Respondents' BMIs were calculated from measured height and weight. Height and weight measurements were performed by trained field interviewers. Height was measured using a portable stadiometer and measurements were recorded to 0.1 cm. Weight was measured using a portable Tanita electronic scale and was recorded to 0.1 kg. Height and weight were measured two times and averaged; if the two differed by more than a pre-specified amount (0.5 cm for height, 0.2 kg for weight), three measurements were performed and the average of the 2 measurements with the smallest difference was taken.

Our primary predictor variable was student race/ethnicity. Participants and participants' parents were asked whether the student was of Hispanic/Latino ethnicity and of which racial group they were a member. When available, the participant-response to Hispanic/Latino ethnicity and participant-chosen racial group was used. Where participant-response of either ethnicity or racial group was unavailable, the parental report was used. We used U.S. Census definitions to initially classify the sample into the following mutually exclusive categories: Asian/Pacific Islander; Black; Hispanic; Multi-racial; Native American/Alaskan Native; and White. We classified any individual who checked Hispanic as Hispanic regardless of additional racial/ethnic categories indicated. We categorized anyone not choosing Hispanic but choosing more than one racial category as Multi-racial. Because of small sample sizes in groups other than Blacks, Whites, and Hispanics, we collapsed groups that consisted of Asian/Pacific Islander, Multi-racial, and Native American/Alaskan into a racial category named “Other.”

Healthy Passages attempted to interview a parent/caregiver for all participants and selected the parent/caregiver with the most knowledge about the child's health and education, usually the mother or female guardian. The parent or caregiver was asked to report all household income by choosing one of the following response categories: <$25,000; $25,000–$49,999; $50,000–$99,999; and >$100,000. They were also asked to report the highest educational level achieved by a household member, including that of step- or foster parents, siblings, grandparents, or significant others of caregivers with the following response options: less than high school graduation; high school graduate or GED; some college or 2 year degree; college graduate or beyond.

Objective measurements of the height and weight of the parent/caregiver participating in the interview—typically the mother or female guardian–were taken at the time of the interview. We calculated BMI from height and weight (weight (kg)/height²(m²)) and then categorized weight status as low/normal weight v. overweight/obese (BMI≥25 kg/m²). Where objective heights and/or weights were missing (9.5% of cases), we inserted the mean parental/caregiver BMI of the sample and added an indicator flag to indicate missing data. We constructed models with and without the mean parental/caregiver BMI substituted to assess bias and found no material difference in the findings. Thus, we present only the models with the mean parental BMI substituted for those with missing data to avoid potential bias and inefficiency from deleting participants.

Age and gender were self-reported by participants.

We examined a number of variables that described school physical activity and dietary resources (Table 1) individually and as part of a school physical activity/nutrition summary score. The school physical activity/nutrition summary score included the following variables: 1) Physical Education (PE) taught in school 4 or 5 days/week; 2) PE taught in school ≥36 weeks/year; 3) 5^th graders have regular recess; 4) physical activity taught in health class; 5) school does not have a pouring contract with a beverage company; 6) nutrition taught in health class; 7) brand name fast food items are not available in the cafeteria. In addition, we explored three additional variables school offers breakfast; an outside food company manages food service; and school participates in the USDA lunch program. We did not include these variables, however, in the summary score as we had no a priori hypothesis regarding the direction of their influence on BMI. Approximately 10% of school personnel surveyed did not answer the question regarding whether or not school nutrition was taught in school. When asked how many weeks/year PE was taught, a number of school personnel survey respondents answered more than 52 weeks, an implausible value. Due to a large degree of missing or implausible data for these two variables (total ∼10%) and because neither was associated with the outcome, we constructed the school summary score with and without these variables and found no material differences in the findings. In order to maximize the sample size and avoid bias, this study presents results with the school summary score not including weeks/year of PE and whether nutrition is taught in school. The school physical activity/nutrition summary score represents a count of the positive factors that apply to each school (range: 0–5).

As an additional sensitivity analysis, we conducted a factor analysis of the variables in the physical activity/nutrition summary score. This analysis resulted in three factors, one related to gym and recess, a second related to foods and beverages available in the cafeteria, and a final one related to the number of weeks gym was taught. None of the factors was associated with individual BMI so we present only findings related to the summary score.

We controlled for additional school-level demographics, including the school administrator report of the percentage of students who were White and the school median household income aggregated from individual parent/caregiver report of individual household income as described above.

Finally, we controlled for the survey site: Birmingham, Los Angeles, and Houston.

Among females, only the school level household income was inversely associated with individual student BMI (β = −0.12, CI: −0.21,−0.02). For males and females, neither the percent of the student body reported to be White nor the summary score representing nutrition/physical activity programming was associated with BMI. In males and females, although each of the ten variables describing the school physical activity and nutrition resources was tested individually in addition to testing the summary score itself, no statistically significant associations were found (data not presented for individual components of summary score).

The parameter estimates for Hispanic and Black race/ethnicity in Model 2 (without school characteristics) and Model 3 (with school characteristics added) were compared. Among females, the addition of school characteristics reduced the effect size of Black race by nearly half (Model 2: β = 1.40, CI: 0.75, 2.06; Model 3: β = 0.77, CI: 0.03, 1.51). In contrast, among males, Black race was not associated with BMI in any model. Hispanic ethnicity was not associated with BMI among females in either Model 2 or 3, while among males the effect sizes were similar in Models 2 and 3.

Among females and males, the unconditional model (Model 1) indicated significant variation in BMI between schools, as reflected in the sigma_u values for both models. These values are the square roots of school-level variance components and can be interpreted as school-level standard deviations, so that the amount by which a typical school differs from the overall average BMI is about 1 unit (1.34 units for females, 0.97 units for males). In contrast, the standard deviation at the student level is approximately 5 units for both males and females. In females, after controlling for race/ethnicity, age, household demographics, and parental BMI, the between-school variation in BMI was no longer statistically significant and was estimated at 0 after accounting for chance variation at the student level. In males, the inclusion of individual level characteristics did not fully explain the between school variation in BMI among males, although their inclusion did result in an approximate one-third reduction in between-school variance (sigma_u²) (see Table 3, Model 2). The addition of school characteristics and programs did not further decrease the between-school variance in BMI.

Black females had BMIs averaging 1.4 kg/m² higher than White females. Mean BMI did not significantly differ from Whites for any other female racial/ethnic group. Females with an obese/overweight parent or caregiver had BMIs averaging >2 kg/m² higher than those who did not have an obese/overweight parent or caregiver.

Hispanic males averaged nearly 1.5 kg/m² higher in BMI than their White peers. Both markers of socioeconomic status—household income and highest educational level achieved in the household—had inverse but non-significant relationships with BMI. Males with an obese or overweight parent or caregiver had BMIs averaging >2 kg/m² higher than those who did not have an obese/overweight parent or caregiver.