Video Abstract
The Community Eligibility Provision (CEP), a universal free school meals policy, increases school meal participation by allowing schools in low-income areas to provide free breakfast and lunch to all students; however, its impact on obesity remains uncertain. The objective of this study is to estimate the association of CEP with child obesity.
School obesity prevalence was calculated using BMI measurements collected annually between 2013 and 2019 from students in California public schools in grades 5, 7, and 9. To estimate the association of CEP with obesity, we used a difference-in-differences approach for staggered policy adoption with an outcome regression model conditional on covariates, weighted by student population size.
The analysis included 3531 CEP-eligible schools using school-level obesity prevalence calculated from 3 546 803 BMI measurements. At baseline, on average, 72% of students identified as Hispanic, 11% identified as white, 7% identified as Black, and 80% were eligible for free or reduced-price meals. Baseline obesity prevalence was 25%. Schools that participated in CEP were associated with a 0.60-percentage-point net decrease in obesity prevalence after policy adoption (95% confidence interval: −1.07 to −0.14 percentage points, P = .01) compared with eligible, nonparticipating schools, corresponding with a 2.4% relative reduction, given baseline prevalence. Meals served increased during this period in CEP-participating schools only.
In a balanced sample of California schools, CEP participation was associated with a modest net decrease in obesity prevalence compared with eligible, nonparticipating schools. These findings add to the growing literature revealing potential benefits of universal free school meals for children’s well-being.
What’s Known on This Subject:
The Healthy, Hunger-Free Kids Act improved the nutritional quality of school meals, which are healthier than alternatives. CEP increases school meal participation by providing free meals to all students; however, its impact on obesity remains uncertain.
What This Study Adds:
In low-income California schools between 2013 and 2019, participation in CEP was associated with a relative decrease in obesity prevalence compared with eligible, nonparticipating schools, suggesting that universal free school meals policies may be effective for addressing childhood obesity.
Childhood obesity is a pressing public health concern for which solutions remain elusive. Obesity often tracks into adulthood1 and increases the risk of chronic conditions and premature death.2 Because obesity disproportionately impacts racially and ethnically minoritized and low-income children,3 effective population-level obesity-reduction strategies must address social determinants of health.4,5 Universal free school meals (UFM) policies represent one potential approach.
The Community Eligibility Provision (CEP) is a federal UFM policy authorized by the 2010 Healthy, Hunger-Free Kids Act (HHFKA). CEP operates through the National School Lunch Program and School Breakfast Program, which provide free and reduced-price meals to qualifying students. Schools that adopt CEP offer free meals to all students, saving schools the administrative burden of processing meal applications. Participating schools receive federal reimbursement based on the identified student percentage (ISP), which is calculated by using the percentage of students directly certified to receive free meals without an application, such as through participation in a means-tested safety net program, multiplied by 1.6 to approximate additional certification via application.6 CEP became an option for high-poverty schools nationwide in 2014. By 2023, >40% of US public schools were participating, reaching nearly 20 million children.6
CEP expands access to and reduces the stigma of receiving free meals,7 thus increasing school meal participation.8 –11 Increased school meal participation may reduce obesity because school meals are more nutritious than meals obtained elsewhere.12 –16 This may be particularly true for low-income children, for whom several income-stratified analyses reveal greater positive associations of nutritional quality and school meal participation relative to those observed in higher-income students.13,17,18 In addition to authorizing CEP, the HHFKA improved nutrition standards for school meals, resulting in improved nutritional quality among school meal participants.15,18 –21 CEP also relaxes budget constraints for low-income families. Students receiving free breakfast and lunch save approximately $4.70 per day ($850 per year) compared with paying full price.22 This allows families to purchase more nutritious groceries and spend less overall,23 potentially increasing disposable income. Evidence also suggests that UFM policies improve food insecurity15,23 and child academic performance.8,11,15,24 Although CEP has the potential to reduce obesity, its impact remains uncertain. Preliminary evidence suggests that UFM policies may be associated with obesity reduction;10,24 –26 however, more rigorous research is needed. To add to this small body of literature, with this study, we seek to evaluate the association of CEP with child obesity.
