The Urban Built Environment and Obesity in New York City: A Multilevel Analysis
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Weight Control; Health Promoting Community Design
The Urban Built Environment and Obesity in New
York City: A Multilevel Analysis
Andrew Rundle, DrPH; Ana V. Diez Roux, MD, PhD, MPH; Lance M. Freeman, PhD; Douglas Miller,
MS; Kathryn M. Neckerman, PhD; Christopher C. Weiss, PhD
Abstract INTRODUCTION
Purpose. To examine whether urban form is associated with body size within a densely- Since the mid-1970s the United
settled city. States has experienced an epidemic of
Design. Cross-sectional analysis using multilevel modeling to relate body mass index (BMI) overweight and obesity.1 Recent re-
to built environment resources. search suggests that built environment
Setting. Census tracts (n 5 1989) within the five boroughs of New York City. characteristics affect rates of obesity by
Subjects. Adult volunteers (n 5 13,102) from the five boroughs of New York City recruited influencing physical activity patterns.2,3
between January 2000 and December 2002. Urban planning and public health
Measures. The dependent variable was objectively-measured BMI. Independent variables research suggests that pedestrian-ori-
included land use mix; bus and subway stop density; population density; and intersection ented environments, characterized by
density. Covariates included age, gender, race, education, and census tract–level poverty and high street connectivity, mixed land
race/ethnicity. use, and high population density,
Analysis. Cross-sectional multilevel analyses. encourage travel by walking and bi-
Results. Mixed land use (Beta 5 2.55, p , .01), density of bus stops (Beta 5 2.01, p , cycling. Also, by reducing reliance on
.01) and subway stops (Beta 5 2.06, p , .01), and population density (Beta 5 2.25, p , privately-owned vehicles, the provision
.001), but not intersection density (Beta 5 2.002) were significantly inversely associated with of public transit promotes pedestrian
BMI after adjustment for individual- and neighborhood-level sociodemographic characteristics. activity. The increase in transportation-
Comparing the 90th to the 10th percentile of each built environment variable, the predicted related activity associated with these
adjusted difference in BMI with increased mixed land use was 2.41 units, with bus stop characteristics of the built environ-
density was 2.33 units, with subway stop density was 2.34 units, and with population ment contributes to an overall increase
density was 2.86 units. in activities related to daily living. In
Conclusion. BMI is associated with built environment characteristics in New York City. accordance with previous work regard-
(Am J Health Promot 2007;21[4 Supplement]:326–334.) ing urban form and travel behavior,
Key Words: Body Mass Index, Land Use Mix, Public Transit, Population Density, recent research finds an association
Prevention Research. Manuscript format: research; Research purpose: modeling/ between environmental characteristics
relationship testing; Study design: nonexperimental; Outcome measure: biometric; such as population density, availability
Setting: local community; Health focus: weight control; Strategy: built environment; of nearby destinations, and intersec-
Target population: adults; Target population circumstances: education/income tion density and both self-reported and
level, geographic location, and race/ethnicity objective measures of physical activity,
walking, and biking.4–6 These features
of the built environment are thought
Andrew Rundle, DrPH, is with the Department of Epidemiology, Mailman School of Public to increase active transportation and
Health, New York, New York. Ana V. Diez Roux, MD, PhD, MPH, is with the Department of provide independence from the need
Epidemiology, School of Public Health, University of Michigan, Ann Arbor, Michigan. Lance M. for private automobiles to accomplish
Freeman, PhD, is with the Graduate School of Architecture, Planning and Preservation; and daily tasks and in turn to lower body
Douglas Miller, MS, Kathryn M. Neckerman, PhD, and Christopher C. Weiss, PhD, are with the size.4,7
Institute for Social and Economic Research and Policy, Columbia University, New York, New York. Such data suggest that compared
with automobile-dependent environ-
Send reprint requests to Andrew Rundle, DrPH, Department of Epidemiology, Mailman School of ments, pedestrian-oriented environ-
Public Health, 722 West 168th Street, Room 730, New York, NY 10032; Agr3@columbia.edu.
