 |
|
 |
|
 |
 |
|
 |
 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
a Division of Adolescent Medicine, Department of Medicine, Children's Hospital Boston, Boston, Massachusetts
b Department of Internal Medicine, Veterans Affairs Medical Center
c Department of Pediatrics, Center for Human Growth and Development, University of Michigan Health System, Ann Arbor, Michigan
d Channing Laboratory, Brigham and Women’s Hospital, Boston, Massachusetts
 |
ABSTRACT |
|---|
 TOP ABSTRACT METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
METHODS. We performed a cross-sectional analysis of 17007 teens in the National Longitudinal Study of Adolescent Health. Using multivariate linear regression, we examined the association between adolescent self-reported physical activity and individual race/ethnicity stratified by gender, controlling for a wide range of sociodemographic, attitudinal, behavioral, and health factors. We used multilevel analyses to determine if the relationship between race/ethnicity and physical activity varied by the school attended.
RESULTS. Participants attended racially segregated schools; 
80% of Hispanic and black adolescent boys and girls attended schools with student populations that were <66% white, whereas nearly 40% of the white adolescents attended schools that were >94% white. Black and Hispanic adolescent girls reported lower levels of physical activity than white adolescent girls. There were more similar levels of physical activity reported in adolescent boys, with black boys reporting slightly more activities. Although black and Hispanic adolescent girls were more likely to attend poorer schools with overall lower levels of physical activity in girls; there was no difference within schools between black, white, and Hispanic adolescent girls' physical activity levels. Within the same schools, both black and Hispanic adolescent boys had higher rates of physical activity when compared with white adolescent boys.
CONCLUSIONS. In this nationally representative sample, lower physical activity levels in Hispanic and black adolescent girls were largely attributable to the schools they attended. In contrast, black and Hispanic males had higher activity levels than white males when attending the same schools. Future research is needed to determine the mechanisms through which school environments contribute to racial/ethnic disparities in adolescent physical activity and will need to consider gender differences in these racial/ethnic disparities.
Key Words: adolescents • physical activity • racial/ethnic disparities • school environments
Abbreviations: Add Health—National Longitudinal Study of Adolescent Health • ICC—intraclass correlation
Child and adolescent overweight is a growing public health problem.1–3 Over the last 2 decades, childhood overweight has increased sharply in all population groups, yet it has affected racial/ethnic minorities disproportionately.4–7 A biological explanation alone seems implausible when one considers that just 40 years ago rates of overweight were higher in white children than black children in almost every age group.5 Although there is increasing evidence of genetic determinants that contribute to obesity, in examining the roots of the racial/ethnic disparity in childhood and adolescent overweight one must consider a variety of environmental, behavioral, socioeconomic, and cultural influences in addition to genetic and metabolic factors.2
Physical activity is one potentially modifiable health behavior that may help to decrease the risk of overweight.8–11 Previous studies have demonstrated that racial/ethnic disparities in adolescents' physical activity levels exist, although few studies have looked separately at males and females.12–15 Despite the robust evidence establishing the association between race/ethnicity and physical activity, research addressing potential causes of the disparity is limited. Most previous research on childhood and adolescent physical activity has focused on individual or family attributes such as the adolescent's body mass index (BMI) or the family's income as potential predictors of the adolescent's activity.11–17 Some recent research has looked beyond the individual and family to the "built environment": ecological factors such as availability of sidewalks or number of playing fields in a school.18, 19 Although previous research has established a link between school physical education programs and physical activity participation,13 no research to date has examined whether other attributes of school environments such as racial/ethnic composition or median household income of the students contribute to the differences in adolescent physical activity by racial/ethnic groups.
