Built and social environments associations with adolescent overweight and activity.
ABSTRACT Little is known about the patterning of neighborhood characteristics, beyond the basic urban, rural, suburban trichotomy, and its impact on physical activity (PA) and overweight.
Nationally representative data (National Longitudinal Study of Adolescent Health, 1994-1995, n = 20,745) were collected. Weight, height, PA, and sedentary behavior were self-reported. Using diverse measures of the participants' residential neighborhoods (e.g., socioeconomic status, crime, road type, street connectivity, PA recreation facilities), cluster analyses identified homogeneous groups of adolescents sharing neighborhood characteristics. Poisson regression predicted relative risk (RR) of being physically active (five or more bouts/week of moderate to vigorous PA) and overweight (body mass index equal or greater than the 95th percentile, Centers for Disease Control and Prevention/National Center for Health Statistics growth curves).
Six robust neighborhood patterns were identified: (1) rural working class; (2) exurban; (3) newer suburban; (4) upper-middle class, older suburban; (5) mixed-race urban; and (6) low-socioeconomic-status (SES) inner-city areas. Compared to adolescents living in newer suburbs, those in rural working-class (adjusted RR[ARR] = 1.38, 95% confidence interval [CI] = 1.13-1.69), exurban (ARR = 1.30, CI = 1.04-1.64), and mixed-race urban (ARR = 1.31, CI = 1.05-1.64) neighborhoods were more likely to be overweight, independent of individual SES, age, and race/ethnicity. Adolescents living in older suburban areas were more likely to be physically active than residents of newer suburbs (ARR = 1.11, CI = 1.04-1.18). Those living in low-SES inner-city neighborhoods were more likely to be active, though not significantly so, compared to mixed-race urban residents (ARR = 1.09, CI = 1.00-1.18).
These findings demonstrate disadvantageous associations between specific rural and urban environments and behavior, illustrating important effects of the neighborhood on health and the inherent complexity of assessing residential landscapes across the United States. Simple classical urban-suburban-rural measures mask these important complexities.
- [Show abstract] [Hide abstract]
ABSTRACT: Purpose There is a need for empirical support of the association between the built environment and disability-related outcomes. This study explores the associations between community and neighborhood land uses and community participation among adults with acquired physical disability. Methods Cross-sectional data from 508 community-living, chronically disabled adults in New Jersey were obtained from among participants in national Spinal Cord Injury Model Systems database. Participants’ residential addresses were geocoded to link individual survey data with Geographic Information Systems (GIS) data on land use and destinations. The influence of residential density, land use mix, destination counts, and open space on four domains of participation were modeled at two geographic scales—the neighborhood (i.e., half mile buffer) and community (i.e., five mile) using multivariate logistic regression. All analyses were adjusted for demographic and impairment-related differences. Results Living in communities with greater land use mix and more destinations was associated with a decreased likelihood of reporting optimum social and physical activity. Conversely, living in neighborhoods with large portions of open space was positively associated with the likelihood of reporting full physical, occupational, and social participation. Conclusions These findings suggest that the overall living conditions of the built environment may be relevant to social inclusion for persons with physical disabilities.Annals of epidemiology 07/2014; · 2.95 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Background: Rural children are at greater risk for physical inactivity and obesity than their urban counterparts. The study purpose was to identify the perceived environmental factors in the home, school, and community that support physical activity among rural children. Methods: A concurrent mixed-methods study involving 99 parent-child (ages 6-11 years) dyads and 17 teachers from four rural communities was conducted in 2007. Qualitative methods included focus group (child only) and PhotoVOICE (parents, teachers) sessions. Quantitative methods included a survey (parent), accelerometry (child), and Geographic Information Systems analysis. All data were merged and results compared. Agreement from all three populations (children, parents, and teachers) provided the highest level of evidence followed by agreement among two populations, and then evidence from each population independently. Results: Within the home, physical activity opportunities include parenting practices (e.g. logistical support), the presence of siblings, and availability of physical activity equipment. At school, opportunities include age-appropriate recreation equipment, adequate recess and physical education time, policies to support physical activity and physical education programming, and school location. In the community, access to destinations (e.g. living close to town center), park and recreation availability (especially indoor opportunities), and resolution of perceived safety issues (e.g. wild or roaming animals) would enhance children's physical activity in rural settings. Conclusions: Opportunities for increasing rural children's physical activity exist. More research, interventions, and, evaluations on ways to promote physical activity in diverse rural settings is needed.140st APHA Annual Meeting and Exposition 2012; 10/2012
Built and Social Environments
Associations with Adolescent Overweight and Activity
Melissa C. Nelson, PhD, RD, Penny Gordon-Larsen, PhD, Yan Song, PhD, Barry M. Popkin, PhD
Little is known about the patterning of neighborhood characteristics, beyond the basic
urban, rural, suburban trichotomy, and its impact on physical activity (PA) and overweight.
Nationally representative data (National Longitudinal Study of Adolescent Health, 1994–
1995, n ?20,745) were collected. Weight, height, PA, and sedentary behavior were
self-reported. Using diverse measures of the participants’ residential neighborhoods (e.g.,
socioeconomic status, crime, road type, street connectivity, PA recreation facilities), cluster
analyses identified homogeneous groups of adolescents sharing neighborhood character-
istics. Poisson regression predicted relative risk (RR) of being physically active (five or more
bouts/week of moderate to vigorous PA) and overweight (body mass index equal or greater
than the 95th percentile, Centers for Disease Control and Prevention/National Center for
Health Statistics growth curves).
Six robust neighborhood patterns were identified: (1) rural working class; (2) exurban;
(3) newer suburban; (4) upper-middle class, older suburban; (5) mixed-race urban; and
(6) low-socioeconomic-status (SES) inner-city areas. Compared to adolescents living in newer
suburbs, those in rural working-class (adjusted RR[ARR]?1.38, 95% confidence interval
[CI]?1.13–1.69), exurban (ARR?1.30, CI?1.04–1.64), and mixed-race urban (ARR?1.31,
CI?1.05–1.64) neighborhoods were more likely to be overweight, independent of individual
SES, age, and race/ethnicity. Adolescents living in older suburban areas were more likely to be
physically active than residents of newer suburbs (ARR?1.11, CI?1.04–1.18). Those living in
low-SES inner-city neighborhoods were more likely to be active, though not significantly so,
compared to mixed-race urban residents (ARR?1.09, CI?1.00–1.18).
