Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Obesity (Silver Spring). Author manuscript; available in PMC 2009 November 30.
Published in final edited form as:
PMCID: PMC2785847

School Level Contextual Factors Are Associated With the Weight Status of Adolescent Males and Females



To determine whether school context influences the BMI of adolescent males and females.

Methods and Procedures

Our sample was 17,007 adolescents (aged 12–19) from the National Longitudinal Study of Adolescent Health (Add Health). We used gender-stratified multilevel modeling to examine the contribution of schools to the overall variance in adolescent BMIs, calculated from self-reported weight and height. We then examined the associations of individual attributes with BMI after controlling for the average BMI of the school and the association of two school-level variables with BMI.


Participants attended schools that were segregated by race/ethnicity and socioeconomic status (SES). In females, when controlling only for individual-level attributes, individual household income was inversely associated (β = –0.043, P = 0.01) while Hispanic (β = 0.89, P < 0.001) and black (β = 1.61, P < 0.001) race/ethnicity were positively associated with BMI. In males, Hispanic (β = 0.67, P < 0.001) race/ethnicity was positively associated with BMI; there was no difference in the BMIs of blacks compared with whites (β = 0.24, P = 0.085). After controlling for the school racial/ethnic makeup and the school level median household income, the relationship between individual race/ethnicity and BMI was attenuated in both male and female adolescents. Higher school level median household income was associated with lower individual BMIs in adolescent girls (γ = –0.37, P < 0.001) and boys (γ = –0.29, P < 0.001) suggesting a contextual effect of the school.


Male and female adolescents attending schools with higher median household incomes have on average lower BMIs. Resources available to or cultural norms within schools may constitute critical mechanisms through which schools impact the BMI of their students.


Child and adolescent obesity continues to be a leading public health problem. Between 1980 and 2002, the prevalence of obesity tripled in children and adolescents (1,2); recent data suggest that the trend continues to worsen (3). Hispanics, African Americans, and those of lower socioeconomic status (SES) continue to have disproportionately higher rates of obesity when compared to whites and those of higher SES (35). Numerous studies have demonstrated racial/ethnic and socioeconomic disparities in both adolescent weight status and weight-related behaviors (37), but only recently have studies looked to environmental contexts, such as neighborhoods and schools for possible explanations (8,9). While the number of studies examining neighborhood variables has grown, there remains a paucity of studies focused on school environments (10,11).

Schools are logical targets of study to identify factors that may contribute to racial/ethnic and socioeconomic disparities in adolescent health behaviors and health outcomes, specifically weight-related behaviors and outcomes. Schools are readily defined social contexts in which students spend a great deal of time. They may influence both energy intake and energy expenditure through a number of different pathways including the food choices available through lunch programs or vending machines, opportunities for physical activity, role modeling by teachers and/or coaches, and/or cultural norms among the student population. It is also conceivable that schools may contribute to racial/ethnic and/or socioeconomic disparities. Recent studies have shown that schools are becoming increasingly segregated along racial/ethnic and socioeconomic lines (12) leading to very different school experiences for students of different backgrounds. Though the deleterious effects of these segregated schools on academic achievement is well documented (13), less is known regarding their potential impact on the health status of their students.

Using data from the National Longitudinal Study of Adolescent Health (Add Health), a nationally representative school-based survey of adolescents (14), we sought to identify demographic characteristics of the school that may influence the weight status of students above and beyond the influence of the characteristics of the individual students. Add Health is uniquely suited for studies of this kind given the nested nature of the data (i.e., the students are nested within schools), as well as the contextual information that was collected from school administrators (i.e., school size, racial/ethnic makeup of the school). We specifically sought to answer the following questions: (i) Do racial/ethnic and socioeconomic disparities in adolescent boys’ and girls’ BMI exist within schools? (ii) Is there variability in adolescent boys’ and girls’ BMI between schools, after accounting for the composition of the student body attending the schools? (iii) Do adolescent girls and boys attending higher income schools have on average lower BMIs? (iv) Does the racial/ethnic makeup of the school influence the BMI of individual students?


