|Home | About | Journals | Submit | Contact Us | Français|
To effectively prevent and reduce childhood obesity through healthy community design, it is essential to understand which neighborhood environment features influence weight gain in various age groups. However, most neighborhood environment research is cross-sectional, focuses on adults, and is often carried out in small, nongeneralizable geographic areas. Thus, there is a great need for longitudinal neighborhood environment research in diverse populations across the lifecycle. This paper describes: (1) insights and challenges of longitudinal neighborhood environment research and (2) advancements and remaining gaps in measurement and study design that examine individuals and neighborhoods within the context of the broader community. Literature-based research and findings from the “Obesity and Neighborhood Environment Database” (ONEdata), a unique longitudinal GIS that is spatially and temporally linked to data in the National Longitudinal Study of Adolescent Health (N=20,745), provide examples of current limitations in this area of research. Findings suggest a need for longitudinal methodologic advancements to better control for dynamic sources of bias, investigate and capture appropriate temporal frameworks, and address complex residential location processes within families. Development of improved neighborhood environment measures that capture relevant geographic areas within complex communities and investigation of differences across urbanicity and sociodemographic composition are needed. Further longitudinal research is needed to identify, refine, and evaluate national and local policies to most effectively reduce childhood obesity.
Numerous local, state, and national obesity prevention initiatives target neighborhood diet and activity environments.1–3 However, for these strategies to effectively reduce and prevent obesity, better understanding of which environment features influence weight gain throughout childhood and into adulthood is needed.4–7 Such understanding necessitates longitudinal neighborhood-level obesity research,8–10 which has only recently emerged in the literature.11–17
The current paper describes: (1) insights and challenges of longitudinal neighborhood environment research and (2) advancements and remaining gaps in measures and study design that examine individuals and neighborhoods within the context of the broader community. Areas of focus are residential selection and mobility, measurement approaches, and subgroup-specific effects. Each issue is illustrated with examples from (1) the longitudinal “Obesity and Neighborhood Environment Database” (ONEdata), which contributes unique longitudinal GIS data tied to individual-level, nationally representative data spanning the adolescent to young adult years; (2) an analogous longitudinal cohort of adults, the Coronary Artery Risk in Young Adults study (CARDIA), as adults are key purchasers of food for children; and (3) key findings from the wider neighborhood environment literature, although there are few comparable studies.
Cross-sectional neighborhood environment studies are particularly problematic because neighborhoods and individual behaviors evolve over time through complex, inter-related processes (Figure 1).18 Individuals move in and out of neighborhoods according to financial or social constraints and residential preferences19 (Figure 1, Pathways A,B). Individuals with physically active lifestyles and healthy diets may prefer and afford neighborhoods that support healthy behaviors20–23 (Pathways C–E). Likewise, physical activity and food resources are placed in areas with the greatest demand, characterized in part by the characteristics of nearby residents (Pathways F,G).24 The largely cross-sectional literature ignores these dynamic interactions and may grossly mis-estimate influences of neighborhood features on obesity-related behaviors.25 With longitudinal data it is possible to address individual characteristics that may contribute to these temporal inter-relationships.20
To illustrate these limitations, findings are highlighted from Add Health, a nationally representative, prospective cohort study of adolescents representative of the U.S. school-based population in Grades 7 to 12 in 1994–1995 (Wave I) followed through 1995–1996 (Wave II, n=14,738) and into adulthood in 2001–2002 (Wave III, n=15,197) and 2007–2008 (Wave IV, n=15,701). As described elsewhere,26 Add Health included a core sample plus subsamples of selected minority and other groupings collected under protocols approved by the IRB at the University of North Carolina at Chapel Hill. ONEdata includes >6000 time-varying built, economic, and social environment variables (Appendix A) from external sources linked to respondent residential locations at Waves I and III.27 Environment measures capture areas within 1-, 3-, 5-, and 8-km straight-line (Euclidean neighborhood buffers) and street network (street network neighborhood buffer) distances around residences and within census block groups and tracts.28 ONEdata captures 3.6% of 2000 census block groups (n=7558).