These findings demonstrate disadvantageous associations between specific rural and urban
environments and behavior, illustrating important effects of the neighborhood on health
and the inherent complexity of assessing residential landscapes across the United States.
Simple classical urban–suburban–rural measures mask these important complexities.
(Am J Prev Med 2006;31(2):109–117) © 2006 American Journal of Preventive Medicine
Built and social environments are important determi-
nants of obesity-related health behavior (e.g., physical
activity [PA]) and targets for intervention strategies.4
Research studying neighborhood effects on health has
relied largely on aggregate socioeconomic status (SES)
verweight and obesity have emerged as na-
tional public health concerns,1,2with adoles-
cence as an important developmental period.3
ments.5–9In an emerging literature exploring how
environment facilitates or restricts health behavior,
however, specific individual-level measures of neighbor-
hood factors (e.g., crime/safety,10–12street connectiv-
ity, road type/traffic,13–14and activity-related recre-
ation facility access15–18) have been independently
associated with PA.
Neither aggregate indices of SES nor specific aspects
of the built environment appear in isolation in neigh-
borhoods. Clearly, factors such as SES, crime, lack of
recreation facilities, and other community-level mea-
sures occur jointly.15In contrast to traditional risk
factor approaches to data analysis that examine the
independent effects of specific neighborhood charac-
teristics, pattern analyses allow examination of the
effects of multiple dimensions of the environment.
Both independent risk factor analysis and pattern anal-
ysis may make important contributions to understand-
ing how the environment affects behavior.
From the Division of Epidemiology and Community Health, Univer-
sity of Minnesota (Nelson), Minneapolis, Minnesota; and Depart-
ment of Nutrition, University of North Carolina (Nelson, Gordon-
Larsen, Popkin), Carolina Population Center (Nelson, Gordon-
Larsen, Popkin), and Department of City and Regional Planning
(Song), Chapel Hill, North Carolina
Address correspondence and reprint requests to: Melissa C. Nel-
son, PhD, RD, Division of Epidemiology and Community Health,
University of Minnesota, 1300 S. 2nd Street, WBOB Suite 300,
Minneapolis MN 55454-1015. E-mail: firstname.lastname@example.org.
The full text of this article is available via AJPM Online at
Am J Prev Med 2006;31(2)
© 2006 American Journal of Preventive Medicine • Published by Elsevier Inc.
0749-3797/06/$–see front matter
While the importance of covariance and joint effects
of neighborhood features has gained recognition in
this growing area of interest in environmental determi-
nants of obesity, the study of patterning to date has
been limited largely to index development as a means
of measuring very specific aspects of the built environ-
ment.13,19There is little empirical evidence describing
the diversity and covariance of community characteris-
tics using data-driven techniques within nationally rep-
resentative data sets, including a wide array of residen-
tial landscapes, most importantly, rural areas that are
By using pattern analyses to measure the effects of
multiple environmental characteristics on behavior,
this research fills an important gap in the literature.
Using data from a nationally representative, ethni-
cally diverse sample of adolescents, the aims of this
study were to (1) identify meaningful patterns of
sociodemographic and built features in neighbor-
hood environments that have been identified as
potentially important determinants of PA, and
(2) describe the cross-sectional associations between
these neighborhood patterns and adolescent resi-
dents’ PA and weight status.
Add Health is a school-based longitudinal survey of youths,
grades 7 through 12. A random sample of 80 high schools and
52 junior high feeder schools was selected. The Add Health
sample was designed to be nationally representative of stu-
dents in grades 7 through 12 in 1995 in the United States.
Survey procedures20were previously approved by the Institu-
tional Review Board at the University of North Carolina at
Chapel Hill. The Wave-I in-home survey (1994–1995) in-
cluded 20,745 adolescent participants. Analyses were con-
ducted in 2005–2006.
(n ?20,612, 99.4%) Wave-I participants were identified and
geocoded, using primarily street-segment matches from com-
mercial geographic information system (GIS) databases
(n ?17,119) or global positioning system (GPS) units
(n ?3242) (when a street-segment match was unavailable).
When neither was available, residential location assignments
used a ZIP?4/ZIP?2 or 5-digit ZIP centroid match, or the
respondent’s school location.15A relational database linked
the location of a participant’s home to (1) neighborhood
attributes, based on buffers around each home; (2) block
group, tract, and county attributes from U.S. Census and
other federal sources; and (3) Add Health participants’
survey responses. Add Health was designed for national
representation of youth, not for geographic representation.
Yet, the data include a wide array of geographic areas across
the United States.
location. Homestreetaddresses ofmost
Buffers for respondent locations. To assess the variety of
neighborhood characteristics, a 3-km buffer was drawn
around each respondent’s residential location using Euclid-
ean distance. While there is some suggestion that 5-mile
catchment areas may be relevant for adult PA,15,17it is likely
that smaller areas influence adolescent PA, where travel is
more limited. There is little empirical data to support appro-
priate buffer size for PA outcomes at a national level. Sensi-
tivity analyses were conducted to determine the appropriate
buffer size (3 km) for these analyses.
Physical activity facilities within 3 km. Commercially avail-
able, retrospective (1995), digitized Yellow Pages, using pro-
prietary 4-digit extensions to the Standard Industrial Classifi-
cation (SIC) codes, were obtained. These 8-digit codes
correspond with those used by the Census, allowing for the
detailed characterization of facility type. A comprehensive list
representing PA facility/resource types (n ?169) was com-
piled for these SIC codes.15SIC code counts were summed to
measure all activity-related facilities, and subdivided to specif-
ically assess parks. Park locations were verified using digital
aerial photographs from the U.S. Geological Survey.