Methods
Because we hypothesize that CEP reduces obesity by increasing participation in school meals, we first conducted a simple, unadjusted, difference-in-differences analysis of the change in free, reduced-price, full-price, and total meals served per child per year, comparing schools that adopted CEP in school year 2018 to 2019 with eligible, nonparticipating schools. The California Department of Education (CDE) provided meals served counts from 2017 to 2019 (the only years available during the study period).
For our primary analysis of the association of CEP with obesity prevalence, we used a difference-in-differences design for staggered policy adoption.27 The design was repeated cross-sectional at the child level and longitudinal at the school level, comparing schools that participated in CEP with eligible schools that did not participate between 2013 and 2019. As secondary analyses, we examined the association of CEP with normal weight and overweight prevalence. The University of Washington Institutional Review Board determined the research exempt from oversight.
Primary Outcome
School-level obesity prevalence from school years 2013−14 through 2018−19 came from California’s physical fitness testing program which collects BMI each spring from fifth-, seventh-, and ninth-grade students.28 Although neither individual nor aggregate continuous BMI measurements are publicly available, they report categorical aggregate measures of students with obesity, defined as BMI greater than or equal to the 95th percentile for age and sex according to the Centers for Disease Control and Prevention growth charts.28,29
Secondary Outcomes
We examined obesity as our primary outcome because it is most clearly associated with future risk of disease.1,30 –32 As secondary outcomes, we assessed change in normal weight (here defined as BMI below the 85th percentile because CDE did not report a separate underweight category) and overweight (BMI at or above the 85th percentile but below the 95th percentile) prevalence.
Exposure
Treatment was defined as school participation in CEP, reported annually by CDE and the National Center for Education Statistics.
Covariates
Cross-sectional studies have revealed CEP adoption to be positively associated with ISP and the percentage of Hispanic and Black students and potentially associated with school size, school level, and rurality.33,34 Accordingly, we controlled for pretreatment school-level characteristics: percentage of students eligible for free and reduced-price meals (approximating ISP which was not available at baseline [2013–14]), percentage of fifth-grade students (of fifth, seventh, and ninth-grade students measured) because some schools group grades differently, percentage of students identifying as Hispanic because they comprise the majority of the sample, and participation in US Department of Agriculture (USDA) Provisions 2 and 3 (described in the Supplemental Information). School covariate data came from CDE and the National Center for Education Statistics. We also included county-level rurality (from the National Center for Health Statistics), unemployment rate (from the Bureau of Labor Statistics), and percentage of adults 25 and older with a bachelor’s degree or higher (from the American Community Survey), which could impact policy adoption and obesity outcomes beyond school-level subsidized meal eligibility. The models were weighted by school population size. As a sensitivity analysis, we replaced unemployment and education variables with a composite measure of socioeconomic status.35
Exclusion Criteria
Figure 1 depicts school exclusion criteria. Beginning with 9841 public schools participating in California’s physical fitness testing program, we excluded 1904 schools that did not report obesity outcomes in all 6 study years, 3744 schools that were ineligible for CEP (ISP <40%) to create comparable treatment and control groups, and 19 schools that switched from participating to not participating in CEP because our analytic approach required treated schools to remain treated. Schools reporting improbable outcomes were excluded from our sample; these included 433 schools reporting zero students with obesity. Obesity prevalence had a near-normal distribution, except for a disproportionate number of zeros. In these schools, obesity prevalence in the years before and after zero was often high. The majority of these schools also reported zero overweight prevalence. The high prevalence of obesity in US children3 makes this large spike at zero improbable and was considered a reporting error. Additionally, we excluded 210 schools reporting a >25-percentage-point (pp) change in obesity prevalence between years, which was beyond the first (−23 pp) and 99th (23 pp) percentiles of the distribution. This was deemed a change too large between years to be plausible and likely a reporting error. To check that our results were robust to how we defined outliers, we conducted sensitivity analyses, changing this cutoff to 20 and 30 pp.
