 1).

TABLE 1.

School characteristics, split by school neighbourhood income and level of segregation.

	Low‐income neighbourhoods				High‐income neighbourhoods
	Within analysis	Between‐school analysis		Excluded a	Within analysis	Between‐school analysis		Excluded a
	Integrated	White segregated	Minority segregated		Integrated	White segregated	Minority segregated
Black–White comparison
Black students, n (% b )	8301 (15.3)	—	9680 (17.9)	—	14 901 (27.5)	—	2633 (4.9)	—
White students, n (% b )	13 209 (4.6)	29 765 (10.3)	—	—	57 009 (19.7)	105 174 (36.3)	—	—
Schools (total n = 9163), n (% c )	230 (2.5)	607 (6.6)	379 (4.1)	1773 (19.3)	492 (5.4)	1692 (18.5)	93 (1.0)	711 (7.8)
School enrolment, n	1300	636	654	469	1400	711	560	423
% FRPM	78.2	66.5	89.0	85.0	41.5	28.7	70.3	53.0
% of enrolled students who are Black	11.2	2.5	22.7	4.4	8.2	2.2	29.4	4.4
% of enrolled students who are White	16.2	38.8	3.3	12.0	28.4	42.0	6.4	19.0
% of enrolled students who are Latino	61.5	46.4	66.0	75.0	37.7	29.8	44.6	51.0
Median neighbourhood household income (mean, $)	35 650	38 056	33 536	35 704	97 543	105 745	90 589	94 860
Neighbourhood % with > 16 years of education	13.6	17.0	11.4	12.01	43.3	49.0	39.8	41.05
Charter school, N	19	112	68	282	35	132	10	69
Schools in lower decile of income (n)	46	90	96	351	—	—	—	—
Schools in upper decile of income (n)	—	—	—	—	114	443	11	97
Lower decile median Neigh'd HH Inc. (mean, $)	29 795	30 104	30 092	29 706	—	—	—	—
Upper decile median Neigh'd HH Inc. (mean, $)	—	—	—		115 774	121 034	115 288	118 618
Latino–White comparison
Black students, n (% b )	101 916 (14.2)	—	173 789 (24.1)	—	147 973 (20.6)	—	28 998 (4.0)	—
White students, n (% b )	36 614 (12.6)	6360 (2.2)	—	—	143 528 (49.6)	18 655 (6.4)	—	—
Schools (total n = 9163), n (% c )	708 (7.7)	129 (1.4)	1724 (18.8)	428 (4.7)	1841 (20.1)	343 (3.7)	506 (5.5)	298 (3.2)
School enrolment, n	901	368	585	166	928	539	512	314
% FRPM	71.2	61.6	88.2	76	34.4	16.3	61.6	44
% of enrolled students who are Black	5.5	1.6	7.1	10.1	3.8	2.3	6.2	9.2
% of enrolled students who are White	28.0	58.2	4.4	33.1	36.7	50.9	10.5	30.5
% of enrolled students who are Latino	55.0	26.3	81.4	41.1	34.9	14.3	59.6	33.7
Median neighbourhood household income (mean, $)	37 351	37 635	35 042	36 452	100 874	120 123	91 186	99 766
Neighbourhood % with > 16 years of education	15.57	18.75	10.85	16.14	45.65	58.61	37.96	45.92
Charter school, n	87	44	267	83	124	43	49	30
Schools in lower decile of income (n)	84	23	313	52	—	—	—	—
Schools in upper decile of income (n)	—	—	—		489	135	56	65
Lower decile median Neigh'd HH Inc. (mean, $)	30 372	30 458	30 179	30 179	—	—	—	—
Upper decile median Neigh'd HH Inc. (mean, $)	—	—	—	—	121 034	124 760	118 358	121 147

^a For the Black–White comparison, excluded schools had no obesity data for White and Black students; for the Latino–White comparison, excluded schools had no obesity data for White and Latino students.

^b Row percent based on total n of Black (n = 57 170), Latino (n = 719 911) or White (n = 289 656) children with available data.

^c Row percent based on n = 9163 total schools.

Even within low‐ and high‐income strata, segregated and integrated schools were located in neighbourhoods with different median household income levels (Table 1). For example, for schools in high‐income neighbourhoods, the mean of the median household income for the neighbourhoods of White‐segregated schools was $15 000 higher than Black‐segregated schools and $8000 higher than Black–White integrated schools; as well as $30 000 higher than Latino‐segregated schools and $19 000 higher than Latino–White integrated schools.