Walkability within 3 km. High street connectivity, or “walk-
ability” (i.e., neighborhood street networks that are continu-
ous, integrated, and maximize linkages between starting
points and destinations, providing multiple route options)
has been positively associated with residential activity pat-
terns.13,21Indices of connectivity include (1) intersection
density (three-way and four-way intersections), (2) alpha
index (ratio of observed to maximum possible route alterna-
tives [circuitry] between nodes, where the maximum possible
circuits is the maximum number of links minus the number
of links in a minimally connected network), (3) gamma index
(ratio of observed node linkages to the maximum possible
links in the network), and (4) cyclomatic index (number of
route alternatives [circuits] between nodes).
Road type within 3 km. Road networks were mapped using
retrospective U.S. Census TIGER (topologically integrated
geographic encoding and referencing) line files (www.
census.gov/geo/www/tiger/). Road types were assessed using
Census feature class codes (www.census.gov/geo/www/tiger/
appendxe.asc) (i.e., feature Class A categories). The presence
of smaller, local roads (category A4x, which are more likely to
have single traffic lanes, sidewalks, and lower speed limits)
were of particular interest for this research, in comparison to
the presence of larger roadways (A1x) on which walking/
biking is more difficult (e.g., primary highways). Road types
were assessed as the proportion of total roadways and the
absolute total length.
Census measures. Census data reflecting individuals’ resi-
dential block groups were extracted from the 1990 Census of
Population and Housing summary tape file 3A (STF3A). A
block group is a relatively small administrative unit in the U.S.
Census (averaging 300 to 3000 residents). Variables used here
were education (proportion of adults aged ?25 years with a
college degree), minority (proportion of nonwhites), poverty
(proportion of people with incomes ?185% of poverty level),
housing units (proportion of housing units occupied by
renters, proportion vacant, median housing unit age), and
mobility (proportion of population living in same housing
unit since 1985, proportion of population working in county
American Journal of Preventive Medicine, Volume 31, Number 2 www.ajpm-online.net
of residence). The metropolitan statistical area (MSA) of Add
Health schools was also identified, and regions were broadly
categorized as urban, suburban, or rural.
Crime. Reported crimes (per 100,000 population) were as-
sessed using 1995 U.S. Federal Bureau of Investigation Uni-
form Crime Reporting county-level data from the National
Archive of Criminal Justice Data (www.icpsr.umich.edu/
NACJD/index.html), which have been shown to be associated
with PA levels in this sample.22For Add Health respondents
(n ?366) in counties with no available 1995 crime data, crime
rates were used from a previous year (1990 to 1994). For three
counties (n ?95 individuals), crime was imputed as average
reported crime in surrounding counties.
Physical Activity/Sedentary Behaviors
Daily PA (e.g., housework, active play, sports, exercise) was
self-reported using standard epidemiologic 7-day recall method-
books) employed questions similar to those used and validated
in other large-scale studies.23–26Questions asked—variations on
“During the past week, how many times did you . . .”—allowed
estimation of activity frequency (bouts/week) by metabolic
to 8 METs.27
Adolescents also reported sedentary behaviors (e.g., watch-
ing/playing TV/videos, video or computer games [hours/
week]), using recreation centers, and playing sports with
parent(s). Overall activity frequency was summed to deter-
mine total weekly MVPA or sedentary behavior, as well as
whether individuals met national recommendations for PA
(i.e., recommendation is five or more bouts/week of
MVPA)28and did not meet recommendations for sedentary
behavior (i.e., recommendation is TO NOT exceed 14 hours/
week “screen time”).29,30
Self-reported height and weight were used to calculate body
mass index (BMI) (kilograms/square meters).31The 95th
percentile of nationally representative data (2000 Centers for
Disease Control and Prevention/National Center for Health
Statistics growth curves), was used to classify overweight.32
Individuals aged 21 years (n ?9) were considered as 20-year-
olds for assessing overweight. Self-reported weight and height
have been shown to correctly classify a majority of Add Health
participants as overweight.33
Adolescents self-reported race/ethnicity; reports were vali-
dated during in-home parent interviews. Parents reported
highest level of achieved education, which was used to
estimate SES. Income was reported in $1000 increments and
imputed where missing, using parent occupation, family
structure, and school community.
Identifying Patterns in Environment
Cluster analyses were used to identify patterns of environmental
characteristics and to specify homogeneous, non-overlapping
clusters (or patterns) of neighborhoods sharing various mean-
partitioning data into different numbers of clusters4–10by
Euclidean distances between observations that were weighted
for national representation, using SAS FASTCLUS, SAS version
9 (Research Triangle Institute, Research Triangle Park NC,
2004). Representing different constructs of the neighborhood,
19 variables were used (Table 1). Z-score transformations of
variables were used to generate clusters, allowing for the appro-
priate weighting of variables with different scales.34
To identify initial cluster centers (i.e., seed values), 1000
iterations of each cluster procedure were conducted.35The
initial group center for each iteration was randomly generated.
The iteration with the largest overall r2value, which allowed for
the evaluation of relative heterogeneity between clusters (vs
homogeneity within clusters), was identified. Clusters best fitting
the data maximized this inter- to intra-variability ratio, yielding a
higher r2. (For the six-cluster solution series—i.e., the final
cluster solution—the maximum r2value identified through this
iterative process was 0.41.) Results of these numerous analyses
were assessed to identify common patterns appearing across
various procedures. The final presented clusters were those
representing the most robust data patterns.