ments should be associated with lower
This manuscript was submitted May 8, 2006; revisions were requested September 14, 2006; the manuscript was accepted for rates of obesity. At the state level,
publication September 17, 2006. Vandegrift and colleagues have found
Copyright E 2007 by American Journal of Health Promotion, Inc. an association between obesity rates
0890-1171/07/$5.00 + 0 and suburban sprawl.8 Other analyses
326 American Journal of Health Promotion
Health Promotion hepr-21-00-07.3d 12/1/07 10:19:18 326 Cust # 06050861Rof county- or metropolitan-level mea- settled areas such as New York City. It is and its suburbs. Data collection took
sures of ‘‘sprawl,’’ typically associated possible that at an extreme high end of place at six community-based health
with automobile-dependent environ- population density, relative variation in centers, two community hospitals, and
ments, find positive associations be- built environment characteristics no six medical centers, and through the
tween sprawl and body mass index longer influences physical activity pat- New York Blood Center. Volunteers
(BMI), adjusting for individual-level terns. A firm understanding of how who presented themselves at these sites
characteristics.3,9,10 physical activity and BMI associate with were enrolled. Research staff con-
Further evidence for the association variation in built environment charac- ducted extensive recruitment efforts in
between the built environment and teristics across the range of built community settings such as health and
body size is provided by two recent environments is required before local neighborhood fairs. The study was also
studies that included neighborhood- policy and planning initiatives can be extensively publicized in the city to
level measures. A study based in Perth, designed in this area. encourage volunteers to participate.16
Australia, found that individuals who Qualifications for enrollment included
METHODS
resided on a highway or a street with literacy level high enough to complete
either no sidewalk or a sidewalk on a follow-up questionnaire and an age
only one side of the street, or who Design
of at least 30 years. At the time of
perceived an absence of paths within Using a cross-sectional design, the
enrollment into the cohort, written
walking distance, were more likely to present study addresses this gap in our
informed consent was obtained in
be overweight.11 Associations were also knowledge by examining whether built
person by research staff. The baseline
found between overweight and re- environment characteristics similar to
data are maintained by the Herbert
duced access to recreational facilities those shown in other locales to be
Irving Comprehensive Cancer Center
associated with physical activity and/or
as well as the perceived absence of at Columbia Presbyterian Medical
body size are associated with BMI in
shops within walking distance.11 An- Center. Analyses of body mass index,
New York City. Results presented here
other study by Frank and colleagues individual demographic variables, and
are from secondary data analyses of
measured body mass, time spent walk- appended neighborhood characteris-
existing survey data on socio-demo-
ing, and time spent in a car among tics were approved by the Columbia
graphic characteristics and objectively
residents of the Atlanta, Georgia, met- Presbyterian Medical Center Institu-
measured height and weight obtained
ropolitan area.7 Positive associations tional Review Board.
from 13,102 residents of New York
were found between intersection den- To investigate how representative
City. The home addresses of survey
sity and walking among white and the NYCP is of New York as a whole,
participants were geographically linked
Black women and white men. An demographic characteristics were
to census tracts and contextual mea-
inverse association was observed be- compared to the 2000 Census.17 Be-
sures were constructed to examine the
tween intersection density and BMI in cause there are selection processes that
association between the individual and
white men. Increasing mixed land use, influence people to take part in health
urban form. The study used multilevel
that is, a greater variety of land uses surveys, comparisons were also made to
analysis to relate common indicators of
within a neighborhood, was also sig- the 2002 New York Community Health
the built environment, such as popu-
nificantly and inversely related to Survey (NYCHS), a random-digit-dial
lation density, land use mix, and access
obesity and BMI. Additionally, more health survey of New Yorkers con-
to public transit, to BMI, while con-
time spent in a car and less time ducted by the New York City Depart-
trolling for individual- and neighbor-
walking were both associated with ment of Health.18
hood-level sociodemographic charac-
obesity.
teristics.
Despite such evidence, many ques- Measures: Individual Level
tions remain about the built environ- Sample Questionnaire data on socio-demo-
ment and its relationship to both New York City, through the Aca- graphic variables and home address
physical activity and obesity. The work demic Medicine Development Compa- were gathered and height and weight
of Cervero and colleagues illustrates ny (AMDeC), set out to establish were objectively measured using clini-
the need for closer examination of a prospective cohort study of residents cal scales and rigid stadiometers avail-
cross-neighborhood variation within of New York City and the surrounding able at the medical centers and hospi-
densely populated cities.12 Most na- suburbs.16 The analyses presented here tals. Demographic data such as age,
tional studies employ measures of the are of data collected during the base- race/ethnicity, gender, pretax income,
built environment at the county- or line enrollment of subjects into the educational attainment, and address of
metropolitan-level, which fail to cap- cohort, which is referred to as the New residence as well as height and weight
ture the variability within cities. Fur- York Cancer Project (NYCP). A conve- measures were available. The questions
ther, most studies that have focused on nience sample of 18,187 volunteers was regarding ethnicity provided two cate-
individual cities are sited in lower- recruited between January 2000 and gories for Blacks: African-American
density places such as Austin, Texas; December 2002. Recruitment took and West Indian/Caribbean. For sta-
Atlanta, Georgia; and Portland, Ore- place across the five boroughs of New tistical analyses, six categories for racial
gon.7,13–15 It is unknown whether vari- York City with the goal of recruiting an background/ethnicity were created
ation in built environment character- ethnically and socioeconomically di- based on study subjects’ response to
istics is associated with BMI in densely verse population reflective of the city questions: Asian American; Black—
March/April 2007, Vol. 21, No. 4 Supplement 327
Health Promotion hepr-21-00-07.3d 12/1/07 10:19:22 327 Cust # 06050861RAfrican American; Black—Caribbean Land Use Mix. Because we were specif- accounting for nonindependence of
American; Caucasian; Hispanic; and ically interested in the balance of observations within tracts.19 Statistical
Other race. Response categories for commercial and residential uses, we analyses of the cross-sectional baseline
income were: less than $15,000, created a measure indexing these two data were performed using SAS Proc
$15,000 to $29,000, $30,000 to $49,000, types of development. Using tax asses- Mixed.19 As this is not a probability
$50,000 to $99,000, and $100,000 or sor data, we obtained precise measures sample, our tests for statistical signifi-
more. Categories for educational at- of the building area in each tract cance indicate the magnitude and
tainment included eighth grade or devoted to commercial or residential precision of our results. After exclud-
less, some high school, high school uses. Using the sum of commercial and ing subjects with missing data for BMI
graduate, vocational school, some col- residential building area in each cen- or individual-level demographic char-
lege, college graduate, and graduate sus tract as the denominator, we acteristics and those with extreme out-
school. Reliability and validity analyses calculated the ratio of building area lying BMI data (BMI .70), 13,102
on the questionnaire were not devoted to commercial use and the subjects residing in 1989 census tracts
conducted by AMDeC. Dummy ratio of building area devoted to were included in the analyses.