Study population

This research uses data from the first of four Waves of the Add Health, a nationally representative school-based study of adolescents enrolled in grades 7 through 12. The primary sampling unit of the Add Health study is schools. Before sampling, schools were sorted by size, school type, census region, level of urbanization, and the percentage of the student body that is white. Add Health used systematic sampling methods and implicit stratification to ensure that the selected schools were representative of US schools. All of the students attending the chosen schools were eligible for the In-home Sample. Students were stratified by gender and grade in school with ~17 students randomly selected from each stratum. The final sample was made up of 132 participating schools with ~200 students per school completing the In-home questionnaire. Wave I data at the individual, family, school, and community level were collected between 1994 and 1996, Wave II data were collected 1 year later, and Wave III data were collected from 2001 to 2002 (14). Wave IV data will be collected in 2008. Because our a priori hypotheses focused on differences between Hispanics and African Americans and whites, we limited our sample to those who self-identified as white, non-Hispanic; black, non-Hispanic; or Hispanic. We also excluded the 502 participants who self-identified as disabled due to concerns that their weight status might be influenced by different factors when compared with their nondisabled peers.

In addition to the above exclusions, we also excluded 962 students who had missing data either for the dependent variable or for >3 of the 13 independent variables. However, before this final exclusion, we addressed a high nonresponse rate (~10%) for the two variables measuring SES—parent reported maternal education and household income. In an effort to avoid selection bias and inaccurate inferences resulting from listwise deletion, we imputed these variables by best subset regression (15,16). After this imputation and all exclusions, our final sample contained 17,007 adolescents. Our study population was nested within 132 schools. On average, ~200 students were sampled from each school with the exception of the 16 schools in which the entire student population was sampled.

Study variables

Outcome variables

The BMI (weight (kg)/height (m2)) of individual participants was our outcome variable, a validated measure commonly used in studies of weight status in children (17,18). BMI was calculated from self-reported weight and height. We also examined relationships with overweight—as defined by the International Task Force on Obesity—as the outcome (19). There were no differences in the relationships when using BMI as a continuous variable vs. a categorized variable (i.e., relationships that were statistically significant when using BMI as a continuous variable were also statistically significant when using BMI as a categorical variable) and so we have chosen to present here only the results with BMI as a continuous variable.

Independent individual variables

Demographic variables controlled for included age, gender, race/ethnicity, maternal education, and household income. Race/ethnicity was classified from two questions, one which asked participants to indicate whether they were of Hispanic/Latino origin and the other which asked them to choose a category of race that best describes them. We constructed six mutually exclusive categories: Hispanic, black or African American (not Hispanic), Asian/Pacific Islander, white (not Hispanic), Native American/American Indian, and other. As described above, we limited our population to those who identified as Hispanic, white, or black. We used the parental report of the highest grade of education achieved by the participant's mother and the total household income over the past year. After performing sensitivity analyses, we dichotomized the maternal education variable into having achieved a college degree or higher vs. not. We performed sensitivity analyses on the household income variable including it as a continuous variable as reported (in dollars/year); transforming it into a measure relative to the poverty level (taking into consideration the household size then comparing it with poverty thresholds in 1995, the year the data were collected); dichotomizing it into living in poverty vs. not living in poverty; transforming it into a measure of equalized household income (household income divided by the square root of the household size), a measure favored by economists to take into consideration the impact of household size but recognizing that each additional household member does not incur the same additional cost to the household. There was no significant difference in the associations between household income and BMI with the different measures; for ease of presentation, we have chosen to present findings using the percent poverty level as a continuous measure. As mentioned above, responses for maternal education and household income were lower than other response rates and so we present findings using the imputed values.

School-level variables

In Add Health, schools were the primary sampling unit. A sample of 80 schools and 52 middle or feeder schools were selected. Administrators of the 132 schools were asked to fill out questionnaires to describe the demographic characteristics of the schools as well as to provide information regarding school policies. We included two variables to describe the demographics of the school: the school level median household income and racial/ethnic makeup. The school level median household income was calculated as a composite of the household income reported by the parents of individual students attending the same schools. We felt this was an appropriate measure of the school level household income because the ~200 students who participated in the in-home survey were randomly sampled as to be representative of the school population. The racial/ethnic makeup was reported by the school administrators as the percentage of the student body that is white and was reported in categories (0%, 1–66%, 67–93%, and 94–100%). We dichotomized this variable to ≤66% and >66%.