Table 1. Measured constructs of the neighborhood environment
Income/wealth Income to poverty level (% less than 185%)
Home age (year structure was built)
Ethnicity (% minority)
Education (% with college degree)
Occupancy status (% owner occupied, % renter occupied, % vacant)
Mobility (% living in same house since 1985), median house age
Proportion working in county of residence
Serious crimes (arrests) per 100,000 persons
Proportion of A1 and A4 roads in 3-km buffer
Total length of A1 and A4 roads in 3-km buffer
Intersection density in 3-km buffer
Gamma index in 3-km buffer
Cyclomatic index in 3-km buffer (total route alternatives)
Alpha index in 3-km buffer (observed total route alternatives)
For each facility type: count in 3-km buffers
Socioeconomic status and environment
Street connectivity (walkability)
Recreation facilities for physical activity
August 2006 Am J Prev Med 2006;31(2)
Demonstrating Cluster Variability
Cluster analytic procedures detect underlying data patterns,
regardless of utility or substantive merit, but statistical meth-
ods for validating cluster analyses are limited.34To show that
clusters fit the data in a meaningful way, clusters are often
tested by predicting external variables not used to generate
the patterns (although associated with the clusters in theory).
To demonstrate meaningful variability between patterns
and to validate these findings, neighborhood clusters were
assessed as independent variables in generalized linear mod-
els predicting adolescent PA, sedentary behavior, and over-
weight. As another tool for comparison, broad neighborhood
characteristics were examined (e.g., broad urbanicity classifi-
cations of urban, suburban, and rural; median household
income; percent college-educated population; percent mi-
nority population), which have been used extensively in
All models controlled for important covariates (age,
race/ethnicity, parent education/income). Observations
with missing covariate or outcome data were excluded.
Participants who were severely disabled (n ?132) and/or
pregnant (n ?379) were also excluded. While logistic
regression is commonly used in health research, the odds
ratios yielded by these analyses substantially over-estimate
risk ratios (RRs) that are ?1.0 (and under-estimate those
that are ?1.0) when the outcome is not rare. The binomial
outcomes included in these analyses were not rare (i.e.,
prevalence ?10%), so Poisson regression was used with
robust variance estimates to generate valid estimates of the
adjusted relative risk (ARR) (and relatively conservative
confidence intervals [CIs]), instead of adjusted odds
Descriptive analyses of individual adolescent characteristics
used post-stratification sample weights, allowing results to be
nationally representative. Survey design effects of multiple
stages of cluster sampling were controlled using the survey
(SVY) procedure series in STATA, version 9.0 (Stata Corp,
College Station TX, 2004).
Descriptive Statistics and Final Cluster Solution
The analysis sample (n ?20,745) generating the neigh-
borhood clusters was composed of 50.1% males as well
as 68.5% white, 15.2% black, 11.4% Hispanic and 4.0%
Asian adolescents. Approximately 14.7% of partici-
pants’ parents had less than a high school education,
32.5% had graduated from high school (or had a
general equivalency diploma), 27.8% had some col-
lege, and 25.0% had a college degree or higher. Mean
participant age was 15.4 (?0.12) years.
Six robust neighborhood pattern types were identified
by the final cluster solution and were observed across
numerous iterations of analyses, representing non-over-
lapping groups of U.S. neighborhoods sharing various
attributes. These clusters include (1) rural working class;
(2) exurban (urban/suburban outgrowth); (3) new sub-
urban developments (referent); (4) older, upper-middle
class suburbia with highway access; (5) mixed-race urban;
and (6) low-SES inner-city neighborhoods (Table 2).
These neighborhood patterns are distinguished by impor-
generate the final cluster solution, including SES, race/
ethnicity, socioenvironment, crime, road type, street con-
nectivity, and recreation facilities (Table 3).
Table 2. Major neighborhood types identified through cluster analysis procedures
Rural working classLow SES, moderate-to-low minority population, little mobility
Low connectivity (i.e., low intersection density and few possible routes between any two points)
Low access to PA facilities, very low overall density of roadways
Moderate SES, low minority population
High % of relatively recently built housing units
High % of population commuting to work outside of county of residence
Low access to PA facilities, low street connectivity and low crime
High proportion of large arterial roadways
High SES, low minority population
Relatively recently built housing units
Low access to PA facilities, very poor street connectivity
Few roadways overall (with high proportion of these roadways being of local)
High SES population with somewhat mixed racial/ethnic composition, little mobility
High % of older housing units
Moderate access to PA facilities, moderate street connectivity in linking intersections (and
moderate-to-high number of alternative street routes between any two points)
High density of roads overall (with a moderate proportion of large arterial roadways)
Low SES, high poverty population
Moderate access to PA facilities, moderate-to-high street connectivity and crime
High density of local roadways
Low SES, very high minority and very high poverty population
Large proportion of older housing units
Very high access to PA facilities, very high street connectivity and intersection density
High crime, high density of local roadways
Low SES, inner city
PA, physical activity; SES, socioeconomic status.
