Google Walkability - Semantic Scholar

Journal of Physical Activity and Health, 2012, 9, 689-697 © 2012 Human Kinetics, Inc.

Google Walkability: A New Tool for Local Planning and Public Health Research? Jason Vargo, Brian Stone, and Karen Glanz Background: We investigate the association of different composite walkability measures with individual walking behaviors to determine if multicomponent metrics of walkability are more useful for assessing the health impacts of the built environment than single component measures. Methods: We use a previously published composite walkability measure as well as a new measure that was designed to represent easier methods of combination and which includes 2 metrics obtained using Google data sources. Logistic regression was used to assess the relationship between walking behavior and walkability metrics. Results: Our results suggest that composite measures of walkability are more consistent predictors of walking behavior than single component measures. Furthermore, a walkability measure developed using free, publicly available data from Google was found to be nearly as effective in predicting walking outcomes as a walkability measure derived without such publicly and nationally available measures. Conclusions: Our findings demonstrate the effectiveness of free and locally relevant data for assessing walkable environments. This facilitates the use of locally derived and adaptive tools for evaluating the health impacts of the built environment. Keywords: active transportation, built environment, physical activity methodology Interest in the connections between land use and health has introduced new methods and tools for describing health-related aspects of the built environment.1 These methods often require a significant investment of time for training and data collection while yielding uncertain returns in terms of accuracy and strength of findings. For example, one instrument for assessing the importance of parks to physical activity incorporates more than 600 attributes.2 A second widely used metric for assessing built form and physical activity is a sprawl index developed by Ewing, et al, which requires for its computation 22 separate measures combined through principal components analyses.3 If built environment and physical activity frameworks are to be useful beyond the limited geographic and temporal extents for which they are often developed, these tools will need to be more readily accessible to a wide range of public health and planning researchers and practitioners. Efficient use of these tools requires that they be less methodologically complex and resource-intensive. Instead, if possible, they should be based on free and easily accessible data sources. This research assesses the relative utility of alternative metrics for measuring the built environment. We seek to determine whether the integration of 2 or more Vargo and Stone are with the School of City and Regional Planning, Georgia Institute of Technology, Atlanta, GA. Glanz is with the Schools of Medicine and Nursing, University of Pennsylvania, Philadelphia, PA.

built environment indicators into multicomponent or composite measures of neighborhood “walkability” yield greater predictive power for walking outcomes than single component metrics. Specifically, we test the predictive power of both composite and single component variables in explaining walking behavior in different neighborhoods of the Atlanta, Georgia metropolitan area. In addition, we evaluate the predictive power of built environment measures developed from open access data to assess the potential for such increasingly available datasets to be employed in planning and public health research and practice.

Background Walkability describes those qualities of the built environment that encourage walking behaviors. Composite measures vary by the components they include (eg, density, land use, connectivity), the scale at which they are measured (eg, 1/4 mile, 1/2 mile, or 1 mile from locations), and the methods used in computation (eg, combining component metrics via principal components analysis). While studies employing composite measures of the built environment have demonstrated associations with physical activity, assessments of the relative utility of such measures in predicting outcomes have not been addressed in a comprehensive fashion. This paper examines walkability measures using 2 main categories of component variables: those related to proximity and those related to connectivity of different land uses.4