We stratified all analyses by gender based on prior literature demonstrating differences in weight status between males and females of different races/ethnicities (3). We used a multilevel modeling approach to assess the impact of school context on individual BMI (20,21). The substantive relevance of these models have been well discussed (2224). Specifically, two-level models were estimated, with a continuous response, BMI (y), for a boy or girl i studying in school j. The outcome yijk was related to a set of individual- and school-level predictors, X, and a random effect for each level as yij = β0 + βX + u0j + e0ij. The linear predictor on the right-hand side of the equation consisted of a fixed part (β0 + βX) estimating the conditional coefficients for the independent variables, and two random terms attributable to individuals (e0ij) and schools (u0j), with each assumed to have an independent and identical distribution and variances (σe02, and σu02, respectively) estimated at each level. The school-variance parameter was σu02 estimated before and after controlling individual compositional variables to assess the relative contribution of school context on individual BMI. Estimates reported from the linear models are maximum likelihood based using the xtreg algorithm as implemented with STATA 9 (25,26).


Disparities in weight status by race/ethnicity exist in males and females as seen in Figures 1 and and2.2. In female adolescents, 32% of blacks and 26% of Hispanics were either overweight or obese compared with 19% of whites. In contrast, in males, 34% of Hispanic males were overweight or obese compared with 28% of blacks and 27% of whites. There were significant differences in the individual SES as well as that of the schools attended by different races/ethnicities (Table 1). The median parent-reported household income of whites was $46,000 while that of blacks was $30,000 and Hispanics was $28,000. Fifty-six percent of white adolescents reported their mother completed at least a college degree, while only 50% of blacks and 25% of Hispanics reported having mothers with similar educational achievement. The median household income of the school attended was substantially higher for whites than either Hispanics or blacks. The schools’ racial/ethnic makeup also varied considerably by race/ethnicity; 75% of whites attended schools where two-third or more of the students were white, while only 20% of blacks and 15% of Hispanics attended schools where two-third or more of students were white.

Figure 1
Weight status for females by race/ethnicity.
Figure 2
Weight status for males by race/ethnicity.
Table 1
Participant characteristics by race/ethnicity—males and females combined (n = 17,007)


The results for females from three models (a null or empty variance component model, a variance component model with individual variables in the fixed part, and a variance component model with individual and school variables in the fixed part) are shown in Table 2. In the null or empty variance component model, the variance of the random effects of schools was significantly different from zero (σu02=0.98, P < 0.001) as was the intraclass correlation coefficient (0.048, 95% confidence interval: 0.034, 0.065) indicating that the school-level variance contributed ~5% to the overall variance in BMI. The addition of the individual-level variables to the fixed part of the model reduced the variance at the school level by approximately half but it remained significantly different from zero (σu02=0.55, P < 0.001). Each of the individual-level variables was found to be significantly associated with BMI. Black females were noted to have a BMI 1.61 units higher and Hispanic females 0.89 units higher than the BMI of white females. Higher levels of both maternal education (β = –0.48, P < 0.001) and household income (β = –0.043, P = 0.011) were found to be negatively associated with BMI (i.e., the higher the familial SES the lower the BMI).

Table 2
Individual and school characteristics’ associations with BMI in females

We next added two variables that described the demographic characteristics of the school. Again the variance at the school level was reduced but remained significantly different from zero (σu02=0.46, P < 0.001). The school level median household income was negatively associated with BMI (γ = –0.37, P < 0.001); students attending schools with higher median household incomes had lower BMIs on average. The racial/ethnic makeup of the school was not associated with BMI (γ = 0.10, P = 0.20). The addition of the two school-level variables attenuated several of the associations noted between individual factors and BMI as can be seen in Table 2. The associations between both black and Hispanic race/ethnicity and BMI were attenuated though they remained significant as did the association between maternal education and BMI. The association between individual household income and BMI was attenuated and was no longer significant.


The results for males from three models (a null or empty variance component model, a variance component model with individual variables in the fixed part, and a variance component model with individual and school variables in the fixed part) are shown in Table 3. In the null or empty variance model, the variance of the random effects of schools was significantly different from zero (σu02=0.82, P < 0.001) as was the intraclass correlation coefficient (0.035, 95% confidence interval: 0.024, 0.049) indicating that the school-level variance contributed ~4% to the overall variance. The addition of the individual-level variables to the fixed part of the model reduced the variance at the school level but it remained significantly different from zero (σu02=0.44, P < 0.001). In the model containing individual-level variables only, Hispanic males were noted to have higher BMIs than whites (β = 0.67, P < 0.001), while there was no significant difference in the BMIs of black and white males (β = 0.24, P = 0.085). Neither maternal education (β = –0.043, P = 0.69) nor household income (–0.010, P = 0.50) was found to be associated with male BMI.