American Journal of Preventive Medicine, Volume 31, Number 2 www.ajpm-online.net
Table 3. Mean frequency of specific neighborhood characteristics by clustera
Cluster 1Cluster 2Cluster 3
(n ? 3371)
(n ? 4280)
(n ? 3609)
(n ? 2582)
Rural working class
(n ? 4725)
(n ? 2178)
% population with a college
education (in block group)
% nonwhite population (in block
% population with income ?185%
poverty line (in block group)
% renter-occupied housing units (in
% vacant housing units (in block
Median house age (years) (in block
% population living in the same
house for ?5 years (in block
group) (i.e., % low morbidity)
% population working in county of
residence (in block group)
Number of facilities for physical
activity (in 3 km)
Number of parks (in 3 km)
Alpha index of road connectivity (in 3
Gamma index of road connectivity (in
Cyclomatic index of road connectivity
(in 3 km)
Intersection density (in 3 km)
Reported crime rate (in county)
Total length (meters) of major arterial
roadways (in 3 km)
Total length (meters) of local
roadways (in 3 km)
% roadways that are major arterial
roadways (in 3 km)
% roadways that are local roads (in 3
14.5 ? 0.1 (?0.56)
21.2 ? 0.3 (?0.11) 34.5 ? 0.2 (0.77)32.9 ? 0.3 (0.67)
17.0 ? 0.2 (?0.39)16.2 ? 0.2 (?0.45)
20.0 ? 0.4 (?0.41)15.7 ? 0.6 (?0.54) 15.0 ? 0.3 (?0.56) 36.5 ? 0.5 (0.06)
50.8 ? 0.5 (0.47)75.1 ? 0.6 (1.17)
38.1 ? 0.2 (0.35)
26.1 ? 0.4 (?0.23)14.7 ? 0.2 (?0.79)14.8 ? 0.1 (?0.78)
45.8 ? 0.3 (0.73)48.5 ? 0.4 (0.86)
20.3 ? 0.2 (?0.44)20.6 ? 0.3 (?0.43)20.0 ? 0.3 (?0.45) 21.6 ? 0.3 (?0.38)
47.9 ? 0.3 (0.87)54.6 ? 0.4 (1.18)
11.6 ? 0.2 (0.44)7.0 ? 0.2 (?0.10)7.5 ? 0.2 (?0.04)3.0 ? 0.1 (?0.57)8.8 ? 0.1 (0.12)8.9 ? 0.2 (0.12)
23.9 ? 0.1 (?0.19) 21.5 ? 0.2 (?0.38)
13.2 ? 0.1 (?1.03)
31.2 ? 0.2 (0.38)28.7 ? 0.2 (0.18)41.1 ? 0.2 (1.15)
62.6 ? 0.1 (0.47)
58.6 ? 0.3 (0.20)45.7 ? 0.3 (?0.65)
62.6 ? 0.2 (0.47)
45.8 ? 0.2 (?0.64) 54.5 ? 0.3 (?0.07)
70.5 ? 0.3 (?0.36)
67.4 ? 0.5 (?0.52)
75.1 ? 0.4 (?0.14)81.9 ? 0.3 (0.19)87.9 ? 0.2 (0.48)83.9 ? 0.4 (0.29)
3.3 ? 0.1 (?0.66) 5.7 ? 0.1 (?0.58) 8.1 ? 0.1 (?0.49)
32.5 ? 0.2 (0.37)
23.5 ? 0.2 (0.05)
68.7 ? 0.9 (1.65)
?0.01 ? 0.001 (?0.21)
0.25 ? 0.001 (0.16)
0.01 ? 0.002 (?0.20)
0.19 ? 0.001 (?0.49)
0.01 ? 0.001 (?0.20)
0.16 ? 0.001 (?0.81)
0.04 ? 0.003 (?0.11)
0.22 ? 0.001 (?0.20)
0.02 ? 0.002 (?0.16)
0.26 ? 0.001 (0.22)
0.55 ? 0.018 (1.21)
0.35 ? 0.002 (1.19)
0.51 ? 0.001 (0.22)0.46 ? 0.001 (?0.51)0.44 ? 0.001 (?0.86)0.48 ? 0.001 (?0.22) 0.51 ? 0.001 (0.21) 0.57 ? 0.001 (1.23)
84.4 ? 1.2 (?0.86) 120.9 ? 2.1 (?0.77)185.8 ? 2.4 (?0.61)548.8 ? 3.5 (0.32)584.7 ? 3.2 (0.41)
1172.3 ? 5.0 (1.91)
5.0 ? 0.1 (?1.00)
3635.5 ? 29.3 (?0.83)
114.6 ? 13.5 (?0.91)
8.5 ? 0.1 (?0.81)
4483.8 ? 46.6 (?0.53)
17,937.3 ? 139.6 (0.69)
14.0 ? 0.2 (?0.51)
6003.6 ? 37.5 (?0.004)
2910.7 ? 89.7 (?0.66)
33.2 ? 0.2 (0.53)
6195.1 ? 30.2 (0.06)
18,365.9 ? 180.9 (0.73)
32.3 ? 0.2 (0.48)
6997.9 ? 35.0 (0.34)
11,229.8 ? 159.4 (0.09)
53.9 ? 0.2 (1.65)
9640.2 ? 61.6 (1.27)
16,951.1 ? 197.7 (0.60)
55,425.6 ? 381.8 (?1.07)79,429.1 ? 810.7 (?0.81)113,899.3 ? 971.7 (?0.44)
206,137.1 ? 682.8 (0.57)204,157.2 ? 845.8 (0.55)291,885.8 ? 960.8 (1.51)
0.1 ? 0.02 (?0.78)
16.1 ? 0.1 (1.99)
1.6 ? 0.1 (?0.51) 6.6 ? 0.1 (0.34)4.1 ? 0.1 (?0.09)4.5 ? 0.1 (?0.03)
77.2 ? 0.2 (0.20)
62.8 ? 0.2 (?1.26)
80.2 ? 0.1 (0.51)
73.7 ? 0.1 (?0.16)
76.8 ? 0.1 (0.16)76.0 ? 0.1 (0.08)
aMean ? standard error (mean z-score) of neighborhood characteristics (unweighted).
Am J Prev Med 2006;31(2)
Using the neighborhood clusters, important differ-
ences in the relative risk of overweight by community
type were identified. Compared to adolescents living in
newer suburban developments, those who lived in
(1) rural working class, (2) exurban, and (3) mixed-
race urban neighborhoods were 30% to 40% more
likely to have a BMI?95th percentile of age- and
gender-specific national growth curves (Figure 1), in-
dependent of SES, adolescent age, and race/ethnicity.
Conversely, analyses using individual components or
traditional measures of neighborhood characteristics
show less-clear associations (Table 4). Using traditional
analyses, no difference in the risk of overweight be-
tween low- and moderate-SES communities was found,
although high-SES communities were less likely to be
overweight, compared to moderate-SES communities.
There was no difference in overweight status by the
race/ethnicity of the community. The traditional ur-
ban–suburban–rural status (using MSAs) suggest a
lower likelihood of overweight in adolescents residing
in urban areas, but no differences in rural and subur-
ban neighborhoods. A comparison of the traditional
urban–suburban–rural breakdown with the approach
presented in Table 3 highlights the differences in the
Among adolescents living in older suburban neigh-
borhoods, 39% reported engaging in five or more
bouts of weekly MPVA, compared to 28% of adoles-
cents living in mixed-race urban areas (data not
shown). These findings indicate notable distinctions
within suburban and urban community types; for
example, independent of SES, race/ethnicity, and
age, adolescents living in older suburban develop-
ments were 11% more likely to be physically active
(ARR?1.11, CI?1.04–1.18) (Figure 2). In addition,
those living in low-SES inner-city areas were more
likely to be active compared to those in mixed-race
urban neighborhoods (RR?1.09, CI?1.00–1.18).