689

690 Vargo, Stone, and Glanz

Proximity describes the number and variety of destinations within a specified distance of any location. Population, employment, and household densities measure the compactness of land uses, while land use mix measures the heterogeneity of different uses in space. Measures of connectivity are related to the physical design and the layout of transportation infrastructure. These measures quantify the network connections between trip ends to describe directness of possible paths and the number of options available. For example, the connectivity of a street network can be quantified by measuring the number of intersections per unit of area.5–9 Considering proximity and connectivity at the scale of the neighborhood is important for describing trips that can potentially be made by walking. Urban design and street amenities, such as sidewalks of a certain width or condition, may also encourage walking trips. Such data are not readily available and must often be created from in-situ observations. Kitamura, et al collected and recorded data on sidewalks and streets from visiting sites in the field.10 Instruments for measuring the walkability of single street segments have been evaluated on their strength in predicting physical activity outcomes.2,11,12 One such tool is the Systematic Pedestrian and Cycling Environmental Scan (SPACES). It codes attributes of sidewalks, intersections, and roads by street segment.13 Such tools are based on evidence of associations between the built environment and walking but go beyond enumeration of features and toward assessments of quality. Composite measures can be used to quantify the proximity, connectivity, and urban design dimensions of a street segment or neighborhood areal unit in a single measure. A number of different methodologies for combining component values in composite walkability measures have been studied. Factor analyses,14,15 principal components analysis,15,16 and structural equation modeling17 have been used to create indices and demonstrate multiple pathways between covariates and walking outcomes. A more straightforward approach has been the ranking of built environment data values by decile followed by linear combination.18,19 Several studies categorize locations into high walkability and low walkability settings based on composite values.20–22 Using composite measures offers benefits such as reducing spatial autocorrelation;23 however, the computation of such variables also involves additional time, data, and expertise for sophisticated assessment. As such, these measures may be difficult to interpret for decision-makers and the public, and could limit their utility for local planning or public health practitioners. To date, work with composite measures has been limited to demonstrating associations between walkability measures and walking outcomes. In this work we will use previously reported findings to create our own composite measure that includes freely available data from the US Census and the Internet search firm “Google” and compare its predictive strength with a similar composite measure from the technical literature.

Through this work we test the following hypotheses: 1. Composite (multicomponent) measures are superior to single component measures in predicting walking behaviors 2. Publicly available measures of the built environment, such as those derived from Google Maps and openaccess satellite data, are comparable in predictive power to proprietary measures published in the technical literature.

Methods SEQOL Study Design This work was completed as part of a U.S. Centers for Disease Control and Prevention (CDC) sponsored study titled the Study of Employee Quality of Life (SEQOL), which focused on the travel behaviors and physical activity of employees at Atlantic Station, a mixed use and transportation oriented development in the center of Atlanta’s Midtown business district.22 Participants were recruited from office, retail, and service-based employers within Atlantic Station. Travel and built environment measures were constructed for Atlantic Station, as well as each participant’s residential neighborhood. Survey data included home address, age, sex, race, marital status, education attainment, household income, and salary. In addition, participants were asked to record their travel activity for 4 consecutive days (2 weekdays and 2 weekend days) using a previously developed travel diary.24 Travel diary information was organized by trip and analyzed to count the number of home-based walking trips that participants made.

Built Environment Measures Characteristics of the built environment hypothesized to be associated with nonautomobile travel (both nonmotorized and transit) are described in Table 1. These included population, employment, and household densities, as well as the number of retail destinations, transit stops, and intersections per unit of area. In addition, the percentage of streets with 1 or more sidewalks was quantified. Each of these variables was calculated using a half-mile radius buffer around each residential and employment location address in ESRI’s ArcMap 9.3. The centers of each buffer were geo-located using the addresses provided by participants in the survey. Within each half-mile radius buffer, population and employment densities were calculated using areaweighted averages of census tract-level data on midcensus (2005) estimates. The destinations within each buffer were identified and counted using the publicly available inventory of ‘places of interest’ from Google’s mapping utility, “Google Earth” (example in Figure 1). The destinations variable quantifies the number of neighborhood-scale trip ends, including groceries, gas stations, pharmacies, restaurants, banks, coffee shops,

Table 1 SEQOL Built Environment (BE) Variable Definitions SEQOL BE variable

Operational definition

Units

Source

Conceptual definition

Proximity / connectivity Proximity

Population density

Area weighted average of Residents / gross population density acre calculated from census tracts intersecting buffer around each location.

Atlanta Regional Commission’s (Atlanta MPO) midcensus estimates

Concentration of residents in a place encourages other land uses to colocate and indicates the proximity of residents to each other. Increased density is also an indicator of shorter distances between destinations.

Employment density

Area weighted average of Employees / gross employment density acre calculated from census tracts intersecting buffer around each location.