variables indicating income and residential use for each tract. The two The initial analyses tested for asso-
education were entered into the statis- ratios were multiplied by one another ciations between BMI and individual-
tical models. Overweight was defined and then scaled by a factor of four so level demographic variables. In sepa-
as a BMI greater than or equal to 25 that the index ranged from zero to rate models, income and educational
and less than 30, and obese was one. In a perfectly mixed area—con- attainment were both negatively asso-
defined as a BMI greater than or equal taining equal areas of residential and ciated with BMI. However, in models
to 30. commercial space—this index is equal including both income and education,
to one. If either type of use dominates income was no longer associated with
Measures: Built Environment the index will tend towards zero.4,7 BMI whereas education remained pre-
Contextual measures were created dictive. Thus, all further models con-
within a geographic information sys- Access to Public Transit. Using data from trolled for education. Finally, land use
tem environment and assigned to the New York City Department of City mix, access to public transit, popula-
census tracts. The census tracts used in Planning, we measured access to public tion density, and intersection density
this project were obtained from a 2003 transit by assigning bus stops and were assessed in separate models as
database created by Geographic Data subway stations to the census tracts in predictors of BMI, controlling for in-
Technology, Inc. (Lebanon, New which they fell and then calculating dividual-level race/ethnicity, gender,
Hampshire) and distributed by Envi- their density per km2 in each tract. age, education, and census tract mea-
ronmental Systems Research Incorpo- sures of percentage in poverty, per-
rated (Redlands, California). Home Population Density. Population density
centage Black, and percentage His-
addresses of participants residing in for each census tract was calculated
panic. All multilevel models included
the five boroughs of New York City using Census 2000 population data.
a random intercept for each census
(Bronx, Brooklyn, Manhattan, Queens, The number of residents of each
tract so that the percentage of total
and Staten Island) were geocoded to census tract was divided by the land
variance in BMI explained by between-
longitude and latitude coordinates and area of the tract. Population density
tract variance could be calculated and
matched to census tracts in New York was expressed in residents per km2.
so that the percentage of between-tract
City. Of the 18,187 respondents in the Intersection Density. The number of variance in BMI explained by individ-
NYCP database, 14,147 subjects were street intersections per census tract was ual- and tract-level variables could be
geocoded to census tracts in New York calculated using the Department of estimated.19
City; the rest lived outside the city. New City Planning LION Single Line Street
York City has 2190 residential census RESULTS
Base Map files of New York City. A
tracts in which the 2000 Census re- street intersection that occurred where
corded the presence of residents who census tracts bordered each other was Descriptive statistics for the New
were 30 years of age or older. Of these assigned to each census tract having York City residents in the study sample
census tracts, 2008 were represented in a border at the intersection. Intersec- are shown in Table 1, as are similar
the NYCP data set. tion density is expressed as intersec- descriptive statistics for New York City
Because neighborhood socio-demo- tions per km2. residents 30 years or older derived
graphic characteristics may predict from the 2000 Census and the 2002
BMI and also may be associated with Analysis NYCHS conducted by the New York
built environment characteristics, the The data had a clustered structure Department of Health.17,18 Compared
analysis included census tract-level with individuals grouped within census to Census 2000 data for New York City
poverty rates and racial and ethnic tracts; any given census tract charac- residents aged 30 and older, the NYCP
composition (percentage Black and teristic had the same value for all sample is slightly younger and includes
percentage Hispanic) from 2000 Cen- subjects in that tract. Multilevel analysis a higher prevalence of women, Cauca-
sus data. Data sources and variable was employed, allowing for the simul- sians, and individuals with higher levels
construction for our built environment taneous estimation of the effects of of educational attainment. Compared
measures are described below. group-level and individual-level factors, to the NYCHS, the NYCP sample has
328 American Journal of Health Promotion
Health Promotion hepr-21-00-07.3d 12/1/07 10:19:23 328 Cust # 06050861RTable 1
Characteristics of the Study Population and New York City Overall, As Depicted in the 2000 Census and the 2002 New York
City Community Health Survey*
Study Census Respondents to the New
Demographic Variables Population Data` Demographic Variables York City Community Health Survey§
Age Age
30–39 31 29 30–39 23
40–49 32 25 40–49 19
50–60 25 19 50–60 14
60+ 12 27 60+ 23
Gender Gender
Men 36 45 Men 41
Women 64 55 Women 59
Race/ethnicity Race/ethnicity
Asian 12 10 Asian 5
Black—African American 14 25 Black 24
Black—Caribbean 5 NA Black—Caribbean NA
Caucasian 47 27 Caucasian 44
Hispanic 20 23 Hispanic 24
Other 2 16 Other 3
Education Education
Eighth grade or less 6 13 Less than high school 18
Some high school 7 16 Some high school/ high school degree 26
High school graduate 22 25
Vocational school 2 NA Vocational school NA
Some college 21 20 Some college 20
College graduate 24 14 College or graduate degree 36
Graduate school 18 12
Body size Body size
Underweight (BMI below 18.5) 1 NA Underweight (BMI below 18.5) NA
Normal weight (BMI 18.5–24.9) 34 NA Normal weight (BMI 18.5–24.9) 53
Overweight (BMI 25.0–29.9) 37 NA Overweight (BMI 25.0–29.9) 33
Obese (BMI 30.0 and above) 28 NA Obese (BMI 30.0 and above) 20
* Values are percentages of the total population for each data source.