The characteristics of included schools and schools excluded due to masking were consistent with schools' segregation classification (Table 1). A total of 2484 schools were excluded from the Black–White comparison, and 726 schools were excluded from the Latino–White comparison. Compared to included schools, excluded schools tended to have smaller enrolment overall, and an ethnic composition that was consistent with the definition of segregation. For example, for the Latino versus White analysis, excluded schools had lower enrolment of Latino students compared to the schools that were included and classified as integrated or Latino‐segregated, and lower enrolment of White students compared to schools classified as White‐segregated. Excluded schools were located in neighbourhoods with median household income generally similar to or slightly lower than included schools classified as integrated.

3.1. Overweight and Obesity Disparities

When using the first segregation definition and the primary categorization of income, we observed substantial variation in the Black versus White disparities in combined overweight and obesity prevalences across school segregation and income levels (Figure 1). The disparity was smallest within integrated schools in low‐income neighbourhoods (3%) and high‐income neighbourhoods (8%). The disparity was larger between segregated schools located in low‐income neighbourhoods (11%), but much wider (22%) between segregated schools located in high‐income neighbourhoods. Black children in Black‐segregated schools located in low‐income neighbourhoods had the highest prevalence (46% overweight and obesity prevalence), whereas White children in White‐segregated schools in high‐income neighbourhoods (23%) had the lowest prevalence, yielding a disparity of 23%.

FIGURE 1.

Overweight/obesity disparities in Black versus White children in California public schools. (Top) Overweight/obesity prevalence among Black and White children in California. (Bottom) Disparities within integrated schools and disparities between segregated schools. Left column: Low‐income schools. Right column: High‐income schools.

In the Latino–White comparison, patterns of overweight and obesity prevalence and disparities were similar to the Black–White comparison, but disparities were larger due to a higher prevalence of overweight and obesity among Latino children regardless of the type of school they attended (Figure 2). Nevertheless, the disparity was smallest within integrated schools located in low‐income neighbourhoods (10%), followed by disparities within integrated schools in high‐income neighbourhoods (13%). The disparity was wider for segregated schools located in low‐income neighbourhoods (14%), and even higher between segregated schools located in high‐income neighbourhoods (26%). The disparity was widest (28%) between Latino children in Latino‐segregated, low‐income schools (who had 53% overweight/obesity prevalence), compared to White children in White‐segregated schools in high‐income neighbourhoods (prevalence 25%).

FIGURE 2.

Overweight/obesity disparities in Latino versus White children in California public schools. (Top) Overweight/obesity prevalence among Latino and White children in California. (Bottom) Disparities within integrated schools and disparities between segregated schools. Left column: Low‐income schools. Right column: High‐income schools.

There were no meaningful differences in combined overweight and obesity disparities when we used different definitions of segregation. While estimates varied depending on the alternative definitions of segregation compared with the first definition, most were not significantly different (Figure S1). For example, when using the second definition that focused on ICE, the overweight/obesity disparity between Black versus White children in low‐income schools was 13% (compared to 11% when using the first segregation definition), and in high‐income schools it was 19% (compared to 22% when using the first segregation definition). The biggest difference seen across segregation measures was 6% (i.e., for the first definition, there was a 14% disparity between Latino vs. White children across low‐income, segregated schools, while it was 20% when using the ICE definition). Yet still, despite the substantial variation in the definitions of segregations we tested, the same patterns emerged across all definitions: Ethnic disparities were smallest when comparing children within integrated low‐income schools, and much larger when comparing children across segregated schools, including those that were high‐income. Furthermore, disparities were largest when comparing minority children in low‐income, minority‐segregated schools to White children in high‐income, white‐segregated schools. Moreover, results were also similar when restricting the sample to schools with larger enrolment (Figure S1).

Disparity estimates were also similar when we used more strict definitions of low‐ and high‐income schools, based on the lower and upper deciles (instead of tertiles) of the income distribution. When focusing on schools in neighbourhoods in the 1st to 10th percentile of median household income, combined overweight/obesity disparities were smaller by only 1%–2%. When focusing on schools in neighbourhoods in the 90th to 99th percentile for median household income, disparities were also relatively similar (reduced by 4% for Black‐segregated schools, increased by 2% for Latino‐segregated schools, and unchanged for integrated schools; Figures S2 and S3, right side). This is despite the fact that this more stringent income categorization greatly reduced the ethnic gaps in median neighbourhood income (Table S1). For example, in this analysis, the biggest gap was about $6500, between White‐segregated schools versus Latino‐segregated schools in high‐income neighbourhoods (while in the main analysis shown above, this gap was $30 000).