Atlanta Regional Commission’s (Atlanta MPO) midcensus estimates

Proximity Concentration of employees serves as a proxy for jobs and indicates the intensity of the land use with regard to commercial activity. This relates to trip generation for both residents and other employees in the area.

Destinations

Number of destinations inside buffers around each location. Eight types of destinations were summed to obtain the total: restaurants, pharmacies, coffee shops, grocery stores, bars, gas stations, retail stores, and banks.

# of destinations

Google Earth’s ‘Places of Interest’ combined with buffers around each location created in AcrGIS

Destinations go beyond employee density by more directly measuring the types of commercial services involved in daily trips. As the number is assessed for identical buffers around each location, it is effectively a concentration.

Proximity

Intersections

Number of intersections inside buffers around each location. Intersections were defined as points where 3 or more street segments converged and were assessed without limited access freeways.

# of intersections

Intersections were determined using the MPO’s road network shapefile, limited access roads such as freeways were removed before analysis

Intersections measure the connectivity of the street network and thus the number of pathways possible for making walking trips and the directness of possible paths. As the number is assessed for identical buffers around each location, it is effectively a concentration.

Connectivity

Transit stops

Number of bus and rail transit stops inside buffers around each location.

# of stops

Metropolitan Atlanta Rapid Transit Association (MARTA) and Georgia Rapid Transit Authority (GRTA)

Transit connects people with destinations by facilitating trips that would otherwise be made by car. In addition, walking to and from transit stops can serve as important components of recommended physical activity. As the number is assessed for identical buffers around each location, it is effectively a concentration.

Connectivity

Sidewalks

Percentage of street length with one or more

%

Google Maps Satellite imagery combined with street network from MPO

Sidewalks are the infrastructure which facilitate walking trips. The presence of sidewalks on paths between origins and destinations can make walking a more attractive mode choice for trips.

Connectivity

691


and general retail establishments. As such, this variable is hypothesized to capture regular daily commercial activity likely to generate walking trips. It should be noted that parks were excluded from this measure because of the limited quality of the greenspace data available through Google. The study’s measures of connectivity were derived from a regional roads data layer provided by the Atlanta Regional Commission (ARC), Atlanta’s metropolitan planning organization. Limited access roads, such as freeways and their on-ramps, were removed from the regional roads file to include only those roads where pedestrian travel was possible. The refined regional roads

network was then used for calculating the number of intersections in each buffer zone, as well as the presence of sidewalks along 1 or 2 sides of the street. To determine which street segments were equipped with sidewalks, we again chose to use a free and openly available source of data: Google’s collection of satellite and surface imagery accessible through Google Maps (example in Figure 2). Street segments in the buffer were coded as having sidewalks on 1 side, both sides, or no sides of the street. We tested the predictive power of each single component measure and 2 composite measures, one adopted from an earlier study and a second measure created as part of the SEQOL study. The first composite measure is the

Figure 1 — An example destination calculation using Google Earth.

Figure 2 — An example sidewalk calculation using Google Street View.

Google Walkability 693

“neighborhood accessibility” (NA) measure developed by Krizek.16 Through a longitudinal study of neighborhoods in Seattle, Krizek created this composite measure based on 3 components: household density, number of retail employees, and block area. The measure used 2 spatial domains together to aggregate values of built environment variables. Grid cells 150m by 150m were used to calculate values of each variable that were then averaged using quarter-mile buffers around each location. The composite measure was found to be a statistically significant correlate of changes in vehicle miles traveled and person miles traveled.16 We derived the Krizek household density component for each grid cell in our study area using area-weighted averages of the 2005 ARC estimates of households by census tract. The number of retail employees in each grid cell was calculated using an inventory obtained from the Georgia Department of Labor for the year 2007. Retail employment was defined using the 2-digit North American Industry Classification System (NAICS) codes for retail trade (44-45). Block area per grid cell was measured through GIS using blocks as defined for the 2000 Census. The Krizek composite walkability index was calculated by linear combination of the values from the 3 Krizek components and adjusted to range from 0 to 100. Coefficients for the components in the linear combination were obtained by factor analysis performed using SPSS 15 statistical software. For comparison with the Krizek NA measure, we created a second composite measure that incorporated proximity and connectivity measures using data obtained from several publicly accessible sources. To do so, the 6 SEQOL built environment variables (population, employment, and intersection density, number of destinations and transit stops, and fraction of street length with sidewalks) were adjusted to cover a range from 0 to 100 and an average of the 6 was calculated for each location using Microsoft Excel.