Sample size of 13,102 study subjects.
` Census data restricted to those 30 years of age or older (N 5 4,612,166).17
§ A random-digit-dial survey of New York City residents aged 18 years or older; analyses of the public use data set were restricted to those
respondents 30 years of age or older (N 5 7410).18
a similar distribution of gender, race, the NYCP sample is geographically .55). The four built environment vari-
and education, but is somewhat youn- representative. ables had relatively modest correla-
ger, and there is a higher prevalence of Table 2 reports means, standard tions with each other, between .03 and
overweight and obesity.18 However, deviations, ranges, and correlation .31.
because prevalence of overweight and coefficients of the census tract–level The built environment measures
obesity in the NYCHS is based on self- variables for the tracts represented in were added to a baseline model that
reported height and weight, there is the study. These statistics indicated included individual-level predictors
likely to be underestimation of over- considerable variation in neighbor- and tract-level demographic descrip-
weight and obesity. The average BMI hood characteristics. Some areas con- tors. Models 1 through 4 add measures
for the study subjects in the NYCP was formed to the popular stereotype of of land use mix, access to public
27.73 with a standard deviation of 5.78. Manhattan, with extremely high popu- transit, population density, and inter-
The percentage of New York city lation density, mixed land use, and section density, examining the effects
residents living in each census tract, as ready availability of bus and subway of each individually. Results of these
measured by Census data, and the lines. Others had a lower density and multilevel models are presented in
percentage of the study sample with much more limited access to transit. As Table 3, which also shows the pre-
complete data living in each census expected, census-tract poverty rates dicted difference in BMI associated
tract were strongly associated (beta 5 were correlated with percentage Black with a 90th to 10th percentile differ-
.96, R 5 .61, p , .001), indicating that (r 5 .30) and percentage Hispanic (r 5 ence in the predictor variable. In an
March/April 2007, Vol. 21, No. 4 Supplement 329
Health Promotion hepr-21-00-07.3d 12/1/07 10:19:23 329 Cust # 06050861RTable 2
Correlations and Mean, Median, Range Values for Census Tract Level Variables for 1989 Census Tracts, New York City
Descriptive Statistics of
Correlation Coefficient Census Tracts
Census Tract % % % Land Bus stop Subway Population Intersection
Characteristic` Black Hispanic Poverty use mix density stop density density Density Mean Median Range
% Black 1 27.53% 9.48% 0.00–100%
% Hispanic 20.10*** 1 24.51% 15.67% 0.00–96.10%
% Poverty 0.30 *** 0.55*** 1 20.00% 16.93% 0.00–100%
Land use mix 20.12*** 0.17*** 0.10*** 1 34.54% 27.52% 0.00–100%
Bus stops/km2 0.01 0.15*** 0.17*** 0.21*** 1 19.82 17.31 0.00–131.87
Subway stops/km2 20.01 0.10*** 0.12*** 0.25*** 0.14*** 1 1.19 0 0.00–37.15
Population density 20.03 0.32*** 0.33** 20.03 0.25*** 0.15*** 1 1.95 1.62 0.00–8.87
(10,000 people/km2)
Intersections/km2 20.06* 0.17*** 0.09*** 0.15*** 0.31*** 0.09*** 0.16*** 1 121.07 116.15 5.91–408.33
* p , 0.05.
** p , 0.01.
*** p , 0.001.
Correlations by Pearson’s product moment except correlations with subway stop density, which are presented as Spearman’s rho.
` Total observations 1989.
Table 3
Adjusted Mean Differences in BMI Associated With One-Unit Differences in Census Tract Built Environment Variables,
New York City, 2000–2002
Model 1 Model 2 Model 3 Model 4 Model 5
Land use mix 20.55** 20.46*
(20.96, 20.14) (20.88, 20.04)
20.41 20.34
Bus stops/km2 20.01** 20.002
(20.02, 20.003) (20.01, 0.01)
20.33 20.07
Subway stops/km2 20.06** 20.04*
(20.10, 20.02) (20.08, 20.004)
20.34 20.23
Population density (10,000 20.25*** 20.24***
people/km2) (20.32, 20.18) (20.31, 20.17)
20.86 20.82
Intersections/km2 20.002 20.0001
(20.005, 0.0002) (20.003, 0.003)
20.19 20.01
Percentage of between–census 77 81 87 77 87
tract variation explained`
Data are expressed as beta, (95% confidence interval), and D BMI. Beta is the mean difference in BMI for a one-unit change in the predictor variable
adjusting for individual level age, race/ethnicity, gender, interactions between gender and race/ethnicity and categories of education, and D BMI is the
predicted difference in BMI for a 10th to 90th percentile change in the independent variable. BMI indicates body mass index.
` Percentage of variation in census tract mean BMI explained by the individual and census tract level variables. Calculated as (between census tract
variance from the null model—between census tract variance from the full model)/ between census tract variance from the null model.19
* p , 0.05.