Although recent research has begun to examine the potential impact of environments generally on health behaviors and health outcomes,20–25 little work has focused on childhood diseases and school environments. One notable exception is an important study that examined the influence of school-level socioeconomic status on adolescents' depressive symptoms and demonstrated significant school-level effects.26 The paucity of research focused on school environments' impact on the health of children exists despite the fact that schools are easily definable, relatively small environments in which students spend a significant portion of their day. One possible explanation for the lack of such previous studies is the difficulty in collecting adequate data to examine individual and school attributes simultaneously. The National Longitudinal Study of Adolescent Health (Add Health) is uniquely suited to examine adolescent health behaviors and health outcomes within school contexts. The Add Health study deliberately collected a nationally representative sample of adolescents clustered within a representative sample of schools from a large number of communities. Other investigators have used data from Add Health to establish the association of student-level variables and neighborhood factors with physical activity.13, 15 This study builds on that research by looking beyond school programs to specifically examine social attributes of school environments such as racial/ethnic composition and their potential to modify the association of race/ethnicity with physical activity.
Using the Add Health database, we sought to build on existing knowledge to answer the following research questions: (1) Do racial/ethnic disparities in adolescent boys' and girls' physical activity participation exist in a nationally representative school-based sample? (2) If so, are there individual attributes that influence differences in physical activity participation, and are these similar for adolescent boys and girls? (3) Are racial/ethnic disparities associated with the schools that students attend or related to differential participation within schools? (4) Are there characteristics of schools that influence participation in physical activity?
|
METHODS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Study Variables
Our outcome variable of interest was physical activity measured in self-reported number of times engaged in physical activity per week. This variable was created by summing the responses to 3 survey questions that assessed activity during the past 7 days, a standard time frame in many questionnaires used in epidemiologic studies.30, 31 The questions used were similar to those used and validated in a number of smaller studies.13 The study participants were asked about participation ("During the past week how many times did you ...") in 3 groups of activities, and responses were categorical (not at all, 1 or 2 times, 3 or 4 times, or 
5). The first of the 3 physical activity questions referred to riding a bike, roller-skating, or roller-blading; the second asked about participation in active sports such as soccer, baseball, or basketball; and the third asked about engaging in active exercises such as walking, running, dancing, or jumping rope. Because the questions did not include specifics about time spent, the number of hours per week spent engaging in activity could not be calculated. In addition, the metabolic equivalent values for the activities grouped in each question were variable.32 Because of these limitations, we chose to treat all activities equally. To create a summation of the 3 questions we assigned a numerical value to represent the number of times that the respondent had participated in each activity over the last week for each response using the mean of the range. For example, if the response was 1 to 2 times in the last week, we assigned a numerical value of 1.5. Because we felt that it was unlikely that a respondent would count the same activity >1 time per day in a week, we assigned the response of 5 a value of 6. The constructed outcome variable had a possible range of response of 0 to 18, and the sample responses had a nearly normal distribution. We chose to treat physical activity as a continuous variable because of the lack of a known threshold effect with respect to number of physical activities participated in per week (ie, there is no known difference between participating in activity 2 vs 3 times per week in adolescents that would necessitate categorization). The residual analysis of our regression models supported this linearity assumption (see below).
The key independent variable was participants' self-reported race/ethnicity, coded as white, black, or Hispanic. Sociodemographic covariates included student's age (continuous), maternal education level (categorized as less than high school, high school, some college, and college and beyond), household income measured as percentage of poverty level (continuous), and family structure, particularly the presence of a father in the home. We used the parental report of both household income and maternal education level. As mentioned above, we imputed those values of household income and maternal education level that were missing to avoid bias. We constructed a measure of percentage of poverty level by taking the ratio of household income to household size and comparing that to the standard poverty index for 1995 (the year the majority of the data were collected).33 We also included health-related covariates including smoking status and BMI (weight [kg]/height [m2]).
To explore possible differences among schools that more or fewer racial/ethnic minority students attend, we examined variables that describe the school. First, we included the percentage of white students attending the school as provided by the school administrators. We then constructed a variable to describe the school-level median household income of students attending each school.