Analysis Measures of covariation between indices and their components were used to assess colinearity between different components. For our analysis of the built environment measures with walking activity, a binary variable to identify walkers from nonwalkers was used as an outcome in logistic regression analyses. Individuals making 10% or more of all their home-based trips by walking were considered walkers. Odds ratios were used to assess the magnitude of the relationship between built environment variables and walking outcomes. Results were assessed using a significance level of 0.05 and statistical analyses were performed using SAS Version 9.2. Each significant association identified using bivariate analysis was tested in multivariate logistic modeling while controlling for sex, race, and income. The demographic covariates were introduced into the models to test the associations initially observed in the presence of attributes believed to influence walking behavior and to control for the possibility

that participants categorized as walkers choose to live in more walkable settings. Odds ratios are derived through logistic regression and are used to compare the odds of an outcome between different groups; in this instance, the tendency of individuals residing in pedestrian-supportive neighborhoods to be categorized as “walkers.” The odds ratio is commonly used as an estimate of relative risk between individuals or populations. The metrics and methods of logistic regression are commonly used with survey research focusing on health behavior outcomes, and they have been used with walking behaviors and built environment.25–27

Results A total of 59 employees at Atlantic Station were recruited for the study. Of these, 56 individuals completed both the initial and follow-up surveys. A description of the 56 participants’ demographic characteristics is included in Table 2. As reported there, our sample was slightly skewed toward females (59%), white/non-Hispanic (56%) and households with incomes above $40,000 (58%). Table 2 Description of Study of Employee Quality of Life (SEQOL) Participants and Walking Outcomes #

%

Sex Male Female

24 35

40.70% 59.30%

Age Under 50 Over 50

54 5

92% 8%

Race White/Non-Hispanic Black/Other/Multiple

33 26

55.90% 44.10%

Marital status Widowed/divorced/separated or single and never married Married or living with partner

38 21

64.40% 35.60%

Highest education High school/some college College/graduate school

22 37

37.30% 62.70%

Household income Under $40,000 Over $40,000

25 34

42.40% 57.60%

24 15

42.90% 26.80%

9

16.10%

10

17.90%

Demographics

Walking outcomes Walking trips* Made any walking trip Made a home walking trip Made both home and work walking trips Made 10% or more of home trips by walking * Only 56 participants completed travel diaries.


During the travel diary period, 15 participants (27%) made walking trips at residential locations. Of these participants, 10 made 10% or more of their home-based trips via walking and were classified as walkers. Bivariate odds ratios for the walkability composite measures and SEQOL components (Table 3) showed positive associations with the individual’s likelihood of being a walker. Results indicate that, for every unit increase in the composite scores measuring neighborhood walkability, individuals are 6–10% more likely to be in the walking category, with the Krizek measure found to have a modestly higher ratio. The individual components of the SEQOL Index were also investigated and several had significant odds ratios larger than 1 (see Table 3). In particular, the bivariate odds ratio for population density was the largest of all observed significant associations (OR = 1.212). The final multivariate models revealed the importance of composite measures over single component variables, particularly population density. These analyses produced a refined list of composite measures and components associated with walking (Table 4). When accounting for demographic factors, population density, which demonstrated the largest association in the bivariate analysis, was no longer shown to be associated with walking. However, the associations initially observed for the Krizek and SEQOL composite measures persisted. Both of the variable components obtained using Google sources for the composite SEQOL measure—proximity of destinations and percentage of sidewalks—were found to be significantly associated with the walking outcome in the presence of demographic data.