Statistical Analysis
In our primary analysis, we estimated the association of CEP with obesity prevalence by using a difference-in-differences approach, which accounts for heterogeneous treatment effects due to staggered policy adoption.27 This approach leverages the same mechanism as a traditional difference-in-differences design (comparing treated and untreated groups before and after policy implementation) but does so for each cohort of schools newly participating in CEP (eg, schools newly participating in school year 2014–15, 2015–16, etc), comparing each year of participation to 1 year before policy adoption (eg, for the 2014–15 cohort, a separate treatment effect is estimated for school year 2013–14 vs 2014–15, 2013–14 vs 2015–16, 2013–14 vs 2016–17, 2013–14 vs 2017–18, and 2013–14 vs 2018–19). Our primary approach used an “outcome regression” model based on ordinary least squares, conditional on covariates, as described by Callaway, Sant’Anna, Heckman, and colleagues27,36,37 (Supplemental Information). This avoids the potential biases of two-way fixed effects models (including fixed effects for time and unit in a linear regression model).38 –40 This difference-in-differences approach relies on a conditional parallel trends assumption: baseline outcomes do not need to be the same for both groups; however, trends in the outcome before policy adoption should be similar to provide evidence that the parallel trends assumption post-policy is reasonable. Accordingly, we assess pretreatment trends in the outcome. Additional assumptions are described in the Supplemental Information.
We performed several additional sensitivity analyses. First, to assess the robustness of parallel pretrends, in the preperiod we designated the reference year as that before policy adoption. This is in contrast to our main analysis, which provides a short-term, year-over-year comparison in the pre-policy period.27 Second, we changed our comparison group to include not-yet-treated schools in addition to eligible, never-treated schools. Third, we used a doubly robust model that combines the outcome regression model from our primary analysis with inverse probability weighting, which yields valid results even if 1 of the 2 models is misspecified but can produce larger standard errors.42,60 Lastly, we repeated our main analysis, dropping all schools that participated in Provisions 2 or 3 during the study period (see the Supplemental Information). We used a P value of <.1 as a threshold for statistical significance, consistent with other studies on CEP.8,10 ,11,43 For the analyses, we used xthdidregress in Stata/MP 18.0.41 The Supplemental Information includes code to produce estimates and figures.
Study Results
The balanced sample included 3531 CEP-eligible schools between school years 2013−14 and 2018−19. Aggregate obesity data came from 3 546 803 BMI measurements from students in grades 5, 7, and 9. Table 1 reveals baseline school and county demographics overall and by the year in which the schools initiated CEP. Nonparticipating schools and those that adopted CEP in later years had higher average percentages of white and non-economically disadvantaged students. At baseline, on average, schools were composed of 72% students identifying as Hispanic, 11% identifying as white, 7% identifying as Black, and 5% identifying as Asian. On average, 80% of students were eligible for free or reduced-price lunch. Supplemental Table 4 reveals the baseline characteristics of CEP-eligible schools included in versus excluded from our analysis. The characteristics were not meaningfully different between the 2 groups, except for the percentage of fifth graders, which was higher among excluded schools. Supplemental Fig 5 reveals the percentage of schools that adopted CEP in our sample by district and year. Figure 2 reveals unadjusted trends in obesity prevalence by cohort (year in which schools began participating in CEP), weighted by school population size. The mean baseline obesity prevalence was 25% overall. Table 2 reveals the unadjusted difference-in-differences in free, reduced-price, full-price, and total meals served in school years 2017−18 and 2018−19, comparing schools that began participating in CEP in 2018−19 with eligible nonparticipating schools. Total meals served increased by an average of 13.4 meals per student per year (95% confidence interval [CI]: 5.7 to 21.2), free meals increased by 44.8 meals (95% CI: 37.8 to 51.8), full-price meals decreased by 11.3 meals (95% CI: −12.9 to −9.7), and reduced-price meals decreased by 20.1 meals (95% CI −21.0 to −19.2).