Longitudinal ONEdata findings suggest that reducing neighborhood crime and, for men/boys, providing for-pay physical activity facilities may promote physical activity as adolescents become young adults (Figure 2),17 while limiting neighborhood fast-food availability may reduce fast-food consumption in low-income men (Figure 3).29 Fewer neighborhood features were associated with obesity-related behaviors in girls/women, suggesting that promotion of healthy lifestyles in girls and women may be particularly difficult. Overall, findings suggest that improvements in landscape diversity, public physical activity facilities, and street connectivity may substantially increase physical activity (Figure 2),17 and providing supermarkets and grocery stores may not result in improved diets (Figure 3).29 These findings are consistent with other longitudinal studies showing inconsistent15 or null associations between urban sprawl with walking or obesity,11,12 and mixed, generally null associations between food resources and obesity-related outcomes.13,14,30
These examples use fixed-effect longitudinal models, which condition on each individual, thereby analyzing variation observed within-person, over time, and adjusting for time-constant unmeasured characteristics;9,33,34 in essence, each individual serves as his/her own control. Adjusting for these unmeasured characteristics is critical because they may be powerful drivers of location selection. Yet they are difficult to measure. For example, unmeasured outcome expectations of adults with school-aged children who are more likely to select a neighborhood with high-quality schools (and coincidentally more recreation resources) may influence adoption of physically active lifestyles and healthy diets (Figure 1, Pathway E).
Fixed-effects models control for such outcome expectations that remain constant between time points, although they do not address unmeasured characteristics that change over time (e.g., time constraints). Formal testing of longitudinal fixed-effects models compared to random-effects models (which do not control for time-constant unmeasured confounders and are thus more comparable to cross-sectional models) indicated that fixed effects were warranted (Hausman–Taylor test31) for a wide range of associations between neighborhood environments and diet16 and physical activity17,32 behaviors. Thus, unmeasured confounders were associated with the independent variables, and therefore random-effects estimates were biased.
Further, random-effects estimates were attenuated toward, to, or past the null.17 This finding contrasts with the typical assumption that location-selection bias results in overestimation of neighborhood health effects due to selection of more-favorable environments by people with healthier lifestyles. Specifically, for-pay physical activity facilities (in boys/men) and crime (in girls/women) were more strongly related to physical activity in fixed-versus random-effects models.17 Such facilities may be more common in commercial centers selected less often by more-advantaged, physically active families, which may explain why controlling for location-selection factors attenuates effects.
In the small body of longitudinal neighborhood environment research, most studies investigate health impacts of changes in neighborhood environments that result from relocation of individuals to new residential neighborhoods (residential mobility).11,12,15,33 In contrast, policies assume that health will improve as a result of changes in neighborhoods around stationary residents. In an examination of both mechanisms, associations between physical activity facilities and physical activity behaviors were generally weaker or equivalent in those who did (versus did not) relocate to new residences.17 However, this pattern could be reversed in adulthood, when residential stability is the norm.
Estimated impacts of neighborhood changes around stationary residents may reflect several processes. In the late teenage years, individuals may remain in their parental homes to care for their own young children, attend a local college, or for other reasons that may also affect physical activity levels. Likewise, neighborhoods change systematically, with disadvantaged groups experiencing more neighborhood economic decline.19 Natural experiments before and after introductions of policies or facilities and other longitudinal studies of neighborhood change around stationary residents are clearly needed to disentangle systematic demographic and environmental changes from influences of specific neighborhood features on behavior.
Studies that examine direct relationships between neighborhood environments and BMI (Figure 1, Pathway I) ignore behavioral pathways (Figure 1, Pathways D and H). Using complex modeling techniques to examine inter-related behavioral pathways,34 greater availability of public physical activity facilities was related to lower BMI 6 years later but unrelated to concurrent physical activity and sedentary behaviors. Living farther from a neighborhood park was related to higher TV/video viewing and leisure computer use, but unrelated to BMI.34 Sophisticated, longitudinal analysis with high-quality behavior data are needed to understand complex pathways underlying relationships between neighborhood environments, individuallevel behaviors, and health outcomes.
The findings above controlled for unmeasured characteristics that were constant over time. Addressing key unmeasured predictors of location selection that vary over time such as change in marital or employment status35 requires better understanding of drivers of location selection. Simultaneous equation strategies20 can explicitly model predictors of location selection in a first modeling stage, and instrumental variables36 can also address time-varying unmeasured confounders. Natural experiments or randomized trials are critical for understanding causal effects of neighborhoods on health25,37 but can be costly, pose ethical dilemmas, and often are not feasible for studying large-scale or combined impacts. Therefore, advances in observational research combined with experimental designs are needed.
The above-described longitudinal models assume that effects of neighborhood features on physical activity and diet behaviors are relatively immediate (within the follow-up period). However, long-term and cumulative effects are possible. For example, neighborhood sports fields may promote youth sports participation, thereby developing skills and preferences for active lifestyles that carry through adolescence and into adulthood.38 Likewise, improvement of diet in response to a new supermarket may occur over months or years as residents develop motivation and skill in preparing fresh produce. Innovative analytic approaches for capturing cumulative effects and investigating lag times between neighborhood modifications and changes in behavior and health are needed to understand such long-term impacts.