Discussion The results of our analysis yield several important insights for practitioners of public health and planning. First, our findings show stronger associations between composite measures and walking outcomes than between single component measures and walking outcomes. This implies

that more inclusive or comprehensive evaluations of the built environment, represented here in composite indices, are of greater relevance for assessing the impact of the built environment on local walking behaviors. This finding is suggestive of the most effective means of measuring walkability in general, and of what types of measures are likely to most accurately predict outcomes resulting from policies designed to enhance neighborhood walkability. While commonly employed in physical activity research, population density alone was not found to be a significant predictor of walking behavior in the presence of demographic data. This finding is in conflict with previous studies showing population density to be the most powerful predictor of walking outcomes.28–30 Second, a composite measure of the built environment derived from free, publicly available, and geographically extensive data sources was found to be a statistically significant predictor of walking outcomes. While not found to have the same predictive power in our study as a previously published measure developed from less accessible data sources, this public domain composite variable showed sufficiently strong predictive power to merit consideration for use by planning and public health researchers or practitioners lacking access to more complex measures. It is important to note that the varying predictive strengths between the SEQOL and Krizek NA composite measures may originate in their differing constructions. The Krizek NA measure uses 2 spatial units (grid cells and quarter-mile buffers) to evaluate the built environment while SEQOL uses a single, larger circular buffer (halfmile). Also though the 2 composite measures contain similar conceptual constructs, they are operationalized differently. For example, both include an urban form component related to connectivity. However, Krizek uses block area while SEQOL includes intersection density. While Krizek includes measures to capture separate aspects of proximity and connectivity, SEQOL includes twice the number of components. Finally, Krizek uses principal components analysis as a variable reduction

Table 3 Bivariate Odds Ratios of Making >10% of Trips via Walkinga Index

Odds ratio

95% CI

Krizek NA Index

1.098

1.023–1.179

SEQOL Walkability Index

1.064

1.010–1.121

Population density index (per/acre)

1.212

1.005–1.460

Employment density (emp/acre)

1.055

0.990–1.123

Destinations (#)

1.027

1.003–1.051

Intersections (#)

1.016

0.994–1.038

Transit (# stops)

1.027

0.999–1.057

Sidewalks (% of street length with sidewalks on 1+ sides)

1.032

1.003–1.061

Component

a Significant

predictors shown in bold; P < .05.


Table 4 Multivariate Model Results for the Odds of Walking Odds ratio Variable

Make 10% of trips by walking

95% CI

0.75

0.089–6.312

Model 1 Gender (ref = female) Race (ref = nonwhite)

0.462

0.067–3.194

Income (ref≤$50K)

1.168

0.182–7.485

Krizek NA Index

1.103

1.075–1.198

0.705

0.099–5.005

Model 2 Gender (ref = female) Race (ref = nonwhite)

0.645

0.115–3.606

Income (ref≤$50K)

0.809

0.170–5.695

SEQOL walkability

1.064

1.005–1.127

Gender (ref = female)

0.504

0.072–3.544

Model 3 Race (ref = nonwhite)

0.555

0.101–3.055

Income (ref≤$50K)

1.05

0.183–6.004

# of destinations

1.027

1.001–1.054

Model 4 Gender (ref = female)

0.886

0.120–6.544

Race (ref = nonwhite)

0.446

0.074–2.705

Income (ref≤$50K)

0.797

0.140–4.543

% of street length with sidewalks on 1+ sides

1.035

1.003–1.068

a Significant

predictors shown in bold; P < .05.