Figure 1. Adjusted risk ratios (95% confidence intervals) of
overweight (BMI ?95th percentile) by data-driven neighbor-
hood cluster definitions (n ? 19,029). Note: adjusted for
household income, parental education, adolescent age, and
Table 4. Adjusted risk ratios (95% confidence intervals) of adolescent overweight (?95th percentile body mass index) using
broad, independent, and traditional measures of neighborhood characteristics
Adjusted risk ratio of overweight by neighborhood features
Low tertile Moderate tertileHigh tertile
Median household incomea
% of population with
% of minority populationc
MSA classification of schoold
Rural: 1.9 (0.94–1.27)
1.0 (ref)1.13 (0.97–1.31)
Urban: 0.85 (0.75–0.97)Suburban: 1.0 (ref)
Note: Adjusting for individual parental education, household income, race/ethnicity, and age of adolescent.
aLow-income tertile: median household income ?$23,775/year, moderate tertile: ?$23,775 to ?$36,440/year, high tertile: ?$36,440/year at the
census block–group level (n ? 19,029 in model).
bLow education tertile: ?14.4% of population college educated; moderate tertile: ?14.4% to ?26.5%; high tertile: ?26.5% at the census
block–group level (n ? 19,025 in model).
cLow minority tertile: ?6.1% of population is minority, moderate tertile: ?6.1% to ?49.4%, high tertile: ?49.4% at the census block–group level
(n ? 19,025 in model).
dMSA, metropolitan statistical area (n ? 18,709 in model).
Figure 2. Adjusted risk ratios (and 95% confidence intervals)
of achieving five or more bouts of moderate-to-vigorous
physical activity (MVPA)/week by data-driven neighborhood
cluster definition and type (rural, suburban and urban) (n ?
19,531, across three models). Note: Adjusted for household
income, parental education, adolescent age and race/ethnic-
American Journal of Preventive Medicine, Volume 31, Number 2 www.ajpm-online.net
Independent of SES, race/ethnicity, and age, teens
living in low-SES inner-city and older suburban areas
were least likely to report playing a sport with a
parent(s), and those living in the low-SES inner-city,
mixed-race urban, and older suburban areas were the
most likely to report using a neighborhood recre-
ation facility. High levels of screen time (?14 hours/
week, which is above recommended levels) were most
likely among those living in low-SES inner-city neigh-
borhoods (Table 5).
Using cluster analysis, six robust patterns in residential
neighborhood characteristics were identified, incorpo-
rating a range of sociodemographic and built environ-
ment characteristics: (1) rural working class, (2)
exurban, (3) new suburban, (4) older suburban, (5)
mixed-race urban, and (6) low-SES inner-city areas. To
our knowledge, this is the first study using data-driven
techniques to characterize neighborhoods by sociode-
mographic and built environment features in a nation-
ally representative survey of adolescents. These results
show differences in neighborhood patterns by adoles-
cent PA and overweight (such as the disadvantageous
associations between rural and urban environments
and health). Rural populations have been particularly
understudied; these findings indicate that these indi-
viduals have unique neighborhood characteristics that
deserve further attention.
These findings show other important differences by
activity and weight status. Contrary to recent reports
supporting positive relationships between sprawl and
adult overweight/obesity,19,37,38the current results in-
dicate some beneficial associations between suburban
living and activity/overweight. Cross-sectionally, these
findings indicate that adolescents living in rural work-
ing class, exurban, and mixed-race urban areas were at
the highest risk of overweight compared to those in the
newer suburban cluster type. Teens living in older
suburban communities were most likely to be physically
active. In adolescence, there may be protective factors
shared by those living in suburban communities (e.g.,
school-based sports/activity facilities, community orga-
nizations, low crime) overriding the deleterious effects
of suburbia, which are believed to be at play in restrict-
ing activity of adult residents (e.g., low walkability,
dependence on automobiles).
The natural patterns of neighborhood characteristics
underlying these data represent diverse settings that
span urban, suburban, and rural regions, and are
shown to be valid through empirical evidence and
theoretical frameworks. For example, these results are
consistent with previous findings that low-income, ra-
cial/ethnic minority, and rural populations are less
physically active and more overweight and obese.3,39–41
In addition, the findings are supported by theoretical
Table 5. Adjusted risk ratios (and 95% confidence intervals) of adolescent physical activity and sedentary behavior by detailed data-driven neighborhood cluster
?5 bouts/week of any
activity (n ? 19,531)
Played a sport with a parent in
the last month (n ? 19,495)
Uses a neighborhood recreation
center (n ? 19,424)
?14 hours/week of TV/video
viewing and video/computer
gaming (n ? 19,521)
Note: Poisson regression, adjusting for parental education, household income, race/ethnicity, and age of adolescent.
SES, socioeconomic status.
August 2006Am J Prev Med 2006;31(2)
constructs of transect planning, an approach to urban
planning emphasizing breadth and range in commu-
nity design to recognize a broad range of human and
environmental needs, rather than a “one size fits all”
approach.42These neighborhood patterns overlap
many of the eco-zones outlined in the transect planning
literature, which to date has been based largely on
theory and empirical evidence from smaller geographic
samples with restrictive analytic approaches. Guided by
these eco-zones in the transect theory literature, in
conjunction with the empirically derived cluster defini-
tions of this study (e.g., described by mean frequency of
specific neighborhood characteristics by clusters),
these neighborhood types may be re-created and ex-
plored in other data sets. In characterizing multidimen-
sional neighborhood features, it is understood that they
vary in structure and sociodemographic composition,
and likely have differential impacts on population
health behavior. By better understanding the complex-
ities of today’s American residential landscape, there is
the opportunity to better conceptualize how to en-
hance built and social environments and tailor solu-
tions that will promote healthy lifestyles.