** p , 0.01.
*** p , 0.001.
330 American Journal of Health Promotion
Health Promotion hepr-21-00-07.3d 12/1/07 10:19:23 330 Cust # 06050861Runconditional means model the be- for individual and neighborhood so- density on BMI holds even well above
tween–census tract variance was statis- cio-demographic characteristics. Al- the threshold suggested to cause major
tically significant (p , .001), indicating though the variation in BMI across changes in transportation behavior.
that census tracts differ significantly in census tracts represents only a modest Population density remained signifi-
average BMI, although census tract– portion of the total variation in BMI, cantly associated with BMI after control
level differences accounted for only individuals living in tracts with higher for land use mix and access to public
a modest 6.0% of the total variance. population density, greater density of transit, suggesting an effect on BMI
Table 3 also shows the percentage of subway and bus stops, and a more even that is not explained by the greater
the between–census tract variation in mix of residential and commercial land use mix and public transit options
mean BMI explained by the individual- land uses— in short, in those tracts promoted by increased population
and census tract–level predictor vari- that were more pedestrian-friendly— density. It is possible that a higher
ables. To provide context, the variables had significantly lower BMI compared population density also supports in-
for age, gender, race, education, and with other New Yorkers. From a public creased recreational opportunities and
the demographics of the census tract health perspective, if each study sub- food outlets offering a better supply of
(percentage black, percentage Latino, ject’s BMI were reduced by a half nutritious foods, elements of the built
and percentage in poverty) together unit—a fraction consistent with the environment not assessed in these
explain 77% of the between–census magnitude of associations reported in analyses that may explain associations
tract variance. Table 3—an appreciable shift in the between BMI and population density.
When the built environment mea- BMI distribution would occur. Such Alternately, as discussed below, it is
sures were introduced separately, land a shift would move 10% of the over- possible that the census tract is not the
use mix, public transit density, and weight subjects in our sample into the appropriate scale at which to best
population density had a statistically normal BMI range, and 10% of obese understand the interrelations between
significant inverse association with subjects would shift into the over- built environment characteristics and
BMI, controlling for the individual- weight category. These findings are their associations with BMI. Increased
level and census tract socio-demo- consistent with current literature re- intersection density is thought to pro-
graphic characteristics. Model 1 shows lating the built environment to travel mote walking because it provides more
that residents of tracts that were more mode, physical activity, and obesity; route options and may calm traffic. In
evenly balanced between commercial their distinctive contribution lies in our analyses, intersection density did
and residential uses (i.e., had higher showing that these relationships hold not predict BMI, perhaps because this
values on the land use mix variable) within a high-density urban environ- single measure alone insufficiently de-
had significantly lower BMI than those ment. That is, the dose responses seen scribes street design. In other respects,
in areas that were predominantly resi- elsewhere with population density and however, the data presented here are
dential or predominantly commercial. land use mix persist even in areas that consistent with the hypothesis that
In model 2, increasing density of both are very densely settled. characteristics of the built environ-
subway and bus stops was negatively Pedestrian behavior and activities of ment impact body size through their
associated with BMI. Results from daily living are believed to be influ- connection with physical activities of
model 3 show that individuals living in enced by variation in characteristics of daily living. Other features of the city
areas with higher levels of population the built environment. Specifically, landscape, for example recreation
density had significantly lower BMI. mixed commercial-residential land use centers, health clubs, and parks, may
Model 4 shows that the density of that places goods and services near impact body size via their effect on
intersections within a census tract was residences and the availability of public planned physical or recreational activ-
not significantly associated with BMI. transit are thought to promote walking ities; our future research will examine
In model 5 the association between and independence from private auto- the influence of these urban features
BMI and all five built environment mobiles. Accumulating research sug- as well.27,28
variables was assessed simultaneously. gests that increased time spent driving The correlations we observed among
In this model land use mix, subway is associated with obesity, whereas in- the four built environment measures
density, and population density were creased walking and active commuting were reflected in weaker associations
inversely associated with BMI and are inversely associated with BMI and with BMI in the model including all
remained statistically significant. Nei- obesity.7,20–26 Mixed land use and pub- four measures. The nature of the
ther bus stop density nor intersection lic transit are thought to be positively relationships among these variables is
density were significantly associated influenced by population density, with not clear; they could be described as
with BMI in this full model. a density of .3500 persons per square mutual confounders, intermediate
mile posited as the threshold at which variables, or both. For example, public
DISCUSSION residents begin to use nonmotorized transit may promote mixed land use;
means of transportation.9 Among the alternatively, new commercial destina-
This study found that measures of 1989 census tracts represented in the tions may invite expansion of public
urban form similar to those commonly data set 98% had a population density transit; or the relationship could be
investigated in the literature had a sig- that exceeded this threshold. The reciprocal. When a variable acts as
nificant association with BMI among analyses presented here suggest that a mediator or as both an intermediate
residents of New York City, adjusting the effect of increasing population variable and a confounder, entering
March/April 2007, Vol. 21, No. 4 Supplement 331