technique while SEQOL indices were created using a scaling approach with equal weighting. These differences in covariate design allowed us to compare different constructs of the same underlying theory of how the built environment influences walkability. Despite these differences, the 2 composite measures were highly correlated (r = 0.79 P < 0.001), and both showed significant association with walking outcomes. This result is important for planners and public health practitioners facing data availability and cost constraints. It implies that different composite measures may be employed with similar effectiveness. This allows practitioners to create locally derived metrics of walkability rather than relying on findings from other settings. Third, our SEQOL index was designed to examine the utility of a composite measure that could be more easily created for specific areas and used local data. Early studies of walkability focused on generalized measures of land use and walking outcomes averaged over large areas (ie, county-level values of population density and regional mode split data). Our disaggregated approach more specifically evaluates the built environment around specific locations of interest and thus minimizes ecological fallacy due to the use of generalized values.31 This was important since we are interested in

the association of walkability with individual walking behaviors and locations. Our half-mile radius buffers capture more of an individual’s walking trips;14 however, buffers created with shorter radii capture built environment attributes located closer to an individual’s place of residence or work and are assumed to have a greater influence on walking mode choice. Walkability studies have followed similar assumptions with regard to buffer size when using GIS measures of the built environment.32 More sophisticated analyses could use buffers created from the street network to combine connectivity of an area with proximity measures such as the number of destinations. Finally, we included several publicly available, geographically extensive, and free data sources, including 2 obtained from Google products: destinations (a measure of proximity) from Google Earth and sidewalks (a measure of connectivity) from Google Maps/Street View. Other metrics in the SEQOL index, such as the number of intersections or transit stops, can also be obtained from Google Maps. We estimate that, on average, measures of the 6 components around a single address could be collected from Google and the Census in less than 2 hours. Our combination methods were performed using Microsoft Excel rather than more sophisticated statistical


methods. Nonetheless, the SEQOL Index was shown to be a consistent predictor of individual walking behavior near residences. This finding suggests that municipalities of various sizes, levels of ability, and with varying access to data and resources should be able to create location specific walkability metrics on which to base decisions. Furthermore, community health research can combine such fine-scale measures of the built environment with meaningful, qualitative data from interviews and focus groups. Further study should be devoted to making such composite measures even simpler to use. As presented here, indices are created from a sample of local sites. The walkability index values can be compared against each other but it would be difficult to perform the assessment for a single location and get a meaningful metric. Future work should create universal index values related to the various components so that individual addresses could be assessed to determine walkability without needing to calculate composite measures for a sample of locations throughout the region. That is, a single address could assess each component and create a composite measure that could immediately be evaluated for walkability based on a uniformly applied scale. The majority of the component data for a location can be collected in a couple of hours from Google and other data providers over the web. This would extend the application of the tool to individual residents and neighborhood associations and allow them to argue for local improvements independent of prior action at the municipal level. The use of publicly available imagery and destination databases will continue to evolve and expand with time. Internet products that make use of Google’s data can already be used to more quickly and cost effectively gather component information for composite measure development. The use of Google data are not only more easily adapted and used for automated assessments of the built environment, but it is also already familiar to many users. One example is the popular real estate tool, Walkscore, which uses the Google Map’s application programming interface (API) to compile retail and restaurant information within a certain proximity to a location. The results are given on a 0 to 100 scale and could easily be incorporated into the SEQOL index. However, such tools should be used as a means of obtaining individual components for further combination into more comprehensive composite measures, and may require validation to assess reliability. In fact, when Walkscore ratings were considered for SEQOL participants’ addresses, we did not find a significant association with walking behavior (OR = 0.967 CI = 0.903, 1.035) in bivariate analysis. Our study did find 2 single components to exhibit significant associations with walking behavior; specifically, the 2 Google-derived components sidewalks and destinations. However, the magnitude of these associations is smaller than those of composite measures. Thus, we suggest that more comprehensive composite measures be used as tools for incorporating walkability into decisions concerning changes in the built environment.