Baseline School and County Demographic Characteristics by Year Schools Adopted the CEP
| Overall | Year Schools Began Participating in CEP | ||||||
|---|---|---|---|---|---|---|---|
| Never | 2014−15 | 2015−16 | 2016−17 | 2017−18 | 2018−19 | ||
| n = 3531 | n = 1618 | n = 127 | n = 426 | n = 264 | n = 128 | n = 968 | |
| School characteristics | |||||||
| Sex | |||||||
| Male | 51% | 51% | 51% | 51% | 51% | 51% | 51% |
| Race and ethnicity | |||||||
| Hispanic | 72% | 69% | 73% | 85% | 71% | 62% | 73% |
| White | 11% | 13% | 8% | 3% | 9% | 13% | 13% |
| Black | 7% | 7% | 7% | 7% | 8% | 12% | 7% |
| Asian | 5% | 6% | 9% | 2% | 7% | 7% | 3% |
| 2 or more races | 2% | 2% | 1% | 0.4% | 2% | 2% | 1% |
| Filipino | 2% | 2% | 0.3% | 1% | 2% | 2% | 1% |
| American Indian | 1% | 1% | 1% | 0.4% | 1% | 1% | 1% |
| Pacific Islander | 1% | 1% | 0.4% | 0.1% | 0.4% | 1% | 0.4% |
| Economic statusa | |||||||
| Disadvantaged | 82% | 79% | 93% | 96% | 89% | 71% | 80% |
| Not disadvantaged | 13% | 16% | 6% | 3% | 8% | 20% | 15% |
| No economic information | 5% | 5% | 1% | 1% | 3% | 10% | 5% |
| Eligible for free or reduced-price meals | 80% | 77% | 86% | 85% | 86% | 80% | 81% |
| Grade level | |||||||
| Percentage of students in fifth grade | 52% | 49% | 66% | 61% | 55% | 50% | 51% |
| County characteristics | |||||||
| Percentage unemployed | 10% | 10% | 12% | 11% | 9% | 11% | 11% |
| Percentage bachelor's degree or higher | 26% | 27% | 22% | 27% | 29% | 24% | 23% |
| Ruralityb | 1.8 | 1.7 | 1.9 | 1.5 | 1.8 | 2.2 | 2.0 |
| Overall | Year Schools Began Participating in CEP | ||||||
|---|---|---|---|---|---|---|---|
| Never | 2014−15 | 2015−16 | 2016−17 | 2017−18 | 2018−19 | ||
| n = 3531 | n = 1618 | n = 127 | n = 426 | n = 264 | n = 128 | n = 968 | |
| School characteristics | |||||||
| Sex | |||||||
| Male | 51% | 51% | 51% | 51% | 51% | 51% | 51% |
| Race and ethnicity | |||||||
| Hispanic | 72% | 69% | 73% | 85% | 71% | 62% | 73% |
| White | 11% | 13% | 8% | 3% | 9% | 13% | 13% |
| Black | 7% | 7% | 7% | 7% | 8% | 12% | 7% |
| Asian | 5% | 6% | 9% | 2% | 7% | 7% | 3% |
| 2 or more races | 2% | 2% | 1% | 0.4% | 2% | 2% | 1% |
| Filipino | 2% | 2% | 0.3% | 1% | 2% | 2% | 1% |
| American Indian | 1% | 1% | 1% | 0.4% | 1% | 1% | 1% |
| Pacific Islander | 1% | 1% | 0.4% | 0.1% | 0.4% | 1% | 0.4% |
| Economic statusa | |||||||
| Disadvantaged | 82% | 79% | 93% | 96% | 89% | 71% | 80% |
| Not disadvantaged | 13% | 16% | 6% | 3% | 8% | 20% | 15% |
| No economic information | 5% | 5% | 1% | 1% | 3% | 10% | 5% |
| Eligible for free or reduced-price meals | 80% | 77% | 86% | 85% | 86% | 80% | 81% |
| Grade level | |||||||
| Percentage of students in fifth grade | 52% | 49% | 66% | 61% | 55% | 50% | 51% |
| County characteristics | |||||||
| Percentage unemployed | 10% | 10% | 12% | 11% | 9% | 11% | 11% |
| Percentage bachelor's degree or higher | 26% | 27% | 22% | 27% | 29% | 24% | 23% |
| Ruralityb | 1.8 | 1.7 | 1.9 | 1.5 | 1.8 | 2.2 | 2.0 |
Percentages weighted by school population size. All schools were untreated in the first study year (2013–14).