Health Promotion hepr-21-00-07.3d 12/1/07 10:19:24 331 Cust # 06050861Rthe variable into a statistical model will characteristics and BMI, signifying that for physical activity. Urban planners not control for potential confounding socioeconomic status acted as a positive assume that pedestrians will readily effects, and may obscure associations confounder. Further control for in- walk one quarter mile (.4 km). Thus, between more distal predictors and the come had little to no effect on the from any point within a median-sized outcome.29,30 Thus, the appropriate- results. However, controlling for these census tract of .18 km2, locations in ness of including all four built envi- indicators of socioeconomic status adjacent tracts are close enough to be ronment measures in the model in cannot rule out the possibility of within walking distance. Likewise, order to provide unbiased estimates of positive residual confounding by other commercial locations are often con- the associations with BMI is in ques- aspects of socioeconomic status. centrated on major avenues; when tion. In response to this problem, It is recognized that the data are census tracts are small, they may not which has been observed elsewhere in derived from a convenience sample of adequately reflect the mixture of land the literature, researchers have em- volunteers and that the idiosyncrasies uses within walking distance of a resi- ployed strategies including principal of volunteer populations may bias dence. Future research will develop component analysis to define ‘‘walk- results. However, to generate bias, alternative contextual measures based ability’’ indexes that merge various selection forces would have to be on larger-size radial buffers or gravity- aspects of the built environment into related to both body size and neigh- type measures. a single scale.3,4,7,12 We expect to borhood characteristics. The analyses We also plan to further examine our pursue this strategy in future research. of population distribution by census measure of mixed land use. As in some A strength of this study is the tract in the 2000 Census and the study previous research, the measure used objective measurement of height and population show that the study subjects here indexes the balance between weight, which allows a more valid are geographically representative of commercial and residential land measure of BMI. Prior descriptive New York City. Table 1 also shows that uses.4,7 As such, tracts that are less studies of obesity in New York City have the socio-demographic characteristics mixed are treated as equivalent relied primarily on self-report31; such of the study subjects are similar to whether they are highly residential or data may understate true BMI, espe- those of the city overall, as measured by highly commercial. Increased mixing cially among individuals who are the 2000 Census.17,18 Our study popu- of commercial and residential land heavier and older.32–34 The lower lation has a higher prevalence of uses is thought to increase active prevalence of obesity indicated by New Caucasians and women than the 2000 transportation because it places goods York City Department of Health data Census, but this is typical of health- and services within walking or bicycling may be because of such underreport- related surveys.38 The prevalence of distance. However, areas that are dis- ing.31 Importantly, underreporting Caucasian respondents in the 2002 tinct in their predominant land use could also result in bias toward the null NYCHS is 47% and the prevalence of may also differ in dietary resources, for analyses of predictors of BMI. women is 60%, and this is consistent in opportunities for exercise, and activi- Additional strengths of this study are subsequent annual surveys.18 Table 1 ties of daily living. Future work will the large sample size, the ethnic di- shows that the NYCP population is very develop neighborhood typologies that versity of the study population, and the similar to the NYCHS respondents, distinguish between predominantly focus on an understudied urban envi- who took part in a random-digit-dial– residential and predominantly com- ronment. It is one of few studies to link based health survey designed to in- mercial areas. the built environment with either clude a representative sample of New Finally, the current study is observa- physical activity or obesity in a large, York City. In fact, despite its conve- tional and based on cross-sectional densely-settled city, and to our knowl- nience sample, it appears that the data, which raises a number of caveats. edge it is the only study of its kind to be NYCP sample represents a reasonable The built environment variables depict sited in New York City. demographic and geographic cross- an individual’s census tract at the time Despite these strengths, the results section of New York City residents. of enrollment, regardless of length of should be interpreted cautiously. An additional concern, particularly residence at that address. A high BMI, There is an inverse association between in the more densely-settled sections of however, may reflect years of positive socioeconomic status and body size in New York City, is that census tracts energy balance; we lack information on developed countries, especially among represent relatively small geographic the neighborhoods in which our study women.35,36 Here we controlled for the areas. The median size of the tracts in subjects may have lived over those potential confounding effects of so- our study area is .18 km2, with the 10th years. Furthermore, we lack data about cioeconomic status using income and percentile being .13 km2 and the 90th the built environment surrounding educational level ascertained by ques- percentile being .60 km2. Past work on individuals’ places of work and other tionnaire, in conjunction with census health disparities has shown that asso- key destinations within the city. An- tract-level poverty rate.37 Simultaneous ciations between area socioeconomic other caveat related to the cross-sec- analyses of variables for income and characteristics and health outcomes tional nature of the data is that it education as predictors of BMI yielded are stronger for smaller area units such cannot be discerned whether built a significant association only for edu- as census tracts.37,39,40 However, the use environment factors cause individuals cation. Controlling for education re- of very small area units may not be to maintain a lower body weight or duced the magnitude of the associa- appropriate and may result in misspe- whether health-conscious individuals tion between built environment cification of the area unit most relevant with lower body weights choose to live 332 American Journal of Health Promotion Health Promotion hepr-21-00-07.3d 12/1/07 10:19:24 332 Cust # 06050861R
in neighborhoods with particular built also promote pedestrian activity by 5. Saelens BE, Sallis JF, Black JB, Chen D.
environment characteristics. Built en- creating lively public spaces. Neighborhood-based differences in
physical activity: an environment scale
vironment characteristics may merely The results of our preliminary re- evaluation. Am J Public Health. 2003;93:
cause a sorting of certain types of search on the built environment in 1552–1558.