Conclusion By encouraging a focus on composite measures, this study advocates for the concurrent assessment of 2 conceptual components of walkability: proximity and connectivity. Evidence of a synergistic relationship between these 2 types of measures suggests a need for considering both connectivity and proximity in decision-making. For example, investments to increase connectivity are most advisable where proximity is greatest. The consistency of findings across different composite indices also encourages practitioners to generate and use local measures rather than relying on single, uniform recommendations (eg, national averages) when assessing the built environment. Planners and public health researchers seeking to describe the built environment accurately and efficiently, as well as practitioners seeking to improve the walkability of their own jurisdictions should evaluate local built environments independently rather than applying findings from other areas. This study suggests that the use of Google imagery and spatial databases facilitates such work without sacrificing quality in assessments of the design of the built environment and walking outcomes. Acknowledgments The authors would like to acknowledge the support and effort of the entire SEQOL team, including Craig Zimring, Nicole Dubruiel, Karen Mumford, Julie Brand, Lu Yi, and Arthur Wendel. This publication was supported by Cooperative Agreement Number U48 DP 000043 from the Centers for Disease Control and Prevention to the Emory Prevention Research Center. The findings and conclusions in this journal article are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.

References 1. Saelens BE, Glanz K. Work Group I: Measures of the Food and Physical Activity Environment Instruments. Am J Prev Med. 2009;36(4 Suppl):S166–170. PubMed doi:10.1016/j. amepre.2009.01.006 2. Saelens BE, Frank LD, Auffrey C, Whitaker RC, Burdette HL, Colabianchi N. Measuring physical environments of parks and playgrounds: EAPRS instrument development and inter-rater reliability. J Phys Act Health. 2006;3:190. 3. 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(1):47–57. PubMed doi:10.4278/0890-117118.1.47 4. Frank LD, Stone B, Bachman W. Linking land use with household vehicle emissions in the central Puget Sound: methodological framework and findings. Transp Res Part D Transp Environ. 2000;5(3):173–196. doi:10.1016/ S1361-9209(99)00032-2 5. Rutt CD, Coleman KJ, Craig CL, et al. The impact of the built environment on walking as a leisure-time activity along the US/Mexico border. J Phys Act Health. 2005;2(3):257–271.


6. Forsyth A, Hearst M, Oakes JM, Schmitz KH. Design and destinations: factors influencing walking and total physical activity. Urban Stud. 2008;45(9):1973. doi:10.1177/0042098008093386 7. Rundle A, Diez Roux A, Freeman L, Miller D, Neckerman K, Weiss CC. The urban built environment and obesity in New York City: a multilevel analysis. Am J Health Promot. 2007;21(4):326–334. PubMed doi:10.4278/0890-117121.4s.326 8. Frank LD, Andresen MA, Schmid TL. Obesity relationships with community design, physical activity, and time spent in cars. Am J Prev Med. 2004;27(2):87–96. PubMed doi:10.1016/j.amepre.2004.04.011 9. Stone B, Rodgers MO. Urban form and thermal efficiency: how the design of cities influences the urban heat island effect. J Am Plann Assoc. 2001;67(2):186–198. doi:10.1080/01944360108976228 10. Kitamura R, Mokhtarian PL, Daidet L. A micro-analysis of land use and travel in five neighborhoods in the San Francisco Bay Area. Transportation. 1997;24(2):125–158. doi:10.1023/A:1017959825565 11. Boarnet MG, Day K, Alfonzo M, Forsyth A, Oakes M. The Irvine-Minnesota inventory to measure built environments reliability tests. Am J Prev Med. 2006;30(2):153–159. PubMed doi:10.1016/j.amepre.2005.09.018 12. Day K, Boarnet M, Alfonzo M, Forsyth A. The IrvineMinnesota inventory to measure built environments development. Am J Prev Med. 2006;30(2):144–152. PubMed doi:10.1016/j.amepre.2005.09.017 13. Pikora TJ, Giles-Corti B, Knuiman MW, Bull FC, Jamrozik K, Donovan R. Neighborhood environmental factors correlated with walking near home: using SPACES. Med Sci Sports Exerc. 2006;38(4):708. PubMed doi:10.1249/01. mss.0000210189.64458.f3 14. Cervero R, Duncan M. Walking, bicycling, and urban landscapes: evidence from the San Francisco Bay Area. American Public Health Association. 2003;93:1478–1483. 15. Cervero R, Kockelman K. Travel demand and the 3Ds: density, diversity, and design. Transp Res Part D Transp Environ. 1997;2(3):199–219. doi:10.1016/S13619209(97)00009-6 16. Krizek K. Residential relocation and changes in urban travel: does neighborhood-scale urban form matter? J Am Plann Assoc. 2003;69(3):265–281. doi:10.1080/01944360308978019 17. Rutt CD, Coleman KJ. Examining the relationships among built environment, physical activity, and body mass index in El Paso, TX. Prev Med. 2005;40(6):831–841. PubMed doi:10.1016/j.ypmed.2004.09.035 18. Owen N, Cerin E, Leslie E, et al. Neighborhood walkability and the walking behavior of Australian adults. Am J Prev Med. 2007;33(5):387–395. PubMed doi:10.1016/j. amepre.2007.07.025 19. Leslie E, Coffee N, Frank L, Owen N, Bauman A, Hugo G. Walkability of local communities: using geographic information systems to objectively assess relevant environmental attributes. Health Place. 2007;13(1):111–122. PubMed doi:10.1016/j.healthplace.2005.11.001