Economically disadvantaged defined as neither student’s parents received a high school diploma and/or student is eligible for free or reduced-price meals.
Mean rurality on a scale from 1 to 6, with 6 indicating most rural.
Unadjusted trends in obesity prevalence over time by the year in which schools began participating in CEP. Sample includes 3531 CEP-eligible schools between school years 2013−14 and 2018−19. All schools were untreated in the 2013−14 school year. Schools are grouped by the year they began participating in CEP. Color-coded vertical lines correspond to the first year that each group adopted CEP. The mean obesity prevalence is weighted by school population size.
Unadjusted trends in obesity prevalence over time by the year in which schools began participating in CEP. Sample includes 3531 CEP-eligible schools between school years 2013−14 and 2018−19. All schools were untreated in the 2013−14 school year. Schools are grouped by the year they began participating in CEP. Color-coded vertical lines correspond to the first year that each group adopted CEP. The mean obesity prevalence is weighted by school population size.
Difference-in-Differences in Free, Reduced-Price, Full-Price, and Total Meals Served per Student per Year, 2017 to 2019
| Mean Meals Served per Student per Year | CEP Schools (n = 965) | Non-CEP Schools (n = 1571) | Difference-in-Differences (95% CI) | ||||
|---|---|---|---|---|---|---|---|
| 2017−18 | 2018−19 | Difference | 2017−18 | 2018−19 | Difference | ||
| Free | 159.9 | 201.0 | 41.1 | 138.1 | 134.4 | −3.7 | 44.8 (37.8 to 51.8) |
| Reduced-price | 19.3 | 0 | −19.3 | 21.1 | 21.9 | 0.8 | −20.1 (−21.0 to −19.2) |
| Full-price | 20.2 | 10.2 | −10.0 | 21.2 | 22.5 | 1.3 | −11.3 (−12.9 to −9.7) |
| Total | 199.4 | 211.1 | 11.7 | 180.4 | 178.8 | −1.6 | 13.4 (5.7 to 21.2) |
| Mean Meals Served per Student per Year | CEP Schools (n = 965) | Non-CEP Schools (n = 1571) | Difference-in-Differences (95% CI) | ||||
|---|---|---|---|---|---|---|---|
| 2017−18 | 2018−19 | Difference | 2017−18 | 2018−19 | Difference | ||
| Free | 159.9 | 201.0 | 41.1 | 138.1 | 134.4 | −3.7 | 44.8 (37.8 to 51.8) |
| Reduced-price | 19.3 | 0 | −19.3 | 21.1 | 21.9 | 0.8 | −20.1 (−21.0 to −19.2) |
| Full-price | 20.2 | 10.2 | −10.0 | 21.2 | 22.5 | 1.3 | −11.3 (−12.9 to −9.7) |
| Total | 199.4 | 211.1 | 11.7 | 180.4 | 178.8 | −1.6 | 13.4 (5.7 to 21.2) |
Meals served data from a subset of 2536 schools that adopted the CEP in the 2018−19 school year or were eligible for but did not participate in CEP during the study period 2013 to 2019.