Table 3
Individual and school characteristics’ associations with BMI in males

With the addition of the variables describing the demographics of the school, the variance at the school level was reduced slightly but remained significantly different from zero (σu02=0.38, P < 0.001). As in females, the school level median household income was negatively associated with BMI (γ = –0.29, P = 0.001); adolescent males attending lower income schools had higher BMIs on average. Similar to females, the racial/ethnic makeup of the school was not associated with BMI (γ = 0.11, P = 0.13). Once we controlled for the attributes of the school, the relationship between Hispanic race/ethnicity and BMI was attenuated but remained significant (β = 0.62, P < 0.001). In males, the individual household income was not associated with BMI in either the model with individual-level variables only or the model with school-level variables and individual-level variables.


In this nationally representative study, we found different patterns of obesity among males and females and among different racial/ethnic and socioeconomic groups, similar to other published findings collected during the same period (1,3,4). Specifically, among females, we found both black and Hispanic adolescents were heavier than their white peers as were those with lower household incomes. However, once we accounted for the school the adolescent attended as well as its demographics, these relationships were partially attenuated and the relationship between individual household income and BMI was no longer significant. In males, we found no differences in BMI between blacks and whites but significant differences were observed between Hispanics and whites. After controlling for the demographics of the school attended by males, the association between Hispanic ethnicity and BMI remained significant though attenuated. Interestingly, neither marker of SES—household income or maternal education—was predictive of adolescent males’ BMI. In both adolescent males and females, we found strong relationships between school household income and BMI even after controlling for individual-level household income and maternal education. To our knowledge, this is the first time that that the contextual effect of the school's SES has been shown to be associated with individual BMI. In contrast, in neither males nor females was the racial/ethnic composition of the school associated with BMI.

Our findings reinforce previous studies that have demonstrated the association of schools with academic achievement as well as health behaviors (13,2729). There is a long-standing body of evidence demonstrating the effect of schools on academic achievement independent of the student composition (13,30). Recent studies have also considered schools as a possible explanation for student risk behavior and have found that school factors such as tobacco policies or social norms can partially explain differential patterns of tobacco and alcohol use (27,31,32). Other studies have linked the SES of the school to higher rates of victimization and student weapon carrying (33,34). Particularly relevant to our study, a single European study has shown that students attending a school with a physical education (PE) specialist had improved cardiovascular endurance relative to students at schools without such a specia list (35). In addition, our previous work has shown that physical activity varies by schools and that school environments explain racial/ethnic differences in physical activity in adolescents girls (10). Although the literature linking school environments to health behaviors has grown recently, we know of only one other study that has demonstrated a school contextual effect on a health outcome (11).

The association of the school-level household income with BMI even after controlling for the individual-level household income indicates that there is a contextual effect of the school above and beyond the compositional. Owing to the limitations of the available data, we were unable to elucidate mechanisms through which the socioeconomic context of schools influences weight status of students. However, we hypothesize that schools may influence the weight status of their students through the availability of both physical activity and healthy food resources.

Our earlier work has demonstrated that physical activity participation varies by schools and that schools can partially explain racial/ethnic differences in physical activity in adolescent girls (10). One potential mechanism through which schools may impact physical activity and thus weight status is through variable physical resources. Schools with students of lower SES may have fewer material resources for physical activity such as adequate playgrounds or gymnasiums. Previous studies have shown that physical resources of the school influence activity levels in children and that schools with more disadvantaged kids are more likely to report negative school environmental conditions (36,37). In addition to potentially different physical resources, schools may have different programmatic resources. Schools with students of lower SES may be less likely to offer regular PE classes or interscholastic sports because of constraints on resources (38). National studies have shown sharp declines in PE enrollment during high school but have demonstrated that the sharpest declines are in African Americans (39). PE programs have been shown to vary by the demographics of schools with schools with more college-bound students being more likely to require daily PE and schools with higher per pupil expenditure and higher percentage of white students having overall better PE programs (39).