20. Saelens BE, Sallis JF, Black JB, Chen D. Neighborhoodbased differences in physical activity: an environment scale evaluation. American Public Health Association. 2003;93:1552–1558. 21. Forsyth A, Oakes J, Schmitz K, Hearst M. Does residential density increase walking and other physical activity? Urban Stud. 2007;44(4):679. doi:10.1080/00420980601184729 22. Frank L, Saelens B, Powell K, Chapman J. Stepping towards causation: do built environments or neighborhood and travel preferences explain physical activity, driving, and obesity? Soc Sci Med. 2007;65(9):1898–1914. PubMed doi:10.1016/j.socscimed.2007.05.053 23. Brownson R, Hoehner C, Day K, Forsyth A, Sallis J. Measuring the built environment for physical activity state of the science. Am J Prev Med. 2009;36(4 Suppl):S99–123. PubMed doi:10.1016/j.amepre.2009.01.005 24. Mumford KG, Contant CK, Weissman J, Wolf J, Glanz K. Changes in physical activity and travel behaviors in residents of a mixed-use development. Am J Prev Med. 2011;41(5):504–507. 25. Bergman P, Grjibovski A, Hagstromer M, Sallis J, Sjostrom M. The association between health enhancing physical activity and neighbourhood environment among Swedish adults - a population-based cross-sectional study. Int J Behav Nutr Phys Act. 2009;6(1):8. PubMed doi:10.1186/1479-5868-6-8 26. Berke EM, Koepsell TD, Moudon AV, Hoskins RE, Larson EB. Association of the built environment with physical activity and obesity in older persons. Am J Public Health. 2007;97(3):486–492. PubMed doi:10.2105/ AJPH.2006.085837 27. Kerr J, Rosenberg D, Sallis JF, Saelens BE, Frank LD, Conway TL. Active commuting to school: Associations with environment and parental concerns. Med Sci Sports Exerc. 2006;38(4):787–794. PubMed doi:10.1249/01. mss.0000210208.63565.73 28. Cervero R. Mixed land-uses and commuting: evidence from the American Housing Survey. Transp Res Part A Policy Pract. 1996;30(5):361–377. doi:10.1016/0965-8564(95)00033-X 29. Frank L, Pivo G. Impacts of mixed use and density on utilization of three modes of travel: single-occupant vehicle, transit, and walking. Transp Res Rec. 1994;1466:44–52. 30. Ross C, Dunning A. Land use transportation interaction: an examination of the 1995 NPTS data. Prepared for US Department of Transportation. Atlanta: Georgia Institute of Technology; 1997. 31. Handy S. Methodologies for exploring the link between urban form and travel behavior. Transp Res Part D Transp Environ. 1996;1(2):151–165. doi:10.1016/S13619209(96)00010-7 32. McGinn AP, Evenson KR, Herring AH, Huston SL. The relationship between leisure, walking, and transportation activity with the natural environment. Health Place. 2007;13(3):588–602. PubMed doi:10.1016/j.healthplace.2006.07.002