While these data show important dimensions of U.S.
communities and their association with health behav-
ior, they are not without limitation. First, the data on
PA and BMI are derived from self-reported measures,
which are subject to error and bias. Second, these
findings are cross-sectional, thus limiting causal infer-
ence, and may be influenced by residual confounding
(due to unmeasured characteristics for which there is
no control). Third, it is difficult to assess the true
validity of the cluster method, given that cluster analysis
will detect underlying patterns in data, regardless of
meaning or utility. To address this issue, this research
utilized several approaches, similar to those used in
other previous applications of cluster methods, includ-
ing (1) assessing numerous iterations of cluster solu-
tions to ensure that the final solution represents robust
patterns in the data, (2) comparing these findings to
established theoretical frameworks, such as the eco-
zones of transect planning, and (3) contrasting these
clusters with external variables (that were not used in
the cluster creation), such as PA and overweight status.
The latter is critical for ensuring meaningful interpret-
ability of the cluster results.
Finally, the strength in this study of using data charac-
terizing multiple sites across the nation may also limit the
all of the neighborhood characteristics are relevant to
individual participants, they represent varying degrees of
“neighborhood,” potentially carrying varying degrees of
influence. Methods for defining one’s neighborhood are
highly debated; there is no consensus on appropriate
buffer sizes to capture relevant exposure areas. The
exposure areas assessed here may not be the most appro-
priate size (e.g., 1-mile buffers, county-level crime). In-
deed, this is an important area for future research,
including detailed analyses on methods of measuring
environmental characteristics and important aspects of
influential neighborhood features.
This is possibly the first research characterizing the
national landscape in this way, understanding how
neighborhoods function as a whole (rather than as
individual components) and how they are associated
with adolescent health behaviors and outcomes. Indi-
viduals live in neighborhoods rather than income
brackets, and are simultaneously affected by factors
such as crime, facility access, and street connectivity,
which work in concert to affect health behavior. While
traditional risk factor analysis provides important in-
sights into the association between environment and
behavior, these findings show that broad, traditional
measures of neighborhood characteristics (e.g., me-
dian household income) may not capture the fine-grain
detail and complexity needed to better understand how
environments influence behavior. Thus, using pattern
analysis with detailed environmental measures supple-
ments the understanding of covariance in the environ-
ment and important environment–behavior relation-
ships. Effective population-wide health promotion
strategies and public policies need to address pre-
existing neighborhoods, which are composed of a
variety of factors, many of which may be important
determinants of activity patterns.
The patterns of neighborhood features identified
here show meaningful variation, supported by urban
planning theory and empirical evidence. Not only do
these findings help illustrate the important effects of
neighborhood on health, but they also demonstrate the
inherent complexity of these relationships. This re-
search highlights the extent to which neighborhoods
vary, as well as how individuals may function differently
in different environments, and points to the challenges
of increasing population-wide PA through community
design. Future research is needed to explore the spe-
cific mechanisms through which neighborhood form
affects population health, as well as the interactive
effects of a spectrum of co-varying community charac-
teristics and the extent to which effective intervention
and policy strategies can be tailored to promote healthy
Funding for this study and the development of the prelimi-
nary spatial measures comes from the National Institutes of
Health, including the National Institute of Child Health and
Human Development (NICHD) (R01-HD39183-01, R01
HD041375-01, and K01 HD044263-01), National Institute of
Diabetes and Digestive and Kidney Diseases (DK56350),
National Instituteof Environmental
(P30ES10126); a Cooperative Agreement with the Centers for
Disease Control and Prevention (CDC SIP 5-00); and a New
Investigator Dissertation Award from the Robert Wood John-
son Foundation’s Active Living Research Program (050752).
American Journal of Preventive Medicine, Volume 31, Number 2 www.ajpm-online.net
We also thank the Spatial Analysis Unit at the University of
North Carolina at Chapel Hill, particularly Phil Page, Jay
Stewart, and Evan Hammer, for their assistance in data
collection and processing.
This research used data from Add Health, a program
project designed by J. Richard Udry, Peter S. Bearman, and
Kathleen Mullan Harris, and funded by the NICHD (P01-
HD31921), with cooperative funding from 17 other agen-
cies. Special acknowledgment is due Ronald R. Rindfuss
and Barbara Entwisle for assistance in the original design.
Anyone interested in obtaining data files from Add Health
should contact Add Health, Carolina Population Center,
123 W. Franklin Street, Chapel Hill NC, 27516-2524
No financial conflict of interest was reported by the authors
of this paper.
1. Flegal KM, Carroll MD, Ogden CL, Johnson CL. Prevalence and trends in
obesity among U.S. adults, 1999–2000. JAMA 2002;288:1723–7.
2. Ogden CL, Flegal KM, Carroll MD, Johnson CL. Prevalence and trends in
overweight among U.S. children and adolescents, 1999–2000. JAMA
3. Gordon-Larsen P, Adair LS, Nelson MC, Popkin BM. Five-year obesity
incidence in the transition period between adolescence and adulthood: the
National Longitudinal Study of Adolescent Health. Am J Clin Nutr
4. Institute of Medicine. Does the built environment influence physical
activity? Examining the evidence. Washington DC: National Academy of
Sciences, 2005 (Transportation Research Board special report 282).
5. Diez Roux AV. Multilevel analysis in public health research. Annu Rev
Public Health 2000;21:171–92.
6. Diez Roux AV. Investigating neighborhood and area effect on health. Am J
Public Health 2001;91:1783–89.
7. Diez Roux AV, Merkin SS, Arnett D, et al. Neighborhood of residence and
incidence of coronary heart disease. N Engl J Med 2001;345:99–106.