Given that children may consume as much as 50% of their daily calories while at school (40), school food programs are likely an equally important mechanism through which schools influence the weight status of students. A number of studies have demonstrated that school food resources impact student diets (4143). Studies have also shown that a large percentage of schools are turning to the sale of generally nutritionally poor “competitive foods” (i.e., a la carte items in the cafeteria or vending machine foods) and pouring contracts with soft drink companies for additional sources of revenue and to balance a school's food budget (40,44). We are unaware of data that examine the availability of competitive foods by the demographics of the schools. However, we hypothesize that schools with lower income students may face additional economic hardship and thus may be at least as likely as schools with higher income students to provide competitive foods or engage in pouring contracts. We plan to examine school vending machine policies by the demographics of the school in the future. We are also unaware of existing data regarding school open campus policies that may impact the availability of alternative sources of food for students. Schools with low income students may be more or less likely to have open campus policies and may be situated in neighborhoods with more or fewer fast-food restaurants or convenience stores. Future research is needed to sort out varying food-related influences by schools.

As the obesity epidemic has impacted an increasing number of children and adolescents, a great deal of effort has been spent attempting to identify school-based interventions to reverse the trend. Although a number of studies have documented success in increasing physical activity and awareness of healthy habits, few have succeeded in impacting the weight status of students (40,45). Perhaps, as has been shown in substance abuse research, the interventions need to be targeted less at individuals and more at changing school environments including the available resources and cultural norms (27). Our findings support the notion that interventions should look beyond focusing exclusively on individual students.

There are several limitations to this study that must be acknowledged. First, these data were collected >10 years before and may not be reflective of the current situation among schools. However, there is evidence that schools have become increasingly segregated both racially and socioeconomically over the past decade (12,46). Thus, our findings likely underestimate the current effect of schools on BMI. In addition, recent data from the Centers for Disease Control and Prevention indicate a decline in daily PE participation from 1991 to 1995 but no change from 1995 to 2003 (39). Thus, if our findings are related to levels of PE participation they should be reflective of the current situation. Given the longitudinal nature of the data, it is imperative to perform analyses on baseline data to understand the factors that may influence the health of adolescents as they transition into young adulthood and beyond. This study lays the groundwork for our future work using additional waves of Add Health. A second limitation is our reliance on self-reported height and weight. However, a study by Goodman et al. using Add Health found the correlation between BMI calculated from self-report vs. measured height and weight was 0.92 and only 3.8% of youth were misclassified as obese using self-reported data (47). It is possible that underreporting of weight may be differentially distributed among schools based on the schools’ demographics which could bias our findings. Given the low level of misclassification found by Goodman et al. this, however, is unlikely to reverse our findings.

In conclusion, we find that adolescents attending lower socioeconomic schools have higher BMIs than their peers attending schools of higher SES. This study extends the large body of research linking neighborhood contexts and health (4851) to schools, an appealing target for policy intervention. Future research is needed to identify mechanisms such as resources for physical activity and/or availability of healthy food choices through which schools impact the weight status of their students. Coupling these findings with the identification of mechanisms operating at the school level could help identify appropriate policies and/or interventions to reverse the rise in child and adolescent obesity and to eliminate the racial/ethnic and socioeconomic disparities.


T.K.R. is supported by the Charles H. Hood Foundation Child Health Research Award and the Maternal and Child Health Leadership Education in Adolescent Health Training Program. S.V.S. is supported by the National Institutes of Health Career Development Award (NHLBI 1 K25 HL081275). This research used data from the Add Health project, a program designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris, and was funded by a grant P01-HD31921 from the National Institute of Child Health and Human Development, with cooperative funding from 17 other agencies. Special acknowledgment is due to Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Persons interested in obtaining data files from Add Health should contact Add Health, Carolina Population Center, 123 W. Franklin Street, Chapel Hill, NC 27516-2524, USA (ude.cnu@htlaehdda).



The authors declared no conflict of interest.