Table 3 and Figs 3 and 4 display results from our adjusted difference-in-differences model. Table 3 reveals the overall average treatment effect estimate, indicating that schools that participated in CEP experienced a 0.60-pp net reduction in obesity prevalence after policy adoption compared with eligible nonparticipating schools, conditional on covariates (95% CI: −1.07 to −0.14 pp, P = .01), corresponding to a 2.40% relative reduction, given baseline obesity prevalence (95% CI: −4.28% to −0.56%). Table 3 and Fig 3 display average treatment effects aggregated by cohort. Although all posttreatment cohort effects were negative, effects were largest for later-treated cohorts. Table 3 and Fig 4 reveal dynamic treatment effects (indexed by time relative to CEP adoption). The joint test that all pretreatment effects are zero was not statistically significant (P = .55), indicating parallel pre-policy trends, which lends support to the parallel trends assumption. In other words, despite differences in average baseline obesity prevalence, the differences in trends in obesity prevalence for CEP-participating versus eligible, nonparticipating schools were not different from zero. All dynamic treatment effect estimates were negative but only statistically significant for the first 2 years of CEP participation.
Aggregate Post-Policy Treatment Effect Estimates of School Participation in the CEP on Obesity Prevalence
| Estimate (pp) | 95% CI | |
|---|---|---|
| Overall | −0.60* | (−1.07 to −0.14) |
| Cohort (y of policy adoption) | ||
| 2014 − 2015 | −0.41 | (−1.60 to 0.78) |
| 2015 − 2016 | −0.62 | (−1.55 to 0.30) |
| 2016 − 2017 | −0.38 | (−1.20 to 0.43) |
| 2017 − 2018 | −0.97 | (−2.16 to 0.21) |
| 2018 − 2019 | −0.78* | (−1.39 to −0.16) |
| Duration of policy exposure, y | ||
| 1 | −0.49* | (−0.90 to −0.08) |
| 2 | −0.95** | (−1.59 to −0.30) |
| 3 | −0.50 | (−1.26 to 0.25) |
| 4 | −0.69 | (−1.66 to 0.28) |
| 5 | −0.09 | (−1.67 to 1.48) |
| Estimate (pp) | 95% CI | |
|---|---|---|
| Overall | −0.60* | (−1.07 to −0.14) |
| Cohort (y of policy adoption) | ||
| 2014 − 2015 | −0.41 | (−1.60 to 0.78) |
| 2015 − 2016 | −0.62 | (−1.55 to 0.30) |
| 2016 − 2017 | −0.38 | (−1.20 to 0.43) |
| 2017 − 2018 | −0.97 | (−2.16 to 0.21) |
| 2018 − 2019 | −0.78* | (−1.39 to −0.16) |
| Duration of policy exposure, y | ||
| 1 | −0.49* | (−0.90 to −0.08) |
| 2 | −0.95** | (−1.59 to −0.30) |
| 3 | −0.50 | (−1.26 to 0.25) |
| 4 | −0.69 | (−1.66 to 0.28) |
| 5 | −0.09 | (−1.67 to 1.48) |
Sample includes 3531 schools eligible for the CEP during the study period 2013 to 2019. Treatment effects were estimated using the Callaway/Sant’Anna difference-in-differences outcome regression estimator,27 weighted by school size and conditional on covariates.
*P < .05
**P < .01
Cohort treatment effects of participation in CEP on obesity prevalence. Difference-in-differences estimates are from the Callaway/Sant’Anna outcome regression estimator.27,36,37 Estimates are grouped by the year schools adopted CEP, conditional on covariates. The reference group is eligible, nonparticipating schools, and the reference year is 1 year before policy adoption.