4. Discussion

This study found that, though Black and Latino children had higher rates of overweight and obesity than NHW children regardless of what type of school they attended, these disparities varied by school segregation and neighbourhood income. Overweight and obesity disparities among both Black and Latino children were smaller within integrated schools than they were between segregated schools. However, both types of disparities were much larger for the schools that were in high‐income neighbourhoods. This was because White children in schools located in high‐income neighbourhoods had markedly lower prevalences than White children in schools located in low‐income neighbourhoods. In contrast, the prevalences among children of colour were relatively unchanged unless schools were both in high‐income neighbourhoods and integrated. The highest overweight/obesity prevalences were among Black and Latino children in minority‐segregated schools located in low‐income neighbourhoods, in which about half had overweight or obesity. Meanwhile, only about one quarter of White students in White‐segregated schools located in high‐income neighbourhoods had overweight or obesity. These findings were robust to the definition of segregation used, despite substantial variation in the thresholds for segregation that were tested. Furthermore, these findings remained even when tighter control for income differences by segregation was achieved through narrowly defined income categories that greatly reduced the income gaps between White‐segregated versus minority‐segregated schools.

This study is among the first to investigate ethnic disparities in overweight and obesity among children by examining both school segregation and varying levels of income in school neighbourhoods, for both Black–White and Latino–White disparities. The findings of this study lend support to the theoretical argument espoused by LaVeist et al.—that it is place, not race, that is a primary driver of health disparities [25, 27, 28]—because we found that when minoritized children were exposed to the same contextual conditions as white children (i.e., integrated schools), disparities were smaller. Our study expands on their seminal work that focused on one integrated low‐income neighbourhood. Using statewide data, we show that their argument holds across a variety of income contexts, as disparities were greatly reduced in integrated schools in both low‐ and high‐income neighbourhoods. Furthermore, the findings among Latino children are particularly novel, given that much of the research on segregation and health outcomes has focused on Black versus White disparities. As of 2025, Latino children comprised approximately 50% of the child population in California and are approximately 1 in 4 nationally [29]. Hence, this finding also has implications for population health given the downstream morbidity associated with obesity, including diabetes, heart disease and other chronic diseases.

We found that healthy weight was more prevalent among White students who attended schools in high‐income neighbourhoods, regardless of integration. We did not observe the same ‘benefit’ among minoritized children unless schools were both high‐income and integrated. These findings suggest that schools and their neighbourhoods may shape the varying obesity disparities we observed. School segregation and neighbourhood income may contribute to obesity disparities through differential access to resources and a range of related mechanisms. Minoritized schools are likely also in minoritized neighbourhoods that have fewer social and health‐based resources than White‐segregated areas, even when neighbourhood income is similar [9, 25, 27, 30, 31, 32, 33, 34]. Under‐resourcing of minoritized neighbourhoods is the result of systematic disinvestment into communities of colour and of trickle‐down effects of a long history of disadvantage [35, 36, 37, 38]. For example, majority Black and Latino neighbourhoods have lower access to venues and destinations that are important for social well‐being and health‐promoting activities (e.g., supermarkets, parks, recreational facilities and social gathering places where social capital can be fostered). This lower access to community‐level amenities results from a multilevel system of disadvantage—including discriminatory practices in the allocation of public resources and banking agencies being less likely to approve loans or give more favourable loans for businesses located within communities of colour [35, 36, 37, 38]. Thus, even in minoritized neighbourhoods of higher income levels, Black and Latino households may still be likely to continue to face adverse place‐level exposures [39]. Furthermore, we found that the pattern of disparities persisted even when income differences were minimized by creating narrow income categories. This is evidence that obesity disparities related to school segregation are not purely due to income disparities, and that additional place‐level factors that are differentially distributed due to segregation need to be furthered examined.

4.1. Health Equity Implications

Importantly, these results suggest that health disparities are context and place‐dependent. Thus, to effectively reduce ethnic overweight/obesity disparities and improve health equity, place‐level interventions are needed, including among higher income communities of colour. Broadly, two approaches to improve individuals' environmental exposures could include housing voucher programs to enable people to move to higher resourced areas, or place‐based interventions to enrich the social, education, economic and health‐based resources in communities. The former approach has shown that such programs can have long‐term benefits on both children and adults' educational, social and economic mobility, thus improving long‐term health outcomes [40, 41]. The second approach includes strategies such as policies and initiatives that foster investment in local businesses that promote health, and redevelopment programs that improve access to health‐related facilities such as parks and playgrounds, supermarkets and walking paths and sidewalks [42]. An example of such a program is the Jordan Downs Redevelopment program in Los Angeles, in which traditional public housing units are being renovated (thus not displacing existing residents), alongside construction of affordable new housing, a shopping centre and other retail spaces, alongside increasing job opportunities for residents, child care, parks and gathering spaces [43].