1. Hedley A, Ogden C, Johnson C, et al. Prevalence of overweight and obesity among US children, adolescents, and adults, 1999–2002. JAMA. 2004;291:2847–2850. [PubMed]
2. Ogden C, Flegal K, Carroll M, Johnson C. Prevalence and trends in overweight among US children and adolescents, 1999–2000. JAMA. 2002;288:1728–1732. [PubMed]
3. Ogden C, Carroll M, Curtin L, et al. Prevalence of overweight and obesity in the US, 1999–2004. JAMA. 2006;295:1549–1555. [PubMed]
4. Miech R, Kumanyika S, Stettler N, et al. Trends in the association of poverty with overweight among US adolescents, 1971–2004. JAMA. 2006;295:2385–2393. [PubMed]
5. Freid V, Prager K, MacKay A, Xia H. Chartbook on trends in the health of Americans: Health, United States, 2003. National Center for Health Statistics; Hyattville, MD: 2003.
6. Gordon-Larsen P, McMurray R, Popkin B. Determinants of adolescent physical activity and inactivity patterns. Pediatrics. 2000;105:E83. [PubMed]
7. Gordon-Larsen P, Adair L, Popkin B. The relationship of ethnicity, socioeconomic factors, and overweight in U.S. adolescents. Obes Res. 2003;11:121–129. [PubMed]
8. Nelson M, Gordon-Larsen P, Song Y, Popkin B. Built and social environments: associations with adolescent overweight and activity. Am J Prev Med. 2006;31:109–117. [PubMed]
9. Gordon-Larsen P, Nelson M, Page P, Popkin B. Inequality in the built environment underlies key health disparities in physical activity and obesity. Pediatrics. 2006;117:417–424. [PubMed]
10. Richmond T, Hayward R, Gahagan S, Field A, Heisler M. Can school income and racial/ethnic composition explain the racial/ethnic disparity in adolescent physical activity participation? Pediatrics. 2006;117:2158–2166. [PubMed]
11. Goodman E, Huang B, Wade T, Kahn R. A multilevel analysis of the relation of socioeconomic status to adolescent depressive symptoms: does school context matter? J Pediatr. 2003;143:451–456. [PubMed]
12. Race in American schools: rapidly resegregating school districts. Harvard University. 2002. [13 September 2006]. < reseg_schools02.php>.
13. Lee V, Bryk A. A multilevel model of the social distribution of high school achievement. Sociol Educ. 1989;62:172–192.
14. Add Health research design: Waves I, II, and III. 2003. [13 September 2006]. < projects/addhealth/files/research_design.pdf>.
15. King G, Honaker J, Joseph A, Scheve K. List-wise deletion is evil: what to do about missing data in political science.. Annual Meetings of the American Political Science Association; Philadelphia, PA. 13 July 1998.
16. Stata help for impute. 2007 [29 March 2007]; <>.
17. Dietz W, Robinson TN. Use of the body mass index (BMI) as a measure of overweight in children and adolescents. J Pediatr. 1998;132:191–193. [PubMed]
18. Pietrobelli A, Faith MS, Allison DB, et al. Body mass index as a measure of adiposity among children and adolescents: a validation study. J Pediatr. 1998;132:204–210. [PubMed]
19. Cole T, Bellizzi M, Flegal K, Dietz W. Establishing a standard definition for child overweight and obesity worldwide; international survey. BMJ. 2000;320:1240–1243. [PMC free article] [PubMed]
20. Raudenbush S, Bryk A. Hierarchical Linear Models; Applications and Data Analysis Methods. Sage Publications; Thousand Oaks, CA: 2002.
21. Goldstein H. Multilevel Statistical Models. Arnold; London: 2003.
22. Blakely T, Subramanian SV. Multilevel studies. In: Oakes JM, Kaufman JA, editors. Methods in Social Epidemiology. Jossey-Bass; San Francisco, CA: 2006. pp. 316–340.
23. Subramanian SV. The relevance of multilevel statistical methods for identifying causal neighborhood effects. Soc Sci Med. 2004;58:1961–1967. [PubMed]
24. Subramanian SV, Jones K, Duncan C. Multilevel methods for public health research. In: Kawachi I, Berkaman LF, editors. Neighborhoods and Health. Oxford University Press; New York: 2003. pp. 65–111.
25. Rabe-Hesketh S, Skrondal A. Multilevel and longitudinal modeling using stata. Stata Press; College Station, TX: 2005.
26. StataCorp . Stata Statistical Software. 9.0 College Station, TX: 2006.
27. Kumar R, O'Malley P, Johnston L, Schulenberg J, Bachman J. Effects of school-level norms on student substance use. Prev Sci. 2002;3:105–124. [PubMed]