Parents drive neighborhood selection and offspring behavior through modeling, supports, and rules,39–41 which may influence later (adult) behavior.42,43 Further, characteristics of previous residential locations are the most powerful predictors of subsequent residential neighborhood characteristics.19,44 Therefore, biases related to residential selection in youth may mimic parental residential selection. Innovative strategies for addressing residential selectivity in youth require greater understanding of parental influences on residential choice and behavior.
Common neighborhood resource availability measures such as raw counts of resources within a given area45–49 or distance to the nearest resource50–55 overlook the array of retail, industrial, educational, and residential facilities that cluster in predictable and inter-related ways. Indeed, pattern analysis of a large set of neighborhood environment variables28 suggested that population density, intersection density, and counts of physical activity facilities represent underlying constructs of development intensity. Thus, intersection density (a common indicator of street connectivity56–59) and physical activity facility counts (a common measure of recreation opportunities) may be proxies for general development intensity.
Using density-scaled resource counts is one strategy to separate availability of physical activity and diet resources from development density. For example, scaling by population (resource counts per 10,000 population)17,28 addresses the strong correlation between density of commercial establishments and population density and incorporates crowding as a facet of availability. Alternatively, roadway-scaled measures (resource counts per roadway mile)3,60,61 represent the concentration of resources along access routes and may adjust for overall commercial activity. These density-scaled measures should be validated and further developed to better isolate the impacts of physical activity and food resources.
Accurate neighborhood measures must capture resources and design features within a relevant area, yet there is little theoretic or empirical guidance for delineating neighborhoods. Creating GIS boundaries and variables is therefore subjective,62–65 although varying neighborhood definitions affect study findings in some studies66 but not others.58,67–69 In contrast to administrative units such as ZIP codes or U.S. census tracts, buffer-defined neighborhoods at specified Euclidean68,70 or street network57,71 distances around individual residential locations are specific to individual residents.
In comparative research across 1-, 3-, 5-, and 8.05-kilometer buffers, physical activity was most strongly related to greater intersection density within 1 km and to physical activity resources within 3 km of adolescents’ homes. While street-based activity such as skateboarding or jogging may occur close to home,43 families may be willing to travel longer distances to recreation facilities. Yet these associations vary by gender and income. For example, fast-food consumption was most strongly related to fast-food availability within 3km of homes in low-income men, who may be less likely to own a car, thereby limiting mobility and enhancing reliance on the immediate neighborhood area.72
In short, appropriate proximities may vary by the type of neighborhood feature and constraints of the target population. Empirical comparisons of varying neighborhood definitions and movement toward standard, objective definitions will facilitate greater measurement accuracy and comparability across studies.73 Studies incorporating diverse geographic scales will also inform multilevel policies aimed at local neighborhoods as well as counties and states.
Much neighborhood environment research has been in major metropolitan areas, yet research within a geographically diverse population suggests variation in results across the urban spectrum.60,74,75 In contrast with the typical dichotomous classification of rural versus urban based on population density, the multidimensional concept of urbanicity and rurality may be better captured by classifying according to U.S. Census–defined “urbanized areas,” augmented with percentage of developed land cover to provide nuanced approximations of non-urban (rural), low-density urban (suburban), and high-density urban (central urban) areas.74,76
Using this strategy, greater physical activity was associated with higher intersection density and in non-urban and high-density urban areas,74 but with physical activity facilities only in low-density urban areas.74 There is likely great variation in neighborhood environment and behaviors across these different settings. For example, walkability might have less relevance in rural areas, where walking to destinations is rare, as well as in urban centers where retail destinations are ubiquitous but personal safety concerns interfere with access. Similarly, individual and household characteristics that drive selection of homes in rural, suburban, and urban areas are poorly understood but may underlie differences in association between environment and obesity-related behaviors.
Neighborhood environment research and policies often assume less access to physical activity and healthy food resources in poor neighborhoods, but evidence is inconsistent.3,10,77 For example, neighborhoods with both higher income and education had more intermixed land use, development density, and physical activity resources.28,78 However, “reverse” disparities have been observed in relation to the social environment: areas with high crime and high racial minority populations had greater intermixed land use, development density and resources,28 and lower availability of convenience stores, which typically provide energy-dense, nutrient-poor foods.75 In addition, racial and income disparities in availability of grocery/supermarkets were more apparent in low-density urban (suburban) areas than in high-density urban areas.79–84 Access to healthy neighborhood resources appears to be driven by complex economic and social influences that may vary across geographic contexts.