4.2. Strengths

We capitalize on a statewide surveillance dataset from public school children in California, one of the most populous and ethnically diverse states in the nation [44]. Due to unique strengths in our data, our novel study can explore the association of school ethnic segregation and childhood overweight and obesity disparities within school neighbourhood income categories. Namely, this dataset includes ethnically minoritized schools that are in higher income neighbourhoods, enabling us to parse out ethnic disparities that are usually difficult to disentangle. Furthermore, this study is strengthened by its novel design, which explicitly examines disparities among children within integrated schools who share similar environments, inherently adjusting for unobserved school‐level factors, and contrasts these with disparities between minoritized versus White children in segregated schools [11, 25].

4.3. Limitations

This study has several limitations. First, the findings may not be generalizable beyond California. Second, a general limitation of the CDE data is that data for small schools tends to be masked, as was seen in Table 1. However, as we use weighted data based on school size, the inclusion of these schools would have likely had little effect on the observed patterns. Third, our primary definition of integration/segregation was relatively simple and maximized the number of children included in the analysis. However, our multiple analyses showed that the results are robust to the definition of integration/segregation. Data masking to protect student confidentiality led to greater exclusion of schools with smaller enrolments and thus may limit generalizability of the study findings to larger schools, although characteristics of excluded schools were consistent with segregation classifications used. Fourth, our minimum threshold of ‘high‐income’ neighbourhoods was relatively low for California's cost of living. However, this threshold ensured an adequate sample size of ‘high‐income’ Black and Latino‐segregated schools, and our analyses of an alternative definition of income that focused on the upper decile of income yielded similar results. Finally, the cross‐sectional design and data limitations precluded causal inferences and further examination of mechanisms. As the data were aggregated at the school‐ethnic group level, accounting for using student‐level variables was not possible. Studies examining these factors, as well as additional school‐ and neighbourhood‐level variables, are needed to further elucidate the pathways driving the disparities observed in this study.

4.4. Conclusion

This study showcases that obesity disparities are greater in segregated schools regardless of whether those schools are located in lower or higher income neighbourhoods, and are smaller or dissipate within integrated schools. Furthermore, disparities persist even when income gaps are accounted for. Hence, the findings help identify the independent role of segregation beyond income, further supporting the ‘place, not race’ theory. The disparities were particularly concerning for Black and Latino children in low‐income and segregated schools, who had about 50% prevalence of overweight and obesity. Disparities in health such as those described in this paper will persist unless place‐based interventions are widespread across our nation's schools and their communities. Given that obesity tracks into adulthood, interventions are needed early in life to reduce the immediate and long‐term consequences of high obesity prevalence.

Funding

This work was supported by the National Heart, Lung, and Blood Institute (R01‐HL136718) and National Center on Minority Health and Health Disparities (R01MD017687).

Ethics Statement

Given that this study utilizes publicly available, de‐identified data, the Institutional Review Board at Drexel University determined it does not meet the definition of human subjects research and did not need human subjects ethics approval (IRB 2501011003).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: School characteristics for schools in the 1st–10th percentiles and 90th–99th percentiles for median neighbourhood income.

Figure S1: Overweight/obesity disparities in Black versus White children in California public schools (top) and Latino versus White children in California public schools (bottom), showing various segregation measures and sensitivity analyses for masked data. Within each panel, the estimates shown are from: (1) the primary segregation definition; (2) segregation definition based on the index of concentration at the extremes; (3) segregation definition based on majority/no majority in the school; (4) sensitivity analyses among schools with larger enrolment.

Figure S2: Overweight/obesity disparities in Black versus White children in California public schools using alternative definitions of school neighbourhood income: low‐income neighbourhoods with median income equal between the 1st and 10th percentile ($26 264–$31 964), and high‐income neighbourhoods with median income between the 90th and 99th percentile ($106 616–$143 000).

Figure S3: Overweight/obesity disparities in Latino vs. White children in California public schools using alternative definitions of school neighbourhood income: low‐income neighbourhoods with median income equal between the 1st and 10th percentile ($27 478–$31 964), and high‐income neighbourhoods with median income between the 90th and 99th percentile ($106 616–$156 000).

Data Availability Statement

The data that support the findings of this study are available in California Department of Education at https://www.cde.ca.gov/. These data were derived from the following resources available in the public domain: California Department of Education, https://dq.cde.ca.gov/dataquest/dqcensus/EnrEthGrd.aspx?cds=00&agglevel=state&year=2018‐19; California Department of Education, https://www.cde.ca.gov/ta/tg/pf/.