28. Sellström E, Bremberg S. Is there a “school effect” on pupil outcomes? A review of multilevel studies. J Epidemiol Community Health. 2006;60:149–155. [PMC free article] [PubMed]
29. Aveyard P, Markahm W, Cheng K. A methodological and substantive review of the evidence that schools cause pupils to smoke. Soc Sci Med. 2004;58:2253–2265. [PubMed]
30. Battistich V, Solomon D, Kim D, Watson M, Schaps E. Schools as communities, poverty levels of student populations, and student's attitudes, motives, and performance: a multilevel analysis. Am Educ Res J. 1995;32:627–658.
31. Maes L, Lievens J. Can the school make a difference? A multilevel analysis of adolescent risk and health behaviour. Soc Sci Med. 2003;56:517–529. [PubMed]
32. Moore L, Roberts C, Tudor-Smith C. School smoking policies and smoking prevalence among adolescents: multilevel analysis of cross-sectional data from wales. Tob Control. 2001;10:117–123. [PMC free article] [PubMed]
33. George R, Thomas G. Victimization among middle and high school students: a multilevel analysis. High Sch J. 2000;84:48–58.
34. Wilcox P, Clayton R. A multilevel analysis of school-based weapon possession. Justice Q. 2001;18:509–541.
35. Zhu W. A multilevel analysis of school factors associated with health-related fitness. Res Q Exerc Sport. 1997;68:125–135. [PubMed]
36. Sallis J, Conway T, Prochaska J, et al. The association of school environments with youth physical activity. Am J Public Health. 2001;91:618–620. [PubMed]
37. U.S. General Accounting Office . School Facilities: America's schools report differing conditions. US General Accounting Office; Washington, DC: 1996.
38. Brener ND, Everett Jones S, Kann L, McManus T. Variation in school health policies and programs by demographic characteristics of US schools. J Sch Health. 2003;73:143–149. [PubMed]
39. Lowry R, Brener N, Lee S, et al. Participation in high school physical education—United States, 1991–2003. MMWR Morb Mortal Wkly Rep. 2004;53:844–847. [PubMed]
40. Story M, Kaphingst K, French S. The role of schools in obesity prevention. Future Child. 2006;16:109–142. [PubMed]
41. Wiecha JL, Finkelstein D, Troped PJ, Fragala M, Peterson KE. School vending machine use and fast-food restaurant use are associated with sugar-sweetened beverage intake in youth. J Am Diet Assoc. 2006;106:1624–1630. [PubMed]
42. Slusser WM, Cumberland WG, Browdy BL, Lange L, Neumann C. A school salad bar increases frequency of fruit and vegetable consumption among children living in low-income households. Public Health Nutr. 2007;10:1490–1496. [PubMed]
43. Neumark-Sztainer D, French SA, Hannan P, Story M, Fulkerson JA. School lunch and snacking patterns among high school students: associations with school food environment and policies. Int J Behav Nutr Phys Act. 2005;2:14. [PMC free article] [PubMed]
44. Anderson PM, Butcher KF. Reading, writing, and Raisinets: Are school finances contributing to children's obesity? 2007. [20 June 2007]. < w11177>.
45. Kahn E, Ramsey L, Brownson R, et al. The effectiveness of interventions to increase physical activity: a systematic review. Am J Prev Med. 2002;22(Suppl 4):73–107. [PubMed]
46. Race in American public schools: Rapidly resegregating school districts. 2002. <>.
47. Goodman E, Hinden B, Khandelwal S. Acuracy of teen and parental reports of obesity and body mass index. Pediatrics. 2000;106:52–58. [PubMed]
48. Diez Roux A, Merkin S, Arnett D, et al. Neighborhood of residence and incidence of coronary heart disease. N Engl J Med. 2001;345:99–106. [PubMed]
49. Pickett K, Pearl M. Multilevel analyses of neighborhood socioeconomic context and health outcomes: a critical review. J Epidemiol Community Health. 2001;55:111–122. [PMC free article] [PubMed]
50. Balfour J, Kaplan G. Neighborhood environment and loss of physical function in older adults: evidence from the Alameda County study. Am J Epidemiol. 2002;155:507–515. [PubMed]
51. Curtis L, Dooley M, Phipps S. Child well-being and neighborhood quality: evidence from the Canadian National Longitudinal Survey of Children and Youth. Soc Sci Med. 2004;58:1917–1927. [PubMed]