Analyses
The data were analyzed by using bivariate and multivariate methods with Stata 8 (Stata Corp, College Station, TX).34 We performed our analyses stratified according to gender, a decision that was based on previous literature showing differential physical activity participation patterns between the genders.13, 15 Preliminary analyses showed a significant association of the interaction term of race/ethnicity and gender with physical activity. We felt that a stratified model would be more sensitive to picking up differences in physical activity patterns between the racial/ethnic groups, because it would not rely on the assumption of a multiplicative relationship between race and gender. We conducted bivariate analyses with t tests, 
2 tests, analysis of variance, and Pearson's correlation. We then used multivariate linear regression to examine the associations between the adolescent's race/ethnicity and their level of physical activity, adjusting for age, BMI, tobacco use, maternal education level, family structure, and household income. Poststratification weights were used for all of the initial analyses at the student level. In addition, survey-design effects of multiple stages of cluster sampling were controlled for in all initial analyses. Analyses using the sampling weights and accounting for survey-design effects produced very similar results to those using unweighted data. Preliminary analyses using 2-level models to compute the intraclass correlation (ICC) coefficients for our principal dependent variables suggested between-school variation for some measures. Thus, to account for the clustering of students within schools, we adjusted the SEs by using the Huber/White heteroscedastic-consistent estimator of the variance/covariance matrix, with the cluster correction as found in Stata 8, in linear-regression models evaluating associations with student attributes.35, 36
Next, we explicitly evaluated the influence of schools on physical activity participation by using multilevel analysis. We constructed a 2-level random-intercepts model to assess whether schools differ in overall levels of physical activity participation and added 2 school-level variables that describe the school's racial/ethnic composition and median income. Because of concern regarding strong associations between the racial/ethnic composition of the schools and the school-level median household income, we opted to add the school-level variables 1 at a time. In our final multilevel model, by simultaneously examining school- and student-level effects, we were able to distinguish between associations related to the attributes of the school versus those related to attributes of individual students. As mentioned above, because including sampling weights did not substantially change our results, we present the results of unweighted analyses to facilitate comparisons between the 1- and 2-level models. Regression diagnostic procedures showed no evidence of multicollinearity, heteroscedasticity, or substantial influence from outliers.
|
RESULTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Table 1 summarizes the survey sample's demographic and health characteristics for the combined male and female population according to race/ethnicity. Age and gender were similar across all 3 racial/ethnic groups. Hispanic and black adolescents had lower average socioeconomic status compared with white adolescents, as demonstrated by their lower household incomes (P < .001) and lower levels of maternal education (P < .001) (Table 1). In addition, black and Hispanic adolescents had higher BMIs than their white peers (Table 1).
|
There were significant differences in the schools that the different racial/ethnic groups attended. Most adolescents attended schools that were racially segregated, with >75% of both black and Hispanic adolescents attending schools with <67% white students and nearly 40% of white adolescents attending schools in which >94% of the students were white (Table 1). In addition, black and Hispanic adolescents attended schools with median incomes that were substantially lower than the median incomes of the schools attended by white adolescents (whites, $45000; blacks, $30000; Hispanics, $32500; P < .001) (Table 1).
In our multivariate model with males and females combined, we found evidence of effect modification of the relationship between race/ethnicity by gender (female x black ß = –2.26, P < .001; female x Hispanic ß = –2.23, P < .001), which reinforced our decision to perform additional analyses stratified according to gender.
Schools
In our t test assessing whether the percentage of the student body identified as white is associated with the school-level median household income, we found these 2 school-level factors to be independently associated (P < .0001).
Females
In our unadjusted linear-regression analyses using the continuous physical activity score, Hispanic girls engaged in physical activity 0.81 fewer times per week (P < .001) and black girls engaged in physical activity 0.69 fewer times per week (P < .001) than did white girls (Table 2, model A). After controlling for individual characteristics (age, socioeconomic status, household structure, BMI, tobacco use) in the multivariate linear models (Table 2, model B), it seemed that black adolescent girls engaged in physical activity 0.46 fewer times than did white girls (P < .001), whereas Hispanic girls engaged in activity 0.29 fewer times per week when compared with white girls (P = .012). Age remained inversely associated with physical activity levels.
|
To explore possible characteristics of the schools that influence the overall level of physical activity participation among students, we also examined 2 school-level variables: racial/ethnic composition of the school and school-level median household income (Table 2, models C and D, respectively). We found the school-level median household income of the student population (P < .001) to be predictive of female physical activity participation; however, the schools' ethnic composition (P = .15) failed to predict physical activity participation.