Improved characterization of neighborhood environments requires investigation of the following knowledge gaps (Table 1):
The literature also suggests needs related to creating policies that effectively promote healthy lifestyles and reduce obesity in youth (Table 1).
While limitations have been discussed here across the field of neighborhood research, longitudinal data have been used to provide many illustrative examples, largely due to the lack of comparable longitudinal data. These data may have error, a well recognized limitation in neighborhood environment research.99–105 In addition, while Add Health offers the advantages of a large, nationally representative study population, these data do not capture unique aspects of small localities throughout the U.S. Combined knowledge from large national studies with studies in focused geographic areas and subpopulations is needed to understand how neighborhoods influence obesity across diverse populations.
Investigation and resolution of methodologic issues raised by the small but growing body of longitudinal neighborhood research will become critical as future longitudinal research is developed and more longitudinal data become available. Advancements in characterization of neighborhood environments that capture relevant geographic areas and scales of influence, differences across urbanicity and neighborhood sociodemographics, and diverse physical and social resources are needed. Finally, further research is needed to test the robustness of existing evidence and refine policies to target neighborhood resources with the greatest benefit to population health.
The authors thank Brian Frizzelle, Marc Peterson, Chris Mankoff, James D. Stewart, Phil Bardsley, and Diane Kaczor of the University of North Carolina, Carolina Population Center (CPC) and the CPC Spatial Analysis Unit for creation of the environmental variables. The authors also thank Ms. Frances Dancy for her helpful administrative assistance.
This work was funded by NIH grants R01 HD057194 and R01 HD041375, R01 HL104580, a cooperative agreement with the CDC (CDC SIP No. 5-00), grants from the Robert Wood Johnson Foundation’s Active Living Research and CDC (R36-EH000380) and The Henry Dearman and Martha Stucker Dissertation Fellowship in the Royster Society of Fellows at the University of North Carolina at Chapel Hill, and the Interdisciplinary Obesity Training Program (T32MH075854-04). This research uses data from Add Health, a program project directed by Kathleen Mullan Harris PhD and designed by J. Richard Udry PhD, Peter S. Bearman PhD, and Kathleen Mullan Harris PhD at the University of North Carolina at Chapel Hill, and funded by grant P01-HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative funding from 23 other federal agencies and foundations. Special acknowledgment is due Ronald R. Rindfuss, PhD and Barbara Entwisle, PhD, both from the University of North Carolina at Chapel Hill, for assistance in the original design. Information on how to obtain the Add Health data files is available on the Add Health website (www.cpc.unc.edu/addhealth). No direct support was received from grant P01-HD31921 for this analysis. None of the acknowledged individuals received compensation for their assistance.
The more extensive restricted-use data, available by contractual agreement, are distributed only to certified researchers who have an IRB-approved security plan for handling and storing sensitive data and sign a data-use contract agreeing to protect data confidentiality.
The publication of this theme article was supported by a grant from the Robert Wood Johnson Foundation.
|Measure||Variable description||Data source|
|Street connectivity||Alpha index: ratio of observed to maximum|
possible route alternatives between nodes
(intersections); high values indicate high
|Intersection density||Number of three- or more-way intersections|
(≥links in a single node) per square kilometer
|Physical activity facilities||Categorized based on exploratory factor analysis|
|Dun & Bradstreet|
|For-pay||Instruction, member, public fee facilities|
|Free public||Public, youth organizations|
|Food resources||Dun & Bradstreet|
|Grocery stores||Independent and chain grocery stores and|
supermarkets (supermarkets and grocery stores
are separate in some analyses)
|Convenience stores||Variety and convenience stores and food stores|
attached to gasoline filling stations
|Fast-food restaurants||Fast-food chain and nonchain restaurants,|
excluding food stands and cafeterias (fast-food
chains are isolated in some analyses)
|ESRI StreetMap Pro,|
|Greenspace||Area of recreational or undeveloped land as|
proportion of total land cover excluding water
|National land cover data set|
|Landscape diversity||Simpson’s diversity index: Represents the|
probability that any two pixels selected at
random would be different patch types.
|National land cover data set|
|Population density||Count of persons per square mile||U.S. Census|
|Below poverty, %||People living in households with income below|
the federal poverty level (or below 150% of
federal poverty level)
|Minority, %||People with race/ethnicity other than white non-|
|Median household income||Median household income||U.S. Census|
|Crime rate||Number of nonviolent and violent crimes per|
|Uniform Crime Reporting data|
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
There were no potential or real conflicts of financial or personal interest with the financial sponsors of the scientific project.
No financial disclosures were reported by the authors of this paper.