individuals into particular areas. Thus, New York City support an association 6. King WC, Belle SH, Brach JS, et al.
proposed public health interventions between variation in neighborhood Objective measures of neighborhood
targeting built environment factors environment and physical activity in older
characteristics and body size. Further women. Am J Prev Med. 2005;28:461–469.
may reshuffle people across neighbor- analyses of other neighborhood char- 7. Frank L, Andressen M, Schmid T. Obesity
hoods without impacting the overall acteristics, such as access to parks and relationships with community design,
prevalence of obesity. The last caveat is recreational facilities, crime rates, and physical activity, and time spent in cars.
that the study subjects vary consider- the distribution of grocery stores and Am J Prev Med. 2004;27:87–96.
ably in their age, and the etiology of 8. Vandegrift D, Yoked T. Obesity rates,
fast food restaurants are likely to result income, and suburban sprawl: an analysis
high BMI may vary across birth co- in additional recommendations for of US states. Health Place. 2004;10:221–229.
horts. In a purely cross-sectional study local policymakers. 9. Lopez R. Urban sprawl and risk for being
birth cohort effects may bias results overweight or obese. Am J Public Health.
and may not be fully controlled for by 2004;94:1574–1579.
SO WHAT? Implications for 10. Kelly-Schwartz A, Stockard J, Doyle S,
adjustment for age and socioeconomic Practitioners and Researchers Schlossberg M. Is sprawl unhealthy?
factors.41 Like nearly all studies in this Variation in land use mix, access A multilevel analysis of the relationship of
literature, our analysis does not dem- to public transit, and population metropolitan sprawl to the health of
onstrate a causal relationship between density across census tracts are individuals. J Plan Educ Res. 2004;24:
environmental characteristics and in- 184–196.
associated with differences in BMI 11. Giles-Corti B, Macintyre S, Clarkson JP, et
dividual outcomes. However, it does after accounting for personal de- al. Environmental and lifestyle factors
provide a strong rationale for further mographic characteristics and the associated with overweight and obesity in
prospective or time series studies that demographic characteristics of the Perth, Australia. Am J Health Promot.
allow for stronger causal inference. census tract. These dimensions of 2003;18:93–102.
A causal demonstration that the 12. Cervero R, Duncan M. Walking, bicycling,
the built environment are thought
urban environment influences pat- and urban landscapes: evidence from the
to impact pedestrian travel activity San Francisco Bay Area. Am J Public Health.
terns of diet, physical activity, and and activities of daily living. The 2003;93:1478–1483.
body mass would provide policy strat- results suggest that transportation 13. Shriver K. Influence of environmental
egies for tackling the problem of policy, zoning, and other city plan- design on pedestrian travel behavior in
obesity and associated health issues. ning policies may offer tools to four Austin neighborhoods. Transportation
Making automobile-dependent envir- Res Rec. 1997;1578:64–75.
promote physical activity and en- 14. Handy SL. Urban form and pedestrian
onments less obesigenic is a daunting courage maintenance of a healthy choices: study of Austin neighborhoods.
and expensive task, but may be cost- body size. Transportation Res Rec. 1996;1552:135–144.
effective in environments such as New 15. Greenwald M, Boarnet M. Built
York City that already have a public environment as determinant of walking
transit infrastructure. Feasible near- Acknowledgments behavior: analyzing nonwork pedestrian
travel in Portland, Oregon. Transportation
term policies might aim to improve This work was supported by the Robert Wood Johnson Res Rec. 2002;1780:33–42.
connectivity of bus routes, enhance Health and Society Scholars Program at Columbia 16. Mitchell MK, Gregersen PK, Johnson S, et
University, a Career Development Award from the National
frequency of bus and subway service, Cancer Institute (K07CA092348-04), and an award from
al. The New York Cancer Project:
and reduce fares. Additionally, propo- the National Institute of Environmental Health Sciences rationale, organization, design, and
sals to construct new subway lines, or (1R01ES014229-01). Dr. Rundle is a National Center on baseline characteristics. J Urban Health.
Minority Health and Health Disparities Health Disparities 2004;81:301–310.
to extend existing lines, may include Scholar. 17. Community Studies of New York Inc.
cost-benefit analyses of the impact on Infoshare online. Available at: www.
obesity. References infoshare.org. Accessed April 20th, 2006.
Cities might aim to promote mixed 1. Hedley AA, Ogden CL, Johnson CL, et al. 18. The New York City Department of Health
land use and increase population Prevalence of overweight and obesity and Mental Hygiene. Community Health
among US children, adolescents, and Survey Web page. Available at:
density through the renovation of www.nyc.gov/html/doh/html/survey/
adults, 1999–2002. JAMA. 2004;291:
vacant or underused land and the 2847–2850. survey.shtml. Accessed April 20th, 2006.
redevelopment of brownfields. En- 2. Sallis JF, Kraft K, Linton LS. How the 19. Singer J. Using SAS PROC MIXED to fit
couraging mixed land use in estab- environment shapes physical activity: multilevel models, hierarchical models,
lished neighborhoods may be accom- a transdisciplinary research agenda and individual growth models. J Educ Behav
[comment]. Am J Prev Med. 2002;22:208. Stat. 1998;24:323–355.