Males
In our unadjusted linear-regression analyses using the continuous physical activity score, we found no difference in physical activity participation among black, Hispanic, or white adolescent males (Table 3, model A). However, after controlling for age, BMI, smoking status, and socioeconomic status in a multivariate linear model (Table 3, model B), we observed that Hispanic adolescent males engaged in 0.34 more activities per week than white adolescent males (P = .008), whereas there was no difference found between black and white adolescent males in terms of physical activity participation. Similar to the findings in females, age remained strongly associated with physical activity levels, with younger adolescent males being more likely to engage in physical activity more times per week.
|
= 0.41 and P = .005 for the black adolescents; = 0.38 and P = .017 for Hispanic adolescents) in the same school. In other words, after controlling for school-level characteristics, race/ethnicity became independently associated with physical activity. In adolescent males, both the racial composition of the school (P = .01) and the school-level median household income (P = .001) were found to be predictive of physical activity participation (Table 3 models C and D). |
DISCUSSION |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Previous studies have shown a robust association between household income and physical activity.13, 14 However, we find that after taking into account characteristics of the school that the adolescent attended, individual household income was no longer significantly associated with adolescent physical activity in either males or females, thus suggesting that children in schools that serve poor families have lower participation in physical activity, but that poorer and richer students attending the same school have similar levels of physical activity. One logical explanation for this result is that schools with lower median incomes have fewer material resources (such as gymnasiums or athletic fields), human resources (coaches or physical education teachers), or programmatic support (such as fewer intramural and extramural sports programs), thus providing fewer opportunities for physical activity. One limitation of our study is that Add Health does not have adequate data to test this hypothesis.
Our findings further reinforce that adolescents today are attending schools that are strikingly segregated along racial/ethnic lines.37 Racial/ethnic minority adolescents are attending schools largely populated with other racial/ethnic minorities, whereas white adolescents are attending schools that are overwhelmingly white. However, our data do not allow us to define what it is about these segregated schools that might explain the different levels of physical activity participation within them. Because schools with more racial/ethnic minorities tend to be those with lower median incomes, it again could be an issue of fewer material resources. Although the racial/ethnic composition of a school was a predictor of lower physical activity in adolescent males independent of income level of students at the school, there are other factors that could lead to fewer resources being available at these schools, such as schools with more minorities being more likely to be clustered within poor school districts or receiving fewer resources for "nonacademic" facilities (such as sports fields, pools, and gymnasiums). On the other hand, one could hypothesize that schools with larger percentages of racial/ethnic minorities may have different cultural norms leading to lower levels of physical activity or fewer appropriate role models providing good examples of physically active adults. Surprisingly, the racial/ethnic composition of the schools that adolescent females attended had no bearing on their physical activity participation.
Because there were too few schools sampled per community, we were unable to distinguish between the influences of the schools versus those of the communities served by the schools. It is reasonable to speculate that the schools might also be acting as a proxy for the community. This seems especially likely given the growing body of evidence linking neighborhood and community of residence to health outcomes.20–25 However, it is equally likely that the schools may influence the health behaviors and outcomes of their students independent of the community in which the student lives. Schools are an appealing target for community-based research and intervention because they are discrete entities amenable to policy intervention and because school children spend 25% of their waking hours at school over the course of a year. It is also encouraging to think that schools could provide a buffer to negative community influences for adolescents growing up in impoverished communities.
An important next step will be to explore the mechanisms through which school environments influence physical activity participation. In doing so, it will be imperative to recognize the different patterns of influence in adolescent boys versus girls and in the different racial/ethnic groups. In this study we found that schools with lower school-level median household incomes have, on average, lower levels of physical activity participation, yet we have not identified the actual mechanisms that explain this association. It will be important to determine if schools with poorer students are less likely to have physical education programs, lower-quality programs, fewer physical resources, or less social capital such as physically fit role models. Interventions should focus on increasing physical activity in adolescent females across all racial/ethnic groups, because all females in this study were noted to have low physical activity participation; additional efforts should be aimed at decreasing the differential in physical activity participation between different racial/ethnic groups.