8. Yen IH, Kaplan GA. Poverty area residence and changes in physical activity
level: evidence from the Alameda County Study. Am J Public Health
9. Sundquist J, Malmstrom M, Johansson S-E. Cardiovascular risk factors and
the neighbourhood environment. Int J Epidemiol 1999;28:841–5.
10. Handy S, Clifton K. Local shopping as a strategy for reducing automobile
travel. Transportation 2001;28:317–46.
11. Handy SL, Clifton KJ, Fisher J. The effectiveness of land use policies as a
strategy for reducing automobile dependence: a study of Austin neighbor-
hoods. Austin TX: Center for Transportation Research, Southwest Region
University Transportation Center, 1998.
12. Booth ML, Owen N, Bauman A, Clavis O, Leslie E. Social-cognitive and
perceived environment influences associated with physical activity in older
Australians. Prev Med 2000;31:15–22.
13. Greenwald MJ, Boarnet MG. Built environment as determinant of walking
behavior: analyzing nonwork pedestrian travel in Portland, Oregon. Trans-
portation Res Rec 2001;1780:33–42.
14. Parsons Brinckerhoff Quade &Douglas. The pedestrian environment.
Portland OR: 1000 Friends of Oregon, 1993.
15. Gordon-Larsen P, Nelson MC, Page P, Popkin BM. Inequality in the built
environment underlies key health disparities. Pediatrics 2006;117:417–24.
Available at: www.pediatrics.org/cgi/doi/10.1542/peds.2005-0058.
16. Giles-Corti B, Donovan RJ. Socioeconomic status differences in recre-
ational physical activity levels and real and perceived access to a supportive
physical environment. Prev Med 2002;35:601–11.
17. Sallis JF, Hovell MF, Hofstetter CR, et al. Distance between homes and
exercise facilities related to frequency of exercise among San Diego
residents. Public Health Rep 1990;105:179–85.
18. Brownson RC, Baker EA, Housemann RA, Brennan LK, Bacak SJ. Environ-
mental and policy determinants of physical activity in the United States.
Am J Public Health 2001;91:12.
19. Ewing R, Schmid T, Killingsworth R, Zlot A, Raudenbush S. Relationship
between urban sprawl and physical activity, obesity, and morbidity. Am J
Health Promot 2003;18:47–57.
20. Popkin BM, Udry JR. Adolescent obesity increases significantly in second
and third generation U.S. immigrants. J Nutr 1998;128:701–6.
21. Saelens BE, Sallis JF, Frank LD. Environmental correlates of walking and
cycling: findings from the transportation, urban design, and planning
literatures. Ann Behav Med 2003;25:80–91.
22. Gordon-Larsen P, McMurray RG, Popkin BM. Determinants of physical
activity and inactivity patterns. Pediatrics 2000;105:e83.
23. Sallis JF, Buono MJ, Roby JJ, Micale FG, Nelson JA. Seven-day recall and
other physical activity self-reports in children and adolescents. Med Sci
Sport Exerc 1993;25:99–108.
24. Andersen RE, Crespo CH, Bartlett SJ, Cheskin LJ, Pratt M. Relationship of
physical activity and television watching with body weight and level of
fatness among children: results from the Third National Health and
Nutrition Examination Survey. JAMA 1998;279:938–42.
25. Baranowski T. Validity and reliability of self-report measures of physical
activity: an information processing perspective. Res Q Exerc Sport
26. Pate RR, Heath GW, Dowda M, Trost SG. Associations between physical
activity and other health behaviors in a representative sample of U.S.
adolescents. Am J Public Health 1996;86:1577–81.
27. Ainsworth B, Haskell WL, Leon AS. Compendium of physical activities:
classification of energy costs of human physical activities. Med Sci Sport
28. Pate RR, Pratt M, Blair SN, et al. Physical activity and public health: a
recommendation from the Centers for Disease Control and Prevention and
the American College of Sports Medicine. JAMA 1995;273:402–7.
29. American Academy of Pediatrics Committee on Public Education. Chil-
dren, adolescents, and television. Pediatrics 2001;107:423–6.
30. American Academy of Pediatrics Committee on Public Education. Media
violence. Pediatrics 2001;108:1222–6.
31. WHO Expert Committee. Physical status: the use and interpretation of
anthropometry. Geneva: World Heath Organization, 1995 (WHO Techni-
cal Report Series 854).
32. Centers for Disease Control and Prevention. Growth charts 2000: United
States. National Center for Health Statistics. Available at: www.cdc.gov/
growthcharts. Accessed April 21, 2003.
33. Goodman E, Hinden BR, Khandelwal S. Accuracy of teen and parental
reports of obesity and body mass index. Pediatrics 2000;106:52–8.
34. Aldenderfer MS, Blashfield RK. Cluster analysis. Beverly Hills CA: Sage,
1984 (Sage University Paper Series on Quantitative Applications in Social
35. Nelson MC, Gordon-Larsen P, Adair LS, Popkin BM. Adolescent physical
activity and sedentary behavior: patterning and long-term maintenance.
Am J Prev Med 2005;28:259–66.
36. McNutt LA, Wu C, Xue X, Hafner JP. Estimating the relative risk in cohort
studies and clinical trials of common outcomes. Am J Epidemiol
37. Lopez R. Urban sprawl and risk for being overweight or obese. Am J Public
38. Vandegrift D, Yoked T. Obesity rates, income, and suburban sprawl: an
analysis of U.S. states. Health Place 2004;10:221–9.
39. Gordon-Larsen P, Nelson MC, Popkin BM. Meeting national activity and
inactivity recommendations: adolescence to adulthood. Am J Prev Med
40. Gordon-Larsen P, McMurray RG, Popkin BM. Adolescent physical activity
and inactivity vary by ethnicity: the National Longitudinal Study of Adoles-
cent Health. J Pediatr 1999;135:301–6.
41. Patterson PD, Moore CG, Probst JC, Shinogle JA. Obesity and physical
inactivity in rural America. J Rural Health 2004;20:151–9.
42. DuanyA, TalenE.Transect
August 2006Am J Prev Med 2006;31(2)