plished by relaxing zoning regulations 20. Stoohs RA, Bingham LA, Itoi A, et al. Sleep
3. Ewing R, Schmid T, Killingsworth R, et al.
to allow for commercial development Relationship between urban sprawl and and sleep-disordered breathing in
and introduction of recreational facil- physical activity, obesity, and morbidity. commercial long-haul truck drivers. Chest.
ities. Zoning may also be useful in Am J Health Promot. 2003;18:47–57. May 1995;107:1275–1282.
avoiding giant superblocks that create 4. Frank L, Schmid T, Sallis J, et al. Linking 21. Bell AC, Ge K, Popkin BM. The road to
objectively measured physical activity with obesity or the path to prevention:
dead space and hinder pedestrian objectively measured urban form: findings motorized transportation and obesity in
activity. The use of civic institutions as from SMARTRAQ. Am J Prev Med. China. Obes Res. Apr 2002;10(4):
focal centers for neighborhoods may 2005;28(2 Suppl 2):117–125. 277–283.
March/April 2007, Vol. 21, No. 4 Supplement 333
Health Promotion hepr-21-00-07.3d 12/1/07 10:19:24 333 Cust # 06050861R22. Gordon-Larsen P, Nelson MC, Beam K. public playgrounds. Environ Plan. 37. Krieger N, Chen JT, Waterman PD, et al.
Associations among active transportation, 1998;30:595–613. Choosing area based socioeconomic
physical activity, and weight status in young 29. Kleinbaum D, Kupper L, Muller K. Applied measures to monitor social inequalities in
adults. Obes Res. May 2005;13:868–875. Regression Analysis and Other Multivariate low birth weight and childhood lead
23. Wen LM, Orr N, Millett C, Rissel C. Driving Methods. Boston, Mass: PWS-Kent poisoning: the Public Health Disparities
to work and overweight and obesity: Publishing Company; 1988. Geocoding Project (US). J Epidemiol
findings from the 2003 New South Wales 30. Robins J. The control of confounding by Community Health. 2003;57:186–199.
Health Survey, Australia. Int J Obes (Lond). intermediate variables. Stat Med. 1989;8: 38. Centers for Disease Control. BRFSS: 2004
May 2006;30:782–786. 679–701. summary data quality report. Available at:
24. Lopez-Zetina J, Lee H, Friis R. The link 31. Thorpe L, Mackenzie K, Perl S, et al. One http://www.cdc.gov/brfss/
between obesity and the built in 6 New York adults is obese. NYC Vital technical_infodata/
environment. Evidence from an ecological Signs. 2003;2(7):1–4. 2004QualityReport.htm. Accessed August
analysis of obesity and vehicle miles of 32. Kuczmarski M, Kuczmarski R, Najjar M. 17, 2006.
travel in California. Health Place. Dec Effects of age on validity of self-reported 39. Krieger N, Chen JT, Waterman PD, et al.
2006;12:656–664. height, weight, and body mass index: Geocoding and monitoring of US
25. Krumm EM, Dessieux OL, Andrews P, findings from the Third National Health socioeconomic inequalities in mortality
Thompson DL. The relationship between and cancer incidence: does the choice of
and Nutrition Examination Survey, 1988–
area-based measure and geographic level
daily steps and body composition in 1994. J Am Diet Assoc. 2001;101:28–34.
matter? : the Public Health Disparities
postmenopausal women. J Womens Health 33. Bostrom G, Diderichsen F. Socioeconomic
Geocoding Project. Am J Epidemiol.
(Larchmt). 2006;15:202–210. differentials in misclassification of height, 2002;156:471–482.
26. Barengo NC, Kastarinen M, Lakka T, et al. weight and body mass index based on 40. Krieger N, Waterman P, Chen J, et al.
Different forms of physical activity and questionnaire data. Int J Epidemiol. Monitoring socioeconomic inequalities in
cardiovascular risk factors among 24–64- 1997;26:860–866. sexually transmitted infections,
year-old men and women in Finland. 34. Spencer EA, Appleby PN, Davey GK, Key tuberculosis, and violence: geocoding and
Eur J Cardiovasc Prev Rehabil. Feb 2006; TJ. Validity of self-reported height and choice of area-based socioeconomic
13:51–59. weight in 4808 EPIC-Oxford participants. measures—the Public Health Disparities
27. Giles-Corti B, Donovan RJ. Socioeconomic Public Health Nutr. 2002;5:561–565. Geocoding Project (US). Public Health Rep.
status differences in recreational physical 35. Sobal J, Stunkard A. socioeconomic status 2003;118:240–260.
activity levels and real and perceived access and obesity: a review of the literature. 41. Jacobs DR Jr, Hannan PJ, et al.
to a supportive physical environment. Prev Psychol Bull. 1989;105:260–275. Interpreting age, period and cohort effects
Med. 2002;35:601–611. 36. Ball K, Crawford D. Socioeconomic status in plasma lipids and serum insulin using
28. Talen E, Anselin L. Assessing spatial equity: and weight change in adults: a review. Soc repeated measures regression analysis: the
an evaluation of measures of accessibility of Sci Med. 2005;60:1987–2010. CARDIA Study. Stat Med. 1999;18:655–679.
334 American Journal of Health Promotion
Health Promotion hepr-21-00-07.3d 12/1/07 10:19:25 334 Cust # 06050861RYou can also read