Future work should also address the differential patterns of physical activity participation among adolescent boys attending the same schools. Although previous studies have shown similar rates of physical activity participation among different racial/ethnic groups of adolescent boys,13, 15 to our knowledge this is the first study to demonstrate increased physical activity participation among black and Hispanic adolescent males when attending the same schools as white males. Although this could be viewed as positive, with more ethnic minorities engaging in health-promoting behaviors, it is also possible that the differential patterns of activity is a result of black and Hispanic males being more likely to be encouraged to participate in team sports rather than excel at academics. More research is needed to better understand these differences.
Additional research should also examine the impact of school environments on the weight status of adolescents, because obesity with its health consequences is an important health outcome that physical activity may modify. Physical activity is only one potential contributor to the obesity problem in general and the racial/ethnic disparity in risk of overweight more specifically. Finding a link between school attended and childhood overweight would further reinforce the importance of targeting schools for intervention.38
Several limitations of this study must be noted. First, the cross-sectional analysis eliminates our ability to establish causality; prospective studies are needed to clarify causal directions. Second, as stated above, because our data had too few schools within each community, we were not able to assess the extent to which the school was acting as a proxy for the neighborhood/community in which the adolescent resides. Additional research should explicitly assess not only the relative importance of schools and communities in influencing adolescent physical activity but also the different mechanisms by which these environments serve as facilitators or barriers to exercise.
An additional limitation to our study is our reliance on self-reported physical activity. Add Health, similar to a number of studies, relied on self-report in measuring adolescents' physical activity.17, 30 The questions rely on a standard 7-day time frame, include a wide variety of activities, and thus, in our minds, provide more information than the 2 questions used in the Youth Risk Behavior Surveillance and many other large surveys. In addition, these questions are subject to recall bias, but there is no reason to expect racial/ethnic differences in recall bias. Of course, it is possible that white adolescents may be more likely to recognize the social desirability of physical activity participation and thus overreport their physical activity relative to their Hispanic and black peers; however, we know of no literature demonstrating racial differences in overestimating activity level. If there are racial/ethnic differences in social-desirability bias, they would seem to vary by gender. This topic deserves additional research. The failure to determine time spent in each activity may lead to measurement bias; however, this would likely underestimate the true associations found. It is true, however, that those who report fewer activities per week could be doing activities of greater intensity, and thus, our measure could be misrepresenting activity overall. However, this difference would have to exist in a large percentage of our population to reverse our associations. Although we were forced to rely on gross measurements of physical activity, we were able to assess incremental differences in physical activity that could make a large difference in the health of adolescents. The magnitude of the effects on physical activity are modest, but we believe that the differences are clinically significant, given that even modest amounts of activity have been shown in adults to decrease the risk of developing diabetes and other chronic diseases.
|
CONCLUSIONS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
This research uses data from the Add Health project, a program designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris, and funded by National Institute of Child Health and Human Development grant P01-HD31921, with cooperative funding from 17 other agencies. Special acknowledgment is due 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 at the Carolina Population Center, 123 W Franklin St, Chapel Hill, NC 27516-2524 (addhealth@unc.edu).
Our thanks go to Sonya Demonner for assistance with data analysis and S. Jean Emans and Elizabeth Goodman for edits to earlier versions of this manuscript.
|
FOOTNOTES |
|---|
Address correspondence to Tracy K. Richmond, MD, MPH, 300 Longwood Ave, LO-635, Boston, MA 02115. E-mail: tracy.richmond{at}childrens.harvard.edu
The authors have indicated they have no financial relationships relevant to this article to disclose.
|
REFERENCES |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
||
|
||
Home | My Pediatrics | Journal Information | Current Issue | Past Issues | Subscriptions & Services | Contact Us Click here for faster international access © 2006 American Academy of Pediatrics. All rights reserved. |
||
|
||