Cohort treatment effects of participation in CEP on obesity prevalence. Difference-in-differences estimates are from the Callaway/Sant’Anna outcome regression estimator.27,36,37 Estimates are grouped by the year schools adopted CEP, conditional on covariates. The reference group is eligible, nonparticipating schools, and the reference year is 1 year before policy adoption.
Dynamic treatment effects of participation in CEP on obesity prevalence. Pre- and posttreatment difference-in-differences estimates are from the Callaway/Sant’Anna outcome regression estimator.27,36,37 Estimates are aggregated relative to the year schools adopted CEP (time = 0). The panel of schools is balanced in calendar time, not event time.
Dynamic treatment effects of participation in CEP on obesity prevalence. Pre- and posttreatment difference-in-differences estimates are from the Callaway/Sant’Anna outcome regression estimator.27,36,37 Estimates are aggregated relative to the year schools adopted CEP (time = 0). The panel of schools is balanced in calendar time, not event time.
We observed no significant association of CEP with overweight prevalence and a relative increase in normal weight prevalence associated with CEP (0.58 pp, 95% CI: 0.08 to 1.08; Supplemental Table 5 and Supplemental Figs 6–9). Our primary results were not substantively changed when replacing county-level education and unemployment measures as covariates with a Social Deprivation Index,35 changing the cutoff for excluding outlier schools, or changing the reference year, comparison group, estimation technique, and sample inclusion criteria (Supplemental Tables 6 and 7 and Supplemental Figs 10–22). Pre-policy trends were not significant in any of our sensitivity analyses, supporting the parallel trends and no anticipation assumptions, with the exception of the analysis which dropped schools that participated in Provisions 2 or 3 during the study period (Supplemental Fig 22). However, this analysis dropped a substantial portion of our sample (36%), including 915 schools that switched from participating in Provisions 2 or 3 to CEP between 2013 and 2019.
Discussion
Among 3531 low-income California public schools, those that implemented universal free meals through CEP experienced an average net reduction in obesity prevalence of 2.4% compared with eligible, nonparticipating schools. The strengths of our study include its use of a balanced sample of schools and a comparison group of eligible nonparticipating schools. In addition, we used a novel difference-in-differences approach for staggered policy adoption, allowing treatment effects to vary over time.27,38 By conditioning on covariates, we accounted for covariate-specific trends.27
Examining treatment effects by year of CEP adoption, we found that although effects were negative for all cohorts, they were largest for later-treated cohorts (school years 2017–18 and 2018–19). In computing dynamic treatment effects by years of policy exposure, the strongest associations were observed in the first 2 years of policy adoption, with weaker negative associations in the third and fourth years of adoption. These findings are consistent because later-adopting cohorts contribute to estimates in the first and second years of policy adoption. For instance, schools adopting CEP in the 2017–18 school year were only observed for 2 years post-policy, so they contributed to the first- and second-year post-policy estimates but not the third-, fourth-, or fifth-year estimates. Alternatively, these findings could indicate a reduction in the policy’s effectiveness over time. The authors of future research should explore the sustainability of CEP’s effects.
In our sample, later-treated cohorts had, on average, fewer students already eligible for free or reduced-price meals before CEP adoption. In these schools, CEP may result in a larger proportion of children newly participating in school meals, which is a plausible mechanism for the observed relative reduction in obesity, as evidence reveals that school meals are healthier than alternatives,13,14,17,44 particularly after stricter nutrition standards introduced by the HHFKA resulted in improved nutritional quality.15,18,19 Newly participating children may substitute up to half of their diet during the school year with more nutritious meals, potentially saving money for families.23 This is supported by the observed increase in meals served in 2018–19 CEP-adopting schools. We are not aware of changes in physical activity or other curriculum requirements in California schools during this time that coincide with the adoption of CEP (and are not a direct result of CEP) that could contribute to the observed effect.
Our results build on findings from other researchers. Kenney et al found that the timing of the HHFKA was associated with a reduction in obesity among low-income children in a nationally representative sample.25 In this study, however, the authors did not examine CEP specifically, nor did they have a control group.25 The authors of another study found that BMI increases associated with school meal participation among low-income students attenuated after the HHFKA’s nutrition standards took effect.26 The authors of this study also did not examine CEP specifically and lacked a comparison group.26 Rothbart and colleagues found that obesity decreased among seventh and 10th graders, but not younger students, in New York state school districts.10 Schwartz and Rothbart found that, in New York City middle schools, UFM was associated with reduced obesity among children from households at >185% of the federal poverty level.24 In a nationally representative sample of elementary schoolchildren, Andreyeva and Sun did not find an effect of CEP on obesity but observed a reduced probability of overweight among students from households at <200% of the federal poverty level.8 Internationally, the authors of several studies have found associations of UFM with reduced obesity.45 –47
In our study, we observed a reduction in obesity prevalence among CEP-participating schools relative to nonparticipating schools; however, prevalence increased over time for CEP and non-CEP schools. Although UFM policies may slow rising childhood obesity rates, they alone will not be sufficient to reverse trends and should be considered in conjunction with other obesity-reduction strategies. Although the observed effect size is modest, few population-level interventions have been successful at reducing obesity prevalence, particularly among lower-income populations;48,49 therefore, even small improvements are noteworthy. With this research, we add to a growing body of literature revealing the potential benefits of UFM to child well-being, including increasing participation in school meals,8 –11 improving diet quality,15,18 –20 reducing food insecurity,15 and improving academic outcomes.8,11,15,24 The coronavirus disease 2019 pandemic prompted the USDA to issue temporary waivers between 2020 and 2022 allowing schools nationwide to provide UFM.50 Beginning in 2022, California expanded UFM to all public schools.51 The authors of future research should explore whether such universal expansions impact health.
Data obtained from CDE were aggregated at the school level rather than the child level. Although California has a large and diverse population, these data are not nationally representative; the majority of students identified as Hispanic, which is representative of low-income children in California. The effects of CEP may vary by geographic region. For example, Davis et al found that CEP was associated with an increase in BMI and the probability of overweight (but not obesity) in 1 Atlanta school district.52 Data are from students in grades 5, 7, and 9 and are not generalizable to younger students. Additionally, data were collected and reported by schools, not necessarily for research; however, we attempted to minimize the perceived erroneous outcome measures. Evidence reveals that teacher-measured BMI is highly accurate,53,54 suggesting that the observed outliers were due to reporting rather than measurement error, adding support to the validity of nonoutlier measurements.
Conclusions
In a balanced sample of low-income, predominantly Hispanic California schools, participating in CEP was associated with a modest net reduction in obesity prevalence compared with eligible, nonparticipating schools. These findings add to the growing literature revealing the potential benefits of UFM for child well-being. Such policies represent a promising strategy for addressing childhood obesity.
Ms Localio contributed to study design, cleaned and managed data, conducted the initial data analyses, contributed to data interpretation, and drafted the manuscript; Drs Knox and Basu conceptualized and designed the study and contributed to data analysis and interpretation; Mr Lindman contributed to the study design and contributed to data analysis and interpretation; Ms Walkinshaw coordinated data acquisition, managed the project, and assisted in data cleaning and management; Dr Jones-Smith conceptualized and designed the study, acquired study funding, coordinated and supervised data acquisition, and contributed to data analysis and interpretation; and all authors critically reviewed and revised the manuscript, approved the final manuscript as submitted, and agree to be accountable for all aspects of the work.
FUNDING: Funded by the National Institutes of Health (NIH). All phases of this study were supported by the Eunice Kennedy Shriver National Institute of Child Health & Development of the NIH under award number R01HD105666. The NIH had no role in the design and conduct of this study. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
CONFLICT OF INTEREST DISCLOSURES: The authors have indicated they have no potential conflicts of interest relevant to this article to disclose